Empirical Studies in Comparative Politics

EMPIRICAL STUDIES IN COMPARATIVE POLITICS Empirical Studies in Comparative Politics Edited by MELVIN J. HINICH University of Texas-Austin and MICHAEL C. MUNGER Duke University Reprinted from Public Choice, Volume 97, No.3, 1998 Springer-Science+Business Media, B.V. A c.I.P. Catalogue record for this book is available from the Library of Congress. ISBN 978-1-4419-5072-7 ISBN 978-1-4757-5127-7 (eBook) DOl 10.1007/978-1-4757-5127-7 Printed on acid-free paper All rights reserved ©1998 Springer Science+Business Media Dordrecht Originally published by Kluwer Academic Publishers in 1998 No part of the material protected by this copyright notice may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying, recording or by any information storage and retrieval system, without written permission from the copyright owner. CONTENTS MJ. Hinich and M.C. Munger, Empirical studies in comparative politics F.U. Pappi and G. Eckstein, Voters' party preferences in mUltiparty systems and their coalitional and spatial implications: Germany after unification 11 N. Schofield, AD. Martin, K.M. Quinn and AB. Whitford, Multiparty electoral competition in the Netherlands and Germany: A model based on multinomial probit 39 S.E. Macdonald, G. Rabinowitz and O. Listhaug, Issue competition in the 1993 Norwegian national election 77 K.T. Poole and H. Rosenthal, The dynamics of interest group evaluations of Congress 105 J.W. Endersby and S.E. Galatas, British parties and spatial competition: Dimensions of party evaluation in the 1992 election 145 J.F-S. Hsieh and D. Lacy, Retrospective and prospective voting in a one-party- dominant democracy: Taiwan's 1996 presidential election 165 M.J. Hinich, M.e. Munger and S. de Marchi, Ideology and the construction of nationality: The Canadian elections of 1993 183 M.M. Kaminski, G. Lissowski and P. Swistak, The "revival of communism" or the effect of institutions?: The 1993 Polish parliamentary elections 211 J.K. Dow, A spatial analysis of candidate competition in dual member districts: The 1989 Chilean senatorial elections 233 A.M.A van Deemen and N.P. Vergunst, Empirical evidence of paradoxes of voting in Dutch elections 257 M. Myagkov and P.e. Ordeshook, The spatial character of Russia's new democracy 273 Public Choice 97: 219-227, 1998. © 1998 Kluwer Academic Publishers. 219 Empirical studies in comparative politics MELVIN J. HINICW & MICHAEL C. MUNGER 2 1Department of Government, University of Texas-Austin, Austin, TX 78712, U.S.A.; e-mail: hinich@maiI.LA.utexas.edu; 2Department of Political Science, Duke University, Durham, NC 27708, U.S.A.; e-mail: munger@acpub.duke.edu The study of political science has been substantially transfonned by the work of public choice scholars over the past 40 years ... [Butl public choice in political science is seen as an offshoot, or related discipline, rather than as a substantive field in and of itself (Dow and Munger, 1990: 604) Nowhere in political science is the above statement truer than in the study of comparative politics. There have been many publications by public choice scholars, and many more by researchers who are at least sympathetic to the public choice perspective. Yet little of this work has been integrated into the main stream of comparative political science literature, and still less has made its way onto graduate readings lists. That is not to say there are not examples of successful recent works on the interplay of economics and politics, because there are. Three notable examples come easily to mind: Verdier (1994), Haggard and Kaufman (1995), and Bates (1997) have each made a substantial, and widely recognized, contribution in political science. However, this work is considered (and probably conceived) to be "international political economy", rather than comparative politics. In general, it is far more difficult to find examples of empirically oriented studies of the politics, bureaucratic organization, and regulated economies of particular nations in the canon of the comparativist. This is not because there is no high quality work. Space limitations preclude a representative literature review, but public choice scholars have produced first rate, book-length studies on budgetary politics (Kraan, 1996), coalition politics (Laver and Schofield, 1991; Shepsle, 1994), and voting systems (Grofman and Lijphart, 1992; Cox, 1997). One difficulty, of course, is that the attention within the public choice community to a genuinely comparative perspective is recent. Another is that empirical testing of comparative models requires sources of data not as easily available as data on U.S. politics. [ 1] 220 If these are the reasons that public choice has not yet had a significant impact on political science, then it may just be a matter of time. There is another problem, one that can be addressed immediately. The scathing critique Green and Shapiro (1994) direct toward "rational choice" focused on a lack of empirical testing. Now, it isn't true that public choiceoriented empirical research has never been published in comparative politics journals (see, e.g., Cox and Niou, 1994; Lin et al., 1996; Montinola et al., 1995; or Ordeshook and Shvetsova, 1997). But there hasn't been very much. That is where this issue of Public Choice comes in. In this special issue, the editors hope to make available to students of comparative politics a compendium of work that meets two criteria: (1) In every case, a model of human behavior or of institutional impact is specified. (2) Also in every case, this model is confronted with data appropriate for evaluating whether this model is useful for understanding politics in one or more nations. Theoretical framework The political systems of the world exhibit many more differences than do economic systems. This variation is both an opportunity and a trap. The fact that political systems contrast so sharply represents a chance to "explain" the differences. On the other hand, it is easy to get trapped into studying unimportant details, because it is by definition those characteristics of a polity that are different that make it interesting. If one takes the conventional descriptive approach to the study of political systems, one may never get beyond these differences. After all, the null hypothesis has to be that each nation, at each point in time, is unique, so that there are no useful generalizations. To make real progress we need to quantify the similarities and differences and build theories which go beyond simple correlations. Of course, political scientists have recognized this since the 1950s, led by such works as Duverger (1951) and Almond (1956). What we want to argue (humbly, we hope) is that there is room for the contributions of public choice scholars in advancing this program, and that the unique abilities of the public choice approach to unify and generalize will make this contribution an important one. There have been significant attempts at high-level synthesis already (see, e.g., Rowley and Vachris, 1994; or Schneider, 1995), but this work has only indicated a direction. We would like to provide a public choice view at the level of nations, not of systems. [2] 221 Organization of this issue Three distinct factors characterize a nation's political system. These factors are as follows: • • • Election procedures: The set of voting rules, in most cases a variant either of plurality or proportional representation. Each "variant", however, has unique and perhaps unexpected properties. Institutions and legislature: The rules that govern agenda formation and voting methods in the legislature, and the relationships between an executive branch (if a separate entity exists) and the legislature. Ideological cleavage: The collection of issues and cleavages that determine the nature of ideological conflict for that nation. Empirically, policy positions cluster, but the way that they cluster can differ dramatically over space and time. The first two of these are well-established as explananda in public choice research; the third has received a great deal of attention in the last few years (Hinich and Munger, 1994; Poole and Rosenthal, 1997). This spatial theory of mass electoral competition implies that "issues" link with a small number of latent ideological dimensions. One dimension links with issues that constrain the economic system. Another dimension links with issues which determine the personal, or civil, freedoms. In some countries this cleavage is over religious beliefs while in others it is more diffuse. When national security issues loom large, there is another dimension linking with these security issues. In contemporary Israel, for example, the major cleavage is about the political status of the Arab West Bank. Spatial theory explains the revealed patterns of political preferences for ordinary votes and elites. The structure allows a theorist to study how changes in voting rules with determine outcome given a spatial configuration. Several of the papers in this special issue show how spatial theory can be applied to study the politics of different nations. We want to encourage other researchers, who may find these results intriguing, to test and refine the claims made in these papers. Papers in the special issue Each of the first two papers uses a classic public choice approach: What if? The first paper, by Franz Pappi and Gabriele Eckstein, asks a deceptively simple question: How should party preferences of voters in a multi-party system be measured, compared, and aggregated? The paper has two parts, with the [3] 222 first focusing on pairwise comparisons of parties from surveys of voters Pappi and Eckstein derive implied coalition preferences, and find that if voters could choose over coalitions, the actual governing coalition of CDU/CSU and FDP is dominated by at least two other combinations of parties. The second part of the paper analyzes the common structure of the party rankings. Interestingly, though many voters do not fit the classical "left-right" ideological scale, there exists no competing means of organizing a one-dimensional ordering of parties that correctly classifies more voters. Further, Pappi and Eckstein find that what was West Germany is more "ideological", by this standard, than the former East Germany. Norman Schofield, Andrew Martin, Kevin Quinn, and Andrew Whitford also use a "what if" approach, though their question focuses more on voter choice among parties than on coalition outcomes. They want to know what if voters had the preferences registered in the 1979 Euro-barometer surveys, and European elite-study data, and parties chose positions so as to maximize vote share. The paper incorporates both pre-election electoral estimation and postelection expectations about coalition formation. Schofield et al. show quite convincingly that maximizing expected vote share is generally dominated as a rational party strategy, and demonstrate further that the parties in the Netherlands and Germany did not appear to follow such a strategy. Every collection should contain a few dissenting voices. Stuart Macdonald, George Rabinowitz, and ala Listhaug advance a longstanding research program, using "directional theory", in their study of the 1993 elections in Norway. Long skeptical of proximity voting and the classical public choice model, the authors find an interesting potential counterexample to their directional approach in the strong showing of a traditional centrist party, the aptly named "Center Party", representing agrarian interests. Directional theory implies that voters ask two questions of a party or candidate: "Are you on my side?" and "Are you responsible?" The results illustrate an interesting tension in directional theory: To answer the first questions successfully, parties must stake out strong positions and look for divisive "wedge" issues. But to answer the second question, which has more to do with the ability to run the country, parties must often converge toward the center, or toward expert opinion, on many other issues. Macdonald et al. aver that the conventional wisdom about the interplay of elections and governing may have it reversed: elections cause conflict, but having to govern forces moderation and compromise. Keith Poole and Howard Rosenthal also have a long-term, and very successful, research agenda. The paper they have contributed to this volume ties up a loose end in this agenda, by considering interest group ratings, and the implied shared (but latent) ideological dimension, as the product of a dynamic data generating process. Using more than 200,000 observations by 59 interest [4] 223 groups over the period 1959-1981, Poole and Rosenthal are able to demonstrate three results not found in any of their other work. First, the spatial coordinates of the interest groups and members of the U.S. Congress can be approximated as polynomial functions of time, otherwise showing surprising consistency. Second, the first recovered ideological dimension accounts for about three-quarters of the variance in group rating of members, with the second dimension adding only about an additional five percent in variance accounted for. Third, both interest groups and members of congress appear to be logically, as well as temporally (see above) consistent: members and groups are generally either liberal on both dimensions, or conservative on both. The next five papers are analyses of elections in Britain, Taiwan, Canada, Poland, and Chile. James Endersby and Steven Galatas employ data from the 1992 British Election Study to analyze the factors that contribute to voter attitudes toward the Conservative and Labour parties. Many scholars of British politics have claimed the foremost features of voters are partisan identification and social class. The 1992 election is a particularly interesting test case, as it saw parties move more toward the ideological center of the British electorate. Perceptions of the parties, however, are found by Endersby and Galatas to be multidimensional, and to have more to do with issues than traditional party identification would imply. John Fuh-sheng Hsieh, Dean Lacy, and Emerson Niou ask a venerable question (are voters prospective, or retrospective?) in an entirely new setting. Taiwan is a fascinating context for a study of retrospective voting, as there has been only one party (KMT, or Kuomintang) for the past five decades. Only recently, in fact, has Taiwan even begun to held genuinely contested elections, with the end of martial law and the lifting of many of the restrictions on dissent and on freedom of the press. How will voters react? Hsieh et al. present the first scholarly analysis of the 1996 election, and find voters appear relatively more prospective in all of the statistically significant variables used to predict vote choice, including managing the economy, ethnic relations, domestic safety, and international security. Melvin Hinich, Michael Munger, and Scott de Marchi examine the Canadian national election of 1993, trying to measure the changing ideological context. Canadian politics has always been different, because Canada is a combination of two very ditIerent cultures: the French-identifying Quebec, and the British or aboriginal peoples in the rest of the nation. The main principle around which political conflict has been organized, at least since 1960, has been regionalism-nationalism-continentalism. The debate over national industrial policy, trade restrictions, and subsidies for agriculture and mining have all been handled fairly well by this cleavage. However, the issue of [5] 224 autonomy for, or even outright secession by Quebec, threatens the stability of a national politics organized around a distributional or redistributional dimension. Hinich et al. find that attitudes toward Quebec sovereignty constitute a coherent dimension of conflict, quite distinct from economic or social issues. Further, they find that the Quebec sovereignty issue/dimension, alone, is capable of classifying more than three-quarters of the vote choice of Quebec voters in supporting, or opposing, the increasingly powerful Bloc Quebeoise. Marek Kaminski, Gregorz Lissowski, and Piotr Swistak analyze the 1993 Polish parliamentary elections. The continued electoral attractiveness of neocommunist parties seems surprising, given the strident, and nearly universal, rejection of communism as a form of social organization in Poland in the 1980s. Using game theory as a simulation tool, Kaminsky et al. demonstrate how sensitive the Polish electoral system is to small, even trivial, changes in expressed preferences. Where some observers have attributed the resurgence of post-communist candidates and candidates of the "Peasant Left" to a shift back to the left by voters, Kaminski et al. argue that the size of the shift was actually too small to have caused the electoral victory by the left, and that this victory would not have happened if (1) previous election laws had still been in place, or (2) the fractious parties of the Right had been more unified. The authors conclude that the need for "fine-tuning" electoral institutions to avoid exaggerated swings in partisan representation may be a generic problem in developing democracies. Jay Dow analyzes privately collected data on the 1989 Senate elections in Chile. These elections use the d'Hondt system to choose two senators from each district. This form of election rule appears to cause candidates on the right to locate on the periphery of the space, and gives them few incentives for more centrist positions. Dow also analyzes the apparent underlying ideological dimensions that organize political conflict in Chile, finding that one, or at most two, latent dimensions characterized voter positions in the election. Ad van Deemen and Noel Vergunst investigate the empirical occurrence of two types of voting paradoxes in Dutch elections. The first type of paradox is the Condorcet paradox, where a party a majority prefers to another party may receive fewer seats, because no majority rule equilibrium exists. The second type is the "majority-plurality" paradox, where a Condorcet winner exists, but does not win a majority of seats because of the idiosyncrasies of voting rules. Van Deemen and Vergunst illustrate the possibility of three alternative majority-plurality paradoxes and seek evidence of the actual occurrence of each type of paradox in the national elections held in the Netherlands in 1982, 1986, 1989 and 1994. In no case did the authors find an instance of a true Condorcet paradox (i.e., the aggregate relation appears transitive), but they did find that voting rules appeared to cause many cases of majority-plurality [6] 225 paradoxes. The most frequent type of voting paradox they found was the type where a party which would receive a majority against another party gets fewer seats as a result of the actual multi-party election. Misha Myagkov and Peter Ordeshook, the final paper in the collection, investigate the broad outlines of political conflict in Russia. The authors note that previous research on Russian elections has used an approach similar to that employed in the U.S. in the 1950s and 1960, focusing on attitudes toward candidates or self-reported participation histories. Myagkov and Ordeshook seek to identify the issues, and candidate locations, in a spatial context, using county (rayon) level data. The primary issue in Russian political conflict continues to be beliefs about "reform", with a potential second main dimension, "nationalism", so far having only limited power to organize opinion in the electorate. The most interesting intercandidate contest the authors identify is between Lebed (pro-reform, mildly nationalist) and Zhirinovsky (antireform, and virulently nationalist). If this type of conflict becomes dominant, the stability of an electoral system built around the reform issue may be questionable. Each of these papers offers evidence on at least one, and in many cases two, of our differentiating factors election rules, legislative institutions, and ideological cleavage. While the papers reach no consensus, there is an impressive array of theory and evidence brought to bear on these differentiating factors. We thank the authors for their impressive contributions. Conclusion In closing, it should be noted without equivocation that we reject Green and Shapiro's (1994) claim that empirical testing is privileged as an epistemological basis for understanding, as this is clearly a confused view of the philosophy of science. It is possible, however, that Green and Shapiro are right for the wrong reasons: public choice studies of comparative politics must build a record of success in both prediction and prescription if this work is to be recognized as a useful model of theorizing. Peter Ordeshook makes a comparison between science (identifying first principles) and engineering (solving practical problems), and claims we have too many "scientists" and too few "engineers" for the rational choice approach to have much relevance to world politics. We may have learned some things about voting systems, but little has been done to make this learning available to people struggling with the problem of reforming a political system, or designing a new democracy. Unfortunately, there are profoundly important problems that require our attention: designing a stable democratic federalism for Russia or Ukraine, [7] 226 understanding the nature of strategic threats in Central Europe, and formulating viable political constitutions nearly everywhere. Such matters are, at present, mostly the purview of politicians and pundits. With some notable exceptions, political scientists have reserved for themselves the role of bystanders who merely describe or offer after-the-fact explanations. Of course, one might object to this characterization by noting that, as political scientists, our job is merely to develop theoretical propositions and methodologies that can assist those who act. But we cannot do the job if we are unwilling to take the bits and pieces of theory and methodology available to us and propose alternative institutional designs and constitutional structures as solutions to specific problems (Ordeshook, 1996: 187-188). In closing, then, let us echo Ordeshook's challenge. Some bridges stand, and some planes fly. Engineers have applied theoretical principles to these practical problems, and solved them. Public choice scholars know a lot about why some governments work, but all too little attention has been focused on the solution of practical problems in other nations. In this volume, we have gathered a collection of papers analyzing the political systems of ten nations. It is our hope that this work will provoke a self-conscious effort to compare, and investigate, the public choice of comparative politics. References Almond, G. (1956). Comparative political systems. Journal of Politics 18: 391-409. Bates, R. (1997). Open economy politics: The political economy of the World Coffee Trade. Princeton, NJ: Princeton University Press. Cox, G. (1997). Making votes count: Strategic coordination in the world's electoral systems. New York: Cambridge University Press. Cox, G. and Niou, E. (1994). Seat bonuses under the single nontransferable vote system: Evidence from Japan and Taiwan. Comparative Politics 26: 221-235. Dow, J. and Munger, M. (1990). Public choice in political science: We don't teach it, but we publish it. PS 23: 604-610. Duverger, M. (1951). Political parties: Their organization and activity in the modern state. London: Methuen. Green, D. and Shapiro, 1. (1994). The pathologies of rational choice theory: A critique of applications in political science. New Haven: Yale University Press. Grofman, B. and Lijphart, A. (1992). Parliamentary versus presidential government. New York: Oxford University Press. Haggard, S. and Kaufman, R. (1995). The political economy of democratic transitions. Princeton, NJ: Princeton University Press. Hinich, M. and Munger, M. (1994). Ideology and the theory of political choice. Ann Arbor: University of Michigan Press. Kraan, D.-J. (1996). Budgetary decisions: A public choice approach. New York: Cambridge University Press. [8] 227 Laver, M. and Schofield, N. (1991). Multiparty government: The political of coalition in Europe. New York: Oxford University Press. Lin, T.-M., Chu, Y.-H. and Hinich, MJ. (1996). Conflict displacement and regime transition in Taiwan: A spatial analysis. World Politics 48: 453-481. Montinola, G., Qian, Y. and Weingast, B. (1995). Federalism, Chinese style: The political basis of economic success in China. World Politics 48: 50-81. Ordeshook, P.C. (1996). Engineering or science: What is the study of politics? In J. Friedman (Ed.), The rational choice controversy, 175-188. New Haven: Yale University Press. Ordeshook, P.e. and Shvetsova, O. (1997). Federalism and constitutional design. Journal of Democracy 8: 27-42. Poole, K. and Rosenthal, H. (1997). Congress: A political economic history of roll-call voting. New York: Oxford University Press. Rowley, C.K. and Vachris, M.A. (1994). Why democracy does not necessarily produce efficient results. Journal of Public Finance and Public ChoicelEconomia Delle Scelte Pubbliche 12: 95-111. Schneider, F. (1995). Is there a European public choice perspective? Kyklos 48: 289-296. Shepsle, K. (1994). Cabinet ministers and parliamentary government. New York: Cambridge University Press. Verdier, D. (1994). Democracy and international trade. Princeton, NJ: Princeton University Press. [9] Public Choice 97: 229-255, 1998. 229 © 1998 Kluwer Academic Publishers. Voters' party preferences in multiparty systems and their coalitional and spatial implications: Germany after unification FRANZ URBAN PAPPI & GABRIELE ECKSTEIN University of Mannheim, Department of Political Science I, P.O. Box, D-68131 Mannheim, Germany; e-mail: fupappi@sowi.uni-mannheim.de;geckstein@sowi.uni-mannheim.de Abstract. How should party preferences of voters in a multiparty system be measured, compared and aggregated? We use city block metric of distances between the pairwise comparisons of the five German parties (1995 survey data for West and East Germany). Neither in West nor in East Germany, a party gains the absolute majority of voters' preferences. We derive coalition preferences from the party rankings; the governing coalition of CDU/CSU and FDP is not the winner, compared with other feasible coalitions of the German party system. But the party rankings of the CDU/CSU-FDP coalition leaners are more homogeneous than other groups of coalition leaners. In the second part of the article, we analyze the common structure of all consistent party rankings. Do voters apply the same criteria to evaluate the political parties? Although only a slight majority of individual rankings fit the often used ideological left-right scale, there does not exist a competing one-dimensional order of the parties that would capture more voters. The joint scale of individual party rankings is interpreted as the collective order which facilitates political orientation of voters. This collective order is more pronounced in West than in East Germany where individuals are almost as consistent in their party rankings but where the rankings fit the collective order less well than in West Germany. 1. Introduction In parliamentary systems with proportional representation and multiple parties, the prime minister and his or her cabinet are normally dependent on the support and confidence of more than one parliamentary party. These coalition governments are formed anew after each election in which voters choose parties and not coalitions or governments. Coalitional options are discussed during election campaigns, but not all options turn out to be feasible once the election:s returns become known. Contrary to two-party parliamentary systems, the electorate's signals for a new government are ambiguous, giving the party leaders some leeway in coalition bargaining. But in modern democracies with structured party systems and attentive publics, there also exist restraints for the coalition game. As far as parties are policy seeking, they look for coalition partners close to them in policy or ideological terms. On the other hand, the re-election motive of politicians guarantees that party [ 11 ] 230 leaders will seldom act contrary to what they have promised during the election campaign. In order to fulfill most of their promises, parties not only have to enter a winning coalition, but they have to reach a coalition agreement that puts the promised policies on the government agenda. The construction of policy spaces (cf. Laver and Hunt, 1992) is a very helpful tool to predict the party composition of coalition governments (cf. Laver and Schofield, 1990; Schofield, 1995), since distance in these spaces indicates the difficulties in reaching agreements. What makes policy spaces even more attractive is the Downsian assumption of parties competing for votes in those spaces, trying to find optimallocations under given distributions of voters' ideal points (Downs, 1957; Enelow and Hinich, 1984; Shepsle, 1991). Students of European multiparty systems picked up the Downsian spatial framework first of all for descriptive purposes: What is the ideological party space underlying sympathy ratings of parties (cf. for Germany Norpoth, 1979; Pappi, 1973, 1983). How do national electorates perceive the positions of their parties on a self-anchoring leftright scale and what is the shape of voters' self placements on this same scale (Inglehart and Klingemann, 1976). What is the predictive power of the distance between a voter's ideal point and the perceived party positions on a self-anchoring left-right scale for party preference (van der Eijk, Franklin and Oppenhuis, 1996: 354)? Concepts as the equilibrium position of parties or of the electoral heart (Schofield, 1993) are not strictly applied in empirical studies, but some authors have systematically analyzed the relationships between changes of party positions and voter reactions (cf. Nannestad, 1994). A strict empirical application of spatial models to determine party positions and voter ideal points aiming finally at good predictions of voting behavior has to be based on a correctly specified statistical model. Hinich and his collaborators have proposed factor analytic techniques to identify latent ideological dimensions with either party sympathy scores or perceptions of party positions on a set of issue scales as input (cf. Enelow and Hinich, 1984: 207-215). We applied both approaches to German data, with mixed results. The two-dimensional configuration of parties is very robust and the first and most important dimension is easy to interpret as a socio-economic left-right dimension (cf. Pappi, 1990, 1992; Eckstein, 1995). But the predictive power of the distance between a voter's bliss point and his or her closest party for electoral choice is poor, the wrong predictions running up to 50%. For party sympathy scores we are able to improve the model fit by adding a nonpolitical factor as suggested by Enelow and Hinich (1984: 82-90): competence to rule the country. This factor favors the two larger parties of the Christian Democrats (CDU/CSU) and Social Democrats (SPD) being the ones whose leaders are perceived as candidates for chancellorship. Even then, however, [12] 231 the necessary second ideological dimension raises problems of interpretation. As our analyses of the perceptions of party positions on issue scales show, a major problem is population heterogeneity. The assumption of a common ideological space may be violated for larger subgroups whose judgements are based on a different - one-dimensional - view of the party system. The result would be a two-dimensional space at the aggregate level, consisting of the combination of the one-dimensional views of two population subgroups. On the other hand, experts (cf. Laver and Hunt, 1992) do indeed diagnose a two-dimensional space for Germany, due mainly to the liberal party (Free Democrats, FDP) with its position right of the CDU in socio-economic terms and in-between CDU and SPD in other domestic policies: law and order issues, abortion and similar issues sometimes categorized as social issues (cf. Enelow and Hinich, 1984: 182-205). In this case, the two dimensions result from different party locations on two ideological or issue dimensions, as more or less unanimously perceived by the same voters. In this paper, we shall use an alternative route to learn about the coalitional and spatial implications of voters' party ratings or rankings focusing on overall party preferences and not on issue perceptions or issue preferences. We will not accept the available data as input for statistical data reduction techniques, but we shall first test the preference judgements of East and West German respondents for individual consistency, and then aggregate only the consistent party preferences to study the coalition leanings of these respondents and we shall then construct party spaces analytically starting from consistent party rankings as opposed to statistical data reduction. Aggregating individual preferences leads to a social choice which, under specific conditions, can be compared with the policy position of the coalition government negotiated by the party leaders. 2. The party preference concept and its measurement Party preference is often used as a self-evident concept which has not to be defined. We depart from this practice and explicate our usage of the concept both in substantive and measurement terms. We assume that citizens evaluate parties all the time and that the results of this ongoing process at a specific point in time are the actual party preferences of these citizens. This is why party preference questions are always meaningful, irrespective of whether an election is upcoming or we are in the middle of the legislative period. Evaluations can be based on different criteria, on policy preferences, ideologies, impressions about the party's performance in government, characteristics of party leaders etc. Whenever political messages have party cues, the citizens add and substract evaluation points from the accounts [13 ] 232 they have for the parties. Fiorina (1981: 84) used this idea to reconceptualize party identification as a "running tally of retrospective evaluations of party promises and performance". Transferring this idea to multiparty systems, calls specific attention to two points: (1) Each party must have a separate account, so that the plus point for one party is not automatically a minus point for the other party, and (2) the evaluation criteria for the different parties need not necessarily be the same. We shall have to come back to this point when we discuss consistent preferences and, later, their spatial implications. Conventional measures of party preference in multiparty systems are often incomplete due to the restriction of asking only for one most preferred party. The frequently used alternative of rating the sympathy towards each party separately, asks for a special criterion of party evaluation and has the disadvantages of measurement by fiat. The assumption of equal intervals cannot be tested, one has to assume that the respondents used the scales consistently. We rely instead on paired comparisons between the parties, asking each respondent which of each pair of parties she prefers or whether he is indifferent. With these data we are able to construct a preference relation over a set of parties, resulting at least in a weak ordering of the parties. The elements of a set are weakly ordered if the pairwise preference relations between the elements are reflexive, complete and transitive. A linear or strict order satisfies the same conditions and is, in addition, antisymmetric, allowing no ties between elements (cf., e.g., van Deemen, 1991). In our case, the elements are political parties and the reflexive, complete and transitive preferences between the parties are those of individual citizens. Restricting our analysis to the subset of respondents with weak or strict preference orders leads to the classical social choice problem of how to aggregate the set or profile of individual preferences to get a collective choice. In this article, we are not interested in voting systems; we shall use, however, the aggregation results of different procedures as information about the support level of single parties and party coalitions within the electorate and as evidence of the dominant policy dimension at a specific period. For this latter purpose, we shall apply a city block metric of distances between strict and/or weak preference orders (Kemeny and Snell, 1963: 9-23) and the notion of betweenness of preferences to recover linear profiles (van Deemen, 1991: 194-204). A preference order can be represented in a dominance matrix A with i (i = 1,2 ... m) rows and j (j = 1,2... m) columns, i and j indexing the parties, and an entry of aij = 1 if party i is preferred to j, aij =-1 if j dominates (is preferred [ 14 ] 233 = to) i, and aij 0 for indifference. Then Kemeny and Snell define the following distance function, in the literature also known as Hamming distance function (cf. van Deemen, 19991: 194-196), between two preferences or dominance matrices A and B. . dlstance(A, B) = 1 "2 L I aij - bijl ij For a group of people, their median preference or consensus ranking is the one to which the sum of distances is a minimum, and the heterogeneity of their preferences can be computed analogously to the mean path distance of a weighted graph. Let d ij be the distance between preference i (i = 1,2... m) and j (j = 1,2 ... m) and fi and fj indicate the respective frequencies of these to preferences (2:: fi = 2:: fj = n), then heterogeneity (h) of the preferences of this group of n people is defined as follows: h= L L d lJj/n i 2 i We shall use this measure to characterize the heterogeneity of various groups of coalition supporters. The last building block we need for our analysis is the concept of a linear profile, that is a set of individual preferences which can be combined to form a path. Following van Deemen's proof (1991: 201), we only exemplify the concept, with preferences for three parties a, b, c, and four voters (see Table 1). For three parties, a strict preference consists of three ordered couples, the set of voter 1 being e.g. {<a,b>,<a,c>,<b,c>}. The preference relation of this voter is between that of voter 2 and that of voter 4, since his or her ordered couples are a subset of the union of the sets of voters 2 and 4 and the intersection of these two sets {<a,c>} is a subset of the set of voter 1 (definition 7.9 in van Deemen, 1991: 200). A simple rule to verify betweenness is the addition rule for the respective distances (van Deemen's theorem 7.2), for our example: = A linear profile then consists of preferences i, j, k 1,2 ... n and i ::: j ::: k such that j is in-between i and k (van Deemen's definition 7.10). The other linear profiles of our example are shown in Table I: 2 is in-between 1 and 3, 4 is in-between 1 and 3. But all four preferences cannot be combined to form a linear profile. [ 15] 234 Table 1. Examples of linear profiles for four voters {1,2,3,4} and three parties {a,b,c} Preferences Distances 1 2 Voter 1: a>-b>-c 1 0 2 Voter 2: b>-a>-c 2 Voter 3: c>-b>-a 3 Voter 4: a>-c>-b 4 2 6 2 Linear profiles 4 0 3 6 4 4 4 0 4 4 4 0 2 Corresponding joint party scale 2 distances 4 2 3 4 2 voters midpoints a voters 1 alb a: c I I I I I I I I I I I I I b:e b , 2 , I I I I I I , 3 distances no joint scale 3 voters 4 2 2 4 I _____________ 4 2 distances J no joint scale 3 voters If the preferences forming a linear profile end up in only two least-preferred parties, not only the preferences but also the parties can be arranged on a straight line. This is Coombs (1964) unfolding idea for one dimension, exemplified by the joint scale a, b, c corresponding to the linear profile of the set of voters {I ,2,3 }. The ideal point of voter 1 is somewhere left of the midpoint between a and b on the joint scale. Voter 2 already went across this midpoint resulting in a preferences with b as the closest party, then a and c being last. Going further right on the joint scale, the next midpoint is the one between a and c. Once we go across this midpoint, we get a preference which is missing in our example: b :>- c :>- a. It would be in-between voter 2 and 3, not violating the linear profile, since it has a distance of 2 to both 2 and 3. We shall use the concept of linear profiles and its correspondence with unfolding to derive spatial implications of party preferences in multiparty systems. Distinguishing between weak and strict preference and indifference, in empirical studies one has to decide which preference judgements the respondents have to make. They Can be asked whether party a is at least as good [16] 235 for them as party b (weak preference) for all ordered pairs of parties, that is both the pair a, b, and the pair b, a. We have chosen the other alternative and asked the respondents whether they (strictly) prefer a to b, b to a or whether they are indifferent, restricting the pairs to be nonredundant dyads. The burden of proof for consistent judgements rests on the indifference relation, since respondents are supposed to distinguish between indifference between two known parties, noncomparability, and lack of knowledge which should prevent a judgement (cf. Fishburn, 1970: 12). Thus a direct question concerning indifference may overestimate the completeness of the answers because respondents hesitate to admit their unfamiliarity with specific parties. But the transitivity of judgements can be tested, both for strict preference and for indifference. We will analyze paired comparison data on the preferences of West and East German respondents concerning give parties. The parties are the Christian Democrats (CDU together with its Bavarian sister party CSU), the Social Democrats (SPD), the Free Democrats (FDP), the united party of The Greens and Biindnis '90, the latter being a branch of the opposition movement of the late GDR, and the Democratic Socialists (PDS) as the offspring of the former East German governing communist party. These five parties hold all the seats in the Federal Parliament (Bundestag) of the 13th legislative period, for which Kohl was elected chancellor by the Christian Democrats (294 seats) and the Free Democrats (47 seats). The SPD is the largest opposition party (252 seats) with the Greens as its future possible ally (49 seats). The PDS has its strong hold almost exclusively in East Germany, i.e., within only a fifth of the all-German electorate, and its small share of seats in the Bundestag (30) makes it a dummy player. The field work for our survey was in December 1995, 14 months after the 1994 federal election. The East German electorate was oversampled resulting in 951 respondents with valid data, compared to the 1900 respondents of the West German representative sample. An important precondition for our research agenda is a substantial amount of consistent or rational respondents with transitive and complete party rankings. American studies report figures between 73 and 81 % of respondents coming up with transitive orderings of five presidential candidates. Radcliff (1993) identified 73% of the 1980 electorate who had to rank five presidential candidates, whereas Brady and Ansolabehere (1989) asked only adult Californians who were either Democrats or Independents about the preferences for five Democratic presidential aspirants in 1976. We expected a higher figure than the 73% of Radcliff (1993), at least for the West German electorate, since we assumed that voters are more familiar with parties as permanent players in politics than with candidates who come and go. [ 17] 236 The percentage of Germans with incomplete or intransitive preferences is within the range expected from the American studies. In West Germany, only 20.5% belong to this group, coming close to the California study of a more politically involved group than the general electorate (see Table 2). The East German figure is higher (27.9%), indicating less familiarity with the party system whose major parts were imported from West Germany. In addition, the East German come up with significantly less strict orders between the five parties than the West Germans (27.5% compared to 37.2%). Tied ranks between parties are especially prone to end up in intransitivities since it is well known that people have difficulties with the equivalence relation for indifference judgements (cf. e.g., Brady and Ansolabehere, 1989). The most common error is to be indifferent between party a and band band c, but not, as it would be necessary for transitivity, between a and c. When a, band c follow each other on a policy scale, one may doubt whether the above example of an intransitivity is an error at all, since the sum of two small intervals may be beyond the threshold of the just discernible differences whereas the two smaller distances are not. And with an increasing number of ties, errors of this type increase, too (cf. Brady and Ansolabehere, 1989: 144). Thus, the outstanding result about the East Germans is the smaller percentage of respondents with a complete ranking of the five parties and not that they seem to be less rational voters in the first place. The percentage of voters ranking all five parties alike is almost the same in both parts of Germany (10%). Sometimes, this category is used as an indicator of political alienation whose level would then be equal in East and West Germany, too. We will omit the respondents of this category and the ones with intransitive and incomplete rank orders in most of the following analyses. The other respondents were able to come up with a weak or strict ordering of five parties, and one reason why they are able to do that is that they could economize the criteria used to evaluate the parties. Doing that as individuals, the range of criteria within the electorate as a whole may still be rather broad and heterogeneous. People have a tendency to be much more indifferent between low than high ranked parties. A little more than one-fourth of all respondents evaluated the last four, three or two parties alike. And among these parties, the smaller parties FDP, Greens or PDS are overrepresented, either in combinations comprising only small parties or paired with one of the two larger parties, the FDP with the CDU or the SPD with both the Greens and the PDS (see footnotes of Table 2). The two larger parties CDU and SPD are never tied at the bottom end of the individual preference scales. This is a first hint of two political cleavages running through the German party system, one between left and [18] 237 Table 2. Types of party preferences of the West and East German electoratea Preference order Linear West Germany East Germany % % 37.2 27.5 Weak Ties between two last parties 6.4b Ties between last three parties Ties between last four parties Other types of ties, except 8.5c 1I.5d lO.od 6.1 9.8 20.5 6.3 10.5 27.9 1900 951 TIes between all parties Intransitive or incomplete preferences n 9.4b 79.5 72.1 8.4c a Representative samples of West and East Germans eligible to vote (18 years and older, German citizenship); field work December 1995. b The three most frequent combinations are GreensIPDS, FDPIPDS and CDUIFDP, summing up to 85% in West and 87% in East Germany, with GreensIPDS alone amounting to 43% in West and CDUIFDP alone amounting to 65% in East Germany. c The two most frequent West German combinations are SPD/GreensIPDS (33%) and CDUIFDPIPDS (35%), whereas the East German combinations are more equally distributed. d The two most frequent combinations in the West are SPDIFDP/GreensIPDS (47%) and CDUIFDP/GreensIPDS (38%), whereas the East Germans distribute their first preference more equally between the parties, with the exception of the FOP. right parties and one between small and large parties. We shall come back to this point when we discuss the spatial implications of party preferences. 3. Aggregating individually consistent party preferences Monitoring public opinion and observing the strengths within the electorate is nowadays a routine job of pollsters working for governments, party headquarters and the mass media. In Germany, the figures regularly reported are the marginals of a vote intention question: If we had a Bundestag election next Sunday, which party would you vote for? We present an overview of the vote intentions in West and East Germany, first, for all respondents, and second, for the subsets of consistent respondents with whose distribution we compare the percentages of first ranked parties (see Table 3). In West and East Germany, the Christian Democrats are ahead of their major rival, the Social Democrats, even at a lower level in the East due to the remarkable third party status there [ 19] 238 Table 3. Vote intentions and highest ranked parties of West and East German respondents Vote intention Party of first preference All respondents Respondents with consistent party rankings West East West East West East % % % % % % CDU/CSU 37.1 30.9 39.4 32.5 40.1 30.0 SPD 33.9 23.2 32.2 22.1 31.5 19.2 FDP 2.7 0.7 2.8 0.9 2.7 1.9 B'90lGreens 9.4 9.1 8.8 9.5 10.3 11.0 PDS 0.7 17.3 0.5 18.4 0.6 20.5 14.8 17.5 1511 684 Parties Other parties No vote intention/no answer 1.3 3.3 1.2 2.6 14.7 15.5 15.1 14.0 First preference tied N 1900 951 1511 684 of the former communist party. The differences between vote intention and most preferred party are small, even if East Germans have a slight tendency to rank the PDS, Greens and FDP a little bit higher than expected from their vote intentions. A first past the post voting system would very likely guarantee the CDU a majority of seats in parliament, turning the opposition role almost exclusively to the Social Democrats as the other major party, at least in West Germany. Proportional representation complicates the aggregation of preferences, especially if we keep in mind the necessity to form coalition governments. In this section, we will first ask how robust the leading position of the CDU would be under various aggregation devices apart from counting first preferences before we discuss coalition learnings of West and East Germans. It is well known that different aggregation devices may result in different collective choices. We tum this theoretical insight around and argue, that, empirically, the leading position of a party within a multiparty system is robust if this position is corroborated by various voting systems, starting from the same individual party preferences. One could coin the phrase "robust plurality winner" for a party with more first preferences than the other parties, if this same party is also the Condorcet winner and the number one according to the Borda rule. [20 ] 239 The Condorcet winner is the party which beats all other parties in pairwise contests. Assuming that respondents who are indifferent between two parties will abstain from this pairwise contest, the Condorcet winner is the pairwise majority party, and, therefore, a strong candidate as the leading party in a democracy. The problem of this system of majority decision (cf. van Deemen, 1991: 64) is that this solution does not always exist because of voting cycles. The Borda count is a positional procedure which is computed as the mean rank of each party. Black has outlined this procedure even for the special case of tied ranks for which he recommends the adjusted Borda count instead of the original one (1987: 61-64). Its operational definition is the number of dominated parties, i.e., parties ranked lower than the focal party, minus the number of parties dominating the focal one. With five parties, the resulting scale has a maximum of +4 and a minimum of -4. In Table 4, we have computed the percentage differences between the respective row and column parties as an evidence for the Condorcet winner, showing at the same time that the relatively high percentage of indifferent respondents (see lower triangle of table) do not hinder the existence of a Condorcet winner. The percentages of indifferent respondents between the respective party pairs show the pattern which we have already discussed in Section 2 and which is most clearly observable in West Germany. The number of indifferent respondents is lowest when the two larger parties have to be compared, i.e., the Christian Democrats and the Social Democrats. The percentage of indifferent respondents increases when we compare one of the two larger parties with one of the three smaller parties and this percentage is the highest for comparisons among the group of smaller parties. From this pattern, we learn that there must be many respondents whose party system consists mainly of a comparison of the two major parties whereas they are indifferent between some or all of the minor parties. The pairwise majority party in Table 4 (upper triangle) has minus signs in its column and plus signs in its row starting from the main diagonal and going to the right. In West Germany, this criterion fits the CDU pattern, whereas in East Germany the SPD is the Condorcet winner, thereby surpassing the CDU as the plurality winner. The margin in favor of the SPD is 7.4 percentage points; this is not trivial, compared to the smaller difference of 3.5 percentage points in West Germany, this time in favor ofthe CDU/CSU. When we now add the insights of the Borda count, we are confronted with even more deviations from the position of a robust plurality winner, a role, we had hypothesized, the CDU could play in both parts of Germany. According to the adjusted Borda count, the SPD is the winner in both parts of Germany, as a party, we might add, which has much less enemies (Le., respondents who rank it last) than the CDU. [21 ] 240 Table 4. Difference of pairwise choice between row and column partya and percentage of consistently indifferent respondentsb West Germany Adjusted Against PDS For: PDS B ' 90/Greens Borda B ' 90/Greens SPD CDUlCSU FDP count -57.9 -69.5 -57.2 -46.8 -2.31 -40.4 -lS.l 4.3 0.04 30.0 1.37 35.7 -3.5 SPD 26.3 25.4 CDU/CSU 25.3 20.1 17.1 FDP 35.1 28.5 23.4 3S.S 27.9 East Germany Adjusted Against PDS For: PDS B '90/Greens Borda B '90/Greens SPD CDU -10.2 -21.2 -8.3 -lS.7 32.8 SPD 27.2 27.9 CDU 23.6 23.1 23.4 FDP 33.1 34.0 29.5 a b l.18 -0.26 FDP count 9.6 -0.30 -0.1 29.8 -0.21 7.4 4S.1 0.95 31.9 0.34 -1.19 39.7 Upper triangle: Percentage of row party choice minus percentage of column party choice in pairwise comparison of consistent respondents (n-West = 1511, n-East = 684). Lower triangle: Percentage of indifferent respondents between respective party pairs. This pattern is quite interesting for East Germany, where about one-fifth of the voters rank the SPD highest, and another fifth the PDS. But according to the Borda count, the SPD is the party with the best result, whereas the PDS is more dominated by other parties than being itself the dominating party (negative Borda count). A glimpse at the percentages of East Germans ranking each party last helps to explain the results of the Borda count. Even in East Germany, 20% of the respondents rank the PDS last, compared to only 1% for the SPD and 9% for the CDO. This makes it plausible that the SPD is the Borda winner, a victory which is not very convincing since it is more based on a lack of opponents than on a plurality of aficionados. Thus, coming out as a plurality winner from a proportional election does not guarantee a strong leading position in a multiparty system. The political [22] 241 interpretation of an election result or of the marginals of vote intention question becomes even more difficult when possible coalition governments have to be taken into account. The addition of the percentages of two parties is just not enough if one is interested in coalition support within the electorate (cf. Downs, 1957: 147 ff.). Concerning coalition support, we argue that the consistent supporters of a coalition are those respondents who rank the respective coalition parties higher than any of the opposition parties. Thus, irrespective of the coalitional options being discussed by political elites and the mass media, we derive a measure of coalition support from the frequencies of the two highest-ranked parties in voters' consistent party preferences. To avoid misunderstandings, we term this derived measure coalition leaning, distinguishing it from an outspoken coalition preference which may be formed during an election campaign. Shortly before elections, parties argue in favor of specific coalitions and many voters form a specific coalition preference which they reveal if asked directly. These coalition preferences are formed with respect to feasible coalitions, i.e., coalitions which have a chance to win the necessary majority and which are at the same time not excluded explicitly by the respective parties. With this background of information, the individual voter can form a coalition preference and choose a party or two parties on election day (Germany has a two-ballot system), thereby hoping to increase the probability that the preferred coalition will indeed be formed. What we call coalition leaning is a measure derived from the ranking of parties in one's party preference which should have an impact on coalition preference, but which is meaningful in itself, especially in-between elections, when coalition possibilities are not publicly discussed. Coalition leaning is supposed to be the support infrastructure of the logically possible coalitions which well restrict in the following to all possible two-party coalitions, omitting coalitions of three or more parties. With five parties, ten two-party coalitions are possible, and even if each party is a member of four coalitions, so that the participating parties overlap between coalitions, the consistent supporters of the ten coalitions form disjunct and complete subsets of all those respondents with clear first and second party preferences, ties for the two top parties included. A second set of respondents strictly prefers one of the five parties to all others, but is indifferent between some of the other four parties so that the second rank is already involved. This criterion leads to another five disjunct subsets of respondents with a larger than usual distance between the first and the other parties. This is the set of the one party adherents. A third set consists then of respondents for whom we cannot tell which one or two parties are at the top of the preference order. Empirically, these are overwhelmingly respondents who are indifferent between all five parties. We omit this third set when we now [23 ] 242 identify the consistent supporters of the ten different two-party coalitions. These respondents consist first of all of the ten subsets of the first set and then the five subsets of the second set. The latter respondents do not have consistent coalition preferences, since they concentrate their party choices more than the first group on a single party. The simplest aggregation mechanism is to identify the plurality winner among the possible two-party coalitions. In Table 5, the total percentages of the ten possible two-party coalitions and the five groups of one-party-Ieaners are presented both for West and East Germany. The outstanding result for West Germany is the concentration of the two top ranked parties to only three possible two-party coalitions, first of all the coalition between the Social Democrats and the Greens (25.7%), second the coalition between the CDU and the FDP (22.9%) which is the governing coalition at the time of the field work, and third the coalition between CDU and SPD (19.4%). All other coalitions would be rather implausible due to the usual vote shares of the parties in general elections. The East Germans do not favor these latter coalitions, even if we have data only on party preferences and thereby coalition leanings and not on actual coalition preferences which may be more influenced by realistic expectations about the feasible coalitions. The East Germans deviate from the West German pattern, since they disperse their coalition leanings much more than the West Germans. Their coalition leanings are less constrained by their realistic coalitional alternatives, i.e., those possible two-party coalitions which can be expected to gain majority support. The plurality winner in East Germany would be the CDU/SPDcoalition with 18.4%, followed by a coalition of the Greens and the former communists with 15.3%, then a coalition between SPD and Greens (13.1 %), and finally a coalition between the Social Democrats and the former communists (12.4%). The governing coalition in the federal capital is placed only at rank 5 (9.3%). What do these relative frequencies of groups of coalition leaners mean for elections when coalition leanings get transformed into coalition preferences? Even if not all respondents will form a specific coalition preference, a more or less large subset will form such preferences and in this process, political communication among citizens is important. We argue that a second important aspect of coalition leaning, besides, its relative frequency among the population, is the political homogeneity of the supporters measured in terms of a consensus concerning the ranks of all parties and not only the top two. Even if two respondents rank the SPD and the Greens first and second, they can disagree more or less with respect to the parties ranked third, fourth and fifth, and with respect to the ordering of the coalition parties. In a dynamic system, citizens continually evaluate parties and a coalition will gain extra [24 ] 243 Table 5. Two-party coalition pluralities and percentages of single party preference profiles a West Germany CDU/CSU SPD FDP CDU/CSU SPD FDP B'90/Greens PDS 9.2 19.4 22.9 4.1 0.1 2.0 19.4 7.3 4.1 25.7 22.9 4.1 0.5 0.9 B'90/Greens 4.1 25.7 0.9 2.6 PDS 0.1 2.0 1.3 1.3 n == 1317 East Germany CDU SPD FDP B ' 90/Greens PDS CDU SPD PDP 8.4 18.4 9.3 6.7 1.9 18.4 4.7 1.6 13.1 12.4 9.3 1.6 0.5 0.7 0.2 B'90/Greens 6.7 13.1 0.7 2.4 15.3 PDS 1.9 12.4 0.2 15.3 4.3 n== 580 aTotal percentages of all respondents with clear first and second (offdiagonal) or only clear first rank preference (main diagonal). leverage out of the political discussions of citizens if its supporters speak with one voice due to the homogeneity of their party preferences. Heterogeneous support groups tend to decay under stress, as, for example, during election campaigns. We use the heterogeneity measure based on the Kemeny-Snell metric and introduced above, to measure the internal heterogeneity of the various coalition leaning groups. In addition, we present in Table 6 the median ranking of parties within the most frequent groups of coalition leaners. This is the ranking of parties with minimum distance to the other rankings present within the respective groups. A first rough indicator of the heterogeneity of a group is the number of different party rankings being represented within the group. The surprising finding here is that the East German respondents whose size as a group within our survey is only half of the West German group entertain the same number of rankings as the more popular West Germans. Thus, the East Germans show much less agreement of party rankings than the West Germans who are part of a more established party system. Agreements between individuals are an [25 ] 244 attribute of social order, not of an individualistic consistent world view. We have already seen that the percentage of East Germans with consistent party rankings is slightly less than the respective percentage of West Germans. But these differences appear to be small compared to the greater incoherence of the East German social order. Irrespective of coalitions, the heterogeneity measure for all West German respondents with clear first and/or second party rankings amounts to 7.36 compared to a mean of 8.94 for East Germans. This is a large difference, especially if one considers the large number of ties within the East German party rankings which should have the consequence of lowering the distances between rankings. When we focus our attention now to the three largest groups of coalition leaners in West Germany and the five largest groups in East Germany, we present the results of the heterogeneity measure with the hypothesis in mind that homogeneity of a coalition leaning group is an advantage for the preference formation concerning coalitions. And from this perspective, it is interesting that the governing coalition of Christian Democrats and Liberals does not enjoy the largest support group within the West and especially not within the East German electorate, but its respective support groups are the least heterogeneous ones in both parts of Germany. For this group, the West German heterogeneity measure is 1.94 and the East German one is slightly higher with 2.28, but this latter figure is nevertheless the lowest heterogeneity measure for all East German groups. The major rival of the present coalition, i.e., the SPD/Green coalition, is characterized by the highest heterogeneity measure, again in both parts of Germany. This result is of course partially due to the fact that there are three left parties and only two right or conservative parties represented in our party population. But since this system delineation is not made ad hoc but is based on the criterion which parties are represented in the Bundestag, the heterogeneity measures are the results of a correctly delineated party system. Besides this advantage for the two incumbent parties, another factor improving the position of incumbents is their better ability to define the terms of public debate than that of the opposition parties, thereby building up internal strength. The median party ranking of the CDUIFDP-coalition leaners is identical in East and West Germany with the CDU preferred most and the PDS preferred least, whereas the SPD/Green leaners in West Germany reject the PDS most and the respective East German support group is indifferent between the PDS on one side and the CDU and FDP on the other side. The attitude of the two established left parties, the Social Democrats and the Greens, towards the former Communists is a probable source of conflict, especially when one takes into account the polarization of the East Germans, one fifth of whom rank the PDS highest and another fifth ranking it lowest. [26] 245 Table 6. Party-preferences of most frequent groups of coalition leaners: Median party ranking and heterogeneity Median ranking a Heterogeneityb 25.7 S >- G >- C,F >- P 3.42 CDU/CSU - FDP 22.9 CDU/CSU - SPD 19.4 C>-F>-S>-G>-P C >- S >- F,G >- P 2.62 Coalition Percent leaners SPD - B'90/Greens West Germany 1.94 East Germany CDU-SPD PDS - B'90/Greens SPD - B'90/Greens 18.4 C >- S >- F,G >- P 3.11 15.3 13.1 P >- G >- S >- C,F S >- G >- C,F,P 2.34 3.68 SPD-PDS 12.4 P >- S >- G >- C,F 3.04 CDU-FDP 9.3 C>-F>-S>-G>-P 2.28 a b Preference order among CCDU/CSU), S(PD), F(DP),G(reens), P(DS). The heterogeneity measure is computed like a mean path distance measure of a weighted graph. Given the distances dij between party preferences i and j with the respective frequencies fi and fj. Heterogeneity is defined as: LLdilifj/Lff, since LLfifj = Lff· i i i i The median rankings within the three largest groups of coalition leaners in West Germany and of the five largest in East Germany reduce the complexity of party preferences to a manageable number. Since we already know that the CDU is the West German Condorcet winner and the SPD the East German one, we could now argue that the majority winner is able to choose among the possible two-party coalitions in whose median ranking it is leading. Thus, the CDU could form either a coalition with the PDP or with the SPD in the West, whereas the SPD as the East German majority winner would only have the option of an SPD/Green-coalition in which it ranks first, compared to its second rank in a coalition with the CDU. These arguments could be more convincing if these coalitions could be chosen from a linear profile of coalition options, for which a majority choice is guaranteed (cf. van Deemen, 1991: 204), and which would answer the coalition question at the same time. We will discuss this problem of how to combine party preferences to form linear profiles or one-dimensional party orderings as the problem of collective order in the next section. [27] 246 4. Combining party preferences to a collective order Individuals are able to rank parties consistently if they apply the same criteria of evaluation to the parties. This task is easiest to solve with only one criterion. If the public debate would be focused overwhelmingly on one outstanding criterion, i.e., the solution of one central policy problem, we can imagine a one-dimensional policy scale on which the parties are placed according to their promised solutions concerning this problem. Citizens would agree that the solution of this one-policy problem is central for the welfare of the country, but they would disagree concerning the ideal solution. They identify parties as providers of different standard solutions, and, depending on their ideal points on their one-dimensional policy scale, evaluate parties according to the proximity of the respective party solution to their respective ideal point. Under these conditions, the median solution would be the majority winner. The more public opinion is focused on the solution of one central problem, the higher the probability of the existence of a one-dimensional collective order, in which both parties and voters hold specific positions. It is important to understand that this collective order is derived from policy or party preferences, not from party perceptions. The policy or ideological space underlying specific multiparty systems is often multi-dimensional. For Germany, for instance, experts and citizens agree that the liberal party is right of the CDU in economic and social welfare terms, but left of the CDU in "social" or new politics issues (cf. supra p. 231), making the perceived policy space two-dimensional. But at a specific point in time, either economic or "social" issues are pressing, and, therefore, at the center of public attention, so that the "decision space" may be one-dimensional, compared to the underlying two-dimensional "perception space". The reduction of the perception space to a decision space is, among other factors, also determined by the coalition government of the day. A conservative government of CDU/CSU and FDP is based on economic policy agreements, whereas the SPDIFDP coalition of the period 1969 to 1982 got started with communalities in "social" issues and foreign policy. German unification brought the socio-economic dimension even more to the forefront than was true for the 1980s when environmental issues challenged the priority of the socio-economic policy of the first Kohl governments. The most plausible left-right placement of the five German parties on the socio-economic policy dimension is as follows: The former communists (PDS) as the left-most party, then the Greens left of the Social Democrats, with the Christian Democrats in-between the SPD, and the classical German party favoring market as against state solutions, i.e., the FDP. We will test whether this party ordering can be derived from the pattern of marginals of [28] 247 pairwise comparisons (see Table 4), from the median party rankings of the most frequent groups of coalition leaners (see Table 6) and finally from the profile of linear orderings of the five parties. In Table 4, we have arranged the parties from left to right in the order of the socio-economic dimension. Applying Feld and Grofman's (1988) concept of collective ideological consistency, the differences of the margins of collective preference for all party pairs are assumed to give hints about their one-dimensional party ordering at the aggregate level. "To the extent that collectivities are ideological, this ideology should be reflected by ideologically consistent regularities among the various vote margins between pairs of mutually exclusive alternatives, as well as by a single-peaked majority preference ordering at the aggregate level" (Feld and Grofman, 1988: 776). If we have ordered the parties correctly from the leftmost to the rightmost alternative, the percentage of respondents choosing a specific left party of a pair does not decrease, when the respective second party becomes more rightist. The cumulative logic should guarantee that once a left party has reached a certain percentage of respondents at one point on the underlying continuum, that percentage has at least to remain the same, if not to increase, when the comparison task is made easier for the left side by a confrontation with the more rightist party. This cumulative logic may apply for a majority of voters, even if the individuals within the group deviate to a certain extent from the joint scale. In Table 4, the cumulative logic has to show up in the figures above the main diagonal. Across rows, this difference has to become less negative or more positive the further we move from the left side of the table to the right side. The second regularity pertains to the columns. The further we move down in one column towards the main diagonal, the less negative or the more positive the percentage differences have to become. Take, for example, the FOP column for West Germany. When contrasted with the PDS, a majority of voters decide in favor of the Free Democrats, making the percentage difference negative. Contrasted with the Greens, this less leftist party has a slight majority which grows larger the less leftist the comparison parties for the most rightist party, i.e., the Free Democrats, become, since the comparison gets more favorable for the leftist side, the less leftist the party is with which we compare the Free Democrats. For West Germany, we observe more or less ideological consistency at the aggregate level, with the exception of the comparison between the former communists and the Greens. The comparison between the PDS and the B '90/Greens should be less favorable for the PDS, making the percentage difference quite large, than the comparison of the PDS with the Social De- [ 29] 248 mocrats. Since this expectation does not hold, there is this one deviation from a perfect aggregate pattern of ideological consistency in West Germany. In East Germany, we observe the same type of deviation for the pair PDS vs. Greens as in West Germany, and, in addition, a large deviation for the pair CDU vs. PDP. What all these deviations from the expected pattern have in common are two things: The respective parties are neighbors on the left-right-scale and the percentage of indifferent respondents for these pairs is especially high. The high number of ties makes the percentages for and against the respective parties lower than the ones for the other pairs, thereby deflating the percentage point differences. Thus, we do not accept the deviations from the expected pattern as evidence against our "decision space" hypothesis. A second test of this hypothesis is based on the Hamming distances between the median rankings of the largest groups of coalition leaners in East and West Germany. The existence of a linear profile made up of these altogether six median rankings in the correct order from the PDS-Ied coalitions on the left to the CDUIFDP-coalition on the right would support out hypothesis, even if this evidence is based on an ad hoc selection of preferences: First, on the selection of the largest groups of coalition leaners and, second, on the identification of the respective median rankings. As shown in Diagram 1, the median rankings form indeed a linear profile with the PDS/Green-coalition at the leftmost and the present CDUIFDPcoalition at the rightmost position, the distance between these rankings being 19 units. This distance can either be computed by the formula given supra page 233 or by summing up all the intermediate distances, thus proving the existence of a linear profile. Due to the leading position of the CDU both in the coalition with the FDP and with the SPD, the distance between these two coalition options is only half the size of the distance between the CDU/SPD and the SPD/Green-coalition. In an all-German party space, the Social Democrats may become the median party, but they will have to pay the price of a very heterogeneous support group. Both collective ideological consistency and linear profiles of median rankings indicate one-dimensionality at the aggregate level. It would be more convincing to build up a linear profile from individual party preferences, in the same way as this is tried with unidimensional unfolding. We rely on the unfolding logic but do not use it heuristically with the original, noisy data. Instead, we restrict the analysis first to respondents with consistent party preferences and second to linear orderings, thereby avoiding tied ranks. The reason for this second restriction is our knowledge of the large/small party split of the German party system which does not fit the left-right logic. The electorate seems to be stratified according to the completeness of its view of [30 ] 249 Median rankings P G P S S G S C,F G C,F,P C,F G S C,F P C C S F,G P G F S P Diagram J. Hamming distances between median party rankings of largest groups of coalition leaners in East and West Germany (see Table 6). the party system. Some voters have only the CDU and the SPD on their choice set, being indifferent between the lower ranked smaller parties. Contrary to this pattern, a left-right logic would only allow ties of neighboring parties on this scale, between Greens and SPD, for example, or CDU and FDP, but not between Greens and FDP. Analytically, it therefore makes sense to test the one-dimensionality of the decision space solely for respondents with linear orderings of the parties. The disadvantage of this procedure is the small size of the group comprising our target population, 37.2% of the West German and 27.5% of the East German respondents. We justify this procedure by pointing at the characteristics of a collective ordering of parties. There exist many possibilities for idiosyncratic worldviews, but as soon as a core group of voters responds to a focused public debate by rearranging its party preferences along the underlying policy dimension, a collective order or linear profile becomes crystallized, which facilitates communication between parties and voters. "Communication is not regarded as a process of exchanging pockets of information ... Rather, it is a kind of guessing game. A person carries with him his cognitive field as a map of the world. He responds not to the world, but to the map. When he receives the communication, the meaning it has is a consequence of how it can be fitted into the map" (Coombs, 1964: 122). Such a common map of a core group of voters and the party elites is a precondition of a meaningful public debate about policies. In Germany after unification, it was and is an even more difficult task than usual to come up with a common, all-German map. East and West Germans do not have to agree in their ideal solutions for the socio-economic policy problems of the country, but they should agree in their views about the major problems, if the parties should be able to compete for votes or support in an all-German campaign. Competing for votes on a single policy dimension is a rational strategy if the common or joint policy scale is based on single-peaked preferences of [31] 250 the voters. And single-peakedness is not a characteristic of individuals, but of the collective order. A linear profile consists of an ordered set of preferences based on the relation of betweenness whereas a joint policy scale positions parties and voters on a straight line so that the preferences of voters can be computed from the varying distances of a voter's ideal point to the parties. As Coombs (1964: 84ff.) has shown, five object of evaluation on a straight line are compatible with 16 individual preference orders if metric implications for the various intervals between parties are ignored. The percentage of respondents with preferences which fit a specific joint scale is a rough indicator of the prevalence of this policy scale within a collective order. In West Germany, 58% of the respondents with linear preferences rank the five parties in a way compatible with our hypothesized socio-economic left-right-scale, in East Germany 42%. These percentages indicate the prevalence of this policy scale, under the condition that rival scales do not receive almost as much support. We tested two other policy scales, one standing for "social" or new politics issues with the Greens and the Christian Democrats at opposing ends (Greens - PDS - SPD - FDP - CDU/CSU) and one where only CDU and FDP switch positions and the PDS remains the leftmost party. The latter scale was more prominent during the SPDIFDP coalition period, and is, even nowadays, the major competitor of the socio-economic left-right scale, with 45% compatible party rankings in West and 27% in East Germany. 21 %, respectively 8%, of the preferences fit both scales. The new politics scale picks up only 2% in West and 18% of the preferences in East Germany. Overall, we interpret these findings as supporting the prevalence of the socio-economic left-right scale, even if the position of the FDP within this collective order is not unchallenged. The most important reason for the rightmost position of the FDP on the socio-economic policy scale is the rejection of this party by left-leaning respondents; they reject the FDP more than the CDU, whereas the liberal selfimage is quantitatively less important. But even if many voters still perceive the Liberals as holding middle-ground in the German party system, they see this party to the right of the CDU in specific economic and social welfare policies. The East German respondents still have more problems than the West Germans to fit into a collective order, since they afford the luxury of a lot of ideo-syncratic world views. But despite this fact, the best candidate for a collective order is the same socio-economic scale in both East and West Germany (see Diagram 2). Diagram 2 allows a closer look at the party preferences compatible with the. socio-economic left-right scale. The Hamming distance between two neighboring party rankings in two units, the lines between preferences in- [32] 251 West East Preferences n n ~ 7 GSCPF 29 GSCFP 73 SGCFP 54 SCGFP 36 SCFGP I I I 4 PGSCF 4 GPSCF 11 GSPCF 4 I I 1 ~ 9 SGPCF 2 9 SGCPF 8 6 SCGPF 1 I I CSGPF 17 I ~ I 109 CSFGP 112 CFSGP 29 FCSGP I ~ 48 I CSGFP 4 17 24 12 I Diagram 2. Hasse-Diagram of the 16 linear profiles compatible with the joint socio-economic scale peDS), G(reens), S(PD), C(DU), F(DP) dicating immediate neighbors. There exist 12 unique paths from the leftmost (top) to the rightmost (bottom) preference and these paths correspond to the 12 quantitative joint scales being compatible with the same qualitative scale (see Coombs, 1964). We choose the paths which pick up most respondents in West and East Germany and show the spatial implications on the interval scales of Diagram 3. In both parts of Germany, the distance between CDU and SPD is larger than the distance of each major party to its junior ally, leading to a bi-polari- [33] 252 West Germany PDS Greens SPD eDU FDP eDU FDP East Germany PDS Greens SPD Diagram 3. The best fitting quantitative joint socio-economic scale for West and East Germany zation of the party system on the socio-economic dimension. Such a configuration would not be supportive for a grand coalition of Christian and Social Democrats. The East German pattern departs from the West German one as far as the PDS is concerned. For West Germans, the distance between PDS and Greens is larger than the distance between SPD and CDU, whereas for East Germans, it is shorter, unifying the left camp of three parties more than in West Germany, where the PDS is clearly an outlier at the extreme left of the party system. On the socio-economic policy scale, only the Christians or the Social Democrats have a chance to become the median party. In an all-German context, the chances of the Social Democrats to hold a median position have been improved by German unification, since the East Germans lean more to the three left parties than to the CDU or FDP. But which of the two major parties will become the median or Condorcet winner in the next election, will also depend on voters whose choice sets are focused on the CDU and SPD. On the socio-economic scale, these parties are neighbors, thereby facilitating the existence of a Condorcet winner. This winner will then be able to decide which coalition should be formed, but its options will be restricted by the prevalent socio-economic policy scale and the policy distances on this scale. 