Information Consumption and Electoral Accountability in Mexico
The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters
Citation Marshall, John Louis. 2016. Information Consumption and Electoral Accountability in Mexico. Doctoral dissertation, Harvard University, Graduate School of Arts & Sciences.
Citable link http://nrs.harvard.edu/urn-3:HUL.InstRepos:33493495
Terms of Use This article was downloaded from Harvard University’s DASH repository, and is made available under the terms and conditions applicable to Other Posted Material, as set forth at http:// nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of- use#LAA Information Consumption and Electoral Accountability in Mexico
A dissertation presented
by
John Louis Marshall
to
The Department of Government
in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the subject of Political Science
Harvard University Cambridge, Massachusetts
May 2016 c 2016 — John Louis Marshall
All rights reserved. Dissertation Advisor: Professor Torben Iversen John Louis Marshall
Information Consumption and Electoral Accountability in Mexico
Abstract
Electoral accountability rests on voters re-electing high-performing and removing low- performing incumbents. However, voters in many developing contexts are poorly informed about incumbent performance, particularly of local politicians. This dissertation asks: how do voters in low-information environments hold local governments to account for their performance in office? I seek to explain when Mexican voters obtain performance information pertaining to their municipal incumbents, and ultimately how it impacts their beliefs and voting behav- ior. I argue that voters are able and willing to sanction local governments upon receiving incumbent performance indicators. However, electoral accountability requires incentives for voters and media outlets to respectively acquire and supply politically-relevant news. Information in the news just before elections, when these incentives align, thus strongly influences electoral accountability. I test these propositions by examining in detail voter responses to two key issues in Mexican politics—malfeasance in office and violent crime. The first chapter, coauthored with Eric Arias, Horacio Larreguy and Pablo Querubín, uses a large-scale field experiment to establish that voters indeed update from and act on malfeasance revelations. Reflecting voters’ negative priors, the distribution of leaflets doc- umenting mayoral malfeasance increases the incumbent party’s vote share on average. However, consistent with Bayesian learning, these rewards decrease with positive prior iii beliefs, the strength of such priors, the severity of malfeasance revelations, and the extent of negative updating. Moreover, surprising information mobilizes turnout, while relatively unsurprising information reduces turnout. The second chapter then explores when and why voters choose to become informed. I argue that voters strategically acquire costly political information to cultivate a reputation among their peers as politically sophisticated. Leveraging a field experiment and observa- tional variation, I demonstrate that social incentives increase political knowledge among voters nested in groups that collectively value political knowledge. This effect is most pronounced among relatively unsophisticated voters seeking to reach a minimum standard within their group, but is also evident among more sophisticated voters seeking to differen- tiate themselves from less-informed peers. The third chapter, coauthored with Horacio Larreguy and James Snyder, shows how broadcast media regulates access to relevant incumbent performance indicators. Supporting our argument that a media station’s potential audience shapes their incentives to provide local news, voters only sanction malfeasant parties in precincts covered by media stations whose principal audience resides in the precinct’s municipality. Conversely, media outlets based outside the municipality do not aid, and in fact crowd out, electoral accountability. The final chapter combines these insights to explain how voters hold incumbents to account for homicides in their municipality. I argue that even short-term performance indi- cators in the news prior to elections shape the voting behavior of poorly informed citizens. I show that voters consume most news before elections and update about incumbent per- formance from pre-election homicide shocks reported at that time. Unlike longer-term homicide trends, pre-election homicides substantially reduce the incumbent party’s proba- bility of re-election. Sanctioning again requires, and increases with, access to local media, and is concentrated where voter priors are weakest.
iv | Contents
1 Introduction1 1.1 Dissertation overview...... 4 1.2 Implications and future research...... 9
2 Priors rule: When do malfeasance revelations help and hurt incumbent par- ties? 15 2.1 Introduction...... 15 2.2 Malfeasance, audits and elections in Mexican municipalities...... 22 2.3 Prior beliefs and voting behavior...... 26 2.4 Experimental design...... 35 2.5 Aggregate election results...... 52 2.6 How information influences voting behavior...... 62 2.7 Conclusion...... 75
3 Signaling sophistication: How social expectations can increase political infor- mation acquisition 77 3.1 Introduction...... 77 3.2 Voter information acquisition within social groups...... 83 3.3 Experimental evidence...... 91 3.4 Observational evidence...... 105 3.5 Conclusion...... 115
4 Publicizing malfeasance: When media facilitates electoral accountability in Mexico 118 4.1 Introduction...... 118 4.2 Why the intensity of media coverage matters...... 125 4.3 Political accountability in Mexico...... 128 4.4 Data...... 135 4.5 Empirical strategy...... 143 4.6 Results...... 154 4.7 Conclusion...... 172
v 5 Political information cycles: When do voters sanction incumbent parties for high homicide rates? 174 5.1 Introduction...... 174 5.2 Theoretical argument...... 180 5.3 Violent crime and political accountability in Mexico...... 187 5.4 Local elections, news consumption, and political perceptions...... 193 5.5 Local violence and electoral accountability...... 204 5.6 The moderating role of local media...... 224 5.7 Conclusion...... 233
A Appendix to Chapter 2 236
B Appendix to Chapter 3 254 B.1 Proofs...... 254 B.2 Variable definitions...... 257 B.3 Additional experimental results...... 262 B.4 Additional observational results...... 265
C Appendix to Chapter 4 270 C.1 Variable definitions...... 270 C.2 Audit reports...... 274 C.3 Additional local media stations and news consumption...... 274 C.4 Lack of balance across media stations...... 278 C.5 Additional results...... 278
D Appendix to Chapter 5 283 D.1 Formal model illustrating the voter updating process...... 283 D.2 Months and years of municipal elections...... 289 D.3 Data description...... 289 D.4 Map of municipalities included in different samples...... 296 D.5 Additional analyses...... 296
Bibliography 318
vi | Acknowledgments
This manuscript could not have been completed without the support and advice of my committee. I cannot imagine a better chair than Torben Iversen. His theoretical engage- ment, attention to the big picture and remarkable clarity of thought have inspired and disci- plined my work. While I have enjoyed our long meetings, the greatest testament to Torben is the energized feeling I’ve always felt when leaving his office. I have learned tremen- dously from Jim Snyder’s incisive and thoughtful advice and passion for academic study. I can only hope to be as respected as Jim on both an intellectual and personal level. Jim Alt has taught me to be an academic, created research opportunities and supported me through- out my graduate studies. It has been an honor to be one of his last students. Most of all, I am indebted to Horacio Larreguy. His rigor and amazing desire to uncover truth have pushed me one step further at every turn. Horacio has influenced the location, scope and methods of my research and his commitment to my cause has exceeded any reasonable ex- pectation. It is hard to imagine what this dissertation would look like without him. I hope that I can repay my committee’s investments in me by producing research that will satisfy the lofty standards that they have instilled. I also owe a great deal to the outstanding scholarly community that has supported me over the past six years. My research has been enriched by many conversations with and insights from Charlotte Cavaille, Adi Dasgupta, Julie Faller, Mauricio Fernández Duque, Nilesh Fernando, Noam Gidron, Michael Gill, Shelby Grossman, Andy Hall, Alex Hertel-
vii Fernandez, Akos Lada, Chris Lucas, Rakeen Mabud, Noah Nathan, Max Palmer, Jonathan Phillips, Solé Artiz Prilliman, Rob Schub, Brandon Stewart, Chiara Superti and George Yin. I am proud to count you all as colleagues and friends. I am also grateful for the advice and critiques of Jorge Domínguez, James Robinson and especially Jeff Frieden. Many people that have made my life as a graduate student enjoyable. My office mates Mauricio Fernández-Duque, Leslie Finger, Jen Pan, Molly Roberts and Chiara Superti have been frequent sources of cheer, although I especially miss debating statistics and life with Brandon Stewart early in the morning. As friends and housemates, Nilesh Fernando and Nils Hagerdal have provided well-placed skepticism and quality food. Greg Conti and James Brandt have provided beers, basketball and political theory. Unfortunately, Greg remains undefeated in our occasionally epic table tennis matches. When not drilling back- hand winners down the line against me on the tennis court, Rob Schub has sampled with me the gamut of lunch and drinking spots within 500 meters of CGIS. My GSAS football team have provided fun, injuries, occasional trophies and, most of all, the opportunity to continue playing the sport I love. I would have never embarked on a Ph.D. without guidance from key formative figures. At university, Nigel Bowles and Stephen Fisher taught me how think and write like an academic. Turning further back still, Neil Commin’s enthusiasm and encouragement cul- tivated my interest in politics that led me away from mathematics. At school, Mr (Mick) Hickman instilled in me a rigor that I have never forgotten. My largest debt is to my wonderful fiancée, Rakeen Mabud. She has been my greatest supporter, tirelessly reading through each paper I write, removing “the claw” from my presentation style, and (just about) tolerating the fact that I am never quite done. Beyond my research, Rakeen has quite simply made me a better person. She has taught me to be silly, professional and worldly, among many other things. Her positive influences on me are innumerable, and I cannot wait to share my life with her. viii Finally, I want to thank my parents, Barbara and Chris, and my brother Stephen. I could not wish for a more supportive and loving family. They provided me with freedom and opportunities, cultivated my inquisitive and independent nature, and from a young age tolerated me incessantly asking “why?” about anything from why we were playing football in the park to why people believe in God. Without them, none of this would exist.
