# Leveling the playing field: How campaign advertising can help non-dominant parties

, Volume 16 (6) – Dec 1, 2018
38 pages

Publisher
Oxford University Press
Abstract We examine how campaign advertising affects electoral support. We propose a simple model where advertising disproportionately benefits non-dominant political parties, because voters are uncertain about and biased against such parties. We test this argument in Mexico, where one of the three main parties dominates in many localities. To identify the effects of exposure to campaign advertising, we exploit differences across neighboring precincts in campaign ad distribution. These differences originate from cross-state media coverage spillovers induced by a 2007 reform that equalized access to ad slots across all broadcast media. We find that, on average, ads on AM radio increase the vote shares of the PAN and the PRD, but not the previously hegemonic PRI. Consistent with our model, campaign advertising is most effective in poorly informed and politically uncompetitive electoral precincts, and against locally dominant parties of intermediate strength. 1. Introduction It is widely believed that campaign advertising can effectively mobilize or persuade voters to support the party behind the ad. In the U.S. 2012 presidential campaign, for example, both parties spent more than $400 million on television ads.1 Despite the perceived wisdom of campaign ads, the extent to which they are actually effective remains unclear (see, e.g., DellaVigna and Gentzkow 2010). Furthermore, in many elections across the world, parties do not compete on a level playing field. In contexts where a dominant party captures the media (e.g., Durante and Knight 2012; Lawson and McCann 2005; McMillan and Zoido 2004) or is well-known due to its local machines or perpetual incumbency, campaign ads could play a key role in informing voters about non-dominant parties. We analyze, using a simple learning model in the spirit of Zaller (1992), the impact of changing a party’s share of campaign advertising on vote choice in contexts where one party is dominant. In our model, a party is dominant in two respects. First, informational dominance entails that the utility a voter will receive if the dominant party wins office—generally reflecting the party’s policy positions, policy emphasis, or competence—is known with more certainty, relative to that of the locally non-dominant party (see also Shepsle 1972). Second, ideological dominance entails a bias toward the dominant party among voters, which could originate from non-performance based factors such as clientelistic ties or voter loyalty. Upon reaching voters, campaign ads are more informative about the utility level associated with the non-dominant party obtaining office. Advertising thus allows voters to learn about the relative benefits of each party and decreases the uncertainty surrounding the utility that they would receive upon electing a party that is not locally established. The model predicts that campaign advertising’s effect on voting behavior is greatest among uninformed voters with imprecise prior beliefs about the consequences of electing the non-dominant party, in locations (or elections) where political competition—and thus other local political activity—is low, and where the ideological bias toward the dominant party is not insurmountable. Campaign advertising is, therefore, most electorally beneficial for non-dominant parties where the locally dominant party is neither very strong nor facing severe competition. However, this non-linear relationship in the level of dominance should only apply to non-dominant parties, since voters are already well-informed about the dominant party. Mexico represents an important application of campaign advertising’s potential to shift votes away from parties that are locally dominant, a common way in which dominance is manifested in developing democracies. Despite losing the Presidency in 2000, after seven decades in power, the Institutional Revolutionary Party (PRI) has continued to dominate poorer and more rural parts of Mexico (Langston 2003, 2006). Mexico’s other main political parties—the National Action Party (PAN) and the Party of the Democratic Revolution (PRD)—have now also developed local strongholds of their own. These are generally located in more urban and developed areas, although the PRD—which split from the PRI—also has a significant rural presence in the southern states. Since informational and ideological dominance predominantly manifests very locally, we consider dominance at the electoral precinct level. Moreover, relatively low levels of voter knowledge about political parties suggest that campaign ads have the potential to substantially shape voters’ partisan preferences (e.g., Greene 2011; Lawson and McCann 2005). To identify the effects of campaign advertising, we leverage a major campaign regulation reform reducing inequalities in access to advertising across the country. Beginning in 2009, the reform mandated that all ads broadcast on radio and television over the course of federal election campaigns be allocated by Mexico’s independent Federal Electoral Institute (IFE) according to a formula reflecting the number of parties standing and their previous vote share.2 This formula is adjusted for media outlets located in states holding concurrent local elections, and thus generates cross-state variation in the share of ads allocated to each party. To variation in the probability of exposure to ads from each political party, we exploit differences in the distribution of political advertising between neighboring electoral precincts that originates from differential access to media outlets from different states. We focus primarily on AM radio because its substantial signal coverage extends beyond urban areas and more frequently crosses state borders than FM radio or television signals. This yields a large and disproportionately poor and rural sample, precisely the locations that our theory predicts that campaign advertising should be most effective. Pooling the 2009 and 2012 federal legislative elections, we first show that greater campaign advertising on AM radio substantially increased the vote shares of the PAN and the PRD. Specifically, a standard deviation increase in the campaign ad exposure share of the PAN and the PRD respectively increased their vote share by 3 and 2.3 percentage points, or 11% and 14%.3 Conversely, we find no evidence that PRI campaign ads affected the average PRI candidate’s vote share. The estimated ineffectiveness of PRI advertising suggests that an important legacy of its time in power may be that voters retain relatively precise beliefs about its suitability for office that are not susceptible to campaign ads. We find no evidence to suggest that campaign ads mobilized turnout. Consistent with our theoretical model, the electoral efficacy of PAN and PRD campaign ads has varied across electoral precincts. First, in less economically developed precincts—where our survey evidence indicates that voters were less politically informed—ads were more effective at winning votes. Second, ads were less effective in more competitive precincts, where voters were more politically knowledgeable. Third, we find some evidence that campaign ads were less effective concurrent to the intensely contested 2012 presidential election. Finally, the effects of campaign ads for locally non-dominant parties were non-linear in the vote share of the dominant party. Specifically, ads were least effective in both the most competitive and most locally dominated precincts. Taken together, our results demonstrate that equalizing access to campaign advertising can significantly increase support for locally non-dominant parties. This suggests that broad-based campaign advertising can help foster multi-party competition and informed political participation. On the other hand, our findings highlight the importance of informational advantages accruing to dominant parties, and thus challenge models of political competition where the policy positions and competence of the major parties are assumed to be equally well-known (e.g., Downs 1957). These main findings are robust to various potential identification and interpretation concerns. First, we use a permutation test to demonstrate that random allocations of advertising shares across states do not produce similar results. Second, a variety of checks indicate that measurement error in signal coverage cannot explain our findings. Third, the results are robust to sensitivity analyses including control variables and sample restrictions. Fourth, we show that the findings are supported in the smaller and more urban FM radio and television samples, where our model implies similar heterogeneous effects but smaller average effects. Finally, contrary to the concern that our effects are driven by partisan news coverage rather than advertising, a placebo test shows that the same media allocation formula does not produce the same results before the reform was implemented. Our findings contribute to the literature identifying the effects of campaign advertising in developed and developing countries. The U.S. literature has generally found a limited impact on electoral outcomes (e.g., Ansolabehere et al. 2006; Huber and Arceneaux 2007; Krasno and Green 2008; Levitt 1994) and short-lived effects on voter perceptions (Gerber et al. 2011; Zaller 1992). However, a recent study utilizing an unusually fine-grained spatial design akin to ours similarly finds that television advertising can meaningfully affect county-level vote share without altering turnout (Spenkuch and Toniatti