5. Conclusions Voter preferences in multiparty systems contain a lot of information beyond the distribution of first preferences, i.e., the type of data retrieved by proportional voting systems. We derived information on coalition leanings from the party rankings and we analyzed their spatial implications. With five parties 5! = 120 linear orderings are logically possible, and allowing tied rankings, this number increases to 541. A null-hypothesis about party preferences would [34] 253 predict a random sample of party rankings for an electorate, the next step being the enforcement of prior probabilities for the first ranked parties according to past election results. Both the first and the second null-hypothesis does not predict the distribution of party rankings in West or East Germany. On the other hand, multi- or uni-dimensional unfolding would have its problems, too, to reveal a space in which both parties and voters are located such that the party rankings of voters can be reconstructed from distances to the parties within the joint space. Normally, the data used as input are too noisy to reach convincing solutions. We have first of all identified the 3/4 to 4/5 of the respondents who were able to form at least a weak order among five parties. But this step of the analysis is not yet enough for a construction of a collective order out of the universe of individually consistent party rankings. Our own earlier attempts to find a spatial representation of the German party system were based on unrealistic assumptions about the joint party space. We underestimated the substantive obstacles working against the formation of a collective order which should be the outcome of correct perceptions of party positions and function as the "decision space" at the same time. In this article, we focused on the "decision space" which can be derived from party rankings and we gave up the assumption of a joint space for all voters. We argued instead that spatial implications from party preferences can be drawn if a core group of voters is able to rank all the parties on the basis of strong preferences and if a substantial amount of these preferences is compatible with a joint or collective order of the party system. The application of spatial models to multiparty systems needs a lot of calibration so that the theoretically powerful models can indeed be adapted to real-world situations. For this purpose, we have distinguished between the perception and the decision space, outlining that party rankings have first of all implications for the latter. Our analysis of the logically possible two-party coalition has shown that the governing coalition is able to structure the party rankings to a higher degree than the other possible coalitions. This structuring power was indicated by the measure of heterogeneity of the coalition leaners' preferences. Extrapolating from this result, we hypothesize that the governing coalition in multiparty systems is, as the main political agenda setter, also able to influence public opinion and, thereby, predetermining the collective ordering of parties. For the German party system after unification, we have shown that these spatial implications are almost identical in West and East Germany. Thus, we are justified to interpret our results as evidence of a common all-German party system, not in the trivial sense that the same parties compete for votes in East and West Germany, but in the sense that the parties compete within the same frame of reference. The party preferences of the East Germans are [35 ] 254 more left-leaning than those of the West Germans and the PDS, especially, is on average higher ranked. The common socio-economic policy dimension is the precondition of a democratic election resulting in a median party and, probably, in a collective preference for a specific coalition government. Under these conditions, the party elites are well advised to pay attention to the coalition leanings and the derived policy preferences of the electorate. References Black, D. (1987). The theory of committees and elections. Boston, Dordrecht, Lancaster: Kluwer Academic Publishers. (First published in 1958). Brady, H.E. and Ansolabehere, S. (1989). The nature of utility functions in mass publics. American Political Science Review 83: 143-164. Coombs, C.H. (1964). A theory of data. New York: Wiley. Downs, A. (1957). An economic theory of democracy. New York: Harper and Row. Eckstein, G. (1995). Rationale Wahl im Mehrparteiensystem: Die Bedeutung von Koalitionen im raumlichen Modell der Parteienkonkurrenz. Frankfurt am Main: Peter Lang. Enelow, I.M. and Hinich, MJ. (1984). The spatial theory of voting: An introduction. Cambridge: Cambridge University Press. Feld, S. and Grofman, B. (1988). Ideological consistency as a collective phenomenon. American Political Science Review 82: 774-788. Fiorina, M.P. (1981). Retrospective voting in American national elections. New Haven, London: Yale University Press. Fishburn, P.c. (1970). Utility theory for decision making. New York: Wiley. Inglehart, R. and Klingemann, H.-D. (1976). Party identification, ideological preference and the left-right dimensions among western mass publics. In I. Budge, I. Crewe, and D. Farlie (Eds.), Party identification and beyond. Representations of voting and party competition. 243-273. London et al.: Wiley and Sons. Kemeny, J.G. and Snell, L.J. (1963). Mathematical models in the social sciences. New York, Toronto, London: Blaisdell. Second edition. Laver, M. and Hunt, B.W. (1992). Policy and party competition. New York and London: Routledge. Laver, M. and Schofield, N. (1990). Multiparty government. Oxford: Oxford University Press. Nannestad, P. (1994). Danisches Wahlverhalten 1971-1979: Ein Modell reaktiven Wahlens und einige empirische Ergebnisse. In F.U. Pappi and H. Schmitt (Eds.), Parteien, Parlamente und Wahlen in Skandinavien, 285-306. Frankfurt, New York: Campus. Norpoth, H. (1979). The parties come to order: Dimensions of preferential choice in the West German electorate, 1961-1976. American Political Science Review 73: 724-736. Pappi, F.U. (1973). Parteiensystem und Sozialstruktur in der Bundesrepublik. Politische Vierteljahresschrift 14: 191-213. Pappi, F.U. (1983). Die Links-Rechts-Dimension des deutschen Parteiensystems und die Parteipraferenz-Profile der Wahlerschaft. In M. Kaase and H.-D. Klingemann (Ed.), Wahlen und politisches System, 422-441. Opladen: Westdeutscher Verlag. Pappi, F.U. (1990). Die Republikaner im Parteiensystem der Bundesrepuhlik: Protesterscheinung oder politische Alternative? Aus Politik und Zeitgeschichte 21/90: 37-44. Pappi, F.U. (1992). Wahrgenommene Parteidistanzen zwischen Parteien und eigene Politikpraferenzen der Wahler: Die Anwendung eines Raumlichen Modells der [36] 255 Parteienkonkurrenz auf das Parteiensystem in West- und Ostdeutschland. In H.-J. Andress et al. (Eds.). Theorie, Daten, Methoden: Neue Modelle und Verfahrensweisen in den Sozialwissenschajten, 317-341. Miinchen: OIdenbourg. Radcliff, B. (1993). The structure of voter preferences. Journal of Politics 55: 714-719. Schofield, N.J. (1993). Party competition in a spatial model of coalition formation. In WA. Barnett, MJ. Hinich, and N.J. Schofield (Eds.), 135-174. Political economy: Institutions, competition, and representation. Proceedings of the Seventh International Symposium in Economic Theory and Econometrics. Cambridge: Cambridge University Press. Schofield, N.J. (1995). Coalition politics: A formal model and empirical analysis. Journal of Theoretical Politics 7: 254-281. Shepsle, K.A. (1991). Models of multiparty electoral competition. Chur, CH, et al.: Harvard Academic Publishers. Van Deemen, A.M.A. (1991). Coalition formation and social choice. Nijmegen: NICI. Van der Eijk, C., Franklin, M. and Oppenhuis, E. (1996). The strategic context: Party choice. In C. van der Eijk and M. Franklin, Choosing Europe?: The European electorate and national politics in the face of union, 332-365. Ann Arbor: University of Michigan Press. [37] Public Choice 97: 257-293, 1998. © 1998 Kluwer Academic Publishers. 257 Multiparty electoral competition in the Netherlands and Germany: A model based on multinomial probit * NORMAL SCHOFIELD, ANDREW D. MARTIN, KEVIN M. QUINN & ANDREW B. WHITFORD Center in Political Economy, Campus Box 1208, Washington University, St. Louis, MO 63130, U.S.A. Abstract. A typical assumption of electoral models of party competition is that parties adopt policy positions so as to maximize expected vote share. Here we use Euro-barometer survey data and European elite-study data from 1979 for the Netherlands and Germany to construct a stochastic model of voter response, based on multinomial probit estimation. For each of these countries, we estimate a pure spatial electoral voting model and a joint spatial model. The latter model also includes individual voter and demographic characteristics. The pure spatial models for the two countries quite accurately described the electoral response as a stochastic function of party positions. We use these models to perform a thought experiment so as to estimate the expected vote maximizing party positions. We go on to propose a model of internal party decision-making based both on pre-election electoral estimation and postelection coalition bargaining. This model suggests why the various parties in the period in question did not adopt vote maximizing positions. We argue that maximizing expected vote will not, in general, be a rational party strategy in multiparty political systems which are based on proportional representation. 1. Introduction Democratic political systems can be distinguished by whether they are based essentially on proportional or plurality electoral rules, and whether political parties are strongly disciplined or not (see Table 1). Most of the polities of Western Europe have electoral systems based on proportional representation, with relatively disciplined parties. Some of these polities (such as Austria) only have two parties, but others (such as Finland) may have five or six or more. In contrast, Britain has a plurality (or firstpost-the-post) electoral system, based on over 600 constituencies, each of which returns one member of Parliament. Although the House includes at * This paper is based on research supported by NSF Grants SBR 94-22548 and 96-17708. Versions of this research have been presented at the Public Choice Meeting, San Francisco, March 1997, at the European Public Choice Meeting, Prague, April 1997, and at the Political Science Seminar, New York University, May 1997. [39] 258 least eight parties, the effective number l is just above two (Schofield, 1997a) as would be expected from Duverger's Law (Duverger, 1954, 1984; Riker, 1982; Fedderson, 1992). The British system is often called a "Westminster style" polity. In contrast, the U.S. Congress is based on a plurality electoral system, but (to judge from the heterogeneity of the voting by its members) the two parties are not disciplined in the sense that British parties are (Poole and Rosenthal, 1991). In some of the new democracies (such as Russia) mixed PR and plurality electoral methods have been adopted, and the polity seems to be highly factionalized and the parties very weak. In Japan, which until recently had a complex electoral system based on multimember constituencies, the dominant liberal Democratic Party was comprised of at least six factions all of whose leaders were contenders for the prime ministerial position (Wada and Schofield, 1996). To construct a formal comparative politics, it would be desirable to build a rational choice or game theoretic model that includes (a) a reasonable description of voters' choices, (b) a coherent description of the behavior by political actors in the pre-election context, and (c) a model of the negotiation or bargaining by political actors, in the post-election phase, over the formation of government. Presumably the type of electoral system would enter into the calculation of political actors under (b) so that it would be possible to see clearly the effect of different motivations. Similarly, the effect of political institutions (such as Congressional committees or cabinet rule) could be understood more readily, were a general model of (c) available. Almost all the theoretical work of which we are aware has concentrated either on pre-election behavior by two candidates under plurality rule, or on post-election coalition negotiation in multiparty situations (where "multiparty" means at least three parties). It is clear enough that understanding "multiparty" competition should also involve calculations made by candidates or parties before the election. This paper will use an empirical analysis of electoral data from the Netherlands and Germany to argue that the usual two-party electoral models do not generalize well to multiparty situations. Instead we argue that a coherent model of pre-election party strategy has to incorporate both an electoral component and prediction over post-election coalition possibilities. We use our model of electoral behavior, together with our account of coalition behavior, to suggest why political competition in multiparty situations does not lead to convergence of party positions. In particular, our model of party policy choice emphasizes the heterogeneity of preferred policies within each party. We argue that the proposed model can in principle be used to understand the different motivations determining political choice in electoral [40] 259 Table 1. Types of politics Discipline Electoral rule Proportional Plurality representation Strong West European model Westminster Weak Factional U.S. House systems based on proportional representation or plurality rule, with or without disciplined parties. 2. Elections: Models of pre-election choice by parties The literature on formal models of elections is vast, but we can briefly review the main results of the spatial model using Table 2 as a guide. The spatial model assumes that political choices are points in a policy space, W, of dimension w. Each voter i E N has an ideal point, Xi E W, and each candidate, j E K makes a "declaration" ¢j E W. The "electoral data" is then an n x k matrix Oij = II Xi-¢j II where n = INI and k = IKI. In the pure spatial deterministic model, voter i's "utility" Uij for the choice ¢j is a monotonically decreasing function of Oij. A typical assumption is that uij = _{302 ij . Voter choice is given by an n x k matrix X = (Xii) where Xij = I (i chooses j) if II x i-<PJi 1>llxi-¢j II for I #: j, and Xii = 0 otherwise. The share of the vote of party j is then Jr/X)(¢) = ~ L~I Xij (¢), (where ¢ = (¢I "',¢k) is the vector of declarations). In two-party deterministic models it is usually assumed that each candidate desires to win. Thus for example we could assume that the utility for candidate I, given X and ¢, is 1 if Jrl (X)(¢) > Jr2(X)(¢) o if Jrl = Jr2 -I if Jrl < Jr2. In the Hotelling (1929) and Downs (1957) model, vote maximizing by the candidates (in the case w = 1 and k = 2) forces each candidate (j) to adopt the electoral median (or Nash equilibrium) position, ¢*j = x~(n+I)' where the voter ideal points are ordered XI .::: X2 .::: ... Xn, etc. In the one-dimensional case this result is robust. Thus, if each candidate has policy preferences (so that 1, for example, prefers to win with a policy near some bliss point z I, say), then [ 41 ] 260 Table 2. Models of election Candidate model Office seeking Voter model Deterministic Probabilistic Low dimension Strong convergence Electoral median Nash equilibrium at electoral mean, High dimension when variance is high. Electoral heart Policy-seeking Weak convergence Weak convergence for but possibly to some parties; Nash equilibrium possible divergence by small parties the Nash equilibrium typically satisfies ¢*, = ¢*2 (Osborne, 1995). The logic is clear: to implement a desired policy, a candidate still must win. Attempts to extend the Hotelling-Downs model to the case with k ::: 3, generally find no pure Nash equilibrium (satisfying ¢*j E W for all j E K), even in one dimension (Osborne, 1993). The motivation of the candidates in such a multiparty model is usually taken to be plurality maximization, so candidate j prefers that declaration, ¢j, which maximizes the differences between 7l'j and 7l'" for all I i= j. The reasoning underlying this assumption is presumably that 7l' j is a proxy for the power of the political agent, j, and that such an agent wishes to maximize its power. We shall say that an agent, j, who acts to maximize 7l'j (subject to some model X of voter behavior) is Downsian. A theme of this paper is that Downsian behavior need not be "rational" in a more general context. Attempts to extend the Hotelling-Downs model, with k = 2 but w ::: 2 ran into the well-known difficulty of "generic" non-existence of a pure strategy Nash equilibrium. (In fact, the formal proof of this comes from results by McKelvey and Schofield, 1986; and Saari, 1997, on non-existence of a core in a spatial committee game. We mention these results in the next section.) This "fact" can be side-stepped either by focusing on mixed strategy Nash equilibria (Kramer, 1978), or by introducing a more general notion such as the "uncovered set" (McKelvey, 1986; Cox, 1987). However, the uncovered set is a concept based on spatial committee voting theory, and it is not entirely obvious that it is appropriate for modeling elections, even when political agents are assumed to be Downsian. Below we shall introduce the notion [42] 261 of the electoral and political "heart" (Schofield, 1995a), and argue that it can be used to understand multiparty competition. Existence of pure strategy Nash equilibria with Downsian candidates is much easier to demonstrate when the electoral model (X) is probabilistic or "stochastic" (see, e.g., Hinich, 1977; Enelow and Hinich, 1984; Coughlin, 1992). There are a number of varieties of such a model, but they all suppose voter behavior is described by the probability Xij that voter i chooses j. Thus Xij = Prob(X ij = 1), so 1:jEK Xij = 1, for each i. A typical assumption is that Uij =Uij + Ej where Ej is a "perceptual" error term associated with candidate j and Uij is the spatial utility of i for j. Usually the entries in the error vector E (E I,... ,Ek) are assumed to be independently and identically distributed (iid). The probability Xij is then the probability that Uij > Ui] for all 1 i= j. This condition is simply that = E] - Ej < t3(8~ - 8ij) for all 1 i= j. (1) If it is assumed that each Ej is normally distributed, with zero expected value and variance a 2j, then it is possible to compute the distribution of the random variable ll'j(X) (cp), given the electoral model and the vector of declarations. Since the expectation of Xij is simply Xij it follows that the expectation of the vote share is simply given by E(JTj)(",) = ~1: Xij(CP). As in the deterministic case, we shall say that a candidate, or agent j, who chooses CPj to maximize E(JTj)(cp), subject to (CPI, ... , CPj-t, cpj+j, ... ,¢k), is Downsian. In two-party competition a more plausible assumption for party I say, would be to maximize the probability that JT t (CPt ,CP2) exceeds JT2(CPt ,CP2). Nonetheless, under the independence assumption, these two motivations are effectively identical (Aranson, Hinich, and Ordeshook, 1974). A standard result for k = 2 is that a pure-strategy Nash equilibrium exists and is strongly convergent, in the sense that cpr = cp~, both at the mean of the voter ideal points. As Lin, Enelow, and Dorussen (1996) have recently shown, even when k ::0: 3 there is a pure strategy Nash equilibrium at the voter mean (at least in the case the variance terms j E K, are sufficiently high). In fact, they show (for high variance) that each expectation E(JTj)(cp) is concave in CPj. Thus the Nash equilibrium is unique, and is at the "welfare maximum" point, cP* (so cpr =cp* v j E K). That is to say, 1:iEN II Xi - cp* II is minimized at the convergent Nash eqUilibrium, CP*. However, for low variance, E(JTj)(cp) may fail concavity or quasi-concavity. Indeed, Nash equilibria may be divergent or fail to exist. It would appear that both deterministic and probabilistic models with Downsian agents are preoccupied with finding convergent Nash equilibria. This seems strange, since in no political system of which we are aware do candidate positions display strong convergence. It is true that one can weaken the degree of convergence in 2-candidate elections, by assuming that candi- al, [43 ] 262 dates have policy preferences. This is done by assuming that the outcome of a pair of declarations, «(jJ,,(/J2), is a lottery, {(Prob(rrd,(jJ,),(Prob(rr2),(jJ2)} say, where Prob(rrj) is the probability that candidate j wins the election. As Cox (1984) has shown, if each candidate (or party) has a Euclidean preference based on the party's preferred point, and is committed to the implementation of its declared position if it wins, then there is a (non-convergent) pure strategy Nash eqUilibrium. As soon as parties have preferred positions, then the validity of the assumption of policy commitment in Downsian models becomes suspect. We shall come back to this question later. 3. Committees: Models of post-election coalition behavior In two-candidate elections the formal analysis is over when the election is over and one party wins. If the candidates are parties and k ::: 3, then the process of government formation involves creating a coalition (which mayor may not be winning). The early work on European coalition governments was much influenced by Riker's (1962) analysis of constant sum voting games and his notion of minimal winning coalitions. It was soon noticed (Herman and Pope, 1973) that coalition governments could be minority (lacking a majority) or surplus (with extra partners). This research program then focussed on coalition negotiation between parties with preferred policies in a onedimensional policy space (Axelrod, 1970; de Swaan, 1973). Essentially, these models implied that the party at the median "legislature" policy would belong to the government. In fact, one could go further and argue that if parties were only concerned with policy, then any party at the median position in the legislature could form a minority government and implement its desired policy. However, empirical analysis (Taylor and Laver, 1973) suggested that these one-dimensional models did not provide a satisfactory explanation of government formation. During the 1980s however, theoretical work on the spatial "committee" model of voting (Schofield, 1985) as well as empirical work on party declarations (Budge, Robertson, and Hearl, 1987) suggested that a two-dimensional analysis of coalition behavior could be fruitful. To illustrate, Table 3 presents the election results of the 1977 and 1981 elections in the Netherlands. A coalition of the CDA and VVD formed in December 1977, controlling 77 seats, and lasted 41 months until the 1981 election. (It should be noted that this coalition took 6 months to form after the May 1977 election.) After 1981, a brief "surplus" coalition of PvdA, D66 and CDA (with 109 seats) formed [44] 263 Table 3. Elections in the Netherlands, 1977 and 1981 Party (acronym) 1977 1981 (Seats) Labor (PvdA) 53 44 Democrats ' 66 (D66) 8 28 26 49 (138) 48 (135) Liberals (VVD) Christian Dem Appeal (CDA) 17 Communists (CPN) 2 3 Dem '70 (D70) 1 0 Radicals (PPR) 3 3 3 2 Pacific Socialists (PSP) Reform Federation (RPF) Reform Pol Ass (GDV) Farmers Party (BP) 0 State Reform Party (SGP) 3 (11) 3 (15) Total 149 ISO and then broke down to a minority, 066, COA coalition. A new election had to be called in September 1982. As described in Laver and Schofield (1990), if the parties do have policy preferences, and if the space is essentially two-dimensional, then it is possible for there to be a single "dominant" party at the "core" position. A "core" is a preferred policy position, Z[, for party 1 say, such that no other position, y say, is preferred by a coalition commanding a majority of the seats. Attempts to use the spatial committee model have foundered on the difficulty of estimating party positions. The work presented in Budge, Robertson and Hearl (1987) attempted to use content analysis of party declarations (manifestos) to estimate party positions. However these estimated positions seemed excessively volatile. In this paper we have used the Euro-Barometer II data (Rabier and Inglehart, 1981) and the European Political Parties Middle Level Elites data (ISEIUM, 1983) to estimate party positions (further discussion of these data is provided in the next section, and in Appendix A. See also Quinn et al., 1996.) These data allow us to represent the positions of the four major [45 ] 264 c:: 0 'iii c: Q) E is 0 0 (ij '0 WD 0 (/) ';" -2 -1 o 2 Economic Dimension Figure 1. Dutch voter and party positions, 1979. parties in the Netherlands in 1979 in a two-dimensional policy space, W (see Figure 1). Assume for the moment that the four positions in this figure are preferred policy choices for the four parties and that each party, j, has a Euclidean policy preference derived from a utility function of the form Uj(x) = -llx-zjI12 (where Zj is its preferred position). Now draw the contract curves between the three pairs {CDA, VVD}, {PvdA, VVD} and {PvdA,CDA}. U sing the election results of 1977 to determine seat strength, it is very readily shown that the "core" is empty. To illustrate, the CDA position in Figure 1 can be beaten by a winning coalition of {PvdA,VVD} adopting a position on their contract curve which is nearer to their preferred positions. This in turn can be beaten by a position on the {PvdA,CDA} contract curve and this again can be beaten by a position on the contract curve of {CDA,VVD}. The three contract curves bound a triangle {CDA,VVD,PvdA} which has been termed the political "heart" (Schofield, 1993). It has recently been shown (Schofield, 1997b) that the "heart" Je can be identified with a local version of the uncovered set. If the heart is viewed as a correspondence Je, from the space of all party declarations and voter behavior, to the policy space, W, then Je is lower hemi-continuous. Thus Je admits a continuous selection (Schofield, 1995a). See Appendix C for details. [46] 265 To be more specific, given a vector of party positions, ¢, and post-election weights (seats), we assume that the coalition outcome is a lottery g, namely a set of government coalitions and associated coalition probabilities, g(X)( ¢ ), determined by ¢ and the stochastic electoral model X. Let Jf(X)(¢) be the realization of the heart once the electorate has responded to ¢, and let it denote all lotteries over Jf. Then our formal model assumes that the political outcome, g(X)(¢), belongs to it(X)(¢). As ¢ varies, then g is a continuous selection from it. If we impute preferences to political agents then, knowing g, we may solve the implicit game, g, to deduce Nash equilibria, ¢*, to this game. This is a model of high generality, but of little applicability if we cannot model party preferences or the electoral response, X, appropriately. To develop the Dutch example resulting from the 1977 election, we shall continue with the assumption that the parties are committed after the election to the various positions in Figure 1, and also suppose that coalitions form to propose policy points inside the realized heart. Without a core in 1977 the parties found great difficulty reaching an agreement. In particular the PvdA and CDA could not find an acceptable compromise. As we have observed, after months of negotiation the CDA and VVD eventually formed a minimal winning coalition government. We see this as providing some degree of empirical justification for the heart. To pursue the example, we can chart the change in the heart resulting from the 1981 election. As Table 3 makes clear, the gain of 9 seats by D66 meant that the {CDA,VVD} coalition lost its majority (dropping to 74 seats). The heart now contracts to become the triangle {PvdA,CDA,D66}. It is hardly surprising that this three-party coalition first formed. Nor is it surprising that it collapsed to the minority {CDA,D66} coalition. The rivalry between the CDA and PvdA led to a political crisis, to a new election in September 1982, and to a relatively long-lived coalition of {CDA,VVD}. Observe that the very small parties in 1977 and 1981 seemed to play no significant role in coalition bargaining. For this reason we ignore their effect on elections and coalition bargaining. Note that if only one dimension (the left-right economic axis) were relevant in the Netherlands then theory would suggest that the CDA (being typically at the one-dimensional core) could form a one-party minority government. To our knowledge, minority governments are very rare in the Netherlands and tend to be short-lived. This suggests, contra de Swaan (1973), that one-dimensional models of coalition behavior are inadequate, at least in understanding Dutch politics. The concept of the "political" heart, used here to interpret these two postelection situations in the Netherlands, is clearly based on the spatial theory of committees, since the heart assumes that the party positions are preferred [47 ] 266 policies, and uses the electoral strength of each of the parties to determine the pattern of winning coalitions. However, we can use the notion of the heart to infer something about the motivations of the parties before the elections. It is apparent that the simple maxim of increasing electoral vote is not adequate. Although the PvdA was the largest party in 1977, it could not find a coalition partner with which to form the government. It should also be observed that the principle of "maximizing electoral vote" ignores attitude to risk by the parties. In the stochastic electoral model discussed below, our estimates for the distribution of vote shares for the four parties show that there is a significant variance. For example, the 95% confidence interval for the number of seats which would have been obtained by the PvdA in an election in 1979 (as given by our estimation based on party positions) ranges from 42 to 54. We shall argue in the next section that no party in the Netherlands could choose an acceptable policy position, in the period under discussion, so as to give it a reasonable probability of being at the core. In the absence of a postelection core, negotiations between the parties were bound to be as difficult as we have observed they were. We can contrast this core-less example with one from Germany, also for 1979. Table 4 presents the election results between 1976 and 1982. Figure 2 gives the party positions obtained from ISEIUM (1983) data (see also Appendix A and Martin and Quinn, 1997). Clearly the three parties are almost, but not quite, colinear. If the positions were colinear, then the FDP would be at the core. However, Schofield (1986) shows that such a core is "structurally unstable", since small perturbations in party positions destroy the core property. Nonetheless, it can be inferred that the FDP is pivotal. The "grand" coalition of {CDU,SPD} was possible, but appears to have had low probability. In fact the {SDP,PDP} coalition formed in December 1976 and lasted until the October 1980 election. This was followed by a {CDU,FDP} coalition, which has persisted to the present. Although the Greens won 27 seats in 1982, it is clear that they have been effectively superfluous in coalition bargaining, at least until the present. (This may change after the forthcoming election [Summer 1998] in Germany.) Essentially the PDP could pick and choose between the two large parties. In this simpler situation, there may be some reason for one of the two large parties to seek a core position. However to be at a "structurally stable" core in this situation means to command a majority of the seats. Our electoral model, to be introduced below, can be used to determine whether this is indeed possible. We may also compare the two cases of an empty core (in the Netherlands) or a "structurally unstable core" (in Germany) with the possibility of a structurally stable core. As with the uncovered set, the heart and the core coincide [48] 267 N c 0 '<Ii cQ) E i:5 0 (ij '0 0 Cf) '7 -2 -1 o 2 Economic Dimension Figure 2. German voter and party positions, 1979. Table 4, Elections in West Germany, 1976, 1980 and 1982 1976 1980 1982 (Seats) Christian Democrats (CDU)a 243 226 244 Free Democrats (FOP) Social Democrats (SPO) 39 214 53 218 34 193 27 496 497 498 Greens (GR) Total aThe Christian Social Union (CSU) seats are included with the COU. when the core is non-empty. The post-election bargaining theory proposed in Laver and Schofield (1990) and Schofield (1993) suggests, moreover, that in a two-dimensional policy space, if the core is non-empty and structurally stable, then it can only be occupied by the "dominant" party (see Schofield, 1995b for the proof, and a more precise definition of "dominant"). To give an indication of how a structurally stable core may arise, consider an artificial situation for the Netherlands where we suppose that the D66 is located within [49] 268 the {CDA,VVD,PvdA} triangle, and controls 39 seats, with each of the other three parties controlling 37 seats. It is clear, in this case, that the D66 policy can be beaten by no other two-party coalition. The analysis provided in Laver and Schofield (1990) suggests, moreover, that such a dominant party at the core is likely to form a minority (one-party) government. In contrast, when a minority government forms that is not based on the core in this fashion, then it is nearly always short-lived. The best evidence is from Scandinavian countries: out of 50 governments in Denmark, Norway and Sweden, twentyeight were minority, and all but three were associated with core, dominant parties. The three minority, non-core examples were all short-lived, caretaker governments. If a party can control policy and obtain government perquisites from locating itself at a core, then the logical choice of any party would be to seek the core position. However, such a strategy is feasible only if that policy position is acceptable to the elite in the party. Note also that whether a position is core-like or not depends on the position (and strengths) of the other parties. Thus maneuvering for the core is much more difficult than simply maximizing expected vote. Indeed, there may be no core. In particular the McKelvey-Schofield-Saari results on core existence imply that the core is typically empty in weighted voting games in three dimensions. Thus, a "counter-core" strategy by a smaller party could be to introduce policy dimensions to destroy any possibility of occurrence of a core. However, for this to be effective the new dimension must be relevant in terms of the true preferences of the parties and of the voters. A second possibility, even in two dimensions, is that the underlying electoral response to party positions prohibits any party from advantageously locating itself at the core. To examine this possibility, we shall construct a "stochastic" model of voting for the Netherlands and Germany, based on multinomial probit (MNP). 4. An empirical model of electoral behavior based on pure spatial theory Appendix A presents the empirical procedure we adopted to construct individual and party ideal points in the Netherlands and Germany. Because of the way the scales were constructed, a respondent agreeing strongly that "greater effort should be made to reduce inequality" would be assigned a negative value, and thus tend to be on the left on the first "economic" dimension. On the other hand, a respondent in the Netherlands agreeing strongly that "women should be free to decide for themselves in matters concerning abortion" would obtain a negative score on the second dimension. The second [50] 269 dimension thus provides a technique for distinguishing Liberal voters (who might tend to vote for the VVD) and others who might tend to vote for the Christian Democratic Appeal. This second dimension in the Netherlands may be viewed as gauging beliefs on individual liberty, with a large negative value corresponding to "Libertarian" views. The second dimension in Germany is slightly different. A respondent with a high negative value on this dimension would tend to agree strongly that "greater public control should be exercised over multinational corporations" but to disagree strongly that "nuclear energy should be developed in order to meet future energy needs." We could identify this second dimension in Germany with "corporatism." However, for convenience, we shall refer to this second dimension in both countries as a "social" dimension. The factor loadings in both countries were renormalized so that 95% of all ideal points in each dimension lay in the range [-2,+2], with the mean position at (0,0). Note that although both "policy" dimensions in the two countries might appear to be correlated, our estimates of the underlying density functions (of voter ideal points) suggest that voter responses are not highly correlated in these dimensions. In particular, Figure 2 suggests, for example, that a German voter may simultaneously approve of egalitarian measures as well as believing that nuclear energy should be developed and multinational companies should be relatively unregulated. By this method we obtain a "profile" (x) = (Xi)iEN of voter ideal positions, where each Xi represents, in some degree, the economic and social beliefs of voter i. Using almost identical questions asked of the party elites, and taking the median of these elite positions (in both dimensions) for each party, we obtain an estimate of the "profile" of party positions 4> = (4)j)jEK as given in Figures I and 2. Given the empirical distribution of voter ideal points, we can smooth this to approximate the underlying density function of the voters' positions. The backgrounds in Figures I and 2 give our estimates of these densities. The combination of (x) and (4)) gives a data matrix 0 = (Oij}if KiEN . In both countries voting intentions were known, so this gives an array y = (Yij)K N where Yij = 1 iff voter i intends to vote for j, and 0 otherwise. The challenge is to estimate a probability matrix (Xij) given (Oij) and (Yij), so that a realization(X ij ) of (Xij) is close to (Yij). Following the standard assumptions of the pure spatial probabilistic model described in Section 2, we assume Prob(uij > Uil for alII i= j) = Prob(El - Ej < f:l(oJ - DB) : I Xij i= j) Let ei = (EI-Ej, ... Ej-l-Ej, Ej+l-Ej, ... Ek-Ej) be a (k-l) dimensional stochastic variable, with probability density function f and let ~{ = «o~), ... , (oi O~»E9k-l. [51] 270 Using Equation 1, and conditioning on our assumptions on the implicit covariance matrix ~ we obtain (2) Here the integral is (k-l) dimensional. The maximum likelihood estimator for X is (3) Unlike the usual model where the E'S are assumed to be iid, with zero expectation, we adopt the multinomial probit (MNP) assumption that f is the with general variance-covariance matrix, ~. (For multivariate normal N(O,~) a discussion of probit models, see Alvarez and Nagler, 1996.) Markov Chain Monte Carlo (MCMC) techniques (Albert and Chib, 1993; McCulloch and Rossi, 1994) were used to estimate ~. The details of our estimation results for the pure spatial model for the Netherlands and Germany are reported in Appendix A. Figure 3 presents our estimate of (Xij) in the case of the Netherlands. Clearly, our estimation of Xij will depend on the ideal point of voter i and the location of party j. For example, a Dutch voter with an ideal point anywhere along the "social" axis and at the left extreme on the economic axis "will vote" for the PvdA with probability exceeding 0.9. On the other hand, any voter on the extreme right of the economic axis "will vote" for the VVD with certainty if near the libertarian extreme on the social axis and "will vote" for the CDA with probability about 0.8 if near the opposite end of the social axis. These probabilities make intuitive sense. To illustrate further, Figure 4 shows the distribution of ideal points of those voters who declared in the survey that they intended to vote for PvdA. The distribution of these voters has been normalized to sum to unity. Clearly the distribution is unimodal, with its mode near the PvdA position. Obviously enough there are voters whose ideal points are nearer the D66 position for example, yet who intend to vote for the PvdA. Clearly a pure deterministic model of voting does not suffice. Our estimation would assign to a voter, with such an ideal point, probabilities of approximately .2 for D66, VVD, and CDA each, and 0.4 for PvdA. Such an assignment is not implausible. Figure 5 shows a similar distribution for the ideal points of "CDU" voters in Germany in 1979. Appendix B presents our estimation results for a joint electoral model for the Netherlands and Germany, based on multinomial probit, using not only the spatial data but also individual characteristics of the [52] 271 CI! o <CD ;to "C:''''*: ~o "! o o co ° S"! 'P 0 ~"': ~ 0 ~ o Figure 3a. Estimates of voter probabilities in the Netherlands for the PvdA and D66. [53] 272 ~ «'It! 00 ~ Q. 0 C'! o o .... !l,. Figure 3b. Estimates of voter probabilities in the Netherlands for the CDA and VVD. [54] 273 C\I c CDA 0 '<;; c Q) E (5 0 0 (ij '0 WD 0 en "';" ·2 ·1 o 2 Economic Dimension Figure 4. Dutch party and PvdA voter positions, 1979. voters (derived from the Rabier, Inglehart, 1981, survey), such as occupation and income, religion, education, size of town of domicile, etc. To briefly summarize the results, we find in both countries that the f3coefficient of the pure spatial model is statistically significant while the estimated covariance matrix is significantly different from the iid model. We also find that the pure spatial model is statistically superior to the joint model in the case of Germany. For the Netherlands, the joint electoral model is superior to the pure spatial model. We attempt to account for this result in the next section. To illustrate the effectiveness of the pure spatial model, Table 5 compares the national, the sample, and the "estimated" vote shares for the four parties in the Netherlands. Since only four parties are considered in the model , we give approximate values for the share going to each party, computed as a percentage of the total national vote going to the four parties. It is clear that the sample shares (based on n = 529) are quite close to the national shares for the 1977 and 1981 elections. The estimated vote shares are computed by finding the expected value, E(n j), of the vote for each party. Since vote shares are derived from (Xij) and these are random variables, each vote share is also a random variable. Empirical histograms for the vote share of the four parties in the Netherlands are given in Figure 6. The 95% confidence intervals on the estimated vote shares include the values for the sample shares for three of the four parties. (See Table N2 of the [55] 274 c 0 '0; c CD E 0 0 (ij '0 0 U) ";" -, -2 o 2 Economic Dimension Figure 5. German party and CDU voter positions, 1979. Table 5. Vote shares in the Netherlands' elections, 1977 to 1981 Party Optimal 1981 Sample share in 1979a Estimated 1977 sharec share % % % % % D66 6.1 12.6 10.4 10.6 II ±4 PvdA 38.0 32.4 36.9 35.3 45 ± 3 CDA 35.9 35.2 33.8 29.9 40 ± 3 40 ± 3 VVD a b c [56] Share of national voteb 20.0 19.8 18.9 24.2 100.0 100.0 100.0 100.0 Based on Euro-barometer sample with n =529. Vote shares are very close to seat shares, because of proportional representation. Shares are calculated on the basis of total vote to the four large parties, using Keesing 's Contemporary Archives. Pure spatial model, as described in Appendix A. 275 0 ,0 0.1 0,2 0.3 0.4 25 20 15 10 (ij 0 5 a c 0 I- (l) () (j) 0.. 25 20 15 10 5 0 0.0 0.1 0.2 0,3 0.4 Vote Share Figure 6. Histograms of estimated vote shares for four parties in the Netherlands (based on 1979 data). Appendix for details.) The one anomaly to note is that, for the VVO, the national vote share (approximately 20%) and sample share (18.9%) lie outside the 95% confidence interval (20.8%, 28%) for the estimated expected vote share. However, the two empirical vote shares lie inside the 95% confidence interval of (14.9%, 24.6%) obtained from the joint model. The two models can be compared by considering Tables N2 and N3 in the Appendix. Table 6 compares the national vote shares for the three parties in the 1976 and 1980 elections, together with the sample shares and our estimates, for Germany. Figure 7 gives the empirical histograms of the random variables 1Tj for each party in 1979 in Germany. The point to note is that 95% confidence intervals for expected vote shares under the pure spatial model for the three parties are approximately: COU (.49, .58), SPO (.34, .46) and FOP (.01, .14). These confidence intervals are very similar to the confidence intervals for the joint model (compare Tables G2 and G3 in the Appendix). Although these estimates do not provide much more information than the sample, they do allow us to draw conclusions about the electoral consequences were parties to adopt "strategic" policy positions. This we do in the next section. [57] 276 Table 6. Vote shares in Germany, 1976 and 1980 Party Share of national voteb Sample share a % in 1976 in 1979, % % in 1980 49.1 45.4 51.2 53.2 FDP 8.0 10.8 5.9 6.3 6.4 SPD 42.9 43.8 42.9 40.5 42.6 100.0 100.0 100.0 100.0 100.0 d 0 .2 0.4 0 .6 CD 30 I- 51.0 Based on Euro-barometer sample with n =543 in 1979. Approximately 320,000 votes (out of over 37 million) for small parties are not included. These parties were not represented in the Bundestag. Vote shares are calculated as % of total vote to the three parties. In particular, the Greens gained 15% of the vote, but no seat in 1980. Based on the pure spatial model described in Appendix A. Using individual characteristics, as described in Appendix B. 00 0 Estimated % shared CDU a b iii Estimated % sharec SPD 25 20 '0 c: .5 Q) ~ <I> a. 10 J 0 0.0 02 0 ,4 [ 1i Il 00 0.6 0.2 04 0.6 Vote Sha re Figure 7. Histograms of estimated vote shares for the three parties in Germany (based on 1979 data). It is worth noting, however, that the model we have constructed here, based as it is on the assumption of "stochastic voters", suggests that the underlying variation in vote shares is higher than might be supposed from simple analysis of electoral samples. 5. Models of party behavior derived from the analysis As we have emphasized, party positions (in the model just described) were estimated by taking the median of party elite positions. Thus the positions of the parties are not the same as "declarations" to the electorate. Nonethe- [58] 277 less, using these estimates of party positions, we found that the pure spatial MNP model of voting that we have just presented gives a plausible account of how Dutch and German voters respond to party positions. Thus we may assume that these party positions did have an effect on electoral response in the period in question. We now show that changing party positions would lead to an increase in expected party vote. Let us suppose that the elite members of the parties in question had beliefs concerning electoral response that are compatible with the spatial electoral model just discussed. Consider the thought experiment where each party, j, considers a possible change in position from ¢j to a "strategic" position, ¢*j, assuming all other party positions are fixed. Based on our MNP model, we compute the optimal ¢*j, chosen as to maximize expected vote. In the Dutch case the optimal position for each party is at the electoral mean (that is, at the mean (0,0) of the sample ideal points). After such a move, our estimates of the vote increase by significant amounts. Table 5 shows the estimated optimal vote shares for the four Dutch parties. It indicates that the CDA, for example, could increase expected vote share by up to 10%, from about 30% to 40%. Had the CDA positioned itself at the electoral mean in 1977, we estimate that its expected number of seats would have been 55, rather than 49. 2 Why do the parties not move to the electoral center, as predicted by the Downsian probabilistic model? To continue the thought experiment, suppose that during the 1977 election, the CDA positioned itself at the electoral mean, so that it gained 55 seats, instead of 49. Let us assume further that parties are committed to the positions they adopt. (We provide, below, an argument for this commitment.) We shall argue that this increase of seats for the CDA (coupled with the change of declared policy positions) would have adversely affected the final policy outcome, from the point of view of the CDA elite. To see this note that even with 55 seats, the CDA position at the mean, (0,0), is not a core position. In all probability, the CDA would have been obliged to form a coalition with the VVD. However, if the CDA were required to bargain from its declared position, then the policy compromise could in fact be further from its true policy position than the actual 1977 outcome. The same is true for the 1981 election situation. More importantly, given our estimate of the stochastic relationship between the CDA position and its share, there is no CDA position at which the CDA could have expected to gain a straight seat majority (given our confidence intervals). Moreover, the probability that the CDA could have occupied the core is essentially zero. In the same way, a thought experiment involving the PvdA suggests that it could have won approximately 60 (±4) seats in the 1981 election, by moving to the origin. If the PvdA were obliged to commit to that position, then it would indeed be at a core with high probability and possibly able [59] 278 to form a minority government. Now compare the situation in 1981 if the PvdA declares its "true" position in contrast to the outcome resulting from the PvdA at a core position at the origin. In the former case there is a significant probability of a {PvdA,D66,CDA} coalition in which the PvdA can presumably affect policy. On the other hand, at the core the PvdA may be able to implement a minority government, control all the perquisites, but be obliged to implement a policy far from its ideal position. It is not implausible that the PvdA party elite would prefer to declare the party's "true" position (we explore this somewhat further below). In a situation such as we have described in the Netherlands, we suggest that there was an insignificant probability that any party could occupy a postelection core position and implement a policy that it found preferable to the lottery resulting from a true declaration of preference. The case in Germany is somewhat different. As we have seen, the FOP is "almost" at the core position and has been in almost every government for decades. It could be dislodged from this position were the SPD to move sufficiently close to the FOP. However, assuming policy commitment again, this would have obliged the SPD to implement a policy that the elite of the party would have found highly undesirable. We infer that if a party adopts a policy position simply to increase the number of seats that it controls, then this may not increase the power that the party has to implement desired policy. The willingness of a party to adopt vote maximizing policies depends, therefore, on the trade-off in its "utility" function between the perquisites of office and the importance of policy. However, to determine the optimal position, it would appear that the party must compute not just the effect of a change of position on its vote share, but must also consider how that change affects the post-election lottery over the choice of the coalition government. If there is no policy change that leads to a potentially "dominant" party being at a core position, with significant probability, then there are plausible arguments to be made that the declarations of a "true" policy preference is "rational" for such a party. To make this argument clearer, we need to examine in more detail our notion of the preferred point, ¢j, of party j. In our estimation, we chose each party's point to be at the median (in both dimensions) of the ideal points of the party elite (using the ISEIUM, 1983, responses of party delegates). To illustrate, Figure 8 represents our estimate of the density function of the ideal points of the PvdA delegates. Comparing Figure 8 with Figure 4 (showing PvdA voter ideal positions), it is clear that the PvdA delegate variance is much less than PvdA voter variance. It is reasonable to suppose that the PvdA has some leeway in declaring a pre-election position to the electorate. However, it is also obvious that the party cannot choose any position at will. If we view the [60] 279 c 0 'u; 0 cQ) E is 0 CDA 0 "iii '5 VVD 0 '66 0 en "';" -2 -1 o 2 Economic Dimension Figure 8, Empirical distribution of estimated ideal points of PvdA delegates (based on 1979 Dutch data), PvdA delegates as a committee, then it is almost certain that an internal PvdA core will exist (under majority decision-making by the delegates). This "internal" party core will lie at the (two-dimensional) median of the delegates' ideal points. It is therefore sensible to regard this "internal" core position as the party's ideal point, as we have done in the estimation just discussed.3 Indeed, we can regard this internal party core point as the sincere choice of the party. Now consider internal party decision-making over the choice of a policy point to declare to the electorate. Each delegate of the party has an ideal policy point, as well as beliefs about electoral response and the nature of postelection negotiation. Thus, each delegate will have a personal choice for the best response the party can make to the policy positions of the other parties. It is reasonable to suppose that a delegate's "best response" is continuous in the delegate's ideal point. Since there is a "sincere" internal party core (under majority rule among the delegates), there will also be an unbeaten choice among the party delegates for the policy point to offer to the electorate. Such a point we can refer to as the strategic choice of the party. Clearly, the party may commit to this position by choosing the leader of the party to be exactly that delegate whose ideal point coincides with the strategic choice of the party. Now, consider a situation where each party makes a strategic best response to the declarations of the other parties. Since the strategic best response will [6] ] 280 Table 7. Maximizing expected vote Sample "Ideal" share D66 PvdA Party Estimated 95% point "Strategic" point vote C.1. 10.4 (-0.1, -0.2) (-0.1, -0.2) 10.8 (4.5, 16.5) 36.9 33.8 (-1.1, +0.0) (+0.6, +0.4) (-1.1, +0.0) CDA (+0.6, +0.4) 36.4 32.8 (33.0, 40.0) (28.0,37.0) VVD 18.9 (+ 1.2, -0.3) (+ 1.5, -1.0) 20.0 (16.5, 23.5) be continuous in the other parties' positions, standard arguments can be used to demonstrate existence of a mutual strategic best response, or Nash equilibrium. (See Schofield, 1996, for an outline of such an existence proof.) Notice that this pre-election Nash equilibrium in party positions is credible after the election, since each party position will be associated with the ideal point of a member of the party elite. Moreover, this member, the "leader" of the party, will have every incentive to bargain effectively with other party leaders in post-election coalition negotiations. Let us continue with this thought experiment and suppose that the four parties in the Netherlands had adopted positions that were strategic Nash equilibria with respect to their beliefs about the electoral response to these positions, circa 1979-81, and also their beliefs about post-election bargaining. Assume further that the parties' electoral beliefs are generally compatible with the model of stochastic voting that we have constructed. If we modify the parties' positions to bring the model's estimated expected vote shares into line with the 1981 election results, then we have a method to estimate the parties' declared Nash equilibrium position. Since we have observed an anomaly in the expected vote share of the VVD, as estimated by our MNP model, this suggests that we consider a position for the VVD that differs from this party's sincere position. If we keep the three other parties at their sincere positions and move the VVD from its sincere position (+1.2,--0.3) to (+ 1.5,-1.0), then (as Table 7 reports) the estimated expected vote share of the VVD drops from 24.2% to 20.0% and the 95% confidence interval for the party's vote share changes from (20.8%,28%) to (16.5%,23.5%). This interval now includes the sample share and the national vote shares of the VVD in the 1977 and 1981 elections. The estimated expected vote shares of the other three parties are still close to the sample shares. Of course we have no direct evidence that the VVD did indeed declare a policy position to the electorate in the 1977 and 1981 elections that was more radical than its sincere position. [62] 281 In our view, the choice between the joint electoral model (as described in Appendix B) and the pure spatial model (incorporating the VVD strategic position) depends on the plausibility of a theoretical argument about why a party such as the VVD could rationally choose to declare a policy point which is more radical or extreme than its sincere internal core position. Appendix C gives a stylized argument, adapted from Schofield (1993) and Schofield and Parks (1996), to show that if post-election coalition formation is represented as a lottery, then a party such as the VVD could indeed choose a more extreme position. The point to note about this model of party calculation is that although the VVD may believe that it would lose a few percent of the vote by declaring a more radical position, it could also plausibly believe that it would, in expectation, pull the policy compromises made either with the PvdA or CDA towards its internal core position. The suggestion that the VVD adopted a more extreme position also allows us to make sense of the 1977 attempt by the CDA to form a coalition with the PvdA. As Figure 1 illustrates, the {CDA,VVD} internal core points are nearer to each other than the {CDA,PvdA} internal core points, and so the former coalition might appear more likely. If, however, the VVD had adopted the more extreme position, as we have suggested, so that a {CDA,VVD} compromise would favor the VVD, then it would indeed be rational for the CDA to approach the PvdA first. Of course, such a strategic declaration by the VVD is inherently risky. Indeed, it may have opened the way for smaller parties to enter the electoral fray in 1981, leading to the loss of the {VVD,CDA} majority, and to the diminution of the attractiveness of the VVD as a coalition partner. Having argued that our analysis suggests why the VVD may have declared a more extreme position, we can also add somewhat to our earlier suggestion that the PvdA and CDA would be likely to declare their true, or internal core, positions. For example, in the case of the PvdA, the more centrist delegates of the party would presumably prefer a move by the party to the core position, hoping for a minority PvdA government. On the other hand, more extreme delegates of the party would prefer a leader of the party to adopt a position on the left of the economic dimension, hoping to influence policy through coalition negotiation. For both the PvdA and CDA we might expect considerable internal disagreement within these parties, resulting however, in policy declarations chosen to be compromises close to the sincere policy points of the two parties. 4 The general model we propose emphasizes the heterogeneity of ideal policy points within each party, j, together with the selection of a leader, [63 ] 282 whose ideal point, Zj, coincides with the majority choice of an equilibrium declaration, c/>j*' made by the party elite. 6. Conclusion Models of elections have generally been based on the "Downsian" assumption of expected vote maximization. The conclusions of such models are at variance with the observation that parties do not converge to an electoral center. The empirical work presented here suggests that a pure spatial electoral model is quite satisfactory in modeling electoral response to party positions. The estimation allows us to determine approximately how much party vote shares would increase as a result of policy convergence to the electoral mean. Although nominal political power may be increased by such a move, we argue that the power to implement desired policies need not increase. For the Netherlands we suggest that the electoral model allows us to distinguish between sincere and strategic party positions. In particular, we infer that the two larger parties adopted policy positions that were close to their sincere, or internal core, positions. We have also provided an argument to explain why a small, non-centrist party such as the VVD, could reasonably adopt a policy position that is more extreme than its sincere policy position. In Germany, on the other hand, we infer that the balance of preferences within each party leads to the declaration of a policy point close to the sincere, internal core, of the party. For both countries, our empirical and theoretical modeling provides an explanation for the evident fact that parties facing proportional representation systems do not converge to the electoral center. We argue that understanding political choice in multiparty democracies can be based on the following research program: (a) build a stochastic electoral model, of the kind we have constructed here, which can be used as a proxy for the beliefs ofthe party elite, (b) assume that each party's declaration is associated with the position of a member of the party elite, who then bargains with other party representatives using the post-election party strength as a political resource, (c) model the post-election coalitional bargaining game in terms of a lottery of expected coalition outcome inside the post-election heart, (d) solve the internal pre-election party negotiation game over the choice of party representative and party declaration. We intend to develop this framework in future research, by rejecting the usual assumption that each party is a unitary actor, and by studying the relationship between party declarations (or manifestos) and the distribution of elite ideal points. We also hope to relate this model of internal party choice [64 ] 283 to current work on post-election coalition negotiation (Schofield, Sened, and Nixon, 1997). We suggest that the combination of stochastic electoral models of elections together with spatial committee analysis can provide a theoretically powerful, and substantively relevant way to understand party dynamics in multiparty polities based on proportional representation. Although plurality electoral systems are more difficult than proportional systems to analyse, it should be possible, in principle at least, to extend the analysis to study strategic behavior in Westminster style polities. Finally, since the underlying model involves delegates, it should also be possible to deal with factional polities without disciplined parties. By concentrating on elite political actors, instead of parties, it may be possible to begin to consider the question of the formation of elite groups and the building of political parties. By this method we may gain a better understanding of the differences between the various types of polities mentioned in the introduction, and contribute to the development of a formal theory of comparative politics. Notes 1. The effective number is the inverse of the Herfindahl index of concentration. 2. Note that the "electoral heart" of the sample, and presumably of the entire electorate, is a very small set lying close to the origin (0,0). The relationship between the "electoral heart" and the "political heart", as generated by party positions and weights, is one way to characterize a multiparty system. 3. Even if the internal PvdA core is empty, the "heart" of the PvdA delegates, viewed as a committee, will be a very small set centered on this two-dimensional median. 4. Of course, one could fine-tune the electoral model to equate all expected vote shares with sample vote shares, leading us to infer that all parties "strategize". This would seem somewhat pedantic, given the margins of error. The indirect evidence for VVD strategizing persists even if we use the delegates' mean position rather than the median. References Albert, J.H. and Chib, S. (1993). Bayesian analysis of binary and polychotomous response data. Journal of the American Statistical Association 88: 669-679. Alvarez, R.M. and Nagler, 1. (1996). Correlated disturbances in discrete choice models: A comparison of multinomial probit models and logit models. Typescript. Pasadena: California Institute of Technology. Aranson, P., Hinich, M. and Ordeshook, P. (1974). Election goals and strategies: Equivalent and non-equivalent candidate objectives. American Political Science Review 68: 135-152. Axelrod, R. (1970). Conflict of interest. Chicago: Markham. [65 ] 284 Budge, 1., Robertson, D. and Hearl, D. (Eds.). (1987). Ideology, strategy and party change. Cambridge: Cambridge University Press. Chib, S. and Greenberg, E. (1996). Multinomial probit by Markov Chain Monte Carlo. Typescript. Washington University in St. Louis. Coughlin, P. (\ 992). Probabilistic voting theory. Cambridge: Cambridge University Press. Cox, G.w. (1984). An expected utility model of electoral competition. Quality and Quantity 18: 337-349. Cox, G.W. (1987). The uncovered set and the core. American Journal of Political Science 408-422. de Swaan, A. (1973). Coalition theories and cabinet formations. Amsterdam: Elsevier. Downs, A. (1957). An economic theory of democracy. New York: Harper and Row. Duverger, J. (1954). Political parties: Their organization and activity in the modern state. New York: Wiley. Duverger, J. (1984). Which is the best electoral system? In A. Lijphart and B. Grofman (Eds.), Choosing an electoral system. New York: Praeger. EneIow, J.M. and Hinich, J.J. (1984). The spatial theory of voting: An introduction. Cambridge: Cambridge University Press. Feddersen, T.J. (1992). A voting model implying Duverger's law and positive turnout. American Journal of Political Science 34: 1005-1016. Herman, V. and Pope, J. (1973). Minority governments in western democracies. British Journal of Political Science 3: 191-212. Hinich, MJ. (1977). Equilibrium in spatial voting: The median voter result is an artifact. Journal of Economic Theory 16: 208-219. Hotelling, H. (1929). Stability in competition. Economic Journal 39: 41-57. ISEIUM. (1983). European elections study: European political parties' middle-level elites. Europa-Institut, Universitat Mannheim. Kass, RE. and Raftery, A.E. (1995). Bayes factors. Journal of the American Statistical Association 90: 773-795. Keesing's Contemporary Archives. Harlow: Longman. Kramer, G.H. (1978). Existence of electoral equilibrium. In P. Ordeshook (Ed.), Game theory and political science. New York: New York University Press. Laver, M. and Schofield, N. (1990). Multiparty governments. Oxford: Oxford University Press. Lin, T., Enelow, J. and Dorussen, H. (1996). Equilibrium in multicandidate probabilistic spatial voting. Typescript. University of Texas, Austin. Martin, A.D. and Quinn, K.M. (1997). Evaluating theories of West German voting behavior. Typescript. Washington University in St. Louis. McCulloch, R and Rossi. P.E. (1994). An exact likelihood analysis ofthe multinomial probit model. Journal of Econometrics 64: 207-240. McKelvey, R.D. (1986). Covering, dominance and institution-free properties of social choice. American Journal of Political Science 30: 283-314. McKelvey, RD. and Schofield, N. (1986). Structural instability of the core. Journal of Mathematical Economics 15: 179-198. Osborne, M.J. (1993). Candidate positioning and entry in a political competition. Games and Economic Behavior 5: 133-151. Osborne, M.J. (1995). Spatial models of political competition under plurality rule: A survey of some explanations of the number of candidates and the positions they take. Canadian Journal of Economics 28: 261-301. Poole, K.T. and Rosenthal, H. (1991). Patterns of congressional voting. American Journal of Political Science 35: 228-278. [66] 285 Quinn, K.M., Martin, A.D. and Whitford, A.B. (1996). Explaining voter choice in a multiparty democracy: A look at data from the Netherlands. Typescript. Washington University in St. Louis. Rabier, J.-R. and Inglehardt, R. (1981). Eurobarometer 11. April 1979: The Year of the Child in Europe Ann Arbor: Inter-University Consortium for Political and Social Research. Riker, W. (1962). The theory of political coalitions. New Haven: Yale University Press. Riker, W.H. (1982). The two-party system and Duverger's Law: An essay in the history of political science. American Political Science Review 76: 753-766. Saari, D. (1996). The generic existence of a core for q-rules. Economic Theory 9: 219-260. Schofield, N. (1985). Social choice and democracy. Heidelberg: Springer-Verlag. Schofield, N. (1986). Existence of a 'structurally stable' equilibrium for a non-collegial voting rule. Public Choice 51: 267-284. Schofield, N. (1993). Political competition in multiparty coalition governments. European Journal of Political Research 23: 1-33. Schofield, N. (l995a). Existence of a smooth social choice functor. In W. Barnett, H. Moulin, M. Salles, and N. Schofield (Eds.), Social Choice. Welfare and Ethics. Cambridge: Cambridge University Press. Schofield, N. (1995b). Coalition politics: A formal model and empirical analysis. Journal of Theoretical Politics 7: 245-281. Schofield, N. (1996). The heart of a polity. In N. Schofield (Ed.), Collective decision making: Social choice and political economy. Boston: Kluwer. Schofield, N. (1997a). Multiparty electoral politics. In D. Mueller (Ed.), Perspectives on public choice. Cambridge: Cambridge University Press. Schofield, N. (1997b). The Cl topology on the space of smooth preference profiles. Social Choice and Welfare (forthcoming). Schofield, N. and Parks, R. (1996). Existence of a Nash equilibrium in a spatial model of coalition bargaining. Typescript. Washington University in St. Louis. Schofield, N., Sened, I. and Nixon, D. (1997). Nash equilibrium in multi-party competition with 'stochastic' voters. Annals of Operations Research (forthcoming). Taylor, M. and Laver, M. (1973). Government coalitions in Western Europe. European Journal of Political Research I: 205-248. Wada, J. and Schofield, N. (1996). Bargaining in the Liberal Democratic Party of Japan. In N. Schofield (Ed.), Collective decision-making: Social choice and political economy. Boston: Kluwer. Appendix A: Spatial model The Euro-barometer questionnaire (Rabier and Inglehart, 1981) asked seven questions covering topics such as income distribution, terrorism, nuclear energy, public control of enterprises, environmental protection, control of multinational corporations and abortion. Seven similar questions were asked of political elites of known party allegiance (through the Institut fur Sozialwissenschaft and the Europa Institute, Universitat Mannheim, 1983). Exploratory factor analysis uncovered two factors in both the Netherlands and Germany. Tables N-I and G-\ give the factor [67] 286 Table N-l. Factor loadings, The Netherlands Issue Dimension 1 Dimension 2 Income distribution +.510 (10.86) -0.148 (1.92) Terrorism -0.232 (4.28) -0.253 (2.51) Nuclear energy -0.297 (6.74) Enterprises +0.526 (12.0) Environment +0.306 (7.46) MNC +0.612 (12.6) -0.229 (2.42) Abortion +0.327 (5.56) +0.39 (2.45) Chi-square over d.oJ. = 1.76. Sample size (n) = 529. Table G-l. Factor loadings, Germany Issue Dimension 1 Dimension 2 Economic "Corporatism" Income distribution .331 (5.8) +0.156 (2.51) Terrorism 0.351 (6.2) -0.273 (3.73) -0.641 (4.78) Nuclear energy Environment 0.668 (9.1) -0.173 (2.62) MNC 0.343 (5.71) +0.222 (3.26) Abortion 0.283 (5.24) Chi-square over d.o.f. = 3.11. Sample size (n) = 543. loadings on the various issues as follows (t-values in parentheses). In Germany, the question on enterprises was redundant. Responses to the seven questions (ranging from strongly favor (I) to strongly oppose (4)) were renormalized to have mean 0, standard deviation I. The factor loadings were then used to generate scoring coefficients for each of the responses for each country. These were combined to give for each voter an ideal point on the two dimensions. The same technique was employed for the political elites. Knowing party allegiance we computed the median of the elite for each party on each dimension. This was used as our estimator for each party's ideal or "sincere" policy point. For reasons of identification of the estimation for the Netherlands we adopted a variant of the probabilistic model, using the D66 party as a baseline. That is, letting ¢4 be the estimated position of the D66 party we assumed that realized utility of [68] 287 voter i for party j, given (411, ... ,414), was Uij(41) fJo - fJ(8ij)2 Ui4(41) = E4· + fJ(8i4)2 + Ej for j = 1,2,3, and Just as in Section 4, Prob (Uij > Uil for alII =1= j) =Prob(ej < fJll\). Under the usual assumption of independence of errors it is an easy matter to show that the variance (EI-Ej) = variance (EI) + variance(Ej) while the covariance terms are cov( EI-Ej,Em-Ej) =variance (Ej). In our analysis we used a statistical routine involving Markov Chain Monte Carlo Simulation and data augmentation, due to Chib and Greenberg (1996), that did not assume iid errors. (Technical details can be found in Quinn, Martin, and Whitford, 1997, and Martin and Quinn, 1997). To present the results for the Netherlands, we use the code {PvdA,VVD,CDA,D66} == {1,2,3,4}. We define the covariance terms aIm = COV(EIE4,E m-E4) and the variance terms a2u = var(El-E4). Table N-2 presents our estimate for these variance, covariance terms, together with the estimated coefficients fJo (constant) and fJ (spatial distance). Note that the model is unchanged if all utility terms are multiplied by a constant. Thus, to identify the model we assume a 211 = 1.0, and present the variance and covariance estimates in terms of a211. The logarithms of the marginal likelihood for these pure spatial models are -545 (for the Netherlands) and -506 (for Germany). The Bayes' factors (Kass and Raftery, 1995) for comparing these models with a null hypothesis (that X is not determined by the matrix 8) gives vanishingly small probability to the null hypothesis. The estimated 95% confidence intervals strongly suggest that the coefficients, fJo and fJ, are significantly non-zero. Moreover, confidence intervals on the covariance matrix indicate that the covariance terms are generally not half variance terms (as would be true under iid). To present the estimation for Germany, we use the code {SPD,CDU,FDP} = { 1,2,3} and normalize with respect to the FDP. Table G-2 presents our findings. Again a211 is set at 1.0. Clearly the two variance terms (a 211 and a 222) are dissimilar. It is possible, but not probable, that the covariance term a 12 is 0.5. Appendix B: Individual factors We also estimated a joint model utilizing the spatial matrix (8ij) but involving individual and demographic characteristics that were recorded in the Rabier-Inglehart Survey. [69 ] 288 Table N-2. Estimation results for the Pure Spatial Model in the Netherlands Parameter Party Spatial distance Constant Posterior mean SD 95%CI Lower Upper 0.456 0.512 0.120 0.064 0.230 0.391 0.710 0.638 1.000 -0.376 -0.107 1.012 0.384 0.696 0.000 0.261 0.188 0.289 0.200 0.248 1.000 -0.769 -0.426 0.618 0.092 0.360 1.000 0.084 0.224 1.546 0.752 1.134 0.044 0.027 0.022 0.027 0.038 0.309 0.208 0.255 0.182 0.397 0.280 0.342 Variance-covariance estimates a 2 11 a12 a13 a 222 a23 a233 Sample vote share D66 PvdA VVD CDA 0.104 0.369 0.189 0.338 Predicted vote share D66 PvdA VVD CDA 0.106 0.353 0.242 0.299 Table N3 records the effect of incorporating these additional characteristics. As the table shows, estimated vote shares for D66 and PvdA hardly shift. The high estimated value of the VVD share now drops from 24.2% to 19.5% (close to the sample share of 18.9%), while the low share for the CDA of 29.9% climbs to 34.9% (close to the sample share of33.8%). The logarithm of the marginal likelihood increases from -545 to -515. To interpret the increase of log likelihood as we change from the pure spatial model, to the joint model, we can compute the Bayes' Factor (Kass and Raftery, (1995) between the two models. This is simply the ratio of the marginal likelihoods of the joint and spatial models, and can be computed as exp (-515 + 545) =exp (30) > 1013. This suggests that the joint model is superior, in a statistically significant sense, to the pure spatial model. As a comparison of the confidence intervals on the VVD vote share suggests, the joint model is statistically superior to the pure spatial model in capturing voting behavior in the Netherlands. Table G3 records the effects of incorporating the individual and demographic characteristics in the model for Germany. Clearly, expected vote shares become close [70] 289 Table G-2. Estimation results for the Pure Spatial Model in Germany. Parameter Party Spatial distance Constant Posterior mean SD 95% CI Lower Upper 0.239 1.078 0.066 0.228 0.120 0.604 0.377 1.509 1.000 0.025 3.735 0.000 0.431 2.009 1.000 -0.644 1.361 1.000 0.727 7.942 0.039 0.038 0.029 0.013 0.337 0.486 0.136 0.462 0.578 Variance-covariance estimates U 2 11 U\2 U 2 22 Sample vote share FDP SPD CDU 0.059 0.429 0.512 Predicted vote share FDP SPD CDU 0.063 0.405 0.532 to the sample shares. In particular the SPD share increases from 40% to 42.6% (close to the sample of 42.9%). The logarithm of the marginal likelihood drops from -506 to -623. The Bayes' Factor between the spatial and joint models for Germany is now exp (-506 + 623) = exp (117) > 1050. This suggests that the pure spatial model is statistically superior to the joint model in Germany, at least in the period in question. This is intuitively obvious from the fact that all sample vote shares lie within the 95% confidence intervals of the predicted vote share. Appendix C: Strategic party choice Here, we briefly outline the model of post election coalition bargaining, proposed in Schofield and Parks (1996). Suppose that three parties {A,B,C} have the ideal points presented in Figure 9, and that the parties are committed to their declarations (in the manner proposed in the text). For ease of presentation suppose that they believe that, whatever their declarations, the electoral response will result in party strengths such that any pair of parties will control a majority. If the parties' declarations are [71] 290 Table N-3. Estimation results for the Joint Model, the Netherlands Variable Party Spatial distance Manual labor Religion Income Posterior mean OA99 SD 95% CI Lower Upper 0.065 0.376 0.627 PvdA 0.926 0.252 VVD CDA -0.714 OA56 OA44 -1.699 0.103 0.239 0.264 -0.274 0.777 PvdA -0.060 0.082 -0.225 0.101 1.421 VVD -0.112 0.114 -0.353 0.092 CDA 0.504 0.098 0.339 0.735 PvdA VVD -0.012 0.022 0.028 -0.054 0.032 0.108 CDA 0.049 -0.010 0.021 -0.001 -0.052 PvdA 0.382 VVD -0.131 0.103 0.137 -OA27 0.582 0.111 CDA -0.046 0.113 -0.279 0.158 PvdA -0.083 -0.007 0.029 -0.141 -0.028 VVD 0.038 -0.080 0.067 CDA -0.077 0.028 -0.140 -0.026 0.256 0.239 -0.185 0.767 1.000 -0.350 0.000 0.265 1.000 -0.772 1.000 0.104 0"13 2 0"22 -0.065 0.193 -0.386 1.782 0"23 2 0"33 0.564 0.671 0.695 0.309 0.933 0.161 0.252 2.948 0.233 0.380 1.067 Town size Education Constant 0.179 0.031 Variance-covariance estimates 2 0" II 0"12 Sample vote share Predicted vote share [72 ] D66 0.104 PvdA 0.369 1.107 VVD 0.189 CDA 0.338 D66 0.108 0.030 0.061 0.161 PvdA 0.347 0.306 VVD 0.195 0.026 0.028 0.149 0.389 0.246 CDA 0.349 0.030 0.301 OA01 291 Table G-3. Estimation results for the Joint Model, Germany Posterior Parameter Party Religion Income Town size Education 95% CI Lower Upper 0.295 0.079 0.154 0.466 SPD 0.358 0.275 -1.174 0.910 CDU -0.269 0.467 -1.251 0.606 Spatial distance Manuallabor SD mean SPD -0.341 0.467 -0.641 -0.015 CDU 0.946 0.393 0.312 1.827 SPD -0.023 0.028 -0.078 0.030 CDU -0.056 0.047 -0.154 0.033 SPD 0.096 0.127 -0.154 0.347 CDU -0.594 0.247 -1.146 -0.165 SPD -0.068 0.043 -0.015 0.015 CDU -0.117 0.082 -0.286 0.037 1.843 0.474 0.978 2.835 1.000 0.000 1.000 1.000 Constant Variance-covariance estimates 0"2\\ 0"\2 0.035 0.379 -0.584 0.670 0"222 7.292 4.106 2.442 14.936 0.026 0.026 0.111 Sample vote share Predicted vote share FDP 0.059 SPD 0.429 CDU 0.512 FDP 0.064 SPD 0.426 0.033 0.368 0.481 CDU 0.510 0.029 0.462 0.558 [73] 292 sincere, then coalition policy outcomes will be a lottery across the "heart" ABC. Let U(ABC) be the von Neumann Morgenstern utility of party C for this lottery, where U is derived from the underlying Euclidean preferences of the party. Consider a possible declaration by party C of the position C', chosen such that the triangles {C' ,C,B}, {C' ,C,A} and {A,B,C} are all equivalent. By symmetry each of these lotteries occurs with probability, ~, so the von Neumann utility of party C for the lottery across ABC' can be represented as U(ABC') = ~U(ABC) + ~U(C'B) + ~U(C'A). C' B Figure 9. Strategic party choice By symmetry again U(ABC)=U(C'CB) = U(C'CA), so U(ABC') = U(ABC). If the lotteries are continuous in the declarations, and U is continuous then there exists a point C" on the arc [C,C'] which maximizes U. This point C" is the best response of party C to the declarations, A and B. Similar arguments show existence of a Nash equilibrium, namely mutual best responses of j}(' by A to B", e", etc. Obviously, the best response C" by e depends on the assumptions made on the nature of the lottery. Schofield and Parks (1996) have made specific assumptions on the nature of this lottery, which permit computation of best response. To illustrate suppose after the declarations {A,B,C}, that coalition {A,B} chooses as a policy compromise the mid-point !(A + B) = D, etc., and that coalitions occur with probability inversely proportional to the distance between coalition members. Thus, when party C declares position C, then coalition {C,B} chooses the midpoint E = ! (B+C), and when party C chooses C', then the same coalition {C,B} chooses E' = !(C' +B). It is evident from Figure 9 that the best response by C to {A,B} is a point C", such [74 ] 293 that the midpoint E" = ~(C" +B) minimizes the distance between C and the arc EE'Thus best response by C is to adopt a more extreme position than its ideal point. Under the above assumptions on the coalition lottery, Schofield and Parks show that with the given configuration of ideal points, parties A and B will converge to one another in Nash equilibrium, while C will "diverge." If non-policy portfolios are added to the calculations, the extent of convergence by A and B will increase, while C's divergence will decrease. To extend this model to the case of heterogeneous preferences within party C, for example, note that any delegate of the party with an ideal point in a neighborhood of C will have a best response that is further from the line AB than is their ideal point. We suggest that such a model of best response accounts for our empirical inference that the VVD adopted a more extreme strategic position than its ideal, or internal core, point. Computation of best response and of Nash equilibrium is highly nonlinear and can only be analytically computed in simple symmetric situations such as described by Figure 9. In general, asymmetries in the configuration of the parties' sincere choices will become even more exaggerated when the parties adopt best responses to each others' positions. These inferences appear to be rohust with respect to the specific assumptions made about the coalition lotteries. [75] Public Choice 97: 295-322, 1998. © 1998 Kluwer Academic Publishers. 295 Issue competition in the 1993 Norwegian national election * STUART ELAINE MACDONALD 1, GEORGE RABINOWITZ 1 & OLA LISTHAUG2 1Department of Political Science, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599-3265, U.S.A.; 2 Department of Sociology and Political Science, Norwegian University of Science and Technology, N-7055 Dragvoll, Norway Abstract. We investigate the role of issues in the 1993 Norwegian election. We are interested in comparing two spatial models of issue evaluation, the directional model and the familiar proximity model. The directional model implies that voters ask two questions of parties: Are you on my side? and Can I trust you to be responsible? This contrasts with the classic proximity question: How close are your positions to mine? Prior analysis of Norwegian voters has favored the directional model. The empirical story in 1993, however, features a traditional centrist party, the agrarian Center Party, running quite strongly, which on the surface, at least, challenges the directional model, and presents an interesting case to observe. We also extend our analysis to examine more generally the impact of issues on the election. This unrestricted analysis adds texture to our understanding of the role of issues, while its results dovetail with the analysis of the specific models. When people evaluated parties on the basis of issues in Norway in 1993, the directional model describes that dynamic well. 1. Introduction In democracies, voters often use issues to make judgments about political parties. In the traditional proximity world, voters ask of parties: "How close are your positions to mine?" (e.g., Downs, 1957; Davis, Hinich, and Ordeshook, 1970; Enelow and Hinich, 1984; Hinich and Munger, 1994).1 In the directional world, voters ask: "Are you on the same side of the issues I am, and can I trust you to be responsible?" (Rabinowitz and Macdonald, 1989; Macdonald, Listhaug, and Rabinowitz, 1991; Macdonald and Rabinowitz, 1993). The directional model argues that voters only have preferences for policy direction, and that parties stimulate support by taking strong stands in * We are grateful to Roar Haskjold for research assistance, and to the Norwegian University of Science and Technology and the Arts and Sciences Foundation of the University of North Carolina at Chapel Hill for financial support. The 1993 Norwegian National Election Study, directed by Bernt Aardal and Henry Valen, was made available by the Norwegian Social Science Data Services. Neither the principal investigators nor the archive bear any responsibility for our interpretation of the data. [77] 296 the direction voters favor. Parties must develop intensity in order to generate issue-based support, but are constrained in their intensity by the need to be viewed as responsible. Like the proximity model, the directional model is both rational and spatial. As will become clearer in the discussion of the theories below, the key difference between the models is the lack of precision in voter preference in directional theory. This leads issues to be dichotomous rather than multipositional, and voters to assess pragmatic capability as well as specific issue agreement from party issue position. What types of party systems are expected to emerge in multiparty systems? The directional model makes the specific prediction that parties located in the center of the directional space - a position which lacks intensity should be electorally weak (Macdonald, Listhaug, and Rabinowitz, 1991). In general in directional theory the strongest positions should be noncentrist and nonextreme. It is more difficult to answer this question for the proximity model. Based on the work of Cox (1990), we can generally assume that parties will tend to be widely dispersed in the political space in a multiparty system. Beyond that, the traditional spatial model has little to say about party strategy in such systems. In Norway seven parties seriously compete for seats in the Storting, the Norwegian Parliament. The two largest vote-getting parties are Labor, a traditional social democratic party, and the Conservatives, the traditional party of the moderate right. On Labor's left is the Socialist Party, and on the Conservatives' right is the Progressive Party, a Norwegian version of a party of the reactionary right. There are three parties that traditionally fall in the center of the left-right spectrum, the Center Party, the Christian People's Party, and the Liberal Party. The Center Party normally receives the bulk of its support from farmers and those sympathetic to farm interests. The Christian People's Party represents the moral perspective of the Church, and draws support from the more religious in the society. The Liberals have traditionally been the one true center party in Norway. They are somewhat like the more familiar Liberals (now Liberal Democrats) in Britain, in that they were originally the main opposition party to the Conservatives only to be outflanked by Labor when the working class became politically active. The Liberals were the only one of the seven major parties to fall below the 5% threshold in 1989; hence they had no representatives in the Storting prior to the 1993 election. In terms of governmental control, after the 1989 election, the Center, Christian, and Conservative parties formed a minority government, which collapsed in the fall of 1990 on the European Union issue. From 1990 to 1993 the Labor Party controlled the government, although with a minority of seats. [78] 297 LEFT i i -4 -3 • I -2 RIGHT i -1 i i 2 3 0 Intensity Neutral Center I 4 . Intensity Figure 1. Political ideology. Traditionally Labor has refused to allow other parties to share government with them, a strategy that has been quite effective. 