Politicians, Bureaucrats and Patronage - iSites - Harvard University

3 downloads 527 Views 234KB Size Report
U.S. local governments to support this claim. Empirical comparison of the performance of appointed and elected public of& ficials is usually complicated by the ...
Politicians, Bureaucrats and Patronage Ruben Enikolopovy Harvard University JOB MARKET PAPER

Abstract The paper compares the policies of elected and appointed public o¢ cials with regard to public employment. I argue that elected politicians are more likely to use patronage jobs to achieve personal political goals than appointed bureaucrats. Evidence using panel data on local governments in the U.S. con…rms this claim. The number of full-time public employees is signi…cantly higher in local governments headed by elected chief executives. For part-time employees, who are less likely to be hired for patronage reasons, the di¤erence is much smaller or nonexistent. In addition, privatization of public service provision leads to a decrease in public employment only in communities with appointed chief executives.

I would like to thank Alberto Alesina, James Alt, Je¤ry Frieden, Ed Glaeser, Caroline Hoxby, Maria Petrova, James Robinson, Kenneth Shepsle, Andrei Shleifer, Ekaterina Zhuravskaya and seminar participants in CEFIR, Harvard University, NBER Student Political Economics Conference, XI Spring Meeting of Young Economists, European Economic Association Meeting in Vienna for helpful comments. y Department of Economics, Harvard University. Littauer Center, Room 200, Cambridge, MA 02138, USA. Email: [email protected]

1

1

Introduction

One of the central issues in political economy is the way di¤erent institutional arrangements a¤ect incentives of public o¢ cials and policies they pursue. An important political institution is a method of selection of public o¢ cials, in particular, whether they are elected or appointed. Although there is a strong theoretical argument that the di¤erence in the method of selection should play an important role, empirical evidence about the systematic di¤erences in the policies pursued by appointed and elected public o¢ cials remains scarce. In this paper, I focus on public sector employment as an example of a policy that is chosen di¤erently by elected and appointed public o¢ cials. It is widely acknowledged that public employment can be used by public o¢ cials to achieve their personal political goals. I argue that although both appointed and elected public o¢ cials are tempted to use patronage jobs, the di¤erence in incentives makes elected public o¢ cials more prone to using public employment to increase their chances of staying in o¢ ce than their appointed counterparts. I use data for the U.S. local governments to support this claim. Empirical comparison of the performance of appointed and elected public of…cials is usually complicated by the fact that for most public o¢ ces there is no variation in the method of selection of candidates, so that bureaucrats and elected politicians di¤er not only in their incentives, but in the tasks they perform. In this paper, I examine the performance of the chief executives in the U.S. local governments. Depending on the form of local government, a chief executive can be either directly elected or appointed by an elected council. In both cases, chief executives have essentially the same responsibilities and have substantial in‡uence over public employment policies. Thus, this example provides a rare opportunity to compare public employment policies pursued by elected and appointed public o¢ cials. Patronage jobs are all those posts that are distributed in exchange for electoral support and the pay for which is greater than the value of the public services performed, (Weingrod, 1968: 379; Wilson, 1961). Patronage is a convenient form of getting electoral support in exchange for transfers to a speci…c group of voters. First, it allows to target transfers to the members of a particular group of voters. Second, it allows to solve the commitment problem, prevalent in vote buying (Robinson and Verdier, 2003). When wages in the public sector are higher than in the private sector, voters are willing to support the candidate who credibly o¤ers them the prospects of working in the public sector. Hiring a person before the elections makes such a promise credible, as long as it is costly for the incumbent to …re a public employee. If the incumbent is not reelected, however, the successful opposition candidate can …nd political bene…ts of hiring her own supporters to be higher than the costs of …ring the supporters of the incumbent. In this case public employees have incentive 2

to support the incumbent to avoid the risk of being …red after the elections. Thus, the incumbent can improve her chances of staying in o¢ ce by raising the number of public employees above the e¢ cient level. This argument holds for both elected and appointed public o¢ cials. For elected politicians it increases their own chances of reelection. For appointed bureaucrats it raises their chances of staying in o¢ ce by increasing the reelection probability of the politicians responsible for their appointment. E¢ ciency costs of excessive public employment, however, limit the extent of patronage. Historically, American local governments are notorious for the widespread use of patronage. Distributing public jobs was the cornerstone of political machines that were in charge of city politics for many years. Jobs were given “as ‘favors’to persons who [were] expected to do no more in return than ‘show appreciation’ by voting the ‘right’way and by seeing to it that close relatives do so as well”(Ban…eld and Wilson, 1963: 208). Politicians themselves admitted the use of patronage. On a press conference mayor Richard J. Daley of Chicago told the journalists: “The governor in his message last week said 36 percent of city employees were patronage, and that is inaccurate, untrue and false. There is less than 8 percent.”1 Although such political machines were dismantled many years ago, there are reasons to believe that contemporary politicians follow their example today. Political patronage plays an important role in local politics and can have a signi…cant impact on the outcome of elections. First of all, local public employees constitute a substantial fraction of the voters. According to Census for the year 2000, more than 7 percent of the total labor force in the U.S. are employed by the local governments. Second, public employees are known to be more politically active and the voter turnout among the public employees is substantially higher than the average (Greene and Nikolaev, 1999; Corey and Garand, 2002). Finally, detailed analysis of the patronage system reveals that the social networks play an important role, and each public employee is likely to bring not only her own vote, but the votes of her relatives and friends as well (Chubb, 1982). All these e¤ects combined make the support of local public employees an important factor in determining the outcome of the elections. In this paper, I argue that directly elected chief executives of local governments are more likely to use patronage than their appointed counterparts. Both marginal bene…ts and marginal costs of using patronage are di¤erent for elected and appointed public o¢ cials. The marginal bene…ts are lower for the appointed bureaucrats since reelection of the public o¢ cials responsible for her reappointment does not necessarily guarantee that she stays in o¢ ce. Agency problems and especially common agency problems prevent the public o¢ cials responsible for her reappointment, who 1

Chicago Tribune, April 2, 1969.

3

bene…t the most from patronage, from completely aligning the interest of appointed bureaucrat with their own interests. The main cost of pursuing ine¢ cient vote buying policies, in addition to decreasing the support from the voters that are not bene…ting form patronage, is that it has a negative e¤ect on the perceived competence of a public o¢ cial and, thus, on the quality of the job she gets after leaving the o¢ ce. Appointed chief executives are much more likely than elected politicians to stay in public service rather than pursue career in private sector. Since performance as a chief executive of a local government is more indicative of the skills required in public sector then of the skills required in private sector, the negative e¤ects of patronage on the quality of the future job is likely be higher for appointed chief executives. Thus, appointed bureaucrats have both higher marginal costs and lower marginal bene…ts of using patronage, which makes them less likely to be engaged in vote-buying activities, as compared to elected politicians. In the paper, I develop a simple model of the local government to illustrate this point. The model produces two testable predictions. First, the level of public employment in the local governments headed by directly elected chief executives should be higher than in the local governments headed by appointed chief executives. Second, the di¤erence in the level of public employment between local governments headed by elected and appointed public o¢ cials increases with an increase in the costs of …ring public employees and the wage di¤erential between public and private sector jobs. Using panel data on the U.S. local governments, I …nd empirical support for both of these claims. To separate the explanation provided above from the alternative explanations for di¤erences in public employment policies in local governments headed by appointed and elected chief executives, I exploit the fact that full- and part-time employees are not equally well suited to be used for vote buying. For full-time public employees both the costs of …ring and the wage di¤erential between public and private sector is hire, which makes them more suitable to be used for patronage. Correspondingly, I …nd that local governments with elected chief executive have more full-time public employees than local governments headed by appointed chief executive. For the number of part-time employees, however, there is almost no di¤erence between the communities with di¤erent forms of government. I provide additional evidence that supports the main argument of the paper using information on the privatization of public service provision. Regardless of the form of government, privatization of a public service decreases the number of public employees involved in the production of the privatized service, but increases the number of public employees involved in the production of all other public services. The resulting e¤ect of privatization on total public employment, however, is negative

4

only in the local governments headed by appointed chief executives. Elected executives …nd a way of privatizing public services provision without a corresponding decrease in the total number of public employees. To address the issue of endogeneity of the institutional choice, I analyze the reasons behind the choice of the form of government. I show that despite the multitude of arguments pro and contra both forms of government, there is no clear prediction as of what form of government is preferable in which situation. There is no indication that there is a systematic di¤erence in the characteristics of the communities with di¤erent forms of government. There is also no evidence that endogeneity of the decision to change the form of government may bias the results in the model with community …xed e¤ects. Communities typically change their form of government in response to a …nancial crisis, but they do so regardless of the initial form of government. Overall, the empirical results are unlikely to be driven by the endogeneity of the form of government. The remainder of the paper is organized as follows: The next section discusses related literature. Section 3 presents the main argument of the paper. Section 4 presents the model. Section 5 describes the data and presents empirical results. Determinants of the form of government are discussed in section 6. Conclusions follow in section 7.

