Partisanship and Representation in Local Politics

Partisanship and Representation in Local Politics: New Evidence from Mid-Size U.S. Cities Abstract Do elected officials in city government have the tools necessary to respond to the preferences of their constituents? Or, is the menu of available policy options effectively determined by forces beyond their direct control? Using what we believe to be the most comprehensive and detailed database linking voter preferences to local policy outcomes covering more than 1,000 mid-size cities, we examine the degree to which local governments adapt local policies in response to voters’ views. Overall, our analysis paints a positive picture of democracy in the city. We document tremendous variation in local policy outcomes and show that voter preferences play an important role in explaining why cities tend to adopt different policies. As they grow more Democratic, cities increase their spending across a number of service areas. In addition, we show that voter sentiment shapes the other side of the ledger, determining the mix of revenues on which cities rely. As they grow more Democratic, cities adopt more progressive forms of taxation, relying on property taxes to a greater extent than regressive sales taxes. In short, cities appear to be quite responsive to the needs and wants of their constituents, even on redistributive questions generally thought to lie beyond the capacity and competency of local government. 1 Introduction Democratic representation in the United States is a complex, multi-level affair. Every few years, voters have the opportunity to select the president, determine the balance of partisan power in Congress, choose officials to fill state offices, pick councilmen and mayors, and, in many cases, participate in a large number of other low-profile races, including elections of judges, school board members, tax assessors, and mosquito abatement commissioners.1 Many normative theorists praise America’s system of federalism and decentralization, arguing that allowing policy to be made by a number of smaller jurisdictions provides constituents with the benefits of choice (Ostrom, Tiebout, and Warren 1961; Tiebout 1956), enhances oversight and accountability by bringing policymaking closer to the people (Fisman and Gatti 2002), and creates “laboratories of democracy” that allow for experimentation and eventual diffusion of effective policies and best practices (Makse and Volden 2012; Volden 2006). Although national and state races often attract the greatest attention from the media and public, most policymaking that directly impacts voters on a day-to-day basis actually takes place at the local level. Cities, in particular, control a number of vital public functions — including public safety, fire protection, land-use regulations, and transportation options — that affect public health, safety, and quality of life in very real and often dramatic ways. Our knowledge of the political dynamics in America’s cities remains very limited, however. Many observers have raised serious questions about whether elected officials in city government have the tools necessary to respond to the preferences of their constituents or whether the menu of available policy options is effectively determined by forces beyond their direct control. In this paper, we assess the quality of democratic representation in America’s cities. Using what we believe to be the most comprehensive and detailed database linking voter preferences to local policy outcomes covering more than 1,000 mid-size cities, we examine the 1 Lowery, Lyons, and DeHoog (1990) among others have raised serious questions about whether the American system is too complex, making the task of matching elected officials with the policy outcomes for which they are responsible impossible for regular voters. Others have noted the potential for collective-action problems among local agencies that compete with one another for scarce public resources (Berry 2008). 3 degree to which local governments respond to the wishes and desires of their voters. Unlike existing studies that rely on rough proxies of local voter preferences (e.g., Craw 2010; Hajnal and Trounstine 2010), we construct direct measures of local partisanship by aggregating precinct-level presidential results at the city level. This allows us to expand our analysis to a much larger number of jurisdictions and to measure voter preferences much more precisely than has previously been the case. Overall, our analysis paints a positive picture of democracy in the city. We document tremendous variation in local policy outcomes and show that voter preferences play an important role in explaining why cities tend to adopt different policies. As they grow more Democratic, cities increase their spending across a number of service areas. In addition, we show that voter sentiment shapes the other side of the ledger, determining the mix of revenues on which cities rely. As they grow more Democratic, cities adopt more progressive forms of taxation, relying on property taxes to a greater extent than regressive sales taxes. In short, cities appear to be quite responsive to the needs and wants of their constituents, even on redistributive questions general thought to lie beyond the capacity and competency of local government. We begin by reviewing the existing record of theoretical and empirical research on city government. Prior research largely agrees that external forces limit the ability of local officials to respond to their constituents’ preferences. Next, we describe our new dataset on voter partisanship in American cities built from precinct-level presidential vote results. Studies on voter behavior show that, given ever-increasing ideological sorting by party, presidential vote share serves as an accurate gauge of voter sentiment on a wide range of policy questions (e.g., Levendusky 2009; McCarty, Poole, and Rosenthal 2006). In the following section, we document the relationship between local partisanship and policy outcomes for a large number of mid-size American cities. The conclusion describes the broader implications of our findings for American democracy. 