2 In prior work (Macdonald, Listhaug, and Rabinowitz, 1991), we have studied the Norwegian system in the 1989 election, and found a politics that looked strikingly directional in character. The 1993 election presents a second look at Norway, a look which might hold some surprises. The largest single change in support for any party from 1989 to 1993 was a strong surge in support for the Center Party. Could this be an instance in which the center has become a lively source of support? We shall see. Before turning to the analysis of the role of issues, it is useful to review the theories and observe how we would expect evaluation to depend on issue position under the two models. 2. An overview of directional theory For illustrative purposes, Figure 1 shows a directional continuum of political ideology. The selection of ideology is simply for convenience - directional theory is oriented toward policy issues, but ideology provides a familiar referent and facilitates the use of the terms "left" and "right" to designate the opposing sides. 3 In the figure notice that the continuum is clearly separated into two sides with a neutral center point. An individual is on one side or the other of an issue, or neutral. Position in directional theory reflects the intensity with which a voter supports her side on the issue. The neutral point is the position of zero intensity where the voter favors neither side. In the figure, the center point is given a score of zero. Negative values signify support for the left and positive values support for the right. The designation of signs is arbitrary, so we follow the normal convention that values go from negative to positive as they move from left to right. The values increase in absolute magnitude to indicate increasing intensity. Parties as well as individuals can be placed on the directional continuum. Parties are positioned more distant from the neutral center as a consequence of how strongly they advocate positions of the left or the right. Parties that [79] 298 are consistent and vocal advocates of left policies would be positioned well to the left, those that are more equivocal yet generally favorable toward the left would be located on the moderate left, and so on across the spectrum. Of course, party position is not solely a function of the party's own efforts; it can be influenced by opposition strategy as well as by the media and ongoing events. Notice on the continuum in Figure 1 a short vertical line near each end of the scale. These lines delimit an area called the "region of acceptability." In directional theory there is the explicit idea that parties may be perceived as responsible and trustworthy in government or as irresponsible and not to be trusted. This perception is tied to the issue positions they advocate. Parties that are outside the region of acceptability are penalized by at least some voters. The placement of the lines on the ideology continuum is a bit artificial because being outside the region of acceptability is a function of the entire set of positions a party advocates. Nevertheless, it serves to illustrate the idea that parties with consistently extreme positions are likely to lose support. In directional theory, voters are attracted to parties by clear cues for their preferred side. Hence, parties generate support from voters on their side by being fairly intense; they need only temper their appeal enough to avoid being perceived as extremist. Voters who are intense are more receptive to party cues. They are more enthusiastic about cues for "their side" and more opposed to cues for "the other side." Voters in the center are neutral, and thus will support all parties within the region of acceptability about the same. Similarly, center parties provide no cues to either side, so they tend to be evaluated the same by people across the issue spectrum. These ideas are reflected in the form of the utility or affect function for a party. In directional theory, affect is a monotonic function of Aij, where (1) where Iik is the position of voter i on the directional continuum for issue k, Ijk is the position of party j on issue k, Sk is the general importance of the issue to the election, and P ij = 0 when bkITk :::: r2; Pij > 0 when bkIJk > r2 where r defines the maximum intensity level of a party before it is penalized. Notice that the function captures the essential ideas of the theory. When the party and voter are on the same side, the product IikIjk will be positive; if the party and voter are on opposite sides, the product will be negative. The more intense either the voter or the party, the higher the absolute value of the product will be. The affect function implies than a party will draw more [80 ] 299 favorable ratings from voters on its side when it is more intense, at least until it reaches the boundary of the region of acceptability. Thus, within the region, utility is dependent on the scalar product between the voter and party vectors, CEkIikIjk). Issues are allowed to have differential salience, but it is a general rather than personal salience. 4 And, finally, notice that a party at the neutral center will be evaluated the same by voters anywhere because the party's scale value Ijk will be zero. 3. Comparing the theories We have said less about the traditional proximity model because it is more familiar. In the model each voter is assumed to have a specific ideal point that represents the policy she most prefers. And each party offers a set of policy alternatives that locate the party at a particular point in a multidimensional issue space. According to the model a voter will evaluate most favorably the party whose position is closest. Therefore, utility declines with Euclidean distance (or, equivalently, distance squared). Thus, utility is a monotone function of A ij , where (2) where 8ik is the position of voter i on the proximity continuum for issue k, 8 jk is the position of party j on issue k, and Sk is the general importance of the issue to the election. Now let us see how parties would be evaluated under the two models as a function of issue position. Figure 2 illustrates the predictions for four parties: a center party (C), a moderate right party (MR), a moderate left party (ML), and a far right party (FR).5 The most marked contrast in prediction is for the center party (C). According to directional theory, this party offers no cues about which side of the issue it supports; thus it will activate no particular evaluation based on the issue. If this issue were the sole criterion of judgment, voters anywhere on the spectrum would feel the same way about the party. Therefore, the directional curve is flat. Under the proximity model, the party at the center matches precisely the desires of center voters and they respond with high evaluations for the center party. Moving away from the center in either direction, voter affinity for the policies of the party decreases and thus their evaluation. The resulting relationship is clearly peaked. Next we consider a party at a moderate right location (MR) on the scale. According to directional theory, this party will be liked by voters on the right and disliked by voters on the left, while voters in the center will be neutral. [81] 300 Proximity Illustration Directional Illustration 100 100 E v a " FR 80 80 MR I U a ~-c I i o ,, 60 60 MR ,, C 40 40 n " 20 "" ML 20 ML 0 ~ 4 ~ 4 ~ 0 1 Left-RighI 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5 Left-RighI Figure 2. Illustrative sketches comparing the two theories. Notice that the highest evaluation comes from voters farthest to the right. These voters are the most enthusiastic for the right side of the issue and thus endorse the party most fully, This is different from the proximity model. The voters who will be the strongest supporters for the moderate right party in the proximity model are the voters who share its particular policy mix. Thus, the highest evaluation will come from voters themselves at a moderate right location. Voters at the far right will not like the party as well because its policies do not match their ideal point. Again the support curve is peaked; the peak has just shifted to the right. Peak support in proximity theory is always at the party location. The directional curve for the center party is flat while the curves for the moderate right and moderate left parties have a clear slope. In general, in directional theory the more extreme the party, the steeper the slope will be. In the figure the curves for the center and moderate parties intersect at the point of neutrality. In directional theory, the support curves of all parties within the region of acceptability cross at the neutral point. This is important for it implies that voters on the right side of the neutral point - even those who are closer to the center - will like the right party more than the center party, and voters on the left will like the left party more. Thus, center parties are clearly disadvantaged in directional theory; they lose voters on the right to moderate right parties and voters on the left to moderate left parties. Center parties are not even advantaged among centrist voters; they are simply on an even footing with the other parties among these voters who care little about the issue. [82] 301 The last curve in Figure 2 is drawn for a far right party (FR). This party is hypothetically beyond the region of acceptability in directional theory and hence has a lower utility for most voters. The proximity curve for the party peaks at the far right position. Like the directional curve, the proximity curve is virtually monotonic in this case, reflecting the fact that the theories make similar predictions for extreme parties. 4. The 1993 Norwegian election In order to test the theories we need several pieces of information. Most critically we need to know where the voters stand on a set of issues relevant to the campaign, where the parties stand on those same issues, and how the voters evaluate the parties. We also need demographic information about the voters, so that the analyses will not attribute to issues, effects that are a function of social circumstance. All this information is available in the 1993 Norwegian Election Study. The hot new issue in the 1993 election centered on whether or not Norway should join the EU. It was not the first time the issue had come before Norwegian voters. There had been a divisive referendum in 1972. In the referendum the Labor Party had supported joining the Common Market, and had paid a price for that in the '73 election. There was good reason to think that history would repeat itself. Once again the leadership of the Labor Party was supporting membership and a substantial block of their supporters were opposed. But this is getting a bit ahead of the data-analytic story. The 1993 election survey contained five issue questions on which respondents were asked to place themselves and the seven parties. Each of these was a lO-point scale with the endpoints labeled with extreme positions. The questions dealt with left-right position, environmental protection, immigration policy, unemployment policy, and EU membership. The full wording of the questions appears in the Appendix. There was reason to think that all of these issues would be important in the election. Left-right is consistently consequential in voter choice, and immigration, environmental protection, and unemployment policy remained controversial. The national referendum on EU was just a year away. The mean positions of the parties and the voters are displayed in Table 1. The positions are on a centered scale which ranges from -4.5 to +4.5, rather than the original 1 to lO. When appropriate, the scales have been reflected so that negative values represent positions on the political left and positive values positions on the right. The parties are arranged in what is generally thought of as their left to right order. This matched their public perception as measured by the surveys in 1989 but not in 1993. Labor was placed to [83 ] 302 the right of the Liberal Party, although remaining distinctly on the left side of the scale. The three center parties (Liberal, Center, and Christian People's) all made a distinct leftward shift in 1993, while both Labor and the Conservatives moved rightward. The population mean also took a clear shift in the left direction. While all the changes were modest (within a half unit on the lO-point scale), there was quite a lot of movement. Perhaps it was not coincidental that every change was consistent with party position on the EU scale. The Center, Liberal, and Christian parties were all on the anti-EU side, which is generally seen as the left side of the issue in Norway. Both Labor and the Conservative Party were on the pro-EU side. The respondent mean position was distinctly anti-EU. Notably on unemployment policy, the specific issue that conforms most closely to traditional left-right distinctions, the parties order in the more traditional way with Labor well to the left of the Liberals. 5. Analytic models We will estimate three models of candidate evaluation, one based on the proximity model, one based on the directional model, and one based on a model that we have elsewhere called the mixed model (Rabinowitz and Macdonald, 1989). We will discuss the mixed model shortly. As a first step in the analysis, we set the issue scales so that the middle value is zero. This is consistent with the way we presented the party positions, and operationally fixes the directional neutral point at the middle value of the scale. In describing the regression equations we will use the notation Xik and Xjk to refer to the position of voter i on issue k and party j on issue k, respectively. This common notation across the models emphasizes that all estimation is made on exactly the same data. The general structure of the estimation for the proximity and directional models is as follows: 5.1. Proximity model Affectij = interceptj + I;mbjm demographYim + I;kbklxik - xjkl + errOrij where Affectij is the evaluation of party j by individual i, interceptj is the intercept for party j, bjm is the effect of demographic variable m for evaluating party j, demographYim is the status of individual i on demographic variable m, bk is the effect of issue k in the election, Xik is the position of individual i on issue k, [84] (3) Table I. Mean position of respondents and mean perceived positions of parties on political issues in Norway in 1993 (SD in parentheses) Issue Left-right -.23 Environment 0.92) -.56 (2.36) Immigration European Union Unemployment 00 Ul Respondent 1.70 (2.36) -1.34 (3.12) .18 (2.78) Socialist Labor Liberal Center Christian Conserv Progress -3.12 (1.26) -2.12 (2.21) -1.24 (1.77) -.50 ( 1.73) -.37 ( 1.60) -1.67 (1.92) -1.92 (2.12) -1.02 (1.61 ) -.54 (1.50) .65 (1.42) -.46 3.43 (1.60) 1.89 (2.32) .10 (1.56) .89 (1.73) 3.72 (1.63) 2.67 (1.94) -1.37 (2.22) -4.10 (1.43) 0.68) -1.08 (1.88) -.63 (2.19) 3.09 ( 1.33) .66 ( 1.77) 3.18 (2.03) 2.17 (2.30) -1.82 (1.87) -.73 -.32 ( 1.99) .09 (1.58) 2.35 (2.11) 3.08 (2.06) -1.90 (1.88) -2.70 (2.29) -2.64 (1.77) ( 1.65) -.90 (2.08) w ow 304 Xjk is the mean perceived position of party j on issue k, and IXik - xjkl is the absolute value of (Xik - Xjk). 5.2. Directional model Affectij = interceptj + Lmbjm demographYim + Lkbk(xik . Xjk) + errOrij (4) where all terms are defined as above. Notice several features of the estimation. The analysis is pooled over all the parties, so the entire political system is analyzed within a common frame. Note, too, that each party is allowed a unique intercept and unique demographic effects. Thus demographic factors and unmeasured general popularity factors are controlled, assuring the model is reasonably specified. A party intercept nullifies the need to add an explicit penalty term, but assumes implicitly that the penalty is a constant across respondents. Finally note that in the proximity model distance is assessed using a city-block metric. In each model we include 12 demographic factors, each as a dummy variable. These are subjective class, union membership, education level, income, sector of employment, farm-related occupation, religiosity, gender, urbanrural, region, language preference, and alcohol use. 6 The dependent variable in each regression is the generalized affect the respondent feels toward the party. This is measured using a 100-point sympathy thermometer. The full wording of the sympathy question appears in the Appendix. 6. Results The issue coefficients and summary statistics from the regression analyses are reported in Table 2. Intercepts and demographic effects were estimated but are omitted in order to make the table more readable. Because each regression is based on an identical set of cases and explains the same dependent variable, the adjusted R-square values provide an appropriate summary of the relative efficacy of the models in explaining party evaluation. The demographic analysis, shown at the top of the table, includes only the party intercepts and the demographic variables. It is presented to give a sense of how well non-issue factors explain evaluation in Norway in 1993. Of greatest interest is the report of the proximity and directional analyses. In both models left-right has the strongest impact, but after that the two models diverge slightly with the EU issue more important than immigration in the proximity specification, but immigration more important in the directional specification. These three issues dominate in both sets of estimates. [86] 305 Table 2. Comparison of models explaining party evaluation: Pooled regression analysis Demographic model Adjusted R-square Model standard error .243 21.502 Proximity model Adjusted R-square .370 Model standard error 19.617 Issue Coefficient Std Error t-ratio Prob Distance -4.384 .126 -34.710 .000 Environment .004 .115 .042 Immigration -1.130 .106 -10.613 .966 .000 European Union -1.969 .083 -23.620 .000 Unemployment -.211 .100 -2.113 .034 Left-right Directional model Adjusted R-square Model standard error 0405 19.074 Issue Coefficient Std Error t-ratio Prob Scalar product Left-right Environment 1.593 .162 Immigration .707 European Union Unemployment .622 .147 N of cases .044 .054 .046 .022 .035 35.512 2.959 15.305 .000 .003 .000 27.387 4.170 .000 .000 12,866 As for the other two issues, environment is estimated to have absolutely no consequence in the proximity specification. However, environment is of clear significance in the directional specification, with a slightly larger effect than unemployment. For each issue, the ratio of coefficient to standard error is higher in the directional than the proximity analysis. This is consistent with the overall better fit of the directional model, which improves significantly on the proximity model. [87 ] 306 To shed more light on the comparative performance of the models, it is useful to consider a model which allows a natural nesting of the theories. 7. Mixed model analysis The mixed model is based on the idea that squared Euclidean distance is readily decomposed into two components, one being the scalar product and the other the length of the party and individual voter vectors. 7 These components tend to be only weakly correlated, providing a particularly clean test. Using the notation from the regression equations, -bk(Xik - Xjk)2 = [2bk(Xik . Xjk)] - [bk(Xik)2 + bk(Xjk)2] (5) where the left-hand term is squared Euclidean distance and the right-hand terms are respectively, twice the scalar product and the sum of the two vector lengths. The model is operationalized as follows: 7.1. Mixed model Affectij interceptj + bmbjm demographYim + bkbsPk(xik . Xjk) + bkblenk[(xrk + x;k)/2] + errOrij (6) where all terms other than bSPk and blenk are defined earlier, bSPk is the effect of the scalar product term associated with issue k, blenk is the effect of the length term associated with issue k. Notice that we preserve the correct 2 to 1 ratio of scalar product to length by halving the lengths. If the analysis conforms to a squared Euclidean metric, then the ratio of bSPk to b1enk will be -1. As the ratio gets higher in absolute value, the results are increasingly directional. A positive length coefficient would entirely violate the proximity model and a negative scalar product term would violate both models. In addition to its use as a statistical device for comparing the theories, Rabinowitz and Macdonald (1989) presented the mixed model as a behavioral model which, they showed, has some of the same implications as directional theory without the region of acceptability concept. Both Iversen (1994) and Merrill (1995) have argued for the mixed model as a generic description of how voters use issues to evaluate parties. 8 In addition, Grofman (1985) has constructed a model in which voters discount party position based on the likelihood of the party's enacting specific policies, which also implies [88] 307 Table 3. Mixed model explaining party evaluation: Pooled regression analysis Mixed model Adjusted R-square .405 Model standard error 19.072 Issue Coefficient Std Error t-ratio Prob Scalar product 1.593 .044 35.525 .000 Environment .154 .055 2.788 .005 Immigration .711 .046 15.362 .000 European Union .622 .022 27.411 .000 Unemployment .147 .035 4.170 .000 .196 Left-right Length -.094 .072 -1.292 Environment .024 .051 .477 .633 Immigration -.084 .046 -1.813 .069 European Union -.033 .043 -.772 .440 .029 .046 .628 .530 Left-right Unemployment N of cases 12,866 a mixed model. Hence, if it offers more than the directional or proximity model in terms of understanding voter behavior, the mixed model could be theoretically consequential. The mixed model analysis for the parties is presented in Table 3. Quite strikingly the mixed model explains virtually no additional variance over and above the directional model. For the five issues, all of the scalar product coefficients are significant, and virtually mirror those of the pure directional analysis. None of the length components has a significant effect, a truly remarkable result given the large number of cases in the pooled regression. With no significant length coefficients, it is not even meaningful to compute the ratios. There is no evidence for either a proximity or a mixed calculus for Norwegian voters in 1993. Directional theory clearly provides a better accounting of issue effects than either of the alternative theories. This corresponds well with the conclusion we reported based on the 1989 election. [89] 308 8. Unrestricted analysis Critical to any rational choice model is the idea that voters use consistent criteria in relating their own issue positions to party preference. We imposed that constraint by the way we operationalized the regression models. Yet in reality nothing forces voters to use a consistent calculus when evaluating parties. Nor is there any guarantee that the models under consideration are adequate for capturing the process voters use. Hence, it is useful and prudent to approach the issue-evaluation linkage in as unconstrained a fashion as possible. 9 To do this, we will estimate separately for each party and each issue two parameters, one representing a linear effect and the other a squared effect of the issue. The following model is evaluated for each party: Affectij = interceptj + ~mbj + ~kbljXi demographYim + Lkb2jkXfk + errOrij (7) where b1jk is the effect of issue k for evaluating party j, b2jk is the effect of issue k squared for evaluating party j, and all other terms are defined earlier. The unrestricted analysis will perfectly match any pattern that can be fit by a parabolic shape (including straight lines), and any single-peaked function should be matched reasonably well. No theory of issue effects that we are aware of implies mUltiple peaks nor is there any empirical evidence for systematic multimodal evaluation patterns. Hence the method should be sufficiently general to approximate most evaluation patterns, and sufficiently parsimonious to provide good insight into the impact of issues. Operationally we will start with each issue centered at the 0 point. This duplicates the centering we did in the prior analyses, but here it has a different purpose - to facilitate the interpretation of the results and to avoid potential collinearity problems. When the variables are centered, the coefficient for the linear term shows the basic direction of the relationship. A positive value means the party receives a higher evaluation from those on the right; a negative value implies those on the left evaluate the party more highly. Centering reduces collinearity because values in the middle of the simple linear scale form one end of the squared issue scale while values from the two extremes of the linear scale form the other end of the squared scale. Thus the methodology allows for approximating any single-peaked pattern and assures there is little correlation between the two variables representing respondent issue position. The coefficients for each issue will give a sense of what the curve looks like, and allow us to relate it to the theories. Proximity theory predicts that the peak evaluation for a party should occur among respondents positioned at the [90] 309 same location as the party, with support declining on either side. This implies a significant negative coefficient for the squared issue variable to insure the curve is concave with a definable maximum. lo Directional theory implies that in general the estimated coefficient for the variable in squared form should be small. A positive coefficient for the squared term is usually not a problem for directional theory - it simply adjusts the shape of the monotonic curve. Now let us use this approach to consider the interplay between issues and evaluation in Norway. 9. Results of the unrestricted analysis An unrestricted analysis was run for each of the parties using the five issues and the demographic controls. Table 4 shows the results. Coefficients for both the linear issue scale and the squared issue scale are displayed in the table, along with the mean party positions. When the coefficient for the squared term is negative and the estimated peak of the curve is within the -4.5 to +4.5 range of the issue scale, the peak position is also shown in the table. The demographic effects are not shown to save space. A good starting point for exploring the issue impacts is the left-right ideology of the parties. To facilitate the interpretation of the various coefficients, a curve reflecting the impact of left-right on evaluation is drawn for each party in Figure 3. The intercepts for the curves reflect the estimated evaluation of the party by a person at the neutral center on all five issues. The intercepts give a flavor of the general popUlarity of the parties. II The three parties that have the most dramatic left-right effects are the Conservatives, the Socialists, and the Progressives. Directional theory predicts that slope should be a monotonic function of party location, and these are the three parties with the most extreme locations on the left-right scale. After these three, the next steepest slope is for the Labor Party. Recall that Labor was perceived to be nearer the center than the Liberals in 1993, so the relative slopes of these two parties run against the prediction of directional theory. It could be that the emotional attraction of Labor to the left remains high even when the perceived position switches. In any event, the relatively steep slope for Labor compared to the Liberals is the only aspect of the left-right results that does not conform well to the expectations of directional theory. The evaluation lines show relatively little curvature. The only party with a peak on the scale is the Center Party, but the impact of left-right for this party is modest, as reflected in the barely significant coefficient for the linear term and the insignificant coefficient for the squared term. The most curvature occurs for the Conservative and Progressive parties, and for both of those [91 ] 310 Table 4. Issue effects from an unrestricted analysis of party evaluation in Norway in 1993 Issue Socialist Labor Liberal Center Christian Conserv Progress Left-right Linear Squared Peak Party -5.42** .04 none -3.12 -3.44** -1.19** -.15 -.11 -.82* 1.41 ** -.23 -.14 off scale off scale -1.79 -1.24 -1.67 off scale -.54 .65 6.()()** 5.21** .28* .43** none none 3.09 3.43 Environment Linear Squared Peak Party -.62* .02 none -2.12 -.73** -.28 -.22 -.02 -.19 .05 .81** .10 .04 .01 -.06 -.07 none -.50 none none none -.14 -1.37 .66 1.89 .95** 2.71** -1.92 -.90 -.46 Immigration Linear -.72* 1.14** -.92** .42 Squared -.12 -.20 -.26** -.15 Peak Party .51 -.35** -.20* .18 none 3.72 -2.91 2.92 -1.80 1.41 .73 2.31 -1.90 -.37 -1.02 .10 -1.08 .89 -1.25** 1.74** -.57** -3.58** -.71** 1.29** -.12 -.11 -.04 -.05 European Union Linear Squared Peak Party -.00 off scale off scale -2.69 -2.70 2.67 -1.37 .12 none -4.10 .32 -.15 off scale off scale -.63 3.18 1.03 2.17 -.14 Unemployment Linear Squared -.96** -.09 .17 -.73** .09 .58** .00 .09 -.07 .07 .04 .08 none none none none none -.55* Peak off scale off scale Party -2.64 -1.82 -.73 -.32 .09 2.35 3.08 Constant 43.925 63.029 49.236 39.064 43.056 56.794 30.718 Adj R-square .424 .240 .139 .420 .185 .468 .339 Model std err 18.577 18.052 17.582 19.078 18.780 16.822 21.031 Note. Analyses included the full set of demographic controls. N of cases in each analysis is 1800. **Significant at the .001 level; *Significant at the .01 level. [92] 311 10~-c 80 c .2 ~ 60 iii :l - . --- . -. -" - -. -_a~ .-:~ W 40 . --- . -- enl oc 20 o+-__________-.__________-,.-__________.-__________ -4.50 -2.25 .00 2.25 ~ 4.50 Left-Right Position Figure 3. Party evaluation by left-right position. parties the curve picks up an accelerating trend on the right side of the scale, reflecting their rapidly increasing popularity among those on the far right. Of some interest is the support curve for the Christian People's Party modestly sloped to the right with a peak off the scale. Two things make the curve of interest. The Christian People's Party is located within one unit of the center of the scale, just slightly right of center. And it is the party with the third sharpest image - the Socialists and the Conservatives were more consistently placed, but it was solidly third. Yet, support for the party shows nothing that resembles a clear peak in the middle of the scale as predicted by proximity theory. This type of flat, slightly rightward leaning slope is a direct prediction of directional theory, and is consistent with the view that matching issue positions is not the key link in the evaluation process. At least, not in Norway in 1993 with regard to left-right. Rather, being weakly to one side seems to draw weak support from those on that side. One other feature of the support curves merits attention. The general pattern of party support fits closely the theoretical pattern shown for the directional model, if we assume that the Progressives were penalized for their extreme positions. The Progressives are by far the furthest from the center of any party in Norway across the full set of issues, so that assumption seems [93] 312 quite credible. There is, however, an exception. While support for most of the parties crisscrosses around roughly the same value, Labor generally sits above the other parties. Unlike the penalty for the Progressives, there is nothing in directional theory that predicts the generally high evaluations that Labor receives. But let us consider the remaining issues before we return to confront the general pattern of results. The most interesting issue in 1993 was the European Union question. The support curves for EU are shown in Figure 4 while the coefficients appear in Table 4. For two parties support peaks on the issue scale - for the Liberals and the Progressives. In neither case is the squared coefficient significant, and in the case of the Progressives the linear coefficient is not significant either. As we might have anticipated, neither party's evaluation is based much on its EU position. The same can be said of the Christian People's Party for which the curve has a very modest slope. Far and away the most potent effect is for the Center Party. The EU issue is more than twice as consequential for the Center Party as it is for Labor, where the issue has its second greatest impact. Unsurprisingly, the Center Party took the most extreme stand on the issue; and its strong opposition to EU membership was clearly established in public perceptions. Tn this regard it is interesting to contrast the relatively weak effect of the EU for the Socialists compared to the Center Party. While the Socialists were strongly anti-Eu, they did not come close to the Center Party in either the extremeness of their position or the clarity with which they projected their views. The Labor and Conservative parties anchored the pro-EU side of the scale. While the Conservatives were seen as somewhat more pro-EU than Labor, the issue had a greater impact on the evaluations of the Labor Party. Perhaps this should not be surprising. In the year prior to the election the Labor government had agreed to participate in the European Economic Area, and was clearly going to actively support joining the EU in the upcoming 1994 referendum. As the governing party, Labor's EU position had particular salience. In addition, the Conservatives pursued an intentionally lOW-key strategy. They recognized that to win the referendum, the pro-EU forces would need to obtain support on the social democratic left. To do this, it was important to downplay Conservative rhetoric as much as possible.1 2 Overall, the EU issue is similar to the left-right issue in showing evaluation patterns that are direction-like, even if they are not perfectly directional. The patterns tend to be either uniformly increasing or decreasing through the range of the scale, and the extremity of the party stand strongly influences the importance of the issue in judging the party. The only issue on which the majority of the parties appear to be judged on a more proximity-like basis is the immigration issue. Indeed, six of the seven [94 ] 313 10~-' 80 __- - - - i C60~ ~ =.---0-- ~> ------ ------ _._.-.-.-.';:::.._ ....::-_ ----------- --- ab ens UJ 40 _---------------------------------------------------Pro9 20 0+------------.------------.-----------.------------4 -4.50 -2.25 .00 2.25 4.50 EU Position Figure 4. Party evaluation by EU position. parties are estimated to have a peak evaluation on the immigration scale. The Progressive Party is the only exception. The immigration issue is a complex one for parties in Norway. Public sentiment clearly runs in favor of making it more difficult to immigrate into Norway. Yet showing out and out hostility to immigrants risks putting the party in a bad light. The general response has been for the Progressives to use the issue to generate support and for the rest of the parties to hedge. This is reflected for the most part in the support curves, which are shown in Figure 5. Noticeably, even when support peaks on the scale, there is little to suggest very favorable evaluations from respondents located at the same position as the party. The sharpest curvature is for the Christian People's Party, but even there the predicted range in evaluation is modest, and the curve peaks right rather than left of the center. Once again there is a surprise in the figure. The Labor Party. which is perceived to be slightly on the pro-immigrant side of the issue, has the second strongest effect, and the effect is to produce generally rising support from those who are anti-immigrant. This runs contrary to theoretical expectations given that Labor is perceived slightly to the left of center. Neither directional nor proximity theory, nor indeed any theory with which we are familiar, predicts rising support for a party from those who are opposed to its position. As [95 ] 314 100r--------------------------------------------------, 80 ~_-tab c: .2 60~ 1ii ::> (ij > w --20 O+-r,.~ -4.50 -2.25 .00 2.25 4.50 Immigration Position Figure 5. Party evaluation by immigration position. with most surprises, there are certainly post-hoc stories to tell. Perhaps the most compelling is that the actual policy of the Labor government has been to toughen immigration standards even as party leaders espouse generally pro-immigrant sympathies. In any event, for an issue which ostensibly shows lots of curvilinearity, the impact of immigration for the parties with the curvilinearity is quite weak, and not sharply peaked. The Progressive and Labor parties are the two for which the issue has the greatest impact on evaluation. For the Progressives the results follow directional expectations quite nicely; for Labor they require post-hoc explanation. The effects for environment and unemployment are weaker than for the other three issues. We show the support curves in Figures 6 and 7, but forego a full discussion. Perhaps of some interest is the fact that Labor's curve for the environment, as with immigration, slopes in the "wrong" direction. This anomaly is also consistent with Labor's policy making. The Labor government has been more inclined to trade off environmental protection against living standards than their pro-environment rhetoric would suggest. Where does this leave us? In general, the unrestricted analysis is consonant with directional theory. For the most part, issue effects tend to be [96] 315 10~-, 80 L_-----i Bb 60 F:-,_~"';7·.? ~oc 40 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Prog 20 o+-__________. -__________~-_. -4.50 -2.25 .00 2.25 4.50 Environment Position Figure 6. Party evaluation by environment position. monotonically increasing or decreasing, or else weak; and when a party draws differential evaluations based on an issue, it is usually associated with a strong stand. While this is impressionistic based on the figures we have examined, the observation has some analytic support as well. We replicated the pooled analysis using the unrestricted methodology.13 The analysis provides a summary measure of fit which is interesting to contrast with those of the earlier pooled analyses. The unrestricted pooled analysis has an adjusted R-square of .429 - greater than the .405 estimated for both the pure directional model and the mixed model, but only a modest improvement given the addition of 60 parameters to obtain unique issue estimates for each party. In sum, after running the unrestricted analysis, a reasonable conclusion is that the directional model fits the data well, if not perfectly. The most intriguing finding from the unrestricted analysis is the evaluation pattern for the Labor Party. There are two features of the result for Labor that are noteworthy. First, the party has a higher base evaluation than the other parties. This is not predicted by either of the issue theories, and is almost surely not due to its issue strategy. Issues simply do not explain why Labor is as generally popular as it is. [97 ] 316 10~-, 80 60 r-----------------Ji8b ... ;;; .;;. ........ .. . .... .:. =~enl 40 r>oc - - - - - - - - - - - - _____________________________________ - - -frog 20 0+-~_r1 -4.50 -2.25 .00 2.25 4.50 Unemployment Position Figure 7. Party evaluation by unemployment position. The second feature is more suggestive_ It appears that Labor - at least with regard to some issues - was judged more on what it did in government than on how it was perceived at the time of the election. This finding - really this conjecture, if we are appropriately cautious in our interpretation - dovetails nicely with an on-line processing model. The thrust of the on-line model is that voters maintain a running evaluation counter that is updated as new information appears (Hastie and Park, 1986; Lodge, McGraw, and Stroh, 1989; Lodge and Steenbergen with Brau, 1995). The new information need not be remembered; what is important is its impact on the running counter. For a party actually making policy it does seem reasonable that people's rolling evaluation would be influenced by what the party does in office. This need not be reflected in their long-term memory but would show up in a specific analysis of issue effects such as we have done. While speculative, this would account for the two anomalous findings we observed for Labor as well as its relatively steep EU curve. [98] 317 10. Labor's popularity For political scientists to note that issue positions do not entirely account for a party's popularity is hardly novel. Issues have always been seen by empirically-oriented researchers as just one of several factors that determine a person's feelings toward a party. The standard triumvirate of factors that came out of the early Michigan studies were issues, candidates, and party identification (Campbell, Converse, Miller, and Stokes, 1960). To those three factors, one must surely add a retrospective consideration of how effectively a party in power has managed government (see, for example, Stokes, 1966; Fiorina, 1981; Lewis-Beck, 1991). Hence, a base assessment of any party should hinge on the quality of the candidate or candidates the party is presenting, long-term attitudes toward the party, and consideration of the party's immediate past performance, all in addition to whatever influence its issue stands might have. The 1993 election took place in a year with low inflation and general economic growth, despite rising unemployment. If not wildly optimistic about the economic future, Norwegians were becoming more upbeat. With its vast supply of energy resources including exportable oil and natural gas, the country seemed on an unusually strong economic footing compared to its Scandinavian compatriots as well as its other European neighbors. The future was likely to be better than the immediate past. Labor had presided over government and was generally seen to have managed things well. 14 Moreover, the Labor Prime Minister, Gro Harlem Brundtland, was extremely popular. Her mean evaluation exceeded the party's by approximately six and a half rating points, an unusually large gap and a factor that certainly helped her party.15 11. Conclusion Our analysis of the Norwegian 1993 election clearly substantiates the greater explanatory power of directional theory in accounting for the role of issues in elections. The surge of the Center Party in the election was caused by its avid opposition to Norway's joining the European Union. The Center Party was the party of choice for those who wanted to signal their disapproval of EU membership for Norway. What the rising strength of the Center Party most clearly did not signal was a surge in support for moderation and centrist policies. Based on the pooled analyses or the set of unrestricted parabolic regressions, there is simply no evidence that centrism produces support. In all instances, the issue politics has a distinctly directional cast. The analysis also reveals an interesting anomaly. In the two instances in which Labor rhetoric did not match its governing behavior, voters as- [99] 318 cribed to Labor a position based on the rhetoric but appear to have judged it based on the reality. In this regard, the results are consistent with a combination of directionally-based evaluation and on-line cognitive processing. As Lodge, Steenbergen with Brau (1995) argue, on-line evaluation actually provides mass politics a higher degree of rationality than is commonly assumed. People do not simply buy a campaign message; they incorporate a stream of information in their political judgments that is less manipulable and more policy-wise than survey instruments geared to the electoral cycle are able to uncover. By the time of the election most voters have forgotten the information on which they based their judgments. More generally, the real world is politically consequential. When a party comes to power, it experiences pragmatic demands quite different from those of a campaign. Frequently parties that have been dramatically opposed on the campaign trail look more alike if they exchange positions in control of government. The role of expertise and the requirement that people impose on parties to perform well, often force real governments to steer a fairly narrow course. We introduce this idea because there is a mythology about elections which merits discussion. The myth is that electoral politics encourages party systems to converge, while the leadership of the various parties is eager for the parties to diverge. The truth is likely more complex. To the extent that directional theory provides better insight into electoral dynamics than proximity theory - and our work in the U.S., Great Britain, the Netherlands, and Norway suggests that it does - then electoral politics in multiparty systems encourages divergence. Basic to directional theory is the view that politics tends to be conflictual, with party leaders staking out fairly strong positions. Staking out sides and looking for "wedge issues" is natural to competitive politics. Actually running a country is a different matter. The pressures toward convergence often arise out of objective situations that provide little latitude for leaders. The fact that ruling parties will be judged at least in part on how they perform encourages parties to follow expert advice fairly closely. In this regard, the conventional wisdom about the interplay of elections and governance might well be entirely backwards. Electoral politics encourages cleavage and conflict; governing itself might be the source of moderation. Notes 1. While the proximity model has extended its scope to include concern for voter cognition and spatial stability, the basic message of the model is unchanged: voters choose based on a direct match between their desired policies and the policies that parties are likely to produce when in charge of government. [ 100] 319 2. For a discussion of coalition formation in European systems, see Laver and Schofield (1990). 3. The discussion of the theories draws extensively from Listhaug, Macdonald, and Rabinowitz (1994). 4. This departs from the tradition in spatial theory for salience to be individual specific. From an empirical standpoint, personal salience across a broad set of issues has not been effective in predicting candidate or party utility (e.g., Niemi and Bartels, 1985), while there is clear evidence that issues generally do take on differential importance. Further, personal salience cannot be estimated without overparameterizing the model. Hence, we specify the salience component of the model in terms that are operationally meaningful. 5. The center party is located at 0; the moderate left and moderate right parties are located at -2 and +2, respectively; the far right party is located at +4. The curves for both theories are drawn to reflect the idea of monotonicity. 6. Detailed information about the operationalization of the variables is available from the authors on request. 7. Howard Rosenthal first suggested this nested testing strategy. 8. Iversen (I 994) interprets the mixed model as a policy leadership model. Merrill and Grofman (1997) present another variant of the mixed model which makes voter intensity a variable component. 9. The most unconstrained analysis would allow any relationship between respondent issue position and evaluation. Using a conventional regression methodology, this would involve "dummying up" each issue so that each category represented a separate variable and then observing the pattern of coefficients. While this approach is common for demographic variables, the disadvantage of the approach with regard to issues is that it entails the estimation of quite a large number of parameters. The five issues we examine in this analysis would generate 45 issue coefficients, along with the 12 controls and an intercept, for a total for 58 parameters per party. This could be challenging to interpret, and would create the potential for odd results that were simply statistical artifacts. Clearly a simpler approach is desirable. 10. Peak evaluationjk = bljkl - 2b2jk for the case (b2jk < 0). II. The intercept was calculated in a separate set of regressions which included all the issues in both linear and squared form but excluded the demographic factors. We used this approach to estimate an intercept that was demographically neutral rather than one which reflected a certain demographic prototype. 12. The Conservative Party had used the same tactic in 1972 (Aardal and Valen, 1995: 131). 13. Of necessity the estimated coefficients are identical to those in the separate party equations. 14. Clearly part of Labor's appeal was focussed on BrundtIand (see note 14 below), but there was also a sense that Labor was the party most trusted to manage government. A little background is helpful here. The Labor Party had lost support in the local elections of 1991 compared to the national election in 1989. The two winners were the Center Party and the Socialists, with the Center substantially increasing its vote share. This was widely interpreted as an effect of the rise of the EU issue (Aardal and Val en, 1995: 15). In the opinion polls after the local elections Labor continued to do badly, and by early 1993 in some polls the Conservative Party received a higher level of support than Labor. Then in June, Mr. Thorbj0rn Jagland, the newly elected party chairman of Labor, issued a statement saying that if Labor did not receive more votes than the Conservatives, the party would resign its position in government. This immediately began to consolidate support [ 101 ] 320 for Labor, as many who were not particularly enamored with Labor policy saw it as the only viable governing party (Aardal and Valen, 1995: 16--18). 15. The rating of Brundtland at the election was a record for any leader. At 71.1 she was far more popular than Lahnstein of the Center Party in second place with 59.4. Using the 1989-93 panel, we were able to estimate the effect of leader evaluation on party evaluation and vice-versa. Our results suggest about a 0.3 change in sympathy rating for the party for each point in the sympathy rating of the leader. That would mean that Brundtland contributed somewhere between 2 and 3 points to Labor's overall evaluation. We should note that leaders were typically evaluated somewhat better than their parties in 1993. The differences in the rating for the leader compared to the party were as follows: Socialist, + 1.44; Labor, +6.59; Liberal, -3.45; Center, +6.46; Christian, +2.38; Conservative, -1.93; Progressive, +5.07. References Aardal, B. and Valen, H. (1995). Konflikt og opinion. Oslo: NKS-forlaget. Campbell, A., Converse, P. E., Miller, W. E. and Stokes, D. E. (1960). The American voter. New York: Wiley. Cox, G.W. (1990). Centripetal and centrifugal incentives in electoral systems. American Journal of Political Science 34: 903-935. Davis, O. A., Hinich, M. 1. and Ordeshook, P. C. (1970). An expository development of a mathematical model of the electoral process. American Political Science Review 64: 426448. Downs, A. (1957). An economic theory of democracy. New York: Harper & Row. Enelow, J. M. and Hinich, M. J. (1984). The spatial theory of voting: An introduction. New York: Cambridge University Press. Fiorina, M. (1981). Retrospective voting in American national elections. New Haven and London: Yale University Press. Grofman, B. (1985). The neglected role of the status quo in models of issue voting. Journal of Politics 47: 230-237. Hastie, R. and Park, B. (1986). The relationship between memory and judgment depends on whether the task is memory-based or on-line. Psychological Review 93: 258-268. Hinich, M. J. and Munger, M. C. (1994). Ideology and the theory of political choice. Ann Arbor: University of Michigan Press. Iversen, T. (1994). Political leadership and representation in West European democracies: A test of three models of voting. American Journal of Political Science 38: 45-74. Laver, M. and Schofield, G. (1990). Multiparty government: The politics of coalition in Europe. Oxford: Oxford University Press. Lewis-Beck, M. S. (1991). Economics and politics: The calculus of support. Ann Arbor: University of Michigan Press. Listhaug, 0., Macdonald, S. E. and Rabinowitz, G. (1994). Ideology and party support in comparative perspective. European Journal of Political Research 25: 111-149. Lodge, M., McGraw, K. and Stroh, P. (1989). An impression-driven model of candidate evaluation. American Political Science Review 83: 399--419. Lodge, M. and Steenbergen, M. R. with Brau, S. (1995). The responsive voter: Campaign information and the dynamics of candidate evaluation. American Political Science Review 89: 309-326. [102] 321 Macdonald, S. E., Listhaug, O. and Rabinowitz, G. (1991). Issues and party support in multiparty systems. American Political Science Review 85: 1107-1131. Macdonald, S. E. and Rabinowitz, G. (1993). Direction and uncertainty in a model of issue voting. Journal of Theoretical Politics 5: 61-87. Merrill, S., III. (1995). Discriminating between the direction and proximity spatial models of electoral competition. Electoral Studies 14: 273-287. Merrill, S., III and Grofman, B. (1997). Directional and proximity models of voter utility and choice: A new synthesis and an illustrative test of competing models. Journal of Theoretical Politics 9: 25-48. Niemi, R. G. and Bartels, L. M. (1985). New measures of issue salience: An evaluation. Journal of Politics 47: 1212-1220. Rabinowitz, G. and Macdonald, S. E. (1989). A directional theory of issue voting. American Political Science Review 83: 93-121. Stokes, D. E. (1966). Some dynamic elements of contests for the presidency. American Political Science Review 60: 19-28. Appendix: Wording of the issue questions and sympathy thermometers 1. Left-right (full wording). "There is so much talk about the conflict between left and right in politics. Here is a scale that goes from I on the left - that is, those who are placed politically furthest to the left - to lOon the right, that is, those who are politically furthest to the right. Were would you place yourself on this scale? (5 and 6 are marked by neither/nor.) Where would you place the various parties on such a scale?" 2. Environmental protection. "Value 1 denotes the wish to give much more priority to the protection of the environment, even if it leads to a considerably lower living standard for everyone, including yourself. Value \0 denotes the position that protection of the environment should not go so far that it hurts our standard of living." 3. Immigration policy. "Value 1 on the scale expresses the position that we should make it easier for immigrants to come to Norway, while value 10 expresses the opinion that the number of immigrants should be restricted more strongly than at present." 4. European Union membership (reverse coded). "Value 1 expresses the opinion that Norway absolutely ought to be a member of EU (EF), while value 10 expresses the opinion that Norway absolutely ought not to be a member of EU (EF)." 5. Unemployment. "Value I on the scale expresses the view that government ought to be active in the creation of new jobs, especially in the public sector. Value 10 denotes the position that permanent jobs are primarily to be created by private business." 6. Sympathy thermometer. "We would like to know how you like the parties. We have a scale from 0 to 100, which we label a 'sympathy thermometer.' At the 50-degree [103 ] 322 mark you locate parties that you neither like nor dislike. A party which you like will be located between 50 and 100 degrees. The more you like the party, the higher the mark you will give it. If the party is disliked, you will place it between 0 and 50 degrees, with 0 indicating the least sympathy." [104 ] Public Choice 97: 323-361, 1998. © 1998 Kluwer Academic Publishers. 323 The dynamics of interest group evaluations of Congress * KEITH T. POOLE l & HOWARD ROSENTHAL 2 lCarnegie-Mellon University, 5000 Forbes Avenue, Pittsburgh, PA 15213, U.S.A.; 2Department of Politics, Princeton University, Princeton, NJ 8544, U.S.A. Abstract. We apply a dynamic spatial model to interest group ratings of the members of Congress over the period 1959-1981. Spatial distances between an interest group and the members of Congress are assumed to be monotonic with the ratings. Our pooled cross-sectional time-series data set consists of 203,387 ratings by 59 interest groups. We restrict the spatial coordinates of the interest groups and members of Congress to be polynomial functions of time. Two significant dimensions are recovered: the first dimension, which accounts for approximately 75% of the variance, represents liberal-conservative positions on economic issues; the second dimension, which accounts for approximately an additional 5% of the variance, represents liberal-conservative positions on social issues. Nearly all the interest groups and most members of Congress are ideologically consistent. They are either liberal on both dimensions or conservative on both. 1. Introduction The spatial theory of voting (Downs, 1957; Enelow and Hinich, 1984) must, if it is to develop into a mature science, now pursue empirical analyses with the intensity previously devoted to theoretical developments. Political scientists and others have for some time created spatial representations based on factor analysis, cluster analysis, and other multivariate techniques. More recent work, however, has emphasized scaling methods that, rather than making ad hoc use of existing techniques, were directly responsive to a spatial model of voting. This work has largely focused on the analysis of the National Election Study surveys of American Presidential elections (Rabinowitz, 1976; Cahoon, Hinich, and Ordeshook, 1978; Hinich, 1978; Poole and Rosenthal, 1984a; Enelow and Hinich, 1984). These analyses have been largely cross-sectional. 1 * This work was supported by a grant from the National Science Foundation. An earlier version of this paper was presented at the "Information and Politics Conference", Austin, Texas, 13-15 February 1986. Minor differences in results between this paper and the earlier version reflect the addition of the 1981 data and improvements to the algorithm used for the two- and three-dimensional scalings. [105 ] 324 In many respects, it is Congress, not the Presidency, that provides the richer environment for the study of spatial models. For one, the roll call record provides a much lengthier time series of observations. There are several hundred roll calls per year as against one observation every four years. Moreover, the Congressional record is not prone to the well-known deficiencies of data collection by interview. Congress seems uniquely suited to allowing us to study spatial voting in a dynamic framework, that is, one in which we can study changes in spatial positions in time. The major problem that would appear to confront a spatial analysis of Congress would be the difficulty of analyzing a situation where logrolling and other forms of strategic voting may predominate. The difficulties should not be exaggerated, however. On the one hand, the roll call record itself fails to provide any massive indication of strategic voting. Krehbiel and Rivers (1985), Romer and Rosenthal (1987), and Ladha (1994) indicate that voting on items with a known quantitative ordering (e.g., minimum wage levels) is in accord with sincere spatial voting. More importantly, interest group evaluations furnish us with all important, non-strategic source of data. Interest groups process the roll call record to provide us with yearly evaluations of the candidates. Typically, 10 or more votes are used to score each member of Congress. Certainly, when an interest group weights a vote positively, the interest group sincerely supports the vote so weighted. It is thus natural to take the spatial distance between the interest group and the member of Congress as being monotonic in the quantity (100-rating)/50. (The ratings always range from 0 to 100; division by 50 is an arbitrary normalization.) Assuming that distances are linear in these transformed ratings, we recover spatial coordinates by using a generalized form of the least squares unfolding algorithm of Poole (1984, 1990). The limitations of applying least squares unfolding to the interest group evaluations are: (1) the assumption that distance is linear in the rating (2) the lower bound of 0 to the ratings limits the maximum possible distance (3) each interest group's evaluations being derived from a relatively small number of roll calls implies that the distances between an interest group and the legislators have substantial measurement error. 2 Fortunately, results from scalings based on the interest group evaluations can be cross-checked with results from the NOMINATE stochastic utility model of roll can voting that operates directly on each legislator's roll call voting record. At least in one dimension, the results of interest group unfolding and NOMINATE are highly comparable (Poole and Rosenthal, 1985a,b; 1997: Ch. 8). In addition, we present, in Appendix A, results of bootstrap Monte [106 ] 325 Carlo experiments that suggest that the recovery of the basic two-dimensional spatial coordinates are quite good. These results supplement evidence reported in Poole (1990) that shows the one-dimensional algorithm to be very precise. Unfolding of interest group evaluations should thus lead to an accurate summary of the spatial configuration of members of Congress. Here we analyze a data set spanning the 23-year period from 1959 through 1981. This data set was formed by adding 1981 ratings to the 1959-1980 data set assembled by Poole (Poole and Rosenthal, 1984b; Poole and Daniels, 1985). This addition not only allows us to perform one additional check on the general tenor of the results contained in earlier research but, more importantly, allows us to consider the external validity of the results in terms of a widely perceived shift in a conservative direction represented by the Reagan presidency and by Republican control of the Senate. There were 1,258 representatives, 261 senators and 59 interest groups in our data set. The interest groups consist of "ideological" organizations such as the Americans for Democratic Action and the Liberty Lobby, unions such as the United Auto Workers and the American Federation of Teachers, and broad trade groups such as the Chamber of Commerce of the United States, the National Federation of Independent Business and the National Farmers Organization. Most notably missing, relative to those organizations with visible Political Action Committees, are specific industry groupings dairy farmers, physicians, realtors - and corporate organizations - Mobil, General Motors, etc. Included in the interest groups are Presidential support scores published by Congressional Quarterly. To the extent that these scores are valid representations of the policy position of the White House, we can use these scores to locate the presidents in the same space as the members of Congress. Of the 59 "interest" groups, 14 represent various Congressional Quarterly ratings, including the well-known Conservative Coalition score. Table 1 presents a complete listing of the interest groups. The interest groups provided 203,387 ratings over our 23 year period. Our scaling procedure pools all these ratings in a simultaneous cross-section timeseries analysis. In our scaling model, the legislators are represented in terms of s Euclidean coordinates Xikt, k = 1, ... , s, where s is the dimensionality of the space, t indexes time, and i indexes the legislators. We impose the restriction that at time t, Xikt is a polynomial function of time. That is: where m is the degree of the polynomial used to approximate the temporal behavior of the legislators and interest groupS. 3 Interest group coordinates Zikt [107 ] 326 Table 1. Description of interest groups Abbreviation Number Years rating Group 79-81 71, 72, 74-80 59-81 60,61,63-81 78,89,81 73-77 73, 75-80 American Civil Liberties Union of years ACLU 3 ACU 9 ACA 23 21 ADA AFBF AFGE 3 5 AFSCME 7 AFT 7 ASC 13 BFW 3 2 BCTD 75-81 69-81 79-81 79,81 American Conservative Union Americans for Constitutional Action Americans for Democratic Action American Farm Bureau Federation American Federation of Government Employees American Federation of State, County and Municipal Employees American Federation of Teachers American Security Council Bread for the World Building and Construction Trades Department (AFL-CIO) CFSCA 3 77-80 75-81 79 76-77, 79-80 79,81 81 73-78 77, 79-81 59-81 79-81 CFSCE 2 79-80 Committee for the Survival of a Free Congress: Economic Issues CFSCD 2 79-80 Committee for the Survival of a Free Congress: Defense Issues CFSCS 2 79-80 Committee for the Survival of a Free Congress: Social Issues CCAUSE 2 Common Cause CARTER 4 CCUS 7 CCUS2 CWLA 4 CV 2 CVVF 1 SANE 6 CFNFMP 4 COPE 23 Carter Support Score (CQ) Chamber of Commerce of the United States CCUS 2nd Rating 1979 Senate Child Welfare League of America Christian Voice Christian Voters Victory Fund Citizens for a Sane World Coalition for a New Foreign and Military Policy Committee on Political Education AFL-CIO Committee for the Survival of a Free Congress: All Issues CW 6 CC 23 CFA 10 IKE 2 78,81 75-80 59-81 71-79, 81 59--60 PFPIKE 2 59--60 Eisenhower Presidential Support Score: Foreign Policy (CQ) PDPIKE 2 59--60 Eisenhower Presidential Support Score: Domestic Policy (CQ) [108 ] Congress Watch Conservative Coalition (CQ) Consumer Confederation of America Eisenhower Presidential Support Score: All Issues (CQ) 327 Table 1. Continued Abbreviation Number Years rating Group of years FCNL 3 4 74--76 77-80 Ford Presidential Support Score CCQ) Friends' Committee on National Legislation LBJ 5 64--68 PFPLBJ 5 64--68 Johnson Presidential Support Score: All CCQ) Johnson Presidential Support Score: Foreign Policy CCQ) PDPLBJ 5 64--68 JFK 3 61--63 PFPJFK 3 61--63 Kennedy Presidential Support Score: Foreign Policy CCQ) PDPJFK 3 61--63 Kennedy Presidential Support Score: Domestic Policy CCQ) 59--68 71-77,79-81 71-75,77-81 61--65,67,69, 73,75,77,81 59 Larger Federal Role Support Score: (CQ) FORD LFR 10 LCV 10 LWV 10 LL II LFS NASC NCSC NEA NFO NFU NFIB NTU NWPC NR NIXON REAGAN RIPON 4 4 12 8 17 5 77, 79-81 77-80 69-79, 81 73,75-81 61-65,69-79,81 77-81 10 71,73-81 79 61-74 69-74 81 69-78, 81 77-78 69-81 79,81 1 14 6 I 11 TWR 2 UAW 13 UMW 2 Johnson Presidential Support Score: Domestic Policy CCQ) Kennedy Presidential Support Score: All (CQ) League of Conservation Voters League of Women Voters Liberty Lobby Lower Federal Spending Support Score (CQ) National Alliance of Senior Citizens National Council of Senior Citizens National Education Association National Farmers Organization National Farmers Union National Federation of Independent Business National Taxpayers Union National Women's Political Caucus New Republic Nixon Support Score CCQ) Reagan Support Score CCQ) Ripon Society Taxation with Representation United Auto Workers United Mine Workers [ 109] 328 are similarly restricted. We experimented with s = 1,2,3 and m = 0, 1,2,3. In a nutshell, we are interested in movements through time that can be approximated by low order polynomials. 4 Imposing a low order polynomial structure represents an important advantage over earlier procedures (Poole, 1984) where each year is estimated separately. On average, we have only 20 interest groups present in a given year. Consequently, taking the interest group coordinates as fixed, we would have only 20 observations with which to estimate the legislator coordinates. Because the data exhibit a strong pattern that allows most of the variation in the ratings to be accounted for by a simple unidimensional model, the small N problem was not a severe handicap in obtaining one-dimensional scalings for each year separately. On the other hand, two- or three-dimensional scalings are more prone to unstable results when only 20 observations can be used to estimate each legislator'S coordinates. Consequently, Poole and Daniels (1985) gave only limited attention to multidimensional results. Pooling, in contrast to yearly scalings, avoids small N problems. On average, a representative was present in our data set for a little over eight years. In eight years, a representative would have about 160 ratings, quite satisfactory, say, in terms of estimating the representative's four parameters of a twodimensional linear trend (s = 2, m = 1) model. The more precise estimation afforded by pooling has disclosed a readily interpretable two-dimensional configuration. Our substantive discussion emphasizes the refinements to earlier analysis developed by the two-dimensional configuration. Nonetheless, it is important not to lose sight of the facts that nearly three-fourths of the variation in the ratings is accounted for by a one-dimensional configuration and that only about one-fifth of the residual variation is eliminated by adding a second dimension. Our analysis further discloses that coordinate positions in these low dimensional configurations are highly stable in temporal terms. On the whole, there are only weak trends in the movements of individual legislators and interest groups. This result is subject to the caveat that our analysis pertains only to relative movements. Our methods cannot detect any identical movements undertaken simultaneously by all legislators and interest groups. For example, a uniform shift to the left along one dimension would not be recovered. More precisely, let Xikt = Xikt + akt and let Zikt = Zikt + akt be coordinates that incorporate a general shift parameter, akt. Then, since Zikt - Xikt = Zikt - Xikt, distances are invariant in the unidentifiable parameter akt. Our scaling procedure assumes all akt are zero. Our inability to identity overall shifts makes it important to benchmark our results. We do this in the next section by identifying those senators, representatives, and interest groups who were present over a large number [ 110] 329 of years and exhibited little temporal movement. If one then believes, say, that this set of political actors has become systematically more liberal over the 23 year time series, one can make an appropriate adjustment to the results of our analysis. Our pooled cross-section time-series analysis was made possible by the advent of supercomputing. The number of possible parameters (coordinates) is given by:5 (number of legislators + number of interest groups) x (number of dimensions) x (order of polynomial + 1) = 1578s(m + 1). Using a Cyber 205 supercomputer at Purdue University, we were able to carry out our largest unfolding, that with s = m = 3, in under 10 minutes. The major substantive findings uncovered from these results are as follows: 1. A very good first approximation to the ratings is provided by a unidimensional model where all actors are assumed to occupy a constant spatial position. This model, the s = 1, m = 0 model, produces an R2 of 0.740 between the actual ratings of legislators by interest groups and the distances between legislators and interest groups computed from the recovered coordinates. 2. Significant marginal improvement is obtained by allowing for a twodimensional model with linear trend in individual positions. The R 2 increases to 0.821. Allowing for a higher dimensional space or a higher degree polynomial did not generate substantial increases in explanatory power. 3. Legislators appear to exhibit greater stability in position than interest groups (including presidential support scores!). Indeed, there is little loss in fit (R 2 = 0.818) when the legislators are constrained to have a constant position on the second dimension. 4. Changes captured by the linear trend tend to be small in magnitude. On the whole, to echo Poole and Daniels (1985), there are few systematic shifts in position. Those legislators with large trends also tend to be those legislators with poor overall fits to the model. 5. When the two-dimensional space is rotated so that the horizontal axis is defined by the regression line connecting the Presidential support scores, the analysis strongly suggests that one axis represents liberal-conservative positions on economic issues and the other liberalconservative positions on social issues. [ 111 ] 330 6. Nearly all the interest groups and most members of Congress are "ideologically consistent". They are either liberal on both dimensions or conservative on both. In contrast to the interest groups, however, a significant number of members of Congress are economic liberals but social conservatives. Social liberals who are economic conservatives appear to be largely absent from Congress. (Note the political fortunes of the Libertarian party.) 7. Liberal interest groups tend to be split into two clusters. A "public" interest cluster, including the ADA, the League of Women Voters, and Common Cause, appears to be, holding economic liberalism constant, relatively liberal on social issues. An economic liberal cluster, which includes all the labor and farm organizations, is relatively conservative on the social dimension. 8. The Democratic presidents are closely grouped and appear to be (on both dimensions) just to the right of the core of the Democratic party. In contrast, Republican presidents are located well away from the core of the Republican party. Indeed, the three presidents prior to Reagan occupy a region of the space that represents economic conservatism combined with social moderation. This region is largely devoid of other actors, be they legislators or interest groups. And while Reagan is, as expected, closer to the party core than any other Republican president, even his support score coordinates show a surprising degree of relative moderation. 9. The standard result that the Senate is more liberal than the House is confirmed - until 1981 when the shift to a Republican Senate majority reversed the relationship. The greater liberalism of the Senate in our results can largely be attributed to the result that Senate Republicans were substantially more liberal than House Republicans. (Also see Kernell, 1973.) 10. Positions appear to have polarized along party lines over the period spanned by the data. (Also see Poole and Rosenthal, 1984b.) In the next section of the paper, we develop these findings in detail. [ 112] 331 Table 2. R2 Between input and recovered distances Degree of polynomial Dimensionality Constant Linear Quadratic Cubic of space m=O m= 1 m=2 m=3 s= 1 .740 .765 .776 .782 .832 .844 .839 s= 2 .793 .821 s=3 .802 .831 .851 2. Results 2.1. Evaluations can be represented in a low dimensional space with minor linear trends An overall view of the recovery is provided in Table 2. It can be seen that the interest group evaluations are essentially unidimensional and stable. The s = 1, m = 0 model "explains" 74.0% of the variation in the original ratings. The biggest marginal change in explained variation is brought about by adding a second dimension. Adding a second dimension to any of the polynomial specifications in the table adds over 5% to the explained variation. In contrast, adding a third dimension adds only around 1%. The second largest marginal change results from adding a linear trend. Including the linear term adds over 2.5% to the explained variation whereas a quadratic term adds only around 1% and a cubic term even less. In the two-dimensional model with linear trend, interest groups have movement along both dimensions whereas legislators appear to demonstrate trend solely on the first, main dimension. The R 2 for the full s = 2, m = 1 model is 0.821. Eliminating 1218 legislator trend parameters (Xi2\) on the second dimension lowers R2 only to 0.818. In contrast, dropping the second dimension trend parameters for just the 43 interest groups with a linear term (rating three or more years), lowers R2 to 0.8138 while dropping the legislator second dimension parameters in addition to those for the interest groups results in negligible further loss, with an R2 of 0.8136. There is more sensitivity to dropping a linear trend for the legislators on the first dimension. If the legislators are constrained to constant positions on both dimensions, the R2 falls to 0.803. We believe that adding polynomial terms beyond linear trend only results in fitting noise in the data. Table 3 presents supporting evidence. We split the sample in terms of whether the interest group or legislator was present in our dataset for 10 or more years vs. less than 10 years. For each set of [113 ] 332 Table3. Magnitude of annual change in spatial coordinate given by estimated linear trend parameters Group Interest groups In sample 3 to 9 years In sample 10 to 23 years Average SD Average SD .0513 0.517 27 .0169 .0206 16 .0397 .0398 27 .0201 .0170 16 .0452 .0488 103 .0173 .0195 110 .0422 .0428 591 .0201 .0199 414 .0378 .0432 81 .0218 .0234 42 .0394 .0387 423 .0198 .0207 215 N N first dimension Interest groups second dimension Senators first dimension Representatives first dimension Legislators entering Congress after 1959 Senators first dimension Representatives Notes. (I) No linear terms was estimated for groups or legislators in sample for only one or two years. (2) Table based on two dimensional model with legislators constrained to constant position on second dimension. interest groups, senators, and representatives, we then computed the average and standard deviation of the absolute value of the annual change in position implied by the estimated trend coefficients. Those present in the sample for relatively short periods had substantially higher average annual movements and exhibited greater variability about this average. 6 The above results might be contaminated by the fact that many individuals in our sample for only a short period were in fact senior members of Congress approaching the ends of their careers in 1959. Both to guard against this possibility and to make a quick examination of possible cohort effects, we consequently redid the computations in Table 3 using only those legislators who began service after 1959. The work indicates that the results are robust to this modification. We therefore conclude that much of what is picked up by the trend estimates computed over short-time intervals is not true trend but simply fit to a (possibly autoregressive) random walk. Put differently, from one year to the next, a legislator'S position will appear to change as the set of interest groups changes, as the roll calls chosen to form the ratings change, [ 114 ] 333 and as the legislator's "true" position changes. In the short run, these nonsystematic changes can be captured as trend, but in a longer run, these effects will cancel out. Additional evidence that our computations exaggerate the degree of trend for individual legislators and interest groups is contained in the Appendix and in Poole (1990). These results strongly indicate that there is substantial long-term stability in spatial positions. Note that, for legislators serving 10 or more years, the annual change in position averages on the order of 0.02 or less than one percent of the length of the main dimension. Overall, linear trend makes only a minor improvement in the R2 fit of the model. But much of the improvement in fit is illusory, reflecting short-term changes rather than true long-term trends. Similarly, adding dimensions will be sensitive to noise in the data. Consequently, we look skeptically at adding 2780 parameters to move from a full s = 2, m = I model to a s = 3, m = I model when R2 increases only by 0.01. Insisting on parsimony, therefore, we restrict our remaining discussion to coordinates estimated in a two-dimensional model with linear trend on both dimensions for the interest groups and on the first (untranslated) dimension only for the legislators. Discussion that does not refer to trends is based on coordinates representing the middle of each political actor's rating period. Our finding of low dimensionality in the interest group ratings matches our similar findings for roll call voting in Congress (Poole and Rosenthal, 1997). These results strongly support a body of spatial theory the posits a fundamental or basic ideological space of low dimensionality that generates individuals' issue positions (Ordeshook, 1976; Hinich and Pollard, 1981; Hinich and Munger, 1994). Indeed, Hinich and Munger (1994) argue that a low dimensional space is virtually inevitable in most real world circumstances of political conflict. 2.2. The Presidential translation The two-dimensional results are more readily interpretable if we perform a simple translation of the space produced by the unfolding algorithm. Noticing that the Presidential support scores for both Democrats and Republicans prior to Reagan form relatively tight party clusters, we defined the horizontal axis of the translated space to be the regression line over the set of pre-Reagan Presidential support scores. 7 The origin was chosen to represent a point on the regression line with a zero first dimension coordinate in the original scaling. The results of this translation can be seen in Figures 1-4. Figure 1 shows the average coordinates for all 59 interest groups. A key to the displayed abbreviations can be found in Table 1. For each interest group, Figure 2 shows the coordinates for its first and last years in the data set. To clarify the plot as much as possible, we have eliminated those interest groups who [ 115] 334 Social Dimension Two Dimensions, Constant Postion 1.2 Cons. LL 0.8 0.4 N~S BCTD NFO REAGAN NIXON 0 PFP KE IKJ:DPIKE S~IPON -0.4 AD~ -0.8 C SAN Lib. NTU V TWR -1.2 -1.5 -1 Liberal -0.5 o Economic Dimension 0.5 1 1.5 Conservative Figure 1. Positions of interest groups on the economic and social dimensions. The positions in the plot are from an estimation where interest groups and legislators are constrained to have a fixed, constant position over the 23 years of the estimation. See Table I for a key to the interest group abbreviations. were present fewer than three years. Groups with negligible trend have only one plotted point. Figures 3 and 4 display average coordinates for senators and representatives, respectively. To provide a basic interpretation of the translation, we return to Figure l. The specification of the translation forces the six presidents prior to Reagan to have an average second dimension coordinate of zero. Moreover, all the presidents are close to the horizontal axis, with Carter and Eisenhower being relatively liberal, Kennedy, Johnson, and Nixon relatively conservative. Although Reagan is, as expected, more conservative than the other Republican presidents (see Figure 1), the differences are minor. If Reagan had been used in the regression defining the translation, the remaining results in this paper would be virtually unchanged. Other political actors with ratings above 0.3 are more Conservative on the second translated dimension than are all the [ 116] 335 Social Dimension 1.2 Cons. r Two Dimensions, Linear Endpoints I . - -.... II ACU LL 0.8 as{:C aca NAd!/ ACA ASC nfo 0.4 NMJ"'hCSFCA ccusalbf ford nfu NFO aE 0 a S~ -0.8 I L -1.2 -1.5 -1 ntu r ADA CFA I Lib. FK "lwv6~ -0.4 nixoncsfca NIXON tN . FORD ripon 1\ Iwv RIPmN I nr -0.5 Liberal o Economic Dimension 0.5 1 1.5 Conservative Figure 2. Positions of interest groups on the economic and social dimensions. The positions in the plot are from an estimation where interest group positions can change linearly in time. The lowercase letters show the group at the beginning of its rating period. The uppercase letters show the group at the end of its rating period. See Table 1 for an interest group's abbreviation and rating period. presidents (or at least their support scores) while those below -0.3 are more liberal. 