ix 1| Introduction
Electoral accountability is the idea that voters hold governments accountable for their performance in office, re-electing high-performing incumbent parties and removing low- performing incumbent parties (e.g. Fearon 1999; Ferejohn 1986). This is widely regarded as a cornerstone of representative democracy. As the means through which incumbents can be appraised, voter information—about both performance outcomes and who is responsible for them—thus plays an essential role in this process. Idealized models of voting behavior typically rest on the assumption that voters cast ballots based on well-formed expectations of what different parties offer (Manin, Prze- worski and Stokes 1999). However, in practice, voters across the world—and particularly in developing democracies—are often poorly informed about incumbent performance in office, especially of local politicians (Keefer 2007; Pande 2011). Beyond a lack of repre- sentation in the abstract, this lack of voter information can have important consequences. By failing to remove incompetent or corruption politicians (see Pande 2011) or incentiviz- ing incumbents to target public funds towards engaged voters (e.g. Besley and Burgess 2002; Casey 2015), a misinformed electorate may have important efficiency and distribu- tive implications for political, policy and socioeconomic outcomes. These concerns motivate the overarching question of this dissertation: how do voters in low-information environments hold local governments to account for their performance in office? I focus on incumbent performance information pertaining to local incumbent par-
1 ties, and address this question at the heart of democratic theory by first asking whether such information indeed influences voter beliefs, and in turn how such beliefs affect choices in the voting booth. Delving deeper, I tackle from both a voter demand and media supply perspective the more foundational question of when, if ever, voters actually become suffi- ciently informed to effectively hold their local government to account. In this dissertation, I thus seek to provide answers to various questions: How do voters become informed? Is an active media necessary for electoral accountability? What types of information affect which types of voters, and by how much? Are news stories before elections more likely to affect voting behavior? I examine these questions through the lens of Mexican municipal politics. Following hegemonic rule by the Partido Revolucionario Institucional from 1929 until the 1990s, Mexico has taken important steps toward democratic consolidation. Incumbent turnover, major institutional reforms, increasing media independence and growing multi-party com- petition are manifestations of an increasingly well-functioning democracy. However, the country still faces major challenges including poverty reduction, corruption and drug vio- lence. These challenges are perhaps most pronounced among Mexico’s c.2,500 municipal- ities, which were empowered by decentralizing reforms in the late 1980s and especially in the mid 1990s (see Wellenstein, Núñez and Andrés 2006). Municipal mayors now preside over 20% of total government spending, and are respon- sible for providing basic public services, infrastructural investments and (in the majority of municipalities) local police forces. At the same time, mayors—who will only become eli- gible for re-election for the first time in 2018—face relatively few checks on their actions in office. Since mayors could not be re-elected, a notable feature of this study is the focus on party-level accountability. Mexican political parties remain the focal point of Mexican politics, and while voters are barely aware of individual candidates they generally know what the main parties represent and which is in power locally and nationally (Chong et al. 2 2015; Larreguy, Marshall and Snyder 2016). To support local social and economic development, voters must avoid electing may- ors that are corrupt, unwilling or unable to implement effective policies, or reliant upon providing expensive clientelistic benefits to retain office. However, many voters remain poorly informed about politics (e.g. Chong et al. 2015), media outlets vary in their engage- ment with and independence from politics (Lawson 2002), and local political competition is often limited (Larreguy, Marshall and Snyder 2016). The task of selecting competent politicians is thus far from trivial. Considering municipal incumbent party performance in terms of addressing three salient valence issues—corruption, poverty reduction and pub- lic security—this dissertation specifically examines how information consumption affects local electoral accountability in Mexico’s low-information political environment. I argue that voters are both able and willing to sanction local governments upon re- ceiving incumbent performance indicators. More specifically, I cast voters as Bayesians updating their beliefs based on how politically-relevant information relates to their beliefs before new information was received. However, the extent to which voters ultimately hold municipal governments to account reflects the incentives for voters to acquire, and media outlets to report, such politically-relevant news. As a consequence, electoral accountability follows what I term political information cycles: information in the news just before elec- tions, when voter acquisition and media reporting incentives align, powerfully influencing electoral accountability. Over the course of the four papers in this dissertation, I develop these theoretical ar- guments and map them to appropriate empirical tests that cement the interpretation of the findings and clarify the mechanisms. Blending a range of empirical techniques, from field experiments and quasi-experimental designs to insights from the field, I aim to illuminate the incentives underpinning the supply and consumption of political news concerning these core issues, how voters update from this information, and ultimately how it impacts vot- 3 ing behavior. Although my analysis exploits subnational variation within Mexico, I do not believe my findings are specific to Mexico; throughout, I highlight parallels with both developed and developing contexts across the world.
1.1 Dissertation overview
In “Priors rule,” which is coauthored with Eric Arias, Horacio Larreguy and Pablo Querubín, my dissertation first seeks to establish how voters update from and act on infor- mation documenting mayoral malfeasance in office. While these questions have rightfully received considerable attention, existing research has struggled to impose a clear theoreti- cal framework on the mixed empirical findings. For example, it remains unclear why media revealing mayoral malfeasance reduces the likelihood an incumbent party is re-elected in some instances (e.g. Ferraz and Finan 2008), but has no impact in others (e.g. Banerjee et al. 2014; de Figueiredo, Hidalgo and Kasahara 2013). Similarly, while malfeasance rev- elations appear to induce systemic disengagement in some contexts (Chong et al. 2015), they instead enhance participation in another (Banerjee et al. 2011). We seek to rationalize these mixed findings in a simple Bayesian model emphasizing voter prior beliefs and incorporating a cost of turning out. Theoretically, we show that electoral punishment may be rare among voters already pessimistic about the incumbent party’s competence, even when relatively substantial malfeasance is revealed. The impli- cations for turnout reflect a more subtle non-linearity when voter partisan attachments are bimodally distributed. While relatively unsurprising information can reduce turnout by in- ducing a large mass of voters to abstain because their preference between the parties no longer exceeds the costs of turning out, sufficiently surprising revelations—whether posi- tive or negative about the incumbent—can increase turnout by inducing supporters around one mode to switch parties.
4 These predictions are tested empirically using a large-scale field experiment in the cen- tral states of Guanjuato, México, Querétaro and San Luis Potosí around the 2015 munic- ipal elections. Specifically, we examined how voters responded to learning the outcome of independent audits assessing the extent to which municipal governments correctly spent federal transfers from a major government program designated solely for social infrastruc- ture projects benefiting the poor. In the average municipality between 2007 and 2015, 8% of funds stipulated for social infrastructure projects benefiting the poor were not spent on projects that actually benefited the poor, while a further 6% were spent on unauthorized projects. Reflecting voters’ bleak prior beliefs, relatively high malfeasance revelations increased the incumbent party’s vote share on average. Consistent with our voter learning model, this increase in incumbent party support decreases with positive prior beliefs, the strength of priors, the severity of malfeasance revelations, and the extent of negative updating. Po- tentially squaring the diverging extant findings regarding information’s impact on turnout, we find clear evidence that surprising information mobilizes turnout while relatively unsur- prising information cultivates reduces turnout. In contrast, we find no evidence to suggest that voters become disillusioned with democratic politics. The preceding experimental evidence illustrates the potential importance of informa- tion for understanding voter beliefs and behavior. However, to understand whether voters respond similarly when information is not directly provided to them, a key question regards whether voters will consume such information without external intervention. This question is important in the face of Downs’s (1957) powerful and pessimistic free-riding logic that individual voters have strong incentives to leave information acquisition to others and thus remain “rationally ignorant.” Since Mexico, like many developing democracies, in prac- tice retains relatively weak institutional protections against misuse of office, an informed electorate is central in sustaining accountable government. 5 In my second paper, “Signaling sophistication,” I address a key dimension explaining when and why voters choose to become politically informed: social approval. To endo- genize political information, I propose a social signaling model of information acquisi- tion where voters strategically acquire costly political information to cultivate a reputation among their peers as politically sophisticated. I thus resolve Downs’s (1957) collective ac- tion problem by arguing that, rather than acquiring information to maximize the probability that the best candidate is elected, individuals consume political information for selfish rea- sons emanating from their desire for approval among their peers. The model highlights two motivations that drive voters with varying levels of political sophistication within a group to acquire political information: (1) the desire to meet a minimum standard that separates less sophisticated voters from the least sophisticated, and (2) a social differentiation mo- tive whereby increases in information acquisition among less sophisticated voters forces causes more sophisticated voters to acquire more information to continue to differentiate themselves. I test this theory using both a small-scale field experiment around the 2015 elections and upcoming local elections to generate exogenous variation in social stimuli. These tests vary the likelihood that an individual’s peers are able observe an individual’s political knowledge. The results show that such social incentives significantly increase political knowledge among voters nested in groups that collectively value political knowledge, and likely assign large reputational benefits to politically sophisticated members. This increase is particularly prevalent among relatively unsophisticated voters within a group seeking to reach a minimum standard, but is also evident among more sophisticated voters acquiring news to differentiate themselves from less-informed peers. Although this social signaling logic likely coexists alongside a mass of voters that likely enjoy consuming political news, these findings provide the first evidence of which I am aware that political information acquisition is caused by social pressure. 