2

Related Literature

According to the classical approach representative democracy produces the policy preferred by the median voter. Since bureaucrats are themselves appointed by elected politicians, they will follow exactly the same policies as directly elected public o¢ cials, so that the method of selection of public o¢ cials has no e¤ect on policy outcomes (Baron, 1988; La¤ont, 1996). More recent literature, however, shows that the method of selection does play a role. According to Besley and Coate (2003) direct elections allow voters to unbundle policy issues and, thus, reduce the in‡uence of special interests. Maskin and Tirole (2004) argue that direct elections allows the public to screen and discipline public o¢ cials better. Thus, politicians follow more closely the preferences of the general population compared to their appointed counterparts, but are more likely to pander to public opinion and choose popular actions even when they know that they are not in the general interest of the society. Similar argument is developed in Vlaicu (2007). Alesina and Tabellini (2007a, 2007b) provide a general theoretical framework to analyze the criteria that guide the allocation of policy tasks to directly elected politicians or appointed bureaucrats. They argue that the main di¤erence lies in their motivation. Politicians care only about winning elections, whereas bureaucrats try to maximize their perceived competence. They

5

show that whichever method of selection of public o¢ cials is preferable depends on the features of the task they have to perform. Empirical studies provide evidence that the di¤erence in the method of selection of public o¢ cials does a¤ect the policies they pursue. Fields et al (1997) demonstrate that elected commissioners from the insurance industry follow more pro-consumer policies than their appointed counterparts. Besley and Coate (2003) …nd similar e¤ect in electricity regulation. There is also evidence that states with appointed judges have higher litigation rates (Hanssen, 1999) and lower levels of discrimination charges (Besley and Payne, 2003) compared to those with elected judges. Earlier works comparing performance of local governments headed by appointed and elected executives use much smaller data sets than the one I use in this paper and usually …nd no di¤erences. Linberry and Fowler (1967) …nd that municipalities headed by appointed executive were characterized by lower taxes and expenditures. Morgan and Pelissero (1980) …nd no signi…cant relationship between the form of the municipal government and …scal policies. Rauch (1995) does not …nd any signi…cant e¤ect on the share of municipal government expenditures allocated to infrastructure investment and Levin and Tadelis (2005) …nd no signi…cant di¤erence in the likelihood of contracting out public services in a cross-section of U.S. cities. A notable exception is Vlaicu (2007) who …nds pronounced electoral cycles in police o¢ cer hiring in mayor-council cities that are absent in council-manager cities. A number of works provide systematic evidence on the importance of political use of public employment in the U.S. cities (Alesina, Baqir and Easterly, 1998; Riordan, 1994; Glaeser and Shleifer, 2005) and southern Italy (Chubb, 1982). There are also theoretical arguments for why ine¢ cient income redistribution through public employment instead of direct transfers is used to get political support (Robinson and Verdier, 2003). Lopez-de-Silanes et al (1997) and Boyko et al (1996) demonstrate that the decision of local governments to privatize provision of public services is to a large extent motivated by political patronage considerations. An important question for the identi…cation strategy used in the empirical part of the paper is the reason for choosing a particular form of government. The only rigorous attempt to study the determinants of the form of government is provided by Alesina, Aghion and Trebbi (2007). The paper shows that ethnic fractionalization plays an important role in determining electoral rules. I take into account their results by controlling for ethnic fractionalization of the communities in the empirical analysis.

6

3

The argument

3.1

Why patronage?

Elected public o¢ cials have incentive to o¤er public policies to a certain group of voters in exchange for their electoral support. Such vote buying, however, is limited by the lack of commitment on both sides (Robinson and Verdier, 2003). Politicians face the commitment problem, since after the elections they lack incentives to implement the policy that was promised to attract the voters. To make it credible, politicians have to adopt the policy prior to elections and it should be costly to reverse it ex post. Adoption of the policies prior to elections solves the commitment problem on the side of the politicians, but raises the commitment problem on the side of the voters. If the policy is set prior to elections, they will vote for their most preferred candidate, rather than the politician who o¤ered the policy. To solve this commitment problem, the costs (bene…ts) of reversing the policy should be higher (lower) for the politician, who adopted the policy, so that the policy would be reversed if and only if she is not reelected. Patronage is an example of a policy that can be used to overcome the double commitment problems in vote buying. When wages in the public sector are higher than in the private sector, voters will support the candidate who credibly o¤ers them the prospects of working in the public sector. Hiring a person before the elections makes such a promise credible, as long as …ring of a public employee is costly. If di¤erent politicians draw support from distinctive groups of population (parties, ethnic groups etc.), the successful opposition candidate can …nd political bene…ts of hiring her own supporters to be higher than the costs of …ring public employees from the other group hired by her predecessor. Thus, public employees face the risk of being …red if the opposition candidate is successful, which gives them incentive to support the incumbent. The equilibrium level of patronage is determined by marginal costs and bene…ts of hiring an extra public employee for the public o¢ cial who determines the level of public employment. The bene…ts come from an increase in the number of votes from public employees and, possibly, their friends and family. The marginal cost of hiring an extra employee comes from a decrease in the e¢ ciency of public goods production, since by de…nition the pay for patronage is greater than the value of the public services performed. This leads to a loss of votes from all the citizens except for public employees. In addition, if there is an uncertainty about the abilities of public o¢ cials, lower e¢ ciency signals lower ability of the public o¢ cial, which can hurt her job-market opportunities.

7

3.2

Di¤erences between politicians and bureaucrats

Both appointed and elected public o¢ cials that have authority over public employment policy have incentive to use patronage to increase their chances of staying in o¢ ce. For elected politicians patronage directly a¤ects their chances of reelection. For appointed bureaucrats patronage a¤ects their chances of being reappointed by increasing the reelection chances of the elected public o¢ cials, responsible for their appointment. Both marginal costs and marginal bene…ts of using patronage, however, are likely to be di¤erent for politicians and bureaucrats. The bene…ts of using patronage are lower for a bureaucrat if there is a positive probability that she will loose o¢ ce even if the public o¢ cials responsible for her appointment are reelected. After the elections take place and all the bene…ts of patronage are reaped, the elected public o¢ cials can select their most preferred candidate disregarding the electoral support provided by the incumbent bureaucrat via patronage. In this situation there is another commitment problem, this time between bureaucrat and elected public o¢ cials responsible for her reappointment. O¤ering side payments conditional on the level of public employment may overcome this problem, but this compensation mechanism is likely to be limited by agency problems between bureaucrat and elected public o¢ cials responsible for her reappointment. As long as the bureaucrat has superior knowledge about the target level of public employment, side payments will not be enough to completely align the interests of the bureaucrat with the interests of elected public o¢ cials responsible for her reappointment. The problem is exacerbated by common agency problems when the bureaucrat is appointed by a body of elected public o¢ cials rather than a single elected public o¢ cial. If the interests of di¤erent public o¢ cials in this body are not aligned, the in‡uence of di¤erent members may cancel each other out (Dixit et al, 1997). The cost of using patronage is higher for the bureaucrats if the negative e¤ect of ine¢ cient policies on job-market opportunities is higher for them than for the politicians.2 Ine¢ ciencies in public goods provision a¤ect the perceived ability of a chief executive to run the government. The e¤ect of this perceived ability on job-market opportunities depends on the type of jobs public o¢ cial will seek after loosing o¢ ce. The closer are the skills at the jobs she will seek related to the skills required to run the government, the stronger will be the e¤ect. In particular, the e¤ect is likely to be stronger for the jobs in public sector than for the jobs in private sector. Since appointed chief executives are more likely to remain in public sector, the negative e¤ect of patronage on their job-market opportunities is likely to be 2 Similar e¤ects will take place if appointed public o¢ cials care more about their perceived competence because they care more about the assessment of their perfomance by professional peers (Wilson, 1989: Chapter 9) or try to secure their autonomy and independence (Carpenter, 2001).

8

higher.3 Overall, there are multiple reasons why appointed bureaucrats are likely to have both lower marginal bene…ts and higher marginal costs of using patronage and detailed discussion of these reasons goes beyond the scope of this paper. What is important for the purposes of this paper, is that the extent of patronage in the local governments headed by appointed bureaucrats is likely to be lower than in the local governments headed by elected politicians.

4

The Model

I develop a two-period probabilistic voting model of partisan politics that illustrates why di¤erences in the marginal costs and bene…ts of patronage between elected and appointed public o¢ cials lead to di¤erent levels of public employment. More importantly, the model demonstrates that the di¤erence in the level of public employment chosen by elected and appointed public o¢ cials is higher if the costs of …ring public employees is higher. These result are used to formulate testable empirical hypotheses.

Set up Population consists of a continuum of voters that belong to one of the two groups labeled A and B. The mass of voters is normalized to unity and both groups are of equal size 12 :4 Identity of public o¢ cials and voters is observable, so that actions can be conditioned on the group they belong to. The budget is …nanced through a lump-sum tax on individuals , which is the same for all the citizens. The level of taxes is determined exogenously and can be thought of either being set by law or determined by interjurisdictional competition. Since the number of people is normalized to unity, the total size of the budget equals :5 Production function for the public good is linear in the size of the budget gt = G(Et ), where Et 2 [0; 1] is the share of the budget spent on public employment and G(Et ) is a smooth function such that G0 (0) = 1; G0 (1) = 3

1; G00 < 0:

Table AI in the appendix provides some evidence that support this claim for the case of local governments. More than 90% of the city managers remain in the public sector, where as the corresponding number for the mayors is only 32%. 4 Alowing the size of the groups to be di¤erent does not change the qualitative results of the model. 5 I assume that there is no budget de…cit. This assumption does not a¤ect the results as long as the ability to run budget de…cit does not di¤er between local governments headed by appointed and elected public o¢ cials. The laws that regulated the ability of the local governments to run budget de…cit, however, are the same for the all the forms of government. In addition, empirically I …nd no signi…cant di¤erence in the sized of the budget de…cit between local governments headed by appointed and elected chief executives.

9

These conditions imply that there is an optimal share of the budget spent on public employment E 2 (0; 1) that maximizes production of the public good.

The wage in the public sector is …xed and equals W , so that public employment

is et =

Et W

:The wage in the private sector is also …xed and equals W (1

); where

2 (0; 1). Important feature of the model is that the wage of public employees is

higher than the wage in the private sector, so that a chief executive can o¤er rents to public employees in exchange for their electoral support. Firing is costly and if a public employee is …red he incurs a cost of F = W f .

6

There is no unemployment

and if a public employee is …red, she moves to the private sector. Utility of the voter j that belongs to group i 2 fA; Bg in period t is: i i Ut;j = wt;j

+ gt

I(f ired)F

(1)

i is her wage and I(f ired) is an indicator function that equals one if she where wt;j

was …red in period t and zero otherwise. Local government There are two forms of local government that di¤er in the method of selection of chief executive. In one case, the voters elect the chief executive directly. In the other case, the voters elect the council, and the council appoints the chief executive. I assume away the problem of collective decision making within the council by assuming that the council consists of only one member. During the elections the incumbent faces a challenger that belongs to the other group. The newly elected council decides whether to reappoint the chief executive. No ex-ante policy commitment is possible, so that once in o¢ ce a candidate chooses her most preferred policy. The reelection probability is determined by a probabilistic voting model. Voter j supports the incumbent public o¢ cial if i U2;j (incumbent wins) +

where

j

uted on

j

+

i U2;j (challenger wins)

(2)

is an ideological bias of voter j towards the incumbent uniformly distrib1 1 2s ; 2s

, and

uniformly distributed on

is a general popularity shock in favor of the incumbent 1 1 2h ; 2h

:

Timing There are two periods.7 In the …rst period the incumbent chief executive chooses B the level of public employment for each group in the …rst period eA 1 and e1 ; the 6

Similar results can be obtained if we assume that …ring leads to a loss in the size of the budget available for public good production and/or the utility of the public o¢ cial. 7 Extending the model to an in…nite horizon does not change qualitative results, but complicates the exposition of the model. For these reasons I consider only a two-periods model.