2.3. A nearly unidimensional space Inspection of the figures immediately discloses that our two-dimensional scaling is basically a refinement of a nearly unidimensional space. The figures all form relatively fiat ovals with the basic, unidimensional orientation running from the lower left to the upper right. This orientation represents the axis of the first dimension before translation. Coordinates on the untranslated first dimension, in turn, are virtually identical to those estimated in a simple unidimensional model with no time trends (s = 1, m = 0) over our 23 year period. Squared correlations between the two sets of results all exceed 0.98 as seen in the first row of Table 4. [ 117] 336 Social Dimension Two Dimensions, Linear Time Model 1.2 Cons. 0.8 0.4 0 -0.4 -0.8 Lib. -1.2 -1.5 -1 o -0.5 Economic Dimension Liberal 1 0.5 1.5 Conservative Figure 3. Positions of senators on the economic and social dimensions. The positions in the plot are from an estimation where legislator positions can change linearly in time. For each legislator, the mean position is shown. Northern Democrats are displayed as "d" tokens, Southern Democrats as "s" tokens, and Republicans as "r" tokens. Table 4. Squared correlations between unidimensional coordinates and two-dimensional results Two-dimensional Interest groups Senators Representatives 0.982 0.989 0.993 0.012 0.009 0.013 0.956 0.929 0.951 0.934 0.668 0.775 scaling Recovered first dimension Recovered second dimension Translated first dimension Translated second dimension [ 118 ] 337 Social Dimension Two Dimensions, Linear Time Model 1.2 5 Cons. 0.8 0.4 0 -0.4 -0.8 Lib. -1.2 -1.5 -1 Liberal o -0.5 Economic Dimension 0.5 1 1.5 Conservative Figure 4. Positions of representatives on the economic and social dimensions. See also the caption to Figure 3. The dominance of the untranslated first dimension, seen graphically in the figures, is confirmed in Table 5 where it is shown that the standard deviation of the distribution of coordinates on the first untranslated dimension is roughly three times as great as on the second dimension. 2.4. Substantive interpretation of the axes In the translated space seen in Figure 1, widely known features of the interest groups readily suggest the content of the new axes. The first dimension clearly represents the major distinction between Democratic and Republican administrations. Those favoring, grosso modo, larger government non-defense spending and government regulation of business are on the left. These interest groups include unions of industrial workers, public employees, teachers, the environmentalists and the Naderites. The opposite, conservative positions are on the right. The second dimension just as clearly distinguishes social liberals from social conservatives. Among the economic conservatives, groups such [ 119] 338 Table 5. Standard deviations of coordinates Coordinate Interest groups Senators Representatives 0.772 0.628 0.650 0.313 0.231 0.200 Recovered First dimension Second dimension Translated First dimension 0.724 0.585 0.592 Second dimension 0.412 0.324 0.334 Translated weighted! First dimension 0.780 0.566 0.581 Second dimension 0.454 0.326 0.337 ! Note. In the case of interest groups, "weighted" means that all Congressional Quarterly ratings were eliminated. The resulting N is 43. In the case of legislators, each legislator is weighted by the number of years he or she appeared in the sample. as Christian Voice which are close to Jesse Helms on abortion, school prayer, and other issues, can be distinguished from the more moderate Ripon Society and National Taxpayers Union. Among economic liberals, the group with the largest, most conservative score on the second dimension is the Building and Construction Trades Union, not known as an avid proponent of minority hiring. In contrast, known social liberal groups like the ADA and ACLU have large negative coordinates on the second dimension. Our contrast between social and economic dimensions based on an analysis of the interest groups is confirmed by the legislator coordinates. On the economic dimension, for all 23 years and for both houses, Northern Democrats have an average position that is more liberal than that of Southern Democrats which is more liberal than that of Republicans. (See Figure 6.) But on the social dimension, again for all years and for both houses, the positions of Southern Democrats and Republicans are reversed. Southern Democrats are the most conservative of the three groups. (See Figure 7.) This reversal suggests that the issue content of the social dimension is largely that of desegregation, voting rights and other civil rights issues. While the social dimension indeed includes civil rights, it also appears that it must be more broadly defined. To see this, consider Table 6, which lists, for each year and party group, the individuals with the most extreme positions on each dimension. In both houses, a Southern Democrat - typically from the deep South - is almost always the most conservative individual on the social di- [120] 339 mension. On the other hand, the most socially liberal individual is frequently not a Northern Democrat with a large minority constituency. For six years in the Senate and ten in the House, the most socially liberal individual was a Republican. Among the social liberals, we find those as identified with opposition to the Vietnam war (McCloskey, Harrington, Drinan) or feminist issues (Fenwick) as with civil rights advocacy. Indeed, the liberal end of the social dimension appears to capture all the issues that have been the cause celebre du jour of White urban liberals. In the early part of our time period, civil rights was perhaps the central issue in the White liberal agenda, but in the middle and later part, antiwar, environmental, and consumer issues also occupied this agenda. The emphasis we have placed on White liberal issues is confirmed by the finding that African-American and Hispanic members of the House do not anchor the social dimension but frequently (Clay, Badillo, Garcia) anchor the economic dimension. Issues such as affirmative action, welfare, and food stamps that involve redistributional benefits to minorities no longer appear as ones of civil rights but as issues along the primary, economic dimension. 8 It is noteworthy that while minority group legislators tend to anchor the liberal end of the economic dimension, among interest groups this dimension is anchored by organized labor. From the viewpoint of many economists, this spatial grouping would seem anomalous since many policies dear to labor, such as minimum wage and Davis-Bacon, would appear to restrict employment in a manner detrimental to the very poor. Two explanations appear worth considering. On the one hand, African-American and Hispanic members of the House may not have the very poor as their critical constituency but instead represent those minority individuals with a stake in the system by holding industrial or civil service jobs. On the other hand, these same members might realize that extreme redistributional policies have little chance of passing and, therefore, coalesce with organized labor in order to obtain the best possible deal for their constituents. The latter case suggests that coalition formation may play a key role in the near unidimensionality in our results. Although the interests of minorities and labor could give rise to two distinct economic dimensions, relatively permanent coalition decisions could show only one dimension in the data. In tum, the permanency of the coalitions will make roll call voting appear to be "sincere" even though it results from a fundamentally strategic process. 2.5. Liberal interest groups: Middle class vs. labor and farming To return to the social dimension, our claim of a White liberal pole is perhaps best documented by the separation of liberal interest groups into two distinct clusters apparent in Figure 1. The bottom cluster is middle class, college [121] 340 Table 6. Most extreme legislators A. Most liberal - Economic dimension - Senate Years 1959-66 1967-68 1969-70 1971 1972 1973-74 1975 1976 1977 1978 1979 1980 1981 All Senators No. Democrat So. Democrat Republican Morse (D-OR) Morse (D-OR) Yarborough (D-TX) Javits (R-NY) Morse (D-OR) Morse (D-OR) Yarborough (D-TX) Hatfield (R-OR) Schwieker (R-PA) Gravel (D-AK) Yarborough (D-TX) Schweiker (R-PA) Schwieker (R-PA) Gravel (D-AK) Fulbright (D-AR) Schweiker (R-PA) Gravel (D-AK) Gravel (D-AK) Fulbright (D-AR) Schweiker (R-PA) Hartke (D-IN) Hartke (D-IN) Huddleston (D-KY) Schweiker (R-PA) Durkin (D-NH) Durkin (D-NH) Huddleston (D-KY) Case, C. (R-NJ) Hartke (D-IN) Hartke (D-IN) Huddleston (D-KY) Case, C. (R-NJ) Durkin (D-NH) Durkin (D-NH) Huddleston (D-KY) Case, C. (R-NJ) Durkin (D-NH) Durkin (D-NH) Allen, M. (D-AL) Case, C. (R-NJ) Durkin (D-NH) Durkin (D-NH) Huddleston (D-KY) Case C. (R-NJ) Kennedy, E. (D-MA) Kennedy, E. (D-MA) Huddleston (D-KY) Javits (R-NY) Dodd, C. (D-CT) Dodd, C. (D-CT) Huddleston (D-KY) Javits (R-NY) Republican Note. A Northern Democrat is always the most liberal Democrat. B. Most liberal - Economic dimension - House Years All Reps. No. Democrat So. Democrat 1959 1960 1961-62 1963-64 1965 1966 1967 1968 1969 1970-72 1973-76 1977 1978 1979-80 1981 Green E. (D-OR) Green, E. (D-OR) Edmondson (D-OK) Fino (R-NY) Hargis (D-KS) Hargis (D-KS) Edmondson (D-OK) Canfield (R-NY) Bailey (D-WV) Bailey (D-WV) Edmondson (D-OK) Halpern (R-NY) Olsen (D-MT) Olsen (D-MT) Edmondson (D-OK) Reid (R-NY) Ford, W. (D-MI) Ford, W. (D-MI) Farnsley (D-OK) Reid (R-NY) Rees (D-CA) Rees (D-CA) Thomas, L. (D-TX) Reid (R-NY) Ford, W. (D-MI) Ford, W. (D-MI) Eckhardt (D-TX) Heckler, M. (R-MA) Podell (D-NY) Podell (D-NY) Eckhardt (D-TX) Whalen (R-OH) Clay (D-MO) Clay (D-MO) Eckhardt (D-TX) Reid (R-NY) Collins (D-IL) Collins (D-IL) Eckhardt (D-TX) Reid (R-NY) Badillo (D-NY) Badillo (D-NY) Young, A. (D-GA) Whalen (R-OH) Badillo (D-NY) Badillo (D-NY) Eckhardt (D-TX) Whalen (R-OH) Garcia (D-NY) Garcia (D-NY) Eckhardt (D-TX) Whalen (R-OH) Garcia (D-NY) Garcia (D-NY) Leland (D-TX) Rinaldo (R-NJ) Savage (D-IL) Savage (D-IL) Leland (D-TX) Rinaldo (R-NJ) Note. A Northern Democrat is always the most liberal Democrat. [122 ] 341 Table 6. Continued C. Most conservative - Economic dimension - Senate Years All Senators No. Democrat So. Democrat Republican 1959-60 1961-64 1965 1966 1967 1968 1969-71 1972 1973-76 1977 1978 1979-80 1981 Goldwater (R-AZ) Lausche (D-OH) Thurmond (D-SC) Goldwater (R-AZ) Goldwater (R-AZ) Lausche (D-OH) Byrd, H.F. (D-VA) Goldwater (R-AZ) Simpson (R-WY) Lausche (D-OH) Byrd, H.F. (D-VA) Simpson (R-WY) Simpson (R-WY) Lausche (D-OH) Robertson (D-VA) Simpson (R-WY) Tower (R-TX) Lausche (D-OH) Byrd, Jr. (D-VA) Tower (R-TX) Curtis (R-NE) Lausche (D-OH) Byrd, Jr. (D-VA) Curtis (R-NE) Goldwater (R-AZ) Bible (D-NV) Byrd, Jr. (D-VA) Goldwater (R-AZ) Goldwater (R-AZ) Bible (D-NV) Edwards (D-LA) Goldwater (R-AZ) Helms (R-NC) Proxmire (0-WI) Allen (D-AL) Helms (R-NC) Helms (R-NC) Zorinsky (D-NE) Allen (D-AL) Helms (R-NC) Helms (R-NC) Hatfield, P. (D-MT) Eastland (D-MS) Helms (R-NC) Zorinsky (D-NE) Stennis (D-MS) Helms (R-NC) Helms (R-NC) Mattingly (R-GA) Zorinsky (D-NE) Stennis (D-MS) Mattingly (R-GA) Note. A Southern Democrat is always the most conservative Democrat, except in 1978. D. Most conservative - Economic dimension - House Years All Reps. No. Democrat So. Democrat Republican 1959 1960 1961 1962 1963-64 1965-66 1967-68 1969-70 1971-72 1973 1974 1975 1976 1977-78 1979-81 Hall (D-NC) Cannon (D-NV) Hall (D-NC) Mason (R-IL) Mason (R -IL) Cannon (D-NV) Haley (D-FL) Mason (R-IL) Mason (R-IL) Cannon (D-NV) Williams (D-MS) Mason (R-IL) Mason (R-IL) Hull (D-MO) Williams (D-MS) Mason (R-IL) Hoffman (R-IL) Hull (D-MO) Williams (D-MS) Hoffman (R-IL) Walker P. (R-MS) Jones, P. (D-MO) Williams (D-MS) Walker, P. (R-MS) Gross, H. (R-IA) Jones P. (D-MO) Rarick (D-LA) Gross, H. (R-IA) Crane (R-IL) !chord (D-MO) Rarick (D-LA) Crane (R-IL) Crane (R-IL) Byron (D-MO) Rarick (D-LA) Crane (R-IL) Treen (R-LA) Byron (D-MO) Rarick (D-LA) Treen (R-LA) Treen (R-LA) !chord (D-MO) Rarick (D-LA) Treen (R-LA) Treen (R-LA) !chord (D-MO) McDonald (D-GA) Treen (R-LA) Paul R. (R-TX) !chord (D-MO) McDonald (D-GA) Paul, R. (R-TX) McDonald (D-GA) Stump (D-AZ) McDonald (D-GA) Symms (R-ID) Paul R. (R-TX) Stump (D-AZ) McDonald (D-GA) Paul R. (R-TX) Note. A Southern Democrat is always the most conservative Democrat. [ 123] 342 Table 6. Continued E. Most liberal - Social Dimension - Senate Years All Senators No. Democrat So. Democrat Republican 1959-62 1963-64 1965-66 1967-68 1969-72 1973 1974 1976 1976 1977-81 Proxmire (D-WI) Proxmire (D-WI) Kefauver (D-TN) Cooper (R-KY) Proximire (D-WI) Proxmire (D-WI) Edmondson (D-NE) Cooper (R-KY) Cooper (R-KY) Proxmire (D-WI) Proxmire (D-WI) Harris (D-OK) Proxmire (D-WI) Proxmire (D-WI) Harris (D-OK) Hatfield, M. (R-OR) Proxmire (D-WI) Proxmire (D-WI) Harris (D-OK) Saxbe (R-OH) Proxmire (D-WI) Proxmire (D-WI) Chiles (D-FL) Saxbe (R-OH) Metzenbaum (D-OH) Metzenbaum (D-OH) Chiles (D-FL) Case, C. (R-NJ) Proxmire (D-WI) Proxmire (D-WI) Bumpers (D-AR) Case, C. (R-NJ) Case, C. (R-NJ) Proxmire (D-WI) Bumpers (D-AR) Case, C. (R-NJ) Chafee (R-RI) Metzenbaum (D-OH) Bumpers (D-AR) Chafee (R-RI) Note. A Northern Democrat is always the most liberal Democrat. F. Most liberal - Social dimension - House Years All Reps. No. Democrat So. Democrat Republican 1959 1960 1961 1962 1963-64 1965 1966 1967-68 1969-70 1971 1972-73 1974 1975 1976-77 1978 1979 1980 1981 Boyle (D-IL) Boyle (D-IL) Johnson, B. (D-TX) Canfield (R-NJ) Canfield (R-NJ) Kastenmeier (D-WI) Johnson, B. (D-TX) Canfield (R-NJ) Kastenmeier (D-WI) Kastenmeier (D-WI) Wickersham (D-OK) Merrow (R-NH) Kastenmeier (D-WI) Kastenmeier (D-WI) Fascell (D-FL) Cameron (D-CA) Cameron (D-CA) Fascell (D-FL) Mathias (R-MD) Jacobs (D-IN) Jacobs (D-IN) Fascell (D-FL) Mathias (R-MD) Jacobs (D-IN) Jacobs (D-IN) Thomas, L. (D-TX) Kupferman (R-NY) McCloskey (R-CA) Jacobs (D-IN) Eckhardt (D-TX) McCloskey (R-CA) McCloskey (R-CA) Harrington (D-MA) Eckhardt (D-TX) McCloskey (R-CA) Frenzel (R-MN) Drinan (D-MA) Eckhardt (D-TX) Frenzel (R-MN) Drinan (D-MA) Drinan (D-MA) Eckhardt (D-TX) McCloskey (R-CA) Drinan (D-MA) Drinan (D-MA) Eckhardt (D-TX) Mosher (R-OH) Fenwick (R-NJ) Maguire (D-NJ) Fisher (D-TX) Fenwick (R-NJ) Maguire (D-NJ) Maguire (D-NJ) Fisher (D-TX) Fenwick (R-NJ) Green (R-NY) Maguire (D-NJ) Fisher (D-TX) Green (R-NY) Green (R-NY) Maguire (D-NJ) Leland (D-TX) Green (R-NY) Maguire (D-NJ) Maguire (D-NJ) Leland (D-TX) Green (R-NY) Schneider (R-RI) Frank, B. (D-MA) Leland (D-TX) Schneider (R-RI) Note. A Northern Democrat is always the most liberal Democrat. [124 ] Merrow (R-NH) 343 Table 6. Continued G. Most conservative - Social dimension - Senate Years 1959 1960 1961 1962 1963-64 1965-71 1972-77 1978 1979-80 1981 All Senators No. Democrat Langer (R-ND) So. Democrat Republican Frear (D-CT) Talmadge (D-GA) Langer (R-ND) Talmadge (D-GA) Frear (D-CT) Talmadge (D-GA) Young (R-ND) Talmadge (D-GA) Talmadge (D-GA) Young (R-ND) Bible (D-NV) Talmadge (D-GA) Bible (D-NV) Talmadge (D-GA) Bottum (R-SD) Talmadge (D-GA) Bible (D-NV) Talmadge (D-GA) Young (R-ND) Ellender (D-LA) Bible (D-NV) Ellender (D-LA) Young (R-ND) Eastland (D-MS) Bible (D-NV) Eastland (D-MS) Young (R-ND) Long, R. (D-LA) Hatfield, P. (D-MT) Long, R. (D-LA) Young (R-ND) Long, R. (D-LA) Exon (D-NE) Long, R. (D-LA) Young (R-ND) Long, R. (D-LA) Exon (D-NE) Long, R. (D-LA) Grassley (R-IA) Note. A Southern Democrat is always the most conservative Democrat. H. Most conservative - Social dimension - House Years All Reps. 1959 1960 1961 1962 1963 1964 1965 1966-68 1969-70 1971-72 1973-74 1975 1976 1977-78 1979-80 1981 Hall (D-NC) Barden (D-NC) No. Democrat So. Democrat Republican Brock, L. (D-NE) Hall (D-NC) Weaver (R-NE) Brock, L. (D-NE) Barden (D-NC) Weaver (R-NE) Riley (D-SC) Cannon (D-NV) Riley (D-SC) Weaver (R-NE) Blitch (D-GA) Cannon (D-NV) Blitch (D-GA) McVey (R-KS) Whitten (D-MS) Cannon (D-NV) Whitten (D-MS) Foreman (R-TX) Winstead (D-MS) Baring (D-NV) Winstead (D-MS) Nygaard (R-ND) Whitten (D-MS) Baring (D-NV) Whitten (D-MS) Reifel (R-SD) Long, S. (D-LA) Baring (D-NV) Long, S. (D-LA) Reifel (R-SD) Caffery (D-LA) Baring (D-NV) Caffery (D-LA) Reifel (R-SD) Baring (D-NY) Baring (D-NV) Long, S. (D-LA) Price (R-TX) Teague (D-TX) Byron, G. (D-MD) Teague (D-TX) Price (R-TX) Teague (D-TX) !chord (D-MO) Teague (D-TX) Harsha (R-OH) Teague (D-TX) !chord (D-MO) Teague (D-TX) Quillen (R-TN) Teague (D-TX) Ichord (D-MO) Teague (D-TX) Young, O. (R-FL) Roberts (D-TX) !chord (D-MO) Roberts (D-TX) Young, O. (R-FL) Chappel (D-FL) Byron, B. (D-MD) Chappel (D-FL) Young, O. (R-FL) Note. A Southern Democrat is always the most conservative Democrat, except in 1971 and 1972. [ 125 ] 344 educated liberals with "public" interest orientations. It includes the ADA, the League of Women Voters, the League of Conservation Voters, the Consumer Federation of America, the anti-nuclear SANE, Congress Watch, the Friends Committee on National Legislation, the Coalition for a New Foreign and Military Policy, Taxation with Representation, Common Cause, and the ACLU. The other cluster is overwhelmingly made up of labor and farming groups. The Democratic presidents fall in this cluster. The distinction between these clusters did not escape Ronald Reagan the campaigner who frequently sought to pose as the candidate of Roosevelt and Kennedy Democrats. These messages might be read as: "My opponent is really up there in the other cluster with all those crazy Brie eaters". In addition to the interest group clusters, three other aspects of the figures of translated coordinates are especially worth noting. 2.5.1. Extreme positions of interest groups First. Interest groups are generally more extreme than legislators. No senator, for example, is as far "Southwest" as either Taxation With Representation or SANE or as far "Northwest" as the BCTD. Similarly, no senator is as economically conservative as the National Taxpayers Union or as socially conservative as the Liberty Lobby. Another way to state this conclusion about the extremity of the interest groups is to note, in Table 5, that the standard deviations of interest group coordinates are greater than those of either house of Congress. This result is especially striking given that the standard deviation of the interest groups is substantially reduced by the moderate positions embodied in the Presidential support scores. As the table shows, when all Congressional Quarterly ratings are removed, including not only the Presidential support score but also the highly conservative Conservative Coalition score, the standard deviations for the interest groups are increased. 2.5.2. Consistent and inconsistent positions Second. The 45 authentic (non-Congressional Quarterly scores) interest groups are largely polarized into two opposite quadrants. Accounting for 82% of the interest groups, these quadrants represent the "consistent" positions of "liberal-liberal" and "conservative-conservative". In contrast, a significant portion of the Congressional membership can be said to have had "inconsistent" positions. While none of the three groups of actors contain significant numbers in the economic conservative-social liberal quadrant (4% of the interest groups, 2% of the senators, and 1% of the representatives), there were many members of Congress in the opposite, economic liberal-social conservative quadrant (28% of the senators and 27% of the representatives as against only 13% of the interest groups.) These individuals were disproportionately Democrats, since 45% of Senate Democrats and 44% of House Democrats [ 126] 345 were economic liberals but more "socially" conservative than the average pre-Reagan presidential position. Within the Democrats, Southern Democrats were overwhelmingly socially Conservative. The two quadrants above the Presidential regression line contain 96% of both House and Senate Southern Democrats. American polities would be more truly two-dimensional if there was a significant representation of the missing quadrant of economic Conservatives and social liberals. The sparse dots in this quadrant in both Figures 3 and 4 all represent Republicans. There is, of course, very little current impetus to the development of socially liberal Republicans. In contrast, Congress is gradually polarizing along the lines of the interest groups? 2.5.3. Republican presidents and political compromise While the Democratic presidents were all close to the high density areas in Figures 3 and 4 that show the core group of liberal legislators, the social positions of postwar Republican presidents preceding Reagan were very far from those of the core conservative group on the social dimension. Of course, this distinction has been echoed in Ronald Reagan's ascendancy to dominance in the Republican party. Nonetheless, Reagan's support score position in the space is still more liberal than that of the conservative core. One readily suspects that Reagan's relative moderation reflects the desire of the White House to take winning positions on legislation. Winning involves building consensus, thereby compromising one's position. The need to win will have the greatest impact on Republican presidents who, up until Reagan, were faced with Democratic control of Congress during the period covered by our data. We thus suspect that the Presidential positions calculated from the support score ratings are less extreme than positions taken in campaigning. Indeed, our earlier work using mass electorate survey data also showed polarization of Presidential campaign positions (Poole and Rosenthal, 1984a). The distinctions we have made between the social and economic realms should not mask the basically unidimensional character of American politics. As seen in the figures, there is a strong correlation between positions on the two dimensions. Moreover, as seen in Table 4, the R2 values between position on the economic dimension and position in the unidimensional scaling all exceed 0.92. The R2 between the social dimension and the unidimensional results for the interest groups is an equally high 0.934. The squared correlations for the two houses of Congress are less strong but still substantial. The Senate has the lowest R2 between the social dimension and the unidimensional coordinate; relative to the House (see Table 5), it also has less variance on the first recovered dimension but more variance on the second. These results echo the finding of Poole and Rosenthal (1985b, 1997) that the [127] 346 Senate fits a one-dimensional model less well than the House. This lesser fit in part reflects the Senate's greater involvement with civil rights issues during this period. Analysis in Poole and Rosenthal (1985b; 1991; 1997: 109111) also indicates that civil rights is basically a perturbation in a long-run unidimensional pattern. Even this perturbation, however, can be reasonably approximated within a one-dimensional structure, as the R 2 values indicate. In other words, the basic story is left-right. The two-dimensional plot simply refines this story. 2.6. Intertemporal change The results presented in Figures 3 and 4 are useful for presenting a longrun or average spatial description of the national political system. But what of change over time? Improvements in fit here are scant compared to what was achieved by expanding the dimensionality from one to two. For onedimensional unfoldings, the improvement achieved by a cubic polynomial is less than that available from assuming constant spatial positions in a twodimensional space. Systematic trends are thus minor. 2.7. Calibration: Actors with no trend Given the central tendency to no trend, it is important to document those interest groups and legislators that exhibited the least movement. It is difficult to do this for interest groups, since we had only five interest groups (see Table 1) that had ratings for 16 or more years. Of these, the Conservative Coalition score exhibited the least total movement, covering a distance of only 0.094 in 23 years. COPE was also relatively stable, moving only 0.161 in 23 years. (The entire space has a span of over 2 units.) COPE also exhibited the least movement on the economic dimension, a total of 0.038 units while the Conservative Coalition was the most stable of the five interest groups on the social dimension, moving only a total of 0.020 units. To put the interest groups in the same framework as the legislators, however, it is best to refer to the untranslated first dimension since the legislators were constrained to move only on this dimension. Of those interest groups rating 16 or more years, only the Americans for Constitutional Action had a total movement of less than 0.02 units. The legislators that had a total movement of less than 0.02 were, in the Senate, Bennett (R-UT), Hart (D-MI), Hruska (R-NE), Metcalf (O-MT), and Williams (D-NJ), and, in the House, Conyers (D-MI), Corman (D-CA), Edwards (D-CA), Gubser (R-CA), Hechler (0WV), Howard (D-NJ), MacDonald (D-MA), and Satterfield (R-VA). These individuals provide reasonable coverage of the political spectrum, the two parties, and the regions of the country, with the notable exception of (see [128] 347 below) Southern Democrats. If one believes that this set of individuals became systematically more liberal or more Conservative during the sixties and seventies, then one would want to accordingly recalibrate our results. Our discussion assumes these individuals were stable. Although the linear trend explains a "significant" chunk of the variance in the ratings, the trends are typically of very limited substantive interest. Plots for individual years would be highly similar to Figures 3 and 4. As shown in Table 3, the average yearly movements on the first untranslated dimension are all similar for interest groups, representatives, and senators and are small in relation to the overall size of the space, especially so for those present long enough to avoid substantial upward bias in the magnitude of trend. Another indication of limited movement is provided by Table 6 which shows substantial stability in the identity of those individuals who hold the most extreme positions in Congress. Of particular interest is the social dimension in the Senate. Among Northern Democrats, only two individuals, Metzenbaum and Proxmire have held the liberal pole position in our 23-year period. On the conservative end, for all three "parties", only four individuals have been in the most conservative position. 2.8. Major trends in interest groups On the whole, interest groups do move somewhat more than legislators. Three interest groups present for 10 or more years had a total movement in excess of 0.5 units. While the Liberty Lobby offset a more conservative social position with a more liberal one on the economic dimension, the National Taxpayers Union lost its social liberalism while also becoming more conservative on economic issues. Similarly, the Ripon Society, initially a middle-of-the-road group along with Common Cause, finished by being clearly in the conservative camp. The net results of such changes was to leave the final positions of interest groups more polarized than their initial ones. Large movements for individual legislators are similar to those discussed by Poole and Daniels (1985) from the results of an alternative methodology. The reader is referred to that paper for substantive discussion of individual cases. 2.8.1. Trend as strategic behavior There is an important additional finding, however, concerning the characteristics of legislators who engage in relatively large movements. Having a large trend is negatively related to the degree to which the legislator is fit by the model. Our measure of fit is, as before, the squared correlation between the ratings and the recovered distances. When fit is correlated with the magnitude of average movement per year, we find (for those serving three or more years), [129] 348 a correlation of -0.17 for the Senate and -0.30 for the House. The magnitude of correlation increases when we restrict ourselves to those legislators for whom we have enough years of data to have relatively reliable estimates. For legislators serving over 9 years, the correlation is -0.37 in both houses. Over 14 years, the figures are -0.54 in the Senate and -0.40 in the House. What this means is that legislators with large amounts of spatial mobility tend, to some extent, not to fit the model very well. The lack of fit to the low-dimensional model does not mean that the movement is spurious. For example, Richard Schweiker had the largest movement of any legislator and John Anderson had the second largest movement in the House. Schweiker's conversion to Reagan-style conservatism and Anderson's liberal transition leading to a Presidential candidacy are both well known. Yet Schweiker's R2 was a quite low 0.58 as was Anderson's 0.54. Thus, those senators who fail to fit the mold of maintaining stable reputations in terms of spatial positions also tend to fail to fit the mold of low dimensionality. Both of these failures might be seen as indications of strategic behavior. On the other hand, the whimsical as well as the manipulative also follow this pattern. Poorly fit by the model, William Proxmire rode his Rosinante across a large chunk of our space. In any event, such behavior is not sufficiently pervasive to prevent our model from capturing most of the variation in the ratings. The linear trend term, however, is far from capturing all the intertemporal variation in individual positions. The 82.1 % of variance explained by the model is substantially lower than the 87% achieved by a separate twodimensional scaling for each of the first 22 years (Poole and Daniels, 1985). While we may have omitted some important systematic effects, such as six year senatorial cycle effects (Poole, 1981; Kalt and Zupan, 1984; Dougan and Munger, 1989), we conjecture that most of the additional 5% picked up in the separate scalings is very short run movements and noise. On the whole, trends in the positions of interest organizations and individual politicians do not appear to be of great importance to a spatial model of American politics. The major source of change is likely to be replacement. 10 2.8.2. Replacement effects The effect of replacement may be to lead to substantial changes in the average positions of parties and geographical groupings within the Congress. 11 We now proceed to a brief analysis of some of the more important changes. In terms of the overall means of the two houses of Congress, for both dimensions there is a picture of a stable political system as shown in Figure 5. (We caution, however, that the all-or-none aspect of majority rule may cause small fluctuations in the means to be associated with major changes in policy.) [ 130] 349 0.40 ~- Social Dimension House Economic Dimension -0.20 -0.40 Senate +----------------- - . - .. - - - - - - - - - - - - - - - - - - - ' 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 Figure 5. Overall average locations on the economic and social dimensions. The averages are computed over all members serving in the relevant year. On the economic dimension, there is very little net change in the House, which has had a Democratic majority throughout the period of our data set. Those fluctuations that occur are readily attributable to the electoral fortunes of the two parties. There is a swing to the liberal side following Johnson's landslide in 1964, a conservative rebound after the 1966 midterm elections, a Watergate induced liberal shift, and a Reagan shift at the end. The Senate, institutionally designed to buffer the electorate's fancy via staggered terms, shows virtually no change until 1981. At that point, the massive Republican gains in the 1980 elections resulted in a strong conservative surge that left the House, for the first time in 23 years, more liberal than the Senate. As to the social dimension, which, in any event, has less variation in coordinates than the major, economic dimension (see Table 2), there are only very minor fluctuations, although there is a faint echo of the electoral fluctuations indicated for the House. A very mild long-term trend to liberalism in the Senate was completely offset by the 1980 elections. 2.9. The (once) liberal Senate The immediate source of the historically greater liberalism of the Senate on both dimensions is readily apparent in Figures 6 and 7. On both dimensions, Senate Republicans have always been more liberal than their counterparts in the House. There is very little difference in the positions of House vs. Senate Northern Democrats on both dimensions and of Southern Democrats [ 131 ] 350 0.80 HouseR Sena~ 0.60 _______________________ House~ 0.40 Senate SD -0.20 -0.40 -0.80 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 Figure 6. Overall average locations of Republicans (R), Northern Democrats (ND), and Southern Democrats (SD) on the economic dimension. on the social dimension. At the same time, House Southern Democrats, initially more liberal than their Senate colleagues on economic matters, have recently been more Conservative. Thus, the only consistent and important difference between the two Houses, in terms of the three-party split, concerns the Republicans. This situation would appear to be the result of, grosso modo, the Republicans being gerrymandered into a relatively small number of highly conservative districts in the House while competing in broader constituencies for the Senate. Note, however, that the election of a highly Conservative contingent of freshman Republican senators in 1980 greatly attenuated House-Senate differences while contributing to the strong Conservative shift shown in Figure 5. 2.9.1. Are Southern Democrats more liberal? The other key story told by Figures 6 and 7 concerns Southern Democrats. Southern Democrats who entered the House since passage of the 1965 Voting Rights Act were considerably more liberal than their predecessors (Bullock, 1981). Southern Democrats in the Senate, in contrast, initially more Conservative than their House counterparts, did not become more conservative in the sixties and then became much more liberal in the seventies. That the switch to liberalism was mainly on the economic dimension, is relevant to the view that the switch was induced by the new African-American electorate. We indicated above that minority members of Congress anchored the economic, not [132 ] 351 0.80 SenateSD HouseSD 0.40 0.20 0.00 HouseR Senate R SenateND HouseND -0.20 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 Figure 7. Overall average locations of Republicans (R), Northern Democrats (NO), and Southern Democrats (SO) on the social dimension. the social dimension. After the passage of the Great Society legislation, one could argue that the Northern White liberal agenda turned to other issues and that the new constituency left Southern Democrats free to pursue traditional positions on these issues. At the same time as the Southern Democrats became more liberal on the economic dimension, a conservative trend among Senate Northern Democrats greatly attenuated the gap between the mean positions of the two regional groupings. This result confirms our contention (Poole and Rosenthal, 1984b) that Senate elections have increasingly become polarized, competitive contests nationwide between two increasingly homogeneous parties. 2.9.2. A polarized political system Nonetheless, our previous analysis of polarization in the Senate requires some modification as a result of using a two-dimensional rather than onedimensional model. The basic thrust of our earlier argument was that there was substantial attenuation of party differences in the sixties followed by polarization in the seventies. A similar effect is shown in Figure 8, which combines information on means and dispersions. The mean on the economic dimension is plotted for each party. Also plotted is one standard deviation from the mean in the liberal direction for the RepUblicans and in the conservative direction for the Democrats. This typically creates two overlapping bands. The extent of the overlap indicates the attenuation of party differences. [133 ] 352 Cons. 0.80 0.60 0.40 0.20 Overlap 0.00 -0.20 -0.40 Lib. -0.60 -0.80 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 - - Rep. Mean --- R Mean - One SD --- Dem. Mean -------6- D Mean + One SD Figure 8. Senate: Party dispersion on the economic dimension. Until the late 1970s, Republicans who were one standard deviation or more liberal than the party mean overlapped Democrats who were one standard deviation or move conservative than the party mean. Cons. 0.80 0.60 0.40 0.20 0.00 -0.20 -0.40 Lib. -0.60 -0.80 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 --- Rep. Mean --- R Mean - One SD ---- Dem. Mean -------6- D Mean + One SD Figure 9. House: Party dispersion on the economic dimension. [134 ] 353 In comparison to our earlier results, attenuation does occur in the sixties, but the overlap is substantially less. Moreover, by the late seventies, the overlap is still greater than it was in the early sixties. The basic story is saved, however, by the 1981 data, not present in our earlier analysis. Interest group ratings for 1981 show, for the first time, separation rather than overlap in the two parties. As with the other figures, a similar plot for the social dimension would show similar, but less pronounced, results. The divergence between these results and Poole and Rosenthal (1984b) is clarified by considering that our earlier work had compressed the twodimensional story into one dimension. During the early sixties, the social dimension was likely to have dominated, for a short period, the economic dimension in saliency on the national agenda. On the social dimension, there is less difference between the parties as a result of the extreme conservatism of Southern Democrats. Thus, when the space is forced into one dimension, party differences are greatly attenuated in the sixties. In any event, the tendency to polarization identified here on the basis of data that does not extend beyond 1981, is confirmed by the NOMINATE analysis of roll call votes for 1947 through 1996 in McCarty, Poole, and Rosenthal (1997). Data from the House, presented in Figure 9, shows a pattern similar to that of the Senate. Party differences are least in the mid-sixties and early seventies, and the greatest separation occurs in 1981. In contrast to the Senate, however, the House has always looked like a purer two-party system since there has never been any overlap. 3. Conclusion We believe that a body of empirical research of sufficient size has accumulated to put to rest the classical spatial theory of convergent equilibrium as a model that is at all descriptive of American national politics. There are substantial differences in the policy positions of Democrats vs. Republicans who alternatively or concurrently represent the same constituency, be it Presidential or Congressional (Poole and Rosenthal, 1984a,b). There is no indication of convergent equilibrium. Moreover, the results presented here suggest that there is little movement or spatial adaptation of the type implicit in Hotelling competition. 12 Once elected, candidates appear to be highly constrained by reputational effects. Bernhardt and Ingberman (1985) have developed a model with divergent equilibrium based on the notion of incumbency advantage. In their model, voters perceive incumbents as less risky than challengers. Movement by the incumbent increases his riskiness. Given that the challenger is a Stackelberg follower in spatial competition, no incumbent will move so [ 135] 354 much that his riskiness exceeds the challenger's. Alesina (1988), following earlier empirical work by Hibbs (1977) and theoretical work by Wittman (1983) and Calvert (1985), argues that parties have policy preferences that are known to the voters. In the Alesina framework, parties do not converge because their campaign promises are not credible to voters who know a party will in fact implement its preferences once in power. Further work on both the Bernhardt-Ingberman and Alesina paradigms appears merited by our empirical results. Our finding of low dimensionality is consistent with recent theoretical work on the spatial model which posits a small number of underlying dimensions that in turn produce individuals' issue positions (Ordeshook, 1976; Hinich and Pollard, 1981; Enelow and Hinich, 1984; Hinich and Munger, 1994). In particular, Hinich and Munger argue that low dimensional spaces are inevitable in the presence of political conflict. However, Hinich and Munger do not address convergence, except to predict that political parties will have difficulty in moving ideologically because of problems with credibility. Integrating, with a large dose of speculation, the results in this paper with those of Poole and Rosenthal (1984a), we would suggest that interest groups are more polarized than legislators who in turn are likely to be more polarized than the electorate. The electorate would have a unimodal distribution with most ideal points interior to those of most legislators. In campaigns, as expressed in voter perceptions, Presidential candidates would also be largely on the periphery of the electorate. But elected Presidents, in their support scores, while still highly differentiated along party lines, may be less polarized. Their campaign positions must be compromised as part of the policy-making process in Washington. The basis for the picture we have painted may well lie in the free-rider problem in political participation. Only relative extremists may be sufficiently committed (read have enough "citizen duty") or have enough at stake to participate as money givers and doorbell ringers. In Ingberman (1985), the presence of such polarized contributors is assumed. Candidates can use these contributions to reduce their riskiness. Consequently, incumbents with large contributions have greater freedom of movement. On the other hand, incumbent movement may be constrained by the need to appeal to contributors. The change that television advertising has made in the "production function" of political campaigns and the need to appeal to contributors may well have resulted in the increasing polarization we have observed. Why do presidents appear to be, although polarized, somewhat closer to the center of the spectrum than legislators? We have suggested that the policy process provides a possible explanation. But there are alternative [136 ] 355 theoretical explanations. First, in the Bernbardt-Ingberman (1985) model, convergent equilibrium does occur in the case of truly open seats. At least every eight years, the presidential election is a relatively open contest. In contrast, as Ingberman (1985) shows, challengers will, ceteris paribus, take increasingly extreme positions as the incumbent's seniority grows. Given the strength of incumbency in Congress, the rare successful challengers are likely to be relatively extreme. Second, in Congress, particularly the House, recognition rather than policy may well be the major aspect of campaigning. Because Congressional candidates are so much less well known than candidates for the Presidency, they may well optimize by being mainly in tune with contributors' preferences. Notes I. See, however, Poole and Rosenthal (1984a) for some limited analysis of changing spatial positions in two National Election Study panels. 2. See Snyder (1992) and Poole and Rosenthal (1997: Ch. 8). A related problem is that there is overlap in the roll calls used by the interest groups. However, the overlap is not overwhelming. Poole and Daniels (1985) report that over 30% of all roll calls in 1979 were used by at least one interest group, indicating that coverage by interest groups in our dataset is broad. 3. In the actual unfolding, we used Legendre polynomials rather than the t shown in the text. The Legendre polynomials differ from this t only for quadratics and higher degrees. We use Legendre polynomials because they are orthogonal on the interval [-I, + I] (Hinich and Roll, 1982; Beck, 1983). This property allows us to pick up as much variance as possible with the linear term before we estimate the quadratic term and so on. Because a specification higher than linear did not greatly improve results, most readers will not need to become concerned with the Legendre method. 4. Obviously, estimating the polynomial models runs into a problem for legislators with brief periods of service. We estimate linear terms only for those serving three or more years; quadratic only for four or more; and cubic only for five or more. These estimates will be imprecise (see below) for those legislators with very few years of service. However, since we have argued elsewhere (Poole and Rosenthal, 1991, 1997) that Congress can basically be represented in terms of a stable, two-dimensional model with the first dimension being much more important than the second, we deliberately sought to err on the side of overestimating trends in individual positions. 5. Because some legislators and interest groups are present for time periods too brief to permit estimation of the polynomial parameters, the actual number of parameters estimated is somewhat less than that given in the following expression. 6. A similar result for a dynamic NOMINATE estimation is presented in Poole and Rosenthal (1997: 73). 7. The second dimension coordinate was regressed on the first dimension coordinate for the six presidential support scores. 8. For an analysis of this trend through the \o4th Congress, see McCarty, Poole, and Rosenthal (\ 997). [ 137] 356 9. Again, for an analysis of polarization through the 100th Congress, see McCarty, Poole and Rosenthal (1997). 10. See Poole and Rosenthal (1984b, 1985b) for a discussion of the replacement literature. 11. See Poole and Rosenthal (1987) for analysis of more detailed regional breakdowns in the context of a unidimensional model. 12. Similar results at the level of individual states can be found in Erikson, Wright, and McIver (1993). 13. See Poole (1984, 1990) for a detailed discussion ofthe squared error loss function for this problem. References Alesina, A. (1988). Credibility and policy convergence in a two-party system with rational voters. American Economic Review 78: 796-806. Beck, N. (1983). Time-varying parameter regression models. American Journal of Political Science 27: 557-600. Bernhardt, M. D. and Ingberrnan, D. (1985). Candidate reputations and the incumbency effect. Journal of Public Economics 27: 47-67. Bullock, C. S. III. (1981). Congressional voting and the 1982 congressional elections. Journal of Politics 45: 767-770. Cahoon, L. S., Melvin, J., Hinich, MJ. and Ordeshook, P. C. (1978). A statistical multidimensional scaling model based on the spational theory of voting. In P. C. Wang (Ed.), Graphical representation of multidimensional data, 243-278. New York: Academic Press. Calvert, R. L. (1985). Robustness of the multidimensional voting model: Candidates' motivations, uncertainty, and convergence. American Journal of Political Science 26: 69-95. Defays, D. (1978). A short note on a method of seriation. British Journal of Mathematical and Statistical Psychology 31: 49-53. Dougan, W. R. and Munger, M. C. (1989). The rationality of ideology. Journal of Law and Economics 32: 119-143. Downs, A. (1957). An economic theory of democracy. New York: Harper. Enelow, J. M. and Hinich, M. J. (1984). The spatial theory of voting: An introduction. New York: Cambridge University Press. Erikson, R. S., Wright, G. C. and McIver, J. P. (1993). Statehouse democracy: Public opinion and policy in the American States. New York: Cambridge University Press. Hibbs, D. (1977). Political parties and macroeconomic policy. American Political Science Review 71: 1467-1487. Hinich, M. (1978). Some evidence on non-voting in the spatial theory of electoral competition. Public Choice 33: 83-102. Hinich, M. J. and Munger, M. I. (1994). Ideology and the theory of political choice. Ann Arbor: The University of Michigan Press. Hinich, M. J. and Pollard, W. (1981). A new approach to the spatial theory of electoral competition. American Journal of Political Science 25: 323-341. Hinich, M. J. and Roll, R. (1981). Measuring nonstationarity in the parameters of the market model. Research in Finance 3: I-51. Hubert, L. and Arabie, P. (1986). Unidimensional scaling and combinatorial optimization. In: de Leeuw et al. (Eds.), Multidimensional data analysis. Leiden: DSWO Press. [138 ] 357 Ingberman, D. (1985). Spatial competition with imperfectly informed voters. Mimeo. University of Pennsylvania. Kalt, J. and Zupan, M. (1984). Capture and ideology in the economic theory of politics. American Economic Review 74: 279-300. Kernell, S. (1973). Is the Senate more liberal than the House? Journal of Politics 35: 332-363. Khrebiel, K. and Rivers, D. (1985). Congressional roll call strategies: An application of a new test to minimum wage legislation. Paper presented at the 1985 annual meeting of the American Political Science Association. Ladha, K. K. (1994). A spatial model of legislative voting with perceptual error. Public Choice 68: 151-174. McCarty, N.M., Poole, K. T. and Rosenthal, H. (\997). Income redistribution and the realignment of American Politics. Washington, DC: AEI Press. Ordeshook, P. C. (1976). The spatial theory of elections: A review and a critique. In I. Budge, I. Crewe, and D. Farlie (Eds.), Party identification and beyond. New York: Wiley. Poole, K. T. (1981). Dimensions of interest group evaluation of the U.S. Senate, 1969-1978. American Journal of Political-Science 25: 49-67. Poole, K. T. (\ 984). Least squares metric, unidimensional unfolding. Psychometrika 49: 311323. Poole, K. T. (1990). Least squares metric, unidimensional scaling of multivariate linear models. Psychometrika 55: 123-149. Poole, K. T. and Daniels, R. S. (1985). Ideology, party, and voting in the U.S. Congress, 1959-1980. American Political Science Review 79: 373-399. Poole, K. T. and Rosenthal, H. (\ 984a). U.S. presidential elections: 1968-1980. A spatial analysis. American Journal of Political Science 28: 282-312. Poole, K. T. and Rosenthal, H. (I 984b ). The polarization of American politics. Journal ()f Politics 46: \062-\079. Poole, K. T. and Rosenthal, H. (1985a). A spatial model for legislative roll call analysis. American Journal of Political Science 29: 357-384. Poole, K. T. and Rosenthal, H. (l985b). The unidimensional Congress. Mimeo. Carnegie Mellon University. Poole, K. T. and Rosenthal, H. (1987). The regional realignment of Congress, 1919-1984. In P. Galderisi and R. Simmons (Eds.), Politics in the Intermountain West: Forerunner to realignment? Boulder CO: Westview. Poole, K. T. and Rosenthal, H. (1991). Patterns of congressional voting. American Journal of Political Science 35: 228-278. Poole, K. T. and Rosenthal, H. (1997). Congress: A political-economic history of roll call voting. New York: Oxford University Press. Rabinowitz, G. (1976). A procedure for ordering object pairs consistent with the multidimensional unfolding model. Psychometrika 45: 349-373. Rivers, D. (\ 987). Inconsistency of least squares unfolding. Paper presented at the Political Methodology Meetings, Durnham, NC. Romer, T. and Rosenthal, H. (1987). Modern political economy and the study of regulation. In E.E. Bailey (Ed.), Public regulation: New perspectives on institutions and politics. Cambridge MA: M.LT. Press. Snyder, 1. M. Jr. (\ 992). Artificial extremism in interest group ratings. Legislative Studies Quarterly 17: 319-345. Wittman, D. (1983). Candidate motivation: A synthesis of alternatives. American Political Science Review 77: 142-157. [ 139] 358 Appendix: Bootstrap experiments The statistical properties of the estimated parameters for the squared error loss function used here (the sum of the squared differences between the observed distances and the reproduced distances) are not well understood.B Rivers (1987) has shown that, under certain conditions in one dimension, the estimates are not consistent. However, in one dimension, the solution can be shown to reduce to a combinatorial optimization problem (Defays, 1978; Hubert and Arabie, 1986; Poole, 1990) in which a rank ordering determines all the parameters simultaneously. This is a much different set of conditions than Rivers employs thus leaving the statistical properties of the parameters unresolved. Monte Carlo evidence shown in Poole (1984; 1990) shows no bias in the recovery of the coordinates in one dimension. Consequently, in order to obtain some insight as to the accuracy of our recovery, we performed a bootstrapping analysis for the two-dimension version of our unfolding procedure. Bootstrapping analysis for the one-dimensional version is shown in Poole (1990). We based our analysis on our recovered coordinates for s = 2 and m = 1 (legislators free to move on both dimensions). To reduce the use of computer time, however, we restricted the legislators to the 261 senators. Using the recovered coordinates and trend parameters, we computed distances between the interest groups and the legislators. Distances were generated only for those years for which the senator-interest group pair bot appeared in the actual data. These "true" distances were then converted to ratings by multiplying them by 50 and subtracting the resulting quantity from 100. These "true" ratings were then perturbed by adding normally distributed error at low, medium and high levels (a = 5, 10, 15 respectively). The medium errorlevel corresponds closely to the actual level of error in the interest group ratings. Negative ratings were truncated to 0 and ratings greater than 100 were truncated to 100. This procedure was employed to simulate some of the factors that might perturb the actual data, such as limited selection of roll calls by the interest groups, omitted dimensions specific to individuals, etc. The perturbed data were then submitted to our unfolding algorithm. At each level of error, we conducted 25 replications. From these 25 replications we were able to compute a root mean square error of recovery for each parameter. A summary of key results is presented in Tables A-I and A-2. Constant parameters Positions on the main, liberal-conservative dimension appear to be recovered with great accuracy. At medium levels of error, the root mean square errors average under 0.09 while the span of the dimension is over 2 units. In contrast, the second dimension is recovered with less accuracy. The span of the second dimension is under 1.5 units, but at medium error levels root mean square errors average just under 0.2 for senators and 0.3 for interest groups. While the inaccuracy in recovery is still moderate with respect to the length of the dimension, discussion of individual positions on this dimension should be taken with caution. Our translated coordinates are linear combinations of the recovered coordinates. Since our algorithm generates minimal [140 ] 359 Table A-I. Bootstrap results: Average root mean square errors, recovered constant parameters Error level Senators (XikO, k = I, 2) I st recovered dimension All N= 261 3 or more years N = 213 2nd recovered dimension All N = 261 3 or more years N = 213 Low Medium High .\09 .197 .306 .045 .089 .190 .044 .082 .179 .\09 .187 .265 Error level Interest groups (ZjkO, k = I, 2) I st recovered dimension 2nd recovered dimension All N= 59 3 or more All N = 59 3 or more years N =43 years N = 43 Low Medium High .042 .074 .218 .040 .073 .2\0 .147 .280 .3\0 .150 .262 .283 Note. All entries in table based on 25 replications, each using an independent set of normal errors. covariance between the two recovered dimensions, it follows that the implicit errors on our economic dimension are somewhat greater than those for the first recovered dimension while the errors on the social dimension are somewhat less than those on the second recovered dimension. One apparent anomaly in Table A-I is that interest groups are recovered less accurately on the second dimension in spite of the fact that the interest group recovery appears to be based on a larger effective sample size (261 versus 59). Two explanations can be suggested. First, sample size depends also on how long the interest group was in the data set. In tum, our time series for interest groups tend to be less lengthy than those for senators. Of the 261 senators, 110 were in the data set for over 10 years as against only 16 of 59 "interest groups". Moreover, an interest group in the sample for only a few years is basically rating a stable set of senators whereas a senator in the data set for only a few years is being rated by a more rapidly changing set of interest groups. Thus, holding length in the data set constant, more information tends to be gained about senators than about interest groups. An indication of how length of time in the data set affects recovery is shown in the table. The root mean square errors are less for those interest groups and senators in the sample over two years than they are for the entire data set. [141] 360 Table A-2. Bootstrap results: average magnitude and root mean square errors, linear trend per year Error level Senators (Xikl, k = 1,2) I st recovered dimension 3-9 years 10 or more N=213 years N = 110 2nd recovered dimension 3-9 years 10 or more N=213 years N = 110 Low magnitude .080 .021 .028 .015 RootMSE .059 .019 .046 .014 Medium magnitude .064 .017 .027 .010 RootMSE .106 .031 .070 .021 High magnitude .062 .174 .018 .026 .007 .046 .099 .028 RootMSE Error level Interest groups (Zjkl, k = 1, 2) I st recovered dimension 2nd recovered dimension 3-9 years 10 or more 3-9 years 10 or more years N = 16 years N = 16 N=27 N=27 Low magnitude .106 .050 .056 .008 .112 Medium magnitude .093 .020 .064 .013 .045 .041 .009 RootMSE .161 .048 .067 .021 High magnitude .068 .143 .018 .023 .009 .064 .069 .025 Root MSE RootMSE Note. All entries in table based on 25 replicas, each using an independent set of normal errors. Second, we note that in the actual recovery, interest groups are clustered at the two poles of the main dimension whereas legislators are more spread out. These distributional differences may affect recovery. Trend The results shown in Table A-2 support our claim that individual time trends are of minimal importance. The average magnitude of the yearly change in position is never more than twice the average root mean square error except for interest groups serving 10 or more years in the simulations with a low degree of error. That the only reasonably precise estimates concern interest groups supports our contention that spatial mobility is greater among interest groups than among legislators. As a result of our using recovered values in the bootstrap, the average magnitudes are systematically larger for interest groups and senators present 3 to 9 years than for [142 ] 361 those present for longer periods. However, the root mean square errors of recovery for these parameters is so large that the recovered values appear highly dubious when trend estimates are based on only a few years of data. Finally, note that the discussion in this Appendix applies to individual parameters. Estimated positions for aggregates, such as Southern Democrats, will be far more stable. [143 ] Public Choice 97: 363-382, 1998. © 1998 Kluwer Academic Publishers. 363 British parties and spatial competition: Dimensions of party evaluation in the 1992 election JAMES W. ENDERSBY' & STEVEN E. GALATAS 2 , Department of Political Science, University of Missouri, 113 Professional Building, Columbia, MO 65211, U.S.A.; Department of Political Science, University of Missouri, 113 Professional Building, Columbia, MO 65211, U.S.A. Abstract. Scholars of British politics traditionally characterize the electorate in terms of partisanship and social class. This paper suggests that ideology and issue preferences also enter into voter perceptions of British political parties and leadership. Using data from the 1992 British Election Study, the paper analyzes the factors that contribute to individual voters' perceptions of the Conservative and Labour parties. The 1992 election saw the major parties move toward the ideological center of British voters. Perceptions of political parties are found to be multidimensional and issue-oriented. A spatial model incorporating issue preferences and perceptions of party positions proves both empirically and theoretically richer than simple models of partisanship. The analysis of British voters complements earlier applications of the general spatial model in the context of the United States. 1. Introduction This paper examines the role of issues in the 1992 British general election. In particular, the paper develops a spatial model of electoral choice in which a citizen chooses to cast a ballot for the party nearest to personal preferences on a range of policy dimensions. The results indicate that ideological preferences are significant factors influencing the choices of voters in the 1992 election. Discussions of its application in British politics typically equate spatial theory with unidimensional Downsian analysis and often reject the usefulness of spatial models for understanding British elections. Electoral behavior is more often linked to party affiliation and social class orientation. Compared to parties in the United States, British parties are strong, and voters can be characterized by party loyalty. Literature on the 1992 election point toward a reestablishment of class voting in British politics. If spatial models of political choice trace their origins to the works of Downs (1957) and Black (1958), then it seems appropriate that the British electorate should serve as a case study for modern tests of spatial assumptions. Both Black (for committees) and Downs (for mass politics) based much of their work on an understanding of British politics. But they restrict their [145 ] 364 spatial analogy to a single dimension. On a single policy dimension defined by single-peaked preferences, parties should seek the equilibrium at the position of the median voter. More recent scholarship has shown that an equilibrium need not exist when preferences are nonseparable or the number of policy dimensions grows to two or more (Hinich, 1977). Improvement in the Downsian spatial model have extended spatial theory to a multidimensional policy space (Davis and Hinich, 1966; Hinich and Pollard, 1981), and a number of empirical tests have confirmed the utility of the spatial model for analysis of mass elections (Enelow and Hinich, 1984). Although this paper focuses on the party positions on particular issues and policies, we do not expect British voters to weigh policy alternatives in a void without a unifying ideology (Hinich and Munger, 1994). Instead, evidence we uncover suggests that British voters do have ideological preferences beyond group and party affiliation and use information to evaluate which political party is more likely to satisfy their preferences. 2. Class and party in Britain Historically, the study of voting behavior in Great Britain grew from similar lines of work among American political scientists. In reaction to existing scholarship, with its emphasis on the uniqueness of the British political system and on the role of ideals in the evolution of British politics (Gamble, 1990), a new wave of political scientists focused attention on the behavior of individuals and groups. With the rise of this new school, scholarship also began to shift slowly toward comparisons between British and non-British politics. Behavioralists, equipped with early empirical models of Britain, quickly latched onto social group theories of political behavior. Britain came to be seen as a political system dominated by two social classes: working class and middle class. The two major political parties, Conservative and Labour, reflected the interests of these two classes. The model seemed simple and efficient: the working class voted for Labour and the middle class for Conservative. The largest "natural" class won election because voters selected a party in an election based upon social class, which in tum reflected economic interests (Alford, 1963; Pulzer, 1967; Butler and Stokes, 1969). While the existing social classes disagreed on the extent but not the existence of the welfare state in modem Britain, changes in electoral behavior reflected either the resurgence of Celtic nationalism in Scotland and Wales (Beer, 1982), or appeared to be associated with the relative closeness of the two major parties and the degree of consensus (Butler and Stokes, 1969). [146 ] 365 Sparked by the rise in election support for third parties, including the Liberal party across Great Britain, the Plaid Cymru in Wales, and the Scottish National party in Scotland, an enormous debate developed among scholars of British voting behavior. The existing model of voting behavior fell into disfavor by the mid-1970s as mounting evidence suggested that class voting had diminished in importance (Crewe et aI., 1977). Parsimonius models of voting behavior no longer required social class, according to Franklin and Mughan (1978). Further signs of the collapse of class voting appeared in the 1983 election when less than half the electorate supported the party of their social class (Butler and Kavanagh, 1983). Such attacks in the literature continued into the 1990s (Franklin, 1992). As early evidence mounted against the social class model, other studies began to follow the Michigan model, examining the degree of partisan identification as an explanation for voting behavior. Alt, Sarlvik, and Crewe (1976) and Crewe et al. (1977) launched this 'Decade of Dealignment' debate, suggesting that fewer working class voters identified with the Labour party, while fewer middle class voters saw themselves as Conservative. Sarlvik and Crewe (1983) state that partisan identification is important for voting behavior but increases in the volatility of partisanship occurs; this pattern finds confirmation in other research (LeDuc, 1985; Crewe, 1983; Clarke and Stewart, 1984). However, other evidence suggested that long-term stability existed in partisan identification despite short-term volatility (Miller et aI., 1986). Thus, scholarship on voting behavior in Great Britain became locked in a battle between supporters of the social class model and partisan identification model. Attempts to resurrect the sociological models of voting behavior occurred. Heath, Jowell, and Curtice (1985) conclude that social class forms the basis of political support but manifest itself through political ideology. However, as an alternative to the pure class model, McAllister and Rose (1984) suggest that territorial socialization within the various nations and regions of the United Kingdom influences forms the basis of voting. A final, innovative attempt to rescue class voting compares two types of class models: latent class models rely solely on traditional concepts of class, comparing actual proportions of each class that vote for each party, and associational class models, incorporating a Downsian unidimensional model in which class replaces issue as the underlying basis of the model (Weakliem, 1995). Statistical tests of the two models suggest that the associational model fits the data more accurately and that a slight trend toward class de alignment occurs. While much of the debate over this class-party nexus sits at the heart of voting behavior in Britain, alternatives to these two approaches have also been posited. While the above approaches might be labeled as primary or longterm factor models, the alternative models focus attention on short-term or [147 ] 366 secondary consideration. The schools that advocate these approaches focus on the role of ideology, the influence of party leader images, and the role of issues in motivating electoral choice. Among these alternatives, the role and influence of issues fall into two categories. Single-issue studies investigate the role of an isolated issue, unique to one election, as the impetus for vote choice. As an example, King (1982) suggests the question of unilateral nuclear disarmament, and the Labour party's position on this issue, as a possibility in the 1979 election. Likewise, political events like the Falklands War may disrupt normal partisan identification to produce shifts in election outcomes (Mishler, Hoskin, and Fitzgerald, 1989). However, the second trend emphasizes the influence of issues across time. General analysis suggests that issues, including the state of the economy, unemployment, and inflation, become more prevalent as class declines as the basis for vote choice (Franklin, 1985; Denver, 1994). Furthermore, Denver and Hands (1990) discover that young voters just entering the electoral arena choose a party as a result of opinions on issues rather than long-term factors like partisan identification and social class. Specific issues also play an important role in vote choice; time-series analysis suggests that perceived priorities of parties on macroeconomic issues is important (Clarke et aI., 1986). Some scholars tum toward leadership issues as an explanation for vote choice, absent long-term factors like class voting. Even in parliamentary systems, party leaders influence the distribution of votes (Bean and Mughan, 1989). Whether the magnitude of leadership effects possess the power to alter the electoral outcome remains uncertain (Graetz and McAllister, 1987; Clarke and Stewart, 1992). In addition, ideology has been suggested as an explanation for vote choice and party electoral strategy. Beer (1982) discusses the political upheavals in the 1970s as related to the shift in ideology from traditional collectivism to individualism of the liberal variety. Furthermore, some authors attempt to relate the rise of Thatcherism to a shift in ideology, breaking the growth of ideological hostility toward industrial success and replacing this line of though with a new pro-business culture (Wiener, 1981; Barnett, 1986). However, the role of ideology has been discounted as a component of the party-class nexus (Heath, Jowell, and Curtice, 1985). In addition, many scholars claim the Conservative party contains many divergent, incompatible tendencies and lacks a coherent, dominant ideology (Whitely et aI., 1994; Vincent, 1994). [148] 367 3. Voting in Britain: Irrational choice? Attempts at explaining British voting behavior using spatial concepts have been made. For example, Budge and Farlie (1977) produce a "Synthesis model" of social class, partisan identification and rational choice. An examination of the role of issues and the dimension of party programs contends that cityblock spatial models contain superior explanatory power over Euclidean distance measures; however, issue and ideological spaces contain cross-national explanatory power. Budge (1994) tests several models of voting behavior, including spatial models, cross-nationally. Concluding that parties do not necessarily converge at election time, Budge also asserts that uncertainty and ideology provide incentives for the failure to converge at the median position. Furthermore, Norris (1994) builds a Downsian model of British voting behavior using a study of politicians and party members for the major parties to test the degree of polarization, extremism, factionalism, and fragmentation present in the Labour party in 1992. Finally, spatial analysis does not necessarily conflict with class-based approaches to the study of voting behavior as suggested above (Weakliem, 1995). Nevertheless, rational choice models of voting behavior, especially Downsian models, continue to be rejected by many scholars of British politics. According to Heath, Jowell, and Curtice (1985), rational choice models fail for three reasons: the electorate lacks comprehension of the policy positions of the parties, voters retain ideology and social class as a guide to voting thereby rejecting pure self-interest motivations, and parties spurn vote-maximization strategies, failing to convince the electorate that the party's position changes. For example, using issue scales from 1983 BES, the party closest to median voter on unemployment and inflation was Labour, then Alliance, then Conservative. Unfortunately, many scholars of British politics associate rational choice models of voting behavior primarily with Downsian unidimensional models (Budge and Farlie, 1977; Heath et aI., 1985; Whitely et ai. 1994). Reduction of a multidimensional issue space to the unidimensional Downsian model may produce a number of problems. Without separability of issues, the median voter result does not hold, and equilibrium may be impossible. Norris (1996: Ch. 7), for instance, conducts a content analysis of Labour and Conservative party manifestos over several decades and notes a number of violations of traditional Downsian assumptions. The British parties have exhibited considerable volatility on some issues, even leaping over one another on some policy dimensions and alternating which parties are on the right and on the left. Although Downsian indicates a relationship between issues and ideology collapsing into a unidimensional model (Downs, 1957: 100), several empirical studies of voting behavior in the United States and [149 ] 368 elsewhere suggest a multidimensional issue space appears appropriate. Budge (1993) and Klingemann et al. (1995) analyze political conflict within western democracies and find that elections simplify to a limited number of issue dimensions, often just two. Another arena for criticism of the application of rational choice and spatial models in British politics focuses on the role of issues in such models. For example, Whitely et al. (1994: 202) believe that: "Spatial models fail because they focus exclusively on issues. Voters do not (nor should they) rely exclusively on issues in an uncertain world, where leadership reputations may be a better guide to policy outcomes than pre-election issue positions". However, they also write that when facing a rational electorate, a party strives for support from majorities on all issues and seeks policies that converge towards the center of the space and/or the median voter. Given the emphasis of the social class and partisan identification approaches to voting behavior that dominate much of the scholarship on British voting behavior, another area for criticism focuses on the absence of long-term forces in most rational choice models. According to Budge and Farlie (1977), group loyalties inhibit movements of leaders in the policy space, while electors rely on group habits and party loyalty to produce vote choice and positions on issues. Other critiques center on the basic assumptions of rational choice models and the relationship between the assumptions and reality. Claiming that followers of economic approaches to voting recognize the lack of realism in their assumptions, Budge and Farlie (1977: 512) say that the problems of economic models of voting behavior, demonstrated in election studies, are that rational choice specialists move from simple models to complex models and that the process of relaxing assumptions destroys the mathematical basis resulting in general, misleading, and paradoxical conclusions. Budge et al. (1983: 27) review rational choice models: "The assumptions which support formal voting models are, however, extremely restrictive and unlikely to apply directly to real life". Yet, they develop a new synthesis to overcome the perceived inconsistencies of pure rational choice approaches while retaining a market approach complete with competition, transaction costs, consumers, and producers. 4. Parties, issues, and voters in 1992 The 1992 British General Election presents an interesting opportunity to examine the ability of issue-based spatial models of voting behavior in Great Britain. First, the election features an apparent reversion to the class-party nexus, or the best approximation since the 1970s (Weakliem, 1995); this situ- [150] 369 ation seems paradoxically as a point to test issue-based approaches. Second, the election featured the continuation of Conservative erosion in the Celtic periphery and North England. Finally, the election occurs at a time of growing methodological and theoretical diversity in the study of British politics (Gamble, 1990). The 1992 British General Election Study consists of a nationally representative, cross-sectional sample contacted shortly after the election (Heath et aI., 1993). A total of 3,534 personal interviews were conducted. In order to better represent Scotland the 1992 study oversamples Scots and requires that individual responses be weighted to produce the nationally representative sample. Throughout this analysis, data are weighted. The 1992 British Election Studies (BES) provides a battery of questions designed to elicit respondent preferences over a variety of issues. Like most surveys of public opinion, however, respondents are offered few opportunities to make judgments concerning the issue orientations of the political parties. Embedded in the questionnaire are several scales which allow both selfplacement of the respondents as well as the perceived placement of the political parties. Respondents identify not only their own ideal positions on a unidimensional policy scale but also where they would locate the Conservatives and Labour (as well as the Liberal Democrats and, for Scotland only, the Scottish Nationals). A respondent in the 1992 BES is shown a left-right scale marked with the letters ranging from A to K. A particular policy alternative characterizes each endpoint, for instance, the European Community question ranges from (A) Unite fully with Ee to (K) Protect independence (the Appendix lists all of the 1992 BES scales). An interviewer asks the respondent "Which [ofthe eleven points on the scale] comes closest to your own views?" A respondent chooses one of the eleven letters or may voluntarily opt for left of A or right of K (though these extreme placements are unusual). The following two questions ask the respondent "Which [letter/point] comes closest to the views of the Conservative Party?" and "Which comes closest to the views of the Labour Party?" The 1992 British survey employs a split-sample design. The survey includes a total of eight issue scales, but not all questions are posed to all respondents. Half of the British sample considers five policy areas - unemployment/inflation, taxation/services, nationalisation/privatisation, redistribution of wealth, and the European Community. The other half of the sample evaluates three policy areas - defence, welfare, and women's rights. For these three scales, the letters range from A to G producing shorter, seven-point scales. There are other important differences between the two samples. The survey instrument for the first half of the sample asks respondents "When you were deciding about voting, how important was this issue to you?" for each [ 151 ] 370 of the five policy areas. Voters could respond one of four ways: Extremely important, Important, Not very important, or Not at all important. The second half of the sample did not evaluate the importance of the three-issue scales. The full sample received identical questions pertaining to party affiliation, voting behavior,l and social class. The procedure utilized here transforms the survey responses to policy distances suitable for a test of the spatial model. The respondent's self- and party-placement on the alphabetic orderings are translated from letters to numbers. Thus, the A to K interval corresponds to a numerical scale from 1 to 11, and the A to G interval corresponds to a 1 to 7 scale. Extremist placements, those to the left of A or to the right of K or G, occur infrequently and are simply recoded to the appropriate end of the eleven- or seven-point scale. Differences between self-placement and location of a particular party are then units counting the numbers of ticks separating the two. Thus, the distance measure is equivalent to a traditional, numbered interval scale. Table 1 reports average locations for self- and perceived party-placement for all those responding to the surveys. The seven-point issue scales of the alternative questionnaire are transformed to eleven-point scales for purposes of comparison2 (the actual values on the seven-point scales are used in the analysis below). In addition, the table identifies the percentage of those responding who evaluated the issue as either 'extremely important' or 'important'. The majority of respondents considered issues pertaining to aggregate economic conditions as more important policy concerns (approximately one-third rated the unemployment-inflation and taxation-services issues as extremely important). The indicators of preferences for the direction of the national economy are important to slightly more than a third of the electorate (around 10% consider each of these goals extremely important). On all of the issue scales, Labour lies to the left of the Tories. The mean perceived positions of the two parties differ by less than a unit on only two policy dimensions, and on these two dimensions both parties are located to one side of the mean voter. The average Briton is situated in the middle of the eleven-point European Community scale at 6.0; the Tories fall slightly to the left (5.6) with Labour a bit further out (5.2). On Women's Rights, the mean among respondents is far to the left at 2.8; the Labour Party places at the right at 3.7 and the Conservatives further at 4.1. On the other scales, the two parties surround the average British voter. The Conservative party is placed nearer the mean respondent on two scales - on the European question and on defense. Labour locates closer to the average voter on the other six scales. [152 ] 371 Table 1. Mean placement of respondent and parties on seven-issue scales (British National Election Study, 1992) Labour Issue Respondent party Conservative party importance Unemployment/inflation 3.56 6.44 2.98 50.6% Taxation/services 4.18 7.03 2.79 56.6% Scale N ationalisation/privati sat ion 5.75 8.35 3.56 32.4% Redistribution of wealth 4.71 7.89 3.03 43.1% European Community 5.98 5.56 5.18 38.7% Defence 5.14 6.12 3.86 Welfare 3.89 7.09 3.23 Women's rights 2.81 4.12 3.71 Mean self- and party-placements are for all respondents. Defence, Welfare, and Women's rights are seven-point scales transformed to eleven-point scales for consistency in this table only. Respondents were not asked to rate the importance of these issues for these three issues. Issue importance denotes the percentage of respondents rating the issue as "impOJ1ant" or "extremely important". 5. A spatial model of the British electorate Our contention is that party positions on issues are important to the contemporary British electorate and that discussions of party and class voting and the unidimensionality of British partisanship mixes the unifying nature of political ideology with preferences on particular issues. Moreover, we contend that voter preferences on issues have an independent effect on many voters beyond the influence of party and class differences. Our goal, then, is to test whether relative placements of respondents and of perceived Labour and Conservative party locations on issue scales are related to vote choice in the 1992 election. We evaluate the effect of relative placements on issue scales independently as well as when the respondent's party affiliation and class are controlled. Unlike American style political systems involving the distribution of power among branches of government, the placement of the ruling (Conservative) party essentially is the placement for the location of parties; this analysis does not attempt to derive estimated party locations based on generalized public perceptions. Instead, we compare each respondent's preferred position on a series of unidimensional policy scales and with that same respondent's perceptions of where the two major parties are positioned on those scales. We assume that issue preferences can be measured as a function of simple linear distance on the policy scales. In particular, for a respondent, i, and [ 1531 372 an issue, j, the perceived relative distance, dij , between the Conservative and Labour parties can be computed as: dij = wij(lxijl - xijl - IXijc - xijl) where Xij is the ilh respondent's self-placement on the jlh policy dimension, Xijc and Xijl are the respondent's perceived placements of the Conservative and Labour parties respectively, and Wij is the value or weight that the ilh respondent places on the jlh issue. Initially, we assume that Wij = 1 for all voters; later, we allow for individual variation among the issue weights. Each dij measures relative party distance. The Conservative policy position is preferred for positive values of dij and Labour is preferred for negative values. The magnitude of the distance variable denotes the intensity of preference for one party's perceived platform on that policy. An individual remains indifferent between the parties if the distance between them is zero. In other words, a voter perceives the value of voting for a political party (or candidate) as inversely related to the policy distance that separates that party from the voter. The voter casts a ballot for the political party which is closest to personal ideological (or issue) preferences. A voter's assessment of the voting decision depends on three items: preferred policies (the voter's ideal points on issue dimensions), the perceived positions of each party, and the weights for each dimension. A logistic regression model tests whether issues - defined as these relative party distances - are factors related to vote choice between the two parties. The simple model uses only distance measures as predictors of party votes. An improved model also controls for other primary factors related to voting in British elections. The full logistic regression equation can be represented as where D is the matrix of policy distances, dij , and Z is the matrix of control variables. 3 If issues matter for British voters, then they should be more likely to cast ballots for parties perceived closer to voters' policy preferences and the estimated coefficients for each policy distance variable should be significantly different from zero and in the direction of the party nearer to ideal preferences. If class, party, and group ties (or other indicators) are responsible for voting decisions, then the estimated coefficients for the issue scales should not be significantly different from zero. The dependent variables predicted by this model is a vote cast for either the Labour or Conservative parties. Since the Tories won, predicted votes are for this category, and positive signs for the estimated parameters suggest a greater likelihood of a vote cast for the Conservative Party. Limiting observations to the two major parties does reduce the size of the 1992 sample in [154 ] 373 areas with strong third party showings. This limitation primarily effects constituencies in which the Liberal Democrats ran well and in Scotland where the Scottish National Party claimed many votes. 