6 From the perspective of supporting electoral accountability, these findings demonstrate the importance of social networks in supporting informed political participation. Con- versely, they also highlight how large groups of voters can get stuck in low-information equilibria, or “information traps,” where they have little incentive to consume news because their peers attach little value to political discussion. A growing body of global research, including from Africa (e.g. Casey 2015), Europe (e.g. Adams and Ezrow 2009) and the United States (e.g. Bartels 2008; Snyder and Strömberg 2010), indicates that such informa- tion differentials can have important policy and distributional consequences as politicians distribute resources toward informed voters. The analysis thus far ignores the reporting of news by broadcast and print media, the primary sources of news for Mexican voters. Given that many voters face weak incen- tives to actively search out political information for themselves, the potential importance of media outlets making relevant and (at least somewhat) impartial information accessible to voters is substantial. However, media outlets did not necessarily maximize the public’s capacity to hold elected politicians to account. Rather, newspapers, radio and television stations compete for audience shares and advertising revenues (e.g. Gentzkow and Shapiro 2006; Mullainathan and Shleifer 2005). In the third paper, “Revealing malfeasance,” which is coauthored with Horacio Lar- reguy and James Snyder, I investigate the impact of broadcast media on the electoral sanc- tioning of the municipal malfeasance examined in the first paper. We argue that media out- lets only supply their audience with incumbent performance information when they face economic incentives to do so, namely a large interested audience. We exploit two sources of plausibly exogenous variation to isolate the effect of broadcast media outlets where audit reports are published: the timing of the release of municipal audit reports with respect to around elections, and within-neighboring precinct geographic variation in media coverage. Our findings emphasize the importance of media market structure in supporting elec- 7 toral accountability. In particular, we identify that voters electorally sanction the party of malfeasant mayors, but only in precincts covered by local media stations emitting from within the precinct’s municipality. Each additional local radio or television station reduces the vote share of an incumbent party revealed to be either corrupt or neglectful of the poor by around one percentage point. Demonstrating the importance of audience-based eco- nomic incentives, the effect of a local media station increases with the share of an outlet’s market located inside the municipality. Moreover, we find no evidence that non-local media stations contribute to the electoral sanctioning of malfeasant mayors. Rather, non-local me- dia crowds out electoral sanctioning by attracting audiences to outlets less likely to report local mayoral malfeasance. The final paper, “Political information cycles,” brings together the insights of the pre- ceding chapters to demonstrate how the forces underpinning news consumption patterns and information processing influence the sanctioning of local homicides. In the context of unprecedented violent crime, alongside the economy, public security has represented the most important policy issue among Mexican voters of the last decade. Consequently, the incumbent party’s effectiveness at addressing homicides rates is a key performance metric. Building on the propensity of voters to acquire, and media outlets to report, politically-relevant news before elections, I argue that short-term performance indicators in the news prior to elections may powerfully shape the voting behavior of citizens. My simple Bayesian learning model argues that even signals only weakly linked to longer-term performance can heavily influence the beliefs of poorly informed voters whose weak prior beliefs reflect high levels of political disengagement outside of election campaigns. Drawing on fine-grained survey and electoral data and multiple identification strategies, I document empirical support for the key elements of this political information cycles the- sis. First, I show that voters indeed consume more news before local elections, and that homicides just before such elections increase the salience of public security concerns and 8 reduce voter confidence in the mayor. Second, electoral returns confirm that pre-election homicide shocks reduce the incumbent party’s re-election probability by around ten per- centage points. In contrast, incumbent electoral performance is barely impacted by longer- term homicide rates. Consistent with voter learning, sanctioning is limited to mayor con- trolling police forces, is greater where opposition parties did not experience such a shock prior to the previous election, and is not evident among state and federal incumbents. Third, the punishment of homicide shocks requires, and increases with, access to local broadcast media stations. These effects are pronounced only among less-informed voters that princi- pally engage with politics around elections. Together, the results illustrate the importance of when voters consume news, and may thus explain the electoral volatility and mixed electoral accountability often observed outside consolidated democracies and in federal systems.
1.2 Implications and future research
This dissertation thus examines, piece by piece, how voters obtain and use informa- tion to hold their local governments to account in a major non-consolidated democracy. Combining these pieces unearths the finding that electoral accountability reflects politi- cal information cycles. With respect to corruption, misallocated spending on projects not actually benefiting, and especially local homicides, I show that voters are highly respon- sive to information in the news just before elections. A key conclusion is therefore that, given the political engagement and media environment in Mexico and other similar de- veloping democracies, incumbent performance indicators are only likely to meaningfully impact voting behavior—and ultimately electoral outcomes—when their coverage in the media coincides with voters actually consuming news. My research suggests that this equilibrium simultaneously reflects the incentives for
9 voters to acquire politically-relevant information and for media outlets to report such infor- mation. On the voter side, I find that social approval drives information acquisition. The incentives to meet a minimum standard or differentiate oneself within a social group are especially pronounced around elections when relatively politically unsophisticated voters consume news for the first time. At the same time, whether the media provides politically- relevant information reflects economic incentives. In particular, local outlets whose au- dience principally resides within the municipality are most likely to provide voters with access to incumbent performance information. While the extant literature has highlighted the importance of access to news (e.g. Banerjee et al. 2011; Casey 2015; Ferraz and Finan 2008), this dissertation illuminates the mechanisms underpinning the production of news and, most importantly, shows that only when access is paired with voter consumption will political news impact vote choice. This dynamic provides a novel explanation for the mixed evidence of electoral sanctioning observed both in Mexico (Larreguy, Marshall and Snyder 2015; Vivanco et al. 2015) and among poorly-informed electorates more broadly (Achen and Bartels 2004b; Brollo 2009; Chang, Golden and Hill 2010; Roberts and Wibbels 1999). Another central conclusion is that Mexican voters generally behave as Bayesians upon receiving political information. With respect to both malfeasance in office and local homi- cides, voters update their beliefs from new information. Perhaps unsurprisingly, given the importance of these political issues, such updating ultimately influences actual vote choices: I consistently find that what voters regard as poor (good) performance induces electoral punishment (rewards). Beyond influence the direction of voter updating, prior beliefs also play a crucial role in conditioning the extent to which voters respond. More surprising is evidence of subtle updating, such as taking into account performance and the partisanship of the incumbent at the previous election, or recognizing that local politicians not controlling local police forces may not be responsible for local violence. There are still limits though: I find that voters do not utilize comparisons with neighboring municipali- 10 ties and may over-estimate the precision of recent news reports. Nevertheless, my analysis highlights how voters in a major developing context are not lacking in their capacity to hold politicians to account. The dissertation similarly cast light on Mexican municipal politics. A simple but im- portant conclusion is that local politics matters to voters: even in a complex federal system, voters perceive local politicians as important actors and sanction them accordingly. My specific findings reinforce the salience of local government spending and violence—two key issues across Latin America. While the importance of rising crime may come as no surprise to scholars of Mexico, my emphasis on homicide spikes before elections through the prism of political information cycles make sense of the lack of correlation between general homicide trends and electoral outcomes. The findings with regard to incumbent malfeasance in office and local homicides similarly add nuance by showing that account- ability requires local media, whose information appears to primarily influence voter beliefs rather than coordinate collective responses by generating common knowledge. These findings suggest mixed normative implications for democracy and electoral ac- countability. On one hand, voters demonstrate consistent willingness to vote on the basis of incumbent performance metrics, and exhibit impressive capacity to process the infor- mation they encounter. There is thus little doubt that voters armed with clear information have the capacity to hold local governments to account on key political issues. On the other hand, only a limited portion of the electorate obtains politically-relevant information. Unfortunately, this likely generates three equilibrium pathologies. First, limited political engagement may sustain the low voter expectations that mean that levels of malfeasance that are far from trivial may even be rewarded. Without holding politicians to higher stan- dards, career-oriented politicians face weak incentives to perform in office while parties face little pressure to root out politicians that could represent electoral liabilities. Second, because low levels of political information foster weak prior beliefs, voters can be strongly 11 influenced by new revelations. Particularly in the case of local homicides, voters may thus “over-weight” relatively uninformative signals at the expense of informative signals they did not consume. This possibility chimes with recent theoretical research arguing that an informed electorate necessarily enhances electoral accountability (Ashworth and Bueno de Mesquita 2014a). Third, certain types of voter are systematically poorly informed, and thus politicians are likely to strategically allocate resources toward informed voters. This potential skew may undermine democratic representation. These normative concerns challenge policy makers seeking to strengthen democracy. One clear implication is that NGOs and government seeking to enhance electoral account- ability must recognize political information cycles to enhance their information dissemina- tion campaigns. Here, it is important to provide longer-term information beyond proximate security or economic shocks that accurately reflect incumbent performance. However, ad- dressing the deeper issue of low expectations is more difficult. To achieve this, efforts to alter voter prior beliefs through civic education campaigns—whether in school, through the media or elsewhere—may be helpful. From an alternative perspective, politicians must believe that competence is an election-winning campaign strategy. The introduction of re-election in 2018 could aid this process by individualizing campaigns, while the indepen- dent candidates first permitted to run in 2015 similarly present an opportunity to alter the political agenda. Furthermore, there is clear evidence for this in Italy (Kendall, Nannicini and Trebbi 2015), although engaging Mexican politicians is likely to require compelling local evidence. My dissertation also raises interesting questions that will drive my future research. In particular, more work is required to understand equilibrium dynamics along several di- mensions. First, while social groups play a fundamental role in explaining voter political information levels, a key question remains: what induces voters to enter high information social networks? This dissertation has focused on illustrating how social groups create in- 12 formation acquisition incentives, but deeper questions concern the drivers of social group formation and how politics becomes a focal discussion topic in some networks rather than others. Second, there is also much to be discovered about the production of politically-relevant news. This dissertation consistently shows that media coverage is a necessary condition for electoral accountability, but a systematic account of how much and what types of political news are reported is still missing. My ongoing research aims to address these questions by accumulating an unprecedented archive of local print and broadcast news coverage. Experi- mental tests will identify the relative importance of prohibitive search costs, news coverage cascades within media markets and upcoming local elections in explaining the extent of political news production. Similarly, I intend to explore how the timing of malfeasance revelations differentially influences electoral outcomes. Nevertheless, profit or audience maximization (e.g. Gentzkow and Shapiro 2006; Mullainathan and Shleifer 2005) may only tell part of the story. A key and often neglected component is the interaction between politicians and journalists. It remains essential to understand the empirical relevance of me- dia capture (e.g. Besley and Prat 2006) in preventing a shift toward a high-accountability equilibrium. Third, an outstanding question is what about broadcast media produces such large ef- fects on voter behavior. Media is distinctive from leaflets in important respects, including providing context, a tone of presentation, informational credibility, and the expectation that others will also have been similarly informed. Separating the sources of media’s effects is an important challenge for our understanding and for policy makers seeking to maximize the reach of the information provided. Finally, little is know about the extent to which one-shot and sustained information provision influence political equilibria. How do politicians respond to and learn from the sanctioning of high rates of crime or corruption revelations? Will voters remember infor- 13 mation from previous elections and update it accordingly before the next election? Can learning of reports documenting malfeasance encourage media outlets to continue report- ing on such issues? These questions have profound consequences for social welfare, but remain notable omissions from this dissertation. Moving forward, I intend to examine how policy decisions, voter consumption choices and media outlet reporting choices are shaped by one-shot and sustained provision of politically-relevant information. Given informa- tion’s significant influence on electoral behavior, illustrated throughout this dissertation, the optimist in me hopes that it can also help to drive improving government provision and increase voter expectations of their politicians.