forthcoming). Moreover, our results complement previous studies arguing that a key function of electoral campaigns—via political advertising in our case—is to reduce voter uncertainty about the policy positions and characteristics of different candidates (Lenz 2009; Peterson 2009). Our findings regarding the importance of party dominance also chime with evidence from Italy that media partisan control can also occur in consolidated democracies (Durante and Knight 2012). In contrast, the relatively nascent developing country literature suggests that campaign ads can be highly effective outside established democracies. Although Da Silveira and De Mello (2011) find that differences in television ad allocations between the first and second round of Brazilian gubernatorial elections influence candidate vote shares, we examine an entire campaign without the risk that strategic behavior between rounds confounds our estimates of advertising’s effects. Surveys exploiting less compelling identification strategies also point to powerful effects of campaign advertising in Mexico (Greene 2011; Lawson and McCann 2005). However, such studies do not explain when and where different parties benefit from campaign ads. Exploiting the random assignment of ad slot times in Mexico, Durante and Gutierrez (2014) also find that vote intentions track prime time television and radio advertisements. The effectiveness of ads in developing democracies also contrasts with authoritarian regimes. In such regimes, the media is often controlled or manipulated by the state (e.g., King et al. 2014; McMillan and Zoido 2004), whereas opposition groups possess few opportunities to express their political preferences and platforms (e.g., Djankov et al. 2003). Given the extant evidence, our findings suggest that campaign ads may be most effective in consolidating democracies with dominant parties like Mexico. In such cases, voters are less well informed—particularly about non-dominant parties—and media markets are less concentrated than advanced democracies. Moreover, unlike authoritarian regimes, political competition is sufficiently robust that credible alternatives to dominant parties exist. These findings provide hope for democrats, given that many other consolidating democracies have recently introduced reforms guaranteeing political parties relatively equitable access to campaign advertising.4 Our findings also complement the literature examining the impact of biased and relatively unbiased news media, as opposed to campaign advertising. Various studies have found that media coverage increases voter punishment of incumbent indiscretions in office (Fergusson et al. 2014; Ferraz and Finan 2008; Larreguy et al. 2017a). Using a similar design to ours, Enikolopov et al. (2011) show that the introduction of an independent television station increases the vote share of opposition parties not supported by Russian state media. Unlike campaign advertising, which our results suggest may be considerably more effective outside the relatively informed electorates of consolidated democracies, the findings in the media literature broadly reinforce prominent studies from the United States (e.g., Chiang and Knight 2011; DellaVigna and Kaplan 2007; Gentzkow et al. 2011; Snyder and Strömberg 2010). Finally, this article contributes to several broader debates. First, it offers a more concrete mechanism for previous studies indicating that campaign spending is effective at winning votes (e.g., Spenkuch and Toniatti forthcoming). Second, our results suggest that a key function of electoral campaigns—via advertising in our case—is to reduce voter uncertainty about the policy positions and characteristics of different candidates (Lenz 2009). Third, complementing the consumer advertising literature (see DellaVigna and Gentzkow 2010), we provide further evidence that advertising through broadcast media can persuade individuals to alter their behavior. In particular, our results reinforce the finding that advertising is most effective among consumers with little prior exposure to a product (e.g., Ackerberg 2001). The paper proceeds as follows. Section 2 provides a brief overview of politics and media in Mexico, focusing on Mexico’s campaign advertising reform. Section 3 develops a simple model to analyze the voting implications of campaign advertising in a democracy with dominant parties. Section 4 details our data and identification strategy. Section 5 presents our main results and robustness checks. Section 6 concludes. 2. Politics and Media in Mexico Mexico is divided into 31 states (and the federal district of Mexico City), and operates a presidential form of government. National legislative elections are held every three years, with members of the Chamber of Deputies (House) and Senate elected to single three- and six-year terms respectively.5 The president is elected to a six-year term simultaneous to every other federal legislative election. We focus on the Chamber of Deputies, which contains 300 members elected via plurality rule from single-member districts and 200 members elected according to the national party’s vote share via proportional representation. Mexico’s circa 67,000 electoral precincts make up the legislative districts (within states) that elect national representatives. Between 1929 and 2000, widespread clientelistic practices and electoral manipulation ensured that the Institutional Revolutionary Party (PRI) maintained a stranglehold on the Presidency and almost always retained Congressional majorities. However, Mexican politics became more competitive over the last two decades as the PRI’s grip on power subsided. In 2009 and 2012, three main political parties competed for political control: the left-wing Party of the Democratic Revolution (PRD), the populist PRI, and the right-wing National Action Party (PAN). In this section, we provide a brief overview of political competition, before describing campaign advertising in Mexico and the 2007 media reforms. 2.1. Political Competition Following Mexico’s revolution in 1929, the PRI retained hegemonic status up until the 1990s. The masses were co-opted into the regime, campaigning relied heavily upon distributing public resources and mobilizing voter turnout, and dissension within the party was minimized by maintaining a high political cost of exit (e.g., Cornelius 2004; Fox 1994; Greene 2007; Magaloni 2006). Nevertheless, PRI politicians frustrated by the party’s hierarchy ultimately formed the left-wing offshoot National Democratic Front. This became the PRD in 1989, and has since built a strong base in Mexico City and among relatively poor southern states. The PRI continued to govern in the 1990s, but conceded constitutional reforms in order to receive the Congressional support from the right-wing PAN required to pass pressing legislation to address economic crises. In the more competitive electoral environment, the PRI first lost control of the House in 1997 before PAN candidate Vicente Fox won the Presidency in 2000 (Greene 2007; Magaloni 2006), based on strong support in Mexico’s business-friendly northern and western states. In 2006, the PAN and the PRD became the largest parties in the legislature, and the PAN narrowly retained the Presidency following a contentious wafer-thin election victory. Although the PRI’s vertical hierarchy dispensing patronage was damaged by losing federal office and the party became fractionalized (Langston 2003), its powerful regional presence remained. In almost one-third of states, the PRI never relinquished gubernatorial control to another party, whereas the reforms designed to ensure fair electoral competition at the national level left local elections—that continued to fall under the jurisdiction of state electoral institutes—relatively untouched. These advantages, combined with new decentralized mechanisms better selecting candidates popular in their local area (Langston 2006) and continued vote- and turnout-buying (Larreguy 2013; Larreguy et al. 2016; Nichter and Palmer-Rubin 2015), helped the PRI reclaim its majority status in 2009 in coalition with the New Alliance Party and the Ecological Green Party (PVEM), and the Presidency in 2012. 2.2. Campaign Advertising and the 2007 IFE Reform Disproportionate access to political advertising in the media became a political issue as Mexico transitioned toward competitive democracy in the 1990s. Although a series of constitutional reforms were approved in 1989 and the operational establishment of the Federal Electoral Institute (IFE) in 1990—which became politically independent in 1996—contributed to substantially reducing vote fraud, the PRI enjoyed privileged access to public resources and lower commercial advertising costs, as well as significantly greater coverage and positive appraisals across media formats (Hallin 2000; Lawson 2002; Lawson and McCann 2005). However, the IFE has since progressively increased its regulation and monitoring of advertising spending by political parties, and become more willing to punish violations with fines. As the PRI’s dominance subsided in the 1990s, the PAN capitalized by dominating media coverage and strategically targeting marginal voters. Lawson (2004) and Lawson and McCann (2005) argue that more equal access to television time was essential to Vicente Fox’s victory in the 2000 presidential election. Similarly, Greene (2011) suggests that differential media access—in particular controlling 66% of television advertising time—was the primary reason behind Felipe Calderón’s victory in 2006 by just 0.56% of the vote. The result was highly contentious, given the PAN’s powerful media attacks against Andrés Manuel