Baqir

Districting and Government Overspending
Author(s): Reza Baqir

Source: Journal of Political Economy , Vol. 110, No. 6 (December 2002), pp. 1318-1354
Published by: The University of Chicago Press
Stable URL: https://www.jstor.org/stable/10.1086/342804
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access
to Journal of Political Economy
This content downloaded from

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
All use subject to https://about.jstor.org/terms
Districting and Government Overspending
Reza Baqir
International Monetary Fund
Theories of government spending driven by a common-pool problem

in the fiscal revenues pool predict that greater districting of a political
jurisdiction raises the scale of government. This paper presents evi-
dence on this and related predictions from a cross section of city
governments in the United States. The main finding is that, when
other plausible determinants of government spending are controlled
for, greater districting leads to a considerably greater scale of govern-
ment activity. The results also show that at-large electoral systems do
not, and forms of government that concentrate powers in the office
of the executive do, break this relationship.
I. Introduction
A central feature of the recent literature on the political economy of

government spending is the prominence given to the role of distributive
politics—the politics of policies that produce benefits concentrated to
a particular group of people and costs that are disbursed over the entire
political jurisdiction. Pork barrel projects are a prime example in which
federally financed projects produce benefits for one geographical com-
munity. As discussed extensively in Weingast, Shepsle, and Johnsen
(1981), such politics leads to a bias toward bigger project size and, in
general, bigger government. Legislators, when making spending pro-
I am grateful to Alberto Alesina, James Alt, Alan Auerbach, Gary Cox, Barry Eichengreen,
Caroline Hoxby, Steven Levitt, David Romer, Pablo Spiller, and seminar participants at
Berkeley, Cambridge, Harvard, Institute for International Economic Studies (Stockholm
University), International Monetary Fund, Massachusetts Institute of Technology, and the
World Bank for helpful comments and discussions. This research was supported in part
by a grant from the MacArthur Foundation. The views expressed in this paper are those
of the author and do not necessarily represent those of the affiliated institution.
[Journal of Political Economy, 2002, vol. 110, no. 6]

䉷 2002 by The University of Chicago. All rights reserved. 0022-3808/2002/11006-0001$10.00
1318

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1319
posals, fully value the benefits of public spending in their district but
internalize only a fraction of the taxation costs.1 Other recent papers
in which the same basic channel affects fiscal performance include Chari
and Cole (1993a, 1993b), Chari, Jones, and Marimon (1997), Hallerberg
and von Hagen (1999), and Velasco (1999).
A central prediction that emerges from this class of models is that
the greater the number of districts, the greater the overspending bias
and, hence, the greater the size of government. The purpose of this
paper is to test this and related predictions from a cross section of city
governments in the United States. These governments exhibit substan-
tial variation in both their fiscal outcomes and political structures and
constitute a good data set for testing theories relating political institu-
tions to fiscal outcomes. City political structures are, and have been,
difficult to change, and the resilience in these institutions bolsters our
faith in making causal interpretations from the regression results. These
data have the additional virtue that cities share a common national
institutional setup, and problems in inference arising from unquanti-
fiable historical and institutional factors, which in general plague cross-
country studies, are likely to be less. The central empirical findings can
be summarized as follows.
First, there is strong evidence that, when city population and other
plausible determinants of government size are controlled for, bigger city
councils are associated with considerably greater government expen-
ditures per capita. Extensive sensitivity analysis of the main results in-
dicates that the finding is robust to a variety of considerations. When
possible concerns of endogeneity are addressed by instrumenting for
council size using the size of the city council 30 years ago, the estimated
effect becomes stronger in magnitude. The findings are also robust to
alternative measures of the size of government: the share of total gov-
ernment expenditures in total city income and local government em-
ployment per capita. The results indicate an elasticity of 0.11 of gov-
ernment size with respect to the number of districts: one more political
district in the average city (of seven districts) is associated with a 1.6
percent increase in government expenditures per capita.2 In terms of
aggregate government spending, this amounts to an increase of about
$0.72 million in the average city. Given a median city budget in the
sample of $17.5 million and given that when cities consider redistricting
they generally consider changes of more than just one district, these
amount to nontrivial effects of political districting on government size.
Given that an overspending bias may arise in legislatures and, more
1
This is also referred to as the “common-pool” problem in the literature on environ-
mental economics. At a more general level, the overspending bias arises because of dis-
tricting and generalized taxation.
2
These estimates are taken from the discussion in Sec. IVD.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
1320 journal of political economy
important, that, ex post, each legislator may prefer a coordinated out-
come that entails less spending for all, a central question that emerges
is what political institutions, if any, we can put into place to achieve
better outcomes. I consider the effects of two such institutions that are
purported to limit the effect of districting on government size: (i) at-
large electoral systems, in which candidates for office are elected from
the entire jurisdiction; and (ii) forms of government that afford strong
powers to the office of the executive in the government—the office of
the city mayor in the case of cities. I am able to present evidence on
both of these mechanisms since there is substantial variation in both of
these institutions across cities in the sample and since evidence suggests
that these institutions are hard to change.
It is commonly believed that at-large systems, compared to district
systems, can curtail pork barrel–type spending by inducing council mem-
bers to treat the entire city as their constituency. If at-large council
members did cater to the good of the entire city, the asymmetry in
sharing the benefits and costs of public expenditures would be removed
and the overspending bias would disappear. The evidence I find con-
tradicts this commonly held view. At-large cities are not less susceptible
to pork barrel–type spending than district cities. Many cities in recent
years have adopted mixed electoral systems—in which some council
members are elected at large and some by district—in an effort to try
to capture the best elements of both district and at-large systems. Results
for these cities indicate that the effects of both district and at-large
council members are slightly exacerbated in mixed electoral systems.
One interpretation that these results admit is that in addition to the
externalities that council members impose on each other within a group,
there are also intergroup externalities that they fail to internalize, hence
leading to greater sensitivity of the size of government to the size of the
council.
The other institution I examine is the strong-mayor form of city gov-
ernment. Recent literature in the area of budget institutions—the study
of how the rules of the game surrounding the budgetary process affect
fiscal outcomes—indicates that political institutions that centralize
decision-making authority in one figure in the government, as, for in-
stance, in the president of a presidential government system, can reduce
the overspending bias.3 A strong executive can internalize the exter-
nalities inherent in legislators’ spending proposals and enforce fiscal
discipline on the legislature. City governments in the United States come
in two predominant forms: (i) the mayor-council form, in which the city
mayor is generally elected directly from the city population and is the
head of the executive branch of the government; and (ii) the council-
3
For a review of the empirical literature, see Alesina and Perotti (1999).

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1321
manager form, in which the legislative and executive functions of gov-
ernment are fused into the city council, which may appoint a city man-
ager to administer city services. The relevant difference between the
two is that the former concentrates powers in the city mayor, who cannot
be fired by the city council and can therefore exert independent influ-
ence on the city council. In addition, cities vary considerably in how
much power they concentrate in their mayors, for instance, by giving
them agenda-setting powers and powers to veto council legislation. Us-
ing data on the form of city government and on indicators of mayor
powers, I find suggestive evidence that city governments with strong
mayors, particularly those that afford their mayors veto powers, are able
to break the relationship between districting and the size of government
spending.
The paper is organized as follows. Section II briefly discusses the
theory and the existing empirical evidence and delineates the contri-
bution of this paper. Section III describes the data used in the paper.
Section IV presents the main results with respect to the impact of dis-
tricting on government size. Section V presents the results from inves-
tigating the role of (a) electoral systems and (b) forms of government
in mediating the relationship between districting and government size.
Section VI concludes the paper.
II. Related Literature
Weingast et al. (1981) provided one of the earlier formalizations of the