14 2| Priors rule: When do malfeasance revela- tions help and hurt incumbent parties?1
Co-authored with Eric Arias, Horacio A. Larreguy and Pablo Querubín
2.1 Introduction
Elected politicians across the world put in place policies to alleviate poverty and sup- port economic development. However, the implementation of these policies is often beset by political rent seeking, including bribery (e.g. Hsieh and Moretti 2006), preferential con- tracting (e.g. Tran 2009) and misallocated spending (e.g. Larreguy, Marshall and Snyder 2015). While policy-makers and NGOs have increasingly sought to design institutions to mitigate such agency losses, political accountability ultimately rests upon citizens choos- ing to elect competent politicians and sanction incompetent politicians (e.g. Barro 1973; Fearon 1999; Ferejohn 1986; Rogoff 1990). Given that malfeasance in office still repre- sents a major challenge in many developing contexts (e.g. Kaufmann, Kraay and Mastruzzi 2009; Mauro 1995), a key question is then: when do voters hold their governments to ac-
1We are extremely grateful to Alejandra Rogel, Adriana Paz, Anais Anderson and the Data OPM and Qué Funciona para el Desarrollo team for their implementation of this project. We thank Chappell Lawson, Ken Shepsle, participants at the WESSI workshop at NYU Florence and the organizers and other team members of the EGAP Metaketa project for illuminating discussions and useful comments. This research was financed by the Evidence in Governance and Politics (EGAP) Metaketa initiative, and was approved by the Harvard Committee on the Use of Human Subjects (IRB15-1068) and the New York University Committee on Activ- ities Involving Human Subjects (IRB15-10587). Our pre-analysis plan was pre-registered with EGAP, and is publicly available here.
15 count by punishing incumbent parties for malfeasant behavior in office? A growing political economy literature has emphasized the importance of providing voters with incumbent performance information. Negative information, such as reports revealing corruption, is typically expected to cause voters to electorally sanction those re- sponsible. However, in practice, the evidence is mixed. On one hand, Ferraz and Finan (2008), Larreguy, Marshall and Snyder(2015) and Chang, Golden and Hill(2010) find that local media revealing mayoral malfeasance reduces their likelihood of re-election in Brazil, Mexico and Italy respectively. On the other, Banerjee et al.(2011, 2014), de Figueiredo, Hidalgo and Kasahara(2013) and Humphreys and Weinstein(2012) find that disseminat- ing negative incumbent performance information in India, Brazil and Uganda often does not affect incumbent electoral prospects. The effects on turnout are similarly mixed: while Chong et al.(2015) suggest that negative information may even induce systemic disen- gagement in Mexico, Banerjee et al.(2011) instead observe increased turnout in India. From both a theoretical and applied policy perspective, it thus remains difficult to antici- pate when or how providing information about incumbent performance will impact voters (see Lieberman, Posner and Tsai 2014). We argue that previously overlooked voter prior beliefs can rationalize these mixed findings, and ultimately explain when and how incumbent performance information im- pacts turnout and vote choice. We highlight the importance of the direction and magnitude of belief updating from new information in a simple two-party model where voters form be- liefs about the competence of the incumbent party, receive expressive benefits from voting and are subject to fixed partisan attachments (see also Kendall, Nannicini and Trebbi 2015). Specifically, if voters already believe their incumbent party is malfeasant, even relatively severe corruption revelations can increase incumbent support because voters positively up- date from information that is better than expected. The implications for turnout reflect a more subtle non-linearity. Under empirically-appropriate bimodal distributions of partisan 16 attachments, relatively unsurprising information reduces turnout by inducing a large mass of voters located around one mode to abstain because their relative preference between the parties no longer exceeds the costs of turning out. However, sufficiently surprising revelations—in either direction—increase turnout by inducing a large mass of supporters around one mode to switch parties. We test these theoretical predictions—which are codified in our pre-analysis plan— using a large-scale field experiment conducted in Mexico around the 2015 elections. Be- yond its large population and shift toward multi-party democracy, Mexico’s relatively high but substantially varying levels of corruption and distrust in elected politicians mark it as an important location to test the implications of our argument. Extending two recent em- pirical studies with markedly different findings (Chong et al. 2015; Larreguy, Marshall and Snyder 2015), we examine how voters respond to learning the outcome of independent au- dits assessing the extent to which municipal governments correctly spent federal transfers from a major government program designated solely for social infrastructure projects ben- efiting the poor. Although parties are strong, Mexico represents an interesting case because mayors could not seek re-election.2 Across 678 electoral precincts in 26 municipalities from 4 central Mexican states, we randomized the dissemination of leaflets indicating the results of the municipal audit re- ports to up to 200 households in each treated precinct. We provided voters with one of two measures of incumbent performance: the share of funds spent on projects not benefiting the poor, or the share of funds spent on unauthorized projects. Such malfeasance ranged from 0% to 58% in our sample, with a mean of 21%. To improve our understanding of when information impacts voter beliefs and behavior, we also experimentally varied (1) whether the leaflet included a comparison with the performance of mayors from other parties within
2Re-election will become possible for incumbent mayors from 2018 in most states.
17 the state, and (2) whether leaflet distribution was accompanied by a loud speaker announc- ing the widespread delivery of the leaflets in the precinct. The former variant seeks to isolate the role of relative performance information, while the public mode of transmission seeks to generate tacit or explicit voter coordination through common understanding that others also received the information (Adena et al. 2015; Yanagizawa-Drott 2014). Consistent with the theory, we find that voter responses depend on how the information relates to their prior beliefs. On average, audit report information increased the incumbent party’s vote share by almost 3 percentage points. This positive effect reflects the common perception that the incumbent party is highly corrupt and relatively unwilling to support the poor. Somewhat depressingly, this suggests that voter expectations are so low that poor performance is often rewarded. However, these average effects mask substantial heterogeneity in the response of a Mexican electorate already highly skeptical that local politicians allocate funds as legally required. Illustrating the importance of understanding voter prior beliefs, we demonstrate that the increase in incumbent support induced by our treatment is concentrated among voters in municipalities with lower levels of malfeasance, voters with relatively negative prior perceptions of the incumbent party, and voters that positively updated about their per- ceptions of the incumbent party’s malfeasance upon receiving the information. Conversely, for egregious cases of malfeasance, and where voters update most negatively about their incumbent’s performance, voters are more likely to punish the incumbent party at the polls. Moreover, we find support for the non-linear prediction of malfeasance information’s impact on electoral turnout. In particular, relatively unsurprising information—20-30% of funds spent on projects not benefiting the poor or unauthorized projects—depresses turnout by around 1 percentage point. Conversely, extreme cases of malfeasance—both 0% and above 50%—mobilize turnout by 1-2 percentage points. This non-linearity, which is in line with a bimodal distribution of voters updating from new information, further underscores 18 the key role of voter priors in explaining voting behavior. In contrast, we find little evidence to suggest that revealing more severe cases of malfeasance to voters reduces turnout or confidence in the capacity of elections to select competent politicians. Turning to our variants of the information treatment, there is little evidence that either comparative performance information or low-intensity public information dissemination moderate voter responses. First, if anything, providing voters with a benchmark from other parties produced weaker effects. This is consistent with previous studies in Mexico fail- ing to observe benchmarking across municipalities (Marshall 2016a), and likely reflects voter comprehension constraints and the fact that our information reaffirmed voters’ prior perceptions that challengers are less malfeasant. Second, the loud speaker—intended to in- duce common knowledge among voters about the information, and thereby facilitate coor- dinated action—neither increased social responses nor accentuated voter sanctioning. This null finding contrasts with studies isolating large magnification effects of local broadcast media (Ferraz and Finan 2008; Larreguy, Marshall and Snyder 2015; Marshall 2016a), and suggests that a more powerful public mechanism is required. However, we find evidence suggesting that voter responses are mediated by party reac- tions to revealing incumbent party performance. In addition to our distribution team, voters in treated precincts recalled that both incumbent and challenger local party organizations were more likely to combat or incorporate malfeasance reports in their campaigns, espe- cially where the leaflets informed voters of high levels of malfeasance. We explore how this impacted election strategies by exploiting a discontinuity in the number of polling sta- tions in a precinct, which previous research has found to enable parties to more effectively extract voter mobilization effort from local brokers by increasing the party’s ability to infer broker performance after receiving the signal of an additional electoral return (Larreguy, Marshall and Querubín 2016). The results indicate that the incumbent party is particularly able to counteract bad information about its performance in the precincts that possess an 19 additional polling station. This suggests that the relatively weak punishment of even severe malfeasance revelations in part reflects the incumbent’s capacity to combat negative belief updating with brokered voter mobilization. In sum, the results provide strong support for our simple prior-oriented theory of voter responses to incumbent performance information. In rationalizing the mixed evidence that malfeasance information impacts voting behavior, we make three main contributions. First, we demonstrate that the nature of voter sanctions (rewards) depends on the extent to which voter priors were relatively positive (negative). While relatively negative priors entailed rewards for many incumbents in practice in our study, lower expectations of incumbent malfeasance may explain the apparent greater willingness of voters to sanction in more developed contexts (e.g. Chang, Golden and Hill 2010; Eggers 2014). These results ex- tend Banerjee et al.