4 Restricting valid observations to votes for the two major parties does circumvent a few difficult, though interesting, problems. Abstention (indifference) and alienation (dissatisfaction) are avoided entirely since only citizens who choose to vote are admitted into the model. Likewise, strategic voting among the range of permissible alternatives can not occur (though voters are included who might otherwise support a minor party in a proportional representation system). Since the model requires respondents to place themselves and the major parties on each policy dimension, the issue scales filter the sample as well. Of course, the use of the scales divides the sample into halves. However, the subsample for each model represents the national constituency of Labour and Conservative supporters. Each weighted split sample used in a logistic regression contains approximately eight hundred voters. s The 1992 BES includes measures of political party and class affiliation. This analysis excludes minor parties, so two partisanship dummy variables are created. Each of these party affiliation variables has a value of 1 to indicate an identification withe (Labour or Conservative) Party and a value of o otherwise. 6 Social class proved more difficult to quantify. An index was constructed to identify association within one of six commonly accepted categories representing socioeconomic status. The social class variables ranges from 5 for professionals to 0 for unskilled laborers and corresponds to a traditional ordering of social class in Britain.7 We experimented with a number of other nominal and interval measures for class and social status, but our findings of relationships among the variables remained consistent and this procedure primarily demonstrated the problems with accurately quantifying social status in contemporary Britain. The social class variable should reflect the increasing propensity to vote with the Conservative Party as social status nses. 6. Issues matter Each of the following three tables presents three models, and each estimated logistic regression equation is listed in a separate column. The first model estimates choices of Conservative and Labour voters using only the relative closeness of the two parties on the issue scales. The second model adds the dummy variable indicating affiliation with the Conservative Party and the ordered variable for social class to relative party placement on the issue scales. The third model incorporates a dummy variable for identification with the Labour Party with these other variables. Each table presents the maxi- [155] 374 mum likelihood parameter estimates for each independent variable and their associated asymmetric standard errors. Also included in each table are statistics testing the null hypotheses of dependence and several measures of the goodness of fit for the overall models. The likelihood ratio chi-square statistic tests the joint significance of the explanatory variables. For each parameter estimate, the Wald chi-square statistic tests whether the estimated coefficients for each independent variable is significantly different that zero, controlling for the other independent variables. The tables include two additional indicators to show the measure of association between observed and predicted responses. The gamma statistic indicates the relative occurrence of concordant to discordant pairs between reported votes of respondents and their predicted votes. Each table also presents the percentage of Conservative and Labour votes correctly classified. This computation predicts a Conservative vote for a respondent in cases for which the probability of the fitted model exceed 0.5 and a Labour vote otherwise. The first model compares reported vote choice and relative perceptions of the parties on the policy questions. Table 2 reports the findings of the models for the split sample responding to the five-issue scales, and Table 3 shows results for those responding to the three-issue scales. The left column gives the estimated parameters for the simple logistic regression model limited to the issue dimensions only. Party proximity on each policy issue appears significantly and independently related to the vote. Coefficients for the scales are universally in the correct direction (here, and in the estimated equations that follow). A party perceived nearer voters' policy preferences benefits by receiving a greater share of their votes. The policy scale comparing jobs and prices shows the weakest relationship with party choice, perhaps due to the multidimensional nature of this "unidimensional" scale. In both samples, proximity on the policy scales is strongly related to voters' loyalty; either bundle of issues can explain about 85% of Labour and Conservative votes, without the inclusion of other voter characteristics. Policy preferences and perceptions, however, should also be related to partisanship and social status. Whether issue preferences either influence or are influenced by ideological, 8 partisan, and group affiliation, controls for respondent party and class reduce or eliminate the independent effects of party proximity. The middle and right-most columns of Table 2 and 3 present estimated equations incorporating variables for party and social class. As expected, variables indicating party affiliation are strongly associated with party votes. Relative party proximity measures, however, remain important. In the first split sample, party and class effects absorb the economic policies of inflation-unemployment and privatisation-nationalisation. The effects of social class on the vote disappear as variables indicating affiliation with each [156 ] 375 Table 2. Issues, party, and social class influences on the vote - logistic regression results - eleven-point scales (British National Election Study, 1992) Estimated Unemployment/inflation Taxation/services Estimated Estimated coefficient coefficient coefficient (Standard error) (Standard error) (S tandard error) .072* .048 .040 (.030) (.043) (.051) .192*** .256*** (.032) Nationalisation .155*** (.027) Redistribution of wealth .276*** European Community (.055) .076* .045 (.037) (.041) .250*** (.030) (.044) .196*** .197*** (.037) Party-Conservative .262*** (.049) .190*** (.047) .182* (.053) (.058) 4.465*** 2.839*** (.397) Party-Labour (.443) -3.459*** (.633) Social class Intercept Likelihood ratio Gamma Correctly classified .413** .286 (.137) (.161) -2.177*** -.389 (.124) (.413) (.529) 621.4*** 851.8*** 895.3*** .778*** .893 .973 .983 85.0% 94.4% 95.7% * Chi-square statistic significant at .05 level. ** Chi-square statistic significant at .01 level. *** Chi-square statistic significant at .001 level. of the parties are included. Other political-economic policies retain close relationships with voting behavior. For the second sample, defence and social welfare policies maintain an association with reported vote, although social orientations, that is, policies on women's rights, fail to show any link with vote choice one major party identification is incorporated into the model. The logistic regression equations adding party affiliation explain a larger share of Labour and Conservative voters; up to 95% of reported votes are predicted correctly by these hybrid models. [157 ] 376 Table 3. Issues, party, and social class influences on the vote -logistic regression results - seven-point scales (British National Election Study, 1992) Defence Welfare Estimated Estimated coefficient coefficient coefficient (Standard error) (Standard error) (Standard error) .613*** (.066) .593*** (.052) Women's rights .380*** (.079) Party-Conservative .369*** (.084) .471 *** (.070) .325** Estimated .296** (.098) .504*** (.085) .173 (.1l2) (.125) 4.459*** 2.811 *** (.431) Party-Labour (.457) -3.681 *** (.500) Social class Intercept Likelihood ratio Gamma Correctly classified .665*** .431 ** (.125) (.146) -2.324*** -.130* (.1l6) (.358) (.454) 583.4*** 827.7*** 910.5*** .830*** .846 .955 .973 85.8% 92.7% 94.8% * Chi-square statistic significant at .05 level. ** Chi-square statistic significant at .0llevel. *** Chi-square statistic significant at .001 level. These findings strongly support the thesis of the multidimensional nature of issues in British electoral behavior. The simple proximity model measures the influence of issues as comparative distance between perceived party locations but does not allow for intensity of preferences for particular issues. Respondents in the first split sample not only describe policy preferences but also rate the importance of these policies. These self-reported assessments of policy importance can be used as a proxy for the weight an issue plays in the relative evaluations of the parties. The weight, Wij, for computation of each relative party distance, d ij , is operationally defined in the following manner. Issues considered not very or not at all important have a unit weight as outlined above. Issues described as "important" receive a weight of 2, and those valued as "extremely important" are weighted by 4. [158 ] 377 Table 4. Issues, party, and social class influences on the vote - logistic regression results - eleven-point scales weighted by intensity (British National Election Study, 1992) Unemployment/inflation Estimated Estimated Estimated coefficient coefficient coefficient (Standard error) (Standard error) (Standard error) .035** (.011 ) Taxation/services .080*** (.013) Nationalisation/privatisation .077*** (.014) Redistribution of wealth .149*** (.017) Party-Conservative .024 .023 (.014) (.018) .096*** .097*** (.017) (.019) .038* .022 (.016) (.018) .138*** .101 *** (.025) (.026) 4.663*** 2.976*** (.411 ) (.448) -3.508*** Party-Labour (.643) .481*** Social class Intercept Likelihood ratio Gamma Correctly classified .826*** .336* (.135) (.158) -2.340*** -.487 (.121) (.415) (.519) 601.9*** 853.2*** 898.0*** .888 .976 .982 85.6% 94.7% 95.6% * Chi-square statistic significant at .05 level. ** Chi-square statistic significant at .01 level. *** Chi-square statistic significant at .001 level. The perceived party differences are multiplied by their respective policy weights and then serve as independent variables in new formulations of the logistic regression models. The results for the weighted models are given in Table 4. In general, the findings for weighted party preference are consistent with the uniform party model. Party proximity on policy issues remains significantly related to vote choice. The two economic scales with weak relationships in the uniformly weighted model also show little association with vote choice in weighted evaluations once partisanship and social class are controlled. The importance attached to British participation in the European Community appears somewhat more important in the weighted model. The predictive power of the logistic models with weighted issues increases, but [ 159] 378 only slightly. Intensity of preferences for certain policies may influence a voter's loyalty for one party over another, but the parsimonious proximity model of policy preference and vote choice explains as much with less information. The distance between voters' policy preferences and perceived party programs accounts for the value attributed to certain policies. 9 7. Ideology and the British electorate Our findings suggest that British voters do take issue preferences into account as they choose among parties in national elections. Our results reconfirm that party affiliation and, to a lesser extent, social class reinforce preferences and the likelihood of vote choice for a political party. Voters, nevertheless, are also able to make reasoned evaluations of party positions on policy questions and compare these to their own preferences. British voters tend to choose the political party which they perceive to be nearest their personal preferences as they cast ballots. Evidence suggests that British party elites are well aware of voters' reliance on issues and ideology when making decisions in an era of partisan dealignment. Within recent years both major parties have tried to position themselves strategically, near the center of the British electorate. Labour Party reforms have repositioned it as a catch-all party, more attractive to voters beyond the working class. The Conservative Party has abandoned some of the more extremist elements of Thatcherism, adopting a moderate ideology. Leaders of both parties recognize the need to appeal to moderate voters to win elections and are now competing for the center of the ideological space. I 0 One element missing from this analysis is the influence of nonpolicy characteristics on voting decisions. Personal characteristics of party leaders (such as Tony Blair) may appeal to British voters, just as these factors may be influential to voters in the United States. The Americanization of political campaigns seems to be affecting democratic elections globally, with its emphasis on technology and personalization of electoral alternatives. British parties, moreover, also have nonpolicy characteristics (perhaps unlike their American counterparts). The Labour Party, for instance, has made great pains to shed its image as the "Looney Left", as the Tory campaign equating "New Labour, New Danger" warns voters against trusting their opponents. The old class divisions between the parties seems to give way to a new political-social ideology epitomised by personalized characteristics of the parties. The nonpolicy component of partisan affiliation in Britain seems an important topic for future research. British political ideology remains an illusive concept, but our findings strongly suggest that issue preferences are an important component underly- [160 ] 379 ing voting decisions in Britain. Voters make assessments of their own policy preferences and of the platforms promoted by the two parties. Voters are more likely to cast ballots for the party perceived as nearer to their own preferences on a wide range of issues. This evaluative process goes beyond simple partisanship and class orientation. Voters take issues into consideration when evaluating parties, and the spatial model serves as an effective means to describe the process of how voters make these judgments. Notes 1. The split sample design also varied question order for items on partisan identification and voting behavior. The split sample rating the five policy areas identified partisanship before the vote. Half of respondents. those rating three policies, identified the 1992 party vote prior to personal party affiliation. 2. The equation transforming a value from the seven point scale (X7) to a corresponding point on the eleven-point scale (Xli) is: XII = [(X7-1)/6]*1O + 1. 3. This construction can-ies the primitive assumption that preferences on issues are unique and separable, since the model tests for independent effects of relative distance on each scale. On the other hand, if issue preferences were completely nonseparable or if private opinions con-espond to party or class affiliations, then we would fail to find statistical significance for rejecting a hypothesis of dependence. A factor analysis of placements on the policy scales suggests multidimensionality in party- and self-placements. 4. In the full sample, the Tories receive 45.6% of the vote. Labour gets 34.3%. The remainder of votes are distributed among the smaller parties: Liberal Democrat 17.1 %. Scottish National 2.2%, Green 0.3%, Plaid Cymru 0.2%, and other 0.3%. The Conservatives win 57.1 % of the two-party vote in the 1992 BES. 5. The sample sizes for each logistic regressions are, in order of basic and hybrid models by raw and weighted observations - Table I and 3: 973, 814: 951, 798: Table 2: 1014.847: 988, 826. Sample size for hybrid models is slightly lower because of missing values for the class variable. 6. The BES prompts for party affiliation and, if none is supplied, presses for the party closer to the respondent. If Labour or Conservative is answered at either stage. the variables measure party affiliation. 7. The categories, in descending order, are: Professional, Managerial and technical, Skilled nonmanual, Skilled manual, Partly skilled, Skilled. 8. Perhaps indicating British scholars' rejection of spatial models of ideology and issue voting, the BES does not include questions regarding the general ideological orientation of political parties or leaders other than the scales reviewed here. 9. Of course, all of the policy scales in the 1992 BES are important to a large segment of the split sample. It is quite reasonable to assume that some issues are of no importance to all but a few voters, and weighting across issues would better reflect such preferences. 10. This movement towards the middle is a search for centrist positions on the ideological dimension and not for a median location on each separate issue dimension, as many British scholars have wrongly suggested. While individual issues are linked to ideology. parties seek the median of voter ideal points induced from preferences on the issue dimensions and perceived linkages to party ideology. [ 161 ] 380 References Alford, R.R. (1963). Party and society: The Anglo-American democracies. Chicago: RandMcNally. Alt, J.E., Sarlvik, B. and Crewe, I. (1976). Partisanship and policy choice: Issue preferences in the British electorate, February 1974. British Journal of Political Science 6: 273-290. Barnett, C. (1986). The audit of war. London: Macmillan. Bean, e. and Mughan, A. (1989). Leadership effects in parliamentary elections in Australia and Britain. American Political Science Review 83: 1165-1179. Beer, S. (1982). Britain against itself. New York: Norton. Black, D. (1958). Theory of committee and elections. Cambridge: Cambridge University Press. Budge, I. (1993). Issues, dimensions, and agenda change in postwar democracies: Long-term trends in party election programs and newspaper reports in twenty-three democracies. In W.H. Riker (Ed.), Agenda formation. Ann Arbor: University of Michigan Press. Budge, I. (1994). A new spatial theory of party competition: Uncertainty, ideology and policy equilibria viewed comparatively and temporally. British Journal of Political Science 24: 443-467. Budge, I. and Farlie, D. (1977). Voting and party competition: A theoretical critique and synthesis applied to surveys from ten democracies. London: Wiley. Budge, I., Farlie, D. and Laver, M. (1983). What is 'rational' choice?: Shifts in meaning within explanations of voting and party competition. Electoral Studies 2: 23-38. Butler, D. and Kavanagh, D. (1983). The British General Election of 1983. New York: St. Martin's Press. Butler, D. and Stokes, D. (1969). Political change in Britain. New York: St. Martin's Press. Clarke, H.D. and Stewart, M.e. (1984). Dealignment of degree: Partisan change in Britain, 1974--83. Journal of Politics 48: 689-718. Clarke, H.D. and Stewart, M.C. (1992). The (un)importance of party leader images and party choice in the 1987 British election. Journal of Politics 54: 447-470. Clarke, H.D., Stewart, M.e. and Zuk, G. (1986). Politics, economic, and party popularity in Britain, 1979-83. Electoral Studies 5: 123-141. Crewe, I. (1983). The electorate: Partisan dealignment ten years on. West European Politics 6: 183-215. Crewe, I., Alt, 1. and Sarlvik, B. (1977). Partisan dealignment in Britain 1964--1974. British Journal of Political Science 7: 129-190. Davis, O.A. and Hinich, M.J. (1966). A mathematical model of policy formulation in a democratic society. In 1.L. Bernd (Ed.), Mathematical applications in political science, Vol. II. Dallas: Southern Methodist University Press. Denver, D. (1994). Elections and voting behavior in Britain. Deddington: Philip Alan. Denver, D. and Hands, G. (1990). Issues, principles, or ideology?: How young voters decide. Electoral Studies 9: 19-36. Downs, A. (1957). An economic theory of democracy. New York: Harper and Row. Enelow, J.M. and Hinich, M.J. (1984). The spatial theory of voting. Cambridge: Cambridge University Press. Franklin, M.N. (1985). Assessing the rise of issue voting in British elections since 1964. Electoral Studies 4: 37-56. Franklin, M.N. (1992). Britain. In M.N. Franklin, T.T. Mackie, and H. Valen (Eds.), Electoral change: Responses to evolving social and attitudinal structures in western countries. New York: Cambridge University Press. [162] 381 Franklin, M.N. and Mughan, A. (1978). The decline of class voting in Britain: Problems of analysis and interpretation. American Political Science Review 72: 523-534. Gamble, A. (1990). Theories of British Politics. Political Studies 38: 404-420. Graetz, B. and McAllister, I. (1987). Party leaders and election outcome in Britain, 1974-1983. Comparative Political Studies 19: 484--507. Heath, A., Jowell, R. and Curtice, J. (1985). How Britain votes. New York: Pergamon Press. Heath, A., Jowell, R., Curtice, J.K., Brand, J.A. and Mitchell, J.e. (1993). British General Election Study, 1992 [computer file], ICPSR version. London: Social and Community Planning Research [producer], 1992. Colchester: ESRC Data Archive/Ann Arbor: Interuniversity Consortium for Political and Social Research, 1996. Hinich, M.J. (1977). Equilibrium in spatial voting: The median voter result is an artifact. Journal of Economic Theory 16: 208-219. Hinich, M.J. and Munger, M.C. (1994). Ideology and the theory ofpolitical choice. Ann Arbor: University of Michigan Press. Hinich, M.J. and Pollard, w.J. (1981). A new approach 0 the spatial theory of electoral competition. American Journal of Political Science 25: 323-341. King, A. (1982). Whatever is happening to the British party model? Parliamentary Affairs 35: 241-251. Klingemann, H.D., Hofferbert, R.1. and Budge, I. (1994). Parties, policies, and democracy. Boulder: Westview Press. LeDuc, L. (1985). Partisan change and dealignment in Canada, Great Britain, and the United States. Comparative Politics 17: 379-398. McAllister, I. and Rose, R. (1984). The nationwide competition for votes. London: F. Pinter and the Center for the Study of Public Policy. Miller, w.L., Tagg, S. and Britto, K. (\986). Partisanship and party preferences in government and opposition: The mid-term perspective. Electoral Studies 5: 31-46. Mishler, w., Hoskin, M. and Fitzgerald, R. (1989). British parties in the balance: A timeseries analysis of long-term trends in Labour and Conservative support. British Journal of Political Science 19: 211-236. Norris, P. (1994). Labour party factionalism and extremism. In A. Heath, R. Jowell, and 1. Curtice (Eds.), Labour's last chance: The 1992 election and beyond. Brookfield, VT: Dartmouth Publishing. Norris, P. (1996). Electoral change in Britain since 1945. Oxford: Blackwell. Pulzer, P.G. (1967). Political representation and elections in Britain. London: Allen and Unwin. Sarlvik, B. and Crewe, I. (1983). Decade of dealignment: The Conservative victory of 1979 and electoral trends in the 1970s. New York: Cambridge University Press. Vincent, A. (1994). British conservatism and the problem of ideology. Political Studies 42: 204--227. Weakliem, D.L. (1995). Two models of class voting. British Journal of Political Science 25: 254--270. Whiteley, P., Seyd, P., Richardson, 1. and Bissell, P. (1994). Thatcherism and the Conservative party. Political Studies 42: 183-203. Wiener, M. (1981). English culture and the decline of the industrial spirit. Cambridge: Cambridge University Press. [163] 382 Appendix Issue scales and endpoints (British National Election Study, 1992) Issue scale I (A) 7 (G) or II (K) Keep prices down Unemployment/inflation Back to work Taxation/services Up taxes and spend more Redistribution of wealth tries Greater effort to equalise in- Less concerned to equalise European Community come Unite fully with European income Protect independence Defence Community Spend less on defence Greatly increase spending Welfare Govt responsible for job and Get ahead on own Women's rights Women have an equal role Cut taxes and spend less Nationalisation/privatisation Nationalise more industries Sell of nationalised indus- on defence good standard of living [ 164 ] Women's place in the home Public Choice 97: 383-399, 1998. © 1998 Kluwer Academic Publishers. 383 Retrospective and prospective voting in a one-party- dominant democracy: Taiwan's 1996 presidential election JOHN FUH-SHENG HSIEH l , DEAN LACy2 & EMERSON M.S. NIOU 3 1Department of Political Sciences, National Chengchi University, Taipei 11623, Taiwan; 2Department of Political Science, Ohio State University, Columbus, OR 43210- 1373, U.S.A.; 3Department of Political Science, Duke University, Durham, NC 27708-0204, U.S.A. Abstract. Several theories of voting behavior suggest that voters evaluate candidates in an election based on the candidates' past performance and future promise. There is a dispute in the theory and ambiguity in empirical evidence about which direction voters look when choosing candidates: do voters weigh past performance or future promise more heavily in the voting booth? This paper contributes empirical support to the prospective voting model by testing both retrospective and prospective voting in a pivotal case: the 1996 Taiwan presidential election. Taiwan's 1996 election represents the first popular election of the president from a field of candidates that included the long-ruling KMT party incumbent, Lee Tent-hui. In the Taiwan presidential election, voter evaluations of Lee's prospects for managing the economy in the future prove statistically significant as a predictor of voter choice. Voter evaluations of recent economic conditions do not appear closely related to voter choice. Voters' perceptions of the candidates' abilities to influence ethnic relations, domestic safety, and international security are better predictors of the vote than past ethnic relations or past security problems, even in the face of Communist China's pre- election aggression toward Taiwan. 1. Introduction Theories of voting behavior disagree on whether voters vote retrospectively or prospectively. According to the retrospective voting model, voters evaluate an incumbent's past performance, typically on economic issues, and vote against the incumbent if conditions worsened during his or her term in office (Key, 1966; Fiorina, 1981). The prospective voting model suggests that even if voters consider a candidate's past performance, they look primarily at each candidate's future promise for managing national affairs, again typically economic (Downs, 1957; Achen, 1992). The empirical evidence weighing these two models is mixed (Kiewiet, 1983; Lewis-Beck, 1990). This paper seeks to adjudicate between prospective and retrospective voting models by testing each in an important pivotal case: Taiwan's first popular presidential election, on 23 March 1996. [165 ] 384 Taiwan represents an important case for the study of prospective and retrospective voting since it is, and has been for five decades, a one-party-dominant regime. Though Taiwan is now a fully functioning democratic state with free and fair elections for both the legislative and executive branches, Taiwan remains a one-party-dominant regime with the ruling Kuomintang (Nationalist Party) controlling government at most levels. Taiwan's unique political situation makes it an ideal test case for retrospective and prospective voting models on three counts. First, Taiwan's legislative and executive branches of government have been unified under one party for decades, thus allowing voters to identify clearly who is responsible for past economic and social conditions. The United States often serves as a test case for retrospective voting models (Fiorina, 1981), but given recent American experiences with divided government (Fiorina, 1992; Jacobson, 1990), it is difficult for American voters to serve as "rational gods of vengeance and reward" (Key, 1966). Second, Taiwan's executive branch has been controlled by President Lee Teng-hui for eight years. Lee's tenure enables voters to attribute economic and social conditions to Lee rather than having to decide whether Lee inherited conditions from recent predecessors. Most tests of retrospective voting models use American electoral data, but when a candidate has served only four years or less, as is the case in every American presidential election since 1948, it is difficult for voters to decide whether the president is truly accountable for national conditions. Finally, Taiwanese voters typically find four issues important in national elections: the economy, relations with Communist China, public safety, and ethnic relations. In other countries many more issues - religious, racial, environmental, and cultural - often cloud any attempts to test a parsimonious issue voting model. By tapping voter evaluations of past performance on a few issues along with their evaluations of candidates' potential to improve or undermine those issues, we are able to determine whether voters in the 1996 Taiwan presidential election simply looked at the past and decided to stay the course or whether they looked to the future and evaluated the many different paths offered by the competing candidates. Our results are surprising. Using data provided by National Chengchi University, we estimate a multinomiallogit model of the vote for Lee, Peng, Chen, and Lin. We find that voter evaluations of past economic conditions - both sociotropic and personal - are not statistically significant as an explanation of voter choice. Instead, voter evaluations of Lee's prospects for managing the economy are statistically significant as an explanation of the vote. Voters also evaluated the candidates' abilities to manage ethnic relations and cross-straits relations in the future. Retrospective evaluations of ethnic relations, straits relations, and public safety are not statistically significant [166 ] 385 as explanations of the vote. However, the prospective and candidate-specific versions of these issues are significant. The paper proceeds as follows. We provide a brief background on Taiwan's first popular election of a president (Section 2). We then outline the competing models of voting and discuss empirical findings to date (Section 3). We describe our model and data, and we present results using multinomiallogit to estimate the model (Section 4). We conclude by discussing the importance of our findings in the general literature on voting behavior and on the future of Taiwanese presidential elections (Section 5). 2. Background On 23 March, 1996, voters in Taiwan went to the polls to participate in the first direct election of the president in the history of the Republic of China (ROC) on Taiwan. According to the original stipulations of the ROC Constitution of 1946, the president and vice-president were elected by the National Assembly. The 1994 amendments to the Constitution, known as the Additional Articles, require direct popular election of the president and vicepresident. The presidential and vice-presidential candidates run as a single ticket, and the ticket that attains a plurality of the popular vote is elected. Although the ROC Constitution provides essentially for a parliamentary form of government, the president has, for most of the past half a century, exercised a great deal of power due to his control over the ruling Kuomintang (Nationalist Party, KMT), which has always commanded a majority of seats in the Legislative Yuan (Parliament) (Hsieh, 1993). The KMT nominated Lee Teng-hui as its presidential candidate. Lee picked Premier Lien Chan as his running mate. The major opposition party, the Democratic Progressive Party (DPP), after a two-stage primary, nominated Peng Ming-min, a long- time political refugee residing overseas. His running mate was Hsieh Chang-ting, a prominent DPP member of the Legislative Yuan. Lin Yang-kang and Hau Pei-tsun, both of whom were KMT vice-chairmen, bolted their party to run as an independent ticket. The fourth ticket in race was Chen Li-an and Wang Ching-feng. Chen was a former member of the KMT and president of the Control Yuan, an organ similar to the parliamentary ombudsman in the Scandinavian countries. His running mate, Wang Chingfeng, a member of the Control Yuan, was the only female candidate in the presidential race. The opposition New Party (NP), which split from the KMT in 1993, did not formally nominate a candidate, but decided instead to endorse the Lin- Hau ticket. The election outcome was clear from the beginning of the campaign. President Lee was well ahead of his opponents according to the polls released by [167 ] 386 Table 1. The result of the 1996 presidential election Party Vote % Lee Teng-hui KMT 54.00 Peng Ming-rnin DPP 21.13 Lin Yang-kang Independent* 14.90 ChenLi-an Independent 9.98 Candidate *indicates endorsement by the New Party. Source: China Times, 24 March 1996. various organizations during the campaign. He ended up, as Table 1 shows, with 54% of the valid vote, about 33 percentage points ahead of the second vote-getter, DPP's Peng Ming-min. 3. Retrospective and prospective voting Retrospective voting has gained widespread appeal as a simple yet powerful explanation of voter choice. V.O. Key argued that voters are "rational gods of vengeance and reward," who observe the performance of the incumbent and "cast him out" if the incumbent's performance is poor. As va. Key, Jr. claimed, "Voters may reject what they have known; or they may approve what they have known. They are not likely to be attracted in great numbers by promises of the novel or the unknown" (Key, 1966: 61). Rather than attempting to anticipate the consequences of various policy proposals, voters look at past results. Elections, then, are plebiscites. Fiorina (1981) further develops and tests the theory of retrospective voting in the United States. He finds support for a theory of retrospection close to Key's. Voters appear to weigh past outcomes heavily and to pay scant attention to policy prescriptions. A very different theory of voting holds that voters look forward rather than backward. Downs (1957) proposes a forward- looking theory of voting that is grounded in, but not limited to, retrospective evaluations of the incumbent. While a pure retrospective voting theory such as Key's is rooted in simple reward and punishment motives in voters, Downs argues that retrospective evaluations of candidates and parties become predictors of future performance in office. In Downs's (1957: 49) terminology: "By comparing the stream of utility income from government activity he has received under the present government (adjusted for trends) with those streams he would have received if the various opposition parties had been in office, the voter finds ... his preference among the competing parties." Instead of evaluating [168] 387 political candidates (or parties) based on their past performance, voters evaluate past policies in order to anticipate future policies. Downs attaches great importance to retrospective voting, but, at its core, his thesis holds that voters are investors maximizing future well-being. Achen (1992) argues that voter rationality implies prospective rather than retrospective voting. The inclusion and success of retrospective evaluations in voting models is due to their correlation with other sources of candidate evaluations. Retrospective evaluations are an instrumental and intermediate step in voter evaluations of candidates, but they are not the ultimate decision rule for voters. Achen, like Downs, believes that voters have the motivation and information necessary to anticipate how different candidates or parties will behave in office. Theories of voting also disagree on whether voters care more about their personal well-being or about national conditions. The theories of Key, Downs, and Fiorina all suggest that a voter's ultimate concern is her own well-being. Fiorina (1981) is a bit less restrictive in casting retrospective voting in purely self-interested terms. He includes measures of voters' estimates of their own condition and national conditions, finding that both are statistically significant as predictors ofthe vote. Kinder and Kiewiet (1981) and Kiewiet (1983) argue that voters are sociotropic rather than self-interested, meaning that they care more about the conditions of the national situation or of specific socioeconomic groups than about their own well-being. Different studies suggest that different indicators of economic growth, unemployment, inflation and personal income (Erikson, 1989, 1990; Fair, 1978, 1988; Hibbs, 1987; Tufte, 1978). A prospective sociotropic voter evaluates the candidates by looking at what they will do rather than what they have done, but, unlike prospective personal voters, they care about national or regional conditions. MacKuen, Erikson, and Stimson (1992) find empirical support for the idea that voters are bankers (sophisticated and forward-looking) rather than peasants (uninformed and backward-looking). MacKuen, Erikson, and Stimson, unlike Downs, believe that voters look at national conditions when evaluating the prospects of different candidates. Their empirical model predicts changes in presidential approval ratings based on fluctuations in inflation and unemployment and on retrospective and prospective evaluations of personal and national conditions. Based on data from 1954 to 1988, they find that voter expectations about national economic conditions are statistically significant across several model specifications. Lewis-Beck (1990) finds cross-national support for prospective voting. His study of voting in Western Europe and the United States during the early 1980s reveals that "Economic voters in these nations do act retrospectively, [169] 388 just as Key (and Downs) have described. However, they also respond to a purely prospective component of economic evaluation" (p. 135). Most of the studies on retrospective or prospective voting have focused on national election in the United States and Western Europe. Only a few studies have focused on a case such as Taiwan. Hwang (1994), in a study of Taiwan's Legislative Yuan election, finds that economic conditions were not significant to voter choice. However, Hsieh, Lacy, and Niou (1996) find that economic conditions did matter in the Taipei mayor's race of 1995. Nevertheless, such studies are rare, and the results are contradictory. The recent presidential election in Taiwan provides another opportunity to extend and refine the previous findings. In the case of Taiwan, one may suspect that economic voting may not be significant since Taiwan's economic condition has been generally good over the past several decades. For several decades, Taiwan's economy grew at an annual rate of 8 to 9% on average, and the distribution of income and wealth gradually improved. However economic growth has slowed recently to an annual rate of about 6%, and even the distribution of income and wealth has deteriorated to some extent. Given the recent changes in economic conditions, voters may have reason to disapprove of Lee Teng-hui's performance as president. The Election Study Center of National Chengchi University conducted a nationwide face-to-face survey of 1396 respondents before and after the election. The survey included questions to tap voter assessments of national and personal economic conditions. One question that addresses retrospective evaluations of the national economy condition is phrased as follows: "Do you feel that the current economic condition of our whole society is better or worse than one year ago, or stays the same?" 01. Much better 02. Better 03. About the same 04. Worse 05. Much worse 98. Don't know 95. Refuse to answer A similar question asked voters to rate their personal finance over the past year. Table 2 shows how respondents in Taiwan perceived national and personal economic performance over the previous year and anticipated economic performance over the next year. Retrospective perceptions of the national economy are quite pessimistic: about two thirds of respondents believe the national economy worsened, 23% believe it stayed the same, and only slightly more than one-tenth believe it got better. Retrospective evaluations of respondents' own economic well-being are much more concentrated: 28% of respondents believe they were worse off, 61 % the same, and 12% better off. Prospective evaluations reveal quite different patterns. The respondents are very optimistic. On the national economy, only 18% of respondents believe [170] 389 Table 2. Perceptions of the economy Much Worse Same Better Much N better worse National Last year 404 (31 ) 467 (36) 297 (23) 114 (9) 21 (2) 1304 Personal 127 (9) 253 (19) 820 (61) 134 )10) 21 (2) 1355 33 (4) 124 (14) 302 (34) 402 (45) 33 (4) 894 12 (1) 78 (8) 516 320 (34) 31 (3) 956 (54) Last year National Next year Personal Next year Note. Cell entries are number of respondents, with row percentages in parentheses. Row percentages may not sum to 100 due to rounding. Based upon survey conducted by the Election Study Center, National Chengchi University. the economy will get worse over the next year, 34% believe the economy will stay the same, and nearly half of respondents believe the economy will get better. Nine percent of respondents believe their own economic situation will worsen, while 54% believe they will stay the same and 37% believe they will be better off. The correlations among these variables are modest. The correlation between national prospective and national retrospective evaluations is .24; between personal prospective and personal retrospective evaluations, .37. The correlations between national prospective and personal retrospective is .34 while the correlation between national retrospective and personal prospective is .34. As Table 2 illustrates, there are obvious discrepancies among various types of evaluations: people are more optimistic about the future than about the past, and their evaluations of personal well-being are more concentrated than their evaluations of the national economy. Moreover, the table reveals that the number of people responding to questions regarding retrospective evaluations is much larger than the number offering prospective evaluations. Such discrepancies are interesting and suggest a line of research into how people form expectations of the economy. However, we do not pursue that here. All these questions in the survey were asked before voters were asked about their vote intentions, thus we do not expect that economic perceptions are rationalizations about a person's choice of the incumbent or a challenger in the election. [ 171 ] 390 Voters may be concerned about issues other than the economy as they go to the polls. National security is a paramount concern of Taiwanese voters due to the threat from the Chinese Communists on the mainland. In the 1995 election survey, two questions tap the status of the evaluations between Taiwan and the mainland in the previous year and in the next year. The wordings are similar to the previous questions concerning economic conditions. Taiwanese voters also care about political and social stability. Four questions asked respondents to evaluate law and order in the previous year, law and order in the next year, ethnic relations in the previous year, and ethnic relations in the next year. These questions are all related to domestic stability, and are certainly very important concerns for the voters in Taiwan. Table 3 shows that, on the situations of cross-strait relations and law and order, most people give quite negative responses to the past, but are generally quite positive about the future. On cross-strait relations, more than threefifths of respondents believe the situation worsened in the previous year, 22% believe it stayed the same, and only 16% believe it got better. This reflects the tensions between Taiwan and the mainland prior to the election. On law and order, 46% think that the situation was worse, 39% the same, and 15% better. But when the respondents look to the future, their evaluations are brighter. Only about 14% of respondents believe the cross-strait relations will get worse, 29% believe the same, and 58% believe it will improve. On law and order, 19% believe the situation will get worse in the future, 48% believe the same, and one-third believe that it will get better. On ethnic relations the respondents are generally optimistic. Retrospectively, only 14% of respondents think ethnic relations worsened, 48% believe relations remained the same, and about 39% believe ethnic relations improved. Prospectively, only 5% believe the ethnic situation will get worse, 46% believe it will remain the same, and nearly half believe that ethnic relations will improve. The correlations among these are modest. The highest correlation appears between retrospective and prospective evaluations of ethnic relations (.69). The relationship between retrospective evaluations and vote choice is straightforward. If voters believe that economic, social, and political conditions worsened, they should vote against the incumbent, Lee Teng-hui. The relationship between vote choice and prospective assessments of national conditions is more complex. If voters believe that conditions will improve during the next year, it is not clear for who they should vote, nor is it clear why they expect conditions will improve. It is possible that a person who believes conditions will improve does so because she believes Lee will win reelection and that he will handle the national affairs capably. Another voter may believe that national conditions will improve because she believes Lee will be defeated. [ 172] 391 Table 3. Perceptions of social and political situations Much Worse Same Better worse Much N better Security Last year 261 (24) 428 (39) 241 (22) 165 (15) (I) Law & order 196 (16) 379 (30) 499 (39) 177 (14) (I) 32 (3) 124 550 (48) 372 (33) 65 (6) 1144 (11) 30 (4) 79 (10) 243 448 (54) 36 (4) 837 (29) 49 (6) 107 (13) 391 (48) 253 14 (2) 814 (31 ) 20 (2) 33 (3) 446 (46) 408 (42) 72 (7) 979 Last year Ethnic reI. Last year Security Next year Law & order Next year Ethnic reI. Next year 16 17 1109 1269 Note. Cell entries are number of respondents, with row percentages in parentheses. Row percentages may not sum to 100 due to rounding. Based upon survey conducted by the Election Study Center, National Chengchi University. Prospective evaluations should be candidate-specific. That is, voters should attach to each candidate an assessment of how well that candidate will perform in office. To tap prospective evaluations of the candidates, the National Chengchi University Election Study asked voters which candidate or candidates would best handle the economy, law and order, ethnic relations, and straits relations. Table 4 presents the percentage of respondents who believe that each of the candidates is best able to handle each of the issues. Lee dominates all other candidates in voters' assessments of his ability to manage national affairs. Lee's strongest issue is the economy, followed by relations with the mainland, public safety, and ethnic relations. Chen and Peng earn their highest marks on ethnic relations; Lin, on public safety. The low marks earned by the three challenges underscores a classic problem for political challengers in one-party regimes or even competitive democracies: until a person has held office, voters have little to evaluate. 4. The model and results The purpose of this paper is to assess the beliefs related to vote choice. Particularly, we are interested in whether voters vote retrospectively or prospec- [173 ] 392 Table 4. Percentage of voters who rate each candidate as best able to manage issues Issue Economy Ethnic relations Public safety Straits Lee 70 Chen 2 Peng Lin 4 3 N 1003 41 14 7 5 5 6 11 1003 48 51 8 3 7 1003 1003 Note. Entries are row percentages. Entries may not sum to 100% since some respondents mentioned multiple candidates or no candidates as best able to handle each issue. tively. The dependent variables in our formulation will be vote choice in the 1996 Taiwan presidential election as reported by survey respondents. The dependent variable has four unordered possibilities: vote for Lee, for Peng, for Lin, and for Chen. We use multinomiallogit to estimate the model, normalizing coefficients for Lee at zero. All other coefficients from the model are interpreted as the impact of that variable on vote choice, relative to a vote for Lee. Independent variables include party identification and evaluations of the economy, public safety, ethnic relations, and cross-straits relations during the past year. To determine the extent of prospective voting, we turn to questions that ask voters which candidate would most capably handle each of four issue areas: the economy, public safety, ethnic relations, and relations across the straits. Respondents could list anyone candidate or combinations of two, three, or four candidates as well as answering that none of the candidates could capably handle an issue. We code a dummy variable for each candidate on each issue. The dummy variable takes a value of (l) if a voter believes that the candidate can handle the issue better than any other candidate. The excluded category of response on each issue includes voters who believe that combinations of two, three, all, or no candidates could capably handle the issue. We include party identification in our model, formulated as three dummy variables, KMT, DPP, and NP, indicating which party a voter generally supports. Table 5 present the results. Partisanship is statistically significant as an explanation of voter choice. KMT supporters are less likely to support candidates other than incumbent President Lee Teng-hui; the DPP supporters are more likely to support Peng Ming-min; and the NP supporters are more likely to support Lin Yang-kang. But even after taking into consideration voters' party identification, retrospective and prospective evaluations of the economy, security, ethnic rela- [174] 393 tions, and straits relations appear statistically significant as predictors of voter choice. Surprisingly, retrospective evaluations of the economy, both national and personal, are not statistically significant, except that people who believe the national economy improved during the last year are more likely to vote for Lin than Lee. One explanation of this result may be that voters who believe the economy improved during the last year are more willing to risk a vote for Lin, who campaigned primarily on non-economic issues. Prospective evaluations of Lee's ability to handle the economy are statistically significant in the comparison of Lee with each of the other candidates. The evidence suggests that to the extent that voters vote economically, they rely more on prospective evaluations than retrospective evaluations. Prospective evaluations are likely a function of retrospective evaluations, but our results suggest that beliefs about a candidate's future performance are more closely related to vote choice than beliefs about past performance. Retrospective evaluations of ethnic relations, straits relations, and public safety are not statistically significant as explanations of the vote. However, the prospective, candidate-specific versions of these issues are significant. People who believe that Peng Ming-min is most able to handle ethnic relations and straits relations are more likely to choose him over Lee. Chen draws support from voters who believe he is best able to handle ethnic problems, and Lin gets the votes of voters who believe he is capable of dealing with straits relations and ethnic problems. Lee appears to draw his support from prospective evaluations of his ability to handle the economy and public safety. Prospective evaluations of candidates' abilities to handle various problems may present an endogeneity problem. That is, voters who vote for Lee may be rationalize doing so by responding that he is best able to handle the economy, ethnic relations, straits relations, and public safety. We believe this cannot explain our findings in Table 5 for several reasons. First, we asked voters to evaluate the candidates on each issue before asking them their vote. In doing so, we intended to minimize the extent to which voters would say the candidate they voted for would be best able to handle each of the issues. Second, if prospective evaluations are rationalizations, then one should expect to find that all prospective evaluations of the candidates are statistically significant. If a voter rationalizes voting for Lee by claiming he is best able to manage the economy, then why should she not also claim he is best able to handle public safety, ethnic relations, and straits relations? The simple fact that only a few prospective evaluations are statistically significant leads us to believe that voters are not automatically claiming that the candidate they voted for is best able to handle each of the issues. Finally, if prospective evaluations of various issues are rationalizations of the vote, then should not retrospective [ 175] 394 evaluations of the economy be rationalizations as well? If a voter claims that Lee is best able to handle the economy simply because she voted for him, then she should also have reason to claim that the economy has not worsened over the past year. In short, voters appear not to simply rationalize their vote by claiming that the candidate they voted for is best able to handle all relevant issues. The prospective variables in our model suggest that voters choose a candidate based on how well thev expect that candidate to handle important issues. A surprising finding to emerge from Table 5 is that non- security issues appear at least as important as cross-straits relations. Before and after the election, commentators argued that security concerns dominated the presidential campaign, overshadowing almost all other issues. One plausible explanation is that voters' concerns about cross-strait relations had been "absorbed" by other factors such as economic concerns and party identification. Voters' evaluations of the cross-strait relations, particularly retrospective evaluations of the situation, are closely related to vote choice. Table 6 shows voters' retrospective evaluation of the cross-strait relations and their votes. President Lee fared better among voters who believed the situation had stayed the same or become better in the previous year than among those who thought it worsened. Other candidates found relatively more support from voters who believed the situation across the straits had worsened. The relationship between prospective evaluations of the straits and vote choice is less clear-cut. Table 7 shows how voters voted according to their prospective evaluations of relations with mainland China. Only Lin acquired more support from voters who thought the situation would get worse in the next year. No clear pattern emerges for other candidates. On balance, prospective evaluations of the candidates' abilities to solve various social, political, and economic problems appear to better explain the vote than retrospective evaluations. Retrospective evaluations of the economy carry little weight as a predictor of voter choice, except to the extent that prospective evaluations of Lee's ability to handle the economy stem from his past performance. This seems unlikely, however, since Table 2 shows that most voters believe the economy worsened during the year preceding the election, yet most voters voted for Lee. The results reported in Table 5 support Achen's (1992) contention that rational voters vote prospectively rather than retrospectively. At least in Taiwan's first popular presidential election, voters appear to look forward rather than backward. For elections in other countries, the findings presented here suggests that economic conditions matter even in such a country as the ROC on Taiwan where economic performance has been generally good by international standards. Taiwan is not the first place one would expect to find [176] 395 Table 5. Multinomiallogit estimates of vote choice Candidates Independent variable Constant KMT DPP Chen.Peng Lin. Coefficient Coefficient Coefficient (S.E.) (S.E.) (S.E.) -2.57 (1.66) -2.05** (0.52) -1.61 ** (0.80) NP 1.65** (0.64) Nat. econ. 0.10 (0.23) Last year Pers. econ. Last year Safety Last year Ethnic ReI. Last year Straits Last year Lee -1.40 (1.55) -3.96** (1.07) 2.60** (0.45) -0.14 (0.28) -0.56 (1.25) -0.06 (0.23) 0.28 (0.24) 0.26 (0.23) -0.11 (0.22) -0.38 (0.24) 0.39** (0.20) -1.29** -0.33 (0.24) 0.30 (0.19) -1.92** (0.59) -5.36** (1.83) -1.26** (0.56) -1.80 (1.16) 3.04** (0.64) 0.67** (0.28) -0.12 (0.30) 0.32 (0.24) -0.15 (0.26) -0.00 (0.22) -1.77** (0.57) -0.62 ( 1.12) 0.03 ( 1.27) -0.63 (l.48) 0.10 (0.76) -0.33 ( 1.01) -0.12 (1.43) -1.12 (1.79) 0.69 (0.55) 2.06* (1.14) 3.74** (0.84) -0.86 (1.96) 3.28** (0.66) 1.66** (0.82) 2.26** (0.89) Lee 1.63* (0.91) -0.47 1.29** (0.63) -2.39* (1.40) -1.40** Safety (0.57) (0.56) Economy Peng Economy Chen Economy Lin Economy Lee Ethnic relations Peng Ethnic relations Chen Ethnic relations Lin Ethnic relations (0.66) -2.49 (1.61) 0.25 ( 1.33) -0.00 (1.12) 0.89 (0.80) -1.30* (0.72) [177 ] 396 Table 5. Continued. Candidates Independent variable Chen.Peng Lin. Coefficient Coefficient Coefficient (S.E.) (S.E.) (S.E.) Peng -10.34 -1.17 -10.46 Safety (11S.04) (0.83) (108.S7) Chen --0.06 -1.84** --0.68 (0.91) Safety (0.7S) (0.92) Lin --0.71 -1.06 0.17 Safety (U.67) (0.77) (0.72) Lee --0.47 Straits (0.S8) Peng -9.66 Straits (136.24) 0.9S** (0.S4) 2.S2** (1.30) Chen 0.63 0.14 (0.74) Straits (0.6S) Lin --0.13 0.76 Straits (0.81) (0.88) 1.07 (0.79) 4.4S** (1.92) I.S2* (0.90) 2.38** (0.93) Coefficients for Lee normalized at zero. Log likelihood: -270.18. Number of observations: 662 Percent correctly predicted: 8S.0S Predicted vote shares: Lee 64.8%, Chen: 8.3%, Peng 16.2%, Lin 10.7%. Actual vote shares: Lee 64.8%, Chen 8.3%, Peng 16.2%, Lin 10.7%. * indicates statistical significance at p < .10, two- tailed. ** indicates statistical significance at p < .OS, two- tailed. widespread economic voting. Yet the economy did matter. The results here also confirm the most recent wave of work demonstrating that prospective evaluations mean at least as much to voters as retrospective evaluations of the economy (MacKuen, Erikson, and Stimson, 1992; Lewis-Beck, 1990). Voters are not merely rational gods of vengeance and reward, as V.O. Key described them. Instead, voters form expectations of how political candidates will shape future economic performance. [178] 397 Table 6. Retrospective evaluation of cross-strait relations and the vote Evaluation Vote N Chen Lee Peng Lin 19 38 (22) 29 (17) 175 (I I) 89 (51) Worse 37 (12) 198 (63) 47 (15) 30 (10) 312 Same 7 (4) 125 (76) 22 (13) 10 (6) 164 Better 5 (4) 91 (72) 15 (12) 16 127 (13) 1 (8) 10 (77) 1 (8) 1 (8) Much worse Much better 13 Note. Cell entries are number of respondents, with row percentages in parentheses. Row percentages may not sum to 100 due to rounding. Based upon survey conducted by the Election Study Center, National Chengchi University. Table 7. Prospective evaluation of cross-strait relations and the vote Evaluation Vote N Chen Lee Peng Lin Much worse 2 (10) 13 (62) 2 (10) 4 (19) 21 Worse 8 (13) 31 (50) 15 (24) 8 (13) 62 Same 18 (10) 104 (60) 33 (19) 17 (10) 172 Better 23 (7) 226 (67) 51 (15) 35 (10) 335 Much better 1 (4) 16 4 4 25 (64) (16) (16) Note. Cell entries are number of respondents, with row percentages in parentheses. Row percentages may not sum to 100 due to rounding. Based upon survey conducted by the Election Study Center, National Chengchi University. [179] 398 5. Conclusion The results presented in this paper carry important implications for elections in Taiwan and for the body of theory in political science that relates election outcomes to retrospective or prospective voting. For Taiwanese elections, these results suggest that the economy is an important factor affecting the vote. Voters evaluate the economic performance and promise of competing candidates. Security concerns are important to Taiwanese voters, but they are not predominant. Public safety and ethnic relations are also important issues, but only to the extent that candidates can credibly commit to future improvements rather than campaigning on past success. The results from Taiwan's first popular presidential election also serve as an important test case for theories of retrospective voting. Taiwan is the ideal case to evaluate retrospective voting since one party, the KMT, has been in power for decades, controlling both the executive and legislative branches of government, and since the incumbent candidate has been in power long enough for voters to evaluate his performance. Retrospective voting appears overshadowed by prospective voting, even in Taiwan. In other countries where government is divided between competing parties or where the incumbent executive has been in power only a short time, it is likely that voters will find retrospective clues less helpful when choosing among candidates. References Achen, C.H. (1992). Social psychology, demographic variables, and linear regression: Breaking the iron triangle in voting research. Political Behavior 14: 195-211. Campbell, A., Converse, P.E., Miller, W.E. and Stokes, D.E. (1960). The American voter. New York: Wiley. Downs, A. (1957). An economic theory of democracy. New York: Harper and Row. Erikson, R.S. (1990). Economic conditions and the congressional vote: A review of the macrolevel evidence. American Journal of Political Science 34: 373-399. Erikson, R. (1989). Economic conditions and the presidential vote. American Political Science Review 83: 567-576. Fair, R.c. (1988). The effects of economic events on votes for the President: A 1984 update. Political Behavior 10: 168-179. Fair, R.c. (1978). The effect of economic events on votes for President. Review of Economics and Statistics 60: 159- 173. Fiorina, M.P. (1981). Retrospective voting in American national elections. New Haven: Yale University Press. Fiorina, M.P. (1992). Divided government. New York: Macmillan. Hibbs, D.A., Jr. (1987). The American political economy. Cambridge: Harvard University Press. Hsieh, J.E-s. (1993). Parliamentarism vs. presidentialism: Constitutional choice in the Republic of China on Taiwan. Chinese Political Science Review 21: 173-202. [180 ] 399 Hsieh, J.E-s., Lacy, D. and Niou, E.M.S. (1996). Economic voting in the 1994 Taiwan elections. American Asian Review 14: 51-70. Hwang, S.-D. (1994). Economic conditions and voters' choices (in Chinese). Soochow Journal of Political Science 3: 97-123. Jacobson, G. (1990). The electoral origins of divided government. San Francisco: Westview Press. Key, V.O., Jr. (1966). The responsible electorate. Harvard University Press. Kiewiet, D.R. (1983). Macroeconomic and micropolitics. Chicago: University of Chicago Press. Kinder, D.R. and Kiewiet, R.D. (1981). Sociotropic politics: The American case. British Journal of Political Science 11: 129-161. Lewis-Beck, M.S. (1990). Economics and elections. Ann Arbor: University of Michigan Press. MacKuen, M.B., Erikson, R.S. and Stimson, J.A. (1992). Peasants or bankers: The American Electorate and the U.S. economy. American Political Science Review 86: 597--611. Tufte, E.R. (1978). Political control of the economy. Princeton: Princeton University Press. [ 181 ] Public Choice 97: 401-428, 1998. © 1998 Kluwer Academic Publishers. 401 Ideology and the construction of nationality: The Canadian elections of 1993 * MELVIN J. HINICH1, MICHAEL C. MUNGER2 & SCOTT DE MARCHI3 1Department of Government, University of Texas-Austin, Austin, TX 78712, USA; e-mail: hinich@mail.la.utexas.edu; 2Department of Political Science, Duke University, Durham, NC 27708, USA; e-mail: munger@acpub.duke.edu; 3Department of Political Science, Washington University, St. Louis, MO 63130, U.S.A. Abstract. Canada is one nation, but it is in many ways two communities, one Francophone and the other Anglophone. We employ a formal model of "ideology" and analyze how nationality is constructed in people's minds. The magnitude of the changes in expressed "preferences" in terms of ideology depends on the salience of the new issue, the extent to which it confirms with the existing ideological cleavage, and the difference between the perceived status quo on the new dimension and the voter's most preferred alternative. Using data from the 1993 Canadian National Election Study, we consider the relative importance of different policy dimensions in explaining voting decisions among educated Canadians. The issue of Quebec sovereignty, alone, is shown to have significant power for predicting vote choice. A plausible explanation, confirmed here by regression analysis, is that Quebec sovereignty "stands" for other issues in voters' conception of Canadian politics. 1. Introduction In 1967, French president Charles de Gaulle appeared on a balcony of Montreal's City Hall, at the height of Expo '67. General de Gaulle always ended his addresses with a series of vivats, crowd p1easers like "Vive fa Canada", or "Vive la France". That day, whether out of plan or passion, he said something astonishing: "Vive Ie Quebec Libre" (Thompson, 1988: xi). The sentiment itself, widely heard in Montreal, was unremarkable. What angered the Canadian government, delighted the audience, and surprised the * Earlier versions of this paper were presented at the annual meetings of the Public Choice Society, April 12-14, 1996, Houston, TX. Seminar participants at Duke University, George Washington University, George Mason University, Princeton University, University of Kentucky, and University of North Carolina all helped clarify the ideas contained herein. Finally, specific comments by John Aldrich, John Brehm, Dennis Coates, Paul Gronke, William Keech, Peter Lange, Mark Peffley, George Rabinowitz, Lee Sigelman, Genia Toma, Mark Toma, and Peter VanDoren were particularly useful. Error are ours, however. [183] 402 world was de Gaulle's call for the secession of a province (i.e., for a "free Quebec") from a nation as part of a state visit. Why is the split between French and British Canada so powerful, and so central to Canadian politics, that even foreign leaders exploit the division? In this paper, we advance a model of political dimensions, and dynamic conflict, with an eye toward explaining some of the events in Canada. The particular election we will analyze was the 1993 Canadian national elections. The deep political divisions in Canada, however, predate the 1993 election, and in fact predate de Gaulle's outburst, by a century. However, it was always a federation "in which the provinces were expected to be little more than glorified local governments" (Kornberg, 1988: 13). The North America Act created a nation, but it has failed to create a community, as we will argue below. The following section describes the theoretical basis for the importance of "community" in the study of the construction of nationality. The key conception of community for our purposes, is the work by Taylor (1976) and by Schofield (1985) on satisfying the requirements for common knowledge in achieving cooperation. Section 3 presents our model of "ideology", which we argue is the repository of culture and community in a society. Section 4 considers the application of the logic of this model to the Canadian national election of 1993, using survey data from the Canadian National Election Study. 2. Conceptions of "nationality" In his controversial book, Alien nation, Brimelow (1995) defines a geographic and political entity (the nation-state) as having an ethnic basis. Ethnicity is much more restrictive than "culture", an elastic and encompassing concept that requires only a variety of shared experiences or symbols. Brimelow (1995: 203) is quite explicit: Let's start with a definition. What is a "nation-state"? It is the political expression of a nation. And what is a "nation"? It is an ethno-cultural community - an interlacing of ethnicity and culture. Invariably, it speaks one language (emphasis in original). Moynihan (1993: I) is no more equivocal: A nation is "a group of people who believe they are ancestrally related. It is the largest grouping that shares that belief". Why equate nationality with ethnicity and language? Because functioning nations are communities, with shared experiences and beliefs. If we think of nations as large "communities", identity becomes fundamental. A community is not a collection of individuals, but a coherent organism made up of [ 184 ] 403 individuals who think they belong. Achieving cooperative "best" solutions is still a problem in such a setting, but it is an easier problem. Taylor (1976, 1982) defines "community" as having three elements: First, members share beliefs in norms, and have similar preferences. Second, people in a functioning community have direct, complex relations with one another, and expect these relations to continue indefinitely. Third, obligations are viewed as reciprocal, so that all members of a community "owe" one another respect or the promise of aid in a crisis. Schofield (1985: 218) makes a compelling theoretical argument for why communities matter: "The fundamental theoretical problem underlying the question of cooperation is the manner by which individuals attain knowledge of each others' preferences and likely behavior". Now, "attain" seems an odd word; why not "obtain"? However, people cannot always obtain more information, in an asymmetric information setting. Rather, reliable information about preferences and expectations is much like the technical condition of common knowledge in game theory, requiring members attain an equilibrium level of information about each other, through shared experiences, norms, and reciprocal commitments. Community is a knife-edge, perhaps ephemeral, condition: information is either common (in a community), or asymmetric (in any other social context). Taylor, of course, argued that communities must be small, and that large nations could rarely meet his conditions. Schofield is more optimistic: nations may in fact be able to solve the common knowledge problem on a scale larger than Taylor's "communities". The problem is that the solutions are fragile. Differences in language, customs, and political iconography strain the common knowledge basis of cooperation, and crack the information nexus. Hinich and Munger (1994) draw together several strands of thought on the way people's political beliefs are structured, and how those beliefs do not evolve in an incremental or continuous fashion. The key theoretical concept for Hinich and Munger is "ideology", or set of intellectual organizing principles reflected in widely shared beliefs. As we shall see in the next section, Hinich and Munger claim that ideologies are the means nations use for communicating about political conflict. However, these conflicts may also end communication, and cause cleavages among groups within a geographic political jurisdiction, tearing nations apart. These centrifugal pressures derive from fault lines in communities, and the common knowledge basis of cooperation in the formation of beliefs about other groups of citizens. [ 185] 404 3. Ideology: An overview of a dynamic process The advantage of a dynamic approach to spatial modeling based on ideology is that it is possible to generate realistic predictions about change even if voters' preferences about specific issues are fixed. There is no claim that new issues are the only source of dynamic changes in political systems. What we will argue is that the changes we have seen in Canadian politics over time, and the tensions over national identity in Canada today, can be represented as ideological conflict. In adressing stability and change in a macropolitical context, we build on a growing tradition of thought on dynamic processes. Intellectually, this view derives from Riker's (1982, 1986, 1990) "heresthetics". Aldrich (1983, 1995) explores the implications of placing party "activists" in the strategic role of choosing which "new" issues to emphasize. Lupia (1992) considers the role of information and attention in determining the agenda. Jones (1994) argues that there may not be very many distinct "issues" at all. Instead, people apply different, relatively stable, frames to problems over time. Dynamic change then results from reframing of the problem, not from a change in fundamental preferences. Poole and Rosenthal (1997) culminates a long series of papers offering a powerful technique for measuring the dimensions of conflict in the U.S. potential system, and examining how those dimensions change over time. In this section, we outline a view of new issues, and change, developed at greater length in Hinich and Munger (1994, 1997). Because of the structure our notion of ideology imposes on political competition, it is possible to derive specific results on the introduction of new issues. Later, in applying this model to the study of Canadian politics in the 1993 national election, we will claim that fundamental changes in the partisan control can result from strategic action by elites, holding voter preferences constant. 3.1. Specifying context: Ideology The basic theoretical claim of the ideology model is that political debate in a stable system has no more than two or three dimensions, although the "policy space" seems complex. Ideology may be defined either as (a) a set of ideas with implications for "good" policy, or (b) a widely shared belief system.' Neither of these notions requires internal consistency among ideas or beliefs. Instead, an ideology provides a means to organize political thinking, and to let people understand each other. Ideology exists, because issue positions cluster. If we know what you think on defense policy, abortion rights, and environmental policy, we can guess (with a small error) what you think of school lunch subsidies. Further, this [186] 405 clustering phenomenon is not atomistic, so that ideological positions such as "liberal" and "conservative" have similar meanings to different people. If the understanding of clusters of issues is shared, we will call this clustering phenomenon "ideology". Ideology sharply constrains the strategies of candidates (and hence the choices for voters) in the policy space are constrained. Poole and Rosenthal (1997: 12) note: "The basic implication of the constraint hypothesis is that all issues tend to be mapped onto a fixed ordering, or placement, of legislators". Our name, of course, for the "fixed ordering" that organizes political conflict is ideology. 3.2. New issues and the ideology model: "Normal change" The ideology model has been developed by Hinich and Pollard (1981), Enelow and Hinich (1984), and Hinich and Munger (1994, 1997).2 We will give only a summary of the model here. Let us begin by fixing notation. Xik The ideal point of voter i on policy k, i E {l, ... , n}, k E {l, ... , m}. Policy position on the issue k. \!k The weight, or ideological translation term, for issue k. bk Status quo policy for issue k. 7rp Ideological position of candidate p. For simplicity, we will assume the ideological space is one-dimensional. Zi Induced ideal point of voter i in ideological space. Ai mxm matrix of weights, describing salience and complementarity of issues in the ith voter's utility function. n n-dimensional policy space, where "dimensions" come from the set of existing, widely discussed issues. The set of possible issues is, of course, infinite. Voters have preferences over "issues", a n-dimensional space. Each voter i, with ideal point Xi in the policy space n, chooses between two candidates (Alpha and Beta) based on their imputed platforms tWa, w,B} in n. 3 We will assume that the choice is based on a quadratic utility function, and for notation use [] to represent simple Euclidean Distance calculated from vector differences: Wk (la) (lb) [ 187] 406 But the imputed platforms Wa and wp have to come from somewhere. What is the source of voters' beliefthat these positions represent the likely policies of Alpha and Beta if elected? Our claim is that, though voters may have preferences defined over the n-dimensional policy space n, political competition takes place in the pdimensional ideological space <1>, where p n. Consequently, the choices offered the voter (at least, the choices voters can identify, and candidates can commit to) are constrained by ideology. The correspondence, or mapping, from the ideological to policy spaces can be expressed as a linear function of the various ideological dimensions. Though the model can handle multiple dimensions (see Hinich and Munger, 1994), assume p = 1 for simplicity. The imputed platform (for candidate Alpha, for example) can then be written: « (2) The (n x 1) b vector is the set of status quo policies. The (n x 1) v is the set of mappings from the ideological space to policy space. The ideological position of each candidate is drawn from the set of feasible positions (that is, Jrj E n, where in this case Jrj is scalar).4 The elements of v reflect the beliefs of voters that the prevailing ideology has implications for policies. For example, if Vk is large (either positive or negative), the voter believes abstract ideological statements are highly meaningful for policy k. Conversely, if Vk is near zero in absolute value, the issue is not accounted for by the ideology of the prevailing party system. This does not mean voters don't care about the issue. Instead, if Vk = 0, issue k is outside the issues voters associate with the orthodox political debate they hear from parties and candidates. We are now in a position to consider the policy space n, which contains the voter ideal points. Suppose we allow voter i to have ideal point Xi = (WI, (2), and be faced with the choice between two candidates ex and f3 based on their platforms, Jra and Jrp, in the ideological space n. Let preferences around Xi be described by separable ellipsoidal, rather than circular, indifference curves. The correspondence between the policy space n and the ideological dimension n, given by equation (2) above, can then be depicted in the graph in Figure 1. The feasible positions in n are constrained to appear on the space n, with the status quo point in the policy space represented by b = (b l , b 2 ) and the status quo ideological position being b + JrOV. The induced ideal point on the b + JrV line is b + ZiV, the point tangent to the highest feasible indifference curve. The corresponding ideal point in n is Zi. The voter would prefer Xi, of course, but such a position is not feasible for politicians to take given the prevailing ideological cleavage in the society. The choice is between the partisans of the right, who favor Issue I, and the [188] 407 Issue 2 ,. b+1tV "'" ..;;,; ---b+ 1t av : """/ I I Induce~ : : Ideal Point I I b+z. v I Status Quo Point b+1[ov=(~' ""'1/ -~', Is) b+1t Issue 1 : V'l"1111 P II11I1I1 Figure 1. partisans of the left, who favor Issue 2.5 Consequently, in Figure voter i chooses candidate ex over {J, because lTOi is closer (in weighted Euclidean space) than IT fl to Xj. We will find it useful to define more formally the induced ideal point (denoted b + Zj v in Figures 1) of voter i on the ideological dimension. Assuming, for the sake of simplicity,6 separable preferences and equal issue salience (i.e. circular, rather than elliptical, indifference curves). Enelow and Hinich (1984) show that the ideal point (in an m-dimensional policy space) of a voter presented with only those choices associated with existing parties, located along a single ideological dimension, is: Zi = L~l Vk (Xik 2 L..k=l v k "m bik ) (4) That is, the voter's induced ideal point on the ideological dimension is a weighted sum of the differences between the voter's ideal point x and the status quo policy vector b. The weights are a function of the ratio of the ideological mapping terms Vk to the sum of squares of the weights on all issues. The expression in Equation (4) is a simple model of voter choice: Choose the candidate whose ideological position is closest to the induced ideal point Zj on n. We can express the induced ideological ideal point Zj under the more realistic assumption of different salience weights. This requires the addition of an idiosyncratic (indexed by voter) matrix Ai of preference weights. Under these circumstances, we can define Zj as follows: Zj = VTA(x - b) (5) where v, x, and b are column vectors. [ 189] 408 As Hinich and Munger (1997) show, it is possible to derive explicit direction and magnitude predictions for the change in the ideological ideal point Zj resulting from a focus on a "new" issue WJc+l' Intuitively, the content of these predictions can be summarized as involving the mapping terms v, the preference matrix A j , and the perceived position of the status quo bk +1 compared to the voter's ideal point Xk+l' • Mapping terms. Somewhat surprisingly, the larger the ideological mapping Vm+l, the smaller is the change in Zj. In fact, as Vm+l grows large, the change in ideology vanishes completely.7 A moment's reflection confirms the intuition behind this result, however: if the new issue is well accounted for by, and tightly linked to, the old ideology, it does not cause the voter to rethink his ideological position. • Utility weights. The larger the salience ofthe new issue (am +l) compared to those of the old issues, the larger the change in the voter's induced ideological ideal point. In other words, if the "new" issue is important enough, it dominates all the other issues and itself becomes the entire ideological dimension. As we will see, there is some evidence that this is exactly what is happening, or happens from time to time, in Quebec. If, instead, the voter attaches no weight to the issue, then of course it does not change his ideological identification. • Status quo. The larger the difference between the voter ideal point and the status quo on the new issue, the larger the change of the voter's ideology. Notice that this has nothing to do with underlying preferences, which have been fixed, though latent, all along. Rather, the set of issues that make up the political world, and the ideology that organizes the information in that world, are transformed by the introduction of the new issue. 3.2.l. The analytics of normal change Because Zj is induced on the ideological dimension by the preferences of voter i in the n-dimensional policy space n, it is possible to dispense with the presentation of n and represent voter choices solely on n, the ideological dimension itself. Figure 2 depicts an arbitrarily drawn example of the ideological dimension, with far left and far right positions. The voter's ideal Zj is drawn as before; the comparison for the voter is against rra and rrf3. Whichever candidate is closer to Zj will receive i's vote. As Figure 2 also shows, changes in Zj can change i's vote. In our example, i votes for {J, given the original Zj. But suppose that a new issue is introduced. The change in Z can be in either direction, depending on the relative values of xm+ 1 and bm+1. The magnitude of the change will depend on the weight (am+l,m+l) of the new issue in the voter's utility function, and the mapping [190 ] 409 z. to z.': Unsuccessful New Issue z'; to z ,\': Successful New Issue (MIDPOINT) Figure 2. (Vm+l) of the policy onto the ideological dimension. In Figure 2, we have drawn z; and z;" as examples of possible changes in z caused by the strategic introduction of a new issue. As is clear from the figure, the change from Zj to does not change the vote; the change from Zj to Zj" causes the voter to switch from candidate Beta to Alpha. Candidate Alpha, by introducing a new issue matching the particular parameters needed to change voters' minds, has won the election. This completes our discussion of the introduction of new issues when the prevailing party system and associated ideological cleavage accounts for the new issue. In the next section, we consider the implications of new issues when the new issue changes the dimensionality of the space of political debate itself. z; 3.2.2. New issues and the :,pace of political competition: Realignment The preceding section addressed new issues that "fit" the prevailing context, or the ideological frame of reference. To understand larger issues, particularly the nature of nationhood, it is necessary to think more fundamentally. The consideration of the strategic addition of new issues may make the political system unstable. What are the implications for political strategy of such divisive, and potentially destructive, "new issues'''? As is well known, majority rule election processes always have a determinate Condorcet winner at the ideal point of the median voter, provided the relevant strategy set is one-dimensional and voter preferences are singlepeaked. While one might quarrel with the assumption of single-peakedness, the obvious problem lies with the assumption of unidimensionality, particularly when we allow for strategic introduction of new dimensions. McKelvey (1976) showed that if the core of the election game is empty, and voting is sincere, then it is possible to choose an agenda (sequence of pairwise votes) that lead to virtually any outcome in the policy space. [ 191 ] 410 If a party out of power, or a cross-cutting coalition of disgruntled members, successfully introduce a new issue, the specter of chaos is raised. Those advantaged by the status quo will try to use the institutions of the legislature to defend stability, of course. Still, institutions cannot generically solve the problem of conflict over dimensionality for very long without (nearly) universal agreement. Riker's (1982) example of the U.S. Civil war makes this point very clearly. By raising the issue of slavery, the coalition that became the Republicans after 1854 managed to annihilate the dimensions of the previously (fairly) stable political space. It wasn't clear they would "win" as a result, but they did ensure that if they lost it would be in a different game. The problem was not so much that there was a new issue, because this happens all the time. Rather, chaos arose because the new issue could not be handled within the existing ideological framework. Thus, slavery effectively raised the dimensionality not just of politics, but of the strategy space. Poole and Rosenthal (1993) claim that only a fundamental change in the primary dimension of the ideological space is a "realignment". Thus, realignments should not be confused with temporary circumstances when a "new" (for Poole and Rosenthal, second) dimension is added to the ideological space. Politics will look different (more chaotic) when there are two dimensions, but the original fundamental cleavage will return unless there is a realignment. The process of realignment when there is a genuinely new issue can then be summarized as follows: [W]ell before a realignment, congressional voting should be stable and organized around the cleavage of the last realignment ... This means that the policy space is stable - the same dimension(s) account for voting over time, and legislator ideal points should show little change from Congress to Congress. A new issue then emerges that splits the political parties internally and begins the process of polarization. This can be modeled as a new dimension, orthogonal to the stable set from the last realignment, across which both parties become increasingly polarized. We should see this polarization take place in two ways ... newly elected representatives from the same party should take relatively polarized positions on the new dimension ... [Incumbents] should exhibit movement that, relative to their earlier positions, resulted in polarization. As the process continues, more and more of the voting is concerned with the new issue, so that the old, stable set begins to wither away (Poole and Rosenthal, 1993: 16). To summarize: the ideological linkages between the new issue and all other issues supplant the old linkages, which supported the old ideology. Not all the roll calls are "about" the new issue, of course. Instead, as other matters [192 ] 411 increasingly link to the new main cleavage, the symbolic relations of other issues to the main cleavage increasingly become the basis for choice. 