common-pool problem in the fiscal revenues pool. They considered the
problem in which representatives from legislative districts propose pub-
lic projects from the national tax revenue pool. One of their results was
that, given district tax shares that are nonincreasing in the number of
districts, “project scale for any district grows as the polity is more finely
partitioned into districts” (p. 654).4 The basic ingredients to Weingast
et al.’s overspending result were districting, a legislative norm of uni-
versalism (under which legislators follow a policy of mutual support,
making for a coalition of the whole), and generalized taxation.5 A sub-
sequent criticism of their approach, however, was the universalism as-
sumption. Although in other work (e.g., Shepsle and Weingast 1981)
they showed that the expected utility of a legislator running for reelec-
4
Recent papers that have the same common-pool mechanism at their heart include
Chari and Cole (1993a, 1993b), Chari et al. (1997), Hallerberg and von Hagen (1999),
and Velasco (1999).
5
They also have a different source of inefficiency, which they call the “politicization of
expenditures” (that some project costs are politically beneficial as the public outlays pro-
vide employment, etc.), but it is not necessary for the result on government scale and
districting.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
tion is higher when the legislature has a norm of universalism than in
the alternative environment of minimum winning coalitions, the lack
of a clear voting game made the theory less appealing.6 Subsequent
work by Baron and Ferejohn (1989) and Baron (1991) on legislative
bargaining helped to fill this gap. Their framework was later adopted
by Persson, Roland, and Tabellini (1997, 1998, 2000) to provide a rich
set of predictions relating political institutions to fiscal outcomes. Their
work showed that overspending is more likely in parliamentary systems
since members of the governing coalition are likely to have veto powers
over budget legislation, making the environment like universalism. By
comparison, presidential systems are expected to have less government
spending because they rely on a separation of powers and afford more
powers to an independent executive. Results in this paper shed light
on both these sets of results. The basic prediction on the effect of the
number of players is readily tested. The analogy between presidential
and parliamentary systems on the one hand and mayor-council and
council-manager forms of government on the other is used to present
evidence on the latter type of issues on separation of powers.
The existing empirical literature is based mostly on cross-country and
U.S. state data.7 The general approach in these papers is to examine
how constructed indices of the fragmentation of the budgetary process
affect fiscal outcomes.8 A common overall theme in this literature is
that institutions that centralize decision-making authority are associated
with smaller budget deficits and quicker fiscal adjustment to adverse
shocks. This paper contributes in the following ways. First, I use a sample
of local governments in the United States that allows me to greatly
increase the degrees of freedom and complements the set of findings
pertaining to countries and states. Second, I focus on providing evidence
on a central prediction of common-pool models that has not received
much attention: the effect of districting on government size. Of the two
central predictions that common-pool type models make, that (i) more
districts and (ii) a more decentralized legislative decision-making pro-
cess worsen the outcome, it is the latter that has received most of the
attention. One reason for this omission may be that direct tests of this
6
Of immediate relevance for this paper, Cox and Tutt (1984) provide micro evidence
from the study of the Los Angeles County Board of Supervisors for a norm of universalism
in the board’s budgetary decision making.
7
Relevant papers in the cross-county literature include Roubini and Sachs (1989a,
1989b), von Hagen and Harden (1994), Alesina et al. (1999), Hallerberg and von Hagen
(1999), Kontopoulos and Perotti (1999), and Bradbury and Crain (2001). For state-level
studies, see Alt and Lowry (1994), Poterba (1994), Bayoumi and Eichengreen (1995),
Gilligan and Matsusaka (1995, 2001), and Bohn and Inman (1996), among others.
8
In a different empirical approach, Inman and Fitts (1990) test the predictions of a
common-pool model using time-series data for U.S. federal expenditures and revenues
for the period 1795–1988. They do not directly test the relationship between districting
and government size, but their findings are consistent with those in this paper.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1323
relationship from cross-county or cross-state data are difficult since bud-
gets at the national level are drafted by committees or cabinets and then
submitted for approval to the full legislature. In the absence of an
explicit theoretical model of these institutions, it is unclear whether by
the number of districts we should mean the total strength of the leg-
islature, the number of members in the federal cabinet (or the number
of members of the relevant committee), the number of political parties
in the government, or some combination of the three.9 However, one
can readily exploit the variation in the size of city councils across U.S.
city governments to test this prediction. City councils are relatively cab-
inet- and committee-free. Hence, they offer a rather clean test of the
relationship between districting and government size. The sample of
city governments also has the advantage that these governments share
a common national institutional environment. There is likely less vari-
ation in unobserved institutions across cities in one country than across
countries in the world. Finally, I provide evidence on a question that
has not yet received much attention: How does a city’s electoral system
affect the extent of the overspending bias in the legislatures?10
III. Data
The basic specification used in the paper is to regress measures of
government size on the size of the city council and other determinants
of government expenditures. The data have been combined from dif-
ferent sources. Fiscal data are taken from the 1992 Census of Govern-
ments conducted by the Census Department. Demographic and income
data are taken from the 1990 Census of Population.11 Data on the po-
litical structure of city governments have been combined from (i) a
1990 survey of city governments conducted by an association of local
governments in the United States, the International City/County Man-
agement Association (ICMA), and (ii) the 1992 Census of Governments,
9
Kontopoulos and Perotti (1999) look at the issue of the number of players as well as
the fragmentation of the budgetary process in affecting fiscal outcomes. They measure
the number of players alternatively as the number of political parties in a coalition gov-
ernment and as the number of spending ministries in a government. Using panel data
on 20 OECD countries for the period 1960–95, they find that the number of players
matters for fiscal outcomes but get some variation in which measure matters: for the 1970s
they find that the number of spending ministries matters whereas for the 1980s the number
of parties matters. Their results also cannot be compared directly since their dependent
variable is the change in expenditures as opposed to the level of expenditures.
10
However, see recent work by Persson and Tabellini (2001) and Milesi-Ferretti, Perotti,
and Rostagno (2002) on national electoral systems and the level and composition of
government spending.
11
The fiscal and demographic data were obtained from the County and City Compendium
1993 (Slater-Hall Information Products, Washington, D.C.), a data product similar to the
Census Department’s County and City Databook 1994 but providing more comprehensive
coverage of U.S. cities.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
TABLE 1
Attempted and Approved Changes in City Government Structure, 1980–90
Attempted Approved
Percentage Percentage
Type of Change Number of Total Number of Total
Any change in structure of
government 230 16.2 114 8.0
Increase council size 40 2.8 21 1.5
Decrease council size 20 1.4 11 .8
Change to district electoral system 82 5.8 35 2.5
Change to a mixed electoral system 34 2.4 17 1.2
Change the mix between the at-
large and district council
members 15 1.1 7 .5
Change the form of government 37 2.6 11 .8
Source.—ICMA.
Note.—Total number of cities in the sample is 1,420. The sample is smaller than that in table 4 below because of
data availability on questions of proposed and approved changes in city government structure.
Government Organization File. The latter source provides information

for fewer variables but for a greater number of cities.
Size of the city council is measured as the number of officials elected
to the chief governing body of the government (Csize) and varies from
a minimum of three to a maximum of 50 in the data set. Central to
the empirical analysis is the assumption that council size is costly to
change. Both theoretical and empirical arguments support this as-
sumption. Theoretically, a change in the number of districts almost
always has to be approved by the incumbent legislators. The redistricting
inherent in such a change reapportions the incumbent council mem-
bers’ constituencies and introduces uncertainty in their reelection pros-
pects. In their influential study of the world’s electoral systems, Taa-
gepera and Shugart (1989) convey this point well when they discuss the
resilience in electoral laws: “Reforms usually require the approval of
current assembly members. But these are by definition the very people
whom the current electoral system has served well. Why should they
want to change a system that got them elected?” (p. 5). At a practical
level, there are significant costs involved in changing a political insti-
tution such as the size of the council. Typically the process involves a
proposal brought forward either directly by the voters if the city has a
provision for initiative or by the council, extensive discussion of the
merits and demerits of change in the size of the council and the likely
impact of a change on representation (with a commission being ap-
pointed sometimes to consider the issue at length), and approval by
the council or the city population (by a referendum) or both.
Direct evidence also shows that the size of the city council is difficult
to change. Table 1 summarizes the information from the ICMA data on

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1325
TABLE 2
Correlation between Measures of Government Size (Np1,987)
Government Expenditure Government Expenditure as

per Capita Share of Total City Income
Government ex-
penditure as
share of total
city income .886
Government
employment
per capita .774 .781
the number of attempted and successful changes in city political struc-

ture in the 10 years preceding 1990. In the table, any change in structure
of government refers to any kind of reform that the council considers
and includes measures unrelated to the size of the council and electoral
laws. More relevantly, table 1 shows very few attempts (4.2 percent of
the sample) to change the council size, and of these, there is roughly
a one-half probability of success. Most of the variation in council size
can therefore be assumed to be coming from historical reasons. The
table also shows that the electoral system and the form of government
display considerable inertia.
I measure the size of government in three alternative ways: (i) gov-
ernment spending per capita (Exppc), (ii) government spending as a
share of total city income (Expsh), and (iii) government employment
per capita (govempl). All three measures have been used in the em-
pirical literature on the size of government. Consistent findings across
these measures would indicate a general relationship, not sensitive to
a particular measure of the size of government. These three measures
are quite highly correlated as shown in table 2.
A number of other plausible determinants of city spending are used
as control variables. City size, as measured by population, is used to
address economies of scale considerations and is considered in detail
in Section IV. I also control for the racial heterogeneity of the city
population using an index of racial fragmentation (ethnic):
ethnic p 1 ⫺ 冘i
si2,
where si denotes the share of population of race i in the total city

population, and
i 苸 {white, black, American Indian, Asian and Pacific Islander, other}.
The index ranges from zero (complete homogeneity) to one (complete
heterogeneity) and can be interpreted as the probability that a randomly
selected person from the city population will belong to a different racial

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
group than another randomly selected person. The racial categories are
taken from the 1990 census. In the sensitivity analysis, I also control for
a similar measure of heterogeneity for the council, using data on council
members by race. The additional control variables are per capita income
(incomepc), educational attainment as measured by the percentage of
population with a bachelor of arts or higher degree (BAgrad), and
income inequality in the city as measured by the ratio of the mean to
median household income (MMI90).12 Income and educational attain-
ment are likely determinants of the demand for public services. The
inequality variable is included since the size of government may respond
to redistributive pressure arising out of income inequality. Table 3 shows
the summary statistics on all the variables used in the study.
It is useful to consider what drives the variation of council size in the
sample. Theoretically, the most obvious determinant is city population.
Bigger jurisdictions should, and do, have bigger councils. Regressing
council size on city population (in millions) yields the following equa-
tion (standard errors are in parentheses):
Csize p 6.62 ⫹ 5.36Pop90, R 2 p .15, N p 1,972.
(0.07) (0 .29)
The slope coefficient indicates that although bigger cities have bigger
councils, the effect is considerably small in magnitude. An increase in
the city population from 10,000 to 100,000 would be associated with an
increase in the council size from 6.7 to 7.2—a fairly small effect.13 The
small magnitude of effect is consistent with the evidence on the infre-
quency of changes in council size: over time, while city populations may
have changed considerably, council size has changed less frequently,
leading to a small slope coefficient in the 1990 cross section. The other
important sample correlates of council size are the state in which the
city is located and the city’s ethnic and income heterogeneity. State
matters because city councils derive their authority from state govern-
ments, which vary in their laws governing local governments. Running
the regression above with a complete list of state indicator variables
yields an adjusted R 2 of .41, whereas the estimated coefficient on pop-
ulation remains virtually unchanged (and highly significant). To the
extent that preferences for public services are correlated along ethnic
12
For a subset of the sample I had data on the income-based Gini coefficient and on
the city unemployment rate. The findings on inequality are robust to either measure used.
The unemployment variable was used to control for government spending responding to
unemployment for standard Keynesian reasons. The results on council size were robust
to including the unemployment rate in the regression.
13
The same equation estimated in log-log form yields an elasticity of council size with
respect to city population of 0.11. Taagepera and Shugart (1989, chap. 5) estimate a similar
equation for a cross section of countries in 1985 and report an elasticity of legislature size
(lower house) with respect to country population of 0.33.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
TABLE 3
Summary Statistics
193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
Standard Number of
Variable Units Minimum Maximum Median Mean Deviation Observations
City government expenditures per capita $1,000 per capita .020 7.836 .641 .791 .539 1,991
City government expenditures, as a percentage

of total city income Percentage .078 44.660 4.933 5.973 4.123 1,991
City government employees per capita Employees per 1,000 population .429 98.873 9.746 12.101 8.835 1,996
Council size Number of people 3 50 6 6.859 2.888 2,342

Council size, 1960 Number of people 5 9 7 6.505 1.517 465
Ethnic Fraction .004 .730 .187 .235 .173 3,146
Council-ethnic Fraction 0 .720 0 .122 .180 1,779
City population Number of people 10,005 7,322,564 21,099 45,540 173,103 3,146
Income per capita $10,000 per capita .438 6.330 1.386 1.528 .597 3,146
%BAgrad Fraction .007 .909 .188 .225 .129 3,146
Ratio of mean to median household income Ratio .986 4.777 1.213 1.248 .185 3,146
Council members elected by district Number of people 0 50 0 2.856 4.048 2,342
Council members elected at large Number of people 0 16 5 4.003 2.731 2,342
Mayor-council form of government Indicator variable 0 1 0 .378 .485 1,696
Mayor-council form of government, 1960 Indicator variable 0 1 0 .359 .480 473
Mayor elected directly from city population Indicator variable 0 1 1 .775 .417 1,696
Mayor proposed budget to council Indicator variable 0 1 0 .173 .378 1,696
Mayor appoints department heads Indicator variable 0 1 0 .232 .422 1,696
Mayor can veto council-passed measures Indicator variable 0 1 0 .345 .475 1,696
Mayor can veto specific items of appropriations Indicator variable 0 1 0 .086 .281 1,696
Source.—The 1992 Census of Governments; 1990 Census of Population; 1992 Census of Governments, Government Organization File (all Census Department); a survey of city
governments conducted by the ICMA; and the County and City Compendium 1993 (Slater-Hall Information Products, Washington, DC).
Note.—Cross-sectional city government data pertain to 1990, unless otherwise stated. Ethnic is an index of racial heterogeneity for the city population, ranging from zero (least
heterogeneity) to one (most heterogeneity). Its construction is described in the text. Council-ethnic is a similar measure for the city council using data on council members by
race. %BAgrad is the percentage population over 25 with a college or higher educational degree. For the indicator variables, the variable equals one (and zero otherwise) if the
corresponding statement is true. The last five variables on the powers of mayors are classified further in table 9 below. The text provides further details on all variables.
and income lines, we should expect greater demand for political rep-
resentation in more heterogeneous jurisdictions for a given population.
Regressing council size on population and two measures of heteroge-
neity (ethnic, as defined above, and income inequality, measured by the
ratio of the mean to median income) as well as a complete list of state
indicator variables gives
Csize p 5.87 ⫹ 5.22Pop90 ⫹ 1.64ethnic ⫹ 1.02MMI,

(0.25) (0.39) (0.40)
R̄2 p .42, N p 1,972.
The regressions for government size reported below control for these
measures of income and ethnic heterogeneity, and the coefficients on
these variables can be interpreted as capturing their direct impact on
government size, when the effects that may go through council size are
controlled for.
IV. Results I: Districting and Government Size

Table 4 presents the results of ordinary least squares (OLS) regressions
for the size of government on the size of the city council and other
variables. The three measures of government size are used in log form
because of the presence of large outliers in each of these series and
because in this form the coefficients on the council size variable can
conveniently be interpreted as the elasticity of government size with
respect to the number of electoral districts. The first specification in-
cludes only city population as a control variable. Subsequent specifi-
cations play close attention to the following sets of factors: (i) other
plausible determinants of government size, (ii) the nonlinear effects of
city population on government size, and (iii) the presence of state-
specific effects correlated with both council size and government size.
I discuss each in turn below.
The first specification indicates that the magnitude of effect of council
size on government size is close across the three measures of government
size. When the control variables are added in the second specification,
these magnitudes remain relatively unchanged, indicating, in particular,
that the council size effect is not going through two other possible
political factors: racial heterogeneity and income inequality. Coefficients
on the control variables indicate that government size increases with
the racial heterogeneity of the city and the ratio of mean to median
income of the city. The first of these two findings is consistent with the
results in Alesina, Baqir, and Easterly (1999) and is not explored further
here except to note that since the index of racial fragmentation goes
up with the effective number of racial groups in the population, it is

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1329
important to determine whether bigger councils might simply be prox-
ying for a more heterogeneous population. Results discussed in sub-
section E below indicate that this is not the case. The positive coefficient
on the measure of income inequality interestingly relates to a long-
standing literature on the relationship between income inequality and
redistributive spending.14 The coefficient on per capita income is con-
sistent with previous studies of local public goods that find the demand
for local public services to be income inelastic.15 Plausible coefficients
on the control variables suggest that the empirical model is not grossly
misspecified.
A. City Size
The variable with the greatest statistical significance is the logarithm of

the city population. Log form was used to capture the presence of
outliers in the series. Since city size is strongly correlated with govern-
ment expenditures and council size, the third specification splits the
1990 population into five quintiles and allows for a different slope co-
efficient for each interval. The results are illuminating. In small and
medium-sized cities, per capita government expenditures decline with
city population, consistent with the presence of economies of scale. An
increase of 10,000 people in the city population for cities at the smallest
quintile is associated with approximately an 18 percent decrease in ex-
penditures per capita—an effect of considerable magnitude. At higher
population quintiles the estimated effect becomes weaker in magnitude,
eventually turning positive for the very biggest cities, suggesting the
yielding of economies of scale to diseconomies.16 The pattern of coef-
ficients suggests a U-shaped relationship between the size of government
and city population. Figure 1, which makes a finer division of city sizes
in the sample, confirms this. The vertical bars show average per capita
government expenditures by decile of the city population.17 The con-
nected line shows the average residuals by population decile from a
regression of per capita expenditures on all the other variables included
in table 4. Both series show a similar pattern and suggest that economies
of scale in government spending eventually yield to diseconomies of
14
See Benabou (1996) for an excellent review of this literature. This relationship is not
explored further here since it is not the focus of the paper, but the findings on this variable
are consistent in all the specifications.
15
See, e.g., the review provided by Mueller (1987).
16
The F-tests for the equality of coefficients across the five population quintiles reject
at p-values of less than .001.
17
The pattern is not markedly different when median per capita expenditure for the
interval is used instead of the mean. Figures for the other two measures of government
size yield very similar results.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
TABLE 4
OLS Regressions for Government Size
(1) (2) (3) (4)

A. ln(Government Expenditures per Capita)
ln(council size) .2760*** .3021*** .3203*** .1127***
(.0384) (.0383) (.0384) (.0373)
ln(city population) .1515*** .1307***
(.0133) (.0139)
Ethnic .1920** .2550*** .5099***
(.079) (.0783) (.0911)
Income per capita .2272*** .2339*** .1631***
(.0355) (.0359) (.0349)
%BAgrad ⫺.6060*** ⫺.5940*** ⫺.3898***
(.1537) (.1557) (.1425)
Mean/median income .6613*** .6543*** .8524***
(.0921) (.0939) (.0957)
City population:
1st quintile ⫺1.8201*** ⫺1.2740***
(.3143) (.2821)
2d quintile ⫺1.2652*** ⫺.9433***
(.2057) (.1796)
3d quintile ⫺.7196*** ⫺.5609***
(.1425) (.122)
4th quintile ⫺.2002** ⫺.1842***
(.0808) (.0691)
5th quintile .0147*** .0188***
(.0047) (.0049)
Constant ⫺2.4819*** ⫺3.3969*** ⫺1.9632*** ⫺.0401
(.1436) (.1825) (.139) (.1633)
State fixed effects no no no yes
Observations 1,972 1,972 1,972 1,972
Adjusted R 2 .10 .15 .14 .39
Standard error of
regression .536 .523 .525 .442
B. ln(Government Expenditures as a Share of Income)
ln(council size) .2855*** .2746*** .2903*** .1100***
(.0393) (.0369) (.037) (.0367)
ln(city population) .1225*** .1058***
(.014) (.0136)
Ethnic .2791*** .3357*** .5955***
(.0772) (.0767) (.0901)
Income per capita ⫺.2913*** ⫺.2852*** ⫺.3332***
(.0338) (.0339) (.0343)
%BAgrad ⫺.7339*** ⫺.7187*** ⫺.5896***
(.1497) (.1515) (.1415)
Mean/median income .9025*** .8921*** 1.0982***
(.0917) (.0937) (.097)
City population:
1st quintile ⫺1.3642*** ⫺.7671***
(.3061) (.2778)
2d quintile ⫺1.0219*** ⫺.6553***
(.201) (.1771)
3d quintile ⫺.6137*** ⫺.4491***
(.1377) (.1194)
4th quintile ⫺.1694** ⫺.1437**
(.0789) (.0679)
1330

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
TABLE 4
(Continued )
(1) (2) (3) (4)

5th quintile .0123*** .0154***
(.0047) (.0045)
Constant ⫺.193 ⫺.6216*** .5390*** 2.2141***
(.1494) (.1804) (.138) (.1669)
Observations 1,972 1,972 1,972 1,972
Adjusted R 2 .07 .28 .28 .45
Standard error of
regression .578 .508 .510 .434
C. ln(Government Employment per Capita)
ln(council size) .5082*** .4998*** .4950*** .1322***
(.042) (.0407) (.0408) (.0374)
ln(city population) .0345** .0226
(.0146) (.0146)
Ethnic .0702 .0678 .3166***
(.0889) (.0887) (.0892)
Income per capita ⫺.0375 ⫺.0358 .0081
(.0376) (.0375) (.037)
%BAgrad ⫺.5437*** ⫺.5205*** ⫺.4514***
(.174) (.1749) (.1561)
Mean/median income 1.4238*** 1.4157*** 1.1636***
(.1079) (.1095) (.1047)
City population:
1st quintile ⫺.357 ⫺.5887**
(.3331) (.2865)
2d quintile ⫺.2609 ⫺.4923***
(.2278) (.1883)
3d quintile ⫺.3917** ⫺.4735***
(.152) (.1179)
4th quintile ⫺.1462 ⫺.1513**
(.0941) (.0714)
5th quintile .0032 .0159***
(.0054) (.003)
Constant .9901*** ⫺.5121*** ⫺.2178 2.1450***
(.1544) (.1984) (.1573) (.1772)
Observations 1,977 1,977 1,968 1,968
Adjusted R 2 .08 .20 .20 .51
Standard error of
regression .590 .551 .552 .432
Note.—The dependent variable for each measure of government size is named in each panel. The first specification
for each measure includes only council size and the natural logarithm of the city population. The last specification for
each measure includes all listed covariates and a complete set of state fixed effects. Ethnic is an index of racial
heterogeneity for the city population: higher values represent greater heterogeneity. %BAgrad is the percentage of
population over 25 with a college or higher degree. All variables are described in table 3. Population quintiles have
population data expressed in 100,000s. The data set is a 1990 cross section of city governments in the United States,
and sources are described in Sec. III in the text. Robust standard errors are in parentheses.
** Significant at the 5 percent level.
*** Significant at the 1 percent level.
1331

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
Fig. 1.—Government size by city size
scale. The estimated relationship with respect to council size is not

affected when we allow nonlinear effects of population size.
B. State-Specific Effects
Since state-specific factors, such as differing degrees of state fiscal de-
centralization, are likely to affect both local government spending and
local political structure, the fourth specification in table 4 controls for
a complete set of state fixed effects. The estimated coefficient on council
size drops to little over a third of its value. Figure 2 plots the logarithm
of the median per capita city expenditure in a state against the median
council size in the state and shows a strong positive correlation.18 Note,
however, the presence of influential observations: Washington, D.C.
(with one local government) and the New England states are clustered
on the upper-right side of the figure (fig. 3 excludes these states). Wash-
ington, being the nation’s capital and having the highest level of ex-
penditures per capita, can deserve special treatment. The New England
states are the oldest states in the United States with a history of liberal
and very democratic local government traditions. Some cities in the
New England states also use the town meeting form of local government,
which is unique to these states. A closer examination shows that the
reduction in the council size coefficient in the fourth specification is
18
Figures using the other two measures of government size look very similar.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1333
Fig. 2.—Council size and government expenditures: all states
accounted for by the New England effect. Running the fourth specifi-
cation with only the indicators for the New England states gives close
to the same reduction in the coefficient on council size.19 The results
are presented in row 1a of table 5. An F-test for the equality of the
coefficients on the six state indicators does not reject at conventional
levels, indicating that we could include one indicator variable for New
England states. Row 1b of table 5, which reports the coefficients and
standard errors on the council size variable when only this indicator is
included in the equation, confirms this. Does the relationship between
council size and government expenditures survive when we look only
at the non–New England states? The subsequent two rows in table 5
show that it does. Row 1c reports the coefficients when only non–New
England states are included in the sample and state indicators are not
included. The estimated coefficient for each measure of government
size is very close to the full sample regression with all state indicators
(col. 4 of table 4). Moreover, inclusion of state indicators in this
non–New England sample (row 1d of table 5) does not alter the coef-
ficient much for two of the three measures of government size, consis-
tent with the discussion above that the original reduction in the coef-
19
The New England states are Connecticut, Maine, Massachusetts, New Hampshire,
Rhode Island, and Vermont. The same holds if we include in addition an indicator for
Washington, D.C.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
Fig. 3.—Council size and government expenditures: excluding Washington, D.C., and
New England.
ficient was coming from the New England states.20 Table 5 also pulls
together results from other specification tests, which are discussed in
detail below.
C. Reverse Causality
Evidence presented in Section III suggested that council size is costly
to change, which should limit concerns that results are contaminated
because of reverse causation. However, it is possible that over long pe-
riods, council size may have adjusted to incorporate spending prefer-
ences of cities.21 To address these concerns, I present results with in-
strumental variables, using the size of the city council in 1960 as an
instrument.22 Since we are going a fairly long period back in time, this
variable is likely exogenous to the spending decisions in 1990. However,
20
There are only 92 observations if we look at the New England sample. Regressions
for government size in this sample do not give any significant variable except racial het-
erogeneity. Note that even population is not significant, which indicates a not very good
fit of the model.
21
It is, though, not clear why greater government spending would require bigger coun-
cils. The council refers to the legislative function in government, whereas government
programs typically fall under the executive branch. Getting more government programs
is more likely to mean more government employees than more legislators.
22
The data come from Aiken and Alford (1972) and were obtained electronically from
the web site of the Inter-university Consortium for Political and Social Research, http://
www.icpsr.umich.edu, study 0028.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
TABLE 5
Sensitivity Analysis
Coefficient on ln(Council Size) When the De-
pendent Variable Is:
ln(Expenditures ln(Government
ln(Expenditure as Share of Employment
per Capita) City Income) per Capita)
0. Baseline coefficient .1127 .1100 .1322
(.037) (.037) (.037)
1. State-specific effects:
a. Indicators only for New England states .1284 .1115 .2982
(.039) (.038) (.041)
b. One indicator for New England .1323 .1147 .2937
(.038) (.038) (.040)
c. Non–New England sample, without state .1410 .1224 .3218
indicators (.041) (.040) (.043)
d. Non–New England sample, with state .1294 .1255 .1496
indicators (.043) (.042) (.042))
e. State share in total revenue, with all state .1200 .1169 .1366
indicators (.038) (.037) (.038)
2. Population growth, 1980–90 .0971 .0972 .1051
(.038) (.037) (.037)
3. Effective number of ethnic groups .1088 .1114 .1378
(.049) (.047) (.049)
4. Ethnic heterogeneity:
a. Ethnic ≥ median (.20) .1374 .1275 .1470
(.039) (.038) (.039)
b. Ethnic ! median .0879 .0924 .1173
(.039) (.038) (.039)
5. Big councils vs. small councils:
a. Council size 1 9 .1361 .1338 .1510
(.04) (.039) (.041)
b. Council size ≤ 9 .1718 .1703 .1797
(.049) (.048) (.052)
6. Big cities vs. small cities:
a. City population ≥ median (25,555) .1294 .1266 .1381
(.038) (.037) (.038)
b. City population ! median .0756a .0730a .1191
(.040) (.039) (.04)
7. Central vs. suburban cities (includes indica- .1464 .1402 .1955
tors for central and suburban cities) (.047) (.045) (.048)
8. Population density: controls for .1114 .1084 .1308
ln(population density) (.037) (.037) (.038)
9. Percentage voting for Democratic president .1144 .1102 .1340
(.039) (.039) (.039)
Note.— The table reports results from variations on the fourth specification in table 4 for each measure of government
size. Row 0 reproduces the original specification for ease of reference. Only the coefficients (and robust standard errors
in parentheses) on the council size are reported to conserve space. Results on the other variables are discussed in Sec.
IVE. Each regression additionally includes all variables of table 4, including population quintiles and state indicators.
Effective number of ethnic groups equals the reciprocal of one minus the racial heterogeneity variable for the city
council. For rows 4–6, the council size variable is split according to the conditions in the two subrows. Numbers of
observations range from 1,455 to 1,972 depending on data availability for the additional variables.
a
p-value !.06 (for all other estimated coefficients associated p-values ! .05).

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
it is available for only 465 cities of the OLS sample. Since there is a
considerable change in the sample size, table 6 first reports the OLS
and then the two-state least-squares (2SLS) results (both with and with-
out the state indicators). The OLS results show that the statistically
significant estimated coefficients (on council size, racial heterogeneity,
per capita income, and the ratio of mean to median income) are quite
close to the estimates in the full sample of table 4. Thus the 465-
observation sample is representative of the full sample. When we in-
strument for the 1990 value of council size, the estimated coefficient
increases in magnitude. If reverse causality was contaminating the OLS
results, we would have expected a reduction in the coefficient with the
instrumental variable specification. One explanation that can account
for the increase is that, over time, cities that have received net population
inflows would have benefited from economies of scale in government
spending while at the same time would have been under pressure to
increase representation of the council. This would lead to a downward
bias in the 1990 cross-sectional relationship between council size and
government spending, and the instrumental variable results help to
recover the causal effect of districting on spending.
D. Discussion
Given these results, it is useful to consider the magnitude of the effect

of council size on government size. Focusing on per capita expenditures
and using the baseline specification of column 4 of table 4, we can take
0.11 as an estimate of the magnitude of effect. An addition of one
council member, then, to an average city council (of seven members)
would be associated with a 1.6 percent (≈ 0.11 # 17 ) increase in per
capita city government expenditures. Given average per capita city
spending of $792, this amounts to an increase of $12 per capita. In
aggregate terms, these coefficients imply that for an average city of
58,000 people, an addition of one political district in the city would be
associated with an increase of $0.72 million (≈ 0.016 # 792 # 58,000)
in the city budget.23 With a median city budget of $17.5 million, and
given that when cities consider changes in the number of seats in the
23
If anything, this estimate is likely to be an underestimate of the effect on aggregate
expenditures. Redoing the regressions with the log of total city government expenditures
as the dependent variable gives an estimated coefficient (standard error) of (i) 0.294
(0.061) in the OLS regression of table 4 (with state indicators), (ii) 0.353 (0.191) in the
2SLS specification of table 6 (with state indicators), and 0.381 (0.090) in the OLS re-
gression corresponding to panel B of table 10 below with indicators and interaction for
cities with strong mayors. When the smallest of these coefficients (0.29) is used, the
estimated effect of an increase of one councilman on the aggregate budget is $3.1 million
at the sample mean and $0.73 million at the sample median of total expenditures.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1337
city council they typically consider changes of more than one council-
man, these are fairly substantial effects.
It is also useful to consider actual government spending in otherwise
similar cities that have very different council sizes. Two of the key de-
terminants of council size are state and population. Grouping big cities
by state and population and looking for big differences in council sizes
gives an illustrative example from Connecticut. Bridgeport and New
Haven had very similar levels of population (142,000 and 130,000, re-
spectively), racial heterogeneity (0.571 and 0.573), per capita income
($13,100 and $12,970), land area (16 and 19 square miles), and other
indicators in 1990 used in the baseline regression. New Haven, however,
has 30 council members, 10 more than Bridgeport. In 1990 it also spent
$600 more than Bridgeport. The predicted difference from the regres-
sion in per capita spending is $200 for these two cities.
E. Additional Sensitivity Analysis
In the rest of this section I present results from other sensitivity analysis
exercises carried out to check the robustness of the basic results. Table
5 pulls together the results from variations on the basic specification of
column 4 of table 4.24 For ease of reference, row 0 repeats the results
from table 4. The subsequent four rows were discussed above in the
context of state-specific effects. Row 1e presents one additional speci-
fication to control for state-specific effects: it controls for the share of
total revenue coming from state government as a proxy for the degree
of state influence in the fiscal affairs of the city. This share varies in the
sample from zero to 88 percent with a mean (median) share of 16
percent (13 percent). Row 1e shows that the council size coefficients do
not change much when we control for this variable. It is important to
note that for a good fit between theory and data, most of the revenue
needs to come from local sources. If most revenues came from, say, the
state government and cities lobbied to get greater revenues transferred
down, the appropriate measure of the “number of players” in the com-
mon pool would be the number of municipal governments in the state.
For the average city, 78 percent of the revenue comes from local sources
(median 81 percent). As a final check, I ran the basic regression of
table 5 with the log of the locally generated per capita government
revenue as the dependent variable. The estimated coefficient (standard
error) on ln(council size) is 0.0856 (0.038) with a p-value of .024, very
close to the original estimates of the effect on the size of government.
24
Coefficients on the control variables are suppressed to conserve space. Significant
differences in any of these variables are noted in the discussion. Complete results are
available on request.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
TABLE 6
2SLS Results for Government Size (Np465)
(1) (2) (3) (4)

A. ln(Government Expenditures per Capita)
Estimation OLS 2SLS OLS 2SLS
State fixed
effects no no yes yes
ln(council size) .4084*** .6448*** .1630** .3173**
(.0832) (.1319) (.0751) (.147)
Ethnic .2327* .3013** .4237*** .4188***
(.1379) (.1422) (.1361) (.1369)
Income per .2496*** .2716*** .1788*** .1796***
capita (.0627) (.064) (.0527) (.053)
%BAgrad .1634 .1233 .2781 .2754
(.294) (.2971) (.2277) (.2288)
Mean/median .4021** .4069** .6143*** .6093***
income (.1883) (.19) (.1558) (.1566)
City population:
1st quintile ⫺.7590** ⫺.7671** ⫺.5950** ⫺.5805**
(.3126) (.3154) (.238) (.2395)
2d quintile ⫺.2626 ⫺.2675 ⫺.1823 ⫺.177
(.2015) (.2033) (.1516) (.1524)
3d quintile ⫺.4312*** ⫺.4429*** ⫺.2719*** ⫺.2804***
(.1395) (.1408) (.1042) (.105)
4th quintile ⫺.1957** ⫺.2023** ⫺.0421 ⫺.053
(.0883) (.0891) (.0671) (.068)
5th quintile ⫺.0012 ⫺.0154 .0275* .0182
(.0192) (.0203) (.0148) (.0167)
Constant ⫺1.7937*** ⫺2.2838*** ⫺1.6949*** ⫺2.0302***
(.3001) (.369) (.4203) (.5036)
Adjusted R 2 .16 .15 .57 .57
Standard error
of regression .474 .478 .338 .340
B. ln(Government Expenditures as a Share of Income)
State fixed
ln(council size) .3956*** .6060*** .1684** .2996**
(.0799) (.1265) (.0739) (.1444)
Ethnic .2595* .3205** .4687*** .4645***
(.1323) (.1363) (.1339) (.1345)
Income per ⫺.2502*** ⫺.2307*** ⫺.3066*** ⫺.3060***
capita (.0602) (.0613) (.0518) (.052)
%BAgrad ⫺.1809 ⫺.2166 ⫺.1271 ⫺.1295
(.2822) (.2848) (.224) (.2249)
Mean/median .7334*** .7377*** .9781*** .9738***
income (.1807) (.1821) (.1533) (.1539)
City population:
1st quintile ⫺.6901** ⫺.6973** ⫺.5116** ⫺.4993**
(.3) (.3023) (.2342) (.2354)
2d quintile ⫺.2486 ⫺.2529 ⫺.161 ⫺.1566
(.1934) (.1949) (.1492) (.1498)
3d quintile ⫺.4062*** ⫺.4166*** ⫺.2542** ⫺.2614**
(.1339) (.135) (.1025) (.1032)
4th quintile ⫺.1853** ⫺.1912** ⫺.0432 ⫺.0525
(.0847) (.0854) (.066) (.0668)
1338

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
TABLE 6
(Continued )
(1) (2) (3) (4)

5th quintile ⫺.0043 ⫺.0169 .0209 .013
(.0184) (.0195) (.0145) (.0164)
Constant .5793** .1431 .6093 .3242
(.2881) (.3538) (.4136) (.4948)
Observations 465 465 465 465
Adjusted R 2 .20 .19 .57 .57
Standard error
of regression .455 .458 .333 .334
C. ln(Government Employment per Capita)
State fixed
ln(council size) .6965*** .9087*** .2104*** .3545**
(.0894) (.1413) (.074) (.1446)
Ethnic .0629 .1244 .3039** .2993**
(.148) (.1523) (.134) (.1347)
Income per ⫺.046 ⫺.0263 .0396 .0403
capita (.0673) (.0685) (.0519) (.0521)
%BAgrad .4991 .4631 .4785** ⫺.4759**
(.3157) (.3182) (.2241) (.2252)
Mean/median 1.0453*** 1.0496*** .7374*** .7326***
income (.2022) (.2035) (.1534) (.1541)
City population:
1st quintile ⫺.6291* ⫺.6364* ⫺.4531* ⫺.4396*
(.3357) (.3378) (.2343) (.2357)
2d quintile ⫺.0984 ⫺.1028 ⫺.0342 ⫺.0294
(.2163) (.2177) (.1492) (.15)
3d quintile ⫺.3780** ⫺.3885** ⫺.1662 ⫺.1741*
(.1498) (.1508) (.1026) (.1033)
4th quintile ⫺.2024** ⫺.2083** .0177 .0075
(.0948) (.0954) (.066) (.0669)
5th quintile ⫺.0376 ⫺.0504** .0057 ⫺.003
(.0206) (.0218) (.0146) (.0164)
Constant ⫺.1096 ⫺.5494 .9063** .5933
(.3223) (.3953) (.4138) (.4955)
Observations 465 465 465 465
Adjusted R 2 .19 .18 .65 .65
Standard error
of regression .509 .512 .333 .335
Note.—For each measure of government size and for specifications with and without a complete set of state indicators,
the table compares the results of OLS and 2SLS regressions. Council size (in 1990) is instrumented with 1960 council
size using historical data from Aiken and Alford (1972). The OLS results are reported since the common sample is
much smaller than the regressions of table 4 because of availability of 1960 council size data. Population quintiles have
population data expressed in 100,000s. Robust standard errors are in parentheses.
* Significant at the 10 percent level.
1339

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
Row 2 of table 5 includes as an additional regressor the growth in
the city population, which can be an important factor because of the
implications for city infrastructure. The estimated coefficient on the
population growth variable is negative (⫺0.202 for the expenditure per
capita regression) and statistically significant. When I split the popu-
lation growth variable in a manner similar to the 1990 population var-
iable as described above, I find that most of the effect is coming from
the upper quintiles: the coefficients on the first three quintiles of pop-
ulation growth are not significant, whereas there is very little change in
the coefficients on other variables. The negative estimated effect is there-
fore coming from rapidly growing cities in the sample.
Although we control for racial heterogeneity of the city, heterogeneity
of the council may be an additional factor affecting government spend-
ing. On the one hand, one could envisage that it is not council size per
se but the number of racial groups in the city council that, potentially
through a similar common-pool type argument, is driving up govern-
ment expenditures. Since bigger councils are likely to be more heter-
ogeneous, council size may simply be proxying for the number of racial
groups in the council. This would imply that government spending
should not be related to the number of districts in racially homogeneous
councils. On the other hand, one could argue that in addition to elec-
toral districts, spending coalitions get formed along racial groups, so
that spending is sensitive to both the number of districts and the number
of racial groups in the council. Row 3 presents the results when we
additionally control for a measure of the effective number of racial
groups in the council.25 There is very little impact on the coefficient for
the council size variable. In addition, the coefficient on the effective
number of ethnic groups variable ranges between 0.10 and 0.12 for the
three specifications and is significant at 5 percent. These results are
more consistent with the latter interpretation: it is not the case that the
size of the council is simply proxying for the ethnic heterogeneity of
the council. The next two rows (4a and 4b) show separate slope coef-
ficients for heterogeneous and homogeneous cities and provide further
evidence that the effect of districting is present in both.
In the government expenditure regressions, 1,777 of the 1,972 cities
in the sample have councils composed of nine or fewer members. To
25
This is in addition to controlling for the racial heterogeneity of the city population.
The effective number of ethnic groups is the reciprocal of one minus the ethnic variable
for the city council using data on council members by race. When racial groups are
distributed equally, this equals the number of racial groups. When groups are not dis-
tributed symmetrically, as in one large group and several small groups, it is less than the
number of groups to capture the “effective” number of groups. This is the same variable
used by, e.g., Taagepera and Shugart (1989), Ordeshook and Shvetsova (1994), and Cox
(1997) in their studies of the effects of electoral systems on the number of effective parties
in the legislature.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1341
see whether the relationship holds separately in big and small councils,
I estimate separate coefficients on council size for these two groups.
Row 5 shows significant effects for both large and small councils. I also
estimate separate coefficients depending on whether the city population
is less than or greater than the median for the sample (25,555). This
is done partly to capture the commonly discussed idea that big cities
have their special problems and may attain poorer outcomes for reasons
other than the externalities inherent in distributive politics. Results show
that (a) the magnitude of the effect is stronger in large cities, and (b)
the statistical significance for the council size variable in small cities
drops somewhat, but the coefficients are significant at 6 percent for two
of the three measures and at 5 percent for the third measure of gov-
ernment size.
The next two regressions include other potentially omitted variables:
inner-city versus suburban versus rural cities and population density.
There may be systematic differences between inner cities and suburbs
that are correlated with both desired government expenditures and
council size. Central city residents typically favor greater public services
and, because they are more heterogeneous, may also desire bigger city
councils. Suburbs generally have the opposite characteristics. The same
effect to some extent can be picked up in the population density var-
iable. Results show that the council size effect is robust to these con-
siderations. I also try to control directly for variation in political pref-
erences by using data on percentages voting for a Democratic president
in the 1992 presidential election, assuming that, though many factors
are likely to affect a voting decision, residents with innate preferences
for big government are, ceteris paribus, more likely to vote for a Dem-
ocratic candidate. Such voting data are available only at the county level.
I mapped each city in the sample to the county it is located in and used
the county electoral data as a proxy for the city electoral variable.26 The
results after I control for this variable are shown in row 9 of table 5.
There is little change for the council size coefficient, and, interestingly,
the coefficient on the voting variable is not significant in any of the
specifications, suggesting that other controls in the equation may already
be accounting for the variation in political preferences.
V. Results II: Electoral Systems and Form of Government

Table 1 demonstrated that the city electoral system and form of gov-
ernment change relatively infrequently. The purpose of this section is
26
For this to be a good proxy, it requires that there be a relatively high correlation
across cities in a county on voting patterns. In the absence of direct information on how
large or small this variation may be, the results on this variable should be interpreted with
caution.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
to examine whether the relationship between council size and govern-
ment size varies systematically across these two sets of political
institutions.
A. Electoral Systems
The key variation in electoral systems across cities is whether candidates
are elected from the entire city or from districts within the city.27 Of the
total number of cities in the sample, 56 percent have at-large systems,
17 percent have district systems, and the remaining 27 percent have a
mixed system in which some council members are elected by district
and some at large.28 Traditionally cities had district-based systems. At-
large systems were introduced in some cities around the turn of the last
century in part because it was believed that they would help to curtail
pork barrel–type spending by inducing council members to treat the
entire city as their constituency. For instance, Richard S. Childs, an early
municipal reformer, noted the following as a criticism of ward systems
(and a recommendation for at-large systems): “ward elections notori-
ously produced political small fry who intrigued in the council for petty
favors and sought appropriations for their wards in reckless disregard
of city-wide interests and the total budget” (1965, p. 37). In their review
of the argument for adopting at-large systems in U.S. cities, Engstrom
and McDonald (1986) note that council members elected at large were
“expected to make decisions on the basis of what they perceived to be
good for the entire city, not just one geographic or social segment of
it” (p. 203).
Alternatively, at-large council members, despite running from the
whole city, may have “home bases” or particular constituencies com-
prising subsets of the city population that they seek to distribute ex-
penditures to in exchange for votes.29 If so, we would expect the same
effect from increasing at-large council members as from increasing dis-
trict council members: an additional at-large council member represents
27
I have so far been using the terms “council size” and “number of districts” inter-
changeably. The two need not be the same in cities in which some council members are
elected at large. The results in this section will justify the use of council size as the relevant
right-hand-side variable.
28
For most cities, when council members are elected at large, they run from the entire
city. A few cities, however, have several multimember districts. Although I do not have the
data to distinguish between single-member and multimember district systems, Welch
(1990) collected these data in a survey and found that 1.9 percent of her sample were
such cities. For empirical purposes, therefore, I take the district electoral systems to mean
single-member district systems.
29
See Uslaner (1985) for a study of Israel’s Knesset, an extreme example of an at-large
system at the national level in which all representatives are elected from the entire country.
He shows that legislators identify themselves with particular constituencies within the
country along geographical, ethnic, and religious lines.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1343
an additional player in the pool. Such home bases can develop along
dimensions such as ethnicity, income, age, and any other characteristic
that can segment the city population. Indeed, in at-large electoral sys-
tems, geography may be one of the dimensions along which constitu-
encies for individual council members may form. It will be useful to
explicitly state the two contrasting views on the role of at-large council
members.
Hypothesis I. At-large council members cater to the common good
of the whole city, and district council members cater to the good of
their respective districts.
Hypothesis II. At-large council members cater to particular constit-
uencies and face the same asymmetry between benefits and costs of
their policy proposals as district council members.
The prediction to be tested is whether the relationship between coun-
cil size and government size is nonexistent in cities with at-large council
members. As a first look at the question, I estimate the following spec-
ification:
ln (g) p a 0 ⫹ a1 DL ⫹ a 2 ln ( J ) ⫹ a 3 DL 7 ln ( J ) ⫹ b 7 Z ⫹ e,
where g is a measure of the size of government, DL is an indicator variable
for a council with a majority of the council elected at large, J is the size
of the council, and Z are all controls used in table 4. The predictions
of hypothesis I are a 2 1 0 and a 2 ⫹ a 3 p 0, and those of hypothesis II
are a 2 1 0 and a 3 p 0.
The differing predictions rest on the estimated coefficient for a3.
Results, presented in panel A of table 7, show that the data reject the
first hypothesis for all three measures of government size. For the gov-
ernment expenditure regressions, the estimate of a3 is not statistically
different from zero, indicating the same relationship between districting
and government size in both district and at-large majority councils. For
the government employment regression, an at-large majority council is
associated with a negative intercept effect but a steeper positive rela-
tionship between council size and government size. At the sample mean,
the net effect of switching to an at-large majority council is an increase
in government size of 15 percent. The coefficients on the other variables
are suppressed to conserve space, but they are close to the estimates in
table 4.
Panel B of table 7 takes a closer look at the data. The specification
above lumped together pure at-large systems with those mixed systems
in which a majority of the council members were elected at large. This
panel separates the coefficients across nonmixed and mixed electoral
systems. The first three coefficients compare cities with district systems
to at-large systems, and the next three coefficients compare mixed sys-
tems with an at-large majority to mixed systems with a district majority.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
TABLE 7
Regressions for Effects of Electoral Systems
ln(Expenditures as Share of Employment
per Capita) City Income) per Capita)
(Np1,972) (Np1,972) (Np1,968)
Panel A
Majority at large ⫺.2056 ⫺.1848 ⫺.4662***
(.1358) (.1316) (.1442)
ln[council size (J)] .1510*** .1475*** .1252**
(.0507) (.0493) (.0504)
Majority at large#ln(J) .075 .0701 .2051***
(.0709) (.0686) (.0759)
Adjusted R 2 .33 .43 .45
Panel B
Pure#majority at large ⫺.1457 ⫺.1324 ⫺.3717**
(.1399) (.1358) (.1474)
Pure#ln(J) .1132** .1107** .0957*
(.0495) (.0483) (.05)
Pure#majority at large .0735 .0721 .1754**
#ln(J) (.0737) (.0716) (.0781)
Mixed#majority at large ⫺.676 ⫺.6248 ⫺.7596
(.457) (.432) (.4716)
Mixed#ln(J) .1743*** .1700*** .1448***
(.0502) (.0489) (.0504)
Mixed#majority at large .2953 .2721 .3939*
#ln(J) (.1961) (.1856) (.2192)
Adjusted R 2 .33 .43 .45
Panel C
Pure system .2651* .2599* .1896
(.1555) (.1515) (.1586)
Pure#district share#ln(J) .1190*** .1136*** .1688***
(.0438) (.0429) (.0461)
Pure#at-large share#ln(J) .1139** .1139** .1510***
(.0494) (.0484) (.0521)
Mixed#district share# .2620*** .2547*** .2561***
ln(J) (.0731) (.0716) (.0725)
Mixed#at-large share# .4129*** .3972*** .4461***
ln(J) (.0805) (.0784) (.0845)
Adjusted R 2 .33 .43 .45
Note.—Each panel corresponds to a different specification. All regressions include the complete set of controls of
table 4, including state indicators and population quintiles. Majority at large is an indicator variable for a city council
with a majority of at-large council members. J represents council size. Pure is an indicator for an electoral system in
which either all council members are elected by district or all at large, and mixed is an indicator for an electoral system
in which some council members are elected by district and some at large. District share (at-large share) is the share
of district (at-large) council members in the council. See text for the interpretation on the transformed variables in
panel C. Robust standard errors are in parentheses.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1345
Results are consistent with those in panel A. All at-large councils have
the same relationship between council size and government spending
as pure district councils, whereas mixed systems with a majority of at-
large council members have the same relationship as those with a ma-
jority of district council members.
The results in panel B of table 7 also indicate that the magnitude of
the relationship becomes stronger when we look at mixed systems. The
F-tests for the equality of coefficients on the council size variable across
mixed and nonmixed systems reject at p-values of less than .01 for all
three measures of government size. This indicates that in addition to
an intragroup externality, mixed systems may be associated with an in-
tergroup externality, leading to even less internalization of the costs of
spending proposals and hence greater sensitivity of government spend-
ing to council size.30
Finally, panel C of table 7 takes another look at the same question.
The two sets of results discussed above focused on the at-large election
of the majority of the council. It is possible that at-large council members
may exert influence on spending decisions even when they do not con-
stitute a majority. Such outcomes can come about in universalistic
decision-making norms in the legislature when the aggregate decision
reflects the desired outcomes of each member of the legislature. The
specification estimated in panel C is
ln (g) p a 0 ⫹ (1 ⫺ DM) 7 a1 ⫹ a 2 [ JD
J
JL
]
ln ( J ) ⫹ a 3 ln ( J )
J
⫹ DM 7 a 4[ JD
J
JL
]
ln ( J ) ⫹ a 5 ln ( J ) ⫹ b 7 Z ⫹ e,
J
where DM is an indicator for a city with a mixed electoral system. The
advantage of transforming the data on council members in this way is
that (a) under the null of a 2 p a 3 (alternatively a 4 p a 5), the inde-
pendent variable reduces to ln(J), allowing the estimated coefficients
to be compared to the previous specifications; and (b) a 2 1 a 3 (alter-
natively a 4 p a 5) if and only if ⭸g/⭸JD 1 ⭸g/⭸JL. Hence, a comparison of
a2 and a3 (and a4 and a5, respectively) allows us to compare the mag-
nitude of the effect for the two types of council members.31 The inter-
30
Chow tests to determine whether coefficients on all control variables should be freed
across mixed and nonmixed cities did not reject.
31
We can also run the regression log (g) p a1 ⫹ a2JD ⫹ a3JL ⫹ b 7 Z ⫹ e, separating the
coefficients across mixed and nonmixed systems as above and testing whether a3 p 0. The
magnitudes, however, cannot be compared with the previous set of results because of the
loglinear specification. The results are qualitatively the same. The estimated coefficients
(p-values) for expenditures per capita are the following, with the first two coefficients for
nonmixed and the second two for mixed systems: 0.120 (0.029), 0.010 (0.088), 0.019
(0.013), and 0.050 (0.000).

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
action with DM estimates the effects separately for mixed and nonmixed
councils. The results in panel C of table 7 show (i) no qualitative dif-
ference between the effects of the two types of council members and
(ii) a different pattern of coefficients for mixed versus nonmixed elec-
toral systems. When districts are compared with at-large systems, there
is very little difference in the effects of the two types of council members:
F-tests for aˆ 2 p aˆ 3 do not reject at conventional levels for all three
measures. In mixed systems, the estimated marginal effect of at-large
council members is greater than that of district council members, and
the difference is statistically significant: F-tests for aˆ 4 p aˆ 5 consistently
reject at 5 percent for all three regressions. Comparing the estimated
coefficients for each type of council members across mixed and non-
mixed systems indicates stronger effects in mixed systems: F-tests for the
joint hypothesis aˆ 2 p aˆ 4 and aˆ 3 p aˆ 5 reject at p-values of less than .001
for all three measures of government size.
1. Discussion
The results in table 7 indicate that although critics of district systems
may have been right in thinking that district systems contribute to more
government spending, they were likely wrong in supposing that at-large
council members would not cater to particular constituencies within the
jurisdiction. The result, however, is all the more surprising since the
electoral system used in at-large city elections is a first-past-the-post sys-
tem, where voters typically are allowed to cast as many votes as there
are seats to be filled and candidates with the largest number of votes
are declared the winners. As discussed elsewhere in the literature on
electoral systems (e.g., Cox 1997), such systems tend to reduce the
number of groups in the legislature. For instance, if there is a majority
group and a racial minority group and people vote only for members
of their own group, it is possible for the legislature to consist entirely
of the majority group. The results show that even though at-large systems
may reduce council heterogeneity, council members still seek to target
government expenditure to particular groups, resulting in the contin-
uation of pork barrel–type spending. This is consistent with evidence
discussed above that the relationship between council size and govern-
ment size exists in both homogeneous and heterogeneous councils.
The results in table 7 also show that the estimated effects are larger
in cities with mixed electoral systems. One concern with this result might
be the following: if mixed systems tend to have fewer numbers of at-
large and district council members than pure at-large and pure district
systems, respectively, and if government size is a concave function of
council size, then we would automatically get bigger coefficients in the
mixed sample. This turns out not to be the case. Although there is

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1347
TABLE 8
Mean Number of Council Members by Electoral System
District System At-Large System Mixed System

Number elected by district 2.0 5.7
Number elected at large 4.5 2.5
suggestive evidence that government size is a concave function of council

size, it is not the case that mixed systems have fewer numbers of both
at-large and district council members, as table 8 demonstrates.32 The
table reports the mean number of council members elected by district
and at large by electoral system. Mixed systems, on average, have a
greater number of district council members (than district systems) and
a smaller number of at-large council members (than at-large systems).
The predicted changes in magnitude would therefore be in opposite
directions; however, the results in table 7 show that for both district and
at-large council members, the effects become stronger in mixed coun-
cils. The t-tests for the equality of means across these two types of ob-
servations reject strongly for both district and at-large council members.
B. Form of Government
Cities vary in their form of government and the powers they afford the
office of the executive. Table 9 gives a breakdown. Of the 1,696 cities
for which the form of government and mayor powers data are available,
roughly one-third (641) have the mayor-council form of government.
Subsequent rows show that this form is systematically associated with
greater powers afforded to the city mayor. Of mayor-council form cities,
98 percent have directly elected mayors and 74 percent give their mayors
veto powers, whereas of council-manager cities, 65 percent have directly
elected mayors and 11 percent give them veto powers.
One theme in the existing cross-county literature on political insti-
tutions and budgetary outcomes is that presidential systems of govern-
ment with strong executives can enforce fiscal discipline on a legislature
otherwise prone to overspending.33 The variation in political form of
government across cities in the United States (mayor-council and coun-
cil-manager systems) maps quite well to presidential and parliamentary
systems, and evidence from cities can relate interestingly to the debate
32
The regression corresponding to col. 4 in table 4, but replacing log(J) with J and J 2,
yields the following coefficients (standard errors) on the linear and quadratic terms: 0.0263
(0.0086) and ⫺0.0006 (0.00027), respectively. Both coefficients are statistically significant
at 5 percent in this regression. However, the quadratic term loses significance in some of
the sensitivity analysis regressions of table 5.
33
See the review by Alesina and Perotti (1999) and other papers in the Poterba and
von Hagen (1999) volume.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
TABLE 9
Distribution of Mayor Powers
All Cities Mayor-Council Council-Manager

Total 1,696 641 1,055
Mayor elected directly from city 1,315 629 686
(78%) (98%) (65%)
Mayor proposes budget to council 293 286 7
(17%) (45%) (1%)
Mayor appoints department heads 394 379 15
(23%) (59%) (1%)
Mayor can veto council-passed 585 473 112
measures (34%) (74%) (11%)
Mayor can veto specific items of 146 126 20
appropriations (9%) (20%) (2%)
Note.—Numbers in parentheses are percentages of total. Sample consists of all available observations and, in par-
ticular, is slightly larger than the regression sample. There are no significant differences in the pattern if the sample
is restricted to the regression sample.
at the cross-county level. In this subsection I examine whether cities

with strong executives are able to break the link between districting and
government spending. The basic specification I estimate is similar to
the one estimated for majority at-large systems above:
ln (g) p a 0 ⫹ a1 DM ⫹ a 2 ln ( J ) ⫹ a 3 DM 7 ln ( J ) ⫹ b 7 Z ⫹ e,
where DM is an indicator variable either for the mayor-council form of
government or for one of the measures of mayor powers in table 9.34
The hypothesis to be tested is a 2 ⫹ a 3 p 0.
Panel A of table 10 reports the results when DM equals one (and zero
otherwise) for the mayor-council form of government. First, note that
when variation in the form of government is taken into account, the
coefficients on the council size variable become stronger than the ones
in table 4. This is not due to the change in sample size since running
the regressions of table 4 for this sample gives coefficients very close to
those reported in table 4. Second, for two of the three measures of
government size, the coefficient on the interaction is negative and sta-
tistically significant at p-values of less than 8 percent. Coefficients on
the other variables are suppressed, but the pattern is very similar to the
ones in table 4. Formal tests for the hypothesis a 2 ⫹ a 3 p 0 do not
reject at conventional levels, but the test is not very powerful because
of the statistically weak coefficients estimated on the interaction term.
I next ran similar regressions using the indicators of actual mayor
powers using the four indicators listed above. The only indicator of
mayor powers that was associated with statistically significant results was
the overall mayor veto indicator: in all other cases, log council size was
34
Chow tests for estimating a separate set of coefficients for the control variables do
not reject at conventional levels for all three measures of government size.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1349
TABLE 10
Regressions for Form of Government
ln(Expenditures per as Share of Employment
Capita) City Income) per Capita)
Panel A
Mayor-council form .2619 .2165 ⫺.0144
(.1716) (.166) (.1771)
ln[council size (J)] .2213*** .2171*** .1758**
(.0701) (.0673) (.0687)
Mayor-council form# ⫺.1601* ⫺.1407* ⫺.0156
ln(J) (.0871) (.0842) (.0881)
Observations 1,451 1,451 1,449
Adjusted R 2 .35 .45 .47
Panel B
Mayor veto .4665*** .4348*** .2978*
(.1746) (.1684) (.1772)
ln[council size (J)] .2051*** .2008*** .2019***
(.0669) (.0655) (.0734)
Mayor veto#ln(J) ⫺.2222** ⫺.2060** ⫺.1391
(.0878) (.085) (.0895)
Observations 1,432 1,432 1,430
Adjusted R 2 .35 .45 .47
Panel C
Specification OLS 2SLS OLS 2SLS OLS 2SLS
Mayor veto .469 1.411 .500 1.358 .806* 1.295
(.405) (1.056) (.393) (1.021) (.426) (1.104)
ln[council size (J)] .263** .624** .268** .612*** .678*** .858***
(.131) (.225) (.126) (.217) (.137) (.235)
Mayor veto#ln(J) ⫺.208 ⫺.668 ⫺.219 ⫺.646 ⫺.359* ⫺.559
(.204) (.513) (.198) (.496) (.214) (.536)
Observations 343 343 343 343 343 343
Adjusted R 2 .11 .08 .18 .15 .19 .17
Note.—Each panel corresponds to a different specification. Mayor-council form is an indicator for whether the
government has declared a mayor-council form of government. J represents council size. Mayor veto is an indicator for
whether the city mayor has powers to veto council-passed measures. Regressions include all other controls of table 4,
except panel C, where state indicators are not included. 2SLS specifications in panel C use 1960 values of council size
and mayor-council form as instruments. Robust standard errors are in parentheses.
close in magnitude to the coefficients reported in panel A of table 10

and was statistically significant, but the interaction term was not. The
results with the mayor veto variable are presented in panel B of table
10. For the expenditure measures of government size the results are
quite striking: government expenditure does not increase with council
size in city governments in which mayors have veto powers. For the
government employment regression, the p-value associated with the co-
efficient on the interaction term is .12.
It is also interesting to note that the coefficient estimated on the
indicator for a mayor veto is positive and statistically significant. When
the number of districts is held constant, the effect of switching to a
strong-mayor form is positive for small councils and negative for big

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
councils. At the sample median of a seven-member council, the switch
from a weak- to a strong-mayor form entails a 3 percent increase in
expenditures per capita. Note that the prediction of common-pool mod-
els refers to the slope effect—government size increases with district-
ing—and it continues to be the case that governments with executive
veto do not exhibit this relationship.
A potential problem with the results might be the possible endoge-
neity of the mayor veto variable. If cities that are more prone to having
fiscal problems choose strong-mayor forms of government, then the
coefficient on the mayor veto is likely to be biased upward. Available
evidence, however, indicates that this is not likely. The Aiken and Alford
(1972) study, which was used to instrument for council size, also has
information on the form of city government in 1960, which is a good
predictor for whether the city’s mayor in 1990 has veto powers. For the
common sample of 473 observations, regressing 1990 mayor veto on
1960 form of government, all control variables used in the regressions
above, and a complete set of state indicators yields highly statistically
significant positive coefficients on the 1960 variable for both the linear
probability model and the probit model. A switch to a mayor-council
form of government in 1960 is associated with a 0.40 increase in the
probability of having mayor veto in 1990 for the linear model and a
0.53 increase in probability for the probit model. I next instrument for
the mayor veto variable using the 1960 form of government and run
the same regression.35 These results are displayed in panel C of table
10. For all three measures of government size, the results indicate that
the coefficient on the mayor veto variable becomes stronger in mag-
nitude in the instrumental variable specifications. The standard errors
on the mayor veto variable are also bigger, but the coefficient increases
by about three times for the per capita expenditures regression.36 If
reverse causality was accounting for the positive coefficients in the OLS
results, we would have expected the opposite. Although this finding is
not pursued further here, exploring the positive coefficient on the
mayor veto indicator would make for interesting future research.37 For
the present purposes, I note that addressing endogeneity considerations
35
The sample size drops to 343 from 473 as government expenditure variables are
additionally included.
36
These regressions include all variables in the standard specification, including pop-
ulation quintiles, but do not include state indicators. With the considerably smaller degrees
of freedom, regressions with a complete list of state indicators do not give any significant
coefficients except per capita income. The direction of the change in the coefficient on
the mayor veto variable, however, is the same when state indicators are included.
37
One possibility is that although strong mayors are able to enforce fiscal discipline on
the council, they may themselves indulge in patronage-related spending by virtue of their
strong position.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1351
using the form of government 30 years ago as an instrument strengthens
the results on the council size variable as well as the interaction.
VI. Conclusion
The purpose of this paper was to test whether the size of government
is sensitive to the number of people making spending proposals out of
a common revenue pool. Evidence from U.S. cities shows that scaled
measures of government size do indeed go up with the number of
legislators in a city government. The finding is robust to consideration
of a number of possibly omitted variables and specifications. When con-
cerns of potential reverse causality are addressed by instrumenting with
the size of the city council 30 years go, the estimated magnitude of the
effect becomes stronger. An estimate for the elasticity of government
size with respect to the number of districts is 0.11. The findings also
show that government size increases with the racial heterogeneity of the
city, with a measure of the skewness of the income distribution, and
that it decreases with city population in small cities but increases in large
cities. Given the basic finding, the paper also looked at whether councils
(a) dominated by legislators elected at large and (b) with strong mayors
are able to break this relationship between the number of players and
government size. The evidence suggests “no” to the first question and
“yes” to the second.
Given the relatively good fit between theory and the data at hand,
these findings are important for policy. They argue, first of all, that for
any government, keeping the fiscal house in order depends in large
part on how many people get to spend out of the tax revenue pool.
When more people are added, everybody, including the incumbents,
raises his or her spending decisions. Second, they show the relative
fruitlessness and relative usefulness, respectively, of the following two
ways to contain this spending tendency: electing council members at
large and giving executives strong powers. Given that these city govern-
ments share many features with national governments, these findings
have interesting implications for countries or states considering changes
in the political process to address chronic fiscal problems. They imply
that whether legislators run for office from the entire jurisdiction or
from geographical districts within it, their decisions with respect to gov-
ernment spending do not differ: giving the executive veto powers may
be a better way to enforce discipline on the legislature.
These findings, however, have left certain areas unaddressed. First is
the hard issue of welfare consequences. Lower spending may be ben-
eficial for all if the point of departure is a state of overspending arising
from a common-pool problem and if it is brought down for all districts.
It may not be beneficial if existing spending is too low or the distribution

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
of reductions across districts is skewed. When some districts lose more
spending than others, the welfare consequences are unclear. Thus
strong executives may reduce the distortion on the size of the budget,
yet they might introduce distortions in the distribution of spending if
they cater to a minimum winning coalition. These issues are hard to
address with the data at hand since city government budgets are not
broken down by district. Such analysis may be possible, however, at the
federal level looking at the relationship between the changes in the size
of the federal budget and its distribution across congressional districts.
Another important area of research is to take one step further back
and examine the question, Why do some cities choose strong-mayor
forms of government? Such an exercise addressing the important ques-
tion of why certain political institutions get chosen is admittedly difficult
to address using cross-country data because of the role of history and
unquantifiable factors, but it can be usefully addressed using the city
government data at hand. Over the course of their history, cities have
occasionally changed their form of government. If the arguments pre-
sented in this paper have some bearing on reality, one factor that would
be important for a city deciding the form of government would be the
size of the council. If the size of the council is relatively exogenous
because it may depend in part on the city population and if it is the
case that bigger councils are going to have greater spending pressures,
cities with large legislatures should choose strong-mayor forms of gov-
ernment. Although OLS and probit regressions show a positive and
significant effect of council size on alternative indicators for strong-
mayor forms of government, the empirical challenge is to identify the
effect of council size on the form of government since both may be the
result of some third outside factor (such as a movement for “reform”
that entailed changes to both council size and form of government).
Further work in this area would shed additional important light on these
issues.
References
Aiken, Michael, and Alford, Robert. “Governmental Units Analysis Data.” Study
no. 0028. Ann Harbor: Univ. Michigan, Inter-university Consortium Polit. and
Soc. Res., 1972.
Alesina, Alberto; Baqir, Reza; and Easterly, William. “Public Goods and Ethnic
Divisions.” Q.J.E. 114 (November 1999): 1243–84.
Alesina, Alberto; Hausmann, Ricardo; Hommes, Rudolf; and Stein, Ernesto.
“Budget Institutions and Fiscal Performance in Latin America.” J. Development
Econ. 59 (August 1999): 253–73.
Alesina, Alberto, and Perotti, Roberto. “Budget Deficits and Budget Institutions.”
In Fiscal Institutions and Fiscal Performance, edited by James M. Poterba and
Jürgen von Hagen. Chicago: Univ. Chicago Press (for NBER), 1999.
Alt, James E., and Lowry, Robert C. “Divided Government, Fiscal Institutions,

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
districting 1353
and Budget Deficits: Evidence from the States.” American Polit. Sci. Rev. 88
(December 1994): 811–28.
Baron, David P. “Majoritarian Incentives, Pork Barrel Programs, and Procedural
Control.” American J. Polit. Sci. 35 (February 1991): 57–90.
Baron, David P., and Ferejohn, John A. “Bargaining in Legislatures.” American
Polit. Sci. Rev. 83 (December 1989): 1181–1206.
Bayoumi, Tamim, and Eichengreen, Barry. “Restraining Yourself: The Implica-
tions of Fiscal Rules for Economic Stabilization.” Internat. Monetary Fund Staff
Papers 42 (March 1995): 32–48.
Benabou, Roland. “Inequality and Growth.” In NBER Macroeconomics Annual
1996, edited by Ben S. Bernanke and Julio J. Rotemberg. Cambridge, Mass.:
MIT Press, 1996.
Bohn, Henning, and Inman, Robert P. “Balanced-Budget Rules and Public Def-
icits: Evidence from the U.S. States.” Carnegie-Rochester Conf. Ser. Public Policy
45 (December 1996): 13–76.
Bradbury, John Charles, and Crain, W. Mark. “Legislative Organization and
Government Spending: Cross-Country Evidence.” J. Public Econ. 82 (December
2001): 309–25.
Chari, V. V., and Cole, Harold. “A Contribution to the Theory of Pork Barrel
Spending.” Staff Report no. 156. Minneapolis: Fed. Reserve Bank, 1993. (a)
———. “Why Are Representative Democracies Fiscally Irresponsible?” Staff Re-
port no. 163. Minneapolis: Fed. Reserve Bank, 1993. (b)
Chari, V. V.; Jones, Larry E.; and Marimon, Ramon. “The Economics of Split-
Ticket Voting in Representative Democracies.” A.E.R. 87 (December 1997):
957–76.
Childs, Richard S. The First 50 Years of the Council-Manager Plan of Government.
New York: Nat. Municipal League, 1965.
Cox, Gary W. Making Votes Count: Strategic Coordination in the World’s Electoral
Systems. Political Economy of Institutions and Decisions Series. Cambridge:
Cambridge Univ. Press, 1997.
Cox, Gary W., and Tutt, Timothy N. “Universalism and Allocative Decision Mak-
ing in the Los Angeles County Board of Supervisors.” J. Politics 46 (May 1984):
546–55.
Engstrom, Richard L., and McDonald, Michael D. “The Effect of At-Large versus
District Elections on Racial Representation in U.S. Municipalities.” In Electoral
Laws and Their Political Consequences, edited by Bernard Grofman and Arend
Lijphart. New York: Agathon, 1986
Gilligan, Thomas W., and Matsusaka, John G. “Deviations from Constituent In-
terests: The Role of Legislative Structure and Political Parties in the States.”
Econ. Inquiry 33 (July 1995): 383–401.
———. “Fiscal Policy, Legislature Size, and Political Parties: Evidence from State
and Local Governments in the First Half of the 20th Century.” Nat. Tax. J. 54
(March 2001): 57–82.
Hallerberg, Mark, and von Hagen, Jürgen. “Electoral Institutions, Cabinet Ne-
gotiations, and Budget Deficits in the European Union.” In Fiscal Institutions
and Fiscal Performance, edited by James M. Poterba and Jürgen von Hagen.
Chicago: Univ. Chicago Press (for NBER), 1999.
Inman, Robert P., and Fitts, Michael A. “Political Institutions and Fiscal Policy:
Evidence from the U.S. Historical Record.” J. Law, Econ. and Organization 6
(special issue; 1990): 79–132.
Kontopoulos, Yianos, and Perotti, Roberto. “Government Fragmentation and
Fiscal Policy Outcomes: Evidence from OECD Countries.” In Fiscal Institutions

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC
and Fiscal Performance, edited by James M. Poterba and Jürgen von Hagen.
Chicago: Univ. Chicago Press (for NBER), 1999.
Milesi-Ferretti, Gian Maria; Perotti, Roberto; and Rostagno, Massimo. “Electoral
Systems and Public Spending.” Q.J.E. 117 (May 2002): 609–57.
Mueller, Dennis C. “The Growth of Government: A Public Choice Perspective.”
Internat. Monetary Fund Staff Papers 34 (March 1987): 115–49.
Ordeshook, Peter C., and Shvetsova, Olga V. “Ethnic Heterogeneity, District
Magnitude, and the Number of Parties.” American J. Polit. Sci. 38 (February
1994): 100–123.
Persson, Torsten; Roland, Gérard; and Tabellini, Guido. “Separation of Powers
and Political Accountability.” Q.J.E. 112 (November 1997): 1163–1202.
———. “Towards Micropolitical Foundations of Public Finance.” European Econ.
Rev. 42 (May 1998): 685–94.
———. “Comparative Politics and Public Finance.” J.P.E. 108 (December 2000):
1121–61.
Persson, Torsten, and Tabellini, Guido. “Political Institutions and Policy Out-
comes: What Are the Stylized Facts?” Manuscript. Stockholm: Stockholm Univ.,
Inst. Internat. Econ. Studies, 2001.
Poterba, James M. “State Responses to Fiscal Crises: The Effects of Budgetary
Institutions and Politics.” J.P.E. 102 (August 1994): 799–821.
Poterba, James M., and von Hagen, Jürgen, eds. Fiscal Institutions and Fiscal
Performance. Chicago: Univ. Chicago Press (for NBER), 1999.
Roubini, Nouriel, and Sachs, Jeffrey D. “Fiscal Policy: Government Spending
and Budget Deficits in the Industrial Countries.” Econ. Policy: A European Forum,
no. 8 (1989), pp. 99–132. (a)
———. “Political and Economic Determinants of Budget Deficits in the Indus-
trial Democracies.” European Econ. Rev. 33 (May 1989): 903–33. (b)
Shepsle, Kenneth A., and Weingast, Barry R. “Political Preferences for the Pork
Barrel: A Generalization.” American J. Polit. Sci. 25 (February 1981): 96–111.
Taagepera, Rein, and Shugart, Matthew S. Seats and Votes: The Effects and Deter-
minants of Electoral Systems. New Haven, Conn.: Yale Univ. Press, 1989.
Uslaner, Eric M. “Casework and Institutional Design: Redeeming Promises in
the Promised Land.” Legislative Studies Q. 10 (February 1985): 35–52.
Velasco, Andres. “A Model of Endogenous Fiscal Deficit and Delayed Fiscal
Reforms.” In Fiscal Institutions and Fiscal Performance, edited by James M. Po-
terba and Jürgen von Hagen. Chicago: Univ. Chicago Press (for NBER), 1999.
von Hagen, Jürgen, and Harden, Ian J. “National Budget Processes and Fiscal
Performance.” European Economy: Reports and Studies, no. 3 (1994), pp. 311–418.
Weingast, Barry R.; Shepsle, Kenneth A.; and Johnsen, Christopher. “The Polit-
ical Economy of Benefits and Costs: A Neoclassical Approach to Distributive
Politics.” J.P.E. 89 (August 1981): 642–64.
Welch, Susan. “The Impact of At-Large Elections on the Representation of Blacks
and Hispanics.” J. Politics 52 (November 1990): 1050–76.

193.205.24.245 on Tue, 18 Apr 2023 16:44:23 UTC

Baqir

Uploaded by

Copyright:

Available Formats

Baqir

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Baqir

Uploaded by

Copyright:

Available Formats

Districting and Government Overspending

Author(s): Reza Baqir

This content downloaded from

Theories of government spending driven by a common-pool problem

A central feature of the recent literature on the political economy of

[Journal of Political Economy, 2002, vol. 110, no. 6]

This content downloaded from

This content downloaded from

This content downloaded from

II. Related Literature

Weingast et al. (1981) provided one of the earlier formalizations of the

This content downloaded from

This content downloaded from

This content downloaded from

Government Organization File. The latter source provides information

This content downloaded from

Government Expenditure Government Expenditure as

the number of attempted and successful changes in city political struc-

where si denotes the share of population of race i in the total city

This content downloaded from

This content downloaded from

City government expenditures, as a percentage

Council size Number of people 3 50 6 6.859 2.888 2,342

Csize p 5.87 ⫹ 5.22Pop90 ⫹ 1.64ethnic ⫹ 1.02MMI,

IV. Results I: Districting and Government Size

This content downloaded from

The variable with the greatest statistical significance is the logarithm of

This content downloaded from

(1) (2) (3) (4)

This content downloaded from

(1) (2) (3) (4)

This content downloaded from

Fig. 1.—Government size by city size

scale. The estimated relationship with respect to council size is not

This content downloaded from

Fig. 2.—Council size and government expenditures: all states

This content downloaded from

This content downloaded from

This content downloaded from

Given these results, it is useful to consider the magnitude of the effect

This content downloaded from

E. Additional Sensitivity Analysis

This content downloaded from

(1) (2) (3) (4)

This content downloaded from

(1) (2) (3) (4)

This content downloaded from

This content downloaded from

V. Results II: Electoral Systems and Form of Government

This content downloaded from

This content downloaded from

This content downloaded from

This content downloaded from

This content downloaded from

This content downloaded from

District System At-Large System Mixed System

suggestive evidence that government size is a concave function of council

This content downloaded from

All Cities Mayor-Council Council-Manager