(2011), who start from a similar theoretical framework, by explic- itly focusing on voters prior and posterior beliefs. Nevertheless, the strongest evidence of sanctioning outside consolidated democracies comes from studies examining the role of broadcast media (Ferraz and Finan 2008; Larreguy, Marshall and Snyder 2015; Marshall 2016a). Although our loud speaker did not accentuate the impact of our treatment, radio and television coverage may entail a considerably more powerful signal that drowns out voter priors and primes voters to vote on the issue. More generally, our emphasis on voter priors complements Kendall, Nannicini and Trebbi(2015), who go to great lengths to elicit prior beliefs and demonstrate their importance for understanding how Italian voters respond to campaign information. Our findings similarly reinforce studies showing that corruption information only affects voter attitudes when delivered by a credible source (e.g. Botero et al. 2015). Second, we reinterpret extant findings suggesting that negative campaigning and malfea- sance revelations can engender disengagement (Ansolabehere and Iyengar 1995; Chong et al. 2015; de Figueiredo, Hidalgo and Kasahara 2013). Our non-linear explanation for the 20 relationship between performance information and turnout instead relies on the distribution of voters and their beliefs. In our context, we demonstrate that information both increases and decreases turnout. This may thus explain the mixed existing findings (e.g. Banerjee et al. 2011; Chong et al. 2015; de Figueiredo, Hidalgo and Kasahara 2013; Humphreys and Weinstein 2012). Nevertheless, our theory cannot explain the simultaneous decline in incumbent and challenger support that Chong et al.(2015) observe. This remains an im- portant question for further study, but could simply reflect an idiosyncratic feature of their data, given that—consistent with our argument—they only observe a decline in incumbent partisanship without a commensurate decline in challenger partisan attachment. Third, that voter behavior may be conditioned by challenger and especially incumbent strategies highlights the importance of integrating political actors into models examining the effects of information provision. Our findings pertaining to incumbent responses com- plement results from the Philippines and Sierra Leone, where respectively Cruz, Keefer and Labonne(2015) observe that distributing information about government projects also increases vote buying and Casey(2015) shows that providing voters with information about their candidates lead politicians to readjust their distributive strategies in favor of a more equitable allocation of resources. Using less nefarious means, Cole, Healy and Werker (2012) also find that Indian voters are less likely to punish incumbents for adverse weather shocks when the incumbent responds effectively to the crisis. The remainder of the paper is structured as follows. Section 2.2 first describes the Mexican municipal context in which we build and test our argument. Section 2.3 presents a simple model highlighting the conditions under which we expect new information to in- crease and decrease a voter’s propensity to turn out and cast a ballot for the incumbent or challenger party. Section 2.4 explains the experimental design, and uses individual-level surveys to validate our interpretation of the treatments. Section 2.5 presents the main find- ings, while section 2.6 examines the channels through which information impacts voting 21 behavior. Section 2.7 concludes.
2.2 Malfeasance, audits and elections in Mexican munici-
palities
Mexico’s federal system is divided into 31 states (and the Federal District of Mexico City) containing around 2,500 municipalities and 67,000 electoral precincts. Following major decentralization reforms in the 1990s (see Wellenstein, Núñez and Andrés 2006), municipal governments—the focus of this paper—play an important role in delivering ba- sic public services and managing local infrastructure, and now account for 20% of total government spending. Municipalities are governed by mayors typically elected to three- year non-renewable terms.3
2.2.1 Federal audits of municipal spending
A key component of a mayor’s budget is the Municipal Fund for Social Infrastructure (FISM), which represents 24% of their total budget. According to the 1997 Fiscal Coordi- nation Law (LCF), FISM funds are direct federal transfers legally designated for infrastruc- ture projects that benefit the population living in extreme poverty. Eligible projects include investments in the water supply, drainage, electrification, health infrastructure, education infrastructure, housing and roads. However, voters remain poorly informed about both the resources available to mayors and their responsibility to provide basic public services (Chong et al. 2015). The use of FISM transfers is subject to independent audits. Responding to high levels of perceived mismanagement of public resources, the Federal Auditor’s Office (ASF) was es-
3Re-election will become possible for incumbents from 2018.
22 tablished to audit the use of federal funds in 1999. Although the ASF reports to Congress, its management autonomy is constitutionally enshrined and it has the power to impose fines, recommend economic sanctions and file or recommend criminal cases against public officials (see Larreguy, Marshall and Snyder 2015). The ASF selects around 150 munici- palities for audit each year, based primarily on the relative contribution of FISM transfers within the municipal budget, historical performance and factors that raise the likelihood of mismanagement, and whether the municipality has recently been audited (including con- current federal audits of other programs) (Auditoría Superior de la Federación 2014). The municipalities to be audited in a given year are announced after spending has occurred. Audits address the spending, accounting and management of FISM funds from the pre- vious fiscal year. Although the ASF’s reports categorize the use of FISM funds in various ways, we focus on two key dimensions of mayoral malfeasance documented in the audit re- ports (that are not necessarily mutually exclusive): the share of funds spent on projects not directly benefiting the poor and the share of funds spent on unauthorized projects. Spending not benefiting the poor entails the allocation of FISM funds to social infrastructure projects that do not principally benefit voters in extreme poverty. Unauthorized spending primarily includes the diversion of resources for non-social infrastructure projects (e.g. personal ex- penses and election campaigns) and funds that are accounted for. Such spending is akin to the corruption identified by Brazilian audit reports (Ferraz and Finan 2008). The results for each audited municipality are reported to Congress in February the year after the audit was conducted, and are made publicly available on the ASF’s website. Mayoral malfeasance is relatively high. Between 2007 and 2015, 8% of audited funds were spent on projects not benefiting the poor, while a further 6% were spent on unautho- rized projects. In one case, the mayor of Oaxaca de Juárez created a fake union to collect payments, presided over public works contracts without offering open tender, diverted ad-
23 vertising and consulting fee payments, and failed to document quantities of spending.4 In another instance, 9 municipal governments in the state of Tabasco—Centro, Balancán, Cár- denas, Centla, Jalapa, Jonuta, Macuspana, Tacotalpa and Tenosique—diverted resources to fund the 2012 electoral campaigns of their parties’ candidates.5 It is thus unsurprising to find that 50% of voters do not believe that municipal governments honestly use public resources (Chong et al. 2015).
2.2.2 Municipal elections
Traditionally, local political competition has been between either the populist Institu- tional Revolutionary Party (PRI) and the right-wing National Action Party (PAN), or the PRI and its left-wing offshoot Party of the Democratic Revolution (PRD). Due to regional bases of political support and highly localized influence within municipalities, local poli- tics is typically dominated by one or two main parties. In the municipal elections we study, the effective number of political parties by vote share at the precinct and municipal levels remain consistently around 2.5. To win office, the three large parties often subsume small parties into municipal-level coalitions.6 In 2015, the National Regeneration Movement (MORENA) stood for the first time, and made headway against this hegemony with 9% of the federal legislative vote. Although economic and criminal punishments for misallocating funds are relatively rare, there are good reasons to believe that voters will hold the incumbent party responsi- ble, even without the re-election of individual mayors. First, voters are considerably better informed about political parties than individual politicians (e.g. Chong et al. 2015; Lar- reguy, Marshall and Snyder 2015). Crucially for political accountability, 80% of voters in
4BBM Noticias, “ASF: desvió Ugartchechea 370.9 mdp,” October 21st 2013, here. 5Tabasco Hoy, “Pagaron pobres campañas 2012,” March 6th 2014, here. 6These smaller parties typically benefit by receiving sufficient votes to maintain their registration. 24 our survey can correctly identify the party of their municipal incumbent. Second, Mex- ico’s main parties retain entrenched and differentiated candidate selection mechanisms (Langston 2003). For example, 74% of voters in our survey believe that if the current mayor is malfeasant another candidate from within the same party is at least somewhat likely to also be malfeasant. Third, Marshall(2016 a) also finds that, at least when local media is available, Mexican voters punish their municipal incumbent parties for elevated pre-election homicide rates. However, existing evidence of electoral sanctions against the incumbent party is mixed. As noted above, Larreguy, Marshall and Snyder(2015) observe large electoral penalties among voters with access to broadcast media outlets incentivized to report local news. Exploiting exogenous variation in the release of audit reports prior to elections and access to radio and television stations, they find that each additional local media station—but especially an additional local television station—decreases the vote share of an incumbent party revealed to be malfeasant just before the election by around 1 percentage point. This evidence supports the standard electoral accountability model (e.g. Barro 1973; Fearon 1999; Ferejohn 1986; Rogoff 1990). Chong et al.(2015) find different results in a field experiment conducted in 12 munic- ipalities across 3 states. Disseminating to voters leaflets containing audit report outcomes, they instead find that while incumbent support declines when the incumbent is revealed as highly malfeasant, challenger support declines at least as much. They speculate that such broad-based disengagement, also observed through reduced partisan attachment to the incumbent, reflects an equilibrium where voters disengage because they believe that all politicians are corrupt.7 The disjuncture between these accountability and disengagement findings, which cover
7In the context of our model below, we could think of this as increasing the cost of turning out or inducing the difference in expected competence between candidate parties to zero.