10

production takes place and the …rst period payo¤s are realized. In the beginning of the second period the voting takes place and the new local government is formed. In the case of elected chief executive she is elected directly. In the case of appointed chief executive the council is elected and after that the council chooses a chief executive. After the new local government is formed the chief executive decides on the B level of public employment in the second period, eA 2 and e2 ; the production takes

place and the second period payo¤s are realized. There is no time discounting. Public o¢ cials All public o¢ cials (chief executives and members of council) have the same utility. Public o¢ cial that belongs to group i 2 fA; Bg has per-period utility 8 R > U i dj > > < j2i t;j R Vti = Ut 1;j dj > > > : 0

if in o¢ ce (3)

if loses o¢ ce never been in o¢ ce

Following standard assumptions of the partisan politics models (Alesina and Tabellini, 1990; Persson and Svensson, 1989; Persson and Tabellini, 2000) public o¢ cials care only about the utility of the voters that belong to their own group. Coe¢ cient

2 (0; 1) captures the di¤erence in the importance of the career

concerns. In the context of the model the results of the work that in‡uence public

o¢ cial’s future payments if she loses o¢ ce is the total utility of the citizens. An important assumption of the model is that for the appointed chief executives the e¤ect of their performance while in o¢ ce on the quality of the job they get if they loose o¢ ce is stronger than for the elected politicians

B

>

P

. Note that the

fact that they were in public o¢ ce may have di¤erent e¤ect on the quality of the job that bureaucrats and politicians get. What is important is that the quality of the job that politicians get once they are out of o¢ ce does not depend as strongly on the quality of their performance while in o¢ ce, so that the marginal costs of patronage are higher for appointed public o¢ cials. I assume that

is small enough

so that for all the relevant parameters’values public o¢ cials prefer to stay in o¢ ce: R i R Ut;j dj > Ut 1;j dj: j2i

Similar results can be obtained if we assume that the utility of a public o¢ cial

if she looses o¢ ce depends on the weighted sum of the utilities of the citizens that belong to both groups, but for politicians the utility of the members of their own group enters with a higher weight, since they are more likely to seek a job in their own constituency (party, ethnic group etc).

11

Equilibrium policy The subgame perfect Nash equilibrium of the game can be found by backward induction. Since the costs of …ring a member of the other group are lower than the bene…ts of hiring a member of her own group, chief executive will always prefer to replace a member of the other group with a member of her own group, so that in each period chief executive hires only the members of her own group.8 Votebuying considerations cause the incumbent chief executive to set the level of public employment above the level that maximizes the utility of the members of her own group. If the chief executive in the second period belongs to the same group as the chief executive in the …rst period, the employment does not change. This happens if the incumbent elected public o¢ cials (chief executive or council members) are reelected. If they loose o¢ ce, the new chief executive belongs to a di¤erent group and …res all the public employees, since none of them belongs to her own group, and hires only the members of her own group. The new chief executive, however, sets the level of public employment at the level that maximizes the utility of the members of her own group, since it is the last period and she no longer takes vote-buying considerations into account. Formally, denote E and E such that G0 (E) = Assume that

E W


f . Appendix 1 shows that the council will always appoint the chief executive from its own group. There is no guarantee, however, that it will reappoint the chief executive, since they may replace him with other bureaucrat belonging to the same group. Denote p the probability that chief executive is reappointed if council is reelected. Assume that p < 1; so that there is some chance that chief executive will not be reappointed even if the council is reelected. In this case the marginal bene…ts of patronage will be lower for appointed chief executive, since an increase in the reelection probability of the elected public o¢ cials does not translate one to one to her chances of staying in o¢ ce. De…ne the function (E) =

1 + h (E ( + f ) 2

E + G(E)

G(E)) :

(4)

Appendix 1 proves that under these assumptions the following strategies constitute the subgame perfect Nash equilibrium of the game. In the …rst period the 8 We can avoid an extreme prediction that the new chief executive replaces all the old public employees with the members of his own group by assuming some distribution of the costs of …ring such that the new chief executive will lay o¤ only the pubic employees with su¢ ciently low costs of …ring.

12

incumbent chief executive that belongs to group i sets the level of public employment for the members of her own group and the group she does not belong to equal to ei1 =

and e1 i = 0 respectively. Here E 2 E; E is such that

E W

[1 + p (E ) + ph f + + G0 (E )

= where

(1 E

p (E ))] +

1 (W (1 2

1 + G0 (E ) 2 )

=

+ G(E )) (1

(5) )

(E ) is the probability of the incumbent elected public o¢ cial (either chief

executive or member of council) staying in o¢ ce. The left-hand side of this expression captures the e¤ect of public employment on the utility of chief executive holding constant the probability of staying in o¢ ce, whereas the right-hand side re‡ects the e¤ect of public employment on the probability of staying in o¢ ce. In the model chief executive increases public employment above the optimal level only to improve her chances of staying in o¢ ce. If the probability of staying in o¢ ce is exogenously given (so that the right-hand side of (5) equals zero), public employment will be set at the level E that maximizes the utility of the members of the group, the chief executive belongs to. If the incumbent chief executive stays in o¢ ce (or if the appointed chief executive is substituted with the member of the same group), in the second period she sets e2 i = e1 i = 0 and ei2 = ei1 = e : If the incumbent is not reelected, the new chief executive sets ei2 = 0 and e2 i = e. Proposition 1 Equilibrium level of public employment in the …rst period is higher if the chief executive is elected than if she is appointed. Proof. In the interval E; E all the conditions of the Implicit Function Theorem for (5) are satis…ed, so that that we can use it to show that Since

P


0:

and p < 1, elected chief executive will choose higher level of public

employment in the …rst period. An important feature of the model that makes using public employment as a vote buying mechanism possible is that the …ring of public employees is costly. This makes it costly to reverse a high level of public employment adopted before the elections and makes the o¤er of post-election rents credible. The costs of …ring, however, should not be too high, since in this case public employees realize, that they will not be …red even if an incumbent is not reelected, which undermines their incentives to support her. In equilibrium, the cost of …ring does not a¤ect the utility of public o¢ cials other than by changing the level of public employment, since …ring occurs only if the incumbent loses o¢ ce. The higher the cost of …ring, the higher the

13

level of public employment that can be credibly supported

dE df

> 0 : In addition,

the following proposition holds: Proposition 2 Di¤ erence in the levels of public employment chosen by appointed and elected chief executives increases with the cost of …ring. Proof. In the interval E; E all the conditions of the Implicit Function Theorem for (5) are satis…ed, so that that we can use it to show that P

Since


0 and

d2 E df dp

< 0:

and p < 1 di¤erence in the levels of public employment chosen by

appointed and elected chief executives will be higher if the …ring cost f is higher. We can also show that bigger di¤erence between the wage in public and private sectors, which makes public employment more attractive, leads to higher level of public employment and bigger di¤erence in the level of public employment between local governments headed by elected and appointed chief executives. Proposition 3 Di¤ erence in the levels of public employment chosen by appointed and elected chief executives increases with the wage di¤ erential between public and private sector. Proof. In the interval E; E all the conditions of the Implicit Function Theorem for (5) are satis…ed, so that that we can use it to show that P

Since


0 and

d2 E d dp

< 0:

and p < 1 di¤erence in the levels of public employment chosen by

appointed and elected chief executives will be higher if the wage di¤erential

is

higher.

4.1

Testable predictions

Proposition 1 implies that local governments headed by elected politicians will have higher levels of public employment than the local governments with appointed chief executives. A necessary condition for using patronage for vote buying is that the …ring of a public employee is costly. This condition is satis…ed for full-time public employees, since their …ring requires paying out substantial bene…ts. Thus, we can formulate an empirical hypothesis that corresponds to Proposition 1. Hypothesis 1. The number of full-time public employees in local governments headed by elected chief executives is higher than in local governments headed by appointed chief executives. This prediction relies on the assumption that absent patronage there are no systematic di¤erences in the level of public employment between local governments with appointed and elected chief executives. Clearly, there are many reasons why this assumption may be violated. Appointed chief executives can be more e¢ cient in public employment management, so that they can provide the same level of public 14

goods using fewer employees. Alternatively, appointed chief executives can be more susceptible to the in‡uence of special interests that lobby for higher expenditures on purchases of equipment, services etc., at the expense of lower level of public employment. For this reason I formulate additional empirical hypotheses that are in line with the vote-buying story, but are not consistent with the alternative explanations. According to Proposition 2 the di¤erence in the level of public employment should be lower for public employees that are easier to …re. For part-time employees the costs of …ring are signi…cantly lower that for the full-time employees, since most of them have short-term contracts and cannot claim the same bene…ts as full-time employees if they are …red. Also, the wage di¤erential is also likely to be higher for full-time employees, since part of it comes from fringe bene…ts that part-time employees are not getting. According to Proposition 3 it will also the di¤erence in the level of public employment bigger for the case of full-time employees. Thus, we can formulate an empirical hypothesis that corresponds to Propositions 2 and 3. Hypothesis 2. The di¤ erence between the number of part-time public employees in local governments headed by elected and appointed chief executives is smaller than in the case of full-time employees. The logic outlined above can be extended to make predictions about the relationship between privatization of public services provision and public employment. The main argument for privatization is that it helps to achieve greater e¢ ciency, primarily by cutting labor costs (Savas, 1987, Donahue 1989, Kemp 1991). During the end of the 1980s and the beginning of the 1990s public opinion was strongly in favor of privatization and there was a general trend towards decreasing in-house provision of public services. Public o¢ cials, however, may …nd a way to satisfy public opinion without damaging their political in‡uence by privatizing the provision of public services without a corresponding decrease in public employment. A way of achieving this goal is reallocation of public employees formerly engage in the provision of the privatized public service to the provision of other public services. If elected public o¢ cials are more concerned with mitigating the e¤ect of privatization on the level of public employment, the e¤ect of the number of privatized public services on the level of public employment should be weaker in local governments headed by elected public o¢ cials. Hypothesis 3. The negative e¤ ect of privatization of public services on the number of full-time public employees is smaller in local governments headed by elected chief executives.