4. Quebec sovereignty and the construction of nationhood The 1993 Canadian general election is a fascinating case of a "new" issue changing the politics of a nation. The issue we have in mind, the sovereignty of the province of Quebec, is of course not really new. Tensions between Francophone and Anglophone Canadians predate the existence of Canada as a nation. However, consider the definition of "issue" in Hinich and Munger (1994: 111): Issues: Social problem large numbers of citizens care about that (1) politicians talk about in (a) public, (b) to contributors, or (c) among themselves, OR (2) the press talks about either because some interest group wants it discussed or because citizens care about it. Many people cared a lot about Quebec independence before the 1993 election. In several previous elections the possibility of Quebec becoming a separate nation had been raised. What happened in 1993 was that large numbers of people seemed to care, politicians debated the "issue", and the media covered the story. 4.1. A brief history of Canadian realignments We hope in a few diagrams to capture some of the essence of the main conflicts in Canadian politics in the last century. There are several excellent histories of Canada, and Quebec, including McRae (1964), Kornberg (1988), and Bothwell (1995), as well as some significant analyses of the problem of Quebec nationality, including Brimelow (1986), Kornberg and Clarke (1992), and Gibson (1994). Our particular view of Canadian political history, and the process ofrealignments, derives from Johnston (1993). Johnston argues that there have been three major political epochs in Canada, the first extending from 1866-1896, the second lasting from 1897-1957, and the present epoch beginning around 1962. We will argue that a new, fourth epoch may be beginning, with the "new issue" being the difficult old issue of the place of Francophone Quebec in the Canadian provincial system. The interesting thing about Johnson's account is that it fits so well with the intuition of the ideological model we have outlined in the preceding sections. The first realignment in Canadian political history began around 1896. From the independence of Canada in 1866 until the late 1890s, the nature [ 193 ] 412 and form of the new country were debated, with a focus on urban-rural policies, monetary policy, and tariffs. According to Johnston, however, the chief organizing cleavage was the conflict over the public school system. The status quo at independence in 1866 was hostility toward a separate Catholic school system, with public schools being primarily affiliated with Protestant denominations or the Church of England. The Catholic-Protestant division corresponds fairly closely to the French-English split, but the actual policy question was whether Catholic schools should be allowed as a parallel public school system. By the early 1890s, this division had become much less important as an organizing principle. The Conservatives (previously the Confederation Tories of Upper Canada, before independence) had always supported lands and funding for both Catholic and Protestant church schools. The opposition had come from the Liberals, but by 1895 the Liberal party, too, had dropped its opposition. More important, after 1897 Canada found itself on the verge of military conflict with the U.S., and other rival nations. Beginning in about 1901, the Venezuelan debt crisis tested the resolve of the U.S. in defining the limits of the "Monroe Doctrine", and by 1904 there were significant tensions between the Americans and Great Britain. 8 Given British interest in Canada, and the strategic importance of the lakes and rivers on Canada's southeastern border, there seemed little question that this war would be fought in part on Canadian soil. Johnston (1993: 276-277) describes the change in political dimensions as follows. In this period Conservatives tended to be the party of separate [publicly funded] schools ... The Liberals were the traditional party of opposition to the religious establishment ... Operating at right angles to all of this was the issue of imperial relations, specifically the moral claims of the British Empire ... Before the 1890s, the rest of the Empire seemed too distant to matter much ... By the 1890s, however, imperialist sentiment was waxing and Britain began to look upon the self-governing colonies and dominions as possible co-principals in imperial ventures. Canada was driven to acquiesce in such ventures by practical necessities of statecraft ... If Canada was to have any hope of negotiating successfully with the USA it could do so only as part of a rival concentration of power, the British Empire ... But if Canada wanted to enlist the Empire on its behalf, it had to be willing to reciprocate. This many Canadians were happy to do ... But not all Canadians saw things this way: Catholics ... were less enthusiastic about the Empire than were Protestants. To Catholics a warm external-policy response [194] 413 Separate Schools No Separate Catholic Schools with Equal Funding Figure 3a. to the Empire paralleled pressures within the country toward cultural homogenization ... (emphasis added). We have adapted Johnston's discussion of this shift in issues, and the consequent realignment in the political system, in Figure 3. In Panel (a), we see a single dimension of conflict, over the "issue" of separate schools. However, many other issues also cleaved along the same line, so this dimension came to be the organizing principle around which groups formed and argued. The extreme Protestant voters favored no separate Catholic schools; the Liberal party for decades opposed "church" schools. The Catholic voters, seeking to control not just their schools but their culture, favored a separate system; the Conservative party tended to favor greater Catholic school autonomy and public funding. In Panel (b), however two things have happened. First, the differences between the parties on the issue of separate schools largely disappeared. Both parties now accepted, or actively favored, separate public schools for Catholics and Protestants. Consequently, the schools issue disappeared as an organizing principle. The "new" issue over which coalitions split was support for the British Empire. Interestingly, the Conservatives (before more closely allied with the Catholics) were the party of Empire; Liberals were largely neutral, and in many cases actively opposed imperial unity and military cooperation. 9 The result, in Johnston's (1993: 278) view, is that "the polarity was reversed": As the schools issue became less important (in terms of our model, the difference between the status quo b and voter's goals Xi was reduced), Catholics came to support the Liberals, and Protestants found the Conservatives much more attractive. By the end of WWII, the "support for Empire" dimension had little meaning for most Canadians, because there was little organized opposition to participation in the Allied war effort. There was still an external vs internal element to political conflict, but once again the party polarity had "reversed": Canadian politics was now largely organized around a dimension where geographic region and industrial interests coincided. The eastern, more urban, [ 195] 414 Active Military Alliance with British Empire ~: 1'~" ~ ff \ \ # .,6-c~ ~(1 .,<;-'11 g \ ./ \GC~ ..... ---4it-----l ----e-----I 'y" No Separate Catholic Schools \ New t \ Ideological § \ Dimension ff \,~g , Separate Schools with Equal Funding . t '..... 1 Take NeutraJ, Passive Military Stance Figure 3b. provinces were much more heavily industrialized, but faced import competition from manufactured "secondary" products from other nations. The eastern provinces (and the Conservative party) favored high tariff barriers and other import restrictions, implying an "internal" focus for the Conservatives in the 1950s. The western provinces were rich in minerals and agriculture, but sought increased export markets for these nascent "primary" industries. The western provinces (and the Liberal party) pursued a policy of lower trade barriers, and subsidies for mining and agriculture. This (pre-1957) situation is depicted in panel (a) of Figure 4. Ease Import Restrictions on Consumer Goods Tariff Barriers for Manufactured Products Figure 4a. Interestingly, the Liberal party had always been, in Johnston's (1993) terms, "the only option for a free trader". But by 1956 there was actually little difference between the parties. The Liberals had earlier proposed (but lost) a set of free trade policies in 1891 and 1911, and had managed to win tariff reductions in 1936 and 1938, "but only back to the levels from which a Conservative government had raised them a few years before" (1993: 279). Conservatives had consistently favored import restrictions of almost all sorts, and viewed the "National Policy" (favoring domestic secondary indus- [ 196 ] 415 Subsidize Mines Agriculture 8< Other Prinuuy Industries Hi~ Tariffs _ _ _ _ _ .....1 _ _ .... , Pre-I9J7 ~ Ideology IIl ';f Ul 1957 Z.... ~-. ~.: f .J' I v" , ! No Subsidies for Pritruuy btdustries Figure 4b. tries through tariffs, quotas, and subsidies) as a key part of their program. Until the 1950s, however, the National Policy had not included much support for the developing primary industries in the western provinces. In fact, when mining and agricultural subsidies had been proposed, the Conservatives had opposed them. Consequently, what happened in the 1957-1962 period is surprising: the Conservatives deemphasized (though they did not abandon) protectionism, and moved from opposing to supporting investment in the Canadian west. As panel (b) in Figure 4 shows, the Liberals' changed their platform very little in this period, but the Conservatives moved away from the eastern provinces and shifted toward the west. We have drawn simple indifference curves in panel (b) for the eastern, and western, provinces, and have included the implied ideological ideal points for the western provinces, Zwest for 1957 and 1962. As the reader can see, given the pre-1957 ideological cleavage along a dimension of free tradeprotectionism, the western provinces clearly favored the Liberals. After the 1957-1962 realignment, however, there is a new ideological dimension defined by the positions of the two parties. Given the new constraints for choice, the induced ideal point z~; is much closer to the new location of the Conservative party. Again, it is useful to summarize Johnston's (1993: 280-281) account: What changed? First, the 1950s saw an agricultural crisis. This suggested that the resource axis [gained salience]. But that alone would not have produced a realignment, at least not one benefitting either old party, [ 197] 416 given their pre-1957 policy orientations. The second key ingredient was change in the Conservative leadership. The Conservative party had, in a sense, run out of its traditional alternatives. It had tried on several business-oriented imperialists. It had genuflected toward Quebec. It had even had several Westerners as leader but none who was prepared to articulate a credible agrarian programme .... Finally, in 1956, the rank and file of the party chose someone who did epitomize the Prairie West, John Diefenbaker. Let there be no mistake: this was not a rejection of the Conservatives' beloved National Policy. In fact, Diefenbaker's program of subsidy, generous credit, and crop supports for agriculture were successful (politically, not economicallylo) precisely because Canadian wheat could not be successfully exported: "For the first time, Central Canadian consumers were forced to subsidize a Western industry which had no realistic prospect of comparative advantage in the world market" (Johnston, 1993: 281). So, there was little economic basis for this new issue emphasis. What happened was that the Conservatives dropped their regional emphasis in favor of a logically coherent national program of "Canada first!" Farmers in the western provinces, who had been a strong Liberal bloc for two generations, now became the Conservatives' strongest electoral base. Conservatives didn't change their position on the dimension (protectionism) that had organized politics in the 1950s, and voters didn't change their preferences in any way. What happened was that the Conservatives emphasized a "new" issue, agrarian support policies, which had before received only tepid and occasional advocacy by the Liberals. The account we have given of these three epochs in Canadian political history is highly selective. However, it captures a significant dynamic aspect of the evolution of policies and coalitions. In the next section, we present some measures of current issues in Canadian politics, taken from survey responses during the 1993 Canadian national election campaign. 4.2. Some measures of issues in the 1993 Canadian National Election Let us recall how we got here. The main theoretical claim of our model is this: Even if voters have fixed preferences, changes in how issues map to the ideological space can result in different electoral outcomes. That is not to say voters never revise their beliefs. Instead, we are pointing out that fundamental changes in political conflict, partisan control of the legislature, and even the essence of nationality, can result from strategic action by political elites, holding mass preferences constant. [198] 417 In essence, by introducing a new issue (or greatly increasing the importance of an issue), parties have an opportunity to change the location of the voter's induced ideal point (Zi) even though the underlying issue positions remain the same. This an important realization for the field of electoral politics, as most of our models are badly blunted when forced to cope with large electoral shifts. The 1993 Canadian general election is fascinating, in large part due to the drastic changes that resulted from the reemergence of the issue of Quebec independence. While it is certainly the case that voters both inside Quebec and in the rest of Canada likely had established positions on Quebec independence, the importance of their positions in the national elections of the decade preceding 1993 was minimal. Regressions performed on the data of the 1988 Canadian Election Survey show that a combination of familiar domestic issues coupled with concerns over NAFTA predict electoral choice, and that Quebec independence was not a salient issue. In the terms of our model, the ideological transform v had a near-zero coefficient for the issue of Quebec independence. Thus it was largely unexplained by the existing ideologies. With the increased importance of the issue in the 1993 elections, our model suggests that the v vector would have a very large coefficient for Quebec independence. This is a result of both increased public attention to the issue of independence, an increase in the flow of information from the parties that was narrowly tailored to the issue, and the fact that (at least one of) the parties clearly diverged in their positions on this question. One would also expect the issue to have the greatest importance inside Quebec, both due to an increase in the salience of the issue across the population (Ai in the model), and the greater proportion of voters who belong to the Quebecois party. For the purposes of analyzing the data from the 1993 survey, we focused on the difference between the residents of Quebec and the rest of Canada to highlight the role of a single new issue in the voting calculus. Though many Canadians had stable positions on the issue of Quebec independence prior to 1993, it was not until it became a greater part of the existing parties' ideologies that the issue could affect choices. In this example of how the model might account for dramatic political shifts, our experiment consisted of two steps. A factor analysis was performed on issues taken from the 1993 Canadian Election survey, and issue bundles were specified according to the results of the factor analysis. What we looked for in this case, however, was not change over time, but rather changes wrought by the campaign on residents of Quebec as differentiated from the rest of Canada. [199] 418 There is some precedent for these differences; Kornberg (1988: 21) claims that responses to a series of surveys on political deference (1979 and 19831984) "divided neatly into two factors. One factor dealt with attitudes toward government and its policies, and the other toward military service and obedience to authority". However, Kornberg confirmed what many others had found: Quebec is different, with much less of political differences in opinion being accounted for by the factors that organize the rest of Canada. Since we do believe that both socialization and the campaign (and the dissemination of information more generally) have an important place in voting (i.e., through the v vector), we hypothesized that residents of Quebec would be particularly sensitive to the issues revolving around Quebec, and that the introduction of Quebec sovereignty into the election would result in the formation of a drastically different ideological mapping. The 1993 data is a good test case for this effect, as issues centering on Quebec independence were particularly salient during the election (Table 1). What we discovered in the factor analysis was interesting - the second factor revolved almost completely around the issue of Quebec sovereignty, while the first factor was inclusive and ordered along a left-right ideological dimension. For the larger population, most of the issues in the first factor were significant and salient in our regressions, with the exception of feelings toward feminists (due in large part to its collinearity with feelings toward minorities and homosexuals). The second step was to see if regressions on electoral choice support our conclusion that this second dimension based on independence actually loomed large in the decision-making calculus of voters. The first regression table shows results for the entire Canadian population, while the second focuses solely on Quebec residents. We focus first on educated respondents (those who report attending at least some college) to get an idea of the overall importance of issues among relatively sophisticated voters. As the first regression shows, the items from the first factor from the above factor analysis are the main predictors of electoral choice (Table 2). Predictably, when we focus on residents of Quebec, all issues from the first factor drop out of the model except for Quebec sovereignty and aboriginal rights. Of these, sovereignty has more salience. In terms of our model, the v vector would be composed of very small coefficients for all issues save independence, even though voter positions on these issues are important (and they may care about them as expressed by the Ai matrix of weights). The two issues that were weakly related to Quebec sovereignty in the factor analysis are somewhat salient, but not significant at the .1 level (crime and feelings toward the U.S.) (Table 3). [200] Table 1. Results of factor analysis on 1993 Canadian data Eigenvalues of the correlation matrix: Total = 8. Average = 1 Eigenvalue 2.6356 Difference Proportion Cumulative 1.3914 0.3295 0.3295 2 3 4 5 5 8 8 1.2442 0.2343 1.0098 0.1515 0.1262 0.8583 0.8224 0.1939 0.1028 0.6285 0.2116 0.0786 0.4169 0.0326 0.0521 0.83843 0.0359 0.1073 0.6112 0.7185 0.8213 0.8999 0.9520 1.0000 0.1555 0.4850 0.0480 Initial factor method: principal components ~ Factor pattern Factor 1 Factor 2 Factor 3 Quebec Sov Aborig.S Less Gov Gay Rights Feminists 0.07265 0.75158 0.15738 0.75574 0.69394 0.83877 0.22994 0.05389 -0.34641 0.01983 0.11131 0.94023 -0.13105 -0.10186 Minorities Crime Feel U.S. 0.78486 0.43735 0.42510 -0.09158 0.11223 -0.41452 0.41476 0.02409 0.08330 -0.27919 +:- \0 420 Table 2. Logistic results a on sample from All of Canada b Variable Parameter Standard Asympt. estimate error t-stateC Intercept 0.8455 0.7277 Aborig 0.0583 0.0072 1.162 8.097** Less Gov 0.0343 0.0753 0.456 Gay Rights --0.0224 0.0075 -2.987** Feminists --0.0030 0.0073 Minorities --0.0172 0.0087 --0.429 -1.977** Crime -0.1970 0.0645 -3.031 ** 0.0211 0.0070 -2.857** Feel U.S. Association of predicted probabilities and observed responses Concordant =84.6% Somer's D = 0.697 Discordant = 14.9% Gamma = 0.700 Tied = 0.4% Tau-a = 0.127 (45936 pairs) c = 0.849 Note. A vote for Bloc Quebicoise was coded "1"; all other votes as "0". a The issue of Quebec sovereignty was not included because it was only asked of Quebec residents. b The sample for this regression is all respondents with "Education" 2: 7. (n = 710). C * implies significance at the .1 0 level; ** implies significance at the .05 level, using a two-tailed test. The large magnitude of Quebec sovereignty, and the overall good fit of the model, supports our general notion of the role of the introduction of a new issue in elections. As a new issue becomes more salient to a given part of electorate, it should be the case, given our model of ideology, that other issues drop out of the decision-making process. This clearly happened in Canada for the residents of Quebec during the 1993, and analysis of the 1988 election shows the same general effect at work for the entire Canadian electorate over the issue of NAFTA. What makes the Canadian election unique is the (nearly complete) independence of Quebec sovereignty as an issue. In terms of the diagrammatic exposition we used to illustrate the 18961897 and 1956-1962 realignments, the Quebec "dimension" might become the new ideological organizing principle. What are the main ideological conflicts in Canadian society today, and over the past twenty years? Interestingly, [202] 421 Table 3. Logistic results a on sample from Quebec Onlyb Parameter Standard Asympt. Variable estimate error t-state C Intercept -4.0865 1.2750 -3.205** 0.8686 0.0217 0.1415 6.143** 0.1220 -0.00248 0.1401 0.0117 0.871 -0.212 -0.0117 0.0111 -1.054 0.0139 -0.806 -\.505 Quebec Sov Aborig.s Less Gov Gay Rights Feminists Minorities Crime Feel U.S. -0.0112 -0.1641 0.0184 0.0114 0.1096 0.0123 1.904* 1.496 Association of predicted probabilities and observed responses Concordant = 89.1 % Somer's D = 0.783 Discordant = 10.8% Gamma =0.784 Tied =0.1% Tau-a =0.393 (5680 pairs) c = 0.892 Note. A vote for Blow Quebecoise was coded "I "; all other votes are "0". a The issue of Quebec sovereignty was asked only on Quebec residents. b The sample for this regression is restricted to Quebec residents with "Education" :::: 7. (n = 151). C * implies significance at the. 10 level; ** implies significance at the .05 level, using a two-tailed test. Canadians do not seem "ideological" in classical left-right economic sense (Kornberg, Mishler, and Smith, 1975; see also Clarke and Stewart, 1992, for a review and update). There are, however, issues and values that appear to organize political conflict in Canada over this period. The single most important dimension appears to be that identified by Clarkson (1978): Continentalism vs nationalism. In paraphrasing this argument, Kornberg (1988: 19-20) says: Nationalism is the opposition ideology. The reigning ideology, whether it is called anti-nationalism or continentalism [is based on arguments by liberal economists] .. , To be sure, they are not the only source of anti-nationalism, but they are the chief intellectual force legitimizing the continentalist economic and political policies pursued by Canadian governments since World War II. [203 ] 422 Table 4. Predicted - Actual patterns for Quebec voters Panel A: Best model- Classification with all variables, from Table 3. Educated voters only. Actual vote For Bloc For some Predicted vote Quebecoise other party For Bloc Quebecoise For some other party 66 14 16 55 Column correctly classified: 83% 77% Panel B: Minimal model- Classification with "Quebec separatism" only. Educated voters only. Actual vote For Bloc For some Predicted vote Quebecoise other party For Bloc Quebecoise For some other party 68 12 14 57 Column correctly classified: 85% 80% Panel C: Minimal model - Classification with "Quebec separatism" only. All voters sample, regardless of education. Actual vote For Bloc For some Predicted vote Quebecoise other party For Bloc Quebecoise 337 71 70 297 For some other party Column correctly classified: 83% 81% What about 1993? Our factor analysis (in Table 1, above) found three distinct factors, or dimensions, appeared to divide public opinion in Canada in 1993. The first was a liberal-conservative division on social issues, such as rights of aboriginal peoples and other minorities, homosexuals, and women. The second dimension appears to be Quebec separatism, though fear of crime and resentment of the u.s. also load on this dimension. The third dimension, though weak statistically, does appear to have some explanatory power. This third dimension is the desire for "less govemment".l1 [204 ] 423 If we are correct in claiming that independence for Quebec is the primary issue that organized choice in the 1993 election, then position on that issue, alone, should perform well in classifying voters' party choices. We can test this claim, though only crudely, by comparing the classificatory power of several models. These comparisons are presented in Table 4, which for sake of conserving space presents only 2 x 2 tables of predicted vs. actual vote choices by citizens in Quebec province. Panel A of Table 4 shows the pattern of classification success of the "best" model (from Table 3) in predicting whether a voter selects Bloc Quebecoise (BQ) or some other party. The model performs adequately, overall classifying 80% correctly, and substantially outperforming the naive model prediction of everyone voting for BQ (53%). Classification is slightly more accurate in selecting those who will vote for BQ, but the difference is not substantial (83% compared to 77%). Panel B presents a stripped-down version of the model: the only right hand side variable is support for Quebec sovereignty, with an intercept term. As sometimes happens in ML estimation, less is more: the classification proportions go up, across the board, with overall classification now equal to 83%. In terms of our model, this means that, even if voters care about other issues, either they aren't using information about the other issues or party positions imply these other issues map onto the "Quebec sovereignty" dimension with little perceived error. As we noted earlier, these results are for educated respondents (reported at least some college). It is not clear whether sophistication (which was not a parameter in the model we advanced in Section 3) should make voters more or less likely to respond to salient issues. Panel C repeats the "strippeddown" version, using all respondents who answered the party choice and Quebec sovereignty questions. Interestingly, there is no qualitative difference in the results: overall classification is 82% (compared to a model prediction of 53%). The reader may shake his or her head at this point, and ask, "Why is it surprising that Quebec voters use Quebec sovereignty to vote for, or against, the BQ?" It is not surprising that voters use sovereignty at all, but we think it is at least interesting that we can predict their vote choices using nothing else. Furthermore, as we see from comparing Panel A and Panel B of Table 4, adding other explanatory variables does not improve classification, even among the most sophisticated voters! 12 What may happen, in terms of the contest over the dimensions the divide Canadian voters? We can make a guess which, though tentative, is consistent with the process of adjustment we have outlined above. If there are, indeed, three separate dimensions developing in Canadian politics, the key to the [205 ] 424 Civil Rights (Aboriginals, Gays, Women, Minority) More ~ 0("> Rights r------- . . . .~&Quebec ;'1 I ~ 1,- ;' ;' ;' ;' I I / -~,t. Less Intervention ;'1-;'/;' ;'1 q,.f. (,0-# \. ~ f,~;'-t ;' ;' ;'/;';' ;';' ~ I;, .------- ;' ~ry 0<i>", I I I / ;' ;' ;' ;' ;' \~ ;' V-'Oq, , . / / ----=-" ~ ;' ;' ;' More Regulation ,--------------, Existing primary dimension of cleavage is the difference(s) between Liberal & Conservative parties on civil rights and economic policy. But the Quebec separatism dimension may come to stand for all other conflicts. I" _____ JI 1-1/ t Canadian Nationalism ;' ;';'/;' r- -J.-- /;' "'c. ;' ;' ."i' o.oc; I --"7----( ,..----I---..J. Extent of Government Intervention in Economy ;' ,.-~ ;' ;' ;';' ;' ;' ;' I I Separatism / / -II No Special Protection Figure 5. future will be the contest over the dimension that dominates. One possible depiction is in Figure 5. The two solid lines depict orthodox political conflict within a viable national construct. More or less civil rights for minorities, and more or less government intervention, are standard conflicts in most democracies. Liberals tend to support more regulation, and more civil rights, and conservatives argue for less regulation and the absence of special treatment or affirmitive action programs. The third dimension, the dotted line in Figure 5, is position on a separate and sovereign Quebec. BQ is liberal, but hardly extremist, in the political space described by the two solid lines. They might easily become part of a liberal coalition, and share power in a national government in Ottawa. If, however, BQ chooses to emphasize the dotted line above all other differences, there is no stable or predictable outcome. In terms of the model we have advanced, a radical BQ might follow one of three strategies: (I) Add Quebec sovereignty to the political debate, with an understanding that the mapping terms Vi for many voters would accommodate this dimension as the primary organizing principle in politics. On its face, this seems unlikely to succeed, since the liberal and conservative alternatives are already well-defined, and appear orthogonal to Quebec [206 ] 425 separatism. It is important to remember, however, that Canadians are at almost only weakly "ideological", in the sense of conforming to standard liberal-conservative cleavages. If Clarkson (1978), and Kornberg (1988), are correct, then nationalism-continentalism is the omnipresent, though latent, potential cleavage. Regionalism is but one more step, and independence for Quebec is nothing but an extreme form of regionalism. (2) Represent Quebec sovereignty as a very large change in policy, one which would radically alter the economy and culture of Quebec province for the better. In terms of our model, this means that BQ would have to persuade voters that the status quo is dangerous, or unsupportable, as insidious encroachments on French culture and language continue. If the difference Xj - bj is large enough, Quebec voters may turn in droves toward some more separatist, autonomous relationship with the rest of Canada, even if Quebec separatism does not become the primary dimension of conflict. (3) Depend on high salience of the sovereignty issue. Though it seems trivial, the driving force behind any issue may be its comparative importance. We have conceived of the Ai as being invariant over time, but voters do alter their beliefs and goals. If the salience of sovereignty is large enough, it will swamp all other issues as the primary organizing principle, even if literally no other issue maps onto it. 5. Conclusions In this paper, we have argued that nationalism is a product of beliefs, more than borders. The dimension of identity, and self-definition is latent in all political communities. We have discussed some of the reasons that community, of language, experience, and identity, are crucial to the functioning of a nation. The model outlined above seeks to give some purchase on the question of how people may change these perceptions of shared goals. In considering the Canadian national election in 1993, we found that there appear to be three distinct dimensions vying for primacy in peoples' minds. The first two, civil rights for minorities and the extent of government intervention in economics, are found in most Western democracies, and are consistent with the survival of Canada as a nation. The third dimension, however, is Quebec sovereignty. Our primary empirical finding is striking: support for or opposition to Quebec sovereignty, alone, can classify upwards of 80 percent of voters' choices in the province. Canada may soon break apart. [207] 426 Notes 1. On the importance of, and difficulty understanding, belief systems see Denzau and North (1994). 2. The first suggestion that there might be several "spaces" of political conflict can be found in Ordeshook (1976). Ordeshook's "basic space" is somewhat different from our notion of ideology, but it was an important step. The contribution of Hinich and Pollard (1981), and later work, was to formalize the relation between the policy space and the ideological space, and suggest a consistent and defensible theory for voter choice. 3. We adopt the convention that lower case Greek letters are points, and upper case letters are spaces. Thus, for policy: W E Q; for ideology: 7r E IT; and so on. 4. If P = 2, the 7rj E IT, IT = IT 1 x IT2, and so on in higher dimensions. 5. This theoretical discussion may sound similar to a related perspective, "directional" theory, advanced by MacDonald and Rabinowitz (1987), and Rabinowitz and MacDonald (1989). The key difference, and the main point of contention, between the two the theories, is whether it makes sense to conceive of issues as being linked (as we have), or claiming issues are evaluated in isolation because individual belief systems are diffuse (as MacDonald and Rabinowitz have claimed). The theory of ideology claims that there are important linkages, across issues, in the way that issues are presented. The reason may be budget problem (the only way to increase spending in one area is to decrease spending elsewhere), or pure ideology (defense spending is bad, social welfare programs are good, for liberal ideologues, and the only way to commit is to claim to share beliefs). 6. The only difference implied by the use of circular indifference curves is that the matrix of utility weights is all ones; that is, A = I. 7. IimV m+l-+OO(L1z) = 0/1 = O. 8. Venezuela had enormous debts to Germany and Great Britain, and refused to submit to the edicts of the Hague Tribunal to pay. The Europeans imposed a naval blockade in 1902, shelled Venezuelan ports, and destroyed several small naval craft. The Venezuelans capitulated, and the debts were restructured in such a way that all creditors had some chance of being paid. In 1904, the situation was virtually repeated in the Dominican Republic. This time, however, the government in Santo Domingo made a separate deal with the U.S., one aspect of which was giving debts to U.S. creditors a "senior", or privileged, position. The Europeans protested, and tensions were high. Roosevelt laid out his "big stick" corollary to the Monroe Doctrine, transforming the U.S. pledge of preventing intervention to an implied threat that the U.S. might intervene unilaterally (Morris, 1982: 349-350). 9. It is worth noting that the Liberals were in control over the period in question (Liberals controlled the government from 1896-1911). 10. Kornberg (1988) notes that the National Policy and its elaboration in agricultural subsidies may have hurt the Canadian economy. Because tariffs kept other products out, American firms could "invest" in branch plants to produce consumer goods without significant competition. There are claims that there has been direct interference in Canadian policies by U.S. firms bent on maintaining Canada as a captive market for consumer products, relegating Canadian-owned industry to relatively labor-intensive industries such as clothing, textiles, and footwear. 11. Migue (1997) claims that both federal and provincial governments are trying to woo voters in Quebec province, with the effect being an expansion of government size compared to optimum. This may explain the importance of this dimension as a separate organizing principle. [208 ] 427 12. Using the full model for all voters (running the regression in Table 3 on a sample that includes college and noncollege-educated respondents) gives a classification of only around 75%, even worse than that in panel (a). References Aldrich, J. (1983). A spatial model with party activists: Implications for electoral dynamics. Public Choice 41: 63-100. Aldrich, J. (\995). Why parties?: The origin and transformation of party politics in America. Chicago: University of Chicago Press. Bothwell, R. (1995). Canada and Quebec: One country, two histories. British Columbia: UBC Press. Brimelow, P. (1986). The patriot game: National dreams and political realities. New York: K. Porter Books. Brimelow, P. (1995). Alien nation: Common sense about America's immigration disaster. New York: Harper Collins. Clarke, H. and Stewart, M. (1992). Light anchors, choppy seas: Social structure and party choice in Canada. In M. Franklin, et al. (Eds.), Electoral change: Responses to evolving social and attitudinal structures in western countries. New York: Cambridge University Press. Clarkson, S. (1978). Anti-nationalism in Canada: The ideology of mainstream economics. Canadian Review of Studies in nationalism 5: 45-65. Denzau, A. and North, D. (1994). Shared mental models: Ideologies and institutions. Kyklos 47: 3-32. Enelow, J. and Hinich, M. (1984). The spatial theory of voting. New York: Cambridge University Press. Flanagan, T. (1994). Invasion from the right: The reform party in the 1993 Canadian election. Papers in Political Economy, No. 42. Political Research Group: University of Western Ontario, London, Ontario. Gibson, G. (1994). Plan B: Thefuture of the rest of Canada. British Columbia: Fraser Institute. Hinich, M. and Munger, M. (1994). Ideology and the theory of political choice. Ann Arbor: University of Michigan Press. Hinich, M. and Munger, M. (1997). New issues and the dynamics of political change. Typescript. Department of Political Science, Duke University. Hinich, M. and Pollard, W. (1981). A new approach to the spatial theory of electoral competition. American Journal of Political Science 25: 323-341. Johnston, R. (1993). Issues and party allignments: A review with Canadian examples. In A. Breton, et al. (Eds.), Preferences and democracy, 265-286. Norwell, MA: Kluwer Academic. Jones, B. (1994). Reconceiving decision-making in democratic politics. Chicago: University of Chicago Press. Kornberg, A. (1988). Politics and culture in Canada. Ann Arbor, Center for Political Studies, Institute for Social Research, University of Michigan. Kornberg, A. and Clarke, H. (1992). Citizens and community: Political support in a representative democracy. New York: Cambridge University Press. [209] 428 Kornberg, A., Mishler, W. and Smith, 1. (1975). Political elite and mass perceptions of party locations in issue space: Some tests of two propositions. British Journal of Political Science 5: 161-185. Lupia, A. (1992). Busy voters, agenda control. and the power of information. American Political Science Review 86: 390-404. MacDonald, S. and Rabinowitz, G. (1987). The dynamics of structural realignment. American Political Science Review 81: 775-796. McKelvey, R. (1976). Intransitivities in multidimensional voting bodies and some implications for agenda control. Journal of Economic Theory 30: 283-314. McRae, K.D. (1964). The structure of Canadian history. In L. Hartz (Ed.), Thefounding of new societies. New York: Harcourt Brace. Migue J.-L. (1977). Public choice in a federal system. Public Choice 90: OOO-Q()(). Morris, R.B. et al. (Eds.). (1982). Encyclopedia of American history. 6th edition. New York: Harper & Row. Moynihan, D.P. (1993). Pandaemonium: Ethnicity in international politics. New York: Oxford University Press. Ordeshook, P.C. (1976). The spatial theory of elections: A review and critique. In I. Budge, et al. (Eds.), Party identification and beyond. New York: Wiley. Poole, K. and Rosenthal, H. (1993). Spatial realignment and the mapping of issues in U.S. history: The evidence from roll call voting. In W. Riker (Ed.), Agendaformation, 12-40. Ann Arbor: University of Michigan Press. Poole, K.T. and Rosenthal, H. (1997). Congress: A political economic history of roll call voting. New York: Oxford University Press. Rabinowitz, G. and MacDonald, S. (1989). A direction theory of issue voting. American Political Science Review 83: 93-121. Riker, W. (1982). Liberalism vs. populism: A confrontation between the theory of democracy and the theory of social choice. San Francisco: W.H. Freeman. Riker, W. (1986). The art of political manipulation. New Haven: Yale University Press. Riker, W. (1990). Heresthetic and rhetoric in the spatial model. In J. Enelow and M. Hinich (Eds.), Advances in the spatial theory of voting. New York: Cambridge University Press. Schofield, N. (1985). Anarchy, altruism, and cooperation. Social Choice and Welfare 2: 207219. Taylor, M. (1976). Anarchy and cooperation. London: Wiley. Taylor, M. (1982). Community, anarchy, and liberty. New York: Cambridge University Press. Thompson, D.C. (1988). Vive Ie Quebec libre. Toronto: Deneau Publishers. [210] Public Choice 97: 429---449, 1998. © 1998 Kluwer Academic Publishers. 429 The "revival of communism" or the effect of institutions?: The 1993 Polish parliamentary elections * MAREK M. KAMINSKI] , GRZEGORZ LISSOWSKI 2 & PIOTR SWISTAK3 ] Department of Politics, New York University, New York, NY 10003, U.S.A.; 21nstitute of Sociology, Warsaw University, PL-00-324 Warsaw 64, Poland; 3Department of Government and Politics, University of Maryland, College Park, MD 20742, U.S.A. Abstract. In several Eastern European countries the breakdown of communism in 1989 was followed by a surprising return to power of post-communist parties. Yet, some electoral victories of post-communists look puzzling when contrasted with a small size of a shift in voters' preferences that has led to them. Such is the case of the 1993 Polish parliamentary elections. Using partition-function form games and results of simulated elections, we estimate the impact of three factors that were blamed, in addition to the "shift to the left" in voters' preferences, for the 1993 victory of the post-communists in Poland. We show that the shift to the left was insufficient to assure post-communists an electoral victory, and that this victory would not have happened under the old electoral law or under a unified coalition of the Right. Our results show the high sensitivity of emerging democracies to the details of their institutional backbones. 1. Introduction A few years after the breakdown of communism in 1989, some of the CentralEuropean political scenes experienced another unexpected and radical transformation. In October 1992, a post-communist party regained power in Lithuania. In September 1993, a similar comeback took place in Poland, and in May 1994, in Hungary. Yet, the electoral victories of the post-communist parties were much more spectacular than one could have predicted from the size of their popular support. In Lithuania, 46.6% of popular votes resulted in 56.7% of lower house seats whereas in Hungary the corresponding numbers were 33% and 55%. In Poland, the disproportion between popular votes and seats * The authors are grateful to the companies Case and SMG-KRC Poland for sponsoring a survey used in this study, to the polling agency CBOS for granting access to their survey data, to Jarek Gryz for implementing the simulation program SEATS, and to Jerzy Bartkowski, Irena Jackiewicz, Lena Kolarska-Bobinska, Krzysztof Kruszewski, Marcin Kujawski, Marcin Penconek, Antoni Sulek, and Jim Vreeland for comments and help with data collection. Swistak thanks the Office of International Affairs and the University of Maryland for a financial support. [211 ] 430 was even greater, a mere 20.4% of popular votes for a post-communist party resulted in a total of 37.2% of Sejm (lower house) seats. The electoral success of the post-communist party in Poland facilitated a coalition with another big winner in the elections, the Left Peasants party (15.4% of votes and 28.7% of seats). Since the Left Peasants party originated under communism as a puppet-ally of the former Polish communist party, the coalition of these two parties was labeled post-communist. For brevity and convenience we will refer to this coalition as the post-communist Left or the Left. l This coalition turned out to be very stable and has ruled Poland since 1993. When the "revival of communism" was detected, the students of EasternEuropean politics focused mainly on the economic and social costs of transition. These costs were believed responsible for the widespread disillusionment with democracy and the increase in the support of the former communists. The "shift to the left" in the voter preferences was typically considered identical with the "shift to the left" in the distribution of political power. The interaction of the "shift to the left" in voters' preferences with the institutional details of the political game (electoral law, coalitions, etc.) was noted but was not closely examined (Gibson and Cielecka, 1995; Racz and Kukorelli, 1995; Tworzecki, 1994; Zubek, 1994). Here we take a more careful look at this interaction. In this paper we focus entirely on the 1993 Sejm (lower House) elections in Poland. We show that the electoral victory of the Left is only in part due to a shift to the left in voters' preferences. Apart from the shift, which indeed took place, three other factors contributed to the electoral results. We analyze their effects through a series of computer simulations. It turns out that the most powerful effect was that of a new electoral formula. The second most powerful factor that contributed to the victory of the Left was the failure of the fragmented post-Solidarity Right to form a coalition. Our simulation results show that had the elections been run under the same formula as the 1991 elections, or had a coalition of the Right been formed, the post-communist block would have remained in minority and, most likely, would have had no part in the formation of the government. Interestingly, both the shift in electoral preferences to the left and another factor that is often blamed for the 1993 electoral failure of the Right, the unexpected last-minute entrance of two new parties, show relatively small impact on the final distribution of power. Our analysis suggests that neither one of the two factors alone would have increased significantly the political power of the Left. In short, then, we separate major from minor effects and show the political mechanism behind the post-communist takeover. The results ultimately point to the power of political institutions. [212] 431 To model the logic of coalition formation, we use cooperative games in the partition-function form. We apply this concept to get explanations, obtain simulated results of elections, and formulate a series of game-theoretic questions. We show how the perception of the electoral game, the electoral law, and the entrance of new players resulted in a massive failure of the postSolidarity Right to convert its large share of the popular vote to parliamentary seats. Finally we simulate a series of potential electoral outcomes under various alternative electoral laws, with two last-minute entrants removed from the political scene, and with a united coalition of the Right. 2. The short story of the 1993 elections The story of the spectacular come back of the Left begins on 28 May 1993. On that day a new electoral law was adopted by the Sejm. The new law was more friendly to larger parties and almost all larger parties in the Sejm supported the change, whereas almost all smaller parties voted against it. 2 The voting on the new law took place in a fervent situation due to another major political problem. The non-confidence vote was on the agenda for the same day. The cabinet fell by one vote, and the next day, unexpectedly, President Lech Wales a dissolved the Sejm and called for new elections. The elections took place in September, and brought a defeat to all post-Solidarity parties. The relatively homogeneous bloc of the post-Solidarity Right did especially poorly and, though its popular vote support was substantial (26.2%), the Right received only a tiny portion, 3.5% of all Sejm seats (see Table 1).3 Before the election, almost 1 voter out of 3 declared that he would vote strategically in the face of a sure electoral defeat of his most preferred party (CBOS, 1993a: 13). Thus, the fragmented bloc of the post-Solidarity Right probably lost some of its votes to the post-Solidarity Center and Left Peasants party due to its supporters' strategic voting. It is quite likely that the potential popular support of the Right in 1993 was higher than the actual one and similar to that of 1991 elections, of about 30% of the electorate. Yet the total number of seats that these parties won in the 1991 (34.4%) and 1993 elections (3.5%) differed dramatically. 3. A model of rational coalition-building in a fragmented party system The new electoral law of 1993 was designed to de-fragment the political system and most of the rightist parties supported the change. The Right had two good reasons to believe that the new law would not be harmful. First, the post-Solidarity rightist parties believed that their popular support would be [213 ] 432 Table 1. Percent of the popular vote and the percent of seats obtained by the Right in the 1991 and the 1993 elections to the Sejm Parties of the post Percent of the popular vote Percent of seats in Sejm 1991 1993 1991 1993 Coalition Fatherland 8.7 6.4 10.9 0.0 Centrum 8.7 4.4 9.6 0.0 6.1 0.0 5.9 0.0 solidarity Right 2.7 Movement Right Peasants 5.5 2.4 Solidarity 5.1 4.9 Other parties 3.5 WalesaBioc Total 0.0 5.4 31.5 3.5 2.0 26.2 34.4 3.5 Source: Monitor Polski (1993). sufficient to meet the qualifying thresholds required by the new electoral law. And, second, if the polls ever indicated possible problems with meeting the thresholds, the Right would always have an option to form a coalition; given the bloc's relative ideological homogeneity, coalescing was seen as a viable option. Indeed, just a few days after the presidential decision to dissolve the Parliament, the coalitional negotiations were well-advanced. Ultimately, however, the rightist parties failed to coalesce. To understand the logic of the failure, we will use the somewhat forgotten concept of a partition-function form game (PF), a generalization of the standard characteristic-function form game. 4 We believe that the PF games provide an intuitive and a proper tool to analyze the formation of pre-electoral coalitions in multiparty systems (Kaminski, 1994). The concept of the PF game naturally emerges in the context of the following decision-making problem. Assume, first, that parties platforms are fixed. Hence, in a multiparty system, the only important decisions to be made by party leaders are whether to form coalitions with other parties. Consider, first, the popular vote that parties and their coalitions may receive. Note that some coalitions may be considerably harmful. For instance, a coalition of a communist and a religious right partyS would require a merger of two largely inconsistent ideologies - a manoeuver that would decrease the credibility of the new party and lower its popular support (cf. Hinich and Munger, 1994). [214 ] 433 Other coalitions, however, like between two parties with identical platforms, may be beneficial. In general, then, if two parties coalesce, then their total payoff, in terms of popular votes in elections, may be smaller, equal, or larger than the sum of the popular votes of the partners. However, the maximization of the popular vote is rarely the objective of a political party, the ultimate goal is to maximize the number of seats won. Though typically vote-maximization is equivalent to seat-maximization, in some instances the two objectives do not agree. Sometimes a party is willing to sacrifice a few seats for a larger popular vote share, but, usually, the seat-maximization objective supersedes the vote-maximization one. Under all circumstances, votes and seats are the two main aspects of coalescing. The decision to enter a coalition depends on the expected number of votes and the expected number of seats. The latter is a function of both the distribution of votes and the electoral law. As we will later see, the impact of the electoral law on the outcome of elections can be profound. Formally, we define an n-person cooperative game in partition-function form as follows. Any family of disjoint and nonempty subsets of players C = {C 1, •.. , C r } is a coalitional structure. 6 The partition function assigns to every coalition C j , included in a coalitional structure C, a real number, or a vector of numbers, F(C i, C). We assume that I;iF(Ci, C) is constant for any coalitional structure C. The number F(C, C) is interpreted as an exact payoff of C i under the circumstances C. (The obvious difference between a PF game and cooperative games in characteristic function form is that the payoff function for a given coalition may vary over various actual coalitional structures whereas in a standard characteristic function game the impact of actual coalitional structures is not represented.) There are two interconnected PF games that describe the parameters of a political situation. The popular votes of various coalitions under all possible coalitional structures in all electoral districts form a partition function of the first game, or the vote-game, G v. The payoffs of this game are vectors of popular votes in the districts. An approximation of these payoffs are popular votes measured at the national level. The electoral law, which converts popular votes in districts into parliamentary seats, defines the partition function of the second game, or the seat-game, GS . Thus, an electoral law EL is a function that assigns to any vote-game G V the corresponding seat-game GS = EL(Gv ). The PF games provide a convenient framework for explaining parties decisions to form coalitions, simulating possible outcomes, and posing normative questions about multi-party systems. We begin with the explanations. [215 ] 434 4. The failure of the Right?: An explanation Media reports would indicate that, in their thinking about benefits and costs of coalescing, the post-Solidarity Right leadership employed simplified versions of PF games. Our analysis proceeds under this hypothetical, though highly plausible, assumption. Players (parties) are assumed to be seat maximizers. They observe results of the polls and use these results to estimate the parameters of the vote-game. The estimated vote-game together with an electoral law provide proxies for payoffs in the seat-game. Two (or more) parties coalesce if they expect the number of seats for the coalition to be greater than the sum of seats they can secure individually. After the fall of communism, polls quickly became an important political tool. Politicians learned that polls are not only useful as a source of information but also as a strategic tool. For instance, the leaders of the fragmented rightist bloc proclaimed polls as biased. The leader of Centrum, Jaroslaw Kaczynski, attributed the bias to "manipulation": "So-called public opinion polls are manipulated. In the previous elections, our survey scores were also very poor, but our election results were quite good".8 Kaczynski was clearly right to claim that the polls seriously underestimated the strength of the rightist parties (see Table 2). He might have been acting strategically, however, when he claimed that the polls had been manipulated. Indeed, the causes of the bias seem to be less Machiavellian. The reliability of polls in a highly fragmented party system is much lower than in a two- or threeparty stable democracy. First, in the same sample, the standard error of the estimated support for a party holding 50% of the popular vote is relatively much smaller (though larger in objective terms) than the standard error for a party holding 5% of the vote. Second, the pre-campaign bias of the media towards larger parties produces a systematic underestimation of the poll predictions for the small- and medium-size parties. When, during the campaign, all parties receive comparable media coverage, small- and medium-size parties improve their poll ratings significantly.9 Finally, voters may misrepresent their preferences in pre-election polls. This effect was well-established in 1990 presidential and 1991 parliamentary elections in Poland when both the Right and the Left's ratings were underestimated while the ratings of the post-Solidarity centrist parties were overestimated. When two parties consider the decision whether to coalesce, they are interested in a fragment of the partition function only. First, they would like to know how their share of votes and seats will be affected by the decision to coalesce given the current coalitional structure. Second, they would like to estimate sensitivity of these payoffs to possible changes in the coalitional structure. Even if the rightist leaders did not consider poll ratings as unbiased estimates of their support, the polls generated the only estimate of the frag- [216] 435 Table 2. Estimated support for the parties of the post-Solidarity Right involved in negotiating coalition formation Parties/coalitions of the Estimated post-Solidarity Right support Fatherland 4.5 2.2 6.37 4.42 2.1 2.70 3,607 2.1 2.37 3.607 Centrum Movement Right Peasants Actual votes Sample size 14,874 14,874 Note. "undecided" excluded. "Estimated support" includes unweighted mean support from 11 surveys by CBOS, Demoskop, and PBS Sopot (Fatherland and Centrum) and 3 surveys by CBOS (Movement and Right Peasants). The surveys were conducted between July and September 1993. Source: Domanski (1995); CBOS (I993a). ment of partition function they needed. Moreover, a poll with an unknown systematic bias, that is proportional to the size of the popular support, still provides a reliable estimate of the ratio of the support for a coalition to the total support for its members. The smaller this ratio, the lower the incentives of the parties to maintain the coalitional arrangement. It is likely that the estimates generated by the polls had a decisive impact on the failure of coalitional negotiations. After the Parliament was dissolved on 29 May 1993, the initial plans for coalition formation were announced just two days later. On 8 June, three coalitions of the rightist parties were formed and further fusions were expected. Two weeks later the first signs of the problem were observed. The polls suggested deep subadditivity for two rightist coalitions and approximate additivity for the third coalition in the vote-game (see Table 3). In fact, the two sub additive coalitions received total support well below the senior partner's support recorded just a few days earlier. The CBOS analysts wrote about a "phenomenon of losing a large part of the electorate due to coalescing" (CBOS, 1993a: 5). The required qualifying threshold of 5% for parties and 8% for coalitions further undermined the incentives for maintaining the coalitions. The additive coalition survived but the two subadditive ones dissolved a few days after the announcement of the poll results. The coalition-formation process did not recover from this failure in time to meet the 10 August deadline for declaring coalitions. It is likely that the observed effect of the "loss of the electorate" was due to the frequent changes of party labels rather than a real loss of the electorate. Each new coalition substituted a new coalition name for two, or more, rela- [217 ] 436 Table 3. CBOS poll standings of the coalescing parties of the Right May 11-14 June Right Peasants 3.7 3.7 Fatherland 3.7 3.7 Polish Convention 1.2 1.2 (three minor 1.2 0.4 Right parties) 2.4 0.2 Movement 3.5 5 Centrum 1.2 1.2 18-21 June 9-12 July 12-17 August 2.4 1.2 1.25 5 2.5 1.2 3.7 4 2.4 2.4 2.4 0.5 Source: CBOS (I993a, 1993b). Sample size N E [1086; 1376]. Note. "undecided" excluded. Cells of the table that are merged correspond to a coalition of the respective parties at the time the poll was conducted; the question concerned this coalition. tively well-known brand names of its members. In addition to the information noise caused by this relabeling, two new political actors, Walesa Bloc and Solidarity trade union, entered the scene and took away some support of the Right. The polls of 18-21 June asked for the support for the new-born Walesa Bloc which had not been included in the preceding poll of 11-14 June (see Table 3). The Bloc received the support of 12.5% of respondents. These votes would have most likely gone to other rightist parties had the Walesa Bloc not entered the scene. In addition to the Walesa Bloc, on 28 June, the Solidarity trade union entered the elections. Both entrances added uncertainty to the estimates and, by increasing the number of negotiating units, increased the negotiation costs. 5. The shift to the left When coalescing efforts failed in June and July of 1993, the post-Solidarity Right might have estimated their potential support at the level of the 1991 elections (see Table 1). The Right was fragmented no more than in the 1991 elections and the polls were close to what they had been in 1991. It seems that the Right parties' estimated probabilities of crossing the respective thresholds required by the new law were high. When the political situation appeared roughly predictable, the entrance of two new parties, Walesa Bloc and the Solidarity trade union, changed the picture. The new configuration of the parties induced substantial shifts in the popular vote. In addition, a significant increase in the support for the Left, was detected in August of 1993. In the meantime, coalition building efforts among the rightist parties were failing. And 10 August was the deadline for registering coalitions and candidates on local voting lists (see Figure 1). [218] 437 .~l ...... . 40-~ .......... ~ 30-- I~l. 25- ............. ~ ........... ..... ~. ................. ~ . 10~ 5-~ 0+ June 11 June 18 ___ ....+... July 9 Post-communists Total August 12 - • - August 27 September 17 Left Peasants Figure 1. The estimated support for the post-communists and the Left Peasants. Source: CBOS (l993a); Domanski (1995). Note. "undecided" excluded. And so, the month of August brought a clearer picture of the strategic properties of the political situation. It became apparent that total defeat of the Right was possible. By 10 August no major coalitions of the Right were formed and the Right entered into the 1993 elections fragmented. At that point the game among the parties of the Right changed. Before 10 August, the game had been cooperative. After the deadline, coalition building was no longer possible (for the purpose of entering the elections, that is) and the game turned into a non-cooperative one. The only two actions available to the parties in the non-cooperative game were "to withdraw from the elections" (and encourage its supporters to vote for the other rightist parties) or "to stay". Since the withdrawing party would not have been able to get any compensation in seats for the withdrawal, and every party attached some nonzero probability to crossing the threshold, staying was a dominant strategy. Also, the withdrawal of any of the parties would have increased the expected payoffs to those remaining in the elections. Not surprisingly, almost all of the rightist parties appealed to the others to withdraw, but were unsuccessful (Sulek, 1995: 114). The equilibrium in dominant strategies was "to stay" for all parties, and all of them chose their dominant strategies. Clearly, the payoff structure of the game, under the parties' beliefs, was that of the Prisoner's Dilemma. [219 ] 438 Other aspects of the strategic environment were such that no apparent solution to the problem of the inefficient equilibrium was readily available. The race was close and the perceived reliability of the poll estimators was low. The lack of an enforcement mechanism did not facilitate solution by a contract. Whatever the potential contract could have been, no player would have had an incentive to keep it - the Right was bound to dwell in the inefficient equilibrium. And it did. It was too late to correct for the effects of the new electoral law, the failure to coalesce, and the fragmenting effect of the entry of Solidarity and the Walesa Bloc. But, what were these effects? How bad was the failure to coalesce? How strong was the effect of the electoral law? Had the entry of the two new parties contributed significantly - as a typical political analysis would want us to believe - to the downfall of the right? What would have been the results of the elections without the two parties entering the scene? The exact answers are not feasible but we can offer some approximate answers. Election results can be simulated under varying parameters of the game. We report and discuss results of some computer simulations in the next section. The results turn out to be robust with respect to the parameters of the PF game. 6. Possible results of the elections In this section we will comment on the results of three sets of simulations. The simulations were done in an attempt to assess the impact of the electoral law and coalitional structure on the distribution of seats in the Sejm and the consequent division of power on the Polish political scene. All simulations use actual district-level election results, data from the CBOS surveys, and data from a special survey that was designed by the authors to reconstruct second preferences of voters (i.e., candidates/parties they would have voted for had their first choice been not available). Simulations are based on the relevant fragment of the partition function of the 1993 elections' vote-game. In all simulations, the underlying vote-game (that is, the popular votes under various coalitional structures) is assumed to be the same and is reconstructed from the electoral and survey data. This assumption simply means that the changes in the electoral law do not induce voters to vote strategically. We comment on the robustness of this assumption and discuss possible adjustments. In all simulations we keep constant all factors of the 1993 elections except for the one factor whose impact we want to assess. We begin by looking at the effect of the new electoral law and then move to assess the impact of the rightist coalitional structure. [220] 439 91+R+A VOTES Post-communists • 91+R 91 Post-Solidarity Right Figure 2. The impact of the electoral law. "Votes" stands for the percent of the popular vote in the 1993 elections. All the other numbers represent the percentages of seats in 1993 elections (93 = 91 +R+A+ T) and in simulated elections with partial changes to the 1991 electoral law. R stands for redistricting. A for a new apportionment method, and T for qualifying thresholds (three changes introduced to the 1991 electoral law). "91 + R", for instance, stands for a simulation in which all electoral rules are that of the 1991 with the exception of R, the new districting. 6.1. The impact of adopting the new electoral law Recall, first, that the 1993 electoral law changed three aspects of the 1991 rules: redistricting (R), a new apportionment method (A), and the introduction of qualifying thresholds for parties and coalitions (T) (see note 2 for a more detailed description). To denote this, we write, 93 = 91 + R + A + T. We also use this convention to label electoral rules in Figure 2. "91 + R", for instance, stands for a simulation in which all electoral rules are that of the 1991 with the exception of R, the new districting. Our objective is to assess the effect of each of the three factors (R, A and T) and their combinations (R + A, R + T, etc.) to see which change in the electoral law had the largest marginal impact on the results of the '93 elections. Figure 2 shows the simulated election results, in percent of seats in Sejm, for selected combinations of factors. (Table A I in the Appendix gives a detailed list of results.) The first observation is that if the 1993 elections had been run under the 1991 electoral law, the post-communist Left would have ended up with 37% of seats and the post-Solidarity Right with 26.7%. Redistricting alone (91 + R) would not have had much of an effect on this distribution. The most likely political consequence would have been a coalition cabinet of the Center and the Right with no post-communist participation at all. Since redistricting would not have had much of an impact on the election results, regardless of [221 ] 440 what other changes were or were not present, we only include results with the new districting (R). Our second observation is that redistricting plus either of the other two factors (A or T) would have been sufficient to secure a post-communist Left majority of seats (55.1% and 57% respectively). The Right would have also increased the number of seats significantly 04.8% and 6.5% respectively) though not sufficiently to block the takeover by the post-communist Left. The most dramatic shift in the distribution of seats is due, however, to the joint effect of all three factors (R + A + T). This resulted in the actual 65.9% vs. 3.5% split of seats between the Left and the right in the 1993 elections. 6.2. The impact of removing Solidarity and Walesa Bloc from the political scene There seems to be an agreement among political analysts that by entering the elections, Solidarity and Walesa Bloc weakened the Right by increasing its already serious fragmentation and led to the Right's infamous electoral failure. However, how strong was the effect of the emergence of the two new players? What would have happened had Solidarity and Wales a Bloc not entered the elections? As before, we can approximate the answer by looking at the simulated results. 10 The effect, as it turns out, is both more complex and more interesting that the "common sense" political analysis asserts. Figure 3 shows the percent distribution of seats with the two actors removed under four auxiliary conditions. (Table A-2 in the Appendix gives a more detailed list of results.) Perhaps the most obvious observation that can be made from Figure 3 is the surprisingly small effect of Solidarity and Walesa Bloc on the percent of seats gained by the Left. It varies from the actual 65.9% of seats to 57.6% in the worst case scenario; the best the Right could have done was 12%. Yet, the most surprising observation is that the effect of the two new players could have been opposite to the expected - the percent of seats gained by the Right could have been smaller without Solidary and Walesa bloc present in the elections! This is due to the fact that Walesa Bloc was the only Right party that actually won any seats. The first simulation (SYMl) shows only 1.7% of seats gained by the Right, as opposed to the actual 3.5%. (The Left would have lost some seats, but not too many, a decline from 65.9% to 62%.) There is a possible problem, however, with the estimates obtained in the first simulation. The estimates have large standard errors (due to a small sample size in the survey on second preferences) and the estimated popular support of two parties of the Right, Centrum and Fatherland, came close to the required thresholds. 11 Thus in the next three simulations we have assumed that Centrum has (barely) crossed the threshold (SYM2), that Fatherland has [222] 441 62 SEATS SYM 1 Post-Communists SYM2 • SYM3 SYM4 Post-Soiidarity Right Figure 3. The impact of the two new entrants. "Seats" stands for the percent of seats received in the 1993 elections. All simulations assume that Solidarity and Walesa Bloc did not participate in the elections. In addition, SYM 2 assumes that Centrum has (barely) crossed the threshold, SYM 3, that Fatherland has (barely) crossed the threshold, and SYM 4 that both Centrum and Fatherland have (barely) crossed the thresholds. (barely) crossed the threshold (SYM3) and that both Centrum and Fatherland have (barely) crossed the thresholds (SYM4). Very much like the first simulation (SYM I) these three scenarios would have affected the power of the Right in a very limited way. In the best case scenario, the Right would have obtained 12% of the seats while the Left would have ended with 57.6%. In each case the Left maintains the majority in the Sejm while the power of the Right does not significantly change. Going against the common beliefs, our results show that the entry of Solidarity and Walesa Bloc would not have had much of an effect on the final distribution of power. 6.3. The importance of the coalitional failure of the Right The last effect that remains to be assessed is that of the failure of the Right to enter the 1993 elections as a united coalition. This failure to coalesce has been widely blamed for the electoral failure of the Right. As it turns out, results of our simulations support the folk wisdom. As opposed to the latter, however, the simulation shows the estimated magnitude of the effect. The results are displayed in Figure 4. If six rightist parties were to form a single coalition what part of the supporters of the six separate parties would have voted for the united coalition? While the answer is not clear, a standard simplifying assumption, that of the additivity of votes casted for all parties of the coalition, seems to be the most likely scenario. While there are reasons to believe that a united right [223] 442 65.9 J47.6L 1J32L 35.4 3.5 SEATS 90% Posl-Communists 100% 110% Post-50lidarily Right Figure 4. The impact of the Right's failure to coalesce. "Seats" stands for the percent of seats received in the 1993 elections. All the other numbers represent the percentages of seats in elections simulated under the assumption that the Right coalesced and the vote for the coalition was additive (100%), 10% more than additive (l1O%), and 10% less than additive (90%). would have both lost some voters and gained some, there are no reasons to believe that one of the two effects would have turned out to be significantly stronger than the other. Hence our first simulation in Figure 4 labeled 100% was run under this assumption. To assess the robustness of the result of this simulation with respect to the assumption of additivity, we have run two other simulations. The simulation labeled as 90% assumes 10% loss in votes as compared to the vote under the additivity assumption and simulation labeled 110% assumes a 10% gain. In both cases we assume that the loss or the gain is absorbed proportionally by all remaining parties. 12 Results turned out to be similar in all three simulations. The post-Solidarity Right secures somewhere between 32 and 38.3% of the seats while the postcommunist Left gets between 44.1 and 47.6%. Consequently, the Left no longer has the majority and the most likely scenario is a coalition cabinet of the Center and the Right with no participation of the post-communists parties. The second most likely scenario is a post-Solidarity coalition that includes the Left Peasant party which was included as a junior partner in all former post-Solidarity cabinets. 7. Conclusions In this article, we have shown that the shift to the left in voters' preferences was only one of a few factors in bringing the Left to power in the 1993 Polish parliamentary elections. We have identified and analyzed three other factors [224 ] 443 that had a significant impact on the course of events. Our simulations show that under the old electoral law or with a grand coalition of the right, the electoral outcome would have been dramatically different. It is quite remarkable that under the old electoral law, or under the united coalition of the Right, the post-communist Left would have been most likely excluded from the cabinet. Only the conjunction of the change in the electoral law, the coalitional failure of the Right, and the increased electoral support for the post-communists could have produced a post-communist controlled cabinet. The key problem for the Right was the inability to cooperate. In part this was a consequence of the absence of reliable procedures for the division of future gains. Creating a coalition in a fragmented party system poses an interesting mixture of game-theoretic and estimation problems for rational players. These are formally addressed in a separate paper (Kaminski and Gryz, 1997). Here, we will briefly describe the main aspects of the problem. The first issue to be resolved by members of a prospective coalition is, what information should be taken into account to represent the power of the members? The source of such information could be a carefully designed poll. The second issue is to settle on a fair solution problem of dividing seats among coalitional partners. In short, the problem is to identify the most preferred solution concept for the partition-function form games that is based only on the available polls information. The solution would have to minimize the incentives of the members to leave the coalition and would have to be adopted by all members on the basis of some uniformly accepted principles. Finally, the last problem is that of implementation. How should the solution be implemented given the constraints of limited transferability of seats and uncertain results of the elections? These are not only workable theoretical problems, but important practical points that enter coalitional negotiations. The science of politics can aid these decision processes significantly. The problem of how to divide seats among coalition parties has continuous and pressing significance for the Polish politics. Before the fall 1997 elections, the post-Solidarity Right and one more pre-Solidarity rightist party managed to create two large coalitions, and they are negotiating further merger. The problem of the fair way to divide the pie continues to be the central issue. Emerging political systems are highly sensitive to the details of their institutional backbones. Electoral laws, coalitional structures, existing survey methodologies, or the timing of elections may strongly influence the results. Under the 1989 electoral law in Poland, in the first semi-free elections in the Soviet Bloc, Solidarity won 100% of the freely available seats 13 in the Sejm with only about 70% of the popular vote. This unexpected victory contributed decisively to the subsequent fall of the communist regimes in Eastern Central Europe (Kaminski, forthcoming). Ironically, four years later the winners and [225 ] 444 the losers, the wise men and the suckers, have switched places although their total popular vote support has not changed too much. Notes 1. Our naming convention is merely meant as a convenient shortcut in identifying the main players. There are other than the post-communist leftist parties on the left of the Polish political scene. Hence, the label "Left" refers to the post-communist Left. Similarly, the label "Right" denotes post-Solidarity Right parties not all parties on the right of the political spectrum. 2. The following main changes were introduced in the new electoral law. A system of country-level thresholds was established. For the 391 seats assigned in districts, the thresholds were 5% for parties and 8% for coalitions. For the 69 seats assigned from the national list, the threshold was 7%. Minority parties or coalitions were granted the privilege of having one of these thresholds, selected by them in advance, waived. The apportionment method (the rule of turning the percentage of votes in a district to the number of seats in a district) has been changed. The largest remainder, or Hamilton-Hare method with Droop quota, was substituted by a divisor d'Hondt-Jefferson method. (For properties of the two methods see Balinski and Young, 1982.) Finally, the 37 districts of the 1991 elections were substituted by 52 districts. The average district magnitude (number of seats per district), decreased from 10.6 to 7.3 (Dziennik Ustaw RP nr 45, 1993). 3. Full names of the parties were simplified to one- or two-word labels. Name changes, minor coalitions, or minor mergers were neglected in the description in an effort to keep the presentation simple. For the list of all major parties and coalitions, see the Appendix. For the results of elections, see Table 4, columns "Votes" and "Seats". 4. PF games were introduced by Thrall (1962) under the name of "generalized characteristic function form games". The name was later changed to PF-games by Thrall and Lucas (1963). 5. Thrall (1962: 158) notes this possibility in his path-breaking paper: "In the political arena [subadditive payoffs of two parties] might represent clashing ideologies ... in which an open union would rather weaken support for both [... ]". 6. We assume a slightly more general concept of the PF here and allow that the coalition structure be not exhaustive. For instance, the coalition structure with no party i denotes the case of i withdrawing from the elections. Thus, players are interpreted as "potential'· rather than "actual" players. 7. Recall that we use the label Right as a shortcut for the so called post -Solidarity Right. 8. Quoted in (Sulek, 1995). Sulek links the hostile opinions of politicians about surveys with the low standings of their parties and assesses the overall performance of polling agencies as good. In fact, CBOS reports bring many methodological caveats regarding the interpretation of the results of their polls and discuss in detail possible biases. 9. The electoral law regulated political advertising. All main parties were allotted equal time slots in the national and regional state-controlled TV and radio; paid advertising was severely restricted. 10. The relevant fragment of the PF function was reconstructed with the use of two polls reconstructing second preferences of the voters. The "omnibus" survey by SMG KRC Poland was conducted in February 1994, and included a stratified non-random sample of adults (N = 1000). The pre-electoral poll by CBOS was conducted on 27-29 August [226] 445 1993, and included a representative sample of adults (N = 1104). Votes of the Walesa Bloc and Solidarity supporters were transferred to other parties according to the distribution of second preferences for this group, adjusted for the biases detected in both samples. 11. The estimate of the popular vote for Centrum was 4.42% and the threshold for this political category (a party) in the elections was 5%. The estimate of the popular vote for Fatherland was 7.7% and the threshold for this political category (a coalition) in the elections was 8%. 12. Before the Fall 1997 elections, the Right Solidarity parties and one Rightist pre-Solidarity party, the Confederates, have formed two large coalitions. The current polls (i.e., as of May 1997) consistently show that the total support for both coalitions is at the level of 35%. This is close to the total of 32% of votes won by the Right and the Confederats in the 1993 elections. Assuming that no significant changes have taken place in voters' preferences, these results support the assumption of an approximate additivity of the votes across members of a coalition. They also support a weaker hypothesis that coalescing does not lead to a significant loss of votes (subadditivity). The weaker hypothesis was necessary for the plausibility of our simulations. 13. The 1989 election was not yet fully democratic. 65% of the Sejm seats were guaranteed to the candidates of the communist block. References Balinski, M. L. and Young, H. P. (1982). Fair representation. New Haven: Yale University Press. CBOS. (1993a). Preferencje polityczne spoleczenstwa trzy tygodnie przed wyborami (BS/132/109/93). Warszawa: CBOS. CBOS. (1993b). Preferencje wyborcze spoleczenstwa w lipcu '93 (BS/102/83/93). Warszawa: CBOS. Domanski, R. (1994). House effect w Polsce. Kultura i Spoleczenstwo 38: 69-84. Dziennik U staw RP No 45. (1993). Ordynacja wyborcza do Sejmu Rzeczypospolitej Polskiej. Warszawa: Urzad Rady Ministrow. Gebethner, S. (1995). System wyborczy: Deformacja czy reprezentacja? In S. Gebethner (Ed.), Wybory parlamentame 1991 i 1993 a polska scena polityczna, 9-48. Warszawa: Wydawnictwo Sejmowe. Gibson, J. and Cielecka, A. (1995). Economic influences on the political support for market reform in post-communist transitions: Some evidence from the 1993 Polish parliamentary elections. Europe-Asia Studies, 47: 765-785. Hinich, M. J. and Munger, M. C. (1994). Ideology and the theory of political choice. Ann Arbor: The University of Michigan Press. Kaminski, M. M. (1994). Coalitional structures, Duverger's law, and 'split-merger stability' hypothesis. Mimeo. IRIS Working Paper No. 108. Kaminski, M. M. (1997). How communism could have been saved: Formal analysis of electoral bargaining in Poland in 1989. Public Choice, forthcoming. Kaminski, M. M. and Gryz, J. (1997). Koalicje wyborcze: Lekcja 1993 roku. In J. Wasilewski (Ed.), Zbiorowi aktorzy polskiej polityki. Warszawa: ISP PAN (forthcoming). Monitor Polski No. 50. (1993). Obwieszczenia Panstwowej Komisji Wyborczej, 610-799. Warszawa: Urzad Rady Ministrow. [227 ] 446 Racz, B. and Kukorelli, I. (1995). The 'second-generation' post-communist elections in Hungary in 1994. Europe-Asia Studies 47: 251-279. Sulek, A. (1995). Do czego sluza sondaze? Wykorzystanie badan opinii publicznej w kampanii wyborczej 1993. Kultura i Spoleczenstwo 38: 103-117. Thrall, R. M. (1962). Generalized characteristic functions for n-person games. In Recent advances in game theory, 157-160. Princeton: Princeton University Conferences. Thrall, R. M. and Lucas, W. F. (1963). n-person games in partition function form. Naval Research Logistics Quarterly 10: 281-298. Tworzecki, H. (1994). The Polish parliamentary elections of 1993. Electoral Studies 13: 180185. Zubek, v. (1994). The reassertion of the left in post-communist Poland. Europe-Asia Studies 46: 801-837. Appendix Polish full names and common acronyms of major parties/coalitions in 1993 elections: Centrum Porozumienie Centrum (PC) Fatherland Right Peasants KKW "Ojczyzna" Polskie Stronnictwo Ludowe-Porozumienie Confederats Ludowe (PSL-PL) Konfederacja Polski Niepodleglej (KPN) Post-communists Left Peasants Liberal Democrats Solidarity Sojusz Lewicy Demokratycznej (SLD) Polskie Stronnictwo Ludowe (PSL) Kongres Liberalno-Demokratyczny (KLD) Solidarnosc Unia Demokratyczna (UD) Democrats Walesa Bloc Bezpartyjny Blok Wspierania Reform (BBWR) Labour Unia Pracy (UP) Libertarians Unia Polityki Realnej (UPR) Movement Koalicja dla Rzeczypospolitej (KdR) Party "X" Partia"X" Polish Convention Konwencja Polska Self-defense German Minority Samoobrona Towarzystwo Spoleczno-Kulturalne Mniejszosci Niemieckiej na Slasku Opolskim [228] 447 Table A1. Actual and simulated results of 1993 elections under various electoral laws Votes Party/coalition Seats 93 91 +R+T 91 +R+A 91 +R 91 a 26.2 3.5 6.5 14.8 24.5 26.8 Fatherland 6.4 0.0 0.0 5.2 8.7 8.3 Centrum Movement Right Peasants 4.4 2.7 0.0 0.0 0.0 0.0 2.0 0.9 3.0 1.5 4.6 0.9 WalesaBloc 2.4 5.4 0.0 3.5 0.0 6.5 0.4 3.5 1.7 4.6 5.9 Solidarity 4.9 0.0 0.0 2.8 5.0 5.4 21.9 25.0 28.7 21.6 25.2 23.3 4.0 10.6 0.0 16.1 0.0 17.0 2.2 12.2 3.0 12.6 3.3 11.5 7.3 8.9 11.7 7.2 9.6 8.5 35.8 65.9 57.1 55.1 37.6 37.0 Post-communists 20.4 37.2 21.3 20.9 15.4 28.7 32.8 24.3 30.1 Left Peasants 25.0 16.3 16.1 13.1 7.4 Post-Solidarity Right (total) Post-Solidarity Left 1.7 and Center (total) Liberal Democrats Democrats Labor Post-communist Left (total) Other (total) 15.2 5.7 7.6 7.8 12.7 Confederats 5.8 4.8 6.7 4.8 7.2 Libertarians German Minority 3.2 0.7 0.0 1.1 2.2 0.9 0.0 0.9 0.6 0.9 Party "X" Self-defense 2.7 2.8 0.0 0.0 0.0 0.0 0.4 0.9 1.3 1.1 2.2 0.9 0.9 1.7 aSource: Gebethner (1995). [229 ] 448 Table A2. Actual and simulated results of 1993 elections under the assumption of the trade union Solidarity and Walesa Bloc not entering the contest Party/coalition Seats SYMI SYM2 SYM3 SYM4 Post-Solidarity 0.0 1.7 4.1 10.0 12.0 0.0 1.7 1.7 10.0 10.0 Right (total) Fatherland Centrum 0.0 0.0 2.4 0.0 2.0 Movement 0.0 0.0 0.0 0.0 0.0 Right Peasants 0.0 0.0 0.0 0.0 0.0 25.0 30.4 29.8 27.1 26.1 0.0 16.1 3.9 16.1 2.8 15.0 2.4 14.8 8.9 10.4 3.7 15.9 10.2 9.3 8.9 65.9 62.0 60.9 58.0 57.6 Post-communists 37.2 33.0 32.4 28.7 29.0 28.5 31.5 26.5 31.3 Left Peasants 5.7 4.8 0.0 0.9 0.0 0.0 5.5 4.6 0.0 0.9 0.0 0.0 5.2 4.3 0.0 0.9 0.0 0.0 4.8 3.9 0.0 0.9 0.0 0.0 4.4 3.7 0.0 0.7 0.0 0.0 Post-Solidarity Left and Center (total) Liberal Democrats Democrats Labor Post-communist Left (total) Other (total) Confederats Libertarians German Minority Party "X" Self-defense [230} 26.3 449 Table A3. Actual and simulated results of 1993 elections under the assumption of a united PSR coalition including Centrum, Fatherland, Right Peasants, Movement, the trade union Solidarity, and Walesa Bloc Party/coalition Seats 90% 100% 110% Post-Solidarity 3.5 32.0 35.4 38.3 25.0 17.0 15.5 14.6 0.0 Right coalition Post-Solidarity Left and Center (total) Liberal Democrats Democrats Labor Post-communist Left 0.0 0.0 0.0 16.1 11.3 10.7 9.8 8.9 5.7 4.8 4.8 65.9 47.6 45.8 44.1 37.2 28.7 27.4 20.2 26.5 19.3 25.6 18.5 5.7 4.8 3.5 2.8 0.0 3.0 2.6 0.0 (total) Post-communists Left Peasants 0.0 0.9 0.7 3.2 2.8 0.0 0.4 Party "X" 0.0 0.0 0.0 0.0 Self-defense 0.0 0.0 0.0 0.0 Other (total) Confederats Libertarians German Minority 0.4 [231] Public Choice 97: 451-474, 1998. © 1998 Kluwer Academic Publishers. 451 A spatial analysis of candidate competition in dual member districts: The 1989 Chilean senatorial elections * JAYK.DOW Department of Political Science, University of Missouri-Columbia, 113 Professional Building, Columbia, MO 65211, U.S.A.; e-mail: polscdow@showme.missouri.edu Abstract. This study uses empirical spatial theory to evaluate candidate and voter behavior in senate elections contested during the 1989 Chilean general election. The study evaluates whether senatorial candidates competing in dual member districts under Chilean d'Hondt locate near the periphery or interior of the electoral space. Spatial analyses demonstrate the Chilean senatorial electoral system is characterized by centrifugal forces. In particular, candidates of the right locate on the periphery of the space and face few incentives to pursue moderate electoral strategies. The study also characterizes bases of party and candidate support and the underlying dimensions of political competition. Spatial analysis reveals both change and continuity in the pre- and post-authoritarian electoral universes. 1. Introduction Spatial models of voter choice are instrumental in the modem study of elections (Black, 1958; Downs, 1957; Enelow and Hinich, 1984; Enelow and Hinich, 1990; Hinich and Munger, 1994, 1996; Rabinowitz and Macdonald, 1989). Despite this, empirical spatial theory has largely been confined to the study of United States presidential elections. Laver and Schofield (1990), Schofield et al. (1988), Schofield (1988), Adams and Merrill III (1988a), use spatial theory to study party government in European parliamentary democracies. In addition, Macdonald, Rabinowitz, and Listhaug (Macdonald et al., 1991; Macdonald et al., 1995; Rabinowitz et al., 1991), Merrill (1994, 1995), Iversen (1994) and Adams and Merrill III (1988b) have recently completed spatial analyses of Western European party systems, but these studies primarily seek to evaluate alternative model specifications rather than provide substantive interpretations of particular elections. * I thank Jorge Gonzalez, Larry Graham, Mel Hinich, Jeanne Kinney, Bob Luskin and Mike Munger for helpful comments on this paper. I am responsible for all errors. This research was partially supported by National Science Foundation grant SES-9022930 [233 ] 452 This study provides a spatial analysis of four senate elections contested during the December 1989 Chilean general election. The Chilean electoral system possesses several characteristics which make it extremely valuable for spatial analysis. Most important, Chile has an institutionalized multiparty system in which senators are elected from dual member districts under a form of d'Hondt proportional representation. Under Chilean d'Hondt at least some candidates have incentives to locate away from the ideal point of the median voter. This study seeks to evaluate expectations of candidate behavior by estimating spatial maps of the senatorial elections using a metric scaling methodology that locates voters and candidates in the same space. The Chilean elections also derive significance by presenting a rare opportunity to study party competition in a transitional election. The 1988 plebiscite on the continuation ofPinochet's rule notwithstanding, the 1989 general election marked the restoration of civil affairs in one of the most long-lived twentieth century dictatorships in the Western Hemisphere. The maps provide information useful for interpreting party competition in the transitional election and the relationship between the pre- and post-authoritarian periods. 2. Chilean senatorial elections The December 1989 Chilean election was initiated by Augusto Pinochet's loss in the 1988 plebiscite on the continuation of his rule. Under terms of the Chilean constitution, the military junta which governed Chile since the September 1973 overthrow of Socialist President Salvador Allende called for elections the following year. Chilean political parties, free to organize and campaign, contested the election in two broad coalitions: The centerleft Concertacion de los Partidos por la Democracia (CPD) coalition of seventeen parties which led the fight against Pinochet in the 1988 plebiscite and the rightist Democracia y Progreso (DP) coalition consisting primarily of the Independent Democratic Union (UDI) of nominally independent presidential candidate Heman Buchi and the National Renewal (RN) party. The latter parties were closely associated with the Pinochet regime. Chilean political parties, despite military rule from 1973 to 1989, remain among the strongest and most developed in the world (Valenzuela, 1985; Valenzuela and Valenzuela, 1985; Caviedes, 1991; Scully 1992; Walker, 1993). The election was largely a victory for the Concertacion and its leading Christian Democratic Party (PDq. Christian Democrat Patricio Aylwin won the presidency over Buchi and independent candidate Francisco Errazuriz, and, through the election of Christian Democrats and members of the Concertacion, secured working majorities in both chambers of the Chilean legislature. 2 There exists a general consensus that the election was fair and [234 ] 453 that turnout and return figures are accurate (Angell and Pollack, 1990; Scully, 1995). The Chilean Senate consists of forty-seven members, thirty-eight of whom are elected from dual member districts. The remaining nine senators are appointed. 3 I study elections contested in the two Santiago senatorial districts, a district encompassing the costal cities of Valparaiso and Vina del Mar, and the district allocated to the south-central Sixth Administrative Region. 4 Approximately one-third of the Chilean electorate lives within the two Santiago districts. Similarly, a disproportionate number of voters live in the Valparaiso - Vina del Mar area. The agricultural Sixth Region is south of Santiago and its principle city is Rancagua. Parties contest legislative elections - the electoral system is the same for the Senate and the 120 member lower chamber - by forming alliances and presenting a joint list of up to two candidates in each district. Citizens receive a single vote that is cast directly for candidates. The d'Hondt rule allocates legislative seats by successively dividing each coalition's vote by integers up to the number of available appointments. Seats are then distributed in order of quotient size. Under the Chilean variant the first seat is awarded to the candidate with the most votes on the list with the largest combined plurality. If the combined vote of candidates on this list is two-thirds of the votes cast, then both candidates on the list win seats. Otherwise, the second seat goes to the candidate with the most votes on the list with the second largest combined plurality (Angell and Pollack, 1990; Angell, 1990).5 The military regime adopted these rules to give parties incentives to form broad coalitions and to give disproportionate representation to the minority parties - understood to be those of the right - in each district. Parties are encouraged to coalesce because the allocation of seats is a function of both individual vote receipts and the combined pluralities of a list. Consequently, a majority party candidate is not likely hurt by a coalition partner so long as he or she ranks first in vote receipts. Further, parties increase the likelihood of securing a legislature more to their liking by forming a coalition with like-minded parties. Disproportionate representation for the minority list is secured because to win both seats a coalition's plurality must be at least twice that of the next closest list. Not only is this unlikely, but it is possible that the second candidate on the majority list may receive more votes than the leading candidate on minority list and still not win a seat. This occurred in the Santiago-Poniente election in which right-wing candidate Jaime Guzman of the Union of Independent Democrats (UDI) was elected to the second district seat despite having obtained just over half vote share of Ricardo Lagos of the Party for Democracy (PPD). Similarly, in the Valparaiso - Vina del Mar elec- [235 ] 454 tion UDI candidate Beltran Urenda won the second seat with approximately sixty percent of the plurality received by Christian Democrat Juan Hamilton. 6 Each map presents the estimated spatial locations of the four Concertacion and Democracia y Progreso candidates, the major Concertacion and Democracia y Progreso parties, the Chilean Communist Party (PPCh) and the Concertacion de los partidos por la Democracia. The scaled Concertacion parties are the Christian Democratic Party (PDC), the party for Democracy (PPD) and the Socialist Party (PS). Democracia y Progreso parties scaled are the Union of Independent Democrats (UDI) and the National Renovation Party (RN). These five parties secured a combined total of 85 percent of the seats in the House of Deputies, the lower chamber of the Chilean legislature. The Chilean Communist Party, which was not a member of the Concertacion and did not obtain legislative representation, is included because of its traditional significance in Chilean electoral politics. The Concertacion is scaled because it is an umbrella organization encompassing several political organizations otherwise not considered in the analysis. Most of the senatorial elections were contested by other lists, but these were not generally competitive. In the elections considered in this study, the combined plurality of the two major coalitions averaged eighty-eight percent? Appendix I overviews the political parties included in this study. 3. Theoretical expectations Spatial analyses demonstrate that the centralizing tendencies of the two candidate model do not typically hold under multicandidate elections (Downs, 1957: 120-124; Hinich and Ordeshook, 1970; Palfrey, 1984; Greenberg and Weber, 1985; Greenberg and Shepsle, 1987; Cox, 1987, 1990a, 1990b; Shepsle, 1991). The primary conclusion from this literature is that multicandidate eqUilibrium generally do not exist. When multicandidate eqUilibria can be supported they are typically non-central in that at least some candidates locate away from the median voter ideal point. These equilibria, however, are often sensitive to model specification. Before characterizing expected candidate behavior under Chilean d'Hondt a brief note regarding candidate objectives is in order. That citizens vote for candidates rather than lists introduces very strong incentives for candidates to compete directly for votes even at the expense of coalition partners. Intracoalition competition was indeed evident. For example, in the SantiagoOriente election Democracia y Progreso candidate Sebastian Pinera and Hermogenes Perez de Arce campaigned almost as much against each other as the opposition (Angell and Pollack, 1990). Further, rather than maximizing votes per se, candidates may instead seek maximize rank finish within [236 ] 455 a coalition since it is unlikely that a coalition's vote share will exceed twothirds of the ballot cast. Only in the Sixth, Seventh and Twelfth Regions did the Concertacion win both senate seats. This further exacerbates the tendency for coalition partners to compete against each other. The scarcity of multicandidate equilibria and expectations for the Chilean Senate elections can be illustrated using a very simple spatial representation. Consider a bounded, uni-dimensional electoral space which, for the sake of simplicity, is set equal to the unit interval. Also assume that voters are distributed uniformly across the space and cast a single ballot for the closest of four candidates (AI A2, BI B2). This corresponds to most Chilean Senatorial elections which were typically contested by four major candidates; two representing the Concertacion and two representing Democracia y Progreso. Cox (1990b), building on Eaton and Lipsey (1975), outlines the motivation for a non-central equilibria illustrated in Example I in which four candidates compete for a single seat. Here the candidate pairing at the outer quartiles of the space is a Nash eqUilibrium as no candidate has a unilateral incentive to relocate as doing so results in a loss in expected plurality. In this example each candidate has an equal probability of winning the seat. Example 1 I o I I Regardless of whether candidates are vote or rank maximizers, neither a centralized eqUilibrium nor the non-central pairing result of Example I is supported under Chilean d'Hondt. Considering Example I again and holding constant the positions of all candidates except, for example, A 2 ; if candidate A2 moves to the right up to position 3/4 her plurality remains unchanged (always 25%) with the remaining votes in the interval redistributed from candidate BI to candidate AI. Despite her constant plurality, candidate A2 faces incentives to move to the right as any location in the interval (7112, 3/4) will guarantee the "1\' coalition at least two-thirds of the vote; hence not only her election but that of her coalition partner as well. More generally, Cox (1990b) shows that with vote maximizing candidates, if the number of votes allotted citizens is less than twice the district magnitude, there is no convergent equilibrium and some candidates will locate away from the ideal point of the median voter. Indeed, four candidate equilibrium degenerate as the distribution of voters changes from uniform to a more plausible unimodal distribution. If candidates seek to maximize ranked placement within the coalition, as seems likely in Chilean legislative elections, Denzau, [237] 456 Kats, and Slutsky (1985), and Greenberg and Shepsle (1987) demonstrate multi-agent equilibria are exceedingly rare and candidates should be dispersed across the electoral space. In either case, one expects centrifugal forces to dominate candidate strategies in Chilean legislative elections and at least some candidates to locate away from the ideal point of the median voter. 8 One might expect coalitions to pair strong and weak candidates to give the strongest candidate more latitude in pursuing electoral strategies. The balanced lists in all districts demonstrate this strategy was not generally followed. In the four elections considered in this study the inter-coalitional pluralities differed by an average of less than six percent, and this is heavily weighted by the single lopsided Concertacion ticket in Santiago-Oriente in which Eduardo Frei won approximately forty-one percent of the district vote. Of the seven remaining lists, the intra-coalition vote shares in five differed by less than two percent. In Santiago-Poniente, for example, the pluralities received by Concertacion candidates Lagos and Zaldivar differed by less than 8500 of the over 800,000 votes cast for the list. Comparable intra-coalition vote differences are found in the Concertacion and Democracy and Progress lists in Valparaiso - Vina del Mar, the Concertacion list in Region Six and the Santiago-Poniente Democracy and Progress list. Simply, as measured by aggregate vote receipts, each list presented two viable candidates. The plurality received by each candidate is presented in the maps. 4. Data and methodology The data used to estimate the spatial maps are seven point candidate and party evaluation scores obtained from cross sectional surveys conducted using in person interviews in each of the four senatorial districts. The polls completed in the last half of October, 1989, and average just over 600 respondents. Scores were obtained for all major candidates and political parties. In addition, the surveys solicited respondent self-placement on a left-right ideological scale and demographic information useful for interpreting the maps.9 The spatial maps locate candidates, parties and voters in the same space, and are estimated using a metric scaling methodology described by Cahoon, Hinich, and Ordeshook (1978), Enelow and Hinich (1984), Endersby and Hinich (1992) and Lin et al. (1996). Scaling methodologies and evaluation scores are commonly used to estimate Euclidian representations of elections (Weisberg and Rusk, 1970; Rusk and Weisberg, 1972; Rabinowitz, 1978; Davis, 1984; Poole and Rosenthal, 1984; Enelow and Hinich, 1984; Brady, 1990; Endersby and Hinich, 1992), although some researchers (Aldrich and McKelvey, 1977; Enelow, 1986, 1988; Enelow and Hinich, 1989, 1994) use [238] 457 perceptual data, consisting of voter self and candidate placement on issue scales, and factor analytic procedures for the same purpose. The details of the methodology are presented in Enelow and Hinich (1984: Appendix 9.1). For this reason the following discussion is brief. The statistical methodology assumes the i-th voter's evaluation score for candidate j, Tij, measures the spatial distance between the respondent and candidate. Specifically: where: njk is candidate j's location on dimension, k, Xik is voter's 1's location on the same axis and eij captures unmeasurable, non-systematic influences on Tij and is distributed independently across voters and candidates. The survey question soliciting Tij translates as: Thinking of you and your family'S well-being, how do you evaluate the following personalities as senator? Respond on a scale of 1 to 7 where 1 is very bad and 7 is very good. The corresponding question for political parties is similarly phrased. The statistical problem is to use the observed Tij to estimate n jk and X ik . Once these parameters are estimated one may use a variety of internal and external measures of validity to assess the quality of the maps. As Brady (1991) notes, estimating njk and X ik is non-trivial because Tij is non-linear in each parameter and restrictions must be imposed on the model to identify and estimate its parameters. The estimation method proceeds as follows. Consider an election with i = L.n voters andj = L.p + 1 candidates and parties. Two data transformations are used to make the observations linear in nj and Xi. First, the scores of a preselected candidate or party, T ip +!, are squared and then subtracted from the other similarly adjusted Tij scores. From this difference is subtracted the difference between each candidate's average score and the mean Ti across respondents. The choice of the candidate Tip +! is mathematically arbitrary, but interpreting and comparing the maps is easier if the candidate represents the status quo and is the same in each map. To this end, the centrist Christian Democratic party is the p + 1 party in all maps and its location is defined to be the origin of the space. This means the estimated positions of all other candidates and parties are relative to the PDC. The statistical methodology uses the variation in the transformed scores to estimate candidate and voter positions in the predictive space. The first step in the estimation procedure is to calculate the maximum likelihood factor [239] 458 LD, analysis of the sample covariance matrix, of the transformed scores. This produces, up to an arbitrary rotation, an estimate of the candidate and party positions in the predictive space. to Parameters of the model still to be estimated including the angle of rotation of the axes are obtained by two least squares regressions, the elements of which are detailed in Enelow and Hinich (1984: Appendix 9.1). Finally, the candidate positions are used in a least squares regression with dependent variable Tie - Tip+1 to estimate voter location, p, the predictive space. In the two dimensional case, this regression is: where JLj = eij - eip+l· The statistical methodology estimates a cartesian representation of the positions of candidates and parties on the defined axes. Few measures of statistical fit have been developed to assess the quality of spatial representations. Poole and Rosenthal (1984) propose the squared Pearson correlation coefficient between the observed thermometer scores used to estimate their spatial maps and those predicted by the recovered maps. This statistic, however, may overstate the quality of the spatial representation because the same data is used to estimate and evaluate the spatial maps. I use two measures of model fit in the same spirit as that proposed by Poole and Rosenthal, but not subject to the same concern. The first is the Pearson correlation coefficient between estimated voter axis location and voter self-placement on a five point left (1) to right (5) ideological scale. This statistic, reported in the maps as p, has two virtues. First, respondent ideological self-placement is not used in the statistical methodology and provides something of an independent check on the maps. If the estimated axes accurately reflects the electoral space, then one expects voter location to be highly correlated with ideological views. Second, because only p observations are used to estimate each voter's spacial location, the voter ideal points are somewhat noisy and, arguably, the weakest link in the methodology. If, despite this, one recovers strong statistical measures of fit based on the estimated voter locations, one is justified in placing considerable confidence in the spatial representation. The second set of statistics used to assess the quality of the spatial maps are obtained from a logistic regression in which the dependent variable indicates whether the respondent anticipates voting for a Concertacion (coded 1), or Democracy y Progreso (coded 0) candidate, and the single independent variable is estimated voter spatial location. The few voters expressing an intention to vote for other coalition candidates are treated as missing values. The logit analysis provides a simple means to [240] 459 assess the extent to which spatial location accounts for voter choice. Statistics reported with these regressions include the percentage of anticipated vote that is correctly predicted and several other standard measures of model fit. I estimated both one and two dimensional maps of the Chilean Senate elections, although only the single dimensional results are presented and discussed. The one dimensional maps are most appropriate for comparing expected candidate behavior derived from spatial analyses of multicandidate elections and observed candidate and voter behavior in the Chilean senatorial elections. The second dimension also added only marginally to the fit of the maps as measured by these statistics. Construction of the covariance matrix of adjusted scores requires that respondents assign evaluation scores to all scaled candidates and parties. Approximately 35% of all respondents were able to do this, but this figure is influenced by the lower response rate in Valparaiso - Vina del Mar. In the remaining districts the percentage of respondents able to assign evaluation scores to be scaled candidates and parties is comparable to that typically found in established democracies. Consistent with what one would expect in polity with a strong party system, the vast majority of missing values are due to voters failing to assign evaluation scores to candidates rather than parties. In the four senate elections, an average of 67% of respondents could assign evaluation scores to be seven scaled parties, while only 40% could do the same for the four Concertacion and Democracy and Progress candidates. In Appendix 2, I provide summary statistics for several respondent demographic characteristics for both the full survey sample and the scaled subsamples. Although respondents scaled in the maps are of slightly higher social class and better educated than those in the full survey sample, the selection mechanism introduces few systematic biases in the types of respondents scaled in the maps. The number of scaled respondents is reported with each map. 5. The spatial maps 5.1. Introduction and overview I present the candidate maps in Figures 1-4 and review them simultaneously. All were estimated for a one-dimensional solution and scale properly as evidence by p which exceeds 0.56 in all maps. I list the senatorial candidates, their party affiliations and pluralities above each map. The election winners are denoted by italics. I also present the standard deviation of the estimated voter ideal points on the axis and label the median ideal point (Xm) in the maps. The vertical axis measures the estimated density of the voter ideal points. I I In all districts the distribution of voters exhibits a major node near [241] 460 Moria II. c ........ Me Uadt) t tt ~:t"'- JII.O _TrII. IIr,PDC!) H_ _ doAno,BPtladm 4L1 14.1 _I'IMro,BPU""tt) Il.8 Canc:orlacl... eondld.1o Coadldolo 8D: - . . . _ _ oroollmolool _ _ poIDlo OD..-....I uiI. u u u n U i -II -0 -II I -IJ rPllrra. ,II;'" I -J.t l1li ... IJIII I I PIJC i ..II -u U IF BP.i U i U J.t IJ II 0 II II U U iAA-R/Pl AlII Figure 1. Santiago-Oriente. the ideal point of the median voter and most maps display one or more secondary nodes near the periphery of the space. There are several noticeable differences in the shapes of the four estimated densities. For example, the distribution of voters in the Santiago districts tends to be flatter than those in Vina del Mar and Region VI. The estimated candidate and party locations present clear left-right alignments on the recovered axis. Parties scale in the expected order, with the Christian Democratic Party located to the right of the other Concertacion parties and the Union of Independent Democrats consistently on the far right of the space. In all districts except Valparaiso - Vina del Mar, the PDC is clearly seen as distinct from the other Concertacion parties, but in relationship to the distribution of voters these differences are relatively minor. The Communist placement is the only possible exception to be natural order on the left-right axis. In all districts except Region VI the Socialists and the Party for Democracy scale to the left of the Communists. This would be expected in Chile prior to the military coup as through most of its history the Communist Party was more moderate than the Socialist Party and sought to achieve its political objectives through the electoral process (Gill, 1966; Burnett, 1970; Faundez, 1988; Garreton, 1989). This commitment to moderation was undermined during the military regime when an important faction, the Frente Patriotico Manuel Rodriguez, resorted to political violence [242] 461 N-S"" P .0.80 BDVolerldoolPolDIc UO 29.8 29.2 14.8 18.4 And,... Zaidi...,.. ItZ (POCf) RI..",. La.... RL(PPDtl M1ruel Ote... MO (RNlt) Joj_ 0 _ JO(1JDlttl t ConCt'rldc:ion Candidate It De"""....10 y ,.,.".... CaDdidate SD: Stondard d.... " . . .r ..Umated ...tor Idool polnte.n _ered ula. u u PrIIrrao PI I 0.0 -II -11 -1.0 -1.1 ij Iiij J. PUC All IlL -u -OJ to Ji OJ 1.1 LO 10 1.6 LD III I \JIll \ 110 III to 1.6 U 11 Figure 2. Santiago-Poniente. N.I03 p.O.8'l BIW..... ldool Poln.. lL81 ClUldld.tcund Juan Hamilton, JH (Poct) LaumBoto,lB(INDtl 8.lIran U..ndG, BU (UDItt) Oonzelo Yuooer. OY (RNtl) ~ofV_ t tt 27.3 'ZI.7 16.8 18.& Com:ertocion Candidate De....raci. y ,.,.",... C.ndidate aD: Standard dnlalion or estimated "tar Ideal poiDta OD rec:mered axil. o.a pro rrao i ~IJaT 0.0 -u -1.6 -1.0 III I ~ i -u -u -u IR II u u 10 \JIll GJ 1.6 II ! I I I1U to u II lAI\-RItht AlIa Figure 3. Valparaiso - Vina del Mar. [243 ] 462 N.I83 p.O.18 8DVoIAIr ...... PaInIoI1.17 CMd__ -..arv",", !=:.'"1~fIm) NICIJI.. DIa,NDIPDCIl 28.0 ::9.. II....... v....... IIV (INDft) t Cnnatinc:iOlll o.nd...... II Oem_:l"""" ean.J....'. , _.... _ _I poID'- DB _ SO: 1IIaad........_ u .... _ . u u fIG ... I", u 1ft 1 Ii AI -u -u -u -u -u I 1jII' lID u u Ii u u u u u II\' u , MU ' , 10 u u u LIA-1IIIId Alii Figure 4. Region VI - O'Higgins. culminating in a 1986 attempt on Pinochet's life, leading to the exclusion of the Communist Party from the Concertacion. Likewise, the Socialist Party moderated its stands toward civil affairs as reflected in its inclusion in the Concertacion. The Communist-Socialist placements demonstrate these changes in policies and tactics are not necessarily reflected in mass political perceptions during early transitional periods, and voters perceive important continuities with the pre-authoritarian political universe. As one expects in a strong party system, candidates scale very close to their respective parties and coalitions. In the Santiago districts Democracia y Progreso candidates Guzman, Perez de Arce, Otero and Pinera scale to the far right near the RN and UOI parties. Likewise, Christian Democratis Eduardo Frei, Juan Hamilton and Nicolas Diaz are nearly adjacent to the PDC locations in the recovered maps. Similar candidate-party proximity is recovered for PPD candidate Ricardo Lagos in Santiago-Poniente and RN candidates Gonzalo Yussef in Valparaiso - Vina del Mar and Alfonso Orueta in Region VI. Independent members of the Concertacion and Democracia y Progreso also scale quite close to those parties with which they are most clearly associated. Valenzuela and Valenzuela (1986), building on Lipset and Rokkan (1967) and Sartori (1976) argue, "the political landscape is more or less impermeable to change once it has been established" and the effect of repressing [244 ] 463 political actors as during the military regime is to hold parties and major political figures in the positions they occupied during the democratic period. The recovered Communist-Socialist placement, and the relative proximity of candidates - especially those who rose to positions of party leadership before the September 1973 coup - to associated political parties substantiates this argument. The candidate-party placements closely resemble what one would expect prior to the cessation of civil affairs, attesting to the strength of the Chilean party system. Table 1 presents the results of the logit regressions in which the dependent variable indicates whether the respondent anticipates voting for Concertacion or Democracy y Progreso candidate, and the single independent variable is voter spatial location. The table reports the percentage of anticipated vote that is correctly predicted by voter spatial location, a pseudo R 2 defined as 1InLpllnFR where InLp is the log-likelihood of the full model and InLR is the log-likelihood of the model estimated with intercept only, and the likelihood ratio statistic. The latter two statistics provide overall measures of model fit. I also report the results of a second logistic analysis in which anticipated voter choice is regressed on respondent spatial location, ideological self-placement and several demographic characteristics. The latter regressions aid in interpreting the estimated maps and place the quality of the spatial representation in perspective. To permit comparisons across the two sets of regressions, each is estimated using the sample of respondents for whom there exist a complete set of observations on the full set of independent variables. The parameter estimates in Table I further confirm the maps accurately reflect voter perceived relations among candidates and parties in the Chilean Senate elections. In all districts the logit coefficient for axis location is statistically significant and in the correct direction. Respondent location in the recovered space correctly predicts an average of about 92% of anticipated vote choice between the two major coalitions. This, in conjunction with the pseudo R 2 ,s and likelihood ratio statistics, verify the estimated maps capture the salient characteristics of the election. Detailed cross map comparison beyond candidate order on the recovered axes are problematic because there is not absolute origin in the space. The political meaning of being a centrist in Valparaiso - Vina del Mar, for example, may different considerably from what it means to be a centrist in the Santiago districts or Region VI, and it is impossible to decipher these differences from the maps alone. Consequently, cross district comparisons of the effect of a change in spatial location on voter behavior as measured from a fixed point, say the origin or the median voter ideal point, convey little useful information. Nonetheless, since the scales on which the spatial positions are based are the same across districts, comparisons of differences in positions are likely in- [245 ] 464 Table 1. Logistic regression of voter choice by voter axis location Santiago- Santiago- Valparaiso - Poniente Oriente VinadelMar Constant 1.62* (0.31) Axis location -1.98* (0.29) N Percent correctly predicted PseudoR2 LR Chi-square 245 91.0 0.64 203.1 1.15* (.030) -2.04* (0.32) 202 90.6 0.64 172.3 Region VI 4.91* (1.27) 3.05* (0.77) --4.09* (1.14) -1.87* (0.63) 132 95.5 0.81 120.3 84 90.5 0.51 35.4 Dependent variable: Anticipated vote choice for coalition candidates. Coded 1 if respondent anticipates voting for either Concertacion candidate; 0 if respondent anticipates voting for either Democracy y Progreso candidate. Estimated standard errors are in parentheses. LR Chi-square equals -2(lnLF-lnLR), and is distributed chi-square with 1 degree of freedom. p<0.05 formative. To provide an indication of the impact of a small change in spatial location on the probability of voting for a Concertacion candidate, apclaXaxis, across the maps I calculated the percentage change in the probability of voting for a concertacion candidate Pc one half unit to the right measured from the point on the recovered axis where probability of voting for a candidate of one coalition or the other equals 0.5. This point, of course, varies across districts depending on the scaled locations of candidates, the shape of the estimated distribution of voter ideal points and the aggregate political preferences in the districts. This measures, in some sense, the "typical" impact of a small change in spatial location on the probability of voting for the Concertacion at the most competitive location in each of the maps. The largest marginal effect is Valparaiso - Vina del Mar. Here a one-half unit change to the right from the point of indifference between candidates of the two coalitions, Xind = 1.2, reduces the likelihood of voting for a Concertacion candidate to 12%, a decrease of approximately 76%. The impact of a comparable change in spatial location is more modest in the remaining districts. In Santiago-Poniente and Santiago-Oriente the points of indifference between the coalitions are Xind = 0.82 and Xind = 0.57, respectively, with a one-half unit shift to the right in both districts reducing the probability of [246] 465 voting for a Concertacion candidate to 0.26. The corresponding figures for Region VI are X ind = 1.63 and 0.28. Table 2 presents the reestimated logistic regression of anticipated voter choice on axis location, this time adding respondent ideological selfplacement and several demographic characteristics to the list of independent variables. The demographic variables include an interviewer assessment of the respondent's social class as measured on a seven point scale, as well as respondent gender, age and level of education. 12 Previous studies of Chilean political behavior indicate that social class is only modestly associated with partisan preferences and women are generally more conservative than men (Langton and Rapoport, 1976; Portes, 1976; Valenzuela, 1985; Panzer and Paredes, 1991). The second set of logistic regressions demonstrate that respondent ideological self-placement and demographic characteristics add little information about voter choice in the senate elections beyond that summarized by the spatial maps. The incorporation of this additional information increases the proportion of correctly predicted vote by an average of only 3% relative to estimates based solely on spatial location. This is reflected in the minimal changes in the corresponding pseudo R2 ,s and likelihood ratio statistics. Only in the Santiago districts does the ideology variable contribute significantly to the explained variation in voter choice, but the relative importance of respondent ideology still lags that for respondent location in the estimated space. Once one controls for estimated spatial location, demographic variables add little information useful for understanding voter behavior in the Chilean senate elections. Neither social class nor gender is associated with voter choice, although in three of the four districts the estimated gender coefficients are in the expected conservative direction. Older and more educated voters in Santiago - Poniente are less likely to vote for the Concertacion, although the impact of age is quite minimal. That social class is not strongly associated with voter choice and the estimated coefficient for women is in three of four cases in the conservative direction further indicates the political universe that emerged from military rule exhibits continuity with that prior to 1973. 5.2. Evaluation of spatial location The Median Voter Result does not hold in the Chilean senatorial elections, and at least some candidates are expected to be located outside the central half of the distribution of voters. In all maps there is a wide gap between liberal and conservative candidates and parties. The estimated maps confirm that the Concertacion and, in particular, the Christian Democratic Party, owns the political high ground at the center of the electoral space. The Union of Independent Democrats, [247] 466 Table 2. Logistic regression of voter choice by spatial location, ideological self-placement and demographic characteristics Santiago- Santiago- Valparaiso - Poniente Oriente Vinadel Mar Constant 14.00 (4.29) 6.36 (3.28) Axis location -1.84* (0.32) -1.98* (0.37) -1.63* (0.52) Social class Female Ideology Age Education predicted PseudoR2 LR Chi-square ~.6* -8.78 (8.29) (2.21) -2.73* (1.25) -1.07* 0.34 -1.14 (0.38) (0.91) (0.67) -0.29 (0.45) -0.27 (0.34) 2.19 (1.15) 2.36 (1.35) -0.92 -0.12 (0.59) (0.58) 1.58 (1.36) -0.85 (1.10) -0.06* (0.02) -1.21 * -0.01 (0.02) 0.09 (0.05) -0.02 (0.03) -0.10 (0.49) 2.76 (1.63) 1.94 (1.26) (0.55) N Percent correctly -17.61 (9.52) Region VI 245 94.3 0.71 227.7 202 90.6 0.68 183.1 132 98.5 0.87 128.9 84 95.2 0.63 43.2 Dependent variable: Anticipated vote choice for coalition candidates. Coded I if respondent anticipates voting for either Concertacion candidate; 0 if respondent anticipates voting for either Democracy y Progreso candidate. Estimated standard errors are in parentheses. LR Chi-square equals -2(lnLF-InLR), and is distributed chi-square with 6 degrees of freedom. p<0.05 the National Renovation and Democracy y Progreso candidates scale to the right periphery of the space. In Valparaiso - Vina del Mar the Concertacion parties and candidates are adjacent to the median voter ideal point, while in the Santiago Districts and Region VI Concertacion parties other than the PDC scale well to the left of the median voter ideal point. Indeed, Eduardo Frei in Santiago-Oriente and Juan Hamilton in Valparaiso - Vina del Mar scale to the right of the origin and the median voter ideal point, but even in these districts the region between the left and right candidates is considerable. [248] 467 The most appropriate measure for assessing whether centrifugal or centripetal forces dominate candidate strategies in Chilean senatorial elections is candidate location relative to the distribution of voters. The standard deviation of the estimated voter ideal points on the recovered axis is a natural measure of distance in the recovered space as expectations of candidate locations derived from the spatial model are defined in terms of the distribution of voters. These figures support the expectation that there are few electoral incentives for all candidates to locate near the center of the voter distribution. The average distance separating the innermost candidates in each map is nearly 1.7 standard deviations. Nearly half of the estimated voter ideal points in the maps are between the interior candidates. This percentage ranges from 40% in Valparaiso - Vina del Mar to 47% in Santiago-Poniente and Region VI. An average of approximately 38% of the voter ideal points are to the left of the exterior Concertacion candidate, while only a nominal proportion, an average of approximately 6% are on the right of the exterior Democracia y Progreso candidate. Finally, as the estimated distribution of voter ideal points suggests, the number of voters between coalition partners, especially those on the right, is minimal. For example, no voters scale between Manuel Valdes and Anselmo Sule in Region VI nor are any voters located between Hermogenes Perez de Arce and Sebastian Pinera in Santiago-Oriente. The proportion of voters between Concertacion candidates is greater; ranging from 5% of the estimated ideal points in Santiago-Poniente to approximately 18% in Santiago-Oriente. 6. Conclusion The empirical spatial analysis largely support the thesis that centrifugal forces dominate candidate strategies in elections held under Chilean d'Hondt in dual member districts. In all districts candidates of the right clearly locate away from the ideal point of the median voter. On the left, several candidates including Maria E. Carrera in Santiago-Oriente, Ricardo Lagos in SantiagoPoniente and Anselmo Sule in Region VI scale further from the median voter than one would expect if candidates enjoyed moderate spatial mobility in an electoral system characterized by centripetal incentives. Collectively, the maps indicate there are few centralizing tendencies in Chilean senatorial elections. This makes sense given under Chilean d'Hondt one can win a senate seat with a nominal percentage of the vote. Jamie Guzman, Beltran Urenda, Alfonso Orueta of the Democracia y Progreso won seats with pluralities of less than 20%, indicating there are few incentives for these candidates to moderate their positions to secure larger pluralities. [249] 468 The existence of centrifugal electoral incentives means candidates most likely to emerge and win Chilean legislative elections are those capable of differentiating themselves from the center of the distribution of voter opinion. Indeed, over half of the candidates who won elections in the four senate elections considered in this study scaled to the exterior of the space. This suggests Chilean legislative candidates will primarily appeal to like-minded voters and electoral competition will largely center on mobilizing supporters in one's region of the electoral space. The maps demonstrate the Chilean polity that emerged from military rule shares both similarities and differences with that predating the Pinochet era. The most important differences center on the breadth of the Concertacion coalition and the lack of voter polarization relative to what one would expect prior to 1973. The relative Socialist-Communist locations with the Christian Democratic Party dominating the center of the political spectrum is similar to what one would expect before 1973. The estimated locations of candidates, parties and voters substantives the arguments of Scully (1992, 1995) and others that the Chilean political landscape reflects important continuities with the pre-authoritarian era, and substantiates the argument that polities with long histories of stable democratic governance are resilient to change. 13 The contemporary Chilean polity enjoys considerable stability and ranks among the most consolidated of democratic transitions. A note a caution is suggested by Chile's use of an extremely disproportional electoral system that provides incentives that, ceteris paribus, reward extremists and punish moderates. Political moderates can win Chilean legislative elections as demonstrated by the Santiago-Oriente election of Eduardo Frei and his subsequent election to the Chilean presidency in 1994. The estimated maps demonstrate, however, that parties, especially those on the right, have few incentives to nominate moderates and, at the margins, candidates face electoral incentives to preserve parochial basis of support rather pursue more centrist campaign strategies. These incentives can be reversed with relatively simple changes in the electoral system such as providing citizens with an additional vote. Notes 1. An important exception is Lin, Chu, and Hinich (1996). 2. Concertacion candidates won approximately 60% of the seats in the House of Deputies, the lower chamber of the Chilean legislature. The presence of nine appointed senators in the Chilean senate left the Concertacion with just under 50% of the chamber seats. 3. Chilean regions are numbered from one in the north to twelve in the south, and are generally referred to by number rather than name. The fifth, seventh, eighth, ninth and tenth administrative regions are allotted two senatorial districts. Of these, regions five [250] 469 4. 5. 6. 7. 8. 9. 10. II. 12. (Valparaiso) and eight (Bio-Bio) have large urban areas. The other regions with two senate district are in traditionally conservative southern Chile. I study these elections because only data for these districts were made available by the survey's principal investigator. See Section 4 and note 9 for a discussion of the survey data and methodology. d'Hondt is among the least proportional of all proportional representation rules (Jones, 1993; Lijphart, 1994). For a discussion of the d'Hondt rule see Taagepera and Shugart (1989: 32-34). All election results are obtained from La Epoca. 16 de diciembre de 1989. In addition to the Concertacion and Democracia y Progreso candidates, the Region Six map was estimated including the two candidates of the rightist Avanzada Nacional coalition. This is because the maximum likelihood estimation procedure described in Section 4 failed produced optimal estimates using the smaller covariance matrix. Perhaps not coincidentally, only in Region Six did the Avanzada Nacional secure more than a nominal plurality, winning nearly II % of the vote. Although these candidates scaled as expected on the far right of the space, their locations are not presented to keep the graphical representation consistent with the other maps. The inclusion of the two additional candidates did not significantly reduce the number of observations used to estimate this map. Consistent with this, Adams' (1996) find greater ideological heterogeneity among Illinois state legislators elected from dual member districts than those elected from single member districts. Polling was conducted by the INVESTMARK company and commissioned by the Buchi presidential campaign. The surveys were designed and implemented to the highest academic standards. Since there may be concerns about the integrity of data collected for partisan purposes, the data, survey questionnaires and translations, and details on the survey methodology are available on request. is a function of candidate positions in the underlying space, and under very weak assumptions, the estimated candidate locations are normally distributed (Anderson and Amemiya (1988). The estimated density is used for graphical purposes only. All data analysis and interpretations are based on the raw estimated voter ideal points. The density is estimated using an Epanechnikov kernel with a smoothing window of approximately 0.60. For a discussion see Scott (1992: 125-145). Social class is coded according to the interviewer's assessment of respondent social class on a high (I) to low (7) scale, age is the respondent's age in years, and education is measures on a I (no formal education), 2 (some or all primary), 3 (some or all secondary), 4 (some or all college) scale. The left-right ideological self-placement question translates as: Ln In politics one normally speaks of right and left. On this chart different political positions may be found (I left - 5 right). Please tell me where you place yourself. 13. See Valenzuela and Valenzuela (1986) for a discussion of the resilience of the Chilean party system to military rule and Scully (1992: Ch. 5). [251 ] 470 References Adams, J. and Merrill III, S. (1998a). A Downsian model of candidate competition in the 1988 French presidential election. Typescript. Santa Barbara: University of California. Adams, 1. and Merrill III, S. (1998b). Party structures and policy representation in multiparty elections: An application to Norwegian politics. Typescript. Santa Barbara: University of California. Adams, G. (1996). Legislative effects of single-member vs. multi-member districts. American Journal of Political Science 40: 129-144. Aldrich, J.H. and McKelvey, R.D. (1977). A method of scaling with applications to the 1968 and 1972 presidential elections. American Political Science Review 71: 111-130. Anderson, T.W. and Amemiya, Y. (1988). The asymptotic normal distribution of estimators in factor analysis under general conditions. The Annals of Statistics 16: 759-771. Angell, A. (1990). The Chilean elections of 1989. Electoral Studies 3: 241-245. Angell, A. and Pollack, B. (1990). The Chilean elections of 1989 and the politics of the transition to democracy. Bulletin of Latin American Research 9: 1-23. Black, D. (1958). Theory of committees and elections. New York: Cambridge University Press. Brady, H.E. (1991). Traits versus issues: Factor versus ideal-point analysis of candidate thermometer ratings. In J.A. Stimson (Ed.), Political Analysis, Vol. II. Ann Arbor: University of Michigan Press. Burnett, B.G. (1970). Political groups in Chile. Austin: University of Texas Press. Cahoon, L.S. and Hinich, M.J. (1976). A method for locating targets using range only. IEEE Transactions on Information Theory 22: 217-225. Cahoon, L.S., Hinich, MJ. and Ordeshook, P.e. (1978). A statistical multidimensional scaling method based on the spatial theory of voting. In P.e. Wang (Ed.), Graphical representation of multivariate data. New York: Academic Press. Caviedes, C.N. (1991). Elections in Chile: The road toward redemocratization. Boulder: Lynne Rienner Publishers. Cox, G.w. (1987). Electoral equilibrium under alternative voting institutions. American Journal of Political Science 31: 82-108. Cox, G.W. (1990a). Centripetal and centrifugal incentives in electoral systems. American Journal of Political Science 34: 903-935. Cox, G.w. (1990b). Multicandidate spatial competition. In J.M. Enelow and M.J. Hinich (Eds.), Advances in the spatial theory of voting. New York: Cambridge University Press. Davis, S.K. (1984). A nonmetric test of spatial theories of elections. Political Methodology 10: 1-28. Denzau, A., Kats, A. and Slutsky, S. (1985). Multi-agent equilibria with market share and ranking objectives. Social Choice and Welfare 2: 95-117. Downs, A. (1957). An economic theory of democracy. New York: Harper and Row. Eaton, B.C. and Lipsey, R. (1975). The principle of minimum differentiation reconsidered: Some new developments in the theory of spatial competition. Review of Economic Studies 42: 27-50. Endersby, J.w. and Hinich, M.J. (1992). The stability of voter perceptions: A comparison of candidate positions across time using the spatial theory of voting. Journal ofMathematical and Computer Modeling 16: 67-83. Enelow, J.M. (1986). The linkage between predictive dimensions and candidate issue positions in American presidential elections: An examination of group based differences. Political Behavior 8: 245-261. [252] 471 Enelow, J.M. (1988). A methodology for testing a new spatial model of elections. Quality and Quantity 22: 347-364. Enelow, J.M. and Hinich, M.J. (1984). The spatial theory of voting: An introduction. New York: Cambridge University Press. Enelow, J.M. (1989). The locations of American presidential candidates: An empirical test of a new spatial model of elections. Mathematical and Computer Modeling 12: 461-470. Enelow, J.M. (1994). A test ofthe predictive dimensions model in spatial voting theory. Public Choice 78: 155-170. Enelow, J.M. and Hinich, M.J. (Eds.). (1990). Advances in the spatial theory of voting. New York: Cambridge University Press. Faundez, J. (1988). Marxism and democracy in Chile. New Haven: Yale University Press. Garreton, M.A. (1989). The Chilean political process. Winchester, MA: Unwin Hyman. Gill, EG. (1966). The political system of Chile. Chapel Hill: University of North Carolina Press. Green, D.P. and Shapiro, I. (1994). Pathologies of rational choice theory New Haven: Yale University Press. Greenberg, J. and Shepsle, K. (1987). The effect of multiparty rewards in multiparty competition with entry. American Political Science Review 81: 525-537. Greenberg, J. and Weber, S. (1985). Multiparty equilibria under proportional representation. American Political Science Review 79: 693--703. Hinich, M.J. and Mungeer, M.e. (1994). Ideology and the theory of political choice. Ann Arbor: University of Michigan Press. Hinich, M.J. (1996). Analytical politics. New York: Cambridge University Press (forthcoming). Hinich, M.J. and Ordeshook, P.C. (1970). Plurality maximization versus vote maximization: A spatial analysis with variable participation. American Political Science Review 64: 772791. Iversen, T. (1994). Political leadership and representation in West European democracies: A test of three models of voting. American Journal of Political Science 38: 45-74. Jones, M.P. (1993). The political consequences of electoral laws in Latin America and the Caribbean. Electoral Studies 12: 59-75. Langton, K.P. and Rapoport, R. (1976). Religion and leftist mobilization in Chile. Comparative Political Studies 9: 277-308. Laver, M. and Schofield, N. (1990). Multiparty government. New York: Oxford University Press. Lijphart, A. (1994). Electoral systems and party systems. New York: Oxford University Press. Lin, T.-M., Chu, Y.-h. and Hinich, M.J. (1996). Conflict displacement and regime transition in Taiwan: A spatial analysis. World Politics (forthcoming). Lipset, S.M. and Rokkan, S. (1967). Cleavage structures, party systems and voter alignments. In S.M. Lipset and S. Rokkan (Eds.), Party systems and voter alignments: Cross national perspectives. New York: The Free Press. Macdonald, S.E., Listhaug, O. and Rabinowitz, G. (1991). Issues and party support in multiparty systems. American Political Science Review 85: 1107-1131. Macdonald, S.E., Rabinowitz, G. and Listhaug, O. (1995). Political sophistication and models of issues voting. British Journal of Political Science 25: 453-483. Merrill, S. III. (1994). A probabilistic model for the spatial distribution of party support in multiparty elections. Journal of the American Statistical Association 1190-1197. Merrill, S. 1lI. (1995). Discriminating between the directional and proximity spatial models of electoral competition. Electoral Studies 14: 273-287. [253] 472 Palfrey, T.R. (1984). Spatial equilibrium with entry. Review of Economic Studies 51: 139-156. Panzer, J. and Paredes, R. (1991). The role of economic issues in elections: The case of the 1988 Chilean presidential referendum. Public Choice 71: 51-59. Poole, K. and Rosenthal, H. (1984). U.S. presidential elections 1968-1980: A spatial analysis. American Journal of Political Science 28: 282-312. Portes, A (1976). Occupation and lower-class political orientation in Chile. In A. Valenzuela and lS. Valenzuela (Eds.), Chile: Politics and society. New Brunswick: Transaction Books. Rabinowitz, G. (1978). On the nature of political issues: Insights from a spatial analysis. American Journal of Political Science 70: 804-809. Rabinowitz, G. and Macdonald, S.E. (1989). A directional theory of issue voting. American Political Science Review 83: 93-121. Rabinowitz, G., Macdonald, S.E. and Listhaug, O. (1991). Old players in a new game: Party strategy in multiparty systems. Comparative Political Studies 24: 147-185. Rusk, J. and Weisberg, H. (1972). Perceptions of presidential candidates. Midwest Journal of Political Science 16: 388-410. Sartori, G. (1976). Panies and party systems. New York: Cambridge University Press. Schofield, N., Martin, A, Quin, K. and Whitford, A (1998). Multiparty electoral competition in the Netherlands and Germany: A model based on the multinomial probit. Public Choice (forthcoming). Schofield, N. (1998). A comparison of majoritarian and proportional electoral systems based on spatial modeling and 'rational' politicians. Typescript. St. Louis: Washington University. Scott, D.W. (1992). Multivariate density estimation. New York: Wiley. Scully, T.R. (1992). Rethinking the center: Party politics in nineteenth- and twentieth-century Chile. Stanford: Stanford University Press. Scully, T.R. (1995). Reconstructing party politics in Chile. In S. Mainwaring and T.R. Scully (Eds.), Building democratic institutions Stanford: Stanford University Press. Shepsle, K.A. (1991). Models of multiparty electoral competition. New York: Harwood Academic. Shepsle, K.A. and Cohen, R.N. (1990). Multiparty competition, entry and entry deterrence in spatial models of elections. In lM. Enelow and M.l Hinich (Eds.), Advances in the spatial theory of voting. New York: Cambridge University Press. Taagepera, R. and Shugart, M.S. (1989). Seats and vote. New Haven: Yale University Press. Valenzuela, A (1985). Origins and characteristics of the Chilean system: A proposal for a parliamentary form of government. Working Paper No. 164, The Woodrow Wilson Center. Valenzuela, S. and Valenzuela, A. (1985). Chile and the breakdown of democracy. In H.J. Wiarda and H.P. Kline (Eds.), Latin American politics and development. Boulder: Westview Press. Valenzuela, A and Valenzuela, S. (1986). Party opposition under the authoritarian regime. In A Valenzuela and J.S. Valenzuela (Eds.), Military rule in Chile. Baltimore: The Johns Hopkins University Press. Walker, I. (1993). Political alliances and the role ofthe centre: The Chilean Christian Democratic Party. In A. Angell and B. Pollack (Eds.), The legacy of dictatorship. Liverpool: Institute of Latin American Studies. Weisberg, H. and Rusk, J. (1970). Dimensions of candidate evaluation. American Political Science Review 64: 1167-1185. [254 ] 473 Appendix 1 Political parties scaled in the spatial maps Party! coalition Map Christian Democratic Party PDC Center Left. Dominant legislative party. Opposes both the far left and Pinochet. The PDC was the primary legislative opposition to president Salvador Allende between 1970 and 1973. Won 12 senate seats. National Renovation RN Party for Democracy PPD Union of Independent Democrats Socialist Party UDI Traditional Right. Economically conservative and moderately sympathetic to political reforms. Won 11 senate seats. Moderate Socialist party. Originally formed as an instrumental party to organize voters to oppose the military government in the 1988 plebiscite on the continuation of Pinochet's rule. Several prominent politicians sought election as PDC candidates in the 1989 legislative elections. Won four senate seats. Right. Economically conservative and closely identified with the Pinochet regime. Opposed to major political reforms. Won a single senate seat. Far left of Chilean Politics. Traditionally more ambivalent towards the electoral process, and subject to internal ideological divisions. Won a single senate seat. Left. Traditionally associated with the Soviet Union, and the pursuit of electoral basis of political power. Not part of the Concertacion because of its support of violence against the Pinochet government. Failed to secure legislative representation. Left and Center-Left coalition consisting primarily of the PDC, PPD and Socialists. Supported Aylwin for the presidency. Characteristics acronym PS Communist Party PPCh Concertacion Coalition CPD [255 ] 474 Appendix 2 Mean and standard deviation of respondent characteristics in survey sample and scaled subsample District Santiago-Oriente Full sample (n = 642) Scaled respondents (n =303) Santiago-Poniente Full sample (n =610) Scaled respondents (n = 242) Valparaiso - Vina del Mar Full sample (n =718) Scaled respondents (n = 133) Region VI Full sample (n =482) Scaled respondents (n = 163) 1Percent female Ideological selfplacement Female 1 Age Social class Education 3.00 (0.91) 57.8% 5.18 (0.84) 2.86 (0.98) 50.8% 40.82 (16.07) 39.48 (14.89) 5.11 (0.91) 2.58 (0.71) 2.70 (0.72) 2.95 (1.00) 57.0% 38.4 (15.09) 4.85 (1.08) 2.80 (0.71) 2.97 (1.08) 52.9% 39.0 (15.00) 4.48 (1.08) 3.03 (0.70) 2.84 (1.17) 54.0% 2.56 (1.20) 45.1% 41.28 (16.49) 36.57 (14.37) 5.59 (0.66) 5.53 (0.71) 2.31 (0.73) 2.47 (0.78) 2.89 (0.95) 63.7% 42.06 (16.87) 5.11 (0.97) 2.62 (0.72) 2.83 (1.12) 50.3% 42.13 (16.89) 4.93 (0.97) 2.82 (0.73) respondents in the sample and scaled subsample. All variables measured as described in Note 12. [256 ] Public Choice 97: 475-490, 1998. © 1998 Kluwer Academic Publishers. 475 Empirical evidence of paradoxes of voting in Dutch elections AD M.A. VAN DEEMEN 1 & NOEL P. VERGUNST2 1 University of Nijmegen, Department of Political Science, P. O. Box 9108, 6500 HK Nijmegen, The Netherlands; e-mail: A.vanDeemen@BW.KUN.NL; 2Free University Amsterdam, Department of Political Science and Public Administration, De Boelelaan 108Ic, 1081 HV Amsterdam, The Netherlands; e-mail: NP.Vergunst@SCW.VUNL Abstract. In this paper we analyze four national elections held in 1982, 1986, 1989 and 1994 in the Netherlands on the occurrence of the Condorcet paradox. In addition, we investigate these elections on the occurrence of three so-called majority-plurality paradoxes. The first paradox states that a party having a majority over another party may receive less seats. The second states that a Condorcet winner may not receive the largest number of seats and even may not receive a seat at all. The third says that the majority relation may be the reverse of the ranking of parties in terms of numbers of seats. 1. Introduction The Condorcet paradox is concerned with voting situations in which no majority winner exists, in spite of the fact that voters have consistent preferences. The paradox was discovered in the eighteenth century by the French philosopher and mathematician Marquis de Condorcet (1743-1794). To illustrate the paradox, we present an example constructed by Condorcet (1789). The voting situation concerns sixty voters and three candidates called Pierre, Paul and Jacques. The distribution of the voters is as follows: 23 voters: Pierre Paul Jacques 2 voters: 17 voters: Paul Pierre Jacques Paul Jacques Pierre 10 voters; Jacques Pierre Paul 8 voters: Jacques Paul Pierre The voter preferences are supposed to be transitive. Thus a voter with for instance the preference (Pierre, Paul, Jacques) prefers Pierre to Paul, Paul to Jacques and, because of transitivity, Pierre to Jacques. According to this preference profile, Pierre has a 33 to 27 majority over Paul, Jacques has a 35 to 25 majority over Pierre and Paul has a 42 to 18 majority over Jacques. [257 ] 476 Hence there is no candidate with a majority over every other candidate; there is no majority winner or Condorcet winner. Condorcet also compared the majority rule with other voting rules, in particular the plurality rule. As is well known, the plurality rule selects the candidate with the highest number of first places in the voter preferences. Thus for the example above, this rule would have selected Pierre as the plurality winner, in spite of the fact that a majority of voters prefers Jacques to Pierre. Condorcet (1785, 1789) also discovered that a plurality winner and a Condorcet winner may differ. Moreover, he came to the general conclusion that different voting rules may yield different outcomes for the same voting situation. Though this insight seems trivial by now, it is in fact at the heart of a controversy between liberalism and populism (or, better, majoritarianism) in the theory of democracy. See, e.g., Riker (1982) and Tlinnsjo (1992). Clearly, if the Condorcet paradox occurs, then either no winner can be selected when the majority rule is used or a minority candidate is selected in the case of another rule, for instance the plurality rule. Thus in the case of a paradox, the plurality rule selects a candidate that can be beaten by a majority. If instances of the paradox frequently occur, then either many decision deadlocks can be observed in the case of the majority rule or many minority candidates will be selected in the case of the plurality rule. Therefore, the question of how frequently the Condorcet paradox occurs is important. In this paper we investigate four Dutch elections on the occurrence of the Condorcet paradox. The elections were held in 1982, 1986, 1989 and 1994. The party preferences of the voters in these elections are constructed by using data of the Dutch Parliamentary Election Studies. We also investigate the occurrence of other paradoxes related to the plurality rule and the majority rule. These so-called majority-plurality paradoxes are presented in Van Deemen (1993). The first paradox of this kind occurs when a party is preferred by a majority of the electorate to another party and yet receives less seats than that other party. The second paradox occurs when a Condorcet winner does not receive the largest number of seats. In fact, it is even theoretically possible in list systems of proportional representation like the one used in the Netherlands that the Condorcet winner receives no seats at all. Also this possibility will be investigated. Finally, the third majority-plurality paradox occurs when the ranking of parties in accordance with the vote or seat distribution is the reverse of the ranking of parties as obtained by the majority rule. The paper is organized as follows. Section 2 shortly reviews existing research results concerning the relevance of the Condorcet paradox. In Section 3, the concept of election matrix is introduced and discussed. An election matrix is a suitable tool for the empirical representation of voter preferences in elections. Of course, we show how to detect a Condorcet paradox from an [258 ] 477 election matrix. Section 4 presents the election matrices of the Dutch elections held in 1982, 1986, 1989 and 1994. Further we investigate the matrices on the occurrence of the Condorcet paradox and majority cycles. In the subsequent section, we investigate the data on the occurrence of majority-plurality paradoxes. The final section contains conclusions and discusses future research. 2. The relevance of the Condorcet paradox After a long period of slumbering existence, the interest in the Condorcet paradox was raised again by Arrow (1963) and Black (1958). Their rediscovery of the paradox has led to two important research fields in public choice. In the first, conditions are formulated that require voter preferences to satisfy certain regularities in such a way that the paradox is avoided. So the aim of this research is to find conditions that forbid the occurrence of the paradox. The most-known condition in this respect is single-peakedness formulated by Black (1958). This condition requires that alternatives (or motions as Black called them) can be linearly ordered in such a way that the graph of each voter preference has one and only one peak. Black was able to prove that in case of single-peaked preferences the optimum of the median voter preference must be the majority winner when the number of voters is odd. This result is called the Median Voter Theorem. Further he showed that singlepeakedness ensures the transitivity of majority decision-making in the case of odd numbers of voters and its quasi-transitivity in the case of even numbers of voters. The second research field does not forbid the Condorcet paradox, but instead tries to find out how often it will occur in real decision-making situations. This research line can be split up into two sub-fields. The task in the first sub-field is to calculate the probability of the Condorcet paradox for several numbers of voters and alternatives. The second line aims at detecting empirically instances of the paradoxes. We discuss both sub-fields. The probability approach was initiated, again, by Black (1958). He arrived at a probability of .0556 for the paradox in the case of three voters and three alternatives and given the fact that voter preferences are linear orderings (i.e., rankings without ties). He was not able to give further calculations but he conjectured that the probability of the paradox "increases rapidly with an increase in the number of motions" (Black 1958: 51). Since Black, several researchers have calculated the probabilities of the Condorcet paradox by means of computer simulations (Campbell and Tullock, 1965; Klahr, 1966) as well as by means of analytical expressions (DeMeyer and Plott, 1970; May, 1971; Niemi and Weisberg, 1968; Garman and Kamien, 1968; Gehrlein [259] 478 Table 1. Probabilities of no Condorcet-winners for Dutch elections (1982-1994) Election year Number of parties Probability of no Condorcet-winner 1982 13 .56869 1986 12 .54706 1989 9 9 .45453 1994 .45453 Source: Gehrlein (1983); Gehrlein & Fishburn (1976). and Fishburn, 1976; Gehrlein, 1983). The general conclusion to be drawn from the calculations is that, indeed, the probabilities rapidly increase with an increasing number of voters but this increase is less thaI). in the case of a growing number of alternatives. To find the probabilities for the four Dutch elections under scrutiny, we suppose that the size of the electorate approaches infinity so that we can use the limit values for the several numbers of parties as calculated in the studies mentioned above. The probabilities for the elections are given in Table 1. However, the results in this table should be interpreted with care. First, the calculations are based on the assumption of equally likely voter preferences. This assumption, which is called Impartial Culture, is highly implausible. Studies in which the probabilities of the paradox under alternative cultures or distribution assumptions are calculated, are scarce. In addition, the limit values under different cultures are still unknown. The second reason to be careful is that most of the calculations start from the assumption that voter preferences are linear orderings, i.e., rankings without ties. This is not .a realistic approach. As we shall see, we found many ties in the voter preferences in the four Dutch elections under scrutiny. Also see Niemi (1970). Recently, Jones, Radcliff, Taber, and Timpone (1995) calculated the probability of the paradox for weak orderings, thus for rankings with ties. However, they did not provide limit values for several numbers of alternatives so that their results cannot be used for national elections. Further they defined the paradox as the absence of a unique Condorcet winner. Now it is clear that in the case of weak orderings, two or more Condorcet winners may exist. In this work, we retain the definition of the paradox as the absence of a Condorcet winner and not as the absence of a unique one. So, situations with two or more Condorcet winners will not be considered as paradoxes. [260] 479 The second field in this research tradition is directed at finding empirical instances of the paradox. This research was initiated by Riker (1965), who found a paradox in the U.S. senate in 1911 and a paradox in the House of Representatives in 1956. It is useful here to make a distinction between voting in committees and voting in large elections. Since Riker's study, a number of other empirical instances of the paradox have been found in committee voting situations. Blydenburgh (1971) found two Condorcet paradoxes in the U.S. House of Representatives in 1932 and 1938. Jamison (1975), using experimental data about preferences in small groups, also detected a Condorcet paradox. More recently, Vergunst (1996) found a Condorcet paradox in a case treated in the Second Chamber of the Dutch parliament in 1994. In contrast to these results, instances of the paradox in large elections are difficult to detect. This is remarkable, since probability calculations indicate that the paradox should occur far more frequently in large elections than in comparatively small committees. So, Chamberlin, Cohen, and Coombs (1984) did not find a paradox in the case of elections for the American Psychological Association. Similarly, Niemi and Wright (1987) did not find a paradox in the Presidential Elections of 1980. Feld and Grofman (1992) investigated preference data of 36 elections held in professional organizations, unions and non-profit organizations. They did not find a Condorcet paradox either. Finally, Radcliff (1994) investigated president elections in the U.S. Neither he was able to detect the paradox. The only and so far unique exception is Niemi (1970), who discovered a paradox in a case of university elections. In this paper we search for Condorcet paradoxes in four national elections held in the Netherlands. Basic in our detection method is the concept of an election matrix. 3. Election matrices Election matrices provide a powerful tool for analyzing election data. In particular, it is a convenient device for detecting paradoxes of voting. Let the number of voters who strictly prefer Xi to Xj be denoted by N(Xi,PXj) and let m be the number of alternatives. An election matrix is defined as a square matrix E = «eij)), where eij = N(XiPXj) for each i,j = 1,2, ... ,m and i =1= j, and where ejj =0 for i = 1,2, ... ,m. Thus each cell «eij)) in an election matrix contains the number of voters who strictly prefer alternative Xi to alternative Xj. The matrix (1/n).E denotes the proportional election matrix. A cell «Xi, Xj)) in this matrix contains the proportion of voters strictly preferring Xi to Xj. Let E' denote the transpose of E. Clearly, the matrix S = E - E' is skewsymmetric, that is, «sij)) = «-Sji)) for i, j = 1,2, ... ,m. The entries in S show the margins of the voters for an alternative over another alternative. If the margin is non-negative, then that alternative is preferred by a majority to the other [261] 480 alternative. It is negative if the other alternative is strictly majority-preferred. Clearly, if a row contains only non-negative numbers, the alternative represented by the row is a Condorcet winner. Thus the matrix S provides an easy method for detecting Condorcet paradoxes: look for rows with only non-negative numbers. If there are none, the paradox has occurred. 4. Empirical evidence in four Dutch elections In this paragraph we present and analyze the preference data concerning the four national elections of 1982, 1986, 1989 and 1994. First, the nature of the preference data will be explained. In order to illustrate our working procedure, we then present and analyze the election matrix of 1994 obtained from these data. Finally, the full majority relation for each of the elections of 1982, 1986, 1989 and 1994 is produced and analyzed. Of course, the data will be investigated on the occurrence of the Condorcet paradox. The data are taken from the Dutch parliamentary election study (NKO) of 1982, 1986, 1989 and 1994. In the period before and after each election of the Dutch parliament, about 1500 respondents are interviewed about a wide range of issues. One of the many questions concerns the probability that a respondent will ever vote for a party. Each respondent can choose a point on a scale from 1 ("certainly never") through 10 ("some time certainly"). We will use the probability votes obtained in this way for the construction of the voters' party preferences. We assume that if a voter gives more points to party x than to party y, she strictly prefers x to y. Notation: Pi(X) > Pi(y) ~ xPiy, where Pi(X) stands for the probability future vote score of respondent i for party x and where P stands for strict preference. Furthermore, p(x) = p(y) for a voter means that this voter is indifferent with respect to parties x and y; notation: xly. The voter preferences constructed in this way are then aggregated by the majority rule. According to this aggregation procedure, party x is majority-preferred to party y if the number of voters who strictly prefer x to y is at least as large as the number of voters who prefer y to x; notation: xMy ~ N(xPiy) ::: N(yPix). In this case, we say that party x has a majority over party y. The binary relation M thus obtained over all pairs of parties is called the majority relation. Notice that according to this definition, indifference does not affect the majority relation between parties. In this context, a Condorcet winner is a party having a majority over every other party. In other words, x is a Condorcet winner if xMy for all parties y. Table 2 contains the election matrix obtained from the individual scores collected in the Dutch national election in 1994. For example, the cell (D66, PvdA) in this matrix contains the number of 655 while the cell (PvdA, D66) contains the number of 580. This means that 655 respondents strictly prefer [262] 481 Table 2. Majority scores for the 1994 elections in the Netherlands: N(XPiY) Party D66 PvdA CDA VVD Groen Links RPF SGP GPV CD D66 PvdA CDA VVD GroenLinks RPF SGP GPV CD 0 580 480 471 208 152 136 146 50 655 0 608 540 300 191 194 191 81 818 715 0 594 508 99 89 92 61 753 767 607 0 505 160 154 153 51 988 905 779 783 0 240 237 238 87 995 914 901 847 723 0 107 74 56 1036 946 930 883 743 114 0 79 59 1026 946 937 885 743 94 95 0 61 1262 1183 1151 1137 996 545 573 562 0 Source: Dutch Parliamentary Election Study (n = 1527). D66 to PvdA and that 580 strictly prefer the PvdA. Since the total number of respondents is 1527, we conclude that 292 of the respondents are indifferent or have not given a future probability score for at least one of the two parties. (This means that the numbers of respondents are not the same for each pair.) The matrices for the elections of 1982, 1986, and 1989 are given in the Appendix. The skew-symmetric election matrix for 1994 is the matrix difference between the election matrix and its transpose. It is given in Table 3. In this matrix, the first row contains only non-negative numbers. Therefore we conclude that the corresponding political party D66 is the Condorcet winner. Hence, the Condorcet paradox did not occur in this national election. Proceeding in the same way for the elections of 1982, 1986, and 1989, we arrive at the conclusion that no Condorcet paradox did occur in any of these elections. Notice that this does not mean that the concerned majority relations are acyclical. A Condorcet winner may beat any other party involved in a cycle, as long as this winner itself is not involved in a cycle. In other words, the existence of a Condorcet winner only implies that there is no top-cycle. To detect majority cycles, we need the majority relation for each election. Such a relation can be constructed by means of the skew-symmetric election matrices by using the fact that a non-negative number in such a matrix implies the row party to have a majority over the column party. Proceeding in this way we arrive at the following majority relations for the concerned elections (see Table 4). [263 ] 482 Table 3. Skew-symmetric election matrix 1994 Party D66 PvdA CDA VVD GRLNKS RPF SGP GPV CD D66 PvdA CDA VVD GRLNKS RPF SGP GPV CD 0 -75 -338 -282 -780 -843 -900 -880 -1212 75 0 -107 -227 -605 -723 -752 -755 -ll02 338 107 0 282 227 -13 0 -278 -687 -729 -732 -1086 780 605 271 278 0 -483 -506 -505 -909 843 723 802 687 483 0 -7 -20 -489 900 752 841 729 506 7 0 -16 -514 880 755 845 732 505 20 16 0 -501 1212 llOI 1090 1086 909 489 514 501 0 -271 -802 -841 -845 -1090 13 Table 4. Social rankings based on the majority rule and on the election results of the Dutch parliament, 1982-86-89-94 Ranking 1 2 3 4 5 6 7 8 9 10 11 12 13 1982 MR CDA VVD PvdA D66 DS70 PPR PSP SGP CPN GPV RPF CP BP PR PvdA CDA VVD D66 PSP SGP CPN PPR RPF CP GPV DS70 BP 1986 MR CDA PvdA D66 VVD PPR PSP CPN SGP GPV EVP RPF CP PR CDA PvdA VVD D66 SGP PPR PSP GPV RPF CPN CP EVP 1989 MR CDA PvdA D66 VVD GRLNKS SGP GPV RPF CD PR CDA PvdA VVD D66 GRLNKS SGP GPV RPF CD 1994 MR D66 PvdA CDA VVD GRLNKS RPF SGP GPV CD PR PvdA CDA VVD D66 GRLNKS CD RPF SGP GPV MR: Majority Relation. PR: Proportional Representation Ranking; ranking according to the proportions of votes in the elections. [264 ] 483 Table 4 also gives the rankings of the parties as yielded by the Dutch system of proportional representation. In the next section, we will use this information for detecting the majority-plurality paradoxes mentioned in the introduction. In addition, we note that the majorities in all the four majority relations are strict. The results are surprising indeed. The majority relations are all transitive; none of them contains a cycle. So not only top cycles are absent, but any majority cycle whatsoever. What is the reason of "so much stability" (Tullock, 1981)? The research line on domain conditions as briefly discussed above may provide an answer. It is possible that there is a underlying pattern that restricts the voter preferences in such a way that the paradox cannot occur. As we have seen, the best-known condition in this respect is Black's condition of single-peaked preferences. We apply this to the Dutch situation in the concerned elections. Remember that voter preferences are single-peaked if there is a linear ordering of parties such that, in passing from one party to the next in this ordering, each voter preference shows only one top (or a plateau in the case of indifferent preferences at the top). In the four Dutch elections under scrutiny we could not find such a linear ordering. First we tried out, of course, the traditional left-right dimension in Dutch politics. For example, using data of the National Parliamentary Election Studies we arrived at the following left-right ranking of the parties for 1994: GroenLinks - PvdA - D66 - CDA - VVD - SGP - GPV - RPF - CD See also Vergunst (1995). For this linear ordering we found that only 35.1 % of the voter preferences were single-peaked. The other possible orderings of parties did not work either, not only for this case but for every case we studied. Our conclusions concerning the observed stability cased by preference regularities are preliminary. As the above results show, single-peakedness of voter preferences in Dutch elections is not a very likely cause. However, other domain restrictions like Sen's (1966) value restriction may be operational. In order to be more conclusive, much more empirical research on preference patterns and domain restrictions is needed. 