Electoral Competition and Government Performance in Mexico

Matthew R. Cleary University of Chicago This is a draft. Please do not cite. Comments welcome at [email protected]

Paper prepared for the XXIII International Congress of the Latin American Studies Association, Washington D.C., September 6-8, 2001. I would like to thank Yemile Mizrahi, Alain de Remes, and Guillermo Trejo for sharing their data, and the Centro de Investigación y Docencia Economicas for providing institutional support, and great lunches, during my time in Mexico City.

ABSTRACT This paper uses recent Mexican political history to evaluate the hypothesis that competitive multi-party elections induce government to perform better than it would in the absence of such elections. I use data on public service provision and municipal finances to construct several measures of government performance. Electoral competition is estimated from a collection of all municipal electoral returns from 1980 to 2000. The preliminary analysis shows that electoral competition does have an independent effect on government performance, but the estimated effect is rather small. I conclude that elections alone are less influential on government performance than is commonly thought, and are not sufficient to induce more responsive or efficient government. At least in Mexico, improved government performance will also require institutional reform and further political changes.

INTRODUCTION The broad aim of my dissertation is to expand our understanding of the consequences of electoral competition in transitional democracies. Previous chapters have outlined a long list of theoretical assertions and empirical findings that impute a variety of beneficial effects to elections. As elections become more competitive and meaningful, for example, scholars argue that governments are less likely to abuse human rights (Diamond 1999, Beer and Mitchell 2001), more likely to implement responsive public policies (Manin et al. 1999), and less likely to resort to clientelism (Kitschelt et al. 1998). This is just a sample of the desirable outcomes that have been linked to the rise of democratic governance in general, or to electoral competition in particular.1 My dissertation addresses a subset of such claims; the current chapter focuses on only one. Specifically, the goal of this chapter is to determine whether the rise of electoral competition in Mexico has improved government performance. To evaluate this hypothesis, I examine the effects of increased electoral competition in Mexican municipal elections over the past 20 years. The Mexican case is ideal for the evaluation of electoral effects for many reasons. First, these elections offer a wide variation on the degree of electoral competition across space and time. In some municipalities, the long-ruling Partido Revolucionario Institutcional (PRI) has won (literally) 100% of the vote in several consecutive elections. In others, three or even four parties have had a realistic chance of winning. Second, these same municipalities show wide variation on indicators of government performance (the indicators will be discussed in detail below). Third, the quality of data for all of the relevant variables is relatively good, and the large number of municipalities in Mexico (over 2400) provides a large pool of cases from which to draw inferences.2 These advantages, along with an intra-national comparative research design, provide the researcher a number of techniques for evaluating the hypotheses of interest.3 The next section of this paper offers a theoretical account of how electoral competition might induce good government performance. I also address several limitations to the theory, though empirical investigation of these complicating factors is undertaken in later chapters. The third section reviews alternative explanations of variation in the quality of government. The fourth section discusses measurement issues in detail, and the fifth presents some preliminary statistics and a test of the main hypotheses. The sixth section concludes.

1

From this point on, I will not refer to regimes as more or less democratic. I find that this term is too broad to be useful conceptually. Therefore, I will refer only to the specific aspect of democracy whose variation is of theoretical interest. In most cases, this means that I will distinguish between regimes that have more or less competitive elections. 2 For example, I estimate that there have been roughly 14,941 municipal elections between 1980 and 2000. Thanks to heroic efforts of data collection by Alain de Remes and other researchers, the actual vote counts are publicly available for 14,702 (or 98%) of them. 3 In the interests of space, I will not discuss in great detail the relative merits of an intra-national research design, or the applicability of various theories of democracy and democratization to subnational units. However, I do discuss these issues elsewhere in the dissertation, and I would be happy to provide the relevant chapters to the reader upon request.