15

5

Empirical evidence

5.1

Background information on the forms of government

In the U.S. local communities can choose their form of government in a local referendum. For the purposes of the paper I consider only one di¤erence in the form of government: whether the chief executive o¢ cer is elected or appointed. Local governments with appointed chief executives include municipalities and counties with a council-manager (administrator) form of government. In this type of government, an elected council sets a policy, adopts legislation, and approves the budget. It appoints a manager or administrator who has full responsibility of the day-to-day operation of the local government and has authority to oversee department heads, hire and …re local governments’sta¤ (often including department heads), recommend policy to the council and prepare the budget. Local governments with elected chief executives include municipalities with a mayor-council form of government and counties with a council-elected executive form of government. In this type of government, an elected council serves as a legislative body with a chief executive being elected separately. As in the previous case, the council sets policy, adopts legislation, and approves the budget. The chief executive implements policies adopted by the council, prepares the budget, and exercises some authority over day-to-day operations of the local government, the extent of which varies from jurisdiction to jurisdiction. Communities with other forms of government9 (approximately one third of the total number of local governments) are excluded from the analysis since their form of government cannot be easily interpreted in terms of comparing appointed and elected public o¢ cials.

5.2

Data

I collect the data on the form of government, local public employment and various other characteristics of the local communities for the years 1987, 1997 and 2002. More information on the data sources, variables’ de…nitions and coverage can be found in Appendix II. I restrict the analysis to the communities with population above 10000 inhabitants for two reasons. First, the sample is not representative for smaller communities (see Appendix II). Second, in smaller communities endogeneity of the form of government is likely to a¤ect the results, since among the smaller communities only the rich ones can a¤ord hiring a city manager . Out of 3514 local governments in the sample, for 1575 governments the information on the form of government is available for at least two years. Of these, 245 9

These include municipalities with commission, town meeting and representative town meeting forms of government and counties with commission form of government.

16

local governments changed their form of government (see Table 1). Out of 1028 local governments headed by an appointed chief executive, 66 local governments changed their form of government to that of elected chief executive. Out of 547 local governments that had elected chief executives, 179 local governments changed their form of government to that of appointed chief executive. For all the local governments in the sample, changes in the form of government occur between 1987 and the subsequent years.

5.3

Methodology

Changes in the form of local governments allow me to conduct analysis using community …xed e¤ects. The bene…t of the within analysis is that it allows controlling for the community characteristics that do not change over time. The main drawback of this approach is that it limits the analysis of the direct e¤ect of the form of government to those communities that have changed it during the period under consideration. Since the reasons for these changes might be di¤erent from the historical determinants of the form of government in total population of the communities, I report the results both with and without community …xed e¤ects. An important feature of the data is that the e¤ect of covariates is nonlinear. To address this issue I use covariates matching estimation. Since large number of covariates leads to a signi…cant bias in simple matching estimation if matching is not exact (Abadie and Imbens, 2006) I use a bias-adjusted matching estimator (Abadie et al, 2004). In addition, I report the results of a more conventional regression analysis for comparison. For matching estimation of the between e¤ect of the form of government on public employment I use the logarithm of the number of public employees (either full-time or part-time) as a dependent variable and perform the nearest neighbor matching on the set of covariates which includes the measures of population, income per capita, total budgetary expenditures, income inequality (measured by the ratio of mean and median household income), level of unemployment, ethnic fractionalization, fraction of population 65 years and older, fraction of population above 25 years old with at least 12 years of education, and fraction of urban population, the number of years since incorporation, a dummy variable that equals one for a homerule community, and a dummy variable for counties. I subtract state averages from the dependent variable prior to matching. I follow this approach, rather than directly matching on state variable because of the data constraints that do not allow me to match communities only within the same state. Matching communities from di¤erent states without explicitly controlling for di¤erences between the states is likely to bias the results, since di¤erent states vary signi…cantly with respect to the forms of government preferred by the local communities and the legislation regulating public 17

employment. The standard errors for the treatment e¤ect will be somewhat biased since they do not take into account the initial manipulation with the data and the panel structure of the data. To address the latter problem I reproduce the analysis for each year separately (see section 5.5). In the case of covariates matching the matching on year and type of community is exact. In the corresponding regression analysis I use the following linear panel model: yit = Electedit

1

+ Xit0

2

+ ds +

t

+ "it

(6)

where yit is a measure of public employment in community i at time j, Electedit tells whether the chief executive of the community i at time j was elected, ds is a state dummy, and Xit is a vector of covariates describing characteristics of community j at time t which includes the same covariates as in the matching estimation. To take into account the panel structure of the data the errors are clustered at the community level. To estimate the within e¤ect of the form of government on public employment using covariates matching I …rst construct di¤erences in all the covariates and the measure of public employment between the year 1987 and the subsequent year (if the data is available for both 1997 and 2002 I take the average). After that, I estimate the average treatment e¤ect of a switch from appointed to elected chief executive and in the opposite direction separately, using nearest neighbor matching on the covariates. The set of covariates includes the measures of population, income per capita, total budgetary expenditures, income inequality (measured by the ratio of mean and median household income), level of unemployment, ethnic fractionalization, fraction of population 65 years and older, fraction of population above 25 years old with at least 12 years of education, and fraction of urban population. The standard errors for the treatment e¤ect will be somewhat biased since they do not take into account the initial manipulation with the data. In the regression analysis of the within e¤ects I use the following linear panel model: yit = BecomesElectedit

1

+ BecomesAppointedit

2

^0 +X it

3

+

i

+

t

+ "it (7)

where yit is a measure of public employment in community i at time j: The variable BecomesElectedit equals one if in community i at time t the chief executive was elected, but previously the chief executive was appointed. The variable BecomesAppointedit equals one if in community i at time t the chief executive was ^ it is a vector of elected, but afterwards the chief executive becomes appointed. X covariates describing characteristics of community j at time t which includes the same covariates as in the matching procedure described above, and

18

i

and

t

are

community and time …xed-e¤ects respectively. Since there are good reasons to expect downward rigidity of public employment the e¤ect of a switch from appointed to elected chief executive might be di¤erent from the e¤ect of a switch in the opposite directions. For these reasons I consider these two types of changes separately, rather than including a single indicator for the from of government. To estimate the e¤ects of the privatization on public employment I use the following linear panel model:

yitg = Electedit P rivatizeditg ~0 +X itg

4

1

+ Appointedit P rivatizeditg

2

+

(8)

+ "it

where yitg is a measure of public employment in public service g (or in all public services but g) in community i at time j, the dummy variable P rivatizeditg tells whether the provision of public services g in community i at time t is priva~ itg includes all the variables in (6) and the tized, Appointedit = 1 Electedit and X dummy variables for community, year, public service, community-public service and communtiy-year in di¤erent combinations. To estimate the e¤ect of privatization on total public employment I use as the dependent variable yit –the measure of the total number of employees in community i at time j.

5.4

Empirical results

Results of the estimation without community …xed e¤ects demonstrate that elections of chief executive are associated with signi…cantly higher numbers of full-time public employees (see Table 2). The results of the matching estimation indicate that the number of full-time public employees is 20 percent higher in communities headed by elected public o¢ cials. The magnitude of the e¤ect in the regression model is smaller, but still statistically signi…cant: public employment is higher in communities headed by elected public o¢ cials by 4 percent. Pairwise analysis of the relationship between the covariates and the dependent variable indicates that it is highly nonlinear, so that the results produced by the matching estimators are likely to be more accurate estimates of the e¤ect. Results for the …xed-e¤ects model demonstrate that a change from chief executive being appointed to being elected is associated with a signi…cant increase in the number of full-time public employees. The magnitude of the e¤ect is sizable and very similar for the regression analysis and covariates matching: the number of public employees increases by 8 percent. Not surprisingly, public employment demonstrates downward rigidity and a change in the form of government in the opposite direction does not lead to a decrease in public employment. Instead it leads to

19

an increase in public employment. which is signi…cant in the regression analysis, but smaller in magnitude and not signi…cant in the case of covariates matching. Comparison of the results with and without community …xed-e¤ects indicates that the di¤erences in the level of public employment between the communities with elected and appointed chief executives is larger than the short-run e¤ects of a change in the form of government in a community. Overall, the results con…rm Hypothesis 1 that local governments headed by elected chief executives are characterized by higher level of public employment. Consistent with Hypothesis 2, analogous results for the number of part-time employees are much weaker. Without community …xed-e¤ects, covariates matching results are signi…cant, but are almost twice smaller in magnitude than in the case of full-time employees (see Table 3), whereas regression results are insigni…cant and almost indistinguishable from zero. Controlling for the community …xed-e¤ects, the results are insigni…cant and have the wrong sign in both regression and matching estimations. Overall, the results support Hypothesis 2 that the e¤ect of the form of government on the number of part-time employees is much smaller than on the number of full-time employees. Table 4 reports the estimates of the e¤ect of the form of government on public employment in each type of public services separately without community …xed e¤ects. The results demonstrate that elected public o¢ cials hire more full-time public employees in all public services. The e¤ect is the strongest for …re…ghters and the smallest for the employees working in utilities provision. The di¤erence in the number of part-time employees is signi…cant in half of the cases. For employees in education and health and social services the di¤erence is approximately twice smaller compared to the di¤erence for full-time employees, whereas for employees in environmental and housing services, as well as government administration the di¤erence is almost the same as for full-time employees. For other services the di¤erence is much smaller and not statistically signi…cant. Results of the estimations with …xed e¤ects are less conclusive. They are signi…cant only for the changes in the number of full-time …re…ghters and employees in transportation and transit resulting from the change from chief appointed being appointed to being elected, but have the wrong sign indicating that the number of public employees in these services decreases after the change in the form of government. For the number of part-time public employees there are no signi…cant results. There is strong evidence that the method of provision of public services has signi…cantly di¤erent e¤ect on public employment in communities with elected and appointed chief executives (see Table 5). Regardless of the form of government privatization of a public service leads to a decrease in the number of public employees involved in the production of this public service and increase in the number of public

20

employees involved in the production of other public services. The e¤ect on the total number of public employees, however, depends on the form of government. In communities with appointed chief executive, privatization leads to a signi…cant decrease in the total number of public employees, whereas in communities with elected chief executive there is no signi…cant relationship between privatization and public employment. In other words, only elected o¢ cials …nd a way to privatize provision of public services without reducing the number of public employees. Thus, the empirical results support Hypothesis 3.