25 the same information over the same period, illustrate the need for a theoretical logic able to identify when different types of information impact voters differently. With a bimodal distribution of voters, our learning model offers an explanation for voter disengagement among incumbent supports that we test using a field experiment.
2.3 Prior beliefs and voting behavior
We now explore how information about incumbent malfeasance may impact electoral accountability. A key insight of our simple learning model is that the impact of information on voter beliefs and vote choice rests on how the information relates to voters’ prior beliefs. While high levels of malfeasance are clearly bad news, it is not obvious whether voters will reward or punish incumbent parties for low but non-zero levels of malfeasance. Moreover, with a cost to voting and a bimodal distribution of voters, information relatively close to voter priors can reduce turnout, while major departures can cause wholesale shifts in sup- port from incumbent to challenger (or vice versa). The model derives the key comparative static predictions that guide our experimental design.
2.3.1 Theoretical model
We consider a simple decision-theoretic model where a unit mass of voters update about party competence for office from informative signals, and choose between voting for the incumbent I,8 voting for the challenger C and abstaining.9 As noted above, two party competition holds in most parts of Mexico. Voters receive expressive utility from voting for their favored party, rather than believe
8The theory can be easily adjusted to allow for imperfect within-party candidate correlations. 9In the model, we do not consider the possibility that parties react to the provision of information to counteract its effect. Empirically, however, we provide some evidence of this below.
26 that their choice will impact the electoral outcome (e.g. Brennan and Hamlin 1998). This includes a competence and a fixed partisan component, such that the utility to voter i of voting for party p ∈ {I,C} is given by:
h i E −exp − (θ + δ ) if p = I p I i Ui = (2.1) E[−exp(−θC)] if p = C
where θp is the competence of party p and δi ∈ Γ ⊆ R is a positive or negative partisan bias toward the incumbent. For analytical simplicity in incorporating risk aversion, we employ a standard exponential utility function satisfying constant absolute risk aversion utility, where competence and partisanship are perfect substitutes. The partisan shock δi is independently and identically distributed across the electorate according to cumulative distribution function F. This shock, which is uncorrelated with perceptions of candidate competence, could reflect durable partisan attachments, clientelistic transfers or shocks occurring before the election and after voters receive information. Let c > 0 be the cost of turning out to vote. Voters thus turn out if the utility from voting for their preferred party p is sufficiently large relative to their utility from voting for the other party −p, i.e. p −p Ui −Ui > c.
However, voters are uncertain about the competence θp of both the incumbent and challenger parties, and learn from common information pertaining to party competence such as malfeasance revelations in a Bayesian fashion. In particular, we assume that all voters share the same normally distributed prior beliefs over the competence of each party
2 p, distributed according to N(µp,σp ). Focusing on the case where voters only receive an audit report documenting malfeasance that pertains to the incumbent, voters observe a
2 signal sI drawn from a normal distribution of signals N(θI,τI ) centered on the incumbent’s true but unknown competence level θI. For simplicity, we consider the case where the
27 competence of each party p is independently distributed; voters thus assume that this signal is uninformative about the challenger (c.f. Kendall, Nannicini and Trebbi 2015).10 The
2 known variance of this signal, τI , reflects the fact that the audit report may only capture one dimension of an incumbent’s competence or malfeasance. We discuss the extension to
also receiving a signal sC of challenger quality below.
After receiving a signal of incumbent competence sI, voters form their posterior belief over the incumbent’s competence using Bayes’ rule:
2 N µI + κI(sI − µI),κIσI (2.2)
2 τI where κI ≡ 2 2 captures the strength of the signal relative to the prior. Higher values σI +τI of κI indicate that the signal is relatively more precise than voter prior beliefs. Voters thus update positively about expected incumbent competence if s > µ, and particularly when the signal is precise relative to the prior. Because the beliefs about the competence of both
parties are independent, voters do not update about θC. New information also reduces the 2 2 variance of voter posterior beliefs, given that κIσI < τI .
A positive incumbent performance signal (i.e. sI > µI), increases the utility of voting for I by both increasing the incumbent’s expected competence and reducing i’s uncertainty over policy outcomes. This is reflected in the following expected utility received by voter i:
2 κIσI −exp − µI + κI(sI − µI) − 2 + δi if p = I p Ui = (2.3) 2 −exp − µ − τC if p = C C 2
10At the cost of mathematical complexity this could be relaxed, and would yield similar results for a sufficiently small correlation between sI and θC.
28 where the negative term inside the parentheses reflects voter risk-aversion. Integrating over the distribution of voter partisan biases, it is easy to derive the share of voters Vp turning out for each party and the share of abstainers A:
2 2 τ − κIσ V = 1 − F c − ∆ − κ ∆ − C I , (2.4) I µ I I 2 2 2 τ − κIσ V = F −c − ∆ − κ ∆ − C I , (2.5) C µ I I 2 2 2 2 2 τ − κIσ τ − κIσ A = F c − ∆ − κ ∆ − C I − F −c − ∆ − κ ∆ − C I ,(2.6) µ I I 2 µ I I 2
where ∆µ ≡ µI − µC and ∆I ≡ sI − µI are respectively the difference in (the expectation of) voter priors between incumbent and challenger and the extent of updating about the 2 2 τC−κIσI incumbent’s competence from the signal, while 2 represents the difference in risk associated with C relative to I. These vote shares are perhaps most intuitively illustrated graphically in Figure 2.1, which plots the distribution of voters by their relative preference 2 2 I C τC−κIσI ∆U ≡ Ui −Ui = ∆µ + κI∆I + 2 +δi for the incumbent. We can analyze how voting behavior is impacted by the key parameters in our model by simply shifting the distribution along the ∆U axis. As illustrated by the dotted distribution, a relatively weak positive signal of incumbent competence relative to voters’ priors beliefs entails a small increase in ∆I and a reduction in the relative risk of voting for I, and thus a commensurate shift in the distribution of relative voter preferences to the right. This un- equivocally increases the number of voters supporting I and decreases the number of voters supporting C. Similarly, an increase in the difference between incumbent and challenger priors (i.e. greater ∆µ ), or the precision of a relatively positive signal (i.e. greater κI, or 2 lower σI ), also shift the distribution to the right and increase the incumbent party’s vote share. While the incumbent vote share results hold for any distribution F of partisan attach-
29 Vote C Abstain Vote I Density of voters
-c 0 c
Preference toward I (∆U)
Prior Small positive update Large positive update
Figure 2.1: Vote choice and distributions of voters
ments, the implications of providing information for turnout depend upon the shape and position of F and the extent to which information induces updating. Without loss of gener-
ality, consider the case of receiving sI > µI.A positive signal about the incumbent has two effects, again by shifting voter expectations and reducing aversion. First, it induces some voters that would not have otherwise voted to turn out for I; and if the signal is sufficiently powerful, some C may also shift to I. Second,the positive signal also induces some vot- ers (that would have otherwise voted for C) not to turn out. The relative masses of these conflicting effects on turnout determine whether turnout increases or decreases. To produce sharp empirical predictions, we focus on the case where voter partisan at- tachments are bimodally distributed. In addition, we assume that the voters at each mode turn out for different parties. In many electoral contexts, including Mexico, this is a rea- sonable approximation. As noted above, the geographic dispersion of party strength that
30 ensures most races are effectively two-party races. Furthermore, our survey indicates that, of the two-thirds of voters with a partisan attachment, 73% rate the strength of their attach- ment at 5 or higher on a 7-point scale. The non-linear effect of information in turnout is best illustrated graphically using the example in Figure 2.1. The dotted distribution shifted slightly to right demonstrates that a small update in favor of the incumbent causes more initial C voters to abstain than initial abstainers to vote I. This is easy to see by comparing the mass under each distribution over the interval [−c,c]. However, a sufficiently large positive update about the incumbent— which shifts the distribution further to the right—induces initial C supporters to vote I rather than abstain. More generally, it is easy to see that such non-linear predictions hold for any bimodal distribution where the voters at each mode initially turn out for different parties.11
2.3.2 Empirical implications
The model generates various comparative static predictions, some more particular to our model than others. We focus on the impact of providing voters with a signal of in- cumbent performance, sI, via a treatment containing incumbent performance information pertaining to mayoral malfeasance. We now enumerate the key hypotheses that our experi- ment is designed to test empirically; all hypotheses were registered in our pre-analysis plan before the experiment was conducted. We first consider how incumbent malfeasance information affects posterior beliefs re- garding the incumbent party’s competence and vote choice. As equation (2.2) shows, the direction of updating from a given signal sI depends on the prior expectation µI. The effect is thus context-dependent, reflecting both the nature of the information provided and voter
11To see this, consider the derivative of the density function at the turn out cutoffs.
31 prior beliefs regarding incumbent competence. Given the relatively negative incumbent performance information we will typically provide voters with, and despite the reduction in incumbent uncertainty that this provides, we anticipated that on average:
H1. Incumbent malfeasance information increases voter posterior beliefs that the incum- bent party is malfeasant and decreases the incumbent party’s vote share.
However, the most important implications of the model capture how the effect of in- cumbent malfeasance information varies with voter prior beliefs. These are theoretically unambiguous in our model. First, if voters already believe that the incumbent party is malfeasant (i.e. low µI), a signal of high malfeasance has a weaker impact on negative posterior beliefs and the incumbent party vote share:
H2. The effect of incumbent malfeasance information on voter posterior beliefs that the incumbent party is malfeasant is decreasing in voter prior beliefs that incumbent is malfeasant, while the effect of incumbent malfeasance information on incumbent party vote share is increasing in such voter prior beliefs.
Second, voters already possessing strong prior beliefs about the incumbent’s competence
(i.e. low κI) are less responsive to new information:
H3. The effect of incumbent malfeasance information on voter posterior beliefs that the incumbent party is malfeasant and incumbent vote share is weaker among voters with strong prior beliefs.