4 2 Constraints and the Growth Imperative in City Government Since the 1970s, political parties have come to play an increasingly central role in American politics. At the national level, parties regularly stake out distinct platforms on the most salient issues of the day, particularly in the modern age of rising partisan polarization (Abramowitz 2010; Gerring 2001; Hacker and Pierson 2010; Jacobson 2003; McCarty, Poole, and Rosenthal 2006). Party leaders have also emerged at the forefront of negotiations between the branches of government and have carefully managed the policymaking taking place within them (see, e.g., Alrich 1995; Cox and McCubbins 1993; Cox and McCubbins 2005; Kiewiet and McCubbins 1991; Rohde 1991). Among voters, party affiliation is one of the most important and consistent predictors of voting behavior on Election Day (Campbell, Converse, Miller, and Stokes 1960). It serves both as an important heuristic for voters sorting through complex policy positions (Popkin 1994) and as a social identity shaped by voters’ views and opinions about the world around them (Green, Palmquist, and Shickler 2002). Despite the documented importance of partisan conflict at the national level, most scholarly accounts of local politics either overlook or expressly dismiss the role of partisanship in local government. In their now-classic textbook on city government, Banfield and Wilson (1963) assert that, in local elections, “the party as such seldom has any concrete program or platform” (p. 277). While many argue that local issues are inherently nonpartisan — Fiorello La Guardia once famously observed that there is no Democratic or Republican way to pick up the garbage — others point to political and economic constraints that limit the discretion of local policymakers and can thus thwart partisan policy goals. Many sources of constraint are vertical, resulting from local governments’ subordinate legal status as “creatures of the state.” This context sometimes places cities at the whim of inhospitable state legislatures (Desmond 1955). Gamm and Kousser (2010) find evidence that such interference 5 is most likely to occur when the preferences of local voters diverge from the political interests of the officials who control the state house. The most binding constraints, however, appear to be horizontal: they emerge through the process of competition among neighboring cities to attract and retain scarce resources. In the most prominent formulation of this argument, Peterson (1981) describes how jurisdictional competition among local agencies can effectively limit the policy discretion available to local officials, particularly on economic issues. Competition, and the mere threat thereof, creates a Darwinian dynamic in local government that pushes all cities, regardless of the preferences of their voters or leaders, toward policy convergence. To succeed, cities must grow, and to grow, cities must compete with one another to attract the mobile capital and high-skilled labor critical to sustaining a successful economy. The growth imperative imposes severe limits on the scope of local policymaking, helping explain the “lack of a distinctive partisan impact on local policy outcomes” (Peterson 1981, p. 174). In particular, Peterson argues that competition among cities eliminates opportunities for economic redistribution by giving wealthy taxpayers and business owners a credible exit option they can use to avoid intolerably high local tax burdens or inadequate support for the interests of the local business community. Anticipation of such tax avoidance, with its concomitant negative impacts on the local economy, gives even the most progressive community good reasons to not enact its preferences into actual policy. “The politics of redistribution at the local level,” Peterson writes, “is thus an arena where certain kinds of citizen needs and preferences seldom become demands; an arena where redistributive questions, even when posed as major political issues, are treated by a variety of strategies designed to forestall, delay, and preclude their implementation” (p. 182). In short, competitive threats push all cities — regardless of the preferences of their voters and residents — to both adopt the same types of policies in order to maximize their communities’ economic opportunities and avoid using local taxes and expenditures to carry out economic redistribution. 6 3 Re-evaluating the RD Design Despite strong theoretical expectations of policy convergence, relatively little empirical work has attempted to measure the influence of partisanship on local policy outcomes. In addition to being small in quantity, much of the existing work points to mixed findings. In two early studies of municipal expenditures, Brazer (1959) and Fried (1975) find little evidence of a relationship between local partisan leanings and patterns in city spending. Minkoff (2012) also documents evidence of convergence, especially among spatially proximate jurisdictions. By contrast, other studies find substantial variation in policy adoption across cities, some of which appears to be explained by local political considerations (e.g., Hajnal and Trounstine 2010). One important drawback in the existing literature — and, we believe, one cause for the