1

ELECTORAL COMPETITION AND GOVERNMENT PERFORMANCE4 Theory. Within the framework of accountability theory, there are only two reasons that elected representatives would faithfully serve the interests of their constituents to the best of their ability. The first is that a representative might have a normative reason for doing so. She could view public service as a vocation, honestly believing that she was chosen to do the people’s work. Or, she might just happen to hold views and beliefs that are in accord with the majority of her constituents, so that she works toward the public interest simply by pursuing policies that she personally finds preferable. The second reason a representative might serve the public interest is instrumental. This is the motive most commonly imputed to politicians, for example when they are assumed (in both formal models and in the vernacular) to value reelection above all else. In this instrumental sense, representatives will perform well in order to maintain their hold on public office, or as a means of attaining higher office. The public’s goal is to fill government offices with politicians who are driven by one or both of these motives.5 Given a sufficient level of competence, a government whose office holders have motives to serve the public interest will perform better than one filled with politicians who lack such motives. The challenge to citizens, then, is to devise some method of selecting representatives that will most efficiently empower properly motivated politicians. Whether the goal is to select normatively committed representatives or to induce good behavior from instrumentally motivated politicians, one method dominates both political practice and academic interest: competitive, multi-party elections. But how exactly do elections function as mechanisms of selection? Some scholars view elections as opportunities to select normatively committed representatives, or “good types” (Campbell et al. 1960; Fearon 1999). Since aspirants to political office operate campaigns prior to elections, voters can observe the politicians’ characteristics, including their personality, policy preferences, and their intellect. Voters can then use elections as an opportunity to select politicians who are most likely (and able) to work toward the public interest. If voters have reliable information on the relevant characteristics, then this mechanism should work, and we would observe that polities in which representatives are elected pursuant to competitive campaigns have better-performing representatives than those that do not. Alternatively, elections may help voters control politicians, so that the latter act in the former’s interest, whether or not they are otherwise disposed to do so (Key 1966; Fiorina 1981). Since elections recur periodically, representatives who are concerned about their future careers will anticipate a day in which their constituents have an opportunity to remove them from office. Hoping to avoid removal, politicians will work toward the public interest, if they believe that doing so is the best way to demonstrate to the voters that they should be returned to office in the future. A representative may be 4

The issues raised in this section and the next are given fuller treatment in a previous chapter, available from the author on request. 5 It might be argued that, given the choice, the first type of motivation is better than the second. But in practice it is hard to distinguish between politicians who really want to “do the right thing” and those who want to get elected; The Federalist, among others, argues that the latter motivation is the best that can be hoped for. It is also probable that the actions of most politicians are influenced by a mix of both motives.

2

tempted to shirk, steal, or otherwise avoid doing the work for which the public empowered them; but if he believes that these activities would lead to his removal from office, he will refrain. These two approaches to electoral accountability are well known, and are usually described as prospective and retrospective control. Of course, it is also well known that elections do not always serve the purposes imputed to them by these two approaches. If elections always and everywhere allowed voters to select good representatives, then the newspapers would not be so full of stories about corrupt, incompetent, or “out-of-touch” politicians. And if elections served as a mechanism of retrospective punishment, good performers would eventually replace poor performers, and would then remain in office for long periods of time. Sometimes elections seem to work in these ways, but very often they do not. When elections fail as mechanisms of control, or to use Powell’s (2000) phrase, when elections fail “as instruments of democracy”, what goes wrong? I have implicitly identified many of the potential problems with electoral control in the discussion above. For example, the prospective account of selecting honest, competent representatives requires that voters have reliable information about the qualities of candidates. Very often this requirement is not met in practice. Politicians can lie during campaigns, and even if some are honest, it may be difficult for the public to know which are which. Also, some characteristics, such as integrity and competence, may be exceedingly difficult to observe in practice. So it may be the case (or, rather, it most certainly is the case in many elections) that voters go to the polls without having a reliable sense of which politician would best serve their interests once in office. In such a situation, voters will not be able to consistently select good representatives, and elections will not function as mechanisms of control. All of the other assumptions involved in the story of how elections induce good government also come into play here. If voters lack information; if they reward and punish representatives based on criteria other than government performance; if politicians are not concerned with reelection (see below); if politicians are not sufficiently competent or if their offices do not carry the power that they need to provide the outcomes the public desires; if any of these things are true, elections may fail as mechanisms of control. The assumption that representatives want to be reelected deserves special attention for the Mexican case. This assumption is not generally accurate even in the U.S. and Europe, where it is most often applied. Politicians frequently retire or change careers, and therefore hold office in full knowledge that they will never face reelection. Others have strong ideological commitments, such that they would rather be voted out of office than help enact policies with which they disagreed. Such problems are well known among Americanists, though they are often treated as a curiosity, an exception to the rule that does not require explanation. Students of Mexican politics cannot afford to be so cavalier regarding the last term problem. According to the Mexican Constitution, no individual can serve two consecutive terms in any elected post. Every elected official in the country, at all times, is serving his or her last term in any particular office.6 It turns out to be the case that while the behavior of political parties may be conditioned by the

6

This is not strictly true, since politicians can re-assume most offices after sitting out one term. In practice, few politicians do so.

3

need to retain elective offices, individual politicians face no such (direct) pressure.7 Accordingly, the theoretical result (discussed in a previous chapter) is that for elections to function as mechanisms of popular control in Mexico, two things must happen. First, voters must reward or punish parties for the performance of their office-holding members, since they cannot punish representatives directly. Second, parties must be able to either 1) control the nomination process such that they put forward “good types” as candidates, or 2) induce good behavior from their office-holders by promising post-term bureaucratic positions or other incentives (see also Carey 1996). These complicating factors call for the testing of more detailed and specific hypotheses, which are discussed below. Hypotheses. I begin with a basic hypothesis. Whatever issues may complicate the story (and such issues will be addressed shortly), one thing is clear: where once Mexico had singularly noncompetitive elections, it now has free, fair, and hotly contested elections in many areas. Whether this increase in competition is due to electoral reform, socioeconomic changes, or a shift toward a more democratic political culture, the fact remains that all parties face a real threat of being removed from office. While each of the main parties, especially the PRI, still has its strongholds, there are relatively few “safe” seats, in which the incumbent party need not worry about the outcome of future elections. This threat is new. It did not exist 20 or 25 years ago, when the PRI could count on winning virtually every election in the country by wide margins. One must wonder, then, what effects this new threat might have on the behavior of politicians while they are in office. For present purposes, I define government performance as the municipality’s success in providing responsive policies and services. (The justification for this definition and a discussion of measurement strategies are discussed below.) The link between government performance and competitive elections is clear. If citizens complain about the lack of responsiveness and service provision from their local governments (and they do), and if political parties know that voters can remove them from office if they are sufficiently dissatisfied (and they do), then all else equal, we would expect that those municipalities in which elections are competitive will exhibit better government performance than municipalities in which the incumbent party does not feel threatened with the loss of office. This is the central hypothesis of this paper, and it will be addressed in detail with the statistical analyses below. But for completeness, I should also mention some other hypotheses that are implied by the theoretical discussion above. Most importantly, the last term problem in Mexico renders the standard story of retrospective electoral accountability inapplicable. As discussed above, if competitive elections induce improved performance in Mexico, it must be because parties are held accountable by voters, and office-holders are held 7

Of course, it is possible that representatives will have an incentive to serve the public interest if they plan to run for another office upon completion of their current term. So, for example, the mayor of a state capital might feel the same electoral pressure as he would under conditions of reelection, if plans to run for the governorship at the end of his term. But generally, most municipal presidents as individuals do not face such pressures, for the simple reason that there are far more mayors than there are governorships or seats in the federal legislature. In practice, especially under the period of PRI hegemony, politicians floated in between elective positions and bureaucratic assignments.

4

accountable by their parties. Specifically, we would expect to find empirical support for the following propositions: 1. In municipalities that perform relatively well, citizens are more likely to judge parties based on the past performance of their office-holders. 2. In municipalities that perform relatively well, citizens are more likely to have good information on candidates, and to use it when they vote prospectively. 3. As elections are more competitive, political parties are more likely to nominate candidates based on likely performance in office. 4. As elections are more competitive, parties reward office-holders who perform well with bureaucratic or appointed positions after their elected term. The remainder of this chapter is dedicated to evaluating the basic hypothesis of interest. But these additional hypotheses are equally important for understanding the effects of electoral competition in Mexico, and I give them close consideration in subsequent chapters. OTHER CAUSES OF GOVERNMENT PERFORMANCE There have not been many studies that attempt to explain the quality of government output, though an emerging literature exists that attempts to offer a general account of (national) government performance.8 Here, however, I am only concerned to address several factors that are likely to influence municipal performance in the Mexican case. The primary goal is to address factors that need to be included as control variables in subsequent statistical analyses, rather than to provide a general theoretical account of the causes of government performance. Theory and Hypotheses. One obvious factor that should influence municipal performance is the municipality’s ability to perform well. Concretely, this means that all else equal, municipalities with larger budgets (per capita) should be able to provide better services. In Mexico, the bulk of municipal budgets comes from federal and state transfers. It has often been suspected that these transfers are politically motivated, but the evidence on this point is not strong (c.f. Molinar and Weldon 1994). Whether federal and state transfers are politically motivated or not, it remains plausible that performance will partly be explained by the amount of money that municipal governments have at their disposal. On the other hand, the size of the budget can also be affected endogenously: municipal governments that are more determined to perform well may take measures to increase their revenues, which they can do by attempting to raise revenue locally. Accordingly, a conceptual distinction should be made between the size of the municipal budget provided by transfers, which is determined exogenously to performance and which could therefore 8

For example, see Putnam 1993 and La Porta et al., 19xx. Of course, there is a large literature that addresses the effect of politics on national economic performance. But political scientists do not often address the political causes of the quality of government output, from the provision of public services, to the quality of police protection, regulation, and social security.

5

be considered a cause of performance, and the amount of locally raised revenue, which can be viewed as an indirect indicator of performance, as well as a cause. The role of local revenue will be discussed below. But the role of the municipal budget in general suggests a clear hypothesis: as the budget (per capita) increases, government performance should improve. I conceptualize performance here in terms of the provision of public services. A second factor that may affect a municipality’s ability to perform is the identity of the party in power. Some scholars believe that municipalities ruled by the PRI, for example, are more likely to exhibit poor performance, either because of the PRI’s historical lack of concern with competition, or for some other reason. Others would argue that PAN-held municipalities are likely to be more efficiently run, because many PANistas have a background in business. Non-PRI parties may also have to perform well to prove that they can govern, since they have no track record on which to base their claims of competence and experience. It is at least worth testing, then, whether there are party-level differences in the quality of government among municipalities, controlling for other factors. By using dummy variables for the different parties in power, I will test the proposition that each political party has a different “natural” level of government performance. A third factor that has been linked to government performance in Mexico is the party relationship between the municipal and state governments. Alain de Remes (1998, 1999) coined the term juxtaposition to refer to those cases in which the municipal president is of a different party than the state governor.9 It is conceivable that the governor could make life difficult for mayors who are of an opposing party, and indeed this charge has often been leveled against PRI governors. However, this issue should be distinguished conceptually from the issue of federal and state transfers, discussed above. Governors may delay or withhold funds from certain municipalities (though not legally). But they may also act to impede the performance of opposing mayors in non-financial ways, for example by directing negative rhetoric at the opposing mayor, acting uncooperatively with bureaucratic programs or assistance, withdrawing police forces, or any other number of strategies. Therefore, the juxtaposition of state and local governments warrants investigation as an effect on performance, independent of the financial relationship between the state and municipality. The hypothesis is that, all else equal, a juxtaposed municipality will have a lower level of government performance than non-juxtaposed municipalities.10 A final factor that has been linked to government performance, and the most sophisticated theoretical alternative to accountability theory, is the theory of social capital (Putnam 1993). According to Putnam’s account, government works better in areas that have high levels of social capital, because social capital allows citizens to successfully confront collective action problems. The link between social capital and government performance is implicit in Putnam’s conception of the state (or at least of Italian regional governments) as a vehicle or tool through which citizens can act collectively. So, while 9

Governors of a different party than the president can also be described as juxtaposed, but this is not a concern of the current investigation. 10 Since charges of this type of foul play are almost always leveled against a PRI governor, I also test the hypothesis only with respect to juxtaposed municipalities with a PRI governor, by using an interaction term in an alternative specification.

6

Putnam offers no real discussion of politics that is analogous to accountability theory’s focus on the incentives facing individual politicians, he is still able to link social capital to government performance by assuming that government output is a response to collective action or public goods problems. In practice, this could work one of two ways. The first, which I will call the exogenous influence model, envisions the government as an actor outside of the rest of a polity. Social capital can help induce good performance by giving citizens an increased ability to deal with the government, either by successfully communicating their preferences to government actors, or by decreasing the costs of organizing opposition to poorly performing governments, or by some other mechanism. The second way that social capital might influence government performance is through an endogenous influence model. In this conception, the government is not clearly separated from the rest of the polity. Good government can result from the fact that those who govern share the same political culture, associational memberships, and social networks as the rest of the polity. As more and more citizens are connected to networks of high social capital, the probability increases that a representative selected from among them, even at random, will also be connected to those networks. And if social networks provide good information about the characteristics, preferences, and competence of candidates, than we would expect polities with high levels of social capital to be able to select “good types” even more frequently. The distinction is best summarized by the following question: assuming that social capital does in fact correlate with government performance, is it because politicians are better controlled by citizens who are can act collectively, or is it because those politicians themselves are able to draw an the reserve of social capital that inheres in their communities? In this sense, social capital theory is not at all incompatible with the two versions of the accountability model presented above. The exogenous influence model is simply a specific case of the retrospective accountability model discussed above; social capital makes the accountability mechanism more effective by providing information and organizational channels for voters. The endogenous influence model is compatible with prospective accountability, in which voters try to choose “good types” based on all available information. Here, social capital improves voter’s knowledge of the candidates, and may also improve the quality of likely candidates. So while Putnam’s hypothesis about social capital and government performance has distinct empirical implications, it is best viewed as a particular type of accountability theory, not as an opposing theory.11 But in any event, the basic hypothesis indicated by social capital theory is clear: municipalities that have higher levels of social capital will exhibit better government performance.

11

An interesting question for future research is whether social capital could induce good performance in the absence of elections. If social capital theory works according to what I have called the exogenous influence model, then it seems likely that elections are still necessary; social capital only increases the skill with which citizens can use elections. But if the endogenous influence model is at work, then it seems possible, at least in theory, that representatives chosen by non-electoral means (such as a lottery, heredity, nepotism, camarilla networks, or selection from within an oligarchy) would still perform better than their counterparts in polities with low levels of social capital.

7

MEASUREMENT This section offers an extensive discussion of measurement issues for all of the variables that will be included in subsequent statistical analyses. Measurement and analysis will focus on the main hypothesis of interest. The other hypotheses (listed above as 1-4), are tested in subsequent chapters. Government Performance. There are any number of ways to measure government performance, depending on what level of government is being evaluated and what aspect(s) of performance are of theoretical interest in a given project. Here, I offer two strategies that I believe are the most appropriate measures of performance for Mexican municipalities. But I do not claim that these measures are appropriate in all situations, such as cross-national studies of the quality of (federal or central) government. So, for example, my measures differ somewhat from both Putnam’s (1993) and La Porta et al.’s (1998), but they differ for good reasons. Provision of Public Services. The first strategy is to directly estimate the municipality’s performance on a bundle of issues that are of primary concern to the voters. Here, there is fairly good evidence that we can use to determine what criteria are most appropriate. To begin with, the Mexican Constitution explicitly assigns responsibility to municipal governments for eight issue areas (see Table 1).12 With the partial exceptions of public security and roads, which in practice are still largely the responsibility of higher levels of government, these issue areas are the exclusive responsibility of municipal governments.13 Table 1. The Constitutional Responsibilities of Municipal Governments in Mexico a. Potable water, drainage, sewage systems, treatment and disposal of waste water b. Public lighting c. Cleaning, collection, removal, treatment, and disposal of waste material. d. Markets and supply centers e. Monuments f. Slaughterhouses g. Streets, parks, gardens, and their equipment h. Municipal and traffic policing Source: Mexican Constitution, Article 115 (author’s translation) Another clue about which issues are integral parts of municipal performance can be found by taking note of what citizens seem to be most concerned about. Happily, the 12

The constitutional reform that assigned these responsibilities to municipal governments was passed in 1983. 13 The separation of police powers between municipal and state forces is complex. The constitution gives municipalities the responsibility for “municipal and traffic policing”. Local police do have arrest powers, but it is typically the state police who investigate crimes reported after-the-fact and arrest suspects.

8

available evidence suggests that citizens have a fairly good sense of what municipalities should be doing, since their most common complaints about municipal governments closely match the list of responsibilities presented in Table 1. Table 2 shows the responses to the question, “what is the most serious problem in your neighborhood or locality?”14 This question was asked in a pre-election poll in 1998, in the states of Chihuahua and Puebla. In both states, the most common answers are public security and road pavement. However, it is not clear that these complaints directly reflect on municipal governments. Regarding public security, it should be noted that the responsibility for police powers is still dominated by higher levels of government. Most municipal security forces are responsible for traffic and neighborhood patrolling, but do not investigate or process serious offenses. Similarly, the survey responses regarding road pavement are ambiguous. While municipalities are responsible for maintaining local roads, state and federal agencies still have responsibilities for interstate roads, as in the United States. So it is not clear whether these responses really reflect dissatisfaction with municipal performance or not. This is especially problematic in the case of Puebla, since the numbers reported are for “pavement and highways”, the latter of which clearly indicates a complaint against the state or federal government. Beyond these two responses, however, the survey results give a clear indication of which municipal responsibilities are most important to the citizenry. Overwhelmingly, citizens identify the provision of public services as a serious problem in their localities. The only other response that garners a significant percentage is unemployment. These survey responses show that lack of water, electricity, and sewers are all serious issues in municipal politics. Numerous case studies of Mexican municipal governance also attest to the importance of public utility provision in local politics.15 Together, the official responsibilities of municipal governments and the survey responses about the most serious problems in municipalities suggest a good method for evaluating municipal performance. If data can be collected that reflect the municipality’s effort to address public security, road pavement, and the provision of water, electricity, and sewers, then these data could be used to as an estimate of overall government performance. While the available data is not perfect, I believe that it does provide a sufficiently accurate and intuitive measure of municipal performance.

14

“¿Cuál es el principal problema de su colonia, barrio, o localidad?” The Mexican literature on local governance places quite a bit of importance on public service provision. For example, a recent edited volume called Gerencia Pública Municipal: Conceptos Básicos y Estudios de Caso [Municipal Public Management: Basic Concepts and Case Studies] includes chapters with the following titles (author’s translation): “The Implementation of a Program to Manage Garbage in the City of Mérida”; “Decentralization of Drinking Water Service in the City of Lerma, México”; and “The Use and Destination of Waste Water in the City of La Paz” (Cabrero Mendoza and Nava Campos, 1999). For other examples, see Ziccardi 1995 and Cabrero Mendoza 1996. It is hard to find studies of local government administration that do not at least mention public services as a significant issue. 15

9

Table 2. Survey Response to the Question, “What is the most serious problem of your locality or neighborhood?”, 1998.

Response

Chihuahua N %

Puebla N %

Insecurity Road pavementa Lack of water Unemployment Public Services Sewers Public Lighting Other (valid)b None Don’t know, don’t answer

208 94 77 57 41 40 34 175 91 185

20.8 9.4 7.7 5.7 4.1 4.0 3.4 17.5 9.1 18.5

127 177 96 71 51 53 30 219 114 56

12.8 17.8 9.7 7.1 5.1 5.3 3.0 22.0 11.5 5.6

TOTALc

1002

100.2%

994

99.9%

Source: Survey responses taken in 1998 pre-election polls in the states of Chihuahua and Puebla, reported by Yemile Mizrahi (1999, 2000). Data provided by Mizrahi with thanks to the Ford Foundation for funding the surveys. Notes: a) In Puebla, this response is reported as “pavement and highways” b) This category sums all other valid responses. Each response was returned by less than 3% of the sample, with the exception of “lack of schools”, which was the response of 3.9% of the sample in Puebla. c) Percentage totals vary due to rounding.

First, I am aware of no data on municipal road maintenance. This is simply not measurable at the municipal level. Second, data do exist on public security, but they are not of good quality, and their applicability to the present measurement problem seems tenuous. The Mexican government has made available statistics on alleged crimes, arrests, and convictions, at the municipal level, since 1987. However, it is widely believed that these data are rendered inaccurate because most crimes are never reported. This is especially true for minor crimes, and probably for sexual assaults. Still, it could be argued that a general sense of the crime rate could be measured by tracking the most serious crimes, which are most likely to be reported. These crimes, including homicide, assault, kidnapping, and armed robbery, are also the types of street crimes or random crimes that are most likely to be responsible for creating fear and a sense of insecurity among citizens. In principle, then, it is possible to use existing data to measure changes the types of crime that most likely to be reflected in citizen’s judgments of government performance. However, the more serious problem with the measurement of crime rates is that it is not clear that the municipality has effective control over municipal police powers, such that it would be fair to judge municipal performance based on changes in the crime rate. On the one hand, municipal governments are not sufficiently empowered with police who have the ability to investigate crimes and make arrests. On the other hand, it is likely that crime rates are determined by a host of socioeconomic factors exogenous to the level of

10

policing, which factors may also covary with other determinants of government performance, such as levels of education or wealth.16 Third, the data on public service provision is comparatively good. The Mexican census, which is taken every ten years (“on the zeros”), reports the number of households in each municipality that do or do not have access to water, electricity, and sewers. In 1995, the government also collected a conteo (often translated as a “quick count”), which used a mix of data collection and statistical estimations to determine the same information. Coverage on these variables is 100% complete. The only significant drawback for using this data is that the hard data are only collected every ten years. Temporal coverage improves somewhat if we accept the 1995 conteo as sufficiently accurate, but it is still not possible to construct an annual indicator of government performance without making several unrealistic assumptions about the rate of change of each variable in the intervening years. For this reason, the best way to estimate the model, when these data are used to construct the dependent variable, is to attempt merely to explain the rate of change in service coverage, between 1980 and 1990, and 1990 and 2000, based on the overall picture of electoral competition in the intervening years. A true time-series analysis is not possible with the available data. Public Finance Data. The second approach to estimating government performance is to use municipal public finance data as a proxy measure of how hard the local government is working to provide public services. At first glance, this may not seem to be the most appropriate measure of government performance. But in fact, the specifics of the Mexican case are such that much can be learned about government performance from examining municipal finance data.17 There are two indicators that are of particular interest. First, we can determine how much money (as a raw total and as a percent of the total budget) is spent on gastos corrientes, or operating expenses. Starting in the mid-1980s, municipal governments in Mexico faced pressure to modernize and become more efficient, in response to decentralizing reforms that gave them more autonomy and responsibility. However, it seems clear that municipalities responded to this pressure in a range of ways, from inaction to swift modernization. In most cases, any efforts that municipalities made to modernize, such as updating office equipment or hiring knowledgeable personnel, is reflected in the operating expenses. According to the main hypothesis of this paper, then, we would expect to find that operating expenses as a percentage of the total municipal budget should be higher in municipalities that had previously faced more competitive elections. A second way to use public finance data to gauge how municipalities were changing is to calculate the amount of money that was raised locally (again, either as a raw total or as a percent of the total budget). All municipalities are able to raise their own revenue through taxes, fees, and fines. However, as described in a previous chapter, these revenue sources were largely ignored before the mid-1980s. As in the previous example, municipalities reacted differently to pressures for modernization and 16

This should not really be the end of the story. Whether it is fair or not for outside observers to judge municipal performance based on the crime rate, it seems clear that the public does make such a judgment. We might therefore expect municipal governments to be punished for high crime rates, whether they are really to blame for the problem or not. Future drafts of this paper will use available crime statistics to evaluate this possibility, though the current draft does not address the issue. 17 Appendix 1 provides a description of the coverage and definitions of the available public finance data.

11

performance. The data suggest that some began to systematize the collection of local taxes, while others did not take advantage of the potential revenue source. Cabrero and Orihuela (2000) find that large cities were much more likely to take advantage of local revenue sources, possibly because cities have a larger tax base from which to draw rents. In any event, the collection of local revenue can be seen as an attempt by the municipality to increase its revenue, modernize, and provide better services. So this statistic has a clear application to our main hypothesis: locally-raised revenue as a percent of the total budget (per capita) should be higher in municipalities that had previously faced more competitive elections. To be sure, these two indicators are merely indirect measures of government performance. In principle, it is possible (though unlikely, I think) that the municipalities that increase their operating expenses and local revenue either waste the money or do not spend it on public projects. The indirectness of the indicator might suggest that the first measurement strategy, an estimation of public service provision, is preferable. However, the two public finance indicators have one other advantage, which should not be overlooked. Whereas data on public service provision is only available from the decennial census, data on municipal revenue sources is available annually, starting in 1989. Of course, this renders other data that are available for the 1980s (and earlier) much less useful, but it provides an excellent opportunity to analyze year-to-year changes in government performance. The analysis below considers both the public service data and municipal finance data as indicators of government performance. Electoral Competitiveness. The concept of electoral competitiveness, or the “thing” that needs to be measured, is some sense of how threatened an incumbent party should feel, or of how worried they should be that they might lose the next election. All of the measures that I offer here are based on the assumption that political parties look to past electoral results in order to get an idea of how things are going. For example, if the PRI has won each of the past four elections with 80% or more of the vote, they should be less worried about winning an upcoming election than if their vote share has decreased by 10% in each of the past four elections, from 80% down to 50%. Or, the PRD should be more worried about remaining in power in a municipality that they won only by a plurality, than in one that they won with 65% of the vote. Of course, looking to past performance will not always tell the whole story: parties should also be concerned about whether a municipal election is to be held concurrently with a federal election, since there are often “coattail” effects in concurrent elections. They might also look to the local results of non-concurrent, but recent, federal elections to get a sense of how the electorate is shifting. Still, overall, municipal electoral returns should provide a reliable metric for how competitive future contests are likely to be. Accordingly, this section introduces several indicators of electoral competitiveness, all based on municipal vote data, and discusses the relevance of each measure to the current analysis. Perhaps the most common measure of how closely parties are competing is the effective number of parties index, developed by Laakso and Taagepera (1979). This index measures the degree of fractionalization in any given election by weighting the

12

relative vote totals for each party (or candidate) and summing them into a single statistic. The index is given by the following formula: N = 1 / Σ pi2, where p is the proportion of the vote for each of i parties. So, for example, if two parties each win half of the vote, the formula returns N = 1/(.52 + .52) = 2, and if three parties win a third of the vote each, N = 1/(.332 + .332 + .332) = 3. If, on the other hand, three parties compete but one wins 90% of the vote, and the others win 5% each, N = 1/(.92 + .052 + .052) = 1.23, indicating that even though three parties entered the election, in fact there is just over one effective party dominating the contest. This measure is widely used in political science for a number of purposes, and at first glance it seems to have an intuitive application to the measurement of electoral competition. For example, from among the examples calculated in the preceding paragraph, it is clear that the first two are competitive elections, and the third is not. Generally, we might expect that as elections are more closely contested, the index will increase monotonically. But one could object that the first and second examples are equally competitive, even if the outcome in the second example is less certain. To be less demanding, then, we might at least expect that any election that could intuitively be called “competitive” would have some minimum score on the index, such that one could choose a threshold of competitiveness and dichotomize the measure into competitive and non-competitive categories. However, it turns out that the mathematical properties of the formula for N make consistent dichotomization impossible. Appendix 2 demonstrates that, once the number of actual parties increases beyond two, the effective parties index does not behave as an intuitively consistent measure of competitiveness. Furthermore, the appendix demonstrates that no consistent threshold can be chosen when the number of actual parties varies from one contest to the next.18 A better measure of electoral competitiveness is the margin of victory (M1), calculated simply as the percentage of votes for the winning party less the percentage won by the second place party.19 This statistic offers a clear, intuitive, and monotonic indicator of how closely contested a given election is. For example, an election that is won 55% to 45% is clearly more competitive than a result of 60-40, or 70-30. However, while the N begins to look inconsistent when more than two parties compete, M still provides an intuitive, consistent measure. An electoral result of 50-45-5 is more closely contested than a return of 50-25-25. The margin of victory matches our intuition here, since the winner in the former case should clearly be more worried about the competition than the winner in the latter case (though neither should be complacent). Note, however, 18

Molinar’s (1991) suggested modification to N seems to alleviate much of the problem, but his measure also has limitations as a measure of competitiveness (see Appendix 2). 19 Cox (1988) criticizes the use of M in studies of voter turnout. His argument is persuasive, but does not apply to the current investigation. For one, Cox’s main methodological objection is specific to cases in which turnout is the dependent variable (the problem is that the total effective vote is used to calculate both the independent and dependent variable). Second, Cox’s objection applies to equally-sized districts, which greatly increase the chance of a spurious relationship. Neither of these conditions holds in the present investigation: turnout (or any other variable in which “total votes cast” is part of the dependent variable) is not used here as a dependent variable; and the case units (municipalities) vary widely in terms of population size and total votes cast.

13

that the effective number of parties measure would lead to the opposite conclusion, since N (50-45-5) = 2.20 and N (50-25-25) = 2.67.20 Given even a modicum of uncertainty about electoral support from one election to the next, it seems clear that the party winning by 5% should be more worried about their future than the party winning by 25%. But in the preceding example, the two winning parties obtained an identical proportion of the vote. The total vote of the losing parties is also identical. One might object, then, that the two winners should be equally worried about the next election, if there is a significant possibility that the losing parties will form a coalition for the next election.21 For this reason, I have also calculated a modified version of the margin of victory (M2), in which the combined vote percentage for the second, third, and fourth place parties are subtracted from the winning percentage. In some cases, this produces a negative value, indicating that a coalition of the losing parties would have defeated the winning party (assuming sincere party voting).22 In any event, M2 correlates extremely well with the simpler measure (see Table 3), making it unlikely that the choice of one indicator over the other would affect the analysis presented below. One problem with any statistic that uses Mexican electoral data, which I have not mentioned to this point, is the widespread suspicion (if not fact) that many elections in the 1980s and 1990s were marred by vote fraud. Fraud may have been extensive in some cases, especially in the 1980s, and especially in rural areas. But all is not lost in the presence of vote fraud. Among other details about how fraud was committed, Molinar (1991) notes that it was much easier for the PRI to add votes to its own total (for example, by sending groups to vote multiple times at different polling stations, or via the taco, several ballots rolled together and introduced into the urn by one voter), than to subtract votes from the opposition. Many election observers also believe that the voting lists were often “padded” by the PRI, either by legitimately registering likely PRI-voters or by adding ineligible names. On the one hand, the possibility of vote fraud may not present a problem for the current analysis. According to the theory, electoral competitiveness is significant because it serves as a sign of how worried the incumbent party should be about future elections. It is entirely possible that the PRI, for example, would account for its own ability to commit fraud when considering the likely outcome of future elections. That is, the party might not distinguish between a “safe” municipality won by fraud and one won by real popular support. If the party felt confident in its ability to continue unhindered with fraudulent practices in the future, it would have no reason to make such a distinction. On the other hand, it turns out to be the case fraud did become more difficult to commit toward the end of PRI hegemony, especially in the 1990s. Opposition parties and the press became more vigilant, and were able to force the PRI into annulling the results of several elections. If, as seems likely, local PRI organizations began to worry that their 20

This is not meant to demonstrate that there is anything inherently inconsistent with the effective number of parties index. It merely shows that employing the index as a measure of electoral competitiveness, as defined here, is not entirely appropriate. 21 As a practical matter, this should be a worry for all political parties in Mexico. During the past few years all of the major political parties have entered into coalitions for either municipal or gubernatorial elections. Even the PAN and the PRD, who are thought to be at opposite ends of the ideological spectrum, have forged alliances, most notably in the gubernatorial race in Chiapas 2000 (which they won). 22 In fact, this is true in 1329 cases, or 9% of all elections.

14

ability to influence elections through fraud would be hampered in the future, they might be interested to know how an election would have turned out in the absence of fraud. Following Molinar’s (1991) suggestions, even in the presence of fraud, we should be able to measure the degree of opposition in a municipality, and therefore the potential for a future opposition victory, by measuring the opposition vote as a percentage of the votingage population. Since the data are available and reliable for both opposition vote total and the voting age population, I have calculated two variables based on this data. Analogously to the margin of victory, I calculate the following: Opposition Vote (O1): 2nd place total vote / population ≥ 18 Opposition Vote (O2): (2nd + 3rd + 4th total vote) / population ≥ 18 Table 3 presents correlations among the five measures of electoral competition discussed here. There are several notable results in the table. First, all of the measures are very closely correlated. Keeping in mind that competitiveness decreases as the margin of victory increases, we would expect M1 and M2 to be negatively correlated with the other measures. So, all of the correlations are in the expected direction, and the smallest coefficient in the table is .67, which is still a rather large. In spite of the conceptual and econometric problems discussed above, it seems to be the case that all of these measures are tapping into the same source of variation. Second, as we might expect, the two measures of margin of victory, and the two measures of opposition votes, are extremely closely correlated. It seems unlikely that the subsequent analysis would be affected by the choice of one of these measures over the other. To simplify, then, I will focus on three measures of competition in the subsequent analysis: N, because it is widely used and easy to interpret; M1, because it is the simplest and most direct measure of the potential future electoral threat, and O2, because it provides an overall sense of “anti-incumbent” vote and should be relatively immune to distortions caused by fraud. Table 3. Pairwise Correlations Among Five Measures of Electoral Competition

N M1 M2 O1 O2

N

M1

M2

O1

O2

1.00 -0.91 -0.95 0.67 0.80

1.00 0.99 -0.82 -0.84

1.00 -0.79 -0.85

1.00 0.96

1.00

Note: N=14702 for the first three rows, and 14640 for the last two. All correlations are significant p