Third, voters update more positively (negatively) about the incumbent party’s competence upon hearing that the incumbent is relatively clean (malfeasant):
H4. The effect of incumbent malfeasance information on voter posterior beliefs that the incumbent party is malfeasant is increasing in the level of reported malfeasance, 32 while the effect of incumbent malfeasance information on incumbent party vote share is decreasing in the level of reported malfeasance.
Finally, given that the extent of voter belief updating reflects the difference between the signal and voter prior beliefs,
H5. The effect of incumbent malfeasance information on voter posterior beliefs that the incumbent party is malfeasant is increasing in the extent to which the information is worse than voter priors, while the effect of incumbent malfeasance information on incumbent party vote share is decreasing in the extent to which the information is worse than voter priors.
In absolute terms, and given voter prior beliefs, the direction of updating produces different responses:
H6. Good (bad) news about the incumbent party decreases (increases) voter posterior be- liefs that the incumbent party is malfeasant and increases (decreases) the incumbent party’s vote share. Neutral news leaves both outcomes unchanged.
As demonstrated above, the effect of new information on turnout is non-linear when voters are bimodally distributed with voters at each mode initially turning out for different parties, as the evidence suggests is the case in Mexico. In particular, shockingly positive or negative revelations induce voters to switch parties, while relatively unsurprising but nevertheless informative positive (negative) information induces challenger (incumbent) partisans to become relatively indifferent between the parties and instead abstain when faced with a cost of turning out. This logic does not yield clear predictions for the average effect of new information or its linear interaction with the level of malfeasance reported. However, a clear prediction of the theory is that:
33 H7. High and low levels of reported incumbent malfeasance information increase elec- toral turnout, while relatively unsurprising middling levels of reported malfeasance decrease electoral turnout.
2.3.3 Variation in information’s impact
Our model can be extended to incorporate the impact of simultaneously providing chal- lenger information, and thus permitting relative performance comparisons. We anticipate that such information will impact voter posterior beliefs about the competence of the chal- lenger akin to equation (2.2). However, the effect on vote choice depends on the relation- ship between incumbent and challenger signals. Empirically, we will restrict attention to the case where challenger party information is provided alongside either notably worse or notably better incumbent party malfeasance information. Comparative performance infor- mation thus always provides a stronger signal, by compounding a single signal with a clear benchmark for comparing sI to sC. For average malfeasance revelations, in our experimental sample where the majority of voters receive information indicating stronger challenger performance, we expected that:
H8. Comparative malfeasance information on average decreases the incumbent party’s vote share more than just incumbent malfeasance information.
Furthermore, this anti-incumbent party effect was expected to be particularly large when voters learn of especially poor incumbent performance—in both absolute and relative terms:
H9. The effect of comparative malfeasance information on incumbent party vote share decreases more quickly in the level of reported incumbent malfeasance, and the dif- ference relative to the reported malfeasance of the challenger, than just incumbent malfeasance information.
34 Another potentially important factor in explaining when information supports political accountability is the method of information transmission. Public modes of transmission— through which voters become aware that other voters have also received a given piece of information—could produce powerful effects by inducing explicit or tacit voter coordina- tion based on their common knowledge (e.g. Adena et al. 2015; Arias 2015; Yanagizawa- Drott 2014). Explicit discussion of the information may result in voters engaging heavily with the information received, and in turn consolidating their beliefs around such infor- mation. This could even result in agreement to harmonize vote choices. Alternatively, tacit coordination only relies on the delivery of information engendering the (higher-order) belief that others also received the information, and will likely act on such information. Such coordination may in part explain the large effects of local media found in Mexico (Larreguy, Marshall and Snyder 2015; Marshall 2016a). Both such cases of coordinated behavior induce shifts that could not be achieved by providing the same information using private modes of information transmission. In the context of our model, where we do not explicitly model voter discussion and coordination, public delivery mechanisms are perhaps best interpreted as a clearer signal
2 of performance (lower σI ). We thus expected the social treatment to magnify any impact of incumbent malfeasance information:
H10. The average and heterogeneous effects of incumbent malfeasance information on the incumbent party’s vote share (H1-H7) are greater when the information is delivered through a public mechanism.
2.4 Experimental design
We designed a field experiment to test our theoretical argument. Specifically, we fo- cus on Mexico’s municipal elections held on Sunday 7th June 2015, and seek to identify 35 the effect of providing voters in 678 electoral precincts with the results of audit reports documenting the municipal use of federal transfers designated for infrastructure projects benefiting the poor. Municipal elections were held concurrently with state and federal leg- islative elections. We first explain our sample selection, before outlining our information interventions and how voters interpreted them.
2.4.1 Sample selection
Our study focuses on 26 municipalities in the central states of Guanajuato (7 municipal- ities), México (14 municipalities), San Luis Potosí (4 municipalities) and Querétaro (1 mu- nicipality). These are shown in Figure 2.2. Beyond holding elections in 2015,12 these states were chosen for logistical reasons, because they contain internal variation in the municipal incumbent party and because they broadly represent Mexico as a whole. The 26 munici- palities were selected from those where an audit was released in 2016 according to three criteria. First, the safety of voters and our distribution and survey teams. After immediately receiving threats upon entering Aquismón and Villa Victoria, these municipalities were re- placed by Atlacomulco, Temoaya and additional precincts from Tlalnepantla de Baz in the state of México. Importantly, our blocking strategy—explained in detail below—ensures that all blocks are contained within the same municipality, and thus excluding these prob- lematic municipalities does not affect the internal validity of our study. Second, we only selected municipalities over-performing, and especially under-performing, relative to the state average of opposition parties by at least 2 percentage points. Finally, municipali- ties were chosen to match the distribution of incumbent parties across audited municipal governments across the 4 states. Of our 26 municipalities, 17 were governed by the PRI (including 16 in coalition with the Teacher’s (PANAL) and Green (PVEM) parties), 5 by
12Municipal elections reflect state electoral cycles, which are staggered across years. On 7th June 2015, 15 states and the federal district held simultaneous local elections. 36 Figure 2.2: The 26 municipalities in our sample
the PAN (including 2 in coalition with PANAL), 2 by the PRD, and 1 by the Citizen’s Movement (MC). Within each municipality, we selected up to one third of the electoral precincts, over- sampling precincts from municipalities with particularly high and low levels of incum- bent malfeasance and strong contrasts with opposition party performance within the state. We first prioritized accessible rural precincts, where possible, in order to minimize cross- precinct spillovers and maximize the probability that voters would not receive the informa- tion through other means. Moreover, to maximize the share of households we could reach with a fixed number of leaflets, attention was restricted to precincts with fewer registered voters (subject to our blocking design detailed below). In urban areas, where we had more precincts to chose from, we restricted our sample to precincts with at most 1,750 registered voters, and designed an algorithm to minimize the number of neighboring urban precincts
37 in our sample.13
2.4.2 Treatments
In partnership with the non-partisan Mexico NGO Borde Político,14 we sought to eval- uate the impact of providing voters with incumbent performance information. Our primary treatment distributed leaflets to voters documenting the use of FISM funds in their munic- ipality. For each municipality, the leaflet focused on either spending that does not benefit the poor or unauthorized spending. All treatments were delivered at the electoral precinct level, Mexico’s lowest level of electoral aggregation. Our leaflet was designed to be non-partisan, accessible and sufficiently intriguing that voters would not discard it.15 The front page of the leaflet explains that Borde Político is a non-partisan organization and that the information contained in the leaflet is based on the ASF’s official audit reports available online. Figure 2.3 provides an example of the main page from a leaflet focusing on a severe case of unauthorized spending in the municipality of Ecatepec de Morelos in the state of México. The leaflet first states that FISM funds should only be spent on social infrastructure projects, and provides graphical examples of such projects on the right. The leaflet then informs recipients of the total amount of money their municipality received (146.3 million pesos, in this case), and the percentage of this money spent in an unauthorized way by their government (45%). To avoid suspicions of
13The algorithm started with the set of neighboring precincts surrounding each precinct and identified all neighboring precincts that were eligible for our sample; we then iteratively removed the precinct with most “in-sample” neighbors until we reached the required number of precincts for that municipality. In most municipalities, the algorithm ensured that our sample contained no neighboring precincts. 14Borde Político is a leading NGO seeking to increase voter knowledge about the actions of their politicians in office, with significant experience in the development of web-based platforms providing politically-relevant information to individuals (see borde.mx). 15The particular design was produced by a graphic artist based on feedback from multiple focus groups. We also sought legal advice to ensure that our leaflets did not constitute political advertisements, and thus were not subject to distribution restrictions.
38 EL DINERO DEL FISM, FONDO DE INFRAESTRUCTURA AGUA DRENAJE CAMINOS LUZ SOCIAL MUNICIPAL, DEBE POTABLE
GASTARSE EN OBRAS DE ESCUELAS CLÍNICAS VIVIENDA INFRAESTRUCTURA
LOS GASTOS QUE NO SEAN EN OBRAS DE INFRAESTRUCTURA DEBEN SER 0%
EN 2013, EL PARTIDO QUE
GOBIERNA ECATEPEC RECIBIÓ AGUA DRENAJE CAMINOS LUZ POTABLE 146.3 MILLONES DE PESOS DEL FISM Y GASTÓ 45% EN COSAS ESCUELAS CLÍNICAS VIVIENDA QUE NO DEBE
GASTÓ COMO NO DEBE
45 PARTIDO QUE GOBIERNA ECATEPEC
DE ¡PIÉNSALO! EL EL VOTO ¡COMPÁRTELO! 7 JUNIO DEPENDE DE TI
Figure 2.3: Example of local-information only leaflet in the municipality of Ecatepec de Morelos in the state of México
political motivation, neither the mayor nor their party is referred to directly, although as noted above the vast majority of voters can correctly identify the party of their incumbent mayor. Figure 2.4 provides an example from the municipality of Salamanca in Guanaju- ato, where all 54.1 million pesos were correctly allocated to social infrastructure projects benefiting the poor. To investigate how the provision of information impacts voters, we varied leaflet dis- semination along two theoretically important dimensions. First, to identify the impact of providing voters with a benchmark against which to compare their municipality’s perfor- mance, we produced a comparative leaflet. In contrast with the local leaflet shown in
39 EL DINERO DEL FISM, FONDO DE INFRAESTRUCTURA SOCIAL MUNICIPAL, DEBE GASTARSE EN OBRAS QUE BENEFICIEN A LOS QUE MENOS TIENEN.