conflicting findings — is insufficient clarity about the proposed mechanisms and poor translation of the theoretical construct into empirical operational measures. It useful to separate the relationship between partisanship and policy into two separate components: partisanship of the voters and partisanship of the officials they elect. While there is clearly a strong relationship between the two — overall, Democratic voters tend to support Democratic lawmakers, even in local elections (Oliver and Ha 2007; Schleicher 2007) — the latter does not always serve as useful proxy for the former. We focus our attention in particular on two recent studies, Ferreira and Gyourko (2009) (F+G) and Gerber and Hopkins (2011) (G+H), which use the regression discontinuity design to examine the relationship between partisanship and policy in American cities. Both sets of analyses build on the work of Lee, Moretti, and Butler (2004) and exploit close mayoral elections as “natural experiments.” These studies represent some of the most methodologically sophisticated work on the subject, with an emphasis on causal inferences. They are, however, flawed in several respects. First, although both studies purport to study the influence of partisanship in local politics in broad terms — F+G describe their results as measuring the “strength of partisan impulses at the local level” (p. 421) while G+H ask, “Does par7 tisanship lead cities to adopt significantly different fiscal policies?” — the design of both studies allows them examine only the party in government while providing few insights about the role played by the party in the electorate (Key 1964). Second, the strength of the RD design rests on the assumption that close elections are indeed decided randomly. We show that, in the case of G+H, this assumption appears to be violated in a way that undermines the authors’ central findings. It is worth considering each of these critiques in more detail. Lee (2008) demonstrates the conditions under which elections provide an as-if random assignment mechanism that can be used to identify causal effects in observational studies. Both F+G and G+H examine mayoral elections in American cities, focusing on close contests in which Republican and Democratic candidates prevail over their opponents by very small margins. When these elections are so close that the winner is determined effectively by chance, this design ensures that elected Democratic mayors are identical to their Republican mayors on most covariates, providing an unbiased estimate of the effect of mayoral partisanship on policy outcomes. While close elections have a number of properties that make them useful for estimating many quantities of interest, one clear disadvantage is that this design estimates only the effect of mayoral partisanship, ignoring other pathways through partisanship may be influencing local policy outcomes. Indeed, the key assumption of the design is that cities with close elections are identical both in terms of the partisan composition of city councils — the legislative branches of local government that generally have greater direct influence over policy than mayors who possess few formal legislative powers — and the partisanship of the electorate. By effectively holding constant other partisan influences, the regression discontinuity design estimates the effect of mayoral partisanship on policy. It cannot, however, speak to the other potential mechanisms through which partisanship might play an active role in local democracy.2 By contrast, our analysis shifts the focus to examine the role of 2 A related concern is that RD studies estimate only the local treatment effect. If expect close elections to be different from other contests — for example, by depriving the winning mayoral candidate of a clear mandate to implement their preferred policies — these local effects may be substantially different from the broader phenomena to which we may wish to generalize the findings. 8 voter preferences, providing evidence that voters’ partisan predispositions have important impacts on the size of local government and the distributive implications of city fiscal policy.3 The limited substantive focus of regression discontinuity studies likely explains why this work produced largely null findings. There is, however, one clear exception: G+H appear to show that Republican mayors allocate a greater share of local resources toward public safety, increasing the share of the budget spent on police relative to their Democratic counterparts. The causal strength of this finding rests on the critical assumptions that very close mayoral contests are indeed decided randomly, something that only rarely appears to be the case in practice (see Caughey and Sekhon 2011; Grimmer, Hersh, Feinstein, and Carpenter 2012). Instead, close electoral contests magnify inherent differences between candidates — in terms of candidate quality, resources, and interest group endorsements — by providing the most likely context in which relatively small electoral advantages are likely to make the difference between winning and losing the election. A similar problem appears to drive G+H’s finding about the relationship between mayoral partisanship and police spending. In Figure 1 below, we replicate the authors’ regressiondiscontinuity analysis to show that there is indeed a difference in police spending between cities with Republican and Democratic mayors. Consistent with the results reported by G+H, cities in which Republican candidates prevail do indeed appear to allocate a greater share of their budget on police over the first three years of each mayor’s term compared to other jurisdictions with Democratic winners. If cities that elect Republican mayors are different than those with Democratic winners on other dimensions, this finding cannot be interpreted causally, however. For example, it is possible that Republican mayoral candidates are more likely to win elections in cities with stronger police unions, which have historically endorsed Republicans, because the endorsement from public safety unions should matter most in a particularly competitive election. However, the same underlying factors that helped Republicans win may also explain the subsequent policy outcomes (e.g., cities with 3 The null findings in the RD studies suggest that mayoral partisanship is probably not the mechanism through which voter preferences get translated into policy outcomes. 9 strong police unions allocate a greater share of their budget toward public safety). Impact on Policing ● ● ● ● ● ●● ● ● 0.00 ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.2 0.4 0.6 0.20 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● 0.8 Democratic Vote Share Figure 1: Effect of mayoral partisanship on change in city police spending (replicating Gerber and Hopkins 2011). To examine whether this type of confounding can explain G+H’s findings, Figure 2 presents results from a simple placebo test. Rather than examining the relationship between mayoral partisanship and relative spending growth over the mayors’ first three years in office (the quantity of interest in the G+H analysis), our placebo test focuses on differences in police spending as a share of local budgets prior to the beginning of each mayor’s term. Specifically, we examine the percent of each city budget spent on policing during the year of the election, a number determined through a budget that was adopted under the previous mayoral administration. For close elections to produce unbiased results, we should expect to see no differences between cities in with Republicans victors and those that elect Democratic mayors prior to their election. However, this is not the case. Instead, Figure 2 shows clear differences, with cities that eventually elect Democratic mayors spending substantially larger share of their budgets on police than cities with Republican winners (p=0.01). 10 ● ● 0.00 −0.04 ● ● ● 0.15 ● 0.10 ● ● Repu W 0.05 Democrat Wins Police Spending Prior to Election Republican Wins 0.04 Change in Police Spending 0.08 ● 0.2 g Placebo Test ● 0.20 Republican Wins ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● 0.8 ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ●● ● ● ●●● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.00 ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.10 ● ● 0.05 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● Democrat Wins ● ● ● ● ● 0.15 Democrat Wins Police Spending Prior to Election ● ● 0.2 re 0.4 0.6 0.8 Democratic Vote Share Figure 2: Placebo test: Effect on mayoral partisanship on police spending under previous administration. The magnitude of the effect, 3 percentage points, is almost identical to the 2.5 percentage point difference between Republican and Democratic mayors reported in G+H. The placebo test points to two possible explanations. First, it may be the case that low levels of police spending create political opportunities for Republican candidates, who can exploit public concern over public safety to their electoral advantage. Alternatively, the G+H result may simply represent regression to the mean, with cities that previously spent a smaller share of their budget on policing increasing their public safety budget at a faster rate rate over time. Regardless of which explanation is the correct one, the placebo tests provides clear evidence that the differences between these cities do not appear to be causally related to the partisanship of their mayors. 11 4 Data and Methods Unlike the RD studies, which focus exclusively on mayoral partisanship, we look at the relationship between the partisan predispositions of voters and local policy outcomes. Thus, our analysis expands the mechanisms through which parties can play a role, including election to legislative offices and even perhaps direct democracy. We discuss the possible mechanisms and offer our speculation about which ones appear to be most consistent with the empirical relationships we document in the conclusion. In our analysis, we use three types of data to estimate the impact of partisanship on local governing: (1) municipal-level partisanship; (2) local spending; (3) local revenue sources. In addition, we include a variety of important controls to address the potentially confounding influences of municipal demography and state-level legal context. 4.1 Municipal Partisanship To calculate our key independent variable — the proportion of a municipality that is Democratic — we use precinct-level presidential returns from the Federal Elections Project dataset (Lublin and Voss 2007). With these precinct results, we generate an original data set that calculates the the percent of each municipality that supported Democratic candidate Al Gore in the 2000 presidential election. Because the Federal Election Project includes precinct-level data from only one election, our results look at cross-sectional variation and do not allow for comparisons within cities over time. Our data’s cross-sectional nature introduces one important limitation into our analysis: we cannot fully separate whether an observed connection between Democratic vote share and spending are a result of preferences shaping spending or preferences being shaped by spending. We suspect both are at play. However, even finding a link between partisanship and local spending/revenue decisions would represent an important improvement in our understanding of local democracy; thus far, no study has uncovered a clear and consistent correlation between voter preferences and local government resource 12 Table 1: Regional Composition of Metropolitan Areas With and Without Municipal-Level Data Available Northeast (%) Midwest South West Municipal Data Available n = 148 39.9 39.2 2.0 18.9 All Metro Areas n = 326 18.8 23.6 38.6 19.0 allocation. We draw our municipalities from the nation’s 148 metropolitan areas that identify their precincts by municipality.4 These labels allow us to aggregate precinct-level data to the municipal level. Our 148 metropolises — while drawn from a wide array of geographic regions — are not representative of the nation as a whole. As Table 1 illustrates, they overrepresent the Northeast and Midwest and underrepresent the South due to regional differences in the importance of municipal governments.5 Indeed, in the South, county governments — rather than their municipal counterparts — largely take on the role of local governing. Therefore, our results may have limited generalizability in the American Southeast. Our use of municipal-level presidential returns represents a significant departure from previous studies of political geography. Most of this literature relies on county-level returns (e.g., Bishop and Cushing 2008; Gimpel and Schuknecht 2004; Glaeser and Ward 2006; Massey, Rothwell, and Domina 2009), although some more recent research has taken advantage of the wider availability of precinct-level data (e.g., Chen 2012; Chen and Rodden 2012; Einstein 2011). Moreover, no research assessing the impact of partisanship on local government spending has, to our knowledge, used partisan data at such a low level of geographic aggregation. These municipal-level presidential returns thus represent a unique opportunity 4 In addition, we include all municipalities from the states of California and Michigan, which make detailed local election data publicly available. Our analysis is limited to municipalities located in metropolitan areas due to the constraints of the Federal Elections Project database. The coverage of our sample, however, is many times larger than other comparable analyses. 5 The data for all national metropolitan areas excludes metropolises with overly aggregated or missing precinct-level data: Kankakee, IL PMSA, Richland-Kennewick-Pasco MSA, Seattle-Bellevue-Everette, WA PMSA, Bremerton, WA PMSA, or Tacoma, WA PMSA. 13 to test whether mass partisanship has an impact on local government spending and revenue decisions. 4.2 Spending and Revenue Sources To measure local government spending and revenues, we take advantage of a wide array of data from the Census of Governments. These include total spending, expenditures broken down by type (police, fire, libraries, development, transit, and welfare), operational expenditures (again broken down by type), staffing and pay records, and details about city revenue sources. Consistent with previous research on national-party politics — but in contrast with previous non-findings on the connection between partisanship and local politics — we expect that more Democratic municipalities will, on average, spend more and derive their revenues from more progressive sources (e.g., property rather than sales taxes). Indeed, a wealth of political science research links the national Democratic Party with redistributive spending and more progressive forms of taxation (Bartels 2008; Petrocik 1996; Petrocik, Benoit, and Hansen 2003) We calculate two different kinds of spending and revenue dependent variables: per capita and proportional. The per capita measures simply calculate the dollars spent (or revenues raised) per person in a given municipality; for example, in our data set, the average municipality spends $951 per person. Conversely, the proportional measures estimate the percentage of, say, local revenues derived from sales taxes (11.7 percent in our average municipality). 4.3 Demographic Controls and State Fixed Effects There are, of course, a myriad of confounding variables that might explain why Democratic cities spend more or tax more progressively. We therefore control for virtually every municipal demographic variable for which we can collect U.S. Census data. These municipal characteristics include: median household income, median housing value, percent of families in poverty, unemployment rate, percent black, percent Hispanic, percent urban, density per 14 square mile, commuting time, and percent of residents that own their homes. Because most of our spending and revenue variables are measured per capita, we do not include population as a control in our main models. However, as a robustness check, we re-ran all of our models with population incorporated, and none of our results changed in terms of substantive effects or statistical significance. We limit our analysis to municipalities with a population of less than 500,000 (n = 1, 351). Existing research suggests that larger cities — which tend to be overwhelmingly Democratic and thus do not vary on our key independent variable — follow different spending priorities and collect revenue in unique and idiosyncratic ways (see Judd and Swanstrom 2012, Chapter 12). Finally, we also include state fixed effects. As legal “creatures of the state,” municipalities only possess the powers to tax and spend allotted to them by their state governments and constitutions. State legal context thus shapes both municipal spending and revenue collection decisions in important ways. Failure to account for this state-level context would therefore clearly bias any empirical investigation. The impact of state laws is most clear in the context of taxation; only 28 states allow cities to levy sales taxes. Thus, local revenue policies vary dramatically across states. In Oklahoma, for example, municipalities use the sales tax as the only source of general tax revenues. Conversely, in other states, cities also rely on the property tax and — where permitted — income taxes to supplement their tax revenues (National League of Cities 2002, p. 4). To ensure adequate degrees of freedom in these state fixed effects models, we limit our analyses to municipalities located in states for which we have more than 50 observations, reducing our sample size to 1,041.6 4.4 Model We use OLS regressions to assess the relationship between Democratic vote share in each municipality and local spending and revenue decisions. To address heteroskedasticity, all 6 All results presented below, however, remain in the same direction and statistically significant in the same direction when we run non-state fixed effects models with our full sample n = 1, 351. 15 standard errors are robust. 5 Results We first present our results for spending — including total expenditures levels, operations spending (all expenditures excluding capital projects), and spending on payroll — before moving to our discussion of revenues and debt. 5.1 Spending Our models reveal that, all else held constant, Democratic cities spend more total, on operations, and on pay than their Republican counterparts across an array of spending types. Table 6 illustrates the magnitude of these results. To allow easy comparison across the many different categories of spending, we only present our OLS coefficients for our key independent variable (the proportion of a municipality that is Democratic) with heteroskedasticityconsistent standard errors in parentheses. All of these coefficients, however, are derived from models containing a full set of controls, including income, poverty, unemployment rate, percent black, percent Hispanic, percent urban, density per square mile, commuting time, percent homeownership, and state fixed effects. Our analysis of staffing focuses on police, fire, and libraries, as these are the most common functions widely provided by city governments in nearly every state. Full model results are available in the Appendix A. The dependent variables in each model can be identified by looking at column and row headings in Table 6. For example, looking at the first row from left to right, the dependent variables are total municipal expenditures and total expenditures on operations (both per capita). Table 6 reveals that, as expected, more Democratic cities spend more — in total, on operations, and on payroll — across a wide array of spending arenas. The coefficient on Democratic vote share is most consistent in three arenas: total expenditures, police spending, and welfare spending. Looking at the per-capita total spending models, an increase in 16 Table 2: Impact of Democratic Vote Share (2000 Presidential Election) on Local Spending by Spending Type. Democratic Vote Share Coefficient Total Expenditures Police Fire Libraries Development Transit Welfare **p < .01 *p < .05 Per 4.940 .919 .250 .210 .813 -.070 .019 Capita Total Per Capita Operations Payroll Costs (Per 1,000) (1.08)** 4.370 (.861)** (.278)** .963 (.268)** 11.024 (3.28)** (.175) .309 (.163) 20.643 (5.91)** (.124) .184 (.056)** 16.863 (4.96)** (.285)** .287 (.198) (.352) .389 (.184)* (.009)* .018 (.009)* Each coefficient represents the impact of the Democratic vote share in the city on a particular spending category in a fully specified OLS model, with demographic controls and state fixed effects. Robust standard errors for the proportion Democrat coefficients are in parentheses. The number of observations for all models except for pay is n = 1, 041. The number of observations in the pay models are n = 613 (police), n = 426 (fire), n = 194 (libraries). the proportion of a municipality that voted for Al Gore in 2000 by one percentage point corresponds with an almost $5 per person increase in total spending, all else held constant. Similarly, a one percentage point increase in Democratic support is associated with a roughly $1 per person increase in police spending and a 2 cent per person increase in welfare spending, again, all else held constant. Moreover, the impact of Democratic vote share on these spending decisions is remarkably similar when only operations spending is considered. The magnitude of the effects in our our welfare and police spending models is particularly interesting. Indeed, after implementing major redistributive initiatives like the New Deal and the Great Society, the Democratic Party has been inextricably linked with federal welfare spending. Our findings suggest that this connection remains as powerful at the local level, with Democratic population concentrations associated with higher local welfare expenditures, all else held constant. Contrary to Peterson’s (1981) argument, there are clearly Democratic and Republican approaches toward redistribution in America’s cities, just as there are at the national level. 17 While social welfare is an issue that has been firmly identified as Democratic at the national level, the Republican Party has historically retained a reputation of being tough on crime (e.g., Petrocik 1996; Petrocik, Benoit, and Hansen 2003; although see Holian 2004). Nonetheless, we see a close positive relationship between the Democratic vote share in each municipality and its levels of per capita police spending. Although Republicans may prioritize public safety programs, Democrats’ support for bigger city government appears to apply to police functions as well. One possibility is that Democratic cities simply experience more crime. We control for many of the most likely drivers of local crime — including unemployment, percent non-white, percent urban, density, poverty, income, and rates of homeownership — but we do not have a municipal-level measure of crime in our models. In addition, it might be that left-leaning political leaders seek to neutralize the inherent Republican advantage on this issue by spending more on the police force. 5.2 Revenue We also discover that, all else held constant, more Democratic cities are more likely to collect their taxes progressively, relying more heavily on the property tax than the sales tax. The impact of partisanship on revenue choices, however, appears to be substantively smaller than in local spending decisions. Table 3 highlights this relationship by displaying the model coefficients for our key independent variable of interest, Democratic vote share at the city level. These coefficients are derived from our full state fixed effects models. The dependent variables in this table can be easily identified by looking at the row names: from top to bottom, we explore the impact of local voter partisanship on total own source revenue (the total revenue derived from local taxation), per capita property tax collection, per capita sales tax collection, sales tax as a percent of total revenue, total per capita debt, and total per capita long-term debt. The only statistically significant impact of Democratic vote share emerges when sales tax as a share of total revenue is the dependent variable. All else held equal, as cities grow more 18 Table 3: Impact of Democratic Vote Share (2000 Presidential Election) on Local Revenues and Debt. Total Own Source Property Tax (Per Capita) Sales Tax (Per Capita) Sales Tax as Percent of Total Own Source Revenue Debt (Per Capita) Long-Term Debt (Per Capita) **p < .01 *p < .05 Democratic Vote Share Coefficient .087 (1.12) .156 (.507) .434 (.245) -.067 (.022)** .728 (2.50) .931 (2.48) Each coefficient represents the impact of Democratic vote share in the city on a particular revenue/debt category in a fully specified OLS model, with demographic controls and state fixed effects. Robust standard errors for the coefficient on proportion of votes cast for Democrats are in parentheses. Democratic, they rely less on the sales tax for their revenue. The impact, however, appears to be relatively modest: all else constant, an increase in the municipal Democrat vote share by ten percentage points corresponds with a .6 percentage point decline in share of revenues derived from the sales tax. In our average municipality, this means that $181,953 in local revenues formerly collected via the sales tax would now be obtained using a more progressive taxation form, like the income or property tax. 5.3 Robustness Checks To ensure that our results hold under multiple specifications, we conducted a number of robustness checks. First, as we mentioned in our data and methods section, we re-ran all of our models — even those with a per capita dependent variable — adding the municipal population (logged for data analysis) as a control. None of our results changed. In addition, to address the possibility of model oversaturation (Achen 2006), we estimated a series of state fixed effects models in which Democratic vote share is the only independent variable. We continued to include the state fixed effects because of the critical role state legal context plays in determining the availability of local spending options and revenue sources. 19 The Democratic vote share coefficient does not change in direction or significance in any of our key spending results (total, police, or welfare spending). However, the coefficient on the Democratic vote share variable does change in both direction and significance when it is the sole explanatory variable and the share of revenue derived from the sales tax is the dependent variable. We added each of our demographic controls one at a time to see which needed to be in the model for the proportion Democrat to become statistically significant; we discovered that the density per square mile has to be controlled for in the the share sales tax model in order for the proportion Democrat to have a negative and statistically significant coefficient. Taken in concert with the substantively small effect on the share of local revenue collected by sales tax, we feel much more confident in our spending results than in our revenue findings.7 6 Discussion The strength of our spending results relative to our revenue findings suggests that local governments have more leeway to represent the interests of their constituents in expenditure allocation than in revenue collection. Why might preferences be better represented in spending than in taxation? One potential answer lies in the inter-jurisdictional fragmentation described in Peterson’s (1981) seminal account. Competing metropolitan municipalities might worry about losing residents and businesses to neighboring communities with lower tax rates. These powerful economic forces may constrain the range of tax rates (and types of taxation) actually available to local political leaders. Another possibility centers on state regulatory power over taxation. We noted in our discussion of state fixed effects that state governments have a particularly potent influence on the taxation options available to local governments, with many municipalities unable to levy 7 The coefficient on the Democratic vote share variable in the sales tax model also fails to remain statistically significant when we estimate a robust regression using iteratively reweighted least squares. Conversely, all of our spending models the coefficients remain in the same direction and statistically significant under this robust specification. 20 any sales tax. Thus, municipal legal context may be working in concert with jurisdictional competition to dampen the representation of political preferences in the taxation realm. 7 Conclusion Overall, our results provide clear evidence that city governments are responsive to the preferences of their voters. As cities grow more Democratic, they increase their levels of service provision across an array of policy arenas in response to their voters’ desires for bigger government. Thus, the partisanship of voters plays an important role in accounting for variation in budgetary policy across cities. In light of the null/mixed findings in the mayoral partisanship literature — indeed, G+H’s are the opposite of those presented here — our results suggest that scholars interested in exploring the process of policy formation in cities should focus their attention on mechanisms outside of the partisan identification of mayors. In particular, we believe that city councils — which are even more powerful than Congress in terms of their ability to shape budgetary policy in the face of executive branch opposition — represent one of the primary mechanisms through which voter partisanship comes to influence local policy outcomes. It is also probably the case that the electoral connection looms large in the minds of elected officials: Republican mayors elected in largely Democratic cities would certainly think twice about not giving their constituents what they want if their intent is to remain in office after the next election. Future research might begin to adjudicate between these two possibilities by taking advantage of ever-improving longitudinal precinct data. Such longitudinal data would also allow scholars to address another potential question presented by this analysis: does public opinion affect spending levels, or does spending shape public opinion? The causal arrow likely points in both directions; we are unable, however, to empirically separate these two separate causal pathways with our cross-sectional data. Nonetheless, by uncovering the first empirical link between partisan preferences and local spending, our results represent 21 an important first step in demonstrating that local public opinion matters and that local governments have the power to respond to these voter preferences. 22 Table 4: Impact of Democratic Vote Share (2000 Presidential Election) on Total Local Spending by Spending Type (Per Capita Total Expenditures, Per Capita Police Expenditures, Per Capita Fire Expenditures, and Per Capita Libraries Expenditures). Prop Dem. Med. Housing Value Med. HHHold Inc. % Poverty % Unemployed % Black % Hispanic % Urban Density per Sq Mile Commute Time % Own Home Per Capita Total 4.941 (1.08)** -.006 (.011) .019 (.005)** .034 (.056) -.052 (.085) -.002 (.003) -.008 (.003)* -.006 (.004) -.034 (.005)** -.931 (.212)** -.015 (.004)** n = 1081 R2 = .135 Per Capita Police Per Capita Fire Per Capita Libraries .919 (.278)** .262 (.175) .210 (.124) -.007 (.002)** .008 (.002)** .03 (.001)** .002 (.001) .003 (.001)** .002 (.000)** -.039 (.013)** -.008 (.009) -.001 (.004) -.011 (.020) .014 (.014) -.007 (.008) .003 (.002) .001 (.001) .000 (.000) -.001 (.001) -.001 (.001) .001 (.000)* -.006 (.002)** -.001 (.000) .(000) (.000) -.005 (.002)** -.001 (.001) .000 (.001) -.074 (.058) -.030 (.033) -.030 (.012)* -.005 (.001)** -.002 (.001)** -.000 (.000) n = 1081 n = 1081 n = 1081 R2 = .203 R2 = .139 R2 = .145 **p < .01 *p < .05 OLS coefficients with heteroskedasticity-consistent standard errors in parentheses. Models include state fixed effects. Table 5: Impact of Democratic Vote Share (2000 Presidential Election) on Total Local Spending by Spending Type (Per Capita Development Expenditures, Per Capita Transit Expenditures, and Per Capita Welfare Expenditures). Per Capita Development Prop Dem. .813 (285)** Med. Housing Value .001 (.003) Med. HHHold Inc. .003 (.002)* % Poverty -.012 (.018) % Unemployed .048 (.029)* % Black .001 (.001) % Hispanic .000 (.001) % Urban -.001 (.001) Density per Sq Mile -.002 (.002) Commute Time -.137 (.060)* % Own Home -.001 (.001) n = 1081 R2 = .135 **p < .01 *p < .05 Per Capita Transit -.070 (.352) -.002 (.004) .003 (.001)* .018 (.018) -.033 (.028) -.001 (.001) -.004 (.001)** -.000 (.001) -.006 (.002)** -.186 (.072)* -.002 (.001) n = 1081 R2 = .109 Per Capita Welfare .019 (.009)* -.000 (.000) -.000 (.000) .001 (.001) .000 (.000) .000 (.000) .000 (.000)* -.000 (.000) -.000 (.000) -.001 (.001) .000 (.000) n = 1081 R2 = .053 OLS coefficients with heteroskedasticity-consistent standard errors in parentheses. Models include state fixed effects. Table 6: Impact of Democratic Vote Share (2000 Presidential Election) on Local Revenue by Revenue Source Type (Per Capita Sales Tax, Share Sales Tax, Per Capita Sales Tax). Prop Dem. Med. Housing Value Med. HHHold Inc. % Poverty % Unemployed % Black % Hispanic % Urban Density per Sq Mile Commute Time % Own Home Per Capita Own Rev. Per Capita Property Tax Per Capita Sales Tax Share Sales Tax(X100) 2.087 (1.12) .156 (.507) .434 (.245) -.067 (.022)** -.017 (.013) -.011 (.005)* .002 (.003) .000 (.000) .028 (.006)** .010 (.003)** .009 (.001)** .000 (.000) .001 (.055) .009 (.003)** -.003 (.001) -.001 (.001) -.056 (.090) -.020 (.033) .051 (.020)* .004 (.002)* .004 (.003) .003 (.002) -.009 (.001) -.000 (.000) -.007 (.004)* -.001 (.001) -.002 (.001)** -.000 (.000) -.002 (.004) -.006 (.002)** -.003 (.001)* -.(000) (.000) -.030 (.006)** -.004 (.002) -.006 (.002)** .001 (.000)** -.735 (.280)** -.056 (.087) -.238 (.061)** -.008 (.005) -.012 (.004)** .001 (.002) -.005 (.001)** -.000 (.000)** n = 1081 n = 1081 n = 1081 n = 1081 R2 = .087 R2 = .325 R2 = .664 R2 = .788 **p < .01 *p < .05 OLS coefficients with heteroskedasticity-consistent standard errors in parentheses. Models include state fixed effects.