5.5

Sensitivity analysis

I consider the estimates obtained by covariates matching as the benchmark results for the estimates of the direct e¤ect of the form of government on public employment and perform sensitivity analysis only for this set of results. I check how well the matching estimator performs in terms of balancing the covariates and test robustness of the results with regard to alternative estimators. 5.5.1

Estimation technique

Table 6 reports di¤erences in covariates between local governments with appointed and elected chief executives in the original sample and within the matched pairs. Even before matching communities with elected and appointed chief executives are not very di¤erent. The di¤erence in the averages between the two groups is never more than 0.5 of standard deviation. Within the matched pairs the di¤erence decreases even more and never exceeds 0.08 of standard deviation. Overall, the quality of matches appears very high. To see if the results are contaminated by the lack of overlap in the covariate distribution between communities with appointed and elected chief executives, I calculate the propensity score for all the observations based on the same set of covariates as in the benchmark model and exclude those observation for which the propensity score is either smaller than 0.1 or greater than 0.9. After that I repeat all the estimations on this restricted subsample. The results prove to be very close to the results on the whole sample in terms of size and signi…cance, so that the lack of overlap in the covariate distribution does not a¤ect the results. To assess the unconfoundedness assumption for the matching results with community …xed-e¤ects, I compare the level of public employment in 1987 in communities that later on experience a change in the form of government with those communities that do not experience such a change. I …nd no signi…cant di¤erence between such communities. Thus, there is no evidence that the unconfoundedness assumption is violated.

21

Since matching estimations do not take into account the panel structure of the data, they are likely to underestimate standard errors of the estimated e¤ects. To address this issue, I restrict the sample to observations in each of the three years separately. The results remain signi…cant in each of the three subsamples (see Table 7). The magnitude of the e¤ect does not change much from year to year, with the estimates for 2002 being somewhat larger. In addition, I compare the results obtained using di¤erent number of matches. Increasing the number of matches does not a¤ect the signi…cance of the results and leads to somewhat larger estimates of the e¤ect. To check the robustness of the results with regard to alternative estimators I estimate the treatment e¤ect on the treated using covariates matching, as well as propensity score matching. The results for the number of full-time employees remain highly signi…cant, but somewhat smaller in magnitude. The results for the number of part-time employees remain signi…cant in the case propensity score matching, but lose their signi…cance in the case of covariates matching. Overall, the estimates prove to be extremely robust to restricting the sample to include only observations from one year, increasing the number of matches and using alternative matching estimators. 5.5.2

Covariates

The set of covariates in the benchmark results does not include the share of privatized public services, since the privatization of public services is itself a choice variable for the chief executives that can also be a¤ected by the vote buying consideration. However, as noted in the discussion of the Hypothesis 3, in some cases privatization can be driven by popular demand, which can be di¤erent in communities with di¤erent forms of government. To take these considerations into account I check whether the results are robust to controlling for the extent of privatization. All the results preserve the same level of signi…cance and remain quantitatively similar after including the share of privatized public services in the set of controls. In the estimation without community …xed-e¤ects the e¤ect becomes stronger for both fulland part-time employees, whereas the di¤erence between them becomes somewhat smaller. There is no noticeable di¤erence in the results with community …xed-e¤ects. Overall, the results prove to be robust to including the share of privatized public services in the set of controls. The results for the e¤ect of privatization on the total number of employees are robust to the inclusion of community-service …xed e¤ects. The results for the e¤ect of privatization of a public service on public employment in other services loose their signi…cance in this case. The same results, however, along with the results for the e¤ect of privatization of a public service on public employment in the same services, 22

prove to be robust to the inclusion of the community-year …xed e¤ects.

6

Determinants of the form of government

An important caveat is that I cannot fully control for the endogeneity of the choice of the form of government. To see whether the endogeneity of the form of government is likely to bias the results, …rst, I analyze the local newspapers’ reports in the communities that experienced an attempt to change the form of government to show the main arguments used by the proponents of the di¤erent forms of government and the reasons that lead to the attempts.10 Second, I provide a regression analysis of the causes that lead to a change in from of government in local communities in my sample. The main argument used by the proponents of the appointment of chief executive is that it helps to depoliticize the work of the local government and decrease corruption. The proponents of the election of chief executives, however, paralleling arguments in the political economy literature, claim that appointed executives are not directly responsible to the population, and thus, the in‡uence of special interests in this case is stronger. Another argument used by the supporters of direct elections is that such a structure separates the executive and legislative branches of government and introduces a system of checks and balances similar to the one at the federal level. Both sides claim that their preferred form of government guarantees a higher level of accountability. Supporters of election of the chief executive stress the fact that direct elections make the o¢ ce-holder responsible to the whole population rather than the several council members. Supporters of the appointment of executives point out that they can be …red whenever their performance is judged unsatisfactory, whereas in the case of an elected executive, voters have to wait until the next election or initiate recall procedure, which is extremely rare at the local level. The ‡ip side of this argument is that the horizon of planning for the appointed executives may be shorter, which would undermine their ability for long-term planning.11 Another argument of the supporters of appointment of the chief executive is that the candidates for the job are better quali…ed, since most of the communities require them to have special education or several years of experience at a similar job. Appointed executives usually are outsiders that have not previously lived in 10 I have analized all the articles in the local newspapers that mention a change in the form of local government using LexisNexis Academic. Of these, 165 acticles provide information on the determinants of the attempted change in the form of local government. 11 It is hard to substantiate the “short horizon”argument and compare the average tenure length of the appointed and elected executives, since there is no readily available data on the average tenure of the elected executives. According to the surveys of the International City/County Management Association for the appointed executives the average tenure in o¢ ce is 6.9 years.

23

the community, which is viewed as a virtue by the supporters of the appointment, since in this case the o¢ cials do not have any local allegiances; at the same time, supporters of elections regard this as a drawback, since the o¢ cials do not know local conditions and speci…c problems of the community. Newspaper reports also provide information on the reasons that lead to the attempts to change the form of government. Typically, a change in the form of government follows a …scal crisis. As political scientist Steve Erie has commented on the change of appointed manager to elected mayor in Richmond, VA: "You don’t get people demanding this kind of change unless you’ve got real corruption or a …scal crisis."12 Importantly, …nancial crisis facilitates the change in the form of government regardless of the actual direction of change. Bromage (1940) analyzes the causes of the abandonment of the city manager form of government in the U.S. municipalities in 1920s and 30s. He …nds that the main reason for changing the form of government is a …nancial crisis in a community, which is often out of control of the local government. Regression analysis of the determinants of change in the form of government shows that large budgetary de…cit, measured as a share of total expenditures signi…cantly increases the probability of a change in the form of government in either direction, whereas higher wealth of the population in the communities decreases this probability (see Table 8). The fact that the number of public employees a¤ects the probability of a change in the form of government indicates that public employees correctly predict the e¤ect of the form of government on public employment which makes them willing to support election of chief executive. Di¤erences in the e¤ect of the number of part-time and full-time employees are also consistent with the model. Since part-time employees are the likely winners of the expansion of full-time public employment after a switch from appointed to elected chief executive, they support such a change in the form of government. Although I do not …nd evidence that a switch from elected to appointed chief executive leads to a decrease in the level of full-time public employment, the full-time employees themselves might fear this possibility, so that they oppose this change in the form of government. Overall, there is no conclusive evidence that one form of government is preferable to another neither in general, nor in any particular situation. This corresponds well with the fact that there is very little di¤erence in the socioeconomic characteristics of the communities with di¤erent forms of government in the data (Table 6). Thus, we might expect the choice of the form of government to be determined by the particularities of the political situation at the moment the choice was made, rather than any underlying characteristics of the communities, which might a¤ect 12

See “Strong feelings and Strong Mayors”by David Ress, Richmond Times-Dispatch, December 19, 2004.

24

the empirical results of the paper. The available data, however, does not allow me to rule out this possibility completely.

7

Conclusions

The paper argues that elected and appointed public o¢ cials pursue systematically di¤erent policies with regard to public employment. Taken together, the empirical results provide a strong support for the claim that elected chief executives in the U.S. local governments use patronage more than their appointed counterparts. More generally, the results suggest that elected politicians are more likely to be engaged in vote buying than bureaucrats. The results can be generalized to other settings in which public o¢ cials can use ine¢ cient policies to gain popular support that would help them to stay in o¢ ce. The argument in the paper, however, is more relevant for the top ranked, rather than lower level public o¢ cials. First, they are more likely to have opportunities to be engaged in vote buying. Second, incentives of the lower level public o¢ cials are likely to be systematically di¤erent and much strongly a¤ected by the relationships with immediate superiors in party or bureaucratic organization. There is an ongoing discussion on the trade-o¤ between electing and appointing public o¢ cials. There are many arguments pro and contra each of the options with none of them having a clear comparative advantage. Maskin and Tirole (2004) argue that the main trade-o¤ is between accountability and intrinsic motivation of the public o¢ cials. Besley and Coate (2003) note that direct election help to unbundle the issues the public o¢ cial is responsible for. Alesina and Tabellini (2007a, 2007b) provide a comprehensive analysis of the relative performance of appointed and elected public o¢ cials in a wide range of tasks. The results of this paper o¤er an additional insight to the discussion, by providing evidence that elected politicians are more prone to using patronage to increase their chances of staying in o¢ ce. The paper, however, does not explicitly model or test empirically the theories that explain the di¤erence in bene…ts and costs of patronage between appointed and elected public o¢ cials, leaving this analysis for the future work.

25

Appendix I Since there is no commitment, chief executive in the second period chooses the level of public employment to maximize her utility V2i = E2i

+

1 (W (1 2

)

+ G(E2 )

F Lt )

(9)

where E2 = E2i + E2 i is the total public employment in the second period. Since 2 > f , the costs of …ring a member of the other group are lower than the bene…ts of hiring a member of her own group, so that regardless of the employment in the …rst period the executive will not hire anyone from the group she does not belong to E2 i = 0 , so that E2 = E2i . Knowing that, the council will always appoint the chief executive from its own group. Since the marginal cost of increasing public employment is di¤erent from the marginal cost of decreasing public employment, the level of employment of the members of her own group will depend on the level of employment in the previous period:

E2i (E1i ) =

8 > < > :

if E1i < E

E E2i = E1i E

if E1i 2 E; E if

E1i

(10)

>E

The outcome for the voters depends on the identity of the incumbent. In addition, if the incumbent is reelected, the outcomes in the second period will depend on the policies chosen in the …rst period. This allows the incumbent to a¤ect her chances of staying in o¢ ce by changing the …rst period policies even in the situation when she can not commit to future policies. If the incumbent that belongs to group i is not reelected, the new executive will set E2 i = E and E2i = 0: It is easy to check that E1i 2 = E; E is not an equilibrium strategy for the incumbent, so that E1i 2 E; E and E1 = E1i = E2i = E2 . In this case Lt = 0 and V1P;i = V2P;i .

Since the council always appoints chief executive from its own group, voting

for the council member becomes equivalent to voting for the group to which the chief executive belongs. Total number of people that would vote for the incumbent consists of the voters that belongs to the other group, public employees (all of whom belong to the same group as the incumbent), and the members of the same group that are not public employees:

26

N =N

i

i + NGi + NN G =

1 2

1 + s ( + [G(E1 ) 2

G(E)

2E ]) +

(11)

1 + s ( + W (f + ) + [G(E1 ) G(E)]) + 2 1 E1 + s ( + [G(E1 ) G(E)]) = W 2

+E1

W 1 + 2 1 = + s [ + (E1 (f + ) 2

E + G(E1 )

G(E))]

E + G(E)

G(E))

So that the reelection probability is given by (E) =

1 + h (E (f + ) 2

(12)

Where E = E1 = E2 . In the case of appointed chief executive the probability that she stays in o¢ ce is p (E). The incumbent chief executive chooses the level of public employment in the …rst period that maximizes her expected utility in both periods U = V1i (E) + p (E) V2i (E) + (1

p (E)) V1i (E)

(13)

where p = 1 in the case of elected chief executive. In the relevant range where E1 2 E; E V2i (E) = V1i (E) ; so that (13) simpli…es to U = V1i (E) [1 + p (E) +

(1

p (E))]

(14)

The level of public employment in both period is determined by the …rst order condition

[1 + p (E ) +

(1

p (E ))]

[1 + p (E ) + =

ph f + + G0 (E )

(1 E

@V1i (E ) = @E

p (E ))] +

1 (W (1 2

@ (E ) i pV1 (E ) (1 @E

1 + G0 (E ) 2 )

),

=

+ G(E )) (1

(15) )

Thus, in equilibrium the incumbent chief executive that belongs to group i will set the level of public employment e1 i = 0 and ei1 = (15) and

E W

: Where E is is given by

(E ) is determined by (12). If the incumbent chief executive stays in

o¢ ce, she will set e2 i = 0 and ei2 = ei1 =

E W

27

: If not, the new chief executive (who

belongs to group

i) will set ei2 = 0 and e2 i =

E W

.

Appendix II Sources The data on the form of local government (including counties, municipalities and townships) comes from the 1987 Census of Governments, Government Organization File by the U.S. Bureau of Census; Municipal Form of Government survey conducted by International City/County Management Association (IMCA) in 2001; and Pro…le of Local Government Service Delivery Choices survey conducted by IMCA in 1997 and 2002. The data on the mode of provision of public services by local governments comes from the 1987 and 2002 Census of Governments, Government Organization Files by the U.S. Bureau of Census and Pro…le of Local Government Service Delivery Choices survey conducted by IMCA in 1997 and 2002. The data on public employment and budgetary data come from the Employment Statistics and Finance Statistics parts of the Census of Governments for the years 1987, 1997 and 2002. The data on economic and social characteristics of municipalities and counties comes from the Census of Population and Housing for the years 1980, 1990 and 2000. I exclude from the analysis communities with zero median household income. Two communities (Holmedel, NJ and Spokane, WA) were excluded from the dataset due to mistakes in the reported form of government. The …nal dataset is constructed by merging the databases using unique FIPS places’and MCD’s Census codes. A small number of cases were merged manually due to typos in the original datasets.

Variables construction The surveys distinguish …ve forms of municipal government (mayor-council, councilmanager, commission, town meeting and representative town meeting) and three forms of county government (commission, council-administrator and council-elected executive). Counties with council-elected executive and municipalities with mayorcouncil forms of government are considered as having elected executive. Counties with council-administrator and municipalities with council-manager forms of government are considered as having an appointed executive. Other forms of government are excluded from the analysis. With respect to the form of government in the Municipal Form of Government survey, I refer to the variable imis_FOG in the database rather than the response to Question 1 in the survey based on the suggestion by IMCA sta¤. I check robustness of the results using the response to Question 1 in the survey instead and …nd no di¤erence in the results. The surveys that provide information on the method of provision of public services ask whether a particular public service is provided by the local government 28

and what the mode of provision of the public service is. To describe the method of provision of public services, I construct a dummy variable which is equal to one if the public service is contracted out. For public services that are not provided the variable is missing. For observations from the Pro…le of Local Government Service Delivery Choices survey it equals one if the service is said to be provided only by "Private for pro…t”, and zero otherwise. Since data on the method of provision of public services come from di¤erent surveys, I check for the comparability of these measures. For the year 2002 there is some intersection between the Census of Governments and Pro…le of Local Government Service Delivery Choices surveys which allows for such examination. The intersection for the variable that describes the method of provision of public services contains 635 observations for 134 local governments. The correlation between the measures from the two datasets is 0.21 (signi…cant at 0.01 level). In the …nal dataset, which contains observation only for the local governments for which the variable describing the form of the government is not missing the intersections contains 470 observations for 92 local governments and the correlation is 0.18 (signi…cant at 0.01 level). The fact that I use dummy variables to describe the method of provision of public services can explain relatively low levels of correlation. Thus, the alternative measures of the methods of provision of public services provided by independent surveys are robustly correlated and may be used jointly in our empirical investigation. In separating the employment by functions education services include all elementary and secondary education, as well as libraries; health and social services include health, hospitals and welfare; environmental and housing includes housing and community development, natural resources, parks and recreation, sewerage, and solid waste management; government administration includes …nancial administration, judicial and legal, and other government administration; transportation and transit includes airports, streets and highways, water transport and terminals, and transit; utilities include water supply, electric power and gas supply.

Coverage The data for the year 1987 is based on information from Census of Governments, Government Organization File which contains information on all counties, municipal and township governments (38932 observations). Some of observations do not contain all necessary information, but overall the sample can be considered as representative. The data for the year 1997 comes from the Pro…le of Local Government Service Delivery Choices survey. According to ICMA “The [1997] Pro…le in Local Government Service Delivery Choice surveys were mailed in Fall 1997 and Spring 1998 29

to the Chief Administrative O¢ cers in municipalities with populations 10,000 and over and to the Chief Administrative O¢ cers of counties with populations 25,000. In addition, a random sample of one in eight municipalities from 2,500 to 9,999 in population and one in eight counties with populations 2,500 to 24,999 and from those local governments under 2,500 that are recognized by ICMA. Of the 4,952 municipalities and counties that received surveys, 1,586 responded (32.0%).”Thus, the sample is biased towards large local governments and might not be representative for the municipalities with population under 2,500. The data for the year 2002 comes both from the Pro…le of Local Government Service Delivery Choices survey and Census of Governments with information from the Census of Governments being limited by the availability of information on the form of local government, which comes from the Municipal Form of Government survey. According to ICMA “The [2002] Pro…le in Local Government Service Delivery Choices survey was conducted in Fall 2002 and Spring 2003. Letters were mailed to the Chief Administrative O¢ cers in municipalities with populations 10,000 and over, in counties with populations 25,000 and over, and to a random sample of one in eight municipalities from 2,500 to 9,999 in population and one in eight counties with populations from 2,500 to 24,999. (...) Of the 5,370 municipalities and counties that received surveys, 1,283 responded (23.9%).”“The 2001 Municipal Form of Government surveys were mailed in Summer 2001 and Winter 2002 to the Municipal Clerks in municipalities with populations 2,500 and over and to those municipalities under 2,500 in population that are in ICMA’s database. Of the 7,867 municipalities that received surveys, 4,244 responded (54.0%).”Thus, for the year 2002 the sample is also biased towards large local governments and might not be representative for the municipalities with population under 2,500. There is also some evidence that the sample is biased towards richer communities. Trebbi, Aghion and Alesina (2007) analyze the di¤erence between respondent and non-respondent municipalities in the 2001 Municipal Form of Government survey. They show that respondent and nonrespondent municipalities are not signi…cantly di¤erent in terms of ethnic division and total population, but median income in the non-respondent municipalities is somewhat lower with di¤erence being statistically signi…cant. Overall, the data for 1987 can be considered representative, whereas for 1997 and 2002 the sample is biased towards larger communities and may be considered as representative of the population of relatively large communities (with population above 2500 inhabitants) and less representative of the population of smaller communities. There is also some evidence that the sample is somewhat biased towards richer communities.

30

References 1. Abadie, Alberto and Guido W. Imbens (2006) “Large Sample Properties of Matching Estimators for Average Treatment E¤ects,” Econometrica, 74(1): 235-267. 2. Abadie, Alberto, David Drukker, Jane L. Herr, and Guido W. Imbens (2004) "Implementing matching estimators for average treatment e¤ects in Stata." Stata Journal 4(3): 290-311. 3. Alesina, Alberto, Philippe Aghion and Francesco Trebbi (2007). "Electoral Rules and Minority Representation in U.S. Cities", Quarterly Journal of Economics, forthcoming. 4. Alesina, Alberto, Reza Baqir, and William Easterly (2000) "Redistributive Public Employment." Journal of Urban Economics 48(2): 219-41. 5. Alesina, Alberto and Guido Tabellini (1990) "A Positive Theory of Fiscal De…cits and Government Debt." Review of Economic Studies 57(3): 403-14. 6. Alesina, Alberto and Guido Tabellini (2007) "Bureaucrats or Politicians? Part I: A Single Policy Task." American Economic Review 97(1): 169-79 7. Alesina, Alberto and Guido Tabellini (2007) "Bureaucrats or Politicians? Part II: Multiple policy tasks." Journal of Public Economics, forthcoming. 8. Ban…eld, Edward C. and James Q. Wilson (1963). City politics. Cambridge„ Harvard University Press: 362 p. 9. Baron, David P. (1988) "Regulation and Legislative Choice." RAND Journal of Economics 19(3): 467-77. 10. Besley, Timothy and Stephen Coate (2003) "Elected versus Appointed Regulators: Theory and Evidence." Journal of the European Economic Association 1(5): 1176-1206. 11. Besley, Timothy and Abigail A. Payne (2005). Implementation of Anti-Discrimination Policy: Does Judicial Selection Matter?, C.E.P.R. Discussion Papers CEPR Discussion Papers: 5211. 12. Bositis, David A. (1998-2002) "Black Elected o¢ cials: A Statistical Summary." The Joint Center for Political and Economic Studies. 13. Boycko, Maxim, Andrei Shleifer, and Robert W. Vishny (1996) "A Theory of Privatisation." Economic Journal 106(435): 309-19.

31

14. Bromage, Arthur W. (1940) Manager plan abandonments. New York„ National Municipal League. 15. Carpenter, Daniel P. (2001) The Forging of Bureaucratic Autonomy: Reputations, Networks, and Policy Innovation in Executive Agencies1862-1928. Princeton, NJ: Princeton University Press. 16. Chubb, Judith (1982) Patronage, power, and poverty in southern Italy : a tale of two cities. Cambridge [Cambridgeshire] ; New York, Cambridge University Press. 17. Colburn, David R. and Je¤rey S. Adler (2001) African-American mayors : race, politics, and the American city. Urbana, IL, University of Illinois Press. 18. Corey, Elizabeth C. and James C. Garand (2002) "Are Government Employees More Likely to Vote?: An Analysis of Turnout in the 1996 U.S. National Election." Public Choice 111(3-4): 259-83. 19. Diermeier, Daniel, Michael Keane and Antonio Merlo (2005) "A Political Economy Model of Congressional Careers." American Economic Review 95(1): 34773. 20. Dixit, Avinash, Gene Grossman and Elhanan Helpman (1997) “Common Agency and Coordination: General Theory and Application to Tax Policy”, Journal of Political Economy 105(4): 752-769. 21. Donahue, John D. (1989) The privatization decision: public ends, private means. New York, Basic Books. 22. East, John P. (1965) Council-manager government: the political thought of its founder, Richard S. Childs. Chapel Hill„University of North Carolina Press. 23. Fields, Joseph A., Linda S. Klein, and James M. S…ridis (1997) "A Market Based Evaluation of the Election versus Appointment of Regulatory Commissioners." Public Choice 92(3-4): 337-51. 24. Glaeser, Edward L. and Andrei Shleifer (2005). "The Curley E¤ect: The Economics of Shaping the Electorate." Journal of Law, Economics, and Organizations, 21: 1-19. 25. Greene, Kenneth V. and Oleg Nikolaev (1999) "Voter Participation and the Redistributive State." Public Choice 98(1-2): 213-26. 26. Hanssen, Andrew (1999) "The E¤ect of Judicial Institutions on Uncertainty and the Rate of Litigation: The Election versus Appointment of State Judges," Journal of Legal Studies 28(1): 205-32 32

27. Kemp, Roger L. (1991) Privatization : the provision of public services by the private sector. Je¤erson, N.C., McFarland & Co. 28. La¤ont, Jean-Jacques. (1996) "Industrial Policy and Politics." International Journal of Industrial Organization 14(1): 1-27. 29. Levin, Jonathan and Steven Tadelis (2005) "Contracting for Government Services: Theory and Evidence from U.S. Cities." Working Paper. 30. Lineberry, Robert and Edmund Fowler (1967) "Reformism and Public Policy in American Cities." American Political Science Review 61: 701-716. 31. Lopez-de-Silanes, Florencio, Andrei Shleifer and Robert W. Vishny (1997) "Privatization in the United States." RAND Journal of Economics 28(3): 44771. 32. Maskin, Eric and Jean Tirole (2004) "The Politician and the Judge: Accountability in Government." American Economic Review 94(4): 1034-54. 33. Persson, Torsten and Lars E. O. Svensson (1989) "Why a Stubborn Conservative Would Run a De…cit: Policy with Time-Inconsistent Preferences." Quarterly Journal of Economics 104(2): 325-45. 34. Persson, Torsten and Guido Tabellini (2000) Political economics: explaining economic policy. Cambridge, Mass., MIT Press. 35. Rauch, James E. (1995) "Bureaucracy, Infrastructure, and Economic Growth: Evidence from U.S. Cities during the Progressive Era." American Economic Review 85(4): 968-79. 36. Riordon, William L. (1994) Plunkitt of Tammany Hall: a series of very plain talks on very practical politics. Boston, Bedford Books of St. Martin’s Press. 37. Robinson, James A., Ragnar Torvik, and Thierry Verdier (2006) "Political Foundations of the Resource Curse." Journal of Development Economics 79(2): 447-68. 38. Robinson, James A. and Thierry Verdier (2002). The Political Economy of Clientelism, C.E.P.R. Discussion Papers CEPR Discussion Papers: 3205. 39. Savas, Emanuel S. (1987) Privatization: the key to better government. Chatham, N.J., Chatham House Publishers. 40. Vlaicu, Razvan (2007) "Executive Performance under Direct and Hierarchical Accountability Structures: Theory and Evidence," working paper.

33

41. Weingrod, Alex (1968) "Patrons, Patronage and Political Parties," Comparative Studies in Society and History 19: 377-400. 42. Wilson, James Q. (1961) "The Economy of Patronage." Journal of Political Economy 69: 369-380. 43. Wilson, James Q. (1989) Bureaucracy. New York: Basic Books.

34

Table 1. Changes in the form of local government across time. Total

More than one observation per community:

Number

Percent

1575

44.72

Switched from Elected to Appointed

179

5.03

Switched from Appointed to Elected

66

1.88

Stayed Elected

368

10.47

Stayed Appointed

962

27.37

Appointed

828

55.16 23.55

Elected

1111

One observation per community:

Total

1939

3514

Only communities with population above 10000 are included.

35

31.61 100

Table 2. Effect of the Form of Government on the Number of Full-Time Employees. ln(Full-Time Employees) First Differences Matching

Fixed Effects Regression

Became Elected After Being Appointed

0.081 [2.84]***

0.078 [2.65]***

Elected Before Becoming Appointed ln(Population)

-0.047

-0.065

Elected

Matching

Regression

0.199 [10.42]***

0.039 [2.40]**

[1.35] [2.13]** 0.445 0.449 [7.52]*** [13.21]*** ln(Per Capita Income) 0.543 0.054 [9.52]*** [1.80]* ln(Expenditures) -0.025 0.181 [0.50] [3.65]*** Urban 0.135 -0.096 [2.06]** [0.90] Ethnic Fractionalization -0.037 -0.006 [2.15]** [0.27] Inequality 0.322 -0.120 [4.23]*** [1.14] Unemployment 1.730 0.179 [4.20]*** [0.51] Over 65 0.904 0.668 [5.49]*** [2.22]** High School Graduates -0.108 0.110 [1.10] [1.13] Year dummies No Yes State-Year dummies Yes No No Yes Community dummies Yes No Community controls Observations 5361 5361 1250 5686 Number of communities 3247 3247 1250 3477 Absolute value of t-statistics in brackets. Community controls include years since incorporation and dummy variables for county and homerule. Covariates in (1) are the same as in (2). In (1) and (3) the average treatment effect using bias-corrected nearest neighbor matching with one match per observation is reported. In (1) and (3) heteroskedasticity-consistent t-statistics using one match in the second matching stage. Errors in (2) and (4) are clustered at the community level. Covariates in (3) are the same as in (4) except for state-year dummies, which are accounted for by subtracting state averages prior to matching. Matching on year and type of community in (3) is exact. Only communities with population above 10000 are included. * significant at 10%; ** significant at 5%; *** significant at 1%.

36

Table 3. Effect of the Form of Government on the Number of Part-Time Employees. ln(Part-Time Employees) First Differences Matching

Fixed Effects Regression

Became Elected After Being Appointed

-0.147 [0.90]

-0.051 [0.43]

Elected Before Becoming Appointed ln(Population)

-0.067

-0.004

Elected

Matching

Regression

0.130 [3.67]***

0.004 [0.11]

[0.89]

[0.05] 0.285 0.415 [5.03]*** [4.19]*** ln(Per Capita Income) 0.477 0.217 [9.98]*** [2.83]*** ln(Expenditures) -0.067 0.387 [0.67] [2.21]** Urban -0.122 -0.083 [0.88] [0.27] Ethnic Fractionalization -0.037 -0.128 [0.73] [2.03]** Inequality -0.090 0.202 [0.48] [0.54] Unemployment -2.615 0.231 [3.07]*** [0.24] Over 65 -0.660 -0.651 [1.66]* [0.85] High School Graduates 0.047 0.073 [0.22] [0.26] Year dummies No Yes State-Year dummies Yes No No Yes Community dummies Yes No Community controls Observations 5361 5361 1250 5686 Number of communities 3247 3247 1250 3477 Absolute value of t-statistics in brackets. Community controls include years since incorporation and dummy variables for county and homerule. Covariates in (1) are the same as in (2). In (1) and (3) the average treatment effect using bias-corrected nearest neighbor matching with one match per observation is reported. In (1) and (3) heteroskedasticity-consistent t-statistics using one match in the second matching stage. Errors in (2) and (4) are clustered at the community level. Covariates in (1) are the same as in) except for state-year dummies, which are accounted for by subtracting state averages prior to matching. Matching on year and type of community in (3) is exact. Only communities with population above 10000 are included. * significant at 10%; ** significant at 5%; *** significant at 1%.

37

Health & Social Services

Firefighters

Police

Environmental & Housing

Government Administration

Transportation & Transit

Utilities

ln (Full-Time Employees)

Education

Table 4. Effect of the Form of Government on the Number of Employees by Function.

0.176 [3.45]***

0.189 [3.83]***

0.242 [5.33]***

0.182 [6.03]***

0.228 [7.29]***

0.136 [5.03]***

0.118 [3.82]***

0.086 [1.75]*

ln (Part-Time Employees)

0.123 0.070 -0.020 0.068 0.216 0.124 0.043 0.017 [2.42]** [1.67]* [0.56] [1.63] [4.44]*** [3.25]*** [1.47] [0.85] Each cell reports the results of a separate estimation. The average treatment effect estimated using bias-corrected nearest neighbor matching with one match per observation is reported. Heteroskedasticity-consistent t-statistics using one match in the second matching stage in brackets. Covariates include measures of population, income per capita, total budgetary expenditures, income inequality (measured by the ratio of mean and median household income), level of unemployment, ethnic fractionalization, fraction of population 65 years and older, fraction of population above 25 years old with at least 12 years of education, fraction of urban population, years since incorporation and dummy variables for homerule. Matching on year and type of community is exact. The number of observations in all the estimations is 5357. Only communities with population above 10000 are included.* significant at 10%; ** significant at 5%; *** significant at 1%.

32

Table 5. Effect of Privatization on Public Employment. Number of Full-Time Employees Same Service Other Total Services Privatized*Elected

-0.855 [14.17]***

0.045 [7.55]***

0.002 [0.64]

Privatized*Appointed

-1.057 [22.77]***

0.038 [7.48]***

-0.003 [1.82]*

0.451 0.440 0.418 [7.05]*** [16.49]*** [14.15]*** ln(Per Capita Income) 0.620 0.243 0.302 [4.71]*** [4.88]*** [5.39]*** ln(Public Expenditures) 0.075 0.071 0.095 [1.95]* [2.90]*** [2.83]*** Ethnic FractioNoalization -0.004 -0.032 0.010 [0.02] [0.56] [0.16] Urban (%) 0.073 0.039 0.046 [1.91]* [2.58]*** [2.58]*** Inequality -0.339 -0.181 -0.268 [1.45] [1.82]* [2.87]*** Unemployment 1.310 0.369 0.166 [1.92]* [0.74] [0.42] Population over 65 0.928 0.763 0.961 [1.74]* [3.33]*** [4.30]*** High School Graduates 0.252 -0.015 -0.136 [1.26] [0.22] [1.99]** Year dummies Yes Yes Yes Community dummies Yes Yes Yes Public service dummies Yes Yes Yes Observations 19313 19313 71362 Number of communities 4277 4277 4334 Only communities with population above 10000 and constant form of government are included. The unit of observation is service-community-year. Absolute value of robust t-statistics in brackets. Errors are clustered on community-year level. * significant at 10%; ** significant at 5%; *** significant at 1%. ln(Population)

32

Table 6. Mean Covariate Differences in Matched Groups. Difference in means/Stand. dev. Original Sample Matched Pairs ln(Population) 0.128 0.018 ln(Per Capita Income) 0.490 0.033 ln(Expenditures) 0.263 0.027 Urban 0.290 -0.006 Ethnic Fractionalization 0.439 0.076 Inequality 0.249 0.037 Unemployment -0.224 -0.035 Population over 65 -0.012 -0.024 High School Graduates 0.239 0.037 Years since Incorporation -0.064 -0.015 Homerule 0.205 -0.001 The numbers represent the difference in averages between the communities with appointed and elected chief executives divided by standard deviation for the whole sample. The averages and standard deviation in the first column are calculated using the original sample. The averages and standard deviation in the second column are calculated using the matched pairs, so that each observation is weighted by the number of times it is used in matching estimation.

33

Table 7. Robustness of the Matching Estimates. Full-time employees Number of Matches

All Years

1987

1997

2002

0.199

0.137

0.231

0.280

[10.42]***

[5.94]***

[4.73]***

[7.21]***

0.210

0.161

0.237

0.285

[11.23]***

[7.02]***

[4.96]***

[6.37]***

0.212

0.169

0.236

0.282

[10.74]***

[7.10]***

[5.07]***

[6.19]***

0.215

0.173

0.240

0.288

[10.79]***

[7.19]***

[5.03]***

[6.24]***

0.130

0.120

-0.030

0.228

[3.67]***

[2.76]***

[0.33]

[3.31]***

0.146

0.159

0.037

0.210

[4.24]***

[3.74]***

[0.43]

[2.84]***

0.153

0.171

0.043

0.204

[4.47]***

[3.95]***

[0.52]

[2.75]***

0.159

0.181

0.055

0.197

Ln(Full-time employees) 1 3 5 7

Ln(Part-time employees) 1 3 5 7

[4.63]*** [4.23]*** [0.68] [2.64]*** Absolute value of robust t-statistics in brackets. The average treatment effect is reported. Covariates include measures of population, income per capita, total budgetary expenditures, income inequality (measured by the ratio of mean and median household income), level of unemployment, ethnic fractionalization, fraction of population 65 years and older, fraction of population above 25 years old with at least 12 years of education, fraction of urban population, years since incorporation and dummy variables for homerule. Matching on year and type of community is exact. Only communities with population above 10000 are included.* significant at 10%; ** significant at 5%; *** significant at 1%.

34

Table 8. Determinants of changes in form of government Appointed becomes elected

Elected becomes appointed

Budget deficit (share of expenditures)

5.049 [2.35]**

1.917 [2.31]**

ln (Full Time Employmees)

0.526 [1.05]

-0.467 [1.78]*

ln (Part Time Employmees)

0.272 [1.66]*

-0.146 [1.17]

ln (Population)

1.322 [3.74]***

-0.366 [1.43]

ln (Income Per Capita)

-1.839 [2.78]***

-2.757 [5.55]***

ln(Public Expenditures)

-0.963 [1.98]**

0.499 [2.11]**

Ething Fractionalization

-1.902 [1.78]*

3.173 [3.27]***

Urban

-2.421 [4.57]***

-0.498 [1.66]*

Inequality

-4.460 [1.86]*

-0.168 [0.11]

Unemployment

-4.676 [0.71]

-19.087 [2.56]**

Population over 65

-0.063 [0.02]

4.106 [1.31]

High School Graduates

-0.334 [0.22]

7.003 [4.79]***

Observations 1646 655 Only communities with population above 10000 are included. Logit regression with errors clustered by community. Absolute value of robust t-statistics in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%

35

Table AI. Description of the variables. Variable

Description

Elected

Dummy variable that equals one for counties with council-elected executive and municipalities with mayor-council form of government. It equals zero for counties with council-administrator and municipalities with council-manager forms of government. Missing for all other forms of government. With respect to the form of government in the Municipal Form of Government survey, I refer to the variable imis_FOG in the database rather than the response to Question 1 in the survey based on the suggestion by IMCA staff. Source: Census of Governments (1987), Government Organization File; Municipal Form of Government (2001) and Profile of Local Government Service Delivery Choices (1997, 2002) surveys by International City/County Management Association (IMCA).

Appointed

Dummy variable that equals one minus Elected.

Became Elected after Being Appointed

Dummy variable that equals one if the variable Elected equals one in the same period and zero in the previous period. Equals zero in all other cases when Elected is not missing.

Became Appointed after Being Elected

Dummy variable that equals one if the variable Elected equals zero in the same period and one in the previous period. Equals zero in all other cases when Elected is not missing.

Full-Time Employees

The number of full-time public employees. Source: Census of Governments, Employment Statistics (1987, 1997 and 2002).

Part-Time Employees

The number of part-time public employees. Source: Census of Governments, Employment Statistics (1987, 1997 and 2002).

Privatized

Dummy variable that equals one if the provision of public service is contracted out and zero if it is provided in-house or by other government. For public services that are not provided the variable is missing. For observations from the Profile of Local Government Service Delivery Choices survey it equals one if the service is said to be provided by "Private for profit" only and zero otherwise. Source: Census of Governments (1987), Government Organization File; Profile of Local Government Service Delivery Choices (1997, 2002) survey by International City/County Management Association (IMCA).

Share of privatized public services

Share of public services for which the variable “Privatized” (see above) equals one.

Population

Total population. Source: Census of Population and Housing (1980, 1990 and 2000).

Per Capita Income

Per capita income in dollars. Source: Census of Population and Housing (1980, 1990 and 2000).

Expenditures

Total budgetary expenditures of the local government. Source: Census of Governments, Finance Statistics (1987, 1997 and 2002).

Inequality

Ratio of mean and median household income. Source: Census of Population and Housing (1980, 1990 and 2000).

Unemployment

Civil labor force unemployment rate. Source: Census of Population and Housing (1980, 1990 and 2000).

36

Variable

Description

Ethnic Fractionalization

Probability that two people randomly drawn from a city will belong to different ethnic groups. I distinguish six ethnic groups: white, black, Hispanic, American Indians and Alaskan, Asian-Pacific, and others. Source: Census of Population and Housing for (1980, 1990 and 2000).

Population over 65

Fraction of population 65 years or older. Source: Census of Population and Housing (1980, 1990 and 2000).

High School Graduates

Fraction of population above 25 years old with at least 12 years of education. Source: Census of Population and Housing (1980, 1990 and 2000).

Urban

Fraction of urban population. Source: Census of Population and Housing (1980, 1990 and 2000).

Budget Deficit

Difference between government total expenditures and total revenues as a share of total expenditures. Source: Census of Governments, Finance Statistics (1987, 1997 and 2002).

Years Since Incorporation

Number of years since the incorporation of the local government. Source: Census of Governments (1987), Government Organization File

Homerule

Dummy variable that equals one for homerule local governments. Source: Census of Governments (1987), Government Organization File

County

Dummy variable that equals one for counties and zero for municipalities and townships.

Black Mayor

Dummy variable that equals one if mayor is black and zero otherwise. Source: Afro-American Mayors (2001) and Bositis, D. (various years). Available at www.jointcenter.org/publications1/BEO.php

Black Population

The number of people identified as Black or African American alone. Source: Census of Population and Housing (1980, 1990 and 2000).

Share of Blacks among Public Employees

The number of people identified as local employees and Black or African American alone. Source: Census of Population and Housing (1980, 1990 and 2000) 5% Public Use Microdata Sample.

Number of Local Employees

The number of local employees estimated using total population and the share of population in the sample identified as local employees. In the regressions with the share of blacks among public employees this variable is used instead of information from the Census of Governments because it minimizes the number of missing observations. Source: Census of Population and Housing (1980, 1990 and 2000) 5% Public Use Microdata Sample.

37

Table AII. Subsequent occupation of former mayors and city managers. Occupation Public office:

City Managers

Mayors

Number

Percent

Number

Percent

112

91.1

22

31.9

City manager

98

79.7

0

0.00

Other city office

2

1.6

8

11.6

County office

6

4.9

4

5.8

State office

1

0.8

5

7.2

Federal office

0

0.0

3

4.3

Other public office

5

4.1

2

2.9

Self-employed

0

0.0

9

13.0

Other private

7

5.7

21

30.4

Non-profit

0

0.0

5

7.2

Jail

0

0.0

2

2.9

Retired

4

3.2

8

11.6

Died

0

0.00

2

2.9

Total

123

100

69

100

The data for city managers and mayors includes information only for the municipalities with council-manager and mayor-council forms of government respectively.

Table AIII. Summary statistics. Appointed Chief Executive N

Mean

Std.Dev.

ln (Number of Full-time Employees) 3592 5.69 1.22 ln (Number of Part-time Employees) 3592 4.03 1.39 ln(Population) 3616 10.44 0.96 ln(Per Capita Income) 3616 9.41 0.57 ln(Expenditures) 3529 10.44 1.25 Urban 3616 0.79 0.37 Ethnic Fractionalization 3616 0.29 0.18 Inequality 3616 1.23 0.12 Unemployment 3616 0.06 0.03 Population over 65 3616 0.12 0.06 High School Graduates 3616 0.71 0.14 Years since Incorporation 3435 117.20 64.35 Homerule 3535 0.41 0.49 Privatized 2994 0.14 0.20 Note: Only communities with population above 10000 are included.

38

Elected Chief Executive N 2191 2191 2197 2197 2157 2197 2197 2197 2197 2197 2197 2018 2184 1912

Mean 5.52 3.86 10.32 9.14 10.08 0.67 0.21 1.20 0.06 0.12 0.68 121.20 0.31 0.14

Std.Dev. 1.40 1.51 1.03 0.50 1.54 0.44 0.18 0.11 0.03 0.04 0.13 59.13 0.46 0.23