¡LOS GASTOS EN OBRAS QUE NO BENEFICIAN A LOS QUE MENOS TIENEN DEBEN SER 0%
EN 2013, EL PARTIDO QUE GOBIERNA SALAMANCA RECIBIÓ 54.1 MILLONES DE PESOS DEL FISM Y GASTÓ 0% EN OBRAS QUE NO BENEFICIAN A LOS QUE MENOS TIENEN.
¡COMPAREMOS CON LOS GASTOS DE OTROS PARTIDOS!
MUNICIPIOS DE TU ESTADO GOBERNADOS POR OTROS PARTIDOS GASTARON EN PROMEDIO 16% EN OBRAS QUE NO BENEFICIAN A LOS QUE MENOS TIENEN.
GASTOS QUE NO BENEFICIAN A LOS QUE MENOS TIENEN
0 16 PARTIDO QUE OTROS GOBIERNA PARTIDOS SALAMANCA EN TU ESTADO
DE ¡PIÉNSALO! EL EL VOTO ¡COMPÁRTELO! 7 JUNIO DEPENDE DE TI
Figure 2.4: Example of a comparative information leaflet in the municipality of Salamanca in the state of Guanajuato
Figure 2.3, the comparative leaflet in Figure 2.4 involved providing information about the mean outcome among all audited municipalities within the same state but governed by a different political party in the second and third panels. In this example, voters in Salamanca were informed that while their government spent all FISM funds on projects not benefiting the poor, audited municipal governments run by other parties in Guanajuato spent 16% of funds on projects not benefiting the poor. Second, to vary the extent to which the distribution of the leaflets is common knowl- edge among voters within the precinct, we also varied whether the leaflets was delivered in a private or public manner. For the public mode of delivery, the door-to-door delivery de-
40 hi w uiiaiyad9 nmncplte ihntersae oendb te par- other by governed states their within within spending municipalities malfeasant in 21% 9% and of municipality informed own was their precinct average The state. municipalities the audited within governing parties other among average the and municipality voter’s and share to provided. them information encouraged the and information discuss mayor, receive municipal also their would of neighbors performance their the concerning that voters on informed message message second The 30 a loop. playing while leaflets distributing members team other alongside single a messages, campaign blaring of back the on member. carried team speaker a loud portable powerful a by accompanied was above scribed oiia apin iial ekn obodattermessage. their broadcast to seeking similarly campaigns political iue25 uiiaiisadpeicsb hr fmlesn pnigi u sample our in spending malfeasant of share by precincts and Municipalities 2.5: Figure 16 sostedsrbto fmlesn pnigi u ape ohwti a within both sample, our in spending malfeasant of distribution the shows 2.5 Figure fpeicsi u apefo htmunicipality. that from sample our in precincts of Notes eprhsdteemdfidrcsc odsekr rmavno nMxc iyta loserves also that City Mexico in vendor a from speakers loud rucksack modified these purchased We ahpiti n for2 uiiaiis h ieo onscrepnst h number the to corresponds points of size The municipalities. 26 our of one is point Each :
Incumbent malfeasant spending (other parties in state) 16 0 .05 .1 .15 .2 knt h eilscmol rvn rudbfr eia elections Mexican before around driving commonly vehicles the to Akin 0 Incumbent malfeasantspending perifonista .2 41 akdtruhtesreso ahprecinct each of streets the through walked .4 .6 ties, while these variables respectively ranged from 0% to 58% and 0% to 18%; 74% of precincts received relatively bad news about their incumbent party.
2.4.3 Block randomization, compliance and data
Our sample of 678 precincts was allocated according to the factorial design with a pure control shown in Table 2.1. To maximize our ability to differentiate the effects of compar- ative information and public dissemination, the 400 treated precincts were equally divided between the four variants of the information treatment. The control group comprising 278 electoral precincts reflects our sampling and block randomization. For the randomization, precincts were first stratified into rural or urban blocks of 6 or 7 similar precincts within a given municipality.17 Precinct similarity was defined by the Mahalanobis distance between 23 social, economic, demographic and political variables provided by Mexico’s National Statistical Agency (INEGI) and the National Electoral In- stitute (INE).18 Within each block, we then randomly assigned precincts to each of the treatment conditions and, depending on the availability of an additional precinct, either 2 or 3 pure control precincts. Block randomization ensures that different municipalities do not receive different treatment dosages and maximizes the power of the experiment by minimizing differences between treated and control precincts. Leaflets were distributed by our implementing partners Data OPM and Qué Funciona para Desarollo using precinct maps provided by state electoral institutes. Starting from a random location, our distribution team delivered one leaflet to a maximum of 200 house- holds within each treated precinct. Where possible, leaflets were delivered in person with
17Subject to there existing sufficient precincts, and the total treated precincts not exceeding one third of all precincts, we used blocks of 7 precincts. 18We used the R package blockTools to assign precincts to blocks. The algorithm is greedy in that it creates the most similar group first. Where a surplus of potential precincts were available, we used the most similar blocks to maximize statistical efficiency.
42 Table 2.1: Factorial design with a pure control Control Private Public Control 278 precincts Local 100 precincts 100 precincts Comparative 100 precincts 100 precincts
a short message explaining the leaflet’s provenance. When no adult was available, leaflets were left in mail boxes or taped to the recipient’s front door in a water-proof bag. This process typically took several hours per precinct, and was implemented over two weeks up until the no-campaigning period a week before the election. Our team recorded where leaflets were distributed for our follow-up survey. While compliance with the delivery of our treatments was very good in general, we nevertheless encountered some issues in the field.19 In a couple of cases, some leaflets were delivered to voters outside the precinct or adverse weather conditions and poor road conditions prevented us from reaching a precinct. To preserve the randomization, we focus on estimating intent to treat (ITT) effects. Such estimates are also most policy-relevant. Finally, we collected two additional sources of data to measure our principal out- comes. First, we downloaded and requested official precinct-level electoral returns from each state’s electoral institute. We drop 3 precincts in our sample that were merged with another precinct, due to having fewer than 100 registered voters, leaving a final sample of 675 electoral precincts.20 Second, we conducted a post-election survey interviewing 10 voters from each of the treated precincts and 100 randomly selected control precincts.21 This survey included questions regarding voting behavior, perceptions of incumbent and
19In addition, a new block in Chimalhuacán replaced a block where treatments were misassigned. We move the misassigned block from our sample, but the results are unaffected by its inclusion. 20In two of these cases, the small precinct was merged with another precinct that remains in our sample; where the treatment condition conflicts, we retain the larger precinct’s treatment status. 21Budgetary constraints prevented a baseline survey.
43 challenger parties, social coordination, the responses of political parties to our information, and receipt of the leaflet.22
2.4.4 Estimation and balance
To estimate the average ITT effect of providing any type information—local or com- parative, or public or private—we estimate OLS regressions of the form:
Ypbm = βTreatmentpbm + ρbm + εpbm, (2.7)
where Ypbm is an outcome for electoral precinct p within randomization block b in mu- nicipality m. For individual-level survey outcomes, Yipbm also includes an i subscript.
Block fixed effects, ρbm, are included to enhance the efficiency of our estimates by en- suring that we only exploit variation in treatment assignment within blocks of similar precincts. Throughout, standard errors are clustered at the municipality-treatment level, while precinct-level observations are weighted by the inverse of the share of voters to whom we delivered a leaflet (in control precincts, we use the average number of leaflets delivered to treated precincts). This weighting scheme—the only departure from our pre-registered regression specifications—permits more precise estimates by de-weighting large precincts where only a small fraction of voters could receive the leaflet. We show similar results weighting each precinct equally as a robustness check in Table 2.6 below. We also use this baseline specification to validate the randomization. Unsurprisingly, as Table A.1 in the Appendix demonstrates, the treatment is well balanced across 46 precinct
22Despite its success in the administration of the Mexico Panel Surveys and Comparative Study of Electoral Systems modules, our attempts to gauge vote choice by simulating the electoral process with an urn during the survey failed. Many voters felt uneasy in the aftermath of surprising electoral results, and refused to participate believing that our survey team were working on behalf of a party to identify individual vote choices or to trick voters into casting a different ballot. Unfortunately, we had to discard this question from our analysis.
44 and survey respondent characteristics. As usual, there are some significant differences, most notably with respect to incumbent vote share. However, in Table 2.6 we demonstrate that the results are robust, and if anything more precisely estimated, when controlling for pre-treatment variables. Our main estimates pool municipalities receiving information about spending on projects not benefiting the poor and unauthorized spending. If voters evaluate these dimensions of malfeasance similarly, as the findings of Larreguy, Marshall and Snyder(2015) suggest, this maximizes the power of the experimental design. Since voters could plausibly respond to negligent and corrupt spending differently, we also examine the types of information separately in Table 2.5 and observe similar responses. To analyze the mechanisms under- pinning the information treatment, we estimate analogous specifications separating out the dimensions of our information treatment. To examine how the effects of providing audit result information vary with how the information is perceived by voters, we estimate interactive specifications of the form: