Outsourcing and Unionization: From Labor Organizations to Re-organized Work

School of Economics UNSW, Sydney 2052 Australia http://www.economics.unsw.edu.au Outsourcing and Unionization: From Labor Organizations to Re-organized Work Elisabetta Magnani and David Prentice School of Economics Discussion Paper: 2007/07 The views expressed in this paper are those of the authors and do not necessarily reflect those of the School of Economic at UNSW. ISSN 1323-8949 ISBN 978 0 7334 2439 7 Outsourcing and Unionization: From Labor Organizations to Re-organized Work. Elisabetta Magnani The University of New South Wales and David Prentice La Trobe University Abstract. While many believe unionization has little to gain by recent increases in outsourcing, this question has not been analysed systematically. In this paper, using a new dataset for US manufacturing from 1973 to 1994, we analyse the e¤ect of outsourcing on unionization. Instrumental Variables estimation shows outsourcing contributes to higher quasi-rents and industry employment. We …nd the union wage premium of substitutable workers is not a¤ected by the extent of outsourcing. However, unionized workers in jobs that are not substitutes of the tasks being outsourced gain from outsourcing. Finally, we …nd no support for the claim that outsourcing reduces unionization. JEL classi…cation codes: J51, J31, J24 1 “The largest employer in America is a temp agency now. So we have to …gure out how to organize this segment, what the glue is. . . Not in my lifetime.” A former organizing director of the American Federation of Labor-Congress of Industrial Organizations (AFL-CIO), cited in Herzenberg (2000). Two major institutional changes are occurring under our eyes, namely the rapid dismantling of work organizations and the evolution of the boundaries of the …rm. In U.S. private manufacturing the union membership density has more than halved, dropping from 38.9 to 18.2 percent from 1973 to 1994, and similar though less dramatic changes have been observed in Europe, Australia and Japan (Blanch‡ower and Freeman, 1992). Many observers have argued it is market competition that is mainly responsible for undercutting union labor in the wealthy OECD countries (Wood, 1994). Employers have contributed to this trend by investing heavily in avoiding unions (Freeman, 2005; Farber, 1990; Freeman, 1986). Indeed, there is little doubt that the pressure of the world economy has forced wholesale changes in the internal organization of the …rm (Piore and Sabel, 1984). A few of the most used strategies to face the increasing product market competition and the workers’ organizations have been the substantial changes in work rules, the partial dismantling of internal labor markets with the exclusion of some activities and workers from the …rm and the remarkable growth in the use of market-mediated employment arrangements, such as temporary labor and outsourcing (Autor, 2003; Kochan, Katz and McKersie, 1994). According to this general argument the emergence of ‡exible work organization has hampered union organizing activities. While the di¤usion of employment arrangements that are mediated through the market rather than through the …rm (and the unions) poses tough challenges of how best to organize workers in non-standard employment relationships, there is still very little understanding of the way outsourcing threatens, if truly it does, the future of unionism. In a time in which there is a widespread feeling that US unionism needs additional knowledge to come out with new forms of unionism (Freeman, 2005), this paper aims to contribute to such a purpose. 2 We analyze if outsourcing is associated with the decline in unionization using a comprehensive data set for US three-digit manufacturing industries between 1973 and 1994. Because the decline occurred over decades, we require compatible data available annually over the whole period.1 The data has been used elsewhere to capture important insights about the evolution of trade unionism (see Magnani and Prentice, 2003; Magnani and Prentice, 2006). The U.S. manufacturing sector over the last three decades provides a particularly good case study. Starting from the late 1960s the US manufacturing sector has witnessed substantial growth in the use of outsourced labor services, especially compared with total employment (see table 1). While in general outsourcing has grown at a brisk pace in all manufacturing industries, sizeable di¤erences in outsourcing both across industries and across time remain and it is such variation that the empirical literature has exploited (e.g., Bartel el al., 2005; Magnani, 2006). To many observers the use of outsourced labor services has greatly contributed to the productivity recovery in US manufacturing in the late 1980s and 1990s (Siegel, 1995, ten Raa and Wol¤, 2001). These stylized facts call for an evaluation of the e¤ects of outsourcing on industry quasi-rents, labor productivity, employment, the union wage premium and unionization. Some of the key …ndings of our study are the following. Outsourcing has contributed to higher quasi-rents and increased quasi-rents per plant. However it has not increased quasi-rents per employee because outsourcing has been accompanied by increased use of industry labor, which, with a visible increase in quasi-rents, has kept quasi-rents per employee stable. Despite the apparent lack of relationship between outsourcing and quasi-rents per employee, we …nd interesting di¤erences in the way outsourcing impacts upon the union wage premium of workers with jobs similar to the task being outsourced and the union wage premium for other workers. While the premium of substitutable workers is not a¤ected by the extent of outsourcing, we …nd evidence that unionized workers employed in jobs that are not 1 This precludes the use of relatively detailed, but infrequently published data, such as that used by Bartel et al. (11). 3 substitutes of the tasks being outsourced gain from outsourcing. Finally, a set of tests on whether there is any negative relationship between unionization and the extent of outsourcing fails to …nd any support to this hypothesis. These results drive our main conclusions according to which US manufacturing unions have misallocated their resistance resources. To be clearer, our results support those views in favor of an an alliance between unionized in-house employees and external workers (e.g., Hiatt and Jackson, 1997; Cobble, 1991) to share the increase in quasirents from outsourcing. I. Labor organizing institutions and reorganized work. There is widespread belief that from the 1970s the combination of globalization, increased domestic competition and a more volatile macroeconomic environment has created a demand by …rms for much greater labor market ‡exibility than during the postwar boom of the 1950s and 1960s in all major industrialized countries, including the US, the European nations, Canada and Australia (Abraham, 1996; Segal and Sullivan, 1997; Blanchard et al. 1995). Before proceeding, it is worthwhile expanding on what is meant by ‡exibility. Labor market ‡exibility is a multidimensional concept that in general measures the ability of the labor markets to rapidly respond and adjust to shocks. In recent times …rms’ demand for ‡exibility has translated into a demand for new work practices and patterns, which aim to increase numerical ‡exibility i.e. ease in adjusting numbers and hours of employees, and functional ‡exibility, i.e. ease in adjusting the tasks performed by the organization or even individual workers, so to increase productivity while relaxing job security regulation (Osterman, 1999; Kalleberg, 2001). One of the ways in which …rms have responded to competitive challenges to increase ‡exibility has been by changing the mix of employment arrangements. Tasks that were previously performed by workers directly hired by the …rm were increasingly done under contract with …rms in the business service sector and through employment arrangements that involved temporary workers and outsourcing (Autor, 2003). 4 The extent of the increase in the use of alternative employment arrangements is evident if we look at the striking performance of the US business service industry (Standard Industrial Classi…cation code 73) in general, and by the performance of the Temporary Help Supply industry (SIC 736) in particular (reported in Magnani, 2006, Table xx). Magnani (2006) documents that between 1949 and 1998, the cost share of purchased services (outsourcing) grows from 4 percent to 12 percent. Interestingly, two studies …nd links between the rise of outsourcing and productivity growth. First, Siegel (1995) argues that improvements in manufacturing productivity cannot be explained by measurement errors but rather by outsourcing, an increase in the rate of investment in computers, and unmeasured changes in the quality of output and the labor force. Similarly, ten Raa and Wol¤ (2001), relate the recovery of standard TFP growth in manufacturing during the 1980s to an increased use of outsourcing of inputs from service industries as well as to technical change. The ability of unions to cash part of these productivity gains depends largely on whether outsourced labor services and labor services provided by the internal permanent workforce are substitutes or complements. If there is substitutability between internal workers and outsourced labor services, outsourcing may a¤ect internal workers’ productivity by acting like a “worker’s discipline device”. Again, if the two labor inputs are substitutes, outsourcing may raise industry quasi-rents but it will have a limited impact on the share of rents accruing to union members. In fact it could even be that employers use outsourcing to limit unions’ wage demands — a variation on the outsourcing to limit labor costs hypothesis often explored in the literature (Abraham and Taylor, 1996). Kalleberg (2001) reviews the existing literature to …nd little consensus on the relationship between numerical and functional ‡exibility strategies …rms have adopted in recent times. According to Lindbeck and Snower (2001), increased functional ‡exibility and reduced task specialization among workers within a …rm has gone hand in hand with increased specialization in production among …rms, a down-sizing process that involves more 5 narrow focus on a …rm’s “core competencies” in production and the outsourcing of “non-core competencies”. These two sets of empirical evidence, the ones relating market-mediated employment arrangements to increased manufacturing productivity, and the ones confronting functional and numerical ‡exibility lead us to two overlooked aspects of the debate. First,to what extent union members (part of the stable component of the labor force) can bene…t, if at all, from the …rm’s decision to outsource part of its activities. Second, to what extent unions as institutions are a¤ected, or can a¤ect, the use and e¤ects of outsourcing. A. What is Labor Organizations’ Attitude Towards Work Reorganization? An Overview of the Evidence. As highlighted earlier, one of the characteristics of the previous workplace organization in manufacturing was its high unionization - which has declined considerably since the 1970s (for a review of this issue see Magnani and Prentice (2003)). Where workplaces have remained unionized, these changes have posed considerable challenges to trade unions. Although it is not possible to summarize the complex …ndings of a large literature, it is important to stress that studies on the unions’ reaction to the introduction of innovative workplace arrangements have greatly contributed to question the view, prevailing in the industrial relation system since World War II, according to which “management manages and the union grieves”.2 Much less known is the unions’ attitude towards the reorganization of the workplace that involves the use of market-mediated employment arrangements such as outsourcing. One approach has focussed on the threat to union power. As Perry (1997) clearly states, at the macro level the dimension of union power heavily depends on its degree of success in “taking labor out of competition" (from outsourced labor services). If internal unionized labor services and outsourced labor services are highly substitutable, the long run e¤ects of outsourcing may be reduced union wages, particularly if there are reductions in employment as well. Rather di¤erent is the view that comes from seminal contributions on outsourcing. If we understand outsourcing, as Drucker (1995) and Quinn and Hilmer (1994) do, as a process of turning over a part or all of those functions (or skills) that fall outside the organization’s chosen core competencies to an external supplier whose core competencies and skills are the functions being outsourced, outsourcing involves highly specialized activities. Since cost factors are not the only long term determinants of outsourcing and given the complementarity existing between core functions and non-core outsourced functions, unions should be able, at least in the short-medium run, to cash sizeable shares of the bene…ts from outsourcing. If unions are rent-extracting institutions, they should favor any change in the internal organization of the …rm that leads to an increase in the quasi-rents produced. Given the abundance of case studies in which outsourcing has led to legal confrontations between unions and management (reviewed in Miscimarra and Schwartz 2 For instance, the reader is referred to Eaton and Voos (1989) who witness a variety of American union attitudes to “team production systems”. There is evidence of mainstream unions having to innovate in contributing to the competitive advantage of the …rms. For instance in the US steel industry the unions played a key role in the implementations of HPWO work systems (Osterman, 1999; p. 173). 6 (1997)) it is obvious that there is much more at stake in the use of outsourced labor than simple e¢ciency gains at no cost for union members. The issue of how unions actually bene…t if at all from the productivity gains, re‡ected in increased quasi-rents, made possible by the reorganization of the …rm towards outsourcing is the object of investigation of the present study. In the next section we propose a theoretical framework to think about it. II. Unionization and Outsourcing: A Theoretical Framework. In this section we review the unions as rent-extracting institutions framework of Abowd and Farber (1990) and introduce and analyze the e¤ects of outsourcing. For an industry with employment, N , and union membership, L, unionization, U , is the proportion of industry employment that are union members, U = L=N . In the standard model, unions maximize the share Qh(U ) of industry quasi-rents accruing to union members less organizing costs, where Q is industry quasi-rents and h(U ), with 0 < h(U ) < 1, is the function that yields the bargaining power of the union. Organizing costs are equal to Lg(U; S). The per worker cost of organizing the industry, g(U; S), depends positively on both unionization U and employer resistance S.3 The de…nition of industry quasi-rents adopted here is both standard and suitable for estimation. As in Abowd (1989) and in Abowd and Allain (1996) industry quasi-rents is expressed as the sum of two items: union members’ and shareholders’ returns. Union members’ returns is the excess of union labor costs over labor costs using a non-unionized work force, or L(W w), where W is the bargained wage, and w is the wage rate received if the workers were not unionized. Shareholders’ returns is de…ned as pro…t (excluding capital costs), [R WL w(N L)]; where R is total revenues. Thus, as in Abowd (1989) industry quasi-rents are Q = [R 3 Abowd and Farber (1990) do not specify a functional form for h(U ) and g(U; S):They assume the organization cost function g(U; S) is increasing and convex in U and increasing and concave in S. Furthermore, they assume that the union bargaining power function h(U ) is increasing and concave in U . They also assume bargaining between the union and …rm is strongly e¢cient. This implies that bargaining does not reduce the size of the total surplus i.e. the size of …rm quasi-rents (Brown and Ashenfelter, 1986). 7 WL w(N L)] + L(W w) = R wN; the value of revenues minus labor costs valued at the opportunity cost of a non-unionized worker’s time.4 We now add outsourcing to this framework. Denote O as the share of workers that are outsourced. Both quasi-rents and organizing costs are assumed to have a positive relationship with outsourcing. The e¤ect of outsourcing on union bargaining power will depend on the nature of outsourcing. If there is extensive outsourcing of substitute workers (which is the usual case considered), union bargaining power will fall. If complementary workers are outsourced, there may be no e¤ect or may even increase the bargaining power of complementary workers (if they become more essential as a result of outsourcing). Formally, we write the problem that unions faced with outsourcing have to solve as: max Q(O)h(U; O) U N U g(U; S; O) (1) A union will have a benevolent attitude towards outsourcing if outsourcing increases quasi-rents without signi…cantly decreasing union strength or increasing organizing costs. Abowd and Farber (1990) show that under certain conditions there is a positive relationship between unionization and quasi-rents. Thus, by increasing quasi-rents, outsourcing may indirectly provide trade unions with growth opportunities. Obviously a very di¤erent case arises if outsourcing increases industry quasi-rents but also reduces union bargaining power, h(U; O), because outsourcing represents an implicit threat to regular, full time employees or because it increases organizing costs, reducing unionization. Union bargaining power may decrease because of the implicit (or explicit) threat to substitute away from regular full-time employees towards outsourced workers. Organizing costs may rise, as the workforce becomes more dispersed, less localized and less full-time. In these cases while outsourcing 4 Obviously the time dimension here is important, since in the very short run, most normal returns are quasi-rents. In other words, quasi-rents include the normal competitive return to the …xed assets in an industry if, as in the short/medium run, employers do not have the chance to protect the normal return component of the quasi-rents by transferring the assets to other industries. This is the time frame assumed in the analysis that follows as well as in Abowd (1989) and Abowd and Allain (1996). 8 may increase the potential gains to union members by increasing quasi-rents Q, the ex-post realized gains L(W w) are reduced because of a smaller L or W or both. To see the e¤ect on W , we start from the de…nition of the share of quasi-rents accruing to the union, h(U; O)Q = L(W Q=R w), and employing the de…nition of quasi-rents, wN; the union wage W set by e¢cient bargaining can be written as: W = w + h(U; O) R wN L (2) By (2) W increases with the bargaining power h(U; O); lowering the share of quasirents accruing to the employer. Expression (2) makes it clear that if outsourcing su¢ciently reduces unions’ bargaining power, larger quasi-rents, through greater revenue or larger overall employment, may not lead to a higher union wage rate. Finally, as has been highlighted in the theoretical literature on e¢cient bargaining (e.g. McDonald and Solow (1981)) unions can also be concerned about employment. While we will not formally consider this possibility, it is also possible that unions may wish to negotiate for higher employment as well as higher wages. A. A Few Testable Empirical Implications on the Relationship between Outsourcing and Unionization. The review of the literature and the model of the relationship between unionization, outsourcing and the outcomes of union bargaining suggests …ve relationships to investigate. First, as any positive e¤ects on unionization depend on outsourcing increasing quasi-rents, we …rst test if there is any signi…cant positive relationship between the two. Assuming that outsourcing has no direct impact on the price of output p or on the non-unionized wage w, outsourcing may increase quasi-rents by improving productivity resulting in increased output, Y . Employment, N , may rise or fall depending on whether the outsourced labor inputs are complements or substitutes for in-house labor services. In either case, quasi-rents will rise. The reduced form equation determining industry quasi-rents, denoted Q, is summarized below: Q = f (px ; w; C; T; O) 9 (3) where, as suggested by the de…nition of quasi rents, we include px is the price of other inputs and w is the wage rate received if the workers were not unionized. Furthermore, economic theory suggests that quasi-rents also vary with competition, measured by the set of variables C, technological change over time, measured by the set of variables T , and outsourcing, measured by O.5 We also consider two more measures of quasi-rents — quasi-rents per plant and quasi-rents per employee. While both of these variables are functions of the same variables as industry quasi-rents, each enables focussing on a particular aspect of the problem. Quasi-rents per plant is probably a more appropriate measure for analyzing the e¤ect of competition, whereas quasi-rents per employee, qe , is a more appropriate measure of what is bargainable for the union. Second, to analyze the origins of any e¤ect of outsourcing on quasi-rents we analyze the relationships between outsourcing, productivity and employment. Average labor product, AP , is modeled, as suggested by economic theory, as a function of capital, K, the non-unionized wage and the price of other inputs. As a substantial empirical literature suggests productivity is in‡uenced by unionization, U , competition and technology so we include controls for these e¤ects. The reduced form equation determining average product is summarized below: AP = f (K; w; px ; U; C; T; O) (4) Similarly economic theory and previous empirical work suggests employment, N , is related to capital, the price of output and inputs, including the non-union wage, as well as unionization, competition and technology, so a similar reduced form is summarized below: N = f (K; p; w; px ; U; C; T; O) (5) where p is the output price. We will test if outsourcing is signi…cantly positively correlated with productivity or not, and whether there is a signi…cant negative or 5 The variables we use to measure competition, technological change and outsourcing are discussed in detail in the data section. 10 positive relationship with employment. This e¤ect of outsourcing on the industry employment level should not be apparent if outsourcing reduces unions’ power for reasons other than the substitutability hypothesis. In addition, we examine the e¤ects of outsourcing on unionization. First, we consider the union wage premium (W w)=w. If outsourcing does not a¤ect bargaining power, then the wage premium will be a function of quasi-rents per employee but not outsourcing. In addition, we also include measures of unionization, and competition, to control for the e¤ects these variables have on bargaining power. Note that if outsourcing has led to a reduction in unionization through increased organizing costs, then this will be picked up through the unionization variable. The general relationship is summarized below: ( W w w ) = f (qe ; U; C; O) (6) where qe is quasi-rents per employee. Finally, we examine if changes in outsourcing are directly correlated with unionization. To test this, we follow Magnani and Prentice (2003) and model unionization as a function of worker characteristics, domestic and international competition, quasi-rents per employee, technological change, with the addition of a measure of outsourcing. This reduced form relationship is summarized below: U = f (W C; C; qe ; T; O) (7) where W C is a set of worker characteristics identi…ed in the literature as being correlated with unionization. To summarize, the results of regressions based on equations (3) - (7) will provide some indication of the e¤ects outsourcing is having on the returns to unionization. The results of regression (3) -(5) suggest whether and how outsourcing is increasing the size of the pie to be bargained over - a source of a potential gain for the union. The results of regressions based on equation (6) and (7) suggest whether unions have been successful in bargaining for improvements in wages as a result of (or despite) outsourcing. The results of regression (7) indicate whether there is a direct rela11 tionship between unionization and outsourcing, controlling for other determinants of unionization. III. The Data and Identi…cation Issues. The dataset for this paper is an unbalanced panel of three-digit US manufacturing industries observed annually from 1973 to 1994 (excluding 1981-1982). It is collected by the authors, mostly from …ve sources: (1) NBER Productivity Database (Bartelsman and Gray, 1996); (2) Current Population Survey (CPS); (3) Employment and Earnings (Bureau of Labor Statistics, various); (4) County Business Patterns (U.S. Census Bureau) and (5) KLEMS data set (Bureau of Labor Statistics). The reader is referred to Magnani and Prentice (2003)) and Magnani and Prentice (2006) for details on the data. Here su¢ce to say that no other dataset exists that would enable an analysis of outsourcing and unionization across a wide set of manufacturing industries. For this purpose we have construct a single compatible data set from different sources using di¤erent industry classi…cation schemes that change over time. These problems are overcome in two steps. First, a single encompassing three-digit industry classi…cation - the Extended Census Industry Classi…cation (ECIC) - is constructed. Then, we construct, using concordances, aggregates compatible with the ECIC for variables over time across all sources, wherever possible.6 With the three-digit ECIC as the unit of observation, we obtain an unbalanced panel of 1439 observations for roughly 72 industries per year from 1973 to 1980 and 1983 to 1994. All variables used are de…ned, with descriptive statistics in Table 3. In what follows we discuss the construction of dependant variables and explanatory variables. A. Dependant Variables. In the various stages of the regression analysis we have used three sets of dependent variables that were not immediately available, or easily calculable, from the data collected, namely industry quasi-rents, union wage premiums and unionization. 6 For more details on the construction on the dataset see Appendix A of Magnani and Prentice (2003) or Magnani and Prentice (2001). 12 Industry Quasi-Rents. We use three versions of the quasi-rents variable. First, we estimate industry quasi-rents, QUASI-RENTS jt , for industry j at time t as follows: QU ASI REN T Sjt = P rice Shipments-(Cost of M aterials and Energy) (8) Employment w bjt where w cjt is the estimated non-union wage of the workers employed in industry j at time t:7 All of the variables used to calculate industry quasi-rents except w; b are obtained from the NBER Manufacturing Productivity Database (Bartelsman et al., 1996). We obtain the estimated industry speci…c non-union wage w b by running hourly wage regressions using CPS data of personal characteristics for all workers (both union and non-union members) Xit and a union coverage dummy variable (coverage)it : log(hourly wage)it = BXit + C(coverage)it + error termit (9) where B and C are vectors of coe¢cients of variables that refer to individual i at time t.8 We follow Card (1996) in the choice of the relevant individual speci…c variables Xit ; which are race, gender, education (completed some college, did not complete high school), experience, experience squared (some of these are described in more detail below). We also add occupational dummy variables among the explanatory variables as the empirical literature shows occupation has a large impact on unionb nonunion wage di¤erentials (Lewis, 1986).9 The vector of estimated coe¢cients B for personal characteristics is used to predict individual speci…c non-union wages b ijt ) w bijt = exp(BX (10) The estimated w bijt is the wage individual i employed in industry j at time t would earn were trade unions absent. We compute average non-union wages, by industry, 7 Note that this de…nition is consistent with Abowd (1989) and with the …rst de…nition of quasirents introduced by Abowd and Allain (1996). 8 Reported hourly wages are used in the majority of cases. When the CPS individual does not report this, the ratio between weekly earnings and average weekly hours is used. 9 Dummy variables for industry of employment are not included in the wage regression. This choice is motivated by the fact that industry premiums may re‡ect rents and we aim to estimate opportunity cost wages net of rents. 13 using the estimated individual w bijt of workers employed in each three-digit industry, weighted by their individual CPS weights. We estimate QUASI-RENTS PER PLANT by dividing QUASI-RENTS by NUMBER OF PLANTS. QUASI-RENTS PER EMPLOYEE, is estimated by dividing QUASI-RENTS by 10,000*EMPLOYMENT. Union Wage Premium. This is constructed as follows: UNION WAGE PREMIUM jt = Wjt w bjt w bjt (11) where Wj is the actual hourly union wage in industry j, calculated from the hourly wages reported for all workers covered by a union agreement from the CPS and w bj is the estimated non-union wage for the workers covered by union agreement in industry j as described in the previous subsection. Individual CPS weights are again used to compute industry average rates from individual rates. We also re-estimate separate union wage premiums, following the same steps, for two groups of workers, divided by occupation. The …rst group, referred to as the substitute workers, are those working in manufacturing whose occupations involve performing tasks similar to those that are outsourced such as cleaners and personal and protective service workers. The second group, referred to as the complement workers, are workers in all other occupations.10 As before industry average rates for each group are calculated for each year. Estimates based on less than 30 observations are discarded - which means about 200 industry-year combinations are lost - with more lost before 1983. Unionization. The unionization rate variable in industry j at time t, UNIONIZATIONjt , is de…ned as the ratio of the number of workers covered by a trade union agreement over the total industry work force. This variable, and worker characteristics, are calculated from the CPS. We use the outgoing rotations for 1983-1994 and the May CPS for 1973-1980. The year 1981 is excluded because union status was asked 10 The two digit occupations classi…ed as substitute workers are listed in the Appendix. 14 only of a CPS May quarter sample. Furthermore, 1982 is also excluded because there were no union status questions in the CPS that year. To compute both the numerator and denominator of this ratio we use CPS individual weights (weightijt ) for each individual i employed in industry j at time t. Thus U N ION IZAT IONjt = P ijt i (coverage) P i weightijt weightijt (12) where (coverage)ijt is a dummy variable that takes value one if individual i employed in industry j at time t is covered by union contract, and zero otherwise. Because the NBER CPS extract only includes the union coverage variable from 1978 (though collected from 1977), we estimate unionization for 1973 to 1977 by using the ratio of union coverage to union membership in 1978 to scale up annual union membership rates, estimated from CPS data.11 B. The Explanatory Variables. In this subsection, we discuss four sets of explanatory variables that require additional details on their construction: (1) Outsourcing; (2) Measures of Domestic and International Competition; (3) Technological Change; (4) Worker Characteristics. Outsourcing. Data on purchased services by manufacturing industries are drawn from the US Bureau of Labor Statistics KLEMS dataset. Outsourced services fall into one of the following categories: Communication, Finance and Insurance, Real Estate and Rental, Personal and Repair Services, excluding autos, Business Services, Auto Repair and Services, Amusements, Medical and Education Services, Government Enterprises. For the US Information Technology has historically been the most important input acquired (OECD, 1986). The empirical speci…cation uses a industry-level panel data set of 2-digit manufacturing industries (SIC 20–39) for which outsourced services are known between 1949 and 1999. The speci…c measure of outsourcing we use is the ratio of the value of purchased services to the value of output, denoted 11 There are very small di¤erences in between the unionization rates published by the BLS and ours due to slight di¤erences in sampling method (we select based on a valid union coverage variable). A table comparing the two over 1983-1997 is available from the authors upon request. 15 OUTSOURCING SHARE. By using this outsourcing measure merged with 3-digit manufacturing dependent variables, it is reasonable to assume that workers treat these measures of outsourcing as measures of risk of outsourcing rather than as resulting from endogenously determined strategies. This allows us to solve an important methodological di¢culty often encountered in this type of exercises, namely the issue of how to identify the direction of causation between union density and outsourcing. Measures of Competition. To measure the extent of domestic competition, we use the industry-speci…c number of plants, NUMBER OF PLANTS, available on an annual basis from the County Business Patterns (coded following the SIC). Economic theory suggests that one reason for oligopoly is economies of scale. This suggests an inverse relationship between NUMBER OF PLANTS and concentration and indeed the two measures are negatively correlated (with -0.45 for each year in which the concentration ratios are calculated). The extent of foreign competition is estimated using trade penetration variables, calculated from the NBER Trade database. The industry-speci…c import penetration variable, IMPORT SHARE, is de…ned as the ratio of imports over the sum of imports and domestic production by industry. The export variable, EXPORT SHARE, is the ratio between export and domestic production (Feenstra, 1996; Feenstra, 1997). Technological Change. We use the share of scientists and engineers over total industry employment, Share of scientists, to proxy for technological change. Berman, Bound and Machin (1998) report strong evidence of pervasive skill-biased technological change in OECD countries. Table 3 con…rms this view for the US manufacturing. Between 1973 and 1994 the share of scientists has risen by 109%. To compute the variable, SHARE OF SCIENTISTS, over the industry workforce, we use CPS weighted data on occupation. 16 Worker Characteristics. CPS data and weights are used to estimate the share of black workers employed in industry j at time t, SHARE OF BLACK, the average age of the industry work force, AVERAGE AGE, the share of female employees, SHARE OF WOMEN, the share of employees younger than 30, SHARE OF YOUNG, the share of employees older than 40, SHARE OF OLD, the share of the work force with a college degree, SHARE WITH COLLEGE, the share of high school dropouts, SHARE WITH < HIGH SCHOOL. Table 3 demonstrates that the main changes in worker characteristics are the increased share of female employees, the increased share of college educated workers and a slight rise in average age, which re‡ects the rise in the number of middle aged workers. C. Identi…cation Issues. To ensure identi…cation of our empirical speci…cations, we address three issues. First, in all speci…cations we attempt to control for two sources of unobservable variables - those that are industry speci…c (which we control through …xed e¤ects), and those that vary over time (using time dummies). To assess the role these play in identi…cation, for each equation we estimated four speci…cations, namely Speci…cation I (Explanatory variables only); Speci…cation II (Explanatory variables and industry dummy variables); Speci…cation III (Explanatory variables and year dummy variables); Speci…cation IV (Explanatory variables and year and industry dummy variables). In the end we report only the full set of speci…cations for the unionization regressions, but complete sets of results are available from the authors on request. All speci…cations in all tables are estimated using instrumental variables regression. Note that the regression on UNIONIZATION is estimated using weights so to yield an estimate for the average worker. Industry employment weights are calculated using CPS earnings weights for all individuals allocated to each industry. Secondly, because the extent of outsourcing is likely to be correlated with unob- 17 servable factors that determine quasi-rents, employment, and possibly, unionization and the union wage premium, we use the outsourcing share at the two-digit rather than the three digit level. Industry wide outsourcing will be correlated with the extent of outsourcing in a particular industry but less likely to be correlated with unobservable shocks to the particular industry. Thirdly, quasi-rents are unlikely to be exogenous when used in the unionization and union wage premium regressions (equations (6) and (7)). In addition, in equation (6), unobservable determinants of the wage premium are likely to be correlated with the extent of unionization. Furthermore, in the employment and productivity regressions (equations (4) and (5)), both UNIONIZATION and ESTIMATED NON-UNION WAGE are also likely to be correlated with unobservable shocks to textitEMPLOYMENT and PRODUCTIVITY. We deal with each of these issues by selecting appropriate instruments. We use the following instruments for quasi-rents: the prices of raw materials, energy and non-union labor and, following Ramey (1989), a set of instruments to capture macroeconomic driven ‡uctuations in rents: the real crude petroleum price, a set of political dummy variables (Republican government, …rst two years of government let by Democrats, …rst two years of government led by Republicans), real defense expenditure and real defense contracts. In addition we use import and export shares interacting with the prices of energy, raw materials and non-union labor. The same set of instruments, average worker characteristics, are used for instrumenting for UNIONIZATION in the Union wage premium regression and the estimated non-union wage in the employment and productivity regressions. IV. Results and Analysis. Table 3, which is constructed using the data we use for the analysis, reports a set of striking changes a¤ecting the US manufacturing sector from 1973 to 1994. First, there is the well known decline of unionization, from 46% to 20%. Simultaneously, the outsourcing share has risen from 7% in 1973 to 10% in 1994. The data also suggests striking increases in the extent of both domestic and international competition, 18 with a 37% increase in the number of plants, a 146% increase in the import share and a 97% increase in the export share. Despite increased competition, quasi-rents per employee have increased substantially - especially during 1983-1994, a fact that may be due to improved productivity growth after the 1970s.12 . Employment has remained roughly constant over this period, but, strikingly, there has been an increase in the unionization premium. With a substantial decline in the unionization rate, this could re‡ect either that only the strongest unions have survived over this period, or a re‡ection of the reorientation of unions to contribute to competitive advantage as suggested in Section II. A. Did Outsourcing Increase Quasi-rents? To address this question, we run three versions of equation (3) as described in section III.A. Selected results for the quasi-rents, quasi-rents per plant and quasirents per employee regressions are reported in columns two, three and four of Table 4, respectively. First, we review the e¤ects of the controls and then discuss the relationship between outsourcing and the di¤erent measures of quasi-rents. Across all three equations, the price variables and interactive terms are mostly insigni…cantly di¤erent from zero, with the exception of the estimated non-union wage for unionized workers. A statistically signi…cant and positive coe¢cient of this variable in the quasi-rents regression (and again in the quasi-rents per employee regression) is consistent with the view that competitively determined higher nonunionized wages are associated with higher productivity and therefore quasi-rents. While the measure of domestic competition, NUMBER OF PLANTS, is positively correlated with QUASI-RENTS, it signi…cantly reduces QUASI-RENTS PER EMPLOYEE in the entire range of variation of this variable. It is possible that the positive correlation could result from increased specialization across plants. Strikingly the measures of international competition are all statistically signi…cant. EXPORT SHARE is positively correlated with quasi-rents while IMPORT SHARE is 12 A decomposition of quasi-rents in its components along the lines suggested by equation (8) shows that in spite of rising costs, primarily due to rising prices of raw materials and energy, it is the value of output, and speci…cally the volume of production, that drives quasi-rents upward. 19 negatively correlated with quasi-rents — consistent with competitive importing. Finally the control for technological change, SHARE OF SCIENTISTS, is signi…cantly positively correlated with QUASI-RENTS and QUASI-RENTS PER EMPLOYEE, but not QUASI-RENTS PER PLANT. This result and the result for ESTIMATED NON-UNION WAGE suggests that industries with a higher share of research workers expand the number of plants (and possibly increase plant specialization) until the increase in quasi-rents is dissipated. Given that the above mentioned variables have the expected impact on quasirents we now turn to the outsourcing variable. Remarkably, Table 4 shows that OUTSOURCING SHARE is signi…cantly positively correlated with QUASI-RENTS and QUASI-RENTS PER PLANT, but insigni…cantly related to QUASI-RENTS PER EMPLOYEE. A possible explanation of such a result is that the increase in quasi-rents is dissipated through a rise in employment, an hypothesis that we examine in the next section. B. Did Outsourcing A¤ect Productivity and Employment? To address this question, we review the results of regressing PRODUCTIVITY and EMPLOYMENT on outsourcing and a set of controls, as speci…ed in equations (4) and (5). The results are summarized in Table 5. First, reviewing the controls reveals input price e¤ects across the two regressions are consistently as expected. For example, the sign on the ESTIMATED NON-UNION WAGE is negative (though insigni…cantly di¤erent from zero) and there is a signi…cant positive relationship between this price and PRODUCTIVITY, suggesting PRODUCTIVITY rises as employment falls. Although the raw materials and energy prices have opposite effects on PRODUCTIVITY, negative and positive respectively, PRICE OF RAW MATERIALS has the expected positive impact on EMPLOYMENT, implying raw materials and labor are substitutes. PRICE OF SHIPMENTS is surprisingly negatively correlated with EMPLOYMENT, a result that cannot be easily attributed to changes in the market structure as we control for domestic and foreign competition. The IMPORT SHARE is signi…cantly negatively correlated with both EMPLOY20 MENT and PRODUCTIVITY. EXPORT SHARE and PRODUCTIVITY (though not EMPLOYMENT ) are signi…cantly positively correlated. This is consistent with exporting taking place from high productivity industries, and imports gaining share from low productivity industries. As expected CAPITAL has positive signi…cant e¤ects on both EMPLOYMENT and PRODUCTIVITY. UNIONIZATION does not signi…cantly a¤ect EMPLOYMENT but does have a mildly signi…cant positive relationship with PRODUCTIVITY. This is consistent with the view according to which “Unionism per se is neither a plus nor a minus to productivity" (Freeman, 2005). As discussed in section II.A above, if there is complementarity rather than substitutability between in-house labor services and outsourced labor inputs, outsourcing should increase quasi-rents by increasing revenues rather than by reducing employment. Table 5 shows that OUTSOURCING SHARE has a signi…cant positive e¤ect on PRODUCTIVITY, consistent with the e¤ects on quasi-rents, but (more surprisingly) it also increases EMPLOYMENT. This result, not only denies support to the hypothesis of substitutability between the internal and the external labor services, but it also suggests that outsourcing enables the …rms to hire more labor. This raises the following important question. C. Did Outsourcing A¤ect the Returns to Unionization? The results for the e¤ects of outsourcing on the returns to unionization are contained in Table 6. First we consider the e¤ect on the union wage premium for all workers in the second column of this table. Strikingly, there is a statistically signi…cant positive correlation between the UNION WAGE PREMIUM and OUTSOURCING SHARE, after controlling for the size of the quasi-rents. We then re-estimate the equation using as a dependant variable UNION WAGE PREMIUM for substitute workers and complement workers. These results are reported in the third and fourth columns. Remarkably, and consistently with our expectations, a OUTSOURCING SHARE is signi…cantly positively correlated with UNION WAGE PREMIUM for the complementary workers but not for the substitute workers. Before discussing 21 this result in detail, we test the reliability of these results by discussing the e¤ects of the controls. For the all-workers regression, surprisingly UNIONIZATION has no signi…cant e¤ect on UNION WAGE PREMIUM. However for both the two sub-samples, UNIONIZATION does have an e¤ect. For substitute workers, at unionization rates below 38%, greater unionization is associated with lower wage premiums, suggesting the union trades o¤ wages for other bene…ts in bargaining. At unionization rates, above 38% UNIONIZATION has the expected positive e¤ect. For other workers, UNIONIZATION has a signi…cant positive relationship with the wage premium over almost the entire range of variation of the union variable, as the turning point occurs at a unionization rate of 90%. Notably, these …ndings are consistent with Blanch‡ower and Bryson (2004)’s …nding according to which there is no single union wage e¤ect, but a set of e¤ects that depend on worker and sectoral characteristics. QUASIRENTS PER EMPLOYEE have a positive e¤ect on the wage premium of substitute workers and a positive (at a declining rate) e¤ect on the wage premium of other workers. While in the aggregate IMPORT SHARE has no e¤ect, this variable is positively correlated with the union wage premium in the two subsamples. Such results could indicate the existence of a reverse causation e¤ect, from high union wage premiums to high import share or that this variable is picking up the greater returns from using imported inputs, in addition to the e¤ects on quasi-rents. It is important to note that the literature has found considerable variation in the relationship between union wage premiums and import competition (Blanch‡ower and Bryson, 2004).EXPORT SHARE is only signi…cantly positive at the aggregate level - although it is not for the two subsamples. NUMBER OF PLANTS also has a positive e¤ect on the union wage premium (except at very high plant numbers). Undoubtly, the most striking result is the positive relationship between OUTSOURCING SHARE and the union wage premiums, in both the all worker regression and the complement workers regression. There are two possible interpretations for this result. First, outsourcing reduces the bargaining power of unionized substi22 tute workers, but not for unionized complement workers. Second, outsourced labor increases the marginal productivity of complement unionized labor, who bargain for higher wages. Clearly, we cannot easily discriminate between these two, not necessarily, con‡icting explanations, because we do not observe the output or the productivity of these groups of workers. However, this set of results challenge us to address the following question. D. Did Outsourcing A¤ect Unionization? The results for the UNIONIZATION regressions, reported in Table 7, are organized in four columns, which refer to four di¤erent speci…cations of the regression equation. Interestingly, the full speci…cation, with industry …xed e¤ects and time dummy variables (Speci…cation IV in the …fth column), suggests that there is no signi…cant relationship between UNIONIZATION and OUTSOURCING SHARE. Before discussing this striking result, we …rst review the e¤ects of the industry and time dummies. Without industry and time dummies, the variables selected to re‡ect quasi-rents and international competition are signi…cantly related (in varying degrees) with UNIONIZATION. However, the inclusion of both sets of dummies rendered the coe¢cients of these variables and the coe¢cient of the outsourcing share insigni…cantly di¤erent from zero. For other variables, such as the NUMBER OF PLANTS, technological change and worker characteristics variables, the signs and (to a lesser extent) signi…cance vary across speci…cations. In speci…cation IV, the NUMBER OF PLANTS is negatively correlated with UNIONIZATION but the NUMBER OF PLANTS SQUARED is signi…cantly positively correlated with UNIONIZATION, suggesting a U shaped relationship. The signs of the worker characteristics variables, are in most cases, consistent with their e¤ects in individual unionization regressions (see Magnani and Prentice (2003) for details). SHARE OF SCIENTISTS has a signi…cantly negative e¤ect on UNIONIZATION, suggesting falling (or always low) unionization in the more research-intensive sectors of manufacturing. The main conclusion we draw from this set of results is that, although the preceding results showed some gains from outsourcing, these gains do not appear to 23 have been utilized to preserve unionization. Furthermore, it con…rms that, unlike the popular perception, outsourcing has not been behind the decline of unionization in U.S. manufacturing. The decline cannot, as argued in Magnani and Prentice (2003), be explained as a simple function of a small number of changes. Neither globalization or domestic outsourcing are unambiguous causes. V. Conclusions. While unions have been actively involved in the substantial reorganization of the workplace in the late twentieth century, the widespread evidence suggests that there has been union resistance to …rms adopting outsourcing. And this strategy would seem sensible if outsourced workers were substitutes rather than complementary to the permanent unionized labor. Using a newly compiled dataset, we demonstrate that outsourcing has been associated with higher quasi-rents per plant but not with higher quasi-rents per employee, which is suggestive of greater employment per plant. Outsourcing is (robustly) positively associated with a rise in the wage premium to union membership, a result consistent with outsourcing increasing the productivity of the unionized permanent workforce following the outsourcing of complementary services. Furthermore, we demonstrate no statistically signi…cant relationship between unionization and outsourcing — neither positive or negative. The substantial decline in unionization over the period of interest suggests some quali…cations to these conclusions. In particular, it could be the case that where unions survived, outsourcing was restricted to complementary services. Possibly elsewhere outsourcing substituted for the workers no longer represented by the union. The results on employment though suggest that outsourcing was positively associated with employment levels and changes in industry employment, which reduces the likelihood of the latter story. Our analysis of the gains to unionization suggest that union resistance to outsourcing may have been misplaced. While we fail to …nd any correlation between union and outsourcing, our results warrant that there is space for an alliance be- 24 tween unionized internal workers and external workers to share the quasirents that outsourcing increases. This would go in the direction that US unions seem already interested in taking. This is not to say that organizing the contingent workforce segment of the labor force is an easy task. In fact it may imply the full consideration of what material conditions numerical ‡exibility implies, namely income uncertainty, precariousness and skill ‡exibility, aspects that our study is unable to address. APPENDIX - Occupations selected as substitutes for outsourced services Table A lists all of the two-digit Census occupations classi…ed as, in manufacturing, being substitutes for outsourced services. All other occupations are classi…ed as being complementary. There are two features to note about this list. First, while we classi…ed all occupations in the CPS, for programming convenience, it is clear that only some occupations will be represented in our sample. This is because we only consider employees in manufacturing, where some of the listed occupations, such as college and university teachers, are absent. Second, note that in general, the 1980 classi…cation is more detailed than the 1970 classi…cation. However, the 1970 classi…cation contains several classes that appear to be more detailed that the 1980 match. For example, the 1970 class“Other professionals" includes the more detailed 1980 class of “College and university teachers". 25 References [1] [2] Abowd, J.M., 1989. The E¤ect of Wage Bargains on the Stock Market Value of the Firm, American Economic Review, 79, 774-809. [3] Abowd, J.M. and L. Allain, 1996. Compensation Structure and Product Market Competition, NBER Working Paper No. 5493. [4] Abowd, J.M. and H.S. Farber, 1990. Product Market Competition, Union Organizing Activity and Employer Resistance, NBER Working Paper No. 3353. [5] Abraham, K., 1996. Restructuring the Employment Relationship: The Growth of Market Mediated Work Arrangements. In: Abraham, K. and R. McKersie (Eds), New Development in the Labor Market: Toward a New Institutional Paradigm. Massachussets Institute of Technology Press, Cambridge, Mass, pp. 85-119. [6] Abraham, K. G. and S. K. Taylor, 1996. Firms’ Use of Outside Contractors: Theory and Evidence, Journal of Labor Economics, 14, 394-424. [7] AFL-CIO Committee 2000, 1999. Creating a new alliance with state federations and central labor councils, Draft proposal, May. [9] Autor, D., 2003. Outsourcing at Will: Unjust Dismissal Doctrine and the Growth of Temporary Help Employment, Journal of Labor Economics, 21, 1-42. [9] Azoulay, P., 2003. A Portfolio Approach to Transaction Costs Economics: Evidence from Clinical Development, mimeo, Columbia University Graduate School of Business. [10] Azoulay, P., 2004. Capturing Knowledge within and across Firm Boundaries: Evidence from Clinical Development, American Economic Review, 94, 15911612. [11] Bartel, A., S. Lach, and N. Sicherman, 2005. Outsourcing and Technological Change, NBER Working Paper No. 11158. [12] Bartelsman, E.J. and W. Gray, 1996. The NBER Manufacturing Productivity Database, NBER Technical Working Paper 205. [13] Berg, P., E. Applebaum, T. Bailey and A. Kalleberg, 1995. The Performance E¤ects of Modular Production in the Apparel Industry, Industrial and Labor Relations Review, 35, 356-74. [14] Berman, E., J. Bound and S. Machin, 1998. Implication of Skill-Biased Technological Change: International Evidence, Quarterly Journal of Economics, 103, 1245-1280. [15] Blanchard, O. J., J. F. Jimeno-Serrano, J. Andres, C. R. Bean, E. Malinvaud, A. Revenga, G. Saint-Paul, D. J. Snower, R. Solow, D. Taguas, and L. Tohana, 1995. Spanish Unemployment: Is There a Solution? Center for Economic Policy Research, London. [16] Blanch‡ower, D. G. and R. B. Freeman, 1992. Unionism in the United States and Other Advanced OECD Countries. In: Bognanno, Mario F and Morris M. Kleiner (Eds), Labor market institutions and the future role of unions. Blackwell, Oxford and Cambridge, pp. 56-79. 26 [17] Blanch‡ower, D. G. and A. Bryson, 2004. What E¤ect Do Unions Have on Wages Now and Would Freeman and Medo¤ Be Surprised?, Journal of Labor Research 25, 383-414. [18] Booth, A., 1995. The Economics of Trade Unions. Cambridge University Press, New York. [19] Brown, J. N., and O. Ashenfelter, 1986. Testing the E¢ciency of Employment Contracts, Journal of Political Economy, 94, S40-S87. [20] Cappelli, P., and D. Neumark, 2001. External Job Churning and Internal Job Flexibility, NBER Working Paper No. 8111. [21] Card, D., 1996, The E¤ect of Unions on the Structure of Wages: A Longitudinal Analysis. Econometrica, 64, 957-79. [22] Cobble, D. S., 1991. Organizing the Postindustrial Work Force: Lessons from the History of Waitress Unionism, Industrial and Labor Relations Review, 44, 419 - 436. [23] Current Population Surveys, 1979-1997. [machine-readable data …les]/ conducted by the Bureau of Census for the Bureau of Labor Statistics. Washington: Bureau of the Census [producer and distributor], 1998. Santa Monica, CA: Unicon Research Corporation [producer and distributor of CPS Utilities], 1998. [24] Drucker, P., 1995. Managing in a Time of Great Change. Truman Talley Books/ Dutton, New York. [25] Eaton, A. E. and P. B. Voos, 1989. The Ability of Unions to Adapt to Innovative Workplace Arrangements, American Economic Review, 79, 172-176. [26] Farber, H.S., 1990. The Decline of Unionization in the United States: What Can Be Learned from Recent Experience, Journal of Labor Economics, 8, S75-105. [27] Feenstra, R.C., 1996. NBER Trade Database, Disk 1: U.S. Imports, 1972-1994: Data and Concordances, NBER Working Paper No. 5515. [28] Feenstra, R.C., 1997. NBER Trade Database, Disk 3: U.S. Exports, 1972-1994, with State Exports and Other U.S. Data, NBER Working Paper No. 5990. [29] Freeman, R. B., 1986. The E¤ect of the Union Wage Di¤erential on Management Opposition and Union Organizing Success. American Economic Review, 76, 92-96. [30] Freeman, R. B., 2005. What Do Unions Do? The 2004-M Brane Stringtwister Edition, NBER Working Paper No. 11410. [31] Herzenberg, S., 2000. Reinventing the US Labor Movement, Inventing PostIndustrial Prosperity: A Progress Report, International Institute for Labour Studies, Labour and Society Programme, International Labour O¢ce, Geneva, in http://www.ilo.org/public/english/bureau/inst/download/dp11900.pdf. [32] Hiatt, J. P., and L. W. Jackson, 1997. Union Survival Strategies for the TwentyFirst Century, Journal of Labor Research, 18, 487-501. [33] Houseman, S. N., 2001. Why Employers Use Flexible Sta¢ng Arrangements: Evidence from an establishment survey, Industrial and Labor Relations Review, 55, 149-170. 27 [34] Ichniowski, C., K. Shaw and G. Prennushi, 1997. The Impact of Human Resource Practices on Productivity, American Economic Review, 87, 291-313. [35] Kalleberg, A. L., 2001, Organizing Flexibility: The ‡exible …rm in a new century, British Journal of Industrial Relations, 39, 479-504. [36] Kochan T. A., H. C. Katz and R. B. McKersie, 1994. The Transformation of the American Industrial Relations. Cornell University Press, Ithaca NY. [37] Lewis, H.G., 1986, Union Relative Wage E¤ects. In: O. Ashenfelter and R. Layard, (Eds), Handbook of Labor Economics, Vol. 2. North Holland, Amsterdam, pp. 1139-1181. [40] Lindbeck, A. and D. Snower, 2001. Centralized Bargaining and Reorganized Work: Are they Compatible?, European Economic Review, 45, 1851-1875. [39] Magnani, E., 2006. Technological di¤usion, the di¤usion of skill and the growth of outsourcing in US manufacturing, The Economics of Innovation and New Technology, forthcoming. [40] Magnani, E. and D. Prentice, 2001. The Unionization Trade and Employment Database, mimeo, The University of New South Wales. [41] Magnani, E., and D. Prentice, 2003. Did Globalization Reduce Unionization? Evidence from US Manufacturing, Labour Economics, 10, 705-726. [42] Magnani, E., and D. Prentice, 2004. Unionization and short-run output ‡exibility: Evidence from US Manufacturing, mimeo, UNSW, April 2004. [43] Magnani, E., and D. Prentice, 2006. Unionization and Input ‡exibility in U.S., 1973 - 1996, Industrial and Labor Relations Review, 59, 386-407. [44] McDonald, I. M., and R. M. Solow, 1981. Wage Bargaining and Employment, American Economic Review, 71, 896-908. [45] Miscimarra, P. A. and K. D. Schwartz, 1997. Frozen in Time — The NLRB, Outsourcing and Management Rights, Journal of Labor Research, 18, 561-580. [46] Novak, S. and S. Stern, 2003. Complementarity among Vertical Integration Decisions: Evidence from Automobile Product Development, Kellogg School of Management working paper, mimeo. [47] Organization for Economic Co-operation and Development, 1986. Flexibility in the Labor Market. OECD, Paris. [48] Osterman, P., 1988. Employment futures: Reorganization, dislocation and public policy. Oxford University Press, New York. [49] Osterman, P., 1999. Securing Prosperity. The American Labor Market: How It Has Changed and What To Do About It. Princeton University Press, Princeton. [50] Osterman, P., 2000. Work Reorganization in an Ear of Restructuring: Trends in Di¤usion and E¤ects on Employee Welfare. Industrial and Labor Relations Review, 53, 179 - 196. [51] Perry, C., 1997. Outsourcing and Union Power, Journal of Labor Research, 18, 521-534. [52] Piore, M. and C. Sabel, 1984. The Second Industrial Divide: Possibilities for Prosperity. Basic Books, New York. 28 [53] Quinn, J. B. and F. G. Hilmer, 1994. Strategic Outsourcing, Sloan Management Review, 37, 43-55. [54] Ramey, V., 1989. Inventories as Factors of Production and Economic Fluctuations, American Economic Review, 79, 338-354. [55] Segal, L. M. and D. G. Sullivan, 1997. The Growth of Temporary Services Work, Journal of Economic Perspectives, 11, 117-36. [56] Siegel, D., 1995. Errors of Measurement and the recent acceleration in manufacturing Productivity Growth, Journal of Productivity Analysis, 6, 297-320. [57] ten Raa, T. and E. N. Wol¤, 2001. Outsourcing of Services and the Productivity Recovery in U.S. Manufacturing in the 1980s and 1990s, Journal of Productivity Analysis, 16, 149-165. [58] Wood, A., 1994. North-South Trade, Employment and Inequality: Changing Fortunes in a Skill-Driven World. Clarendon University Press, Oxford. [59] Wooden, M., 2003. Long-Hours Working and Enterprise Bargaining, Agenda, 10, 259-271. 29 Table 1: Employment levels in the US business service sector and its sub-sectors in 1972 and 2000 and percentage rate of change between 1972 and 2000 Industry 1987 SIC Employment % Change 1972 2000 Business Services 73 1491 9858 561.2 Advertising 731 122 302 147.5 Credit Reporting and Collection 732 76 158 107.9 Mailing, Reproduction and 733 82 328 300.0 Stenography Services to buildings 734 336 994 195.8 Personnel Supply Services 736 214 3887 1716.4 Computer and Data Processing 737 107 2095 1857.9 Services Engineering and Architectural 871 339 1017 200.0 Services Accounting, Auditing and Book- 872 204 669 227.9 keeping Total Employment - Services 50007 106050 112.1 Total Employment - Nonfarm 73675 131759 78.8 Source: Data are in thousands and come from the Bureau of Labor Statistics. SIC= Standard Industrial Classi…cation code. Table 2: Cost share of purchased services in US manufacturing, 1949-1998 Year 1949 1957 1967 1972 1979 1982 1988 1992 1998 Cost 0.044 0.055 0.072 0.073 0.079 0.080 0.096 0.11 0.12 share Source: Bureau of Labor Statistics (KLEMS data set). 30 Table 3: The Data— Definitions Name of Variable De…nition UNIONIZATION Share of employees covered by collective bargaining agreement (CPS) SHARE OF BLACK Share of employees that are black (CPS) AVERAGE AGE Average age of all employees (CPS) SHARE OF WOMEN Share of female employees (CPS) SHARE OF YOUNG Share of employees with age less than 30 (CPS) SHARE OF OLD Share of employees with age more than 40 (CPS) SHARE WITH COLLEGE Share of employees with a college degree (CPS) SHARE WITH < HIGH Share of employees with less than high school degree (CPS) SCHOOL NUMBER OF PLANTS Number of plants (CBP) EXPORT SHARE Exports/domestic output (NBER Trade) IMPORT SHARE Imports/(domestic output + imports) (NBER Trade) SHARE OF SCIENTISTS Share of employees scientists or engineers (CPS) EMPLOYMENT Number of employees (in 10000s) (NBER) PRODUCTIVITY Real shipments divided by Employment (NBER) OUTSOURCING SHARE Ratio of values of purchased services to output (KLEMS) (a) CBP denotes County Business Patterns, NBER MP and Trade denotes NBER Manufacturing Productivity and Trade Databases respectively. CPS data is from outgoing rotations (1983-1994) and May CPS (1973-1980). 31 32 Table 3a: The Data— Summary Statistics(a) Whole Sample Mean Std.dev Min UNIONIZATION 0.31 0.17 0.02 QUASI-RENTS (b) 923.5 881.4 36.6 QUASI-RENTS 4.6 4.2 1.0 PER EMPLOYEE (c) QUASI-RENTS 67.1 136.5 1.51 PER PLANT (d) NUMBER OF PLANTS 4469.2 6841.1 76 EXPORT SHARE 0.09 0.08 0.0002 IMPORT SHARE 0.11 0.11 0.0003 SHARE OF SCIENTISTS 0.04 0.05 0 OUTSOURCING SHARE 0.08 0.04 0.02 EMPLOYMENT 240.8 237.2 19.5 ESTIMATED NON-UNION 758.0 267.2 259.7 WAGE (w) b UNION WAGE PREMIUM 0.10 0.16 -0.45 PRODUCTIVITY 155.2 169.79 5.6 SHARE OF WOMEN 0.30 0.16 0 AVERAGE AGE 38.3 2.1 30.1 SHARE WITH COLLEGE 0.29 0.14 0 Max 1 5762.0 49.5 1973 Only Mean Std.dev 0.46 0.17 778.5 693.0 3.6 2.3 1983 Only Mean Std.dev 0.31 0.14 828.4 762.1 4.2 3.0 1994 Only Mean Std.dev 0.20 0.12 1181.0 1198.1 6.2 5.9 1541.5 57.0 89.5 58.8 111.1 80.8 171.0 55628 0.41 0.71 0.37 0.50 1251.1 1580.3 3708.8 0.06 0.06 0.03 0.07 250.1 359.6 5052.7 0.06 0.06 0.04 0.03 243.8 44.1 4463.9 0.08 0.09 0.05 0.08 231.5 774.7 6813.7 0.07 0.08 0.05 0.04 229.4 97.5 5085.9 0.12 0.16 0.06 0.10 234.2 931.2 8216.9 0.09 0.15 0.07 0.05 238.5 77.8 1.03 1703.8 0.90 48.5 0.77 0.07 122.8 0.27 38.4 0.20 0.14 122.7 0.17 2.40 0.10 0.11 141.9 0.30 38.3 0.29 0.15 137.7 0.16 1.90 0.12 0.28 206.6 0.30 38.9 0.41 0.20 219.1 0.14 1.89 0.14 (a) For all variables there are 1439 observations. (b) Units for QUASI-RENTS is $10 million dollars (real) (c) Units for QUASI-RENTS PER EMPLOYEE is $10,000 per employee (real). (d) Units for QUASI-RENTS PER PLANT is $100,000 per plant (real). Table 4: Instrumental Variables Regressions of Quasi-rents Quasi-Rents Per Plant and Quasi-rents Per Employee Selected explanatory variables (a) Quasi-rents Quasi-rents Quasi-rents per plant per employee PRICE OF RAW MATERIALS -41.559 14.569 -0.326 (125.923) (23.979) (0.766) PRICE OF ENERGY -111.707 49.565 1.113 (160.317) (30.528) (0.975) ESTIMATED NON-UNION WAGE 0.405 -0.046 0.0019 (0.166) (0.032) (0.001) NUMBER OF PLANTS 0.081 -0.0032 -0.0003 (0.017) (0.0032) (0.0001) NUMBER OF PLANTS SQUARED 1.11e-07 6.27e-08 4.11e-09 (2.05e-07) (3.90e-08) (1.25e-09) IMPORT SHARE -1748.267 -353.118 -9.501 (537.015) (102.260) (3.268) EXPORT SHARE 2090.883 393.052 9.858 (498.449) (94.916) (3.033) SHARE OF SCIENTISTS 756.804 -25.927 70.344 (311.120) (59.245) (1.651) OUTSOURCING SHARE 1925.698 2367.559 1.171 (271.398) (51.681) (1.893) CONSTANT -1065.299 -110.281 0.649 (201.660) (38.401) (1.227) Number of observations 1439 1439 1439 F-Test 212.42 137.72 125.40 All regressions include time and industry …xed e¤ects, and use annual observations from 1973-1980 and 1983-1994. Standard errors are in parentheses. *, **, *** indicate statistical signi…cance at the 10%, 5% and 1%, respectively. Notes: (a) Interactions between import share and input prices as well as export share and input prices are included among the explanatory variables. These results are available upon request. 33 Table 5: Instrumental Variables Regressions of Productivity and Employment Productivity Employment ESTIMATED NON-UNION WAGE 0.11 -0.078 (0.051) (0.049) UNIONIZATION -451.759 -319.422 (290.803) (282.025) UNIONIZATION SQUARED 518.973 215.908 (305.208) (295.995) CAPITAL 0.003 0.003 (0.0005) (0.0005) PRICE OF SHIPMENTS -9.297 -31.778 (4.721) (4.579) PRICE OF RAW MATERIALS -144.831 47.260 (25.289) (24.719) PRICE OF ENERGY 73.957 -33.806 (30.470) (29.550) IMPORT SHARE -71.187 -317.105 (29.880) (28.978) EXPORT SHARE 107.798 10.787 (61.709) (59.845) NUMBER OF PLANTS -0.019 0.022 (0.004) ((0.004) NUMBER OF PLANTS SQUARED 1.65e-07 -9.21e-08 (4.38e-08) (4.25e-08) SHARE OF SCIENTISTS -34.498 22.637 (64.623) (62.673) OUTSOURCING SHARE 233.881 99.246 (57.781) (56.037) CONSTANT 297.64 -70.008 (56.499) (54.793) Number of observations 1439 1439 F-Test 302.37 449.66 All regressions include time and industry …xed e¤ects, and use annual observations from 1973-1980 and 1983-1994. Standard errors are in parentheses. *, **, *** indicate statistical signi…cance at the 10%, 5% and 1%, respectively. 34 Table 6: Instrumental Variables Estimation of Union Wage Premiums For all workers, substitute workers and complement workers Explanatory Variables All Substitute workers Complement workers UNIONIZATION 0.171 -0.689 1.849 (0.726) (0.309) (0.297) UNIONIZATION SQUARED 0.285 0.903 -1.179 (0.704) (0.316) (0.249) QUASI-RENTS PER 0.028 0.037 0.041 EMPLOYEE (0.022) (0.011) (0.010) QUASI-RENTS PER -0.002 -0.0004 -0.001 EMPLOYEE SQUARED (0.0006) (0.0003) (0.0003) IMPORT SHARE 0.077 0.305 0.158 (0.117) (0.083) (0.076) EXPORT SHARE 0.647 -0.047 0.062 (0.230) (0.143) (0.137) NUMBER OF PLANTS 4.96e-06 0.00002 0.00002 (0.00001) (7.56e-06) (7.00e-06) NUMBER OF PLANTS SQUARED -5.36e-11 -3.35e-10 -2.63e-10 (1.61e-10) (9.28e-11) (8.12e-11) OUTSOURCING SHARE 4.737 -0.633 1.295 (1.420) (0.688) (0.642) CONSTANT -0.105 0.073 -0.309 (0.258) (0.109) (0.130) Number of observations 1439 1217 1296 F-Test 9.90 19.47 36.76 All regressions include time and industry …xed e¤ects, and use annual observations from 1973-1980 and 1983-1994. Standard errors are in parentheses. *, **, *** indicate statistical signi…cance at the 10%, 5% and 1%, respectively. 35 Table 7: Weighted Instrumental Variables Estimation of Unionization Explanatory variables No Fixed Industry Fixed Year Fixed Industry and Year E¤ects E¤ects Only E¤ects Only Fixed E¤ects QUASI-RENTS PER 0.081 -0.033 -0.006 0.014 EMPLOYEE (0.020) (0.009) (0.023) (0.014) QUASI-RENTS PER -0.001 0.001 -0.001 -0.0003 EMPLOYEE SQUARED (0.0009) (0.0003) (0.0008) (0.0004) SHARE OF BLACK -0.409 -0.096 0.321 0.079 (0.120) (0.072) (0.181) (0.049) AVERAGE AGE 0.017 0.012 0.030 0.011 (0.015) (0.004) (0.012) (0.003) SHARE OF WOMEN -0.315 -0.140 -0.376 -0.117 (0.038) (0.053) (0.040) (0.047) SHARE OF YOUNG 0.550 0.150 -0.907 -0.063 (0.273) (0.072) (0.304) (0.067) SHARE OF OLD 0.843 0.032 -0.558 -0.219 (0.332) (0.084) (0.361) (0.067) SHARE WITH COLLEGE -1.273 -0.086 -0.229 -0.162 (0.218) (0.056) (0.267) (0.043) SHARE WITH < HIGH 0.124 0.393 -0.347 -0.052 SCHOOL (0.131) (0.045) (0.126) (0.055) NUMBER OF PLANTS -4.82e-07 -0.00001 -6.19e-06 -4.05e-06 (1.94e-06) (3.84e-06) (1.71e-06) (2.99e-06) NUMBER OF PLANTS 6.85e-11 1.07e-10 4.81e-11 5.68e-11 SQUARED (3.71e-11) (4.08e-11) (2.31e-11) (3.04e-11) IMPORT SHARE 0.273 -0.182 0.147 -0.005 (0.063) (0.059) (0.055) (0.046) EXPORT SHARE 0.026 0.153 -0.398 0.035 (0.084) (0.072) (0.098) (0.063) SHARE OF SCIENTISTS 1.111 -0.329 -0.610 -0.193 (0.318) (0.115) (0.402) (0.080) OUTSOURCING SHARE -0.238 -1.098 2.207 0.046 (0.707) (0.218) (0.663) (0.381) CONSTANT -0.555 -0.308 -0.035 -0.029 (0.609) (0.149) (0.455) (0.116) Number of observations 1439 1439 1439 1439 Annual observations from 1973-1980 and 1983-1994 are used. Standard errors in parentheses. *, **, *** indicate statistical signi…cance at the 10%, 5% and 1%, respectively. 36 Table A: Census Occupations classi…ed as equivalent to those in outsourced services. Two Digit Code 01 02 03 04 05 06 07 14 15 16 17 23 24 37 38 39 40 41 1970 Classi…cation Name Engineers Physicans etc. Health workers Teachers, excl college Engineering and science technicans Other professionals (salaried) Other professionals (self employed) Bookkeepers O¢ce machine operators Stenographers etc. Other clerical workers Mechanics - auto Mechanics excl auto Cleaning services Food service Health service Personal service Protective service 1980 Classi…cation Two Digit Code 03 04 05 06 07 08 09 10 11 12 13 14 15 23 24 25 26 27 28 29 30 31 32 33 37 Name Management related occupations Engineers Mathematical and computer scientists Natural scientists Health diagnosing occupations Health assessment and treating occupations College and university teachers Teachers, excl college and university Lawyers and judges Other professional speciality occupations Health technologists and technicans Engineering and science technicans Technicans excl health, engineering and science Secretaries etc Financial records, processing occupations Mail and message distributing Other administrative support occupations Private household service occupations Service occupations excl. protection and household Food service occupations Health service occupations Cleaning and building service occupations Personal service occupations Mechanics and repairers