5. Majority-plurality paradoxes in the four elections The main goal of this paper is to analyze Dutch elections on the occurrence of the Condorcet paradox. In order to detect the paradoxes and, as a by-product, possible majority cycles, the majority relation for each election under scrutiny was reconstructed (see Table 4). These majority relations now will be used [265 ] 484 Table 5. Comparing rankings by majority rule to rankings by the proportional system in the Netherlands (1982-1994) Election year 1982 1986 1989 1994 Kendall's tau .6410 .7576 .9444 .6667 Source: Rankings Table 4. for detecting three possible majority-plurality paradoxes as discussed in the introduction. The three paradoxes are: 1. The More-Preferred-Less-Seats Paradox, which states that a party having a majority over another party may receive less seats; 2. The Condorcet-Winner-Turns-Loser Paradox, which states that a Condorcet winner need not receive the largest number of seats and even may not receive a seat at all, and; 3. The Majority-Reversal Paradox, which states that the majority relation for an election may be the reversal of the ranking of the parties in correspondence with their number of seats as assigned by the system of proportional representation. It can be shown that each of these paradoxes may occur in the Dutch system of proportional representation (Van Deemen, 1993). However, do they occur in reality? Table 4 contains, besides the majority relations in the left part of each column (called MR), the rankings in terms of the number of seats as actually assigned to the parties by the system of proportional representation. See the right part of each column (called PR) in Table 4. A first look at this table immediately shows the differences between the two rankings MR and PR. For the elections of 1982, 1986 and 1994, the More-Preferred-Less-Seats Paradox occurs abundantly. It also occurs in 1989, but to a less extent. In that year, there was only one pair of parties with reversed positions, namely VVD and D66. To gain more insight into the coherence of both relations for each election, we calculated Kendall's tau (Table 5). Clearly, the smaller Kendall's tau, the more majority-plurality reversals can be observed. We conclude that the election of 1982 contains the most reversals. [266 ] 485 The Condorcet-Winner-Turns-Loser Paradox occurred twice. In 1994, D66 was the majority winner, but PvdA received the largest number of seats. Also CDA and VVD received more seats than D66, in spite of the fact that D66 had a majority over both parties. The second case occurred in 1982 when CDA received more seats than the majority winner PvdA. We did not observe the fact that a Condorcet winner may not receive a seat at all. The Majority-Reversal Paradox did not occur in any of the four elections. Although the rankings based on the majority rule can be quite different from the rankings based on the system of proportional representation, there is no election year in which the majority ranking is completely reversed. 6. Conclusions Voting in accordance with the majority rule is often considered as a necessary condition for real democracy (Dahl, 1989). The major problem of the majority rule, however, is the possibility of cycles that prevent the existence of a Condorcet winner. If this is the case, we speak of the Condorcet paradox. In this paper we tried to find instances of the Condorcet paradox in four national elections held in 1982, 1986, 1989 and 1994 in the Netherlands. We did not find the paradox. The majority relation for every election appeared to be fully transitive. This result is in line with previous research on cycles in large elections. For some reason or another, cycles in large elections are scarce. Now it is well-known from theoretical studies that regularities in voter preferences may prevent the occurrence of the paradoxes. Single-peaked preferences constitute the most clear example of this. However, we did not find evidence for single-peakedness in the four elections investigated by us. But of course, other preference regularities we did not investigate yet, may be operational. In order to reveal the real causes of the absence of Condorcet paradoxes, much empirical research has to be done. We also tried to detect instances of majority-plurality paradoxes. Two of the three paradoxes formulated by us actually did occur. First, we found many instances of the paradox that a party having a majority over another party received less seats than that other party. A good indication for this paradox is Kendall's tau. Especially the elections of 1982 and 1994 are characterized by a small Kendall's tau and hence by a high number of reversals. Further, we found two instances of the paradox that a Condorcet winner does not receive the largest number of seats. In the election of 1994, three parties received more seats than the Condorcet winner and in the election of 1982 only one. We did not find instances of the paradox that the majority relation over the [267 ] 486 parties is the complete reverse of the ranking of the parties as yielded by proportional representation. References Arrow, K.J. (1963). Social choice and individual values. New Haven and London: Yale University Press. Second Edition. Black, D. (1958). The theory of committees and elections. Cambridge: Cambridge University Press. Blydenburgh, I.e. (1971). The closed rule and the paradox of voting. The Journal of Politics 33: 57-71. Campbell, C. and Tullock, G. (1965). A measure of the importance of cyclical majorities. Economic Journal 75: 853-857. Chamberlin, J.R., Cohen, J.L. and Coombs, C.H. (1984). Social choice observed: Five presidential elections of the American Psychological Association. The Journal of Politics 46: 479-502. Condorcet, M. (1785). Essai sur ['application de l'analyse a la probabiliti des decisions rendues ala pluralite de voix. Paris. Condorcet, M. (1789). Sur laforme des elections. Paris. Dahl, R. (1989). Democracy and its critics. New Haven: Yale University Press. DeMeyer, F. and Plott, C. (1970). The probability of a cyclical majority. Econometrica 38: 345-354. Feld, S. and Grofman, B. (1992). Who's afraid of the Big Bad Cycle? Evidence from 36 elections. Journal of Theoretical Politics 4: 231-237. Garman, M.B. and Kamien, M.l. (1968). The paradox of voting: Probability calculations. Behavioral Science 13: 306-316. Gehrlein, WV. (1983). Condorcet's paradox. Theory and Decision 15: 161-197. Gehrlein, W and Fishburn, P. (1976). The probability of the paradox of voting: A computable solution. Journal of Economic Theory 12: 14--25. Jamison, D.T. (1975). The probability of intransitive majority rule: An empirical study. Public Choice 23: 87-94. Jones, B., Radcliff, B., Taber, e. and Timpone, R. (1995). Condorcet winners and the paradox of voting: Probability calculations for weak preferences orders. American Political Science Review 89: 137-147. Klahr, D. (1966). A computer simulation of the paradox of voting. The American Political Science Review 60: 384--390. May, R.M. (1971). Some mathematical remarks on the paradox of voting. Behavioral Science 16: 143-151. Niemi, R.G. (1970). The occurrence of the paradox of voting in university elections. Public Choice 91-100. Niemi, R.G. and Weisberg, H.F. (1968). A mathematical solution for the probability of the paradox of voting. Behavioral Science 13: 317-323. Niemi, R.G. and Wright, J.R. (1987). Voting cycles and the structure of individual preferences. Social Choice and Welfare 4: 173-183. Radcliff, B. (1994). Collective preferences in presidential elections. Electoral Studies 13: 5057. [268 ] 487 Riker, W.H. (1965). Arrow's Theorem and some examples of the paradox of voting. In J,.M. Claunch (Ed.), Mathematical applications in political science, 41-60. Dallas: Southern Methodist University Press. Riker, W.H. (1982). Liberalism against populism. San Francisco: Freeman. Sen, A. (1966). A possibility theorem on majority decisions. Econometrica 34: 491-499. Tiinnsjo, T. (1992). Populist democracy: A defence. London: Routledge. Tullock, G. (1981). Why so much stability? Public Choice 37: 189-202. Van Deemen, A.M.A. (1993). Paradoxes of voting in list systems of proportional representation. Electoral Studies 12: 234-241. Vergunst, N.P. (1995). De empirische bruikbaarheid van de stemparadox. Master thesis. Nijmegen: Katholieke Universiteit Nijmegen (In Dutch). Vergunst, N.P. (1996). Besluitvorming over kemcentrale Borssele: Een analyse van de stemparadox in de Nederlandse politiek. Acta Politica 31: 209-228. Data sources Dutch Parliamentary Election Dutch Parliamentary Election Dutch Parliamentary Election Dutch Parliamentary Election Studies Studies Studies Studies 1982. 1986. 1989. 1994. Steinmetz Archive/SWIDOC. Steinmetz Archive/SWIDOC. Steinmetz Archive/SWIDOC. Steinmetz Archive/SWIDOC. [269] >- N -..l 't:l 't:l 0 ~ = .... Q. ~ Table App. 1. Majority scores for the 1982 elections in the Netherlands: N(XPiY) Party CDA VVD PvdA D66 DS70 PPR PSP SGP CPN GPV RPF BP CP CDA VVD PvdA D66 DS70 PPR PSP SGP CPN GPV 0 351 460 320 110 232 189 44 154 47 20 21 583 513 0 364 194 116 85 130 50 129 119 62 37 600 528 480 0 140 180 144 100 117 95 88 40 35 634 596 544 498 0 242 192 106 153 108 102 44 35 675 607 551 508 251 0 115 143 79 138 125 73 45 702 633 597 581 303 276 0 161 93 152 139 75 45 727 656 615 593 306 342 249 0 194 77 83 50 43 760 690 665 637 346 348 254 195 0 191 171 103 53 743 672 636 614 329 354 272 108 209 0 76 46 45 733 659 634 599 326 356 267 123 207 RPF 486 0 473 303 105 239 191 78 156 74 73 27 18 778 715 661 648 376 384 299 196 222 171 161 0 40 788 720 671 660 397 408 321 228 226 218 194 110 0 BP CP 44 Source: Dutch Parliamentary Election Panel Study 1981-86 (n = 1206). 92 0 55 40 -I':- 00 00 Table App. 2. Majority scores for the 1986 elections in the Netherlands N(XPiY) Party CDA PvdA D66 VVD PPR PSP CPN SGP GPV EVP RPF CP CDA PvdA D66 VVD PPR PSP CPN SGP GPV EVP RPF CP 0 575 456 286 301 234 212 56 53 85 53 645 0 472 471 168 120 90 174 171 114 141 28 624 536 0 403 162 154 122 109 106 86 86 17 668 672 565 0 355 283 254 121 113 145 100 14 762 727 706 590 0 152 109 179 163 121 132 20 800 760 757 622 349 0 146 177 163 133 137 21 844 805 825 661 445 329 0 247 224 206 196 25 855 756 804 634 445 314 265 0 77 163 77 23 860 769 815 642 467 333 281 133 0 155 55 21 848 794 830 653 477 334 275 201 164 846 772 821 646 474 334 367 148 100 160 0 21 944 875 930 759 588 467 367 328 314 330 285 0 11 0 139 23 Source: Dutch Parliamentary Election Study 1986 (n = 1356). t-.) -- -..I +>- 00 1.0 490 Table App. 3. Majority scores for the 1989 elections in the Netherlands: N(xPiY) Party CDA PvdA D66 VVD GRLNKS SGP GPV RPF CD CDA PvdA D66 VVD 0 637 581 335 365 90 88 80 43 728 0 573 472 226 187 204 180 48 698 613 0 435 229 152 154 137 47 807 763 737 0 498 186 195 169 43 GRLNKS SGP GPV RPF CD 889 845 873 646 0 214 226 201 67 990 925 1002 761 646 0 151 125 61 989 931 1006 759 646 154 0 72 62 975 931 1008 762 651 181 149 0 62 1113 1074 1146 912 806 443 443 408 0 Source: Dutch Parliamentary Elections Study 1989 (n = 1506). [272] Public Choice 97: 491-523, 1998. © 1998 Kluwer Academic Publishers. 491 The spatial character of Russia's new democracy MISRA MYAGKOV 1 & PETER C. ORDESROOK2 1 Department of Political Science, University of Oregon, Eugene, OR 97403-1284, U.S.A.; 2Division of Humanities and Social Sciences, California Institute of Technology, Pasadena, CA 91125, U.S.A. Abstract. To date, virtually all research on Russian elections, beginning in 1991, have used tools and methodological approaches akin to voting research from the 1950s and 1960s. Researchers have relied either on public opinion polls that try to tease out correlations between a standard menu of socio-economic characteristics, attitudes about candidates, and self-reports of voting history; or on journalistic assessments of aggregate election returns, coupled with substantive expertise of Russian politics. Here, then, we try to gain an understanding of those elections in more contemporary theoretical terms - in terms ofthe spatial analysis of elections and voting. Although our analysis relies on a less-than-optimal source of data - election returns aggregated up to the level of individual rayons (countries) - we are able to draw a spatial map of those elections that is not too dissimilar from what others infer using less explicit methodologies. Specifically, we find that throughout the 1991-1996 period, a single issuereform - has and continues to dominate the electorate's responses to candidates and parties. On the other hand, we find little evidence of the emergence of "nationalism" as an issue, but conclude that to the extent we can detect this issue in the 1996 presidential contest, one candidate, General Alexander Lebed, did succeed in differentiating himself from other nationalist candidates (most notably, Vladimir Zhirinovski) without abandoning the reformist camp. In general, then, this preliminary analysis suggests that the same tools used elsewhere to uncover the spatial map of elections and the connection between basic and actionable issues (individual level thermometer score rankings of candidates and parties) can be applied to Russia with the promise of coherent, understandable results. 1. Introduction Russia's transition to democracy and its electoral processes in particular are presumed to be bedeviled by a multitude of problems, including voters who are not well-acquainted with democratic as opposed to Soviet-style elections, an expansive menu of candidates and parties with ill-defined platforms and policy positions, a political elite well-schooled in the mechanics of electoral fraud and other forms of corruption, an economy that encourages acceptance of radical policy proposals, and an international environment including the expansion of NATO and the loss of super-power status that feeds the flames of nationalism. Nevertheless, despite Vladimir Zhirinovski's meteoric rise [273 ] 492 in the 1993 parliamentary election, and an apparent communist resurgence accompanied by the virtual disappearance of the traditional electoral standard bearer of reform, Russia's Choice, in the 1995 parliamentary contest, the Russian electorate appears to be remarkably stable. Although Yeltsin's vote in the 1996 presidential election did not equal his landslide victory in 1991, he not only emerged as the strongest candidate in the first round of balloting, but he also largely succeeded in the second round in pulling together all proreform voters to his side while simultaneously securing a disproportionate and critical share of votes from Lebed's first round supporters (Myagkov et aI., 1997). This essay explores this apparent stability further by looking at what we can infer about the structure of issues and candidate positions from the approximately 2,OOO-per-election rayon (county) level election returns. What we want to reevaluate in particular is the electoral "map" of Russia drawn by political commentators who have relied thus far on their general understanding of politics and candidates and on vote totals aggregated typically at the regional and even national level. That map offers a picture of an electorate first dominated by a single issue (attitudes toward reform) and that, with the emergence of Zhirinovsky as a national figure in December 1993, saw the addition of nationalism as a second relevant dimension. Subsequent elections (the 1995 parliamentary and 1996 presidential elections) are assumed to have reenforced this two-dimensional picture, with Zhirinovsky and a handful of minor candidates and parties staking out nationalism as "their" issue, Yeltsin, the communists led by Zyuganov, various splinter anti-reform parties, and a fractured coterie of reformers jostling on a "reform" issue that seemed mostly to be a referendum on Yeltsin's administration, and General Alexander Lebed offering a compromise (centrist?) alternative on both issues. Indeed, Lebed's emergence in 1995-96 as a potential successor to Yeltsin was attributed largely to his apparent ability to position himself on these issues so as to secure a share of both pro-reform and nationalist voters in such a way as to allow him to throw his support in one direction or the other between the two rounds of presidential balloting.! While public opinion surveys give credence to this electoral map (see, e.g., Wyman et al., 1995; and Hough et al., 1996), its basic character derives largely from inductive attempts to make sense of the ebb and flow of official election returns from one election to another, aggregated up to the national or regional levels, in the context of some strong priors as to the general orientation of parties and candidates (see, e.g., McFaul, 1993; McFaul and Fish, 1996; Boxer et aI., 1996; and Fish, 1995). There is, of course, much to be learned from informed interpretations of events. But such interpretations, when based on highly aggregated data, are [274 ] 493 subject to the traditional statistical degrees-of-freedom problem and to the informational limitations of the data analyzed. Thus, they cannot uncover potential nuances that can profoundly affect the future course of events such as the viability of Lebed's presumed electoral strategy in the event that only he and some communist candidate compete in the nest presidential election. Here, then, we reassess the extent to which a simple one- or two-dimensional portrayal of the Russian electorate is appropriate using a less aggregated data set and a methodology (factor analysis) whose limitations are more readily deciphered than journalistic inference. Briefly, we find that although a shift from a uni- to a two-dimensional structure characterizes the period 19911993, both the 1995 and 1996 elections give evidence of the emergence of a more complex spatial structure - a structure that is not inconsistent with the view that Russians, like voters elsewhere, not only judge on the basis of ideological predispositions, but also on the basis of specific characteristics of candidates (Enelow and Hinich, 1984; Hinich and Munger, 1994). We support the general explanation for Lebed's ability to emerge as the third largest vote-getter in 1996, but here at least we conclude that the extent of his "compromise" is less than has been otherwise assumed. Perhaps as a portent of the future, we argue that the compromise he achieved was no so great as to alienate those who might support reform in the next presidential election. Finally, we also see evidence of the emergence in 1995 and 1996 of pro-reform voters who nevertheless refuse to vote for Yeltsin or those associated with his administration, except when their choice is between that administration and a return to communist rule. It is these voters, in combination with Lebed's overall strategy, that has thus far lent Lebed the aura in today's mass media as "the man to beat" in the year 2000. 2. Methods and data A spatial (Euclidean) construction in which issues are represented as segments of real lines, candidate platforms as points on these lines, and voters as rational decision makers with well-defined issue preferences who vote on the basis of their distance from candidate positions, is by now a wellaccepted representation of elections and a base paradigm for formal mathematical analyses of elections and candidate strategies (Ordeshook, 1997). Correspondingly, statistically estimating the parameters of this representation - the dimensionality of the issue space, the relative salience of issues, candidate positions, and distributions of voter preferences - using individual level polling data in the form of responses to requests for cardinal (thermometer score) or ordinal rankings of candidates has also advanced considerably (see, e.g., Cahoon et al., 1978; Dow, 1997; Lin et al., 1996). This methodolog- [275 ] 494 ical literature, though, reveals the difficulties associated with estimating a complex nexus of parameters from even rich data: not only does the mathematical complexity of the underlying structural models make it difficult to establish the statistical properties of estimates, but indeterminacies quickly appear when estimating the requisite parameters. Although these difficulties can be overcome with appropriate data drawn from polls of individual voters and a few structural assumptions, they appear to be insurmountable when the only numbers at our disposal are aggregate election returns. For example, although it is tempting to conclude that if candidate X's share of the vote exceeds candidate Y's share, then X must on average rank higher than Y on the preference scales of voters, such an inference is invalid unless a variety of restrictive assumptions about spatial structure are satisfied. Candidate Y, may be sandwiched between two other, more viable opponents so that despite aggregate returns, Y ranks higher than X for a majority of the electorate. The fact is that aggregate data tells us, in the absence of strategic voting, only the first-ranked alternative in each voter's preference scale, whereas, as the methods of multidimensional scaling reveal, reliable inferences about the parameters that concern us require data on each voter's overall rank-ordering of the candidates and perhaps even some cardinal measure of that ranking. This is not to say that inferences from aggregate data are impossible or injustifiable. The analysis of such data can often provide useful guidance to further research, provided we appreciate fully their informational limitations and the assumptions that must be imposed in order to proceed. To illustrate, suppose a single issue dominates campaign discourse, suppose the electorate's preferences on that issue are normally distributed, and suppose the candidates' (or parties') policy positions are uniformly dispersed so that each occupies a unique position and each is approximately equidistant from the candidate to his or her right and left. If all voters now vote for candidates nearest their preferred positions, then the candidates' vote totals should decrease monotonically with their distance from the electorate's median preference. Notice also that as the overall distribution of preferences shifts from one side to another, the votes for candidates on the same side of the issue space will tend to rise and fall together whereas the votes for candidates on opposite sides will tend to correlate negatively. Thus, if we have aggregate data from a cross-section of election districts, each of which satisfies these assumptions, and if the mean preferences across districts are themselves distributed so as to give some "meaningful" variance in election returns, then we can use a simple factor analysis of aggregate returns to estimate the issue space (i.e., recover its unidimensionality) and the candidates' relative positions on it. [276 ] 495 Unfortunately, these assumptions (or their multidimensional extension) need not be satisfied in any specific election. First, electoral rules such as Russia's requirement that parliamentary seats can be won under proportional representation only by those parties whose vote exceeds 5% of the total can induce voters to act strategically - to vote for a competitive candidate or party rather than the alternative closest to their ideal. Strategic voting can also characterize Russia's presidential contest, since there, through the use of majority rule with a runoff, voters might be concerned that a sincere vote will result in a second-round choice between two wholly unacceptable alternatives. Second, absent any well-defined equilibrium and an evident clustering of splinter parties at various issue positions, candidate positions need not be uniformly distributed across the issue: a candidate who performs poorly because he is squeezed out by adjacent opponents may actually be close to the overall median preference. Third, nonvoting might correlate with policy preferences so as to distort any statistical estimates: and in Russia in particular there is evidence that turnout rates declined most rapidly among those who supported Yeltsin in 1991 (Myagkov et aI., 1997). Fourth, preference distributions from one election district to another might not be sufficiently varied, as when districts are either predominantly "pro" or "anti" reform, in which case vote shares across regions will correlate in such a way as to produce a spatial map in which candidates tend to be located at one of only two positions. Finally, any monotonicity in the relationship between vote share and distance from an electoral median can vanish if preferences are not unimodally distributed - something we cannot wholly discount for a country such as Russia in which voters can be reasonably assumed to be polarized between pro- and anti-reform policies. Despite these limitations and indeterminacies, commentators and politicians find it hard or impossible to resist drawing inferences about the geography of an election using such data - to infer salient issues, voter perceptions of the candidate's positions on those issues, and shifts from one election to the next in issue salience, voter preferences, and candidate positioning. This fact and the fact that the requisite individual level polling data remain unavailable or unreliable for countries like Russia compels us to examine more closely the things we can infer from aggregate data, including a factor analytic treatment of that data. After all, even though such an analysis proceeds on tenuous theoretical grounds, our advantage here is that our assumptions can be made wholly transparent. Thus, we can suggest alternative hypotheses for the patterns we find in the aggregate data that can be explored subsequently in future elections with more refined data. So even though we cannot offer definitive conclusions, we can give direction to the gathering of more precise individual-level data. Briefly, the data we use consist of official [277 ] 496 rayon-level election returns for all Russian elections beginning with 1991 through the 1996 presidential election, including the April 1993 referendum that amounted to a vote of confidence on Yeltsin's administration in his conflict with the increasingly recalcitrant Congress of People's Deputies. The sole exception to the comprehensiveness of our data is the December 1993 parliamentary contest. There, since Russia's Central Election Commission has, suspiciously, never published official election returns except at the regional level, we rely on unofficial returns covering approximately half the rayons and which appear to represent a relatively unbiased cross-section of the country (for additional discussion of this data see Myagkov and Sobyanin, 1995). Thus, with the exception of 1993, we have approximately 2,000 observations for each election, where the specific elections considered are these: the 1991 presidential contest held before the dissolution of the Soviet Union in which Boris Yeltsin first assumed the office of the presidency of the Russian Republic, the April 1993 referendum in which voters were asked to answer four questions that directly or indirectly amounted to a vote of confidence in Yeltsin's reform efforts, the December 1993 parliamentary election in which the neo-Nazi Vladimir Zhirinovsky out-polled all other parties with 23% of the vote but in which Yeltsin succeeded in securing majority approval for his constitution for the now-sovereign Russian Federation. the December 1995 parliamentary contest that saw a resurgence in communist party support, the virtual disappearance of the old standard bearer of reform, Russia's Choice, and the first electoral appearance of General Alexander Lebed, the 1996 June-July presidential contest in which Yeltsin won reelection, but only after being required to confront his communist challenger, Gennady Zyuganov in the second round of balloting after forming an explicit coalition with Lebed and an implicit and grudging one with another reform candidate, Gregory Yavlinski. In the next section, 3, we briefly review the results of a simple factor analysis of the returns for each of these elections in order to see whether our results correspond to the electoral maps drawn by commentators on those elections. Here we see evidence of increasing dimensionality in the issue space from one election to the next (excepting the April 1993 referendum) although the issue of reform remains predominant throughout the period. The issue [278] 497 of nationalism, on the other hand, barely registers on our radar screen, and then only in 1996. In Section 4 we divide our data into three categories - rayons that strongly support reform (or Yeltsin), those that support antireform candidates or parties, and "others". What we infer from this analysis is that although the Russian electorate appears to be remarkably stable and consistent in terms of the dimensionality of the issue space and in terms of the spatial positioning of candidates, there is interesting evidence that voters in June 1996 (the first round of balloting in the presidential contest) voted strategically for the primary contenders and that Yeltsin's strategy - allowing Lebed full reign to campaign in that round in the expectation that he would draw votes from other nationalist candidates but that these votes would be won over by Yeltsin in the second round - largely succeeded. Section 5 concludes with a variety of cautionary notes about the inferences we can draw from the aggregate data. 3. Trends from 1991 to 1996 3.1. The 1991 presidential contest Beginning with the 1991 presidential election in which Yeltsin won with 57.3% of the votes and in which his nearest rivals, the communist Ryzhkov and the neo-Nazi Zhirinovski, won only 16.9% and 7.8% respectively, Table 1 reports the results of a factor analysis of the data that considers these three candidates, two minor contestants, and the category "against all", while Figure 1 offers a spatial portrayal of their estimated positions. Table 1. 1991 presidential election Factor Eigenvalue % Variance Cumulative % "explained" 2.71 45.2 19.4 45.1 64.6 2 1.16 3 4 .77 12.9 77.4 .67 11.1 88.6 5 .49 8.1 96.7 The picture drawn by Figure 1 should come as no surprise to those with even a passing familiarity with the 1991 election. As that figure shows, the 1991 contest was largely a competition between Yeltsin and the rest of the [279] 498 1.0 0.8 0.6 0.4 =~:+ ...VELJSiN..............+....................................;.................................../.............................Q........~L .......................... C\I g 0.2 () <II u.. 0.0 Z1RINOVS o -0.2 -0.4 -0.6 -1.0 -0.6 -0.2 0.2 0.6 1.0 Factor 1 Figure 1. 1991 Presidential election, national data. field, where the first and dominant dimension corresponds closely to the issue of pro- versus anti-reform (or at least to a referendum on Yeltsin's "stick-itin-your-eye" stance with respect to anyone associated with the old regime). Rizhkov and Makashov were clear stand-ins for the old communist regime, Bakatin, although not an opponent of reform, was nevertheless likely to be associated with the status quo owing to his position as Gorbachev's Interior Minister, and Zhirinovski was by then a vocal opponent of nearly everything and anything. Admittedly, Table 1 presents us with the usual dilemma in determining the dimensionality of the issue space. There is a rapid drop in percent variance explained after the first factor, but thereafter it is a matter of taste whether, in reporting these results, we should pay heed to factors 2, 3 or 4. In this regard it is worth noting that with only six candidates, including the Against All ballots, we will necessarily see an appreciable increase in "variance explained" as additional factors are considered. Thus, it is best to postpone the interpretation of this table until we examine additional elections, except to say that the first dimension takes clear precedence over all others in that it alone accounts for nearly half the variance in election returns. [280] 499 3.2. The April 1993 referendum As part of the increasingly nasty conflict between Yeltsin and Russia's parliament, the Congress of People's Deputies, Russians were asked in Aprill993 to answer four questions: Do you have confidence in President Boris Yeltsin? Do you approve of the socioeconomic policy carried out by the president of the Russian Federation and the government of the Russian Federation since 1992? Do you consider it necessary to carry out early elections for the president of the Russian Federation? Do you consider it necessary to carry out early elections for the deputies of the Russian Federation? Clearly, the pro-Yeltsin position on these questions was to vote Yes-YesNo-Yes, and it seemed clear that the only issue under consideration in this vote would be people's attitudes toward reform. This presumption is largely verified in Figure 2 and Table 2, where the positioning of the "Yes" response to the third question is diametrically opposite that of "Yes" responses to the other three questions and where the first factor alone accounts for 73% of the variance. There is, nevertheless, some residual variance explained by a second factor that differentiates questions three and four from the first two, thereby reflecting the fact that although a voter might support Yeltsin and his policies in general, he or she might also prefer new elections for all national offices. Table 2. 1993 April referendum Factor Eigenvalue % Variance Cumulative % "explained" 2 2.20 73.4 73.4 .22 7.4 80.8 [281] 500 0.45 , . . . . - - - - . . . - - - - - , - - - - - , . . - - - - - - - , , - - - - - - , , - - - - - - , , , 0.35 ...... , ·y~I t\I j ~05- ~-!I+lr -0.15 L -_ _ _' - -_ _ _' - -_ _ _' - -_ _ _' - -_ _ _' - -_ _---.J -1.2 -0.8 -0.4 0.0 0.4 0.8 1.2 Factor 1 Figure 2. 1993 April referendum. 3.3. The 1993 parliamentary election Following Yeltsin's artillery assault on and termination of the Congress of People's Deputies, Russia's first party-based election as a sovereign state occurred in December 1993. Thirteen parties vied for the 225 seats allocated by national party list proportional representation, and eight succeeded in surpassing the 5% threshold for representation. 2 Judging by their platforms and the public utterances of those who headed the list, the most visible pro-reform party lists were the ones headed by Yeltsin's ex-Prime Minister and economic reform guru, Yegor Gaidar, the explicitly pro-Western economist Yavlinski, and Yeltsin's legal advisor, Sergi Shakrai (for an ordering of the parties and candidates based on their public utterances and reputations see, e.g., McFaul and Fish, 1996). The centrist or moderate pro-reform positions were presumed to included Travkin's list and Women of Russia. The Communist Party (CPRF) and their fellow-traveling Agrarians (APR) anchored the opposition, while Zhirinovski's Liberal Democrats (LDPR) sought to claim nationalism as its issue along with ambiguous utterances on reform. Table 3 and Figure 3 summarize the results of our analysis, which in this instance, is limited to unofficial data and which excludes the important heavily pro-reform regions of Moscow and St. Petersburg. . As with 1991 and April 1993, we see again that the first factor cleanly separates pro- from anti-reform lists ("Small" in Figure 3 refers to the four small- [282] 501 Table 3. 1993 parliamentary election Eigenvalue % Variance "explained" Cumulative % 3.93 43.7 43.7 2 1.35 15.0 58.8 3 4 1.05 11.7 70.5 .83 .67 9.3 79.8 7.4 87.2 Factor 5 0.6 TRAVKIN 0 C';,RF YAVLINSK 0 0.4 GAlpAR 0.2 SM~h 0.0 C\I ~ 01 LL LDPR 0 -0.2 .......... ................. ~ APR 0 FORC93 0 SH, iKHRAI -0.4 ......... !O R""wr".."", -0.6 -0.8 -1.0 0 -0.6 -0.2 0.2 0.6 1.0 Factor 1 Figure 3. 1993 Duma election. est parties who failed to win any seats) and confirms to some extent the commonly perceived ideological orientation of the parties. Specifically, the ordering "Agrarian (APR), Communist (CPRF), Zhirinovski (LDPR), Women of Russia (Ruswomen), Travkin, Shakrai, Yavlinksi, and Gaidar" is probably not much different, if at all from, what most commentators would offer as the ideological ordering of the parties. The position of Zhirinovski's LDPR, though, is interesting in that we do not see here any evidence of "nationalism" as an issue that differentiates the LDPR from the rest of the field. Instead, the LDPR appears as a centrist party on both the first and the second dimensions. We cannot, then, reject the argument that Zhirinovski's success derived more [283 ] 502 from his relative positioning on the issues than from his presumed ability to introduce new issues into the electoral debate. That is, the LDPR's firstplace finish in 1993 appears to derive more from Zhirinovski's abilities as a campaigner - an ability that most commentators believed surpassed that of his rivals, especially Yegor Gaidar, who believed that arguing for "macroeconomic stability" was a viable campaign platform with which to attract people "experiencing" his reforms. Insofar as the dynamics of dimensionality are concerned, we should keep in mind that we are treating four more "candidates" here than in 1991, which should require additional factors to achieve an equivalent level of significance. Correspondingly, Table 3 shows that it is increasingly difficult to assert a unidimensional or even a 2-dimensional issue space. Although the first factor, as above, clearly dominates the rest, the cumulative percent variance explained increases a bit more slowly in 1993 than 1991, and after the first factor, there is no clear cutoff point with which to argue that two dimensions are "better" than three or that three is "better" than four. Notice though that although this contrast with 1991 may be primarily an artifact of the number of candidates considered, this same methodological problem would also be likely to characterize journalistic assessments of election returns, especially ones that are dependent on even a higher level of aggregation of the data. This fact might account for the ease with which events in 1991 were interpreted as unidimensional, while in 1993 - especially when explanations were offered for Zhirinovski's success - the search commenced for additional issues, including that of nationalism. These methodological issues, though, should not detract from the conclusion that attitudes toward reform remained the primary basis for classifying parties. 3.4. The 1995 parliamentary election The picture for the 1995 election to the State Duma portrays a contest in which one dimension is no longer adequate to represent matters, but in which the issue of reform remains, as in 1993, the primary basis for structuring perceptions of candidates. Here, because of the great number of party lists (a total of 43 were listed on the ballot and only four, with a fraction less than half of the vote, surpassed the 5% threshold for representation 3 ), to compare our results with previous elections, we offer two sets of eigenvalues and two sets of spatial representations - one that considers only 7 parties, and the second which considers 9 plus the "party of nonvoters". 4 First, looking at Table 4, we see that with only seven parties, three factors account as before for over 70% of the variance. However, the first factor accounts for only 36.2% as against 45.2% in 1991 (when we consider 6 candidates) and 43.7% in 1993 (when we consider 10 party lists).5 Moreover, when we add three parties - two [284 ] 503 party lists plus nonvoters - so as to make our analysis here comparable to our analysis of the 1993 parliamentary contest, the importance of the first factor declines to 30.7% and it becomes virtually impossible to choose between a 3,4 or greater-dimensional representation - the decline in variance explained from one factor to the next exhibits no discernable step. Clearly, then, the dimensionality of the issue space in 1995 exceeds that for 1993 and 1991. Table 4. 1995 parliamentary election Eigenvalue % Variance "explained" Cumulative % Factor 1 2.53 36.2 36.2 Factor 2 1.37 19.5 55.7 Factor 3 1.04 15.0 70.7 Factor 4 .81 11.5 82.2 Factor 5 .58 8.3 90.5 Factor 1 3.06 30.7 30.7 Factor 2 1.47 14.7 45.3 Factor 3 1.17 11.7 57.0 Factor 4 .90 9.0 66.0 Factor 5 .87 8.7 74.7 Factor 6 .81 8.1 82.8 7-candidate recovery lO-candidate recovery Turning now to the graphical representation of these factors, note first from Figures 4a,b and 5a,b that the recovery of spatial positions on the first three dimensions is largely insensitive to the number of candidates considered (a similar picture obtains with a 5-candidate recovery). Insofar as the substantive interpretation of these dimensions is concerned, we see that once again, the first dimension differentiates between pro-reform and anti-reform parties, with Yeltsin's stand-in, Chernomyrdin's Our Home Is Russia (OHR), being the most centrally located "reform" party. And as with 1993, the positioning of candidates on the second dimension again fails to reveal nationalism as a salient issue. Although Zhirinovski and Lebed after him occupy one end of that dimension, Chernomyrdin occupies the other extreme. Indeed, Chernomyrdin's position here suggests that this second dimension has less to do with nationalism per se, but instead corresponds to something more like "the party in power versus everyone else". We suspect, in fact, that to the extent [285 ] 504 1.0 ....------,----,----r---...,----..,-----r---,----o----, OHR ° i ! .................. ;....................;.. ·············1······················j·· 0.8 . ..... \......................1.. .................. ;......................i......................~. 0.4 (\I ~ ~. ....... 1...................... 1. ................... 0.6 0.2 .....................: ....................:.................... :.................... :..... . . ..... :..... ............... j..•... ,............... ~ ..................... j j ....... "....................,............... ····+···········AMPIl-oV·············· . ~ 0.0 ···········G:4:iDAR·········j··············!·· ·············i··········· ..... ! io -0.2 YIl'v~NL ......... ~ .... °i ..... j.................. .~ .................... . ~gr . . . . . . ; . . . . + ......! .;+~RF -0.4 ..................... :...................... :........... ·····... ·lj.~- ... , .... :..... ....................... .~ ..................... uj)PR -0.6 ...... ····· ........ ··;·······....··· .... ·· ..;..·.. ·· ....·· ...... ···i.. .................... i..·.···· .....·....·.··i.......····· .....·)l.···················t··· -0.8 L-_--'_ _--'-_ _-'-_ _-'-_ _-'--_--'_ _--'-_ _-'-_ _.J 0.4 0.8 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.6 Factor 1 Figure 4a. December 1995 Duma election. that this second dimension serves essentially as a way to differentiate the current regime, represented by Chernomyrdin, it is much like a valence issue in spatial theory (Enelow and Hinich, 1984) - something like "either you believe that Chernomyrdin is the least risky alternative, or you prefer change, left or right". However, whatever interpretation we give this second dimension, it seems clear that a simple left-right conceptualization of political competition became less relevant in 1995 than it had been in previous elections. Figures 4a and 5a are interesting in two other respects. First, note that nonvoters in Figure 4a fall on the reform side of the first factor, which is to say that those who voted in 1996 but not in 1995 were predominantly proreform voters. Thus, if we combine this fact with the finding that the general decline in turnout between 1991 and 1993 came largely at the expense of pro-reform candidates (M yagkov et al., 1997), we see some of the stability of the Russian electorate. That is, it appears as though a significant share of the shifting fortunes of Yeltsin and his allies between 1991 and 1996 derived from decreased turnout among pro-reform voters as opposed to fundamental shifts in preference, and that when those voters returned to the polls in 1996, they reappeared once again as Yeltsin voters. Thus, Yeltsin's strategy in 1996 of encouraging turnout appears to have been more than mere civic-mindedness. The second interesting thing to notice about Figures 4a and 5a is Lebed's relative positioning. Generally identified in the Moscow media as a sane alternative to Zhirinovski's nationalism, Lebed in fact appears to belong to [286 ] 505 0.6 l. . . . . . . . . . lf>. . . . . .·Or. ·. . . . . . . . L$PR 0.4 . . . . . . . . . . r. . . . . . ·. . . ·!· .................. 0.2 ...... i .... ·YAV~NK !.................... ......................'.. ·........ ~· tC~ ·I"....·............T................ l .. l AMPI~OV r--....·............·T. . . . . . . . . . ·. . . . . . . . . . . ·. . . . . . . . . . ·r. . . . . . . . . 0.0 ·····················!······················r·····················!······················1······················!······················r·····················r·····················r···················· . . . .~IDAR ·r. . . . . . . . :i ·. . . . . . . . ;. . . . . . . . . . . !'......................!....................·T:, ....................-r. . . . . . . . . . r. . . . . . . t') ~ LL -0.4 ··....······.. ·······;······················1······ ..............,............ -0.6 -0.8 l... i ,0 -0.2 ' i ········1················· ...:......................,......................1".................... ]"................... -T~;=r: -1.0 ~-' -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 Factor 1 Figure 4b. December 1995 Duma election. 1.0 1 0.8 0.6 0.4 C\I ~ LL 0.2 : ....................·!..........· ........··i.... ·, ................!..............~ 1 ......................:......................:......................:...................... t .................... --i-,.!---t---,---l-I----l-f-~ ·. :. :·.rt~J 0.0 ...... ·.FA~'R 1 : ~ r-· ........·...... ...... : YAVLINSK, 1 . . . . .:·.·.::::"·:::::J..:::::::.::::"::::J.:.:·::..:.·r~,;f"'j , . !...... ...... -·i,~. : 1 1 : 1 .................--... ~"., ....................... i , , , i,,:::. . . . . . . . . . . . . . . . . . . . . , -0.2 .................. o .. ·.................. ·!......................!......................!......................!......................!...................... !....................1"................... -0.4 . . . . . . . . . . .,. . . . . .~l . . . . . . . . . .!......................i"....................,......................l...................~ . . . . . . . . . . ,. . . . . . . . . . . :. . . . . . . . . . . (. . . . . . . . . ,. . . . . . . . . . ·r. . . . . . . . . ·r. . . . . . . . . , -0.6 -0.8 " .............. +. . . . . . . . . . T. . . . . . . . ·. LOP-R' 9.................... 1--_ _' - -_ _' - - _ - - - '_ _---'_ _---'_ _---'-_ _---'-_ _---'-_ _- ' -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 Factor 1 Figure 5a. December 1995 Duma election. [287 ] 506 , 0.8 0.6 0.4 0.2 :=Ifl-i1"~t CO) j 0.0 -0.2 -0.4 -0.6 -0.8 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 Factor 1 Figure 5b. December 1995 Duma election. the camp of refonners on the first dimension, whereas his distance from Our Home Is Russia on the second dimension suggests his success at establishing himself as an outsider to Moscow's politics or at least as distancing himself from Yeltsin's administration. 3.5. The 1996 presidential election Finally we come to June 1996, the first round of balloting in Russia's first presidential election since the dissolution of the USSR. Focusing again on the primary contenders - the five candidates who account for approximately 95% of the ballots (Yeltsin 35.3%, Zyuganov 32%, Lebed 14.5%, Yavlinski 7.3% and Zhirinovski 5.7%) - Table 5 reports the eigenvalues for the first three factors, while Figures 6a,b given the associated spatial positions. The first thing to notice is that the dimensionality here is approximately the same as what we estimate for 1991 when considering 6 candidates. Thus, whatever evidence we thought we saw through 1995 of an increase in dimensionality appears to have reversed itself somewhat in 1996. 6 More interesting, though, is the spatial configurations portrayed in Figures 6a,b. Here, again, we see Zyuganov to one side of the issue space and Yeltsin, Yavlinski and Lebed on the other. The second dimension, though, more clearly than in 1995, appears to correspond to nationalism with Zhirinovski and Lebed anchoring one end of this dimension and Yeltsin the other. Arguably, though, this second dimension, given Zyuganov's position, might be interpreted as [288 ] 507 0.8 , , , 0, ............................................j................................. +................................. j.......................¥Q;l'$w .................... 0.6 0.4 ..·· .......................... t ......................·........ ..!................................·+ .................................,.............................................. 0.2 ................................ ":'...............................i·································t····............................. j..................................}................................ ZVUGANOV , C\I 5 () If , , , t. . . . . . . . . . . . . . . ·l.··································}·····........................... i i ...............................+.................................!. . . . . . . . . . . . . . . +...............................!. . . . . . . . . Y.~N!L ........................ . 0.0 .................9. ............ ~ ..............................·!................................ t -0.2 : j ·········· . ·t·· -0.4 -0.6 ............................................. ~ : : : : L... .._...._..._.... ..._...._..._ ...._... -1.2 : · _ ···············1·.. ···· ·~ ZlRINOVS -0.4 ·~ ; ·-i....i .. ·.. _·.... __.. ...._..·.... __...._.. 0.0 1 ·······1·································l············.................... ................. .LEBED ....................1......... ..........,.................. ........... ....i .. -_ ...._._...._.. ·_ ...._..·_ ...._...._.. ·.i.·i.._.. ...._.. ·_ ...._...._..·_·I?_ ..·_ .. -0.8 . ···t························ 0 ; ' , ~ ·....i·l_ .. ·_ ...._...._..·_ ...._..·_ ...._..·_ .....i. ...._......._..._ ...._..._ ...._.. ---I 0.4 0.8 1.2 Factor 1 Figure 6a. June 1996 presidential election. 0.6r-~, , lEBED 0.4 i ........· .................. ·.. f................................ ·:..............·................ t ...............................1" ............................+ .............. 0.2 .......Z'tU~NOV 0 .. , ·t. . . . . . . . ·............!. . . . ·. . . . . . . . . . . ·t. . ·. . . . . . . . . . . ·. . ·!.................. ~·l : 0 ..............· i j ; i i 0.0 ................................ y..................... ············!·································t·································i·································f································ L~ -0.2 . -0.4 .................. ........·+-..............................·i.. ·............................ . , , 0, , , ..~ ............................. ··j· ....··· ....·· ......··VeiTtiij.. .. . ................................ '\" ................................. ! .......... . ·t. . . . . . . . . . . . . . . . : 0, I. . . . . . . . . . . . . . . · ~ 1 : j.................................. -0.6 ................................ +........· ...................... ·, ................l.\RINO.Vi ................................. j...............................,) ................................ !----..!. ----"-!-----' -0.8 ' - - - - - - ' - - - - - - - " - - - -...... -1.2 -0.8 -0.4 0.0 0.4 0.8 1.2 Factor 1 Figure 6b. June 1996 presidential election. [289] 508 Table 5. 1996 presidential election, first round Factor Eigenvalue % Variance Cumulative % "explained" 1 2.29 45.8 45.8 2 1.33 26.7 72.6 3 .77 15.5 88.1 "old faces versus new faces" and, absent other evidence, should not then be taken as evidence that Russian's are increasingly sympathetic to allowing their sense of nationalism to overtake their attitudes toward reform as the primary basis for evaluating candidates. The third factor, moreover, primarily differentiates between Zhirinovski and Lebed, and suggests that even if the second dimension captures some part of the issue of nationalism, Russian voters can differentiate a sane alternative (ostensibly Lebed) and one who, although admittedly a skilled campaigner, is willing arguably to make any utterance to gain media attention. Equally interesting is the inference we can draw from this figure as to Yeltsin's strategy for defeating Zyuganov in the second round. If we take Figure 6a and, after scaling to reflect relative eigenvalues, draw the bisecting lines so as to map out the areas of support for each candidate, we would find that once Zhirinovski, Lebed, and Yavlinski are eliminated, Zyuganov wins most of Zhirinovski's voters, while Yeltsin gets all of Yavlinski's and Lebed's. Making an explicit appeal for Zhirinovski's vote by promising him a position in his government was, of course, out of the question for Yeltsin, and of the two other possibilities, the most uncertain block is Lebed's even though, ceteris paribus, they should vote for Yeltsin in the second round. 7 But an interim coalition between Yeltsin and Lebed, as events revealed, came at no cost to Yeltsin (indeed, he arguably gained not only votes, but a cease fire in Chechnya whose unraveling could be blamed on Lebed) and rendered the support of Lebed's voter more secure. 4. The "anomaly" of the pro-reform regions Aside from the substantive importance of the contest, Russia's 1996 presidential election is interesting because of its strategic complexity, the uncertain response of voters to the war in Chechnya, and the fact that "reform" had brought something other than beneficial results to an increasingly large part [290] 509 of the population. First, with respect to strategic complexity, we know that majority rule with a runoff gives candidates ample incentive to try to position themselves so that, if they fail to progress to the second round, they can use their support as currency in negotiations with one or both of the two leading candidates. Lebed in particular understood this strategic environment and, with an eye to both the current contest and to the next presidential election, appeared to have successfully positioned himself to become pivotal in 1996 and Yeltsin's heir apparent in 2000. The strategic environment for voters is no less interesting since for many of them the choice in the first round is between a sincere vote for a most-preferred candidate versus a strategic vote for the least objectionable viable one. Pro-reform and anti-reform voters, though, confronted different environments in 1996. The clear leader of the anti-reform camp was Zyuganov, who was virtually certain to be on the ballot in the event of a second round. Indeed, until Yeltsin's resurgence in the polls, the question here was whether Zyuganov could win an outright victory on the first ballot, and, if not, how great his margin of victory would be over a candidate such as Zhirinovski. Thus, the average voter in a rayon in which most voters are opposed to Yeltsin or any reform candidate has the lUXury of not having to cast a strategic ballot. They can vote straightforwardly for Zyuganov, or, if nationalism is most salient, for Lebed. In contrast, pro-reform voters needed to decide which "reform" (i.e., anti-communist) candidate - Yeltsin, Yavlinski, or Lebed - had the best chance of defeating Zyuganov. And although Yeltsin, as the product of a skilled campaign staff and not a little "maneuvering" (e.g., briefcases filled with non-consecutively marked $100 bills) emerged in the end as the front runner, that position may have been hidden from view for all but those voters with access to and a taste for weekly public opinion polls. Thus, it is reasonable to speculate that one concern of voters who ranked Yeltsin below any reform candidate but above Zyuganov or Zhirinovski is that absent strategic voting on their part, the second round of voting would present them with the Hobson's choice of a communist, Zyuganov, versus a neo-Nazi, Zhirinovski. Rendering 1996 more complex still is the fact that the definitions of "reform" and "anti-reform" also arguably blurred. Yeltsin, the clear leader of the pro-reform camp in 1991 and 1993, could readily be seen to have eroded his position owing to the general increase in corruption within his administration, policies that benefitted only a handful of "New Russians", and the unsuccessful yet bloody pursuit of a military victory in the secessionist republic of Chechnya. And although Yavlinski could unambiguously be classified as "pro-reform", and Zyuganov as "anti-reform",8 it remained unclear even as voters entered the voting booths how they would perceive Yeltsin and [291] 510 how their perceptions would interact with the electoral system's strategic imperatives. Finally, there is the argument that by 1996 at least, the Russian electorate had become wholly polarized between reform and anti-reform positions, reflecting the attitudes of those who had by then benefitted directly or indirectly from privatization versus those who continued to wait for pension checks that would not even pay for a below-subsistence grocery list. Yeltsin's support, however defined, had been in continual decline since 1991, whereas Zyuganov and the old guard appeared to have been able to maintain all of their original support while taking full advantage of Zhirinovski's diminished standing. If anyone occupied "the middle" is was Lebed, but few gave him much of a chance of entering the second round of balloting by countering Zyuganov's communist party organization or Yeltsin's influence over the regional authorities who would administer the election or the bankers who might fund a campaign. With these ideas in mind, let us now consider June 1996, but rather than look at the national sample taken as a whole, we first divide regions into three categories: "strong pro-Yeltsin" (n = 510), "strong anti-Yeltsin" (n = 510), and "moderate" (n '" 1000) so that we can try to get a sense of how, on average, voters of different types responded to the candidates. 9 Table 6, then, begins by giving the eigenvalues and variance "explained" for the first three factors for each of these subpopulations and reveals that although rayons from "moderate" and "anti-Yeltsin" regions give approximately the same results in terms of the relative importance of the first and second factors, with the first factor explaining two or three times the variance of the second, the rayons from strongly pro-Yeltsin regions produce two factors that are nearly equal in terms of variance explained. This, of course, is the first instance in our analysis in which the first factor does not wholly outweigh the second. The spatial portrayal of these factors and the candidates' positions on them, given in Figures 7a,b, 8a,b, and 9a,b, is no less interesting. Specifically, comparing Figures 8a and 9a notice that the relative positions of the five candidates considered are approximately equivalent and not much different from Figure 6a when we considered the national data set as a whole. The third factor in Figures 8b and 9b primarily differentiates Lebed from the other candidates, and, in particular, distinguishes between the two "nationalist" candidates - Lebed and Zhirinovski. Overall, then, it is apparent from these figures that, as in our analysis of the national sample, Lebed is something more than a refined version of Zhirinovski. His vote came predominantly from the "moderate" and "pro-reform" parts of the popUlation, and, as such, was a natural coalition partner for Ye1tsin going into the second round. [292] 511 1.4 1.0 . :. . . . . . . . . . .:. . . . . . . . . . . !. . . . . . . . . . +. . . . . . . . . . j.....................+. . . . . . . . . . . . . . . . . . . . :. . . . . . . . . ..lv~O i 1 r-r-i--'--r-+---'---I.:····················· 0.6 <'Ii i 1 0.2 .....................1...................... \...................... \...................... j ....•••..•.••••.. ,•••. : ..•••.••••.•••••••.•••i ..,................... I II- -0.2 . . . . . . . . . . .\. . . . . . . . . .+. . . . . . . -0.6 -1.0 . ; . : : , I : ; ZI~NOVS t.....................~ ..................... L~ED AV~INSKY 10 ·········1······················1······················1··················+····················1···.9.········....... j.................... I : ·l:~;!1+ : .........:....................,................... -:-................... ! 0 L--_ _- ' -_ _- ' -_ _ -0.8 -0.6 -0.4 _ _-'--_ _-'-_ _-'-_ _---J'--_---' -'_~L -0.2 0.0 0.2 0.4 0.6 0.8 1.0 Factor 1 Figure 7a. June 1996 presidential election, pro-Yeltsin regions. 0.8 : -tI;+*~ 0.6 : : : 0.4 ·····················:·····················r··················:····················r································································l·············......... ]""" ................. . 0.2 C') ~ ~ ................. ·····!··········· .. ·;~r 0.0 .................... .YEL.TSIN.......l. ...... ~ : 0 1 -j-...... ·!l~; ............ . ............i......................l ......................t.....................i......................~ ......................l.?................. ' : : • i : +. . . . . . . . . . -0.2 ·····················,·····················r··················r··················i····················j·····............... 1...................... ,..................... -0.4 -0.6 :~;-I= ! ! !0 -0.8 L-_ _.l--_ _- ' -_ _-!' -_ _- ' -_ _- '!-_ _......L_ _- 'i-_ _-.i-_ _--' -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 Factor 1 Figure 7b. June 1996 presidential election, pro-YeJtsin regions. [293 ] 512 0.4 0.2 "YA~NSKr'l,V 0.0 ·'[~ri!I . . . i t ! i j i I I .......................... ..............i.................................;.................................~ ................................. J......•......•....................i................................ -0.2 Iii C\I j YELJSIN. --I----t--I---I---+. . . . . . . . . . ·. . . . . l. . . . . ·. . . . . . ·. . . . . . . . . . .r. . . . . . . . . . . -0.4 -0.6 ••••••••••..••••••.••••••••••••• .; •••••••••••••••.••••••.•••••••••. j •••.•••••••••••••••••••••••••••••.j. ••••••••••••••••••••••••••••••••• j ..•••••...••••...•••..•••...••....j. •••••••••••••••••••••••••••••••• -0.8 1"...............................1".............................. ·:~; ································t·································!····Q··························t..·······························1·································t································ -1.0 -1.2 L-_ _ _--'-_ _ _---'1 _ _ _ _--'--_ _ _ _.i....-_ _ _- l 1 _ _ _-'-_ ~ ·1 l -1.0 -0.6 -0.2 0.2 0.6 1.0 1.4 Factor 1 Figure 8a. June 1996 presidential election, anti-Yeltsin regions. 0.6 0.4 0.2 .., j 0.0 -0.2 -0.4 -0.6 ~l=-I: •••••.••••••••..••••••.••••••••• +..••••••••••••••••••. -0.8 -1.0 -0.6 ············1·································0)···············..·····..·········1·····························..... ................................ ~ -0.2 0.2 0.6 1.0 Factor 1 Figure 8b. June 1996 presidential election, anti-Yeltsin regions. [294 ] 1.4 513 0.8 0.6 ·.. ·· ..·..·...... ·.. ·····.. ·· .. ·:. ··....·...... ·· .. ·· ......·..· .. ·1'·. ....... ····.. ·· ..··· ....·.... 0.4 f· ........···· ....·. ··....···....j..·.... ···......·· ..·'Yr;:[f~N : : 0: ..·....··.... ·· .. ····· .... 0.2 0.0 N ~ 01 u. -0.2 -0.4 -0.6 ................................L. .............................. j......................·.....··...t··....··.·· . ····..····..·..··.··I..····.... -0.8 ................................+.................................1................ ~vL· ·~ ...........······..f..·····..·..·····..···· ....····· . . · ·. . . · · . · · . ·. I·....···..·. ·····. ·········....t. . . . . . . . . . . . . . . .. -1.0 L._ _ _-'-_ _ _---''--_ _ _-'-_ _ _--'-_ _ _ _......._ _ _....I -0.8 -1.2 -0.4 0.0 0.4 0.8 1.2 Factor 1 Figure 9a. June 1996 presidential election, moderate regions. 0.6 0.4 0.2 .., 0.0 ~ 01 u. -0.2 ~.=:i-rf :!~=tI -0.4 -0.8 -1.2 -0.8 -0.4 0.0 0.4 0.8 1.2 Factor 1 Figure 9b. June 1996 presidential election, moderate regions. [295 ] 514 Table 6. 1996 presidential election, first round Eigenvalue % Variance Cumulative % "explained" Pro-Yeltsin regions Factor I 1.91 38.3 38.3 Factor 2 1.62 32.4 70.6 Factor 3 0.83 16.8 87.4 60.1 Anti-Yeltsin region Factor I 3.00 60.1 Factor 2 1.16 23.2 83.3 Factor 3 0.44 8.8 92.1 2.48 49.7 49.7 Factor 2 1.20 23.9 73.6 Factor 3 0.79 15.8 89.5 Moderate regions Factor I However, Figures 7a and 7b, which concern the pro-Yeltsin regions, stand in sharp contrast to these results. First, factor 1 seemingly inexplicably has Yeltsin and Zyuganov occupying approximately the same position, while the other three candidates, including Zhirinovski, occupy the polar opposite point. Second, although factor 2 divides Yeltsin and Zyuganov so that it is tempting to think of it as corresponding to an anti- versus pro-reform dimension, unlike factor 1 in Figures 8a and 9a, it leaves the remaining three candidates at the same position near the center. Figure 7a, then, is wholly unlike any earlier counterpart, including Figures 8a and 9a for the anti-Yeltsin and Moderate subpopulations. Moreover, even the third factor for the pro-Yeltsin regions is unlike that factor in the remaining population. In Figure 7b, factor 3 essentially differentiates Yavlinski from Zhirinovski, whereas in Figures 8b and 9b, it more or less differentiates Lebed from the others. There are any number of explanations for such results, ranging from a violation of the assumptions necessary to allow factor analysis to accurately recover spatial maps to speculations about strategic voting under Russia's runoff rules for presidential elections. However, before we speculate, it is useful to reconsider the December 1995 parliamentary contest, which preceded the first round of presidential balloting by only six months. Dividing our sample as before on the basis of the final round of balloting into Pro- [296] . 515 1.2 ,-'iI~:T_ C'I ~-'r",. : LDPR : 0 0.4············ ·.. l. . . . . . . . . . ·, . . . . . . . . . . ! O.~-!I:=Lr; -0.4 ..........· ...... T. ·OHR.... ! 0 r 1...................... ,.............. ......j ........ ..,... j j j , ·l.......... · ......··\ ....·.... · ....:--· ...... ·....t .. .. . -0.8 L-_--'_ _--'-_ _--'-_ _-'-_ _-'--_ _'---_---i_ _--'-_ _-I -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 Factor 1 Figure JOa. December 1995 Duma election, pro-Yeltsin regions. Ye1tsin, Anti-Yeltsin and Moderate regions, Table 7 reproduces Table 6, while Figures 1Oa,b, lla,b, and 12a,b parallel those we show for the presidential election. Notice first with respect to Table 7 that the data from all three sub-populations look essentially equivalent - and in particular, unlike Table 6, there is no sub-population for which the first two factors are of approximately equal importance. The plots here of candidate positions also seem more in keeping with initial expectations. Nevertheless, there is an important exception: although factor 1 appears to correspond to an anti- versus pro-reform dimension and factor 2 appears primarily to separate Zhirinovski's LDPR from other party lists, the list headed by prime minister Chernomyrdin, Our Home Is Russia (OHR), is, like Ye1tsin himself in 1996, located far to the right in pro-Yeltsin regions but, in keeping with our intuition about things, is given a more centrist and pro-reform position in all other regions. Thus, if we treat Yeltsin and Chernomyrdin as equivalent, the parallelism between Figures 7a and 10a leads us to reject any speculation about strategic voting and the like - speculations that relate to voting rules, since the rules in 1995 (party-list PR) and 1996 (majority rule with a runoff) are wholly distinct. Insofar as other potential explanations are concerned, perhaps the one most consistent with the usual journalistic interpretations of events concerns the attitudes toward Yeltsin and his administration among pro-reform voters in combination with the distribution of preferences in pro-reform regions. [297 ] 516 0.6 0.4 ~ 0.2 ..,... 0.0 -0.2 -0.4 -0.6 -0.8 -1.0 L-_.l~ -0.8 _ _-L._ _...L._ _. i . - _ - - - i_ _--i.._ _- ' -_ _.J -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 Factor 1 Figure lOb. December 1995 Duma election, pro-Yeltsin regions. 1.0 , , CPAF o ! 0.6 GAlbAA q 0.2 C\I ~ ?f. -0.2 -0.6 ························r························;·························r·······················i························!························r·······················AMPiiov········ , , ! ii, I 0 -1.0 L-_ _..L-_ _..L._ _--'-_ _--i.._ _- - '_ _ _' - -_ _.L-_ _.J -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 Factor 1 Figure 11Q. December 1995 Duma election, anti-Yeltsin regions. [298] 0.4 0.6 517 1.0 0.6 ........................t........................ L........................;. .........................;.........................,c. •••••••••••••••••••••••• (. ••••••••••••••••••••••• i ......................... ·"GA~R 0.4 r C') j . . . . .···········t·..···················· ··l··············..·····.. ·t................... ~J-B. 0.8 ........................;..................... ··l·························t ~=r:,l;I[- 0.2 0.0 -0.2 : 0: ~R' . .l........................~ ........................l. . . . . . . . . . . . l. . . . . . . . ~ . . . . . . . .. : ::: I .......•...................... : . : : AMPILOV i : l". . . . . . . . . . . . .t................ ____.. 1 ; -0.4 ........................j.........................;...·····················t························ : . 0 1 ........................1 .. __ _ 1 _0.61....--_ _-'-_ _ _i........_ _--'--_ _---'_ _ _...i...._ _- - '_ _ _....i....._ _--' -1.0 -0.6 -0.8 -0.4 -0.2 0.0 0.2 0.4 0.6 Factor 1 Figure 11 b. December 1995 Duma election. anti-Yeltsin regions. 0.8 , _=:It~! 0.6 0.4 0.2 ... j . . ·!GA1D~R-i . o· t\I .s 0.0 ..................... i ..................... i :6 LL -0.2 .... ·.. ·'iAVlJ~Sr 0: -0.4 -0.6 ; i ; ~ I j j : : ............... j.................... ~ ...................... !.................... j"...................... t ... . .l......................L.....................!......................t .....................;,.....................J...................... ' , . ........... ( ................... +..................... ;......... I : ..,......................,... I .j.................... . AMPiLOV j.....................q.................. .. LEBED : --__~!,i"-lj;.T+ . . . . . . .,. . . . . ,. . . . . . . . . ., . . .. . . . ,. ..rJR.........,...... _0.81....--_ _.i..-._ _..i..-_ _....i....._ _...i...._ _--'--_ _--'--_ _- ' -_ _- L_ _--' -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 Factor 1 Figure 12a. December 1995 Duma election, moderate regions. [299] 518 0.8 0.6 0.4 0.2 C') ~ :s 0.0 -0.2 -0.4 -0.6 -0.8 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 Factor 1 Figure 12b. December 1995 Duma election, moderate regions. First, in earlier elections, pro-reform voters, unlike their pro-communist counterparts, demonstrated a clear willingness to divide their vote among a variety of party lists (Yabloko, Russia's Choice, Our Home Is Russia, and a double handful of smaller parties led by individual personalities) or, when dissatisfied, to simply stay home. Indeed, reform candidates themselves in 1993 and 1995, viewed sometimes as a collection of prima donnas with presidential aspirations, seemed to spend more time focusing on the policy and personality differences among themselves than the things that set them apart from their communist or anti-reform opposition. Thus, Yavlinski and Gaidar, both proWestern economists, found it impossible to coalesce in 1995, while it did not seem at times in 1996 that Yavlinski cared whether his candidacy might help elect Zyuganov. Add in economic and social circumstances along with the war in Chechnya, and we can reasonably suppose that in December 1995 and June 1996 the more liberal and pro-reform a voter might be, the greater would be the dissatisfaction with Yeltsin and his administration, and the more likely would that person be to cast a ballot in 1995 for anyone of a handful of splinter democratic parties or, in June 1996, for someone like Yavlinski or the enigmatic General Lebed. This behavior, of course, would move Yeltsin (and, extending that argument backwards, to December 1995, Chernomyrdin's Our Home Is Russia), to the right, in the direction pro-reform electorate - one that would never vote communist but that also would not support Yeltsin [300] 519 Table 7. 1995 parliamentary contest Eigenvalue % Variance Cumulative % "explained" Pro-Yeltsin regions Factor 1 2.70 38.7 38.7 Factor 2 1.42 20.3 59.0 Factor 3 1.18 16.9 75.9 Factor 1 2.76 39.4 39.4 Factor 2 1.52 21.8 61.3 Factor 3 1.00 14.3 75.6 Anti-Yeltsin regions Moderate regions Factor 1 2.46 35.2 35.2 Factor 2 1.30 18.6 53.9 Factor 3 1.01 14.3 68.2 unless the choice were between him and a communist - would generate data in which Zyuganov and Yeltsin appear on the same side of the first factor. In effect, then, the supposition here is that among pro-reform regions, with voters sensitive to the nuanced differences between pro-reform candidates and parties, the electorate's mean preference consistently falls between Yeltsin in 1996 or Our Home Is Russia in 1995 and all other pro-reform parties and candidates, in which case a significant share of any increase in Yeltsin's or Chernomyrdin's vote there comes from the reformist camp and not exclusively from Zyuganov or any of his communist, anti-reform counterparts. In this event (i.e., absent adequate variance in mean preferences) factor analysis sets Yeltsin and Our Home Is Russia apart from the remaining proreform alternatives and near the anti-reform bloc. On the other hand, because Yeltsin and Zyuganov's vote still correlate negatively, a second nearly equally important dimension is required to distinguish these two candidates just as the second dimension in 1995 distinguishes Our Home Is Russia from the other party lists. In the second round of the presidential election, however, with the population mean to the right of Yeltsin and Yeltsin to the right of Zyuganov, these regions vote most heavily for Yeltsin. For anti-reform or communist voters, of course, Yeltsin, Chernomyrdin, and the rest of the administration are little different than any other advocate [301] 520 of reform, democracy, and a market economy. But insofar as explaining the relative consistency in Zyuganov's estimated spatial position in contrast to the inconsistency in Yeltsin's or Chernomyrdin's, we note that Zyuganov appears to have been much more successful at holding the communist and anti-reform voting blocks together, so that voters in anti-reform regions saw no nuanced difference among anti-reform alternatives. Especially in 1996, Zyuganov stood virtually alone on the right. His only semi-serious competition there was Anpilov, whose vote appears largely to have been limited to winning some disaffected voters who had abstained in previous elections (Myagkov, Ordeshook, Sobyanin, 1997). And although in 1995 the Communist Party under Zyuganov confronted nearly as many splinter anti-reform challengers as did Our Home Is Russia, Zyuganov largely succeeded in consolidating the anti-reform vote, as witnessed by the fact that much of his party's increased support came from voters who had cast ballots in 1993 for the fellow-traveling Agrarian Party. Put differently, the speculation here is that the median preference in 1995 and 1996 in anti-reform regions varied around Zyuganov and not merely to his right or left, and thus allowed factor analysis to recover a relatively undistorted estimate of his positn~ Or to put matters differently still, Zyuganov in 1995 succeeded in positioning himself at the overall median of voters likely to support him; in contrast, Yeltsin and Chernomyrdin found themselves closer to the overall population median (hence Yeltsin's eventual victory), but consistently to the right of preferences among voters most supportive of reform. 5. Conclusions Naturally, any conclusions we offer here about spatial positions need to be tempered by the methodological limitations of our analysis - most notably, the fact that we rely here on data with relatively low information content. Nevertheless, the broad sweep of our conclusions is consistent with the intuition that throughout the 1991-1996 period Russia remained divided between pro- and anti-reform positions and that no candidate succeeded in establishing a viable middle or centrist position. Our Home Is Russia appears to come closest to establishing that position in 1995 as its identity, but even then it is best identified as pro-reform except perhaps among those voters who made sharp distinctions among such parties. This finding may, of course, be an artifact of our methodology. But we suspect that it is also a consequence of a sharply polarized electorate and the fact that no candidate or party has thus far attempted or succeeded at being the happy face of democratic capitalism - the "Social Democratic" alternative to communist and reformist camps. It was assumed during the early stages of the 1996 presidential campaign that [302] 521 such a platform was available to Zyuganov. But apparently out of fear of alienating other more radical members of his coalition (or perhaps because he IS a communist of the old school, pure and simple), Zyuganov ran the lackluster campaign of a person more familiar with old communist election procedures and slogans. That center, then, appears to remain unoccupied. It is unlikely, moreover, that any of the current crop of reform personalities can change this spatial map. Throughout the 1991-1996 period, reformersmost notably Gaidar and Yavlinski - have anchored one end of the first dimension of competition and it is unlikely that any of their kind can transform the electorate's view of them so that they are perceived in more moderate terms. Of course, the difficulty with filling the center is that if the electorate is bipolar, then under majority rule with a runoff, no person is likely to advance such a platform and advance to the second round of presidential balloting. But here we need to enter another note of caution with respect to drawing conclusions about the future. Political leadership does not always involve acting like a Bill Clinton so as to mold one's policies and campaign pronouncements on the basis of the latest public opinion polls. Leadership also entails shaping issues and public opinion to one's own purposes - in this case, reformulating the issue of reform so as to create a viable center. Unfortunately, Russia today appears to have few if any leaders capable of performing this task. Insofar as the feared issue of nationalism is concerned, it is nowhere to be found until 1996, and even then it (the second factor) does not wholly set Lebed and Zhirinovski apart from all other candidates or at least from pro-reform candidate like Yavlinski whose pro-Western orientation manifests itself with frequent visits to Washington D.C. and Cambridge (Mass.). Interestingly, though, notice that although nationalist rhetoric in 1996 may have resonated with voters in the anti-Yeltsin and "Moderate" regions (see the second dimension in Figures 8a and 9a), that issue within pro-Yeltsin regions appears to take a backseat to the structuring of preferences around reform and Yeltsin himself. Of course, none of our conclusions or spatial maps can be approached with confidence until they are confirmed or otherwise replicated (in subsequent elections) with richer data - with individual poll results that measure each respondent's cardinal rankings of the candidates or parties. Only then can we ascertain the relationship between the "fundamental" dimensions uncovered by a multidimensional scaling methodology and the more narrowly construed issues discussed in a campaign (see, e.g., Enelow and Hinich, 1984). Nevertheless, the most general albeit tentative conclusion we reach here is that the Russian electorate is not altogether infertile ground for the development of a coherent party system that would primarily pit a left-learning and rightleaning party against each other. Although we cannot reject the hypothesis [303 ] 522 of a polarized electorate, the relative constancy and coherence of the issue space over five years of tumultuous socio-economic change seems a firm basis upon which parties can fashion coherent ideological positions. And even if secondary issues in the form of valence dimensions (e.g., ORR's position on the second dimension in 1995) arise from time to time, Lebed's apparent ability to moderate other secondary Euclidean issues (i.e., nationalism) without negating his pro-reform positioning suggests that parties can accommodate this multi-dimensionality. It is indeed unfortunate, then, that Russia's electoral system discourages such a development. National partylist PR in parliamentary contests followed by a majority rule plus runoff presidential election can only sustain the highly fragmented and personalized "party system" we see today. This fragmentation may present social scientists with interesting data, but it is unlikely to facilitate political stability or a viable federal state (Ordeshook, 1996; Ordeshook and Shvetsova, 1997). Notes 1. For presidential elections, Russia uses majority rule with a runoff. For its parliamentary contests to the State Duma, it uses a split system in which one-half of the deputies (225) are elected from single-mandate constituencies and one half using national party list proportional representation. For the party list contests, only those parties with at least 5% of the vote are awarded seats. The data we use here from the parliamentary elections are restricted to the party list totals. 2. Zhirinovski's LDPR won 21.1 %, Yegor Gaidar's Russia's Choice won 14.3%, the Communist Party (CPRF) won 11.5%, the Agrarians won 7.4%, Women of Russia 7.5%, Yavlinksi's Yabloko 7.3%, Shakrai's list 6.2%, and Travkin's DPR 5.1%. 3. Those four were the CPRF with 22.7%, the LDPR with 11.4%, OHR with 10.3%, Yabloko with 7.0%. 4. Since turnout increased in 1996 from 1995, "the party of nonvoters" here corresponds to those who voted in 1996 but not in 1995. That is, the share of the vote attributed to this "party" in each rayon equals the percentage increase in turnout between 1995 and 1996. 5. And even if we limit the analysis to 5 candidates (CPRF, LDPR, OHR, Yavlinski, and Nonvoters), the first factor accounts for a mere 37.6% of the variance. 6. If for purposes of comparison we perform a 5-candidate recovery for 1995 (OHR, Nonvoters, Yavlinski, CPRF, and Zhirinovski), the cumulative variance explained by the first four factors is, in sequence, 37.6%, 59.5%, 78,5% and 94.3%. Thus, even if we limit the analysis to the same number of candidates, the variance explained increases more rapidly in 1996 than 1995. 7. Elsewhere we estimate that Yeltsin won 5 out of every 8 of Lebed's voters, and 6 out of every 8 of Yavlinski's, whereas Zyuganov won 6 out of every 7 of Zhirinovski's voters (Myagkov, Ordeshook, and Sobyanin, 1997). 8. Out of fear of alienating the more extreme members of his coalition (and perhaps even as a matter of ideological preference) Zyuganov never embraced a "Social Welfare Democratic" platform and instead left broad hints of massive renationalization of industries, and other wholesale reversals of Yeltsin's policies. In any event, it was clear that Yeltsin's [304 ] 523 coterie of pro-Western economic reformers would have no role to play in a Zyuganov administration, except perhaps as residents of one gulag or another. 9. Pro-Yeltsin regions are those that gave Yeltsin more than 60% of the vote in the second round of balloting; whereas anti-Yeltsin regions are those that gave him less than 42%. References Boxer, Y., McFaul, M. and Ostashev, V. (1996). Elections 1993-1996: Stable and metastable electorates. Working Paper. Moscow: Carnegie Institute. Cahoon, L., Hinich, M.J. and Ordeshook, P.e. (1978). A statistical multidimensional scaling method based on the spatial theory of voting. In P.C. Wang (Ed.), Graphical representation of multidimensional data. New York: Academic Press. Dow, J. (1997). A spatial analysis of the 1989 Chilean presidential election. Working Paper, Department of Political Science, University of Missouri at Columbia. Enelow, J. and Hinich, M.J. (1984). The spatial theory of voting. New York: Cambridge University Press. Fish, M.S. (1995). The advent of multipartism in Russia, 1993-1995. Post-Soviet Affairs 11: 340-383. Hinich, M.J. and Munger, M.e. (\994). Ideology and the theory ofpolitical choice. Ann Arbor: University of Michigan Press. Hough, J.F., Davidheiser, E. and Lehmann, S.G. (1996). The 1996 Russian presidential election. Washington DC: The Brookings Institution. Lin, T.M., Chu, Y.-h. and Hinich, M.J. (1996). Conflict displacement and regime transition in Taiwan: A spatial analysis. World Politics 48: 453-481. McFaul, M. (1996). Russia between elections: What the 1995 results really mean. Moscow: Carnegie Endowment for International Peace. McFaul, M. and Fish, M.S. (1996). Russia between elections. Journal of Democracy 19: 90118. Myagkov, M., Ordeshook, P.e. and Sobyanin, A. (1997). The Russian electorate, 1991-1996. Post Soviet Affairs 13: 30-62. Myagkov, M. and Sobyanin, A. (1995). Irregularities in the 1993 Russian election. Working Paper. Humanities and Social Sciences, California Institute of Technology. Ordeshook, P.e. (1996). Russia's party system: Is Russian federalism viable? Post-Soviet Affairs 12: 195-217. Ordeshook, P.e. (1997). The spatial analysis of elections and committees: Four decades of research. In D.e. Mueller (Ed.). Perspectives on public choice. Cambridge: Cambridge University Press. Ordeshook, P.C. and Shvetsova, O. (1995). If Madison and Hamilton were merely lucky, what hope is there for Russian federalism? Constitutional Political Economy 6: 107-126. Wyman, M., White, S., Miller, B. and Heywood, P. (1995). Public opinion, parties and voters in the December 1993 Russian elections. Europe-Asia Studies 47: 591-614. [305 ]