López Obrador and the 240 cases of electoral irregularities highlighted by the PRD. Despite upholding all such irregularities, the IFE nevertheless declared that they did not impact the electoral outcome. Ultimately, the IFE overhauled campaign advertising regulations in 2007, following the passage of major electoral reforms after the contentious 2006 elections. The new regulations, in force in federal elections since 2009, specify that neither political parties nor independent groups can buy campaign advertising on radio and television stations. The IFE is instead responsible for allocating all advertising slots to political parties during the pre-campaign and full electoral campaign that span the five-to-six months preceding federal elections. Every media station in the country is required to provide 41 min of 30-s campaign advertising slots throughout each day (until the final week of the campaign). The ordering of individual ad slots is randomly allocated by the IFE (see Durante and Gutierrez 2014). Media stations are legally bound by the distribution applied in the state from which their signal is emitted. The IFE determines the number of slots available to each political party using a clearly-defined formula that varies across states (see Online Appendix for details). In states not holding concurrent state-level elections, each party is allocated a minimum advertising share split equally between all parties or full coalitions (30% of total advertising time) and additional time according to their vote share at the previous national legislative election (70% of total advertising time).6 In states holding concurrent state-level elections, 15 of the 41 daily minutes available for advertising are apportioned according to the number of parties or full coalitions standing (30% of state advertising time) and the vote share at the previous state legislative election (70% of state advertising time). In 2009, 11 states simultaneously held state-level elections, whereas 15 states held concurrent elections in 2012.7 The distribution of campaign advertising shares thus varies across states but is fixed across all media stations broadcasting from within each state. Our hand-coded transcription of the 682 unique federal ads broadcast on radio and television during the 2012 election campaign indicates that parties principally used relatively uniform positive messages to convey their policy positions, the salience of particular issues, and emphasize their candidate’s competence.8 Of the 70% of ads that mentioned policy issues, the vast majority focused on valence issues like public security and employment and economic development. Education, health, corruption, and rural development also received significant attention. Although ads emphasized particular issues, and in some cases detailed policies to address them, parties did not generally seek to distinguish their proposals from those of other parties. Explicitly negative ads were outlawed by the 2007 reforms, although 7% of ads still solely attacked opposition parties. For example, some PRD ads alluded to the PRI’s history of corruption during their 70 years in power, whereas some PAN ads attacked the record of the PRI’s presidential candidate Enrique Peña Nieto. Although notably less frequent than policy issues, the competence of individual candidates—predominantly the principles, previous experience, and specific skills of federal candidates—was mentioned in 48% of ads. Consistent with a relatively uniform advertising strategy across the country, Table A.1 in the Online Appendix also shows that candidate mentions were skewed toward presidential candidates: a presidential candidate was mentioned in 56% of ads, whereas the many legislative candidates were mentioned in 44% of ads. The emphasis on the party and its presidential candidates likely reflects low name recognition for federal deputies. For example, the 2009 Comparative Study of Electoral Systems (CSES) survey found that only 18% of voters knew even one federal legislative candidate in their district. These relatively nationally uniform advertising strategies differ significantly from those used up until 2006. Before the reform, parties targeted clearly defined audiences, such as women watching afternoon telenovelas, and bought the corresponding air time to reach such audiences. After the reform, as one political strategist explained, parties were forced to fill many more slots catering to more diverse audiences, and instead adopted a more homogeneous strategy involving less advertising segmentation. 3. Campaign Advertising and Vote Choice with Dominant Parties Theories of special interest politics have typically assumed that greater campaign effort translates into votes (e.g., Baron 1994; Grossman and Helpman 1996; Snyder 1989). In these models, campaign contributions increase the probability that any voter supports the party being campaigned for. However, there now exists considerable evidence that providing factual and partisan politically relevant information affects voters very differently (e.g., Greene 2011; Lupu 2013). Where electorates are poorly politically informed about non-dominant parties, and voters are beholden to parties through local ties, the effects of campaign advertising could differ substantially across voters. Using a simple model to guide the empirical analysis, we thus ask: when is campaign advertising effective at winning votes in the presence of dominant parties? 3.1. Theoretical Model To examine the role of campaign advertising in the presence of dominant parties, we use a simple two-party decision-theoretic model of vote choice where one party is dominant to guide our empirical analysis. Specifically, political parties N and D compete for voters in a given precinct, where party D is dominant. This asymmetric treatment of parties is similar in spirit to models where incumbent politicians face challengers (Shepsle 1972), but contrasts with models of political competition where uncertainty is assumed to be symmetric across parties (e.g., Downs 1957). Parties. Party D is dominant in two respects: information and ideology. First, D’s “policy” outcome xD—which we construe broadly to include D’s policy position, emphasis on particular programs, and valence factors such as expected competence in office—is known with certainty by all voters.9 Conversely, the outcome associated with party N is uncertain. The prior belief of all voters is normally distributed according to |$\mathcal {N}(\delta ,\tau ^2)$|⁠, where δ is the prior distribution’s mean and τ2 > 0 is its variance. This stark difference in policy uncertainty simplifies the model and clarifies D’s informational dominance. However, similar results hold if party D’s position is known with relatively greater certainty than party N’s. Second, every voter i receives an ideological bias v + bi inclining them to vote for D. This represents favorability toward the dominant party, including factors such as loyalty biases, clientelistic benefits, and candidate-specific attributes. Although the average bias v is fixed across voters, bi allows bias to vary across voters. bi has a mean of 0 and is distributed according to cumulative distribution function F. In what follows, we examine how the effect of campaign advertising varies with v, which we interpret as the extent of party competition. To capture D’s ideological dominance we assume that F″ > 0, such that the mass of voters at a given bi is increasing in bi. By allowing the non-dominant party to overcome steadily more biased voters as it becomes more popular, this assumption ensures that the second-order effects of information complement the first-order effects.10 Voters. For simplicity, all voters share the same policy preferences but differ in their ideological bias toward party D. Assuming full turnout,11 voters must decide whether to vote for party D or party N. Each voter maximizes their expected utility, where their utility from policy outcome x is given by u(x) = −exp (−x).12 We thus assume that i’s ideological bias substitutes for policy benefits. Normalizing xD = 0, voter i therefore chooses party N over party D when |$\mathbb {E}[u(x_N)]\ge u(v + b_i)$|⁠. However, voters also learn about xN from campaign advertising. Campaign Advertising. Voters update their beliefs in response to campaign advertising according to their prior beliefs and the persuasiveness of the information they receive. Each voter receives n signals, or ads from party N.13 Each signal xj is independently drawn from the normal distribution |$\mathcal {N}(x_N,\sigma ^2)$|⁠, where xN is party’s N true (but unknown) policy and σ2 > 0 is the known variance of the signal distribution. We assume that σ2 > nτ2, which ensures that the signal does not overwhelm the prior. The mean signal received by a voter is |$\bar{x} = n^{-1}\sum _{j=1}^n x_j$|⁠, and voters use these signals to update their posterior belief about the benefits of N winning office. Some voters may have an optimistic prior about non-dominant party N and could then update negatively about N’s policy outcome. However, these are likely to be sufficiently few in number, since the share of voters biased toward the non-dominant party is likely to be small (given F″ > 0). Moreover, voters that are already biased toward the dominant party will not change their voting behavior. We then focus on the behavior of fairly representative voters for whom N’s policy outcome is sufficiently beneficial relative to D’s, and thus on those voters for whom information could cause them to switch away from the dominant party. Consequently, we let xN > 0 and |$\bar{x}-\delta >\sigma ^2/2n$|⁠.14 In words, N’s true policy outcome xN is better for voters than D’s, whereas voters’ prior beliefs are centered on an expectation sufficiently below N’s true policy outcome. Although this is an important driver of the model’s results, campaign ads would otherwise only play a limited role in influencing voter behavior. Applying Bayes’ rule, each voter’s posterior belief about the policy outcome if party N wins is distributed according to, \begin{eqnarray} \mathcal {N}\left( \frac{\frac{\delta }{\tau ^2} + \frac{n\bar{x}}{\sigma ^2}}{\frac{1}{\tau ^2} + \frac{n}{\sigma ^2}} , \bigg ( \frac{1}{\tau ^2} + \frac{n}{\sigma ^2} \bigg )^{-1} \right). \end{eqnarray} (1) Consequently, each voter’s expected utility from party N winning office is given by \begin{eqnarray} \nonumber \mathbb {E}u(x_N) &=& -\exp \left[ -\left(\frac{\frac{\delta }{\tau ^2} + \frac{n\bar{x}}{\sigma ^2}}{\frac{1}{\tau ^2} + \frac{n}{\sigma ^2}} - \frac{1}{2} \left( \frac{1}{\tau ^2} + \frac{n}{\sigma ^2} \right)^{-1}\right)\right] \nonumber\\ &=& -\exp \left[ -\left(\frac{\delta \sigma ^2 + n\bar{x}\tau ^2}{\sigma ^2 + n\tau ^2} - \frac{1}{2}\frac{\tau ^2\sigma ^2}{\sigma ^2+n\tau ^2}\right)\right] , \end{eqnarray} (2) where the first term is the voter’s expectation of N’s policy outcome, and the negative second term reflects their disutility from risking the election of a candidate whose policy outcomes are uncertain.15 Defining \begin{equation*} R := (\delta \sigma ^2 + n\bar{x}\tau ^2)/(\sigma ^2 + n\tau ^2) - (\tau ^2\sigma ^2)/[2(\sigma ^2+n\tau ^2)], \end{equation*} voter i chooses to vote for party N over party D when R > v + bi. Equation (2) highlights several implications of campaign advertising. First, voters are more likely to believe that party N’s policy outcome will benefit them as the number of ads, n, increases. Second, as in Zaller (1992), voters with precise priors—smaller τ2—are less responsive to an additional ad from N. Third, the effect of campaign advertising on the belief that N will be beneficial increases with the precision of the signal, or as σ2 decreases. Combining voter beliefs with the decision to vote for party N over party D yields our main result determining when a voter supports a non-dominant party. Proposition 1. The proportion of votes for party N, the non-dominant party, is VN ≔ F(R − v). The following comparative statics hold: The vote share of N is increasing in n (i.e., ∂VN/∂n > 0). The effect of n on the votes share of N is decreasing in v and σ2, and increasing in τ2 (i.e., ∂2VN/∂n∂v < 0, ∂2VN∂n∂σ2 < 0, and ∂2VN/∂n∂τ2 > 0). Proof. See Online Appendix. Intuitively, part (a) of the proposition demonstrates that increasing party N’s campaign advertising increases N’s vote share by strengthening voters’ posterior belief that N would implement a desirable policy if elected. However, the results in part (b) imply that this effect will vary depending on contextual campaign advertising and party characteristics. First, increasing the valence bias v toward party D reduces the effectiveness of N’s ads because campaign advertising is less able to overcome strong biases in favor of D. Second, where the precision 1/τ2 of voters’ prior belief that party N will implement δ(<xN) is high, voters will positively update their posterior beliefs less substantially in favor of N. Third, voters are less confident that party N will implement a desirable policy, and thus relatively less likely to vote for N, when ads are relatively imprecise (i.e., high σ2). 3.2. Applicability to Mexico Mexico is an appropriate context to test the model’s implications at the local level. First, despite possessing three main parties, most parts of the country experience two-party competition. This largely follows from the regional concentration of Mexico’s three main political parties. As noted previously, the PRI remained dominant in many states despite losing its stranglehold on national offices, the PRD inherited and retained strong support in southern areas after breaking away from the PRI, and the PAN now has control over many urban areas.16 Furthermore, Larreguy et al. (2016) show a stark rural-urban divide, where the PRI dominates rural areas, and the PRD and especially the PAN both win votes in urban settings. These differences ensure that, as captured by our model, most voters experience two-party competition locally. In 2009 and 2012, the third-placed party only received more than 20% of the vote in 7% of electoral precincts and 8% of districts. Second, party dominance is often manifested at a lower level of aggregation than the district served by each incumbent. Most of Mexico’s 300 federal legislative districts—especially those outside Mexico’s largest cities—contain a mix of urban areas and rural localities often scattered across municipalities with very different local political leaders. Furthermore, Larreguy et al. (2016) show that political brokers typically operate at the precinct-level, whereas Larreguy (2013) and Larreguy et al. (2017b) find that clientelism is particularly marked in small rural and urban localities. There is thus substantial variation in both the extent of dominance and which party dominates within Mexico’s federal legislative districts. Consequently, our analysis focuses on precinct-level dominance rather than district incumbency. Third, as in the model, local dominance generally reflects informational and ideological advantages. For example, where the PRI is dominant, voters often receive material benefits from the PRI, which they expect to receive if they continue voting for the PRI. In contrast, voters are likely to be uncertain of the benefits of voting for the PAN or the PRD. To assess this characterization of dominance, Table 1 examines indicators of an individual’s knowledge of party candidates in the Comparative Study of Electoral Systems (CSES) surveys conducted after Mexico’s 2006, 2009, and 2012 federal elections.17 Consistent with parties possessing local informational dominance, the bivariate correlations suggest that voters are 5 percentage points more likely to know the PAN’s presidential candidate in precincts where the PAN was the party with the largest vote share in 2006, and 2 percentage points more likely to know the PRD’s candidate in precincts where the PRD was the party with the largest vote share.18 Although merely correlational, such differences are especially stark in the context of high rates of reported knowledge. However, voters are not significantly more likely to know the PRI’s candidate when the PRI was the party with the largest vote share in 2006. This lack of a significant difference could potentially reflect decades of PRI rule. Table 1. Correlation between local party dominance and political knowledge. Knows PAN candidate Knows PRD candidate Knows PRI candidate (1) (2) (3) PAN largest party (2006) 0.052*** (0.016) PRD largest party (2006) 0.021* (0.012) PRI largest party (2006) 0.007 (0.016) Observations 12,332 12,332 12,332 Outcome mean 0.91 0.92 0.90 Outcome standard deviation 0.29 0.27 0.30 Largest party (2006) mean 0.28 0.18 0.54 Largest party (2006) standard deviation 0.45 0.38 0.50 Knows PAN candidate Knows PRD candidate Knows PRI candidate (1) (2) (3) PAN largest party (2006) 0.052*** (0.016) PRD largest party (2006) 0.021* (0.012) PRI largest party (2006) 0.007 (0.016) Observations 12,332 12,332 12,332 Outcome mean 0.91 0.92 0.90 Outcome standard deviation 0.29 0.27 0.30 Largest party (2006) mean 0.28 0.18 0.54 Largest party (2006) standard deviation 0.45 0.38 0.50 Notes: Each regression pools the 2006, 2009, and 2012 Comparative Study of Electoral Systems surveys. For the outcome, we code indicators for respondents that both know a given party’s candidate and has an opinion about that candidate. The independent variables indicate whether the PAN, the PRD, or the PRI received the most votes in the precinct in 2006. All specifications are bivariate OLS regressions. Standard errors are clustered by state; our sample contains 32 clusters. *p < 0.1; ***p < 0.01. View Large Table 1. Correlation between local party dominance and political knowledge. Knows PAN candidate Knows PRD candidate Knows PRI candidate (1) (2) (3) PAN largest party (2006) 0.052*** (0.016) PRD largest party (2006) 0.021* (0.012) PRI largest party (2006) 0.007 (0.016) Observations 12,332 12,332 12,332 Outcome mean 0.91 0.92 0.90 Outcome standard deviation 0.29 0.27 0.30 Largest party (2006) mean 0.28 0.18 0.54 Largest party (2006) standard deviation 0.45 0.38 0.50 Knows PAN candidate Knows PRD candidate Knows PRI candidate (1) (2) (3) PAN largest party (2006) 0.052*** (0.016) PRD largest party (2006) 0.021* (0.012) PRI largest party (2006) 0.007 (0.016) Observations 12,332 12,332 12,332 Outcome mean 0.91 0.92 0.90 Outcome standard deviation 0.29 0.27 0.30 Largest party (2006) mean 0.28 0.18 0.54 Largest party (2006) standard deviation 0.45 0.38 0.50 Notes: Each regression pools the 2006, 2009, and 2012 Comparative Study of Electoral Systems surveys. For the outcome, we code indicators for respondents that both know a given party’s candidate and has an opinion about that candidate. The independent variables indicate whether the PAN, the PRD, or the PRI received the most votes in the precinct in 2006. All specifications are bivariate OLS regressions. Standard errors are clustered by state; our sample contains 32 clusters. *p < 0.1; ***p < 0.01. View Large 3.3. Hypotheses We now derive specific empirical predictions by aggregating voters at the electoral precinct level, which our empirical analysis focuses on. The model’s most obvious prediction—from Proposition 1(a)—is that greater campaign advertising (n) by a non-dominant party increases the probability that an individual votes for that party. Since no party dominates all parts of the country, campaign advertising has the potential to help all political parties wherever they are not locally dominant. Consequently, we thus hypothesize that for all parties, on average: Hypothesis 1. An increase in a party’s campaign advertising increases its vote share. The model also identifies the types of precincts where a party’s campaign advertising is most effective. Proposition 1(b) predicts that less well-informed voters—those with a weak prior, or large τ2—are the most responsive to new information provided by political parties. Greene (2011) and Lawson and McCann (2005) argue that a legacy of Mexico’s recent competitive authoritarian past is low levels of political knowledge. Low levels of political knowledge are concentrated among poor and rural voters, which are easier to measure empirically and can thus serve as precinct-level proxies for the precision of voters’ prior beliefs.19 We thus test whether impoverished voters, who are the least well informed, are most likely to internalize campaign ads, and therefore most likely to change their vote as a response:20 Hypothesis 2. Campaign advertising is most effective at winning votes in less developed parts of the country. However, campaign advertising is only one tool deployed by political parties. In competitive localities where multiple political parties use a variety of tactics to win votes, the effect of campaign advertising—which is fixed in quantity by law—may be crowded out by other activities.21 In terms of our model, alternative sources of information and persuasion may also reduce the marginal effect of a given ad by increasing the precision of voters’ prior beliefs (i.e., reduce τ2). Supporting this argument that local competition proxies for the precision of voters’ beliefs about parties, voters in more competitive precincts are more knowledgeable about their local candidates and political institutions.22 Proposition 1(b) therefore also suggests that: Hypothesis 3. Campaign advertising is most effective at winning votes in less politically competitive parts of the country. Similarly, although local political competition may differentially crowd out the effects of campaign advertising across electoral precincts, some elections are more salient than others. As in many other developing democracies, presidential elections in Mexico are particularly hard fought, and political parties dedicate more resources to their electoral strategies. Given that the quantity of campaign advertising is constant across national elections, even though the content may change, we also hypothesize that τ2 is larger and σ2 is smaller in mid-term elections, and thus predict that: Hypothesis 4. Campaign advertising is most effective at mid-term elections. Finally, and bringing together the key insights of our theoretical model, we do not expect the relationship between campaign advertising and local dominance to be linear. When there is little bias toward the locally dominant party, there are fewer votes for the locally non-dominant party to win and the election is likely to be more competitive (decreasing τ2 and increasing σ2). At interim levels of local dominance, voters are more susceptible to campaign advertising because they possess weaker prior beliefs about the non-dominant party (larger τ2) and advertising is not crowded out as much by political competition (smaller σ2). However, Proposition 1(b) shows that advertising ultimately becomes less effective once the ideological bias (v) toward the locally dominant party becomes sufficiently large that no amount of advertising can convince voters to abandon that party. Together, these insights imply that the effects of a non-dominant party’s advertising are non-linear in the level of local dominance: where a dominant party is relatively strong, but not completely commanding, we expect advertising to be most effective.23 In contrast, since the model assumes that the policies of locally dominant parties are well known, we expect to find weaker effects of campaign advertising among locally dominant parties. Hypothesis 5. Campaign advertising by locally non-dominant parties is most effective at intermediate levels of local dominance, whereas campaign advertising by locally dominant parties is relatively ineffective. 4. Research Design To identify the effects of campaign advertising on party vote share, we compare neighboring electoral precincts receiving differential exposure to campaign advertising due to differences in coverage by broadcast signals from out-of-state media stations. We first describe our data and explain our focus on AM radio ads, before detailing our identification strategy. 4.1. Data We collected data from various sources to produce a dataset combining campaign advertising shares for each political party, local economic, and demographic characteristics, and federal election vote shares for each electoral precinct. Electoral precincts—which typically contain 750–1,500 voters—are the smallest area for which media coverage and electoral data could be matched. Given that campaign advertising and signal coverage data at the media outlet-level were first collected after Mexico’s media reforms, we examine the 2009 and 2012 elections. We now describe our main variables; more detailed definitions and sources are provided in the Online Appendix. 4.1.1. Dependent Variable: (vote share) Our main outcome is the legislative vote share in the 2009 and 2012 elections, as a proportion of all votes cast, for each of Mexico’s three main political parties—the PAN, the PRD, and the PRI.24 We aggregate up to the precinct level the polling station-level returns for the 2000–2012 federal legislative elections provided by the IFE.25 4.1.2. Independent Variable: (party campaign advertising share) In their new regulatory role, the IFE collected data from every media station in the country after the 2007 media reforms.26 This data includes the location of the signal’s antennae, which allows us to identify the advertising distribution mandated in the associated state, and the coverage area for each station. The IFE defines the boundary of the coverage area using a 60 dBμ threshold for signal strength.27 This threshold is commonly used to determine a radio station’s audience and sell advertising space commercially.28 Inside a station’s coverage area the signal is of high quality, ensuring that interior precincts have good access to the station’s broadcasts. Precincts outside the coverage area experience sharply decreasing coverage quality as the distance from the boundary increases. We exclude the Federal District given that the small size of its electoral precincts reduces the validity of this comparison, whereas our identification strategy ensures that our sample is disproportionately rural. The number of media stations has not recently changed .29 Our principal independent variable is the share of campaign advertising from a given party to which an electoral precinct has access. Specifically, we compute the average share of campaign advertising for party i across all media stations g covering precinct j at election t: \begin{eqnarray} \mathit {advertising {\,\,} share}_{ijt} = \frac{1}{|\mathcal {G}_j|}\sum _{g\in \mathcal {G}_j} \mathit {media {\,\,} share}_{igt}, \end{eqnarray} (3) where |$\mathcal {G}_j := \lbrace g : g \text{ covers } j \rbrace$| is the set of stations covering precinct j and |$\mathit {media {\,\,} share}_{igt}$| is the share of ads allocated to party i in the state from which media station g emits. We compute |$\mathit {advertising {\,\,} share}_{ijt}$| separately for AM, FM, and television ads. We focus on the share of ads, rather than the total number of ads they could access, because by regulation the number of ads is constant across all media stations and voters cannot listen to multiple radio broadcasts simultaneously. Moreover, the random allocation of slots ensures that differences in access to prime time slots quickly averages out over the campaign (Durante and Gutierrez 2014). Our main analysis focuses on differences in campaign advertising from AM radio stations for several reasons. First, as Figure 2 indicates, AM radio’s large signal range ensures that 87% of electoral precincts in the country are covered by at least one AM radio station. In contrast with the weaker signals of FM radio and television antennae based in urban areas (see Figures A.1 and A.2 in the Online Appendix), AM radio reaches more rural and less well-informed voters (see Table 3).30 Our theory thus suggests that AM ads possess the greatest potential to diminish locally dominant parties. Second, such greater reach of AM signals substantially increases the power of our identification strategy, relative to FM and television signals. Although FM radio and television stations are more numerous, they emit weaker signals that are substantially less likely to travel across state borders, which decreases our sample. Nevertheless, our robustness checks in what follows show qualitatively similar results for ads on FM radio and television. 4.1.3. Precinct-Level Variables We also collected precinct-level data to test the heterogeneous effects predicted by the model. To examine Hypothesis 2, we measure local socioeconomic development, as a proxy for voter knowledge of politics (see panel A in Table A.2 of the Online Appendix), using 5 variables: 2006 electorate density; the proportion of the precinct population that has non-dirt floors, running electricity, running water, a toilet, and drainage; the employment rate; the literate proportion of the population aged above 15; and the share of the population aged above 15 that completed primary school.31 Given the strong correlation between these variables, we combine them by taking the first factor from a factor analysis.32 We refer to this standardized variable as “basic development”. To examine Hypothesis 3, we use the (lagged) effective number of political parties by vote share (ENPV) at the precinct level as a proxy for political competition, and thus other electoral strategies that might lead to more information about party policies (see panel B in Table A.2 of the Online Appendix). One effective party represents complete local dominance by a single party, whereas larger values represent greater political competition.33 To ensure that competition is not affected by campaign advertising during or following the 2009 or 2012 elections, we calculated ENPV using the vote share of every party that stood in each precinct in the 2006 legislative election.34 To assess Hypothesis 4, we use an indicator for the 2012 presidential election.35 Finally, to test Hypothesis 5, we define the locally dominant party as the party that received the most votes in the precinct in the 2006 election. As noted previously, we prefer a local measure of dominance to district-level incumbency because federal deputies serve large districts, whereas local political control, information, and partisan preferences vary substantially within districts. We use linear and quadratic terms to capture the non-linearity in the locally dominant party’s vote share—which proxies for the extent of local dominance—implied by Hypothesis 5. Moreover, we interact these terms with an indicator for whether the party is itself the largest local party, in order to test for differential responses to campaign ads from locally dominant and non-dominant parties. 4.2. Identification Strategy To address the concern that electoral precincts receiving different campaign advertising distributions differ in other electorally relevant respects, our identification strategy exploits within-neighbor variation in campaign advertising shares. In particular, we compare neighboring electoral precincts that receive a different distribution of campaign advertising because they receive a different mix of radio signals from AM stations based inside and outside the state. Our design thus relies on differences in advertising shares that originate from cross-state spillovers in AM radio coverage.36 Specifically, we focus on “treated” precincts that differ from at least one neighboring “control” precinct in terms of the distribution of campaign advertising that they receive from AM radio stations. To ensure the comparability of media access, we use all neighboring control precincts located within 1 kilometer (km) of a coverage boundary. Since broadcast signal strength decays gradually with distance, the commercial coverage boundary is not a sharp difference between receiving or not receiving a station’s signal.37 Rather, some households beyond the boundary can nonetheless receive signals from the media outlet (perhaps not regularly, or depending on time of day), whereas signal quality may be erratic for some households inside the boundary. Figure 3 illustrates this for two radio stations and two adjacent precincts in Campeche. We thus rely on a measure of exposure rather than consumption (see also Huber and Arceneaux 2007). This is because we cannot accurately measure media station audiences, and the decision to listen to political ads likely correlates with other relevant variables.38 Consequently, by identifying the effect of an increase in the probability of exposure to AM radio signals, we estimate the “intent to treat” effect of campaign advertising. It is nevertheless clear that access translates into ad consumption and recall. Exploiting within-state variation and data from the 2009 CSES post-election survey, columns (1)–(3) of Table 2 demonstrate that the likelihood that a voter recalls a televised ad by a particular party increases with their precinct’s television campaign advertising share for that party.39 Furthermore, columns (4)–(12) show that the probability that a respondent can recall a feature of the PAN’s, the PRD’s, or the PRI’s ad campaign over the course of the campaign is positively and generally significantly correlated with the precinct’s average AM, FM, and television share for that party. This correspondence is especially important for radio stations, given that radio ad consumption could occur as citizens commute to and from work across precincts. Moreover, although such cross-border commuting is common in metropolitan areas, our primary AM advertising sample is predominantly rural, and thus less subject to this concern.40 Table 2. Correlation between campaign advertising and voter television ad recall within last week and campaign recall. Recall PAN television ad Recall PRD television ad Recall PRI television ad Recall PAN campaign content Recall PAN campaign content Recall PAN campaign content Recall PRD campaign content Recall PRD campaign content Recall PRD campaign content Recall PRI campaign content Recall PRI campaign content Recall PRI campaign content (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) PAN AM advertising share 0.454* (0.254) PAN FM advertising share 0.630* (0.348) PAN television advertising share 0.487** 0.454* (0.202) (0.225) PRD AM advertising share 0.216 (0.926) PRD FM advertising share 1.204 (0.816) PRD television advertising share 0.824* 0.592 (0.413) (0.526) PRI AM advertising share 0.356** (0.133) PRI FM advertising share 0.674*** (0.234) PRI television advertising share 0.457 0.445* (0.473) (0.258) Observations 738 739 738 1,189 1,189 1,189 1,189 1,189 1,189 1,188 1,188 1,188 Outcome mean 0.79 0.70 0.77 0.43 0.43 0.43 0.46 0.46 0.46 0.42 0.42 0.42 Outcome standard deviation 0.41 0.46 0.42 0.50 0.50 0.50 0.50 0.50 0.50 0.49 0.49 0.49 Advertising share mean 0.26 0.15 0.19 0.28 0.27 0.26 0.16 0.15 0.15 0.20 0.19 0.18 Advertising share standard deviation 0.09 0.06 0.07 0.05 0.08 0.10 0.03 0.05 0.06 0.04 0.06 0.07 Recall PAN television ad Recall PRD television ad Recall PRI television ad Recall PAN campaign content Recall PAN campaign content Recall PAN campaign content Recall PRD campaign content Recall PRD campaign content Recall PRD campaign content Recall PRI campaign content Recall PRI campaign content Recall PRI campaign content (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) PAN AM advertising share 0.454* (0.254) PAN FM advertising share 0.630* (0.348) PAN television advertising share 0.487** 0.454* (0.202) (0.225) PRD AM advertising share 0.216 (0.926) PRD FM advertising share 1.204 (0.816) PRD television advertising share 0.824* 0.592 (0.413) (0.526) PRI AM advertising share 0.356** (0.133) PRI FM advertising share 0.674*** (0.234) PRI television advertising share 0.457 0.445* (0.473) (0.258) Observations 738 739 738 1,189 1,189 1,189 1,189 1,189 1,189 1,188 1,188 1,188 Outcome mean 0.79 0.70 0.77 0.43 0.43 0.43 0.46 0.46 0.46 0.42 0.42 0.42 Outcome standard deviation 0.41 0.46 0.42 0.50 0.50 0.50 0.50 0.50 0.50 0.49 0.49 0.49 Advertising share mean 0.26 0.15 0.19 0.28 0.27 0.26 0.16 0.15 0.15 0.20 0.19 0.18 Advertising share standard deviation 0.09 0.06 0.07 0.05 0.08 0.10 0.03 0.05 0.06 0.04 0.06 0.07 Notes: All data are from the 2009 Comparative Study of Electoral Systems survey. Recall television ad is an indicator coded 1 for respondents that recall having seen a campaign ad from the PAN/PRD/PRI on television in the past week. Recall campaign content is an indicator coded 1 for respondents that recall a feature of the PAN/PRD/PRI’s ad campaign on either radio or television over the course of the campaign. Do not know and did not answer were coded as 0. All specifications are estimated using OLS and include state fixed effects. Standard errors are clustered by state; our sample contains 28 clusters. *p < 0.1; **p < 0.05; ***p < 0.01. View Large Table 2. Correlation between campaign advertising and voter television ad recall within last week and campaign recall. Recall PAN television ad Recall PRD television ad Recall PRI television ad Recall PAN campaign content Recall PAN campaign content Recall PAN campaign content Recall PRD campaign content Recall PRD campaign content Recall PRD campaign content Recall PRI campaign content Recall PRI campaign content Recall PRI campaign content (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) PAN AM advertising share 0.454* (0.254) PAN FM advertising share 0.630* (0.348) PAN television advertising share 0.487** 0.454* (0.202) (0.225) PRD AM advertising share 0.216 (0.926) PRD FM advertising share 1.204 (0.816) PRD television advertising share 0.824* 0.592 (0.413) (0.526) PRI AM advertising share 0.356** (0.133) PRI FM advertising share 0.674*** (0.234) PRI television advertising share 0.457 0.445* (0.473) (0.258) Observations 738 739 738 1,189 1,189 1,189 1,189 1,189 1,189 1,188 1,188 1,188 Outcome mean 0.79 0.70 0.77 0.43 0.43 0.43 0.46 0.46 0.46 0.42 0.42 0.42 Outcome standard deviation 0.41 0.46 0.42 0.50 0.50 0.50 0.50 0.50 0.50 0.49 0.49 0.49 Advertising share mean 0.26 0.15 0.19 0.28 0.27 0.26 0.16 0.15 0.15 0.20 0.19 0.18 Advertising share standard deviation 0.09 0.06 0.07 0.05 0.08 0.10 0.03 0.05 0.06 0.04 0.06 0.07 Recall PAN television ad Recall PRD television ad Recall PRI television ad Recall PAN campaign content Recall PAN campaign content Recall PAN campaign content Recall PRD campaign content Recall PRD campaign content Recall PRD campaign content Recall PRI campaign content Recall PRI campaign content Recall PRI campaign content (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) PAN AM advertising share 0.454* (0.254) PAN FM advertising share 0.630* (0.348) PAN television advertising share 0.487** 0.454* (0.202) (0.225) PRD AM advertising share 0.216 (0.926) PRD FM advertising share 1.204 (0.816) PRD television advertising share 0.824* 0.592 (0.413) (0.526) PRI AM advertising share 0.356** (0.133) PRI FM advertising share 0.674*** (0.234) PRI television advertising share 0.457 0.445* (0.473) (0.258) Observations 738 739 738 1,189 1,189 1,189 1,189 1,189 1,189 1,188 1,188 1,188 Outcome mean 0.79 0.70 0.77 0.43 0.43 0.43 0.46 0.46 0.46 0.42 0.42 0.42 Outcome standard deviation 0.41 0.46 0.42 0.50 0.50 0.50 0.50 0.50 0.50 0.49 0.49 0.49 Advertising share mean 0.26 0.15 0.19 0.28 0.27 0.26 0.16 0.15 0.15 0.20 0.19 0.18 Advertising share standard deviation 0.09 0.06 0.07 0.05 0.08 0.10 0.03 0.05 0.06 0.04 0.06 0.07 Notes: All data are from the 2009 Comparative Study of Electoral Systems survey. Recall television ad is an indicator coded 1 for respondents that recall having seen a campaign ad from the PAN/PRD/PRI on television in the past week. Recall campaign content is an indicator coded 1 for respondents that recall a feature of the PAN/PRD/PRI’s ad campaign on either radio or television over the course of the campaign. Do not know and did not answer were coded as 0. All specifications are estimated using OLS and include state fixed effects. Standard errors are clustered by state; our sample contains 28 clusters. *p < 0.1; **p < 0.05; ***p < 0.01. View Large Pooling across the 2009 and 2012 elections, our design yields a total of 31,969 neighbor-year groups containing a single “treated” unit and up to 23 neighboring “control” units. This produced 146,140 observations in total, whereas Figure 1 shades in gray the 16,239 unique electoral precincts included in our sample. The range of PAN, PRD, and PRI AM advertising shares are respectively 21%–35%, 9%–20%, and 19%–35%. Unsurprisingly, this sample is clustered around the borders of states holding concurrent state-level elections. Accordingly, the summary statistics in Table 3 show that the electoral precincts in our sample are more rural and less economically developed than the national average, as well as the analogous samples based on differences in FM radio and television ad distributions. As noted previously, we expect the effect of campaign advertising in the predominantly urban areas comprising the smaller FM and television samples to be lower than in the more rural AM sample where prior exposure to the PAN and the PRD is lower. Figure 1. View largeDownload slide AM radio neighboring precinct sample used in our main analysis. Figure 1. View largeDownload slide AM radio neighboring precinct sample used in our main analysis. Table 3. Comparison of neighboring precinct samples and population summary statistics. AM radio neighboring precinct sample FM radio neighboring precinct sample Television neighboring precinct sample National population (2009 and 2012) Obs. Mean Std. dev. Obs. Mean Std. dev. Obs. Mean Std. dev. Obs. Mean Std. dev. Dependent variables PAN vote share 146,140 0.272 0.158 60,142 0.257 0.149 53,892 0.261 0.148 131,346 0.264 0.144 PRD vote share 146,140 0.160 0.143 60,142 0.162 0.122 53,892 0.170 0.124 131,346 0.164 0.147 PRI vote share 146,140 0.366 0.140 60,142 0.370 0.128 53,892 0.350 0.128 131,346 0.362 0.142 Treatment variables PAN advertising share 146,140 0.277 0.024 60,142 0.274 0.024 53,892 0.274 0.024 131,369 0.272 0.040 PRD advertising share 146,140 0.148 0.033 60,142 0.154 0.034 53,892 0.153 0.032 131,369 0.145 0.037 PRI advertising share 146,140 0.261 0.059 60,142 0.253 0.057 53,892 0.252 0.056 131,369 0.258 0.069 Covariates Largest party vote share (2006) 146,140 0.475 0.109 60,142 0.457 0.105 53,892 0.458 0.104 128,406 0.482 0.103 PAN largest party (2006) 146,140 0.415 0.493 60,142 0.373 0.484 53,892 0.399 0.490 128,406 0.396 0.489 PRD largest party (2006) 146,140 0.272 0.445 60,142 0.348 0.476 53,892 0.396 0.489 128,406 0.310 0.463 PRI largest party (2006) 146,140 0.313 0.464 60,142 0.279 0.449 53,892 0.205 0.404 128,406 0.294 0.456 Electorate density (2006, log) 146,140 4.484 2.263 60,142 5.189 2.314 53,892 5.624 2.330 126,452 6.480 2.840 Share basic necessities 146,140 0.663 0.300 60,142 0.743 0.262 53,892 0.765 0.243 120,184 0.747 0.302 Share illiterate above 15 146,140 0.114 0.082 60,142 0.092 0.069 53,892 0.087 0.068 120,178 0.086 0.090 Share employed 146,140 0.946 0.059 60,142 0.947 0.044 53,892 0.944 0.046 120,176 0.954 0.045 Share primary complete 146,140 0.474 0.093 60,142 0.485 0.092 53,892 0.483 0.096 120,178 0.432 0.114 Effective number of political parties (2006) 146,140 2.726 0.529 60,142 2.881 0.539 53,892 2.887 0.523 128,406 2.729 0.493 2012 presidential election 146,140 0.530 0.499 60,142 0.503 0.500 53,892 0.502 0.500 131,369 0.506 0.500 AM radio neighboring precinct sample FM radio neighboring precinct sample Television neighboring precinct sample National population (2009 and 2012) Obs. Mean Std. dev. Obs. Mean Std. dev. Obs. Mean Std. dev. Obs. Mean Std. dev. Dependent variables PAN vote share 146,140 0.272 0.158 60,142 0.257 0.149 53,892 0.261 0.148 131,346 0.264 0.144 PRD vote share 146,140 0.160 0.143 60,142 0.162 0.122 53,892 0.170 0.124 131,346 0.164 0.147 PRI vote share 146,140 0.366 0.140 60,142 0.370 0.128 53,892 0.350 0.128 131,346 0.362 0.142 Treatment variables PAN advertising share 146,140 0.277 0.024 60,142 0.274 0.024 53,892 0.274 0.024 131,369 0.272 0.040 PRD advertising share 146,140 0.148 0.033 60,142 0.154 0.034 53,892 0.153 0.032 131,369 0.145 0.037 PRI advertising share 146,140 0.261 0.059 60,142 0.253 0.057 53,892 0.252 0.056 131,369 0.258 0.069 Covariates Largest party vote share (2006) 146,140 0.475 0.109 60,142 0.457 0.105 53,892 0.458 0.104 128,406 0.482 0.103 PAN largest party (2006) 146,140 0.415 0.493 60,142 0.373 0.484 53,892 0.399 0.490 128,406 0.396 0.489 PRD largest party (2006) 146,140 0.272 0.445 60,142 0.348 0.476 53,892 0.396 0.489 128,406 0.310 0.463 PRI largest party (2006) 146,140 0.313 0.464 60,142 0.279 0.449 53,892 0.205 0.404 128,406 0.294 0.456 Electorate density (2006, log) 146,140 4.484 2.263 60,142 5.189 2.314 53,892 5.624 2.330 126,452 6.480 2.840 Share basic necessities 146,140 0.663 0.300 60,142 0.743 0.262 53,892 0.765 0.243 120,184 0.747 0.302 Share illiterate above 15 146,140 0.114 0.082 60,142 0.092 0.069 53,892 0.087 0.068 120,178 0.086 0.090 Share employed 146,140 0.946 0.059 60,142 0.947 0.044 53,892 0.944 0.046 120,176 0.954 0.045 Share primary complete 146,140 0.474 0.093 60,142 0.485 0.092 53,892 0.483 0.096 120,178 0.432 0.114 Effective number of political parties (2006) 146,140 2.726 0.529 60,142 2.881 0.539 53,892 2.887 0.523 128,406 2.729 0.493 2012 presidential election 146,140 0.530 0.499 60,142 0.503 0.500 53,892 0.502 0.500 131,369 0.506 0.500 Notes: Summary statistics are for the AM radio, FM radio, and television neighboring precinct samples (all neighboring control precincts within 1 km of a coverage boundary) and full 2009 and 2012 national population of electoral precincts. Party advertising shares in the national population sample are for AM radio, and are coded as zero for the precincts not covered by a media outlet. View Large Table 3. Comparison of neighboring precinct samples and population summary statistics. AM radio neighboring precinct sample FM radio neighboring precinct sample Television neighboring precinct sample National population (2009 and 2012) Obs. Mean Std. dev. Obs. Mean Std. dev. Obs. Mean Std. dev. Obs. Mean Std. dev. Dependent variables PAN vote share 146,140 0.272 0.158 60,142 0.257 0.149 53,892 0.261 0.148 131,346 0.264 0.144 PRD vote share 146,140 0.160 0.143 60,142 0.162 0.122 53,892 0.170 0.124 131,346 0.164 0.147 PRI vote share 146,140 0.366 0.140 60,142 0.370 0.128 53,892 0.350 0.128 131,346 0.362 0.142 Treatment variables PAN advertising share 146,140 0.277 0.024 60,142 0.274 0.024 53,892 0.274 0.024 131,369 0.272 0.040 PRD advertising share 146,140 0.148 0.033 60,142 0.154 0.034 53,892 0.153 0.032 131,369 0.145 0.037 PRI advertising share 146,140 0.261 0.059 60,142 0.253 0.057 53,892 0.252 0.056 131,369 0.258 0.069 Covariates Largest party vote share (2006) 146,140 0.475 0.109 60,142 0.457 0.105 53,892 0.458 0.104 128,406 0.482 0.103 PAN largest party (2006) 146,140 0.415 0.493 60,142 0.373 0.484 53,892 0.399 0.490 128,406 0.396 0.489 PRD largest party (2006) 146,140 0.272 0.445 60,142 0.348 0.476 53,892 0.396 0.489 128,406 0.310 0.463 PRI largest party (2006) 146,140 0.313 0.464 60,142 0.279 0.449 53,892 0.205 0.404 128,406 0.294 0.456 Electorate density (2006, log) 146,140 4.484 2.263 60,142 5.189 2.314 53,892 5.624 2.330 126,452 6.480 2.840 Share basic necessities 146,140 0.663 0.300 60,142 0.743 0.262 53,892 0.765 0.243 120,184 0.747 0.302 Share illiterate above 15 146,140 0.114 0.082 60,142 0.092 0.069 53,892 0.087 0.068 120,178 0.086 0.090 Share employed 146,140 0.946 0.059 60,142 0.947 0.044 53,892 0.944 0.046 120,176 0.954 0.045 Share primary complete 146,140 0.474 0.093 60,142 0.485 0.092 53,892 0.483 0.096 120,178 0.432 0.114 Effective number of political parties (2006) 146,140 2.726 0.529 60,142 2.881 0.539 53,892 2.887 0.523 128,406 2.729 0.493 2012 presidential election 146,140 0.530 0.499 60,142 0.503 0.500 53,892 0.502 0.500 131,369 0.506 0.500 AM radio neighboring precinct sample FM radio neighboring precinct sample Television neighboring precinct sample National population (2009 and 2012) Obs. Mean Std. dev. Obs. Mean Std. dev. Obs. Mean Std. dev. Obs. Mean Std. dev. Dependent variables PAN vote share 146,140 0.272 0.158 60,142 0.257 0.149 53,892 0.261 0.148 131,346 0.264 0.144 PRD vote share 146,140 0.160 0.143 60,142 0.162 0.122 53,892 0.170 0.124 131,346 0.164 0.147 PRI vote share 146,140 0.366 0.140 60,142 0.370 0.128 53,892 0.350 0.128 131,346 0.362 0.142 Treatment variables PAN advertising share 146,140 0.277 0.024 60,142 0.274 0.024 53,892 0.274 0.024 131,369 0.272 0.040 PRD advertising share 146,140 0.148 0.033 60,142 0.154 0.034 53,892 0.153 0.032 131,369 0.145 0.037 PRI advertising share 146,140 0.261 0.059 60,142 0.253 0.057 53,892 0.252 0.056 131,369 0.258 0.069 Covariates Largest party vote share (2006) 146,140 0.475 0.109 60,142 0.457 0.105 53,892 0.458 0.104 128,406 0.482 0.103 PAN largest party (2006) 146,140 0.415 0.493 60,142 0.373 0.484 53,892 0.399 0.490 128,406 0.396 0.489 PRD largest party (2006) 146,140 0.272 0.445 60,142 0.348 0.476 53,892 0.396 0.489 128,406 0.310 0.463 PRI largest party (2006) 146,140 0.313 0.464 60,142 0.279 0.449 53,892 0.205 0.404 128,406 0.294 0.456 Electorate density (2006, log) 146,140 4.484 2.263 60,142 5.189 2.314 53,892 5.624 2.330 126,452 6.480 2.840 Share basic necessities 146,140 0.663 0.300 60,142 0.743 0.262 53,892 0.765 0.243 120,184 0.747 0.302 Share illiterate above 15 146,140 0.114 0.082 60,142 0.092 0.069 53,892 0.087 0.068 120,178 0.086 0.090 Share employed 146,140 0.946 0.059 60,142 0.947 0.044 53,892 0.944 0.046 120,176 0.954 0.045 Share primary complete 146,140 0.474 0.093 60,142 0.485 0.092 53,892 0.483 0.096 120,178 0.432 0.114 Effective number of political parties (2006) 146,140 2.726 0.529 60,142 2.881 0.539 53,892 2.887 0.523 128,406 2.729 0.493 2012 presidential election 146,140 0.530 0.499 60,142 0.503 0.500 53,892 0.502 0.500 131,369 0.506 0.500 Notes: Summary statistics are for the AM radio, FM radio, and television neighboring precinct samples (all neighboring control precincts within 1 km of a coverage boundary) and full 2009 and 2012 national population of electoral precincts. Party advertising shares in the national population sample are for AM radio, and are coded as zero for the precincts not covered by a media outlet. View Large The key identifying assumption is that neighboring precincts differ only in their AM radio campaign advertising shares. There are good reasons to believe this assumption. First, by restricting attention to within-neighbor comparisons, variation in access to radio signals is in large part determined by fixed signal impediments such as terrain and salt water that inhibit or enhance ground-level electrical conductivity (see Strömberg 2004). Second, given that out-of-state AM radio stations are unlikely to specifically target audiences at the extremities of their coverage area, both because such audiences represent a small share of their potential listenership and because they lack the technology to precisely differentiate precincts,41 the direction and reach of cross-state spillovers are unlikely to be correlated with precinct characteristics. Third, if voters choose where to live according to media availability, they would likely choose a location much closer to the antennae, rather than near the commercial quality coverage boundary where high-quality signal coverage cannot be guaranteed. The balance tests discussed in what follows support this identification assumption. 4.2.1. Estimation Provided that differences in campaign advertising originating from cross-state spillovers in AM signals occur effectively randomly, we can estimate the average effect of exposure to campaign advertising from each political party using the following OLS regression: \begin{eqnarray} \mathit {vote {\,\,} share}_{ijt} = \beta {\,\,} \mathit {advertising {\,\,} share}_{ijt} + \mu _{mt} + \varepsilon _{ijt}, \end{eqnarray} (4) where |$\mathit {vote {\,\,} share}_{ijt}\$| is the vote share of party i ∈ {PAN, PRD, PRI} in precinct j at election t ∈ {2009, 2012}. Since our treatment is a party’s advertising share, equation (4) identifies the effect of greater exposure to a party’s advertising relative to a commensurate decline among all other parties.42 We include neighbor group-year fixed effects, μmt, to ensure that our estimates are only identified out of differences within neighboring precincts at a given election. In all specifications, we weight by the inverse of the number of precincts per neighbor group to ensure that each group is weighted equally.43 Standard errors are clustered by state throughout.44 To examine the heterogeneous effects of media conditional on Xijt, we estimate: \begin{eqnarray} \mathit {vote {\,\,} share}_{ijt} &=& \beta {\,\,} \mathit {advertising {\,\,} share}_{ijt} + X_{ijt} {^{\prime }} \gamma \nonumber \\ &&+\, (\mathit {advertising {\,\,} share}_{ijt} \times X_{ijt}) {^{\prime }} \delta + \mu _{mt} + \varepsilon _{ijt}. \end{eqnarray} (5) We test Hypothesis 2 by interacting a party’s campaign advertising share with basic development, Hypothesis 3 by interacting the advertising share with the ENPV at the 2006 legislative election, Hypothesis 4 using an interaction for the 2012 election, and Hypothesis 5 by interacting the advertising share with quadratic terms in the vote share of the largest party in the precinct in 2006 and an indicator for whether party i was the party with the largest local vote share. 4.2.2. Balance on Demographic, Economic, and Political Covariates The key concern for designs exploiting differences between neighboring locations is sorting. The previous discussion argued that neither strategic sorting (on the part of either voters or radio station owners) nor incidental sorting are plausible in this case. Supporting this claim empirically, Table A.3 in the Online Appendix demonstrates that the PAN’s, the PRD’s, and the PRI’s AM campaign advertising shares are each well-balanced across 29 potentially confounding demographic, economic, and political variables; 9 of 87 regressions yielded coefficients significant at the 10% level.45 These tests lend credibility to our design generating exogenous variation in campaign advertising shares. A variety of robustness checks in what follows further reinforce this claim. 5. Results We now test the implications of our theoretical model. We find that campaign advertising was effective at winning votes for the PAN and the PRD. Consistent with the model, advertising’s effects were greatest in less developed and less competitive precincts. Furthermore, where the PAN and the PRD were not locally dominant, the effect of ads increased non-linearly with the vote share of the locally dominant party. However, we find no evidence that PRI advertising was effective. 5.1. Average Effects of AM Radio Campaign Advertising on Party Vote Share Table 4 reports the average and heterogeneous effects of campaign advertising on AM radio. Respectively, the dependent variable in panels A, B, and C are the precinct-level vote shares of the PAN, the PRD, and the PRI. As noted previously, all estimates of equations (4) and (5) include all possible neighboring precincts located within 1 km of an AM coverage boundary. To save space, lower order interactions terms are omitted from the tables. Table 4. Effect of AM radio campaign advertising on PAN, PRD, and PRI vote share. Party vote share (1) (2) (3) (4) (5) (6) Panel A: PAN vote share PAN advertising share 1.224*** 1.076** 5.006*** 1.542*** −1.314 4.654*** (0.346) (0.400) (0.944) (0.496) (0.791) (1.451) ×Basic development (factor) −0.249** −0.136 (0.116) (0.098) ×Effective number of political parties (2006) −1.548*** −1.145*** (0.396) (0.347) ×2012 presidential election −0.581 −0.614 (0.513) (0.416) ×Largest vote share 9.852*** 2.842 (3.061) (2.855) ×Largest party vote share (2006) squared −9.963*** −7.456*** (2.872) (2.565) ×PAN largest party (2006) 2.772** 2.510** (1.134) (1.140) ×Largest party vote share (2006) −13.153*** −11.714**  × PAN largest party (2006) (4.654) (4.654) ×Largest party vote share (2006) squared 13.967*** 12.382***  × PAN largest party (2006) (4.507) (4.458) Panel B: PRD vote share PRD advertising share 0.702 0.603 1.592*** 1.266*** −0.689 3.444*** (0.424) (0.462) (0.561) (0.362) (0.452) (0.848) ×Basic development (factor) −0.139** −0.099** (0.053) (0.047) ×Effective number of political parties (2006) −0.369** −0.748*** (0.144) (0.187) ×2012 presidential election −0.845 −0.757* (0.560) (0.439) ×Largest party vote share (2006) 5.030*** 0.729 (1.130) (1.041) ×Largest party vote share (2006) squared −4.492*** −3.269*** (0.973) (0.824) ×PRD largest party (2006) −0.110 −0.133 (1.026) (1.015) ×Largest party vote share (2006) 0.191 0.291  × PRD largest party (2006) (4.484) (4.460) ×Largest party vote share (2006) squared −0.718 −0.795  × PRD largest party (2006) (4.698) (4.689) Panel C: PRI vote share PRI advertising share −0.257 −0.247 −0.292 −0.549 −0.041 −0.467 (0.295) (0.297) (0.318) (0.651) (0.356) (0.716) ×Basic development (factor) −0.030 −0.064 (0.042) (0.039) ×Effective number of political parties (2006) 0.015 0.044 (0.055) (0.070) ×2012 presidential election 0.516 0.390 (0.697) (0.728) ×Largest party vote share (2006) −0.190 −0.089 (1.059) (1.044) ×Largest party vote share (2006) squared 0.467 0.533 (1.061) (1.066) ×PRI largest party (2006) −0.215 −0.255 (0.362) (0.365) ×Largest party vote share (2006) 1.049 1.210  × PRI largest party (2006) (1.569) (1.576) ×Largest party vote share (2006) squared −1.751 −1.925  × PRI largest party (2006) (1.628) (1.626) Party vote share (1) (2) (3) (4) (5) (6) Panel A: PAN vote share PAN advertising share 1.224*** 1.076** 5.006*** 1.542*** −1.314 4.654*** (0.346) (0.400) (0.944) (0.496) (0.791) (1.451) ×Basic development (factor) −0.249** −0.136 (0.116) (0.098) ×Effective number of political parties (2006) −1.548*** −1.145*** (0.396) (0.347) ×2012 presidential election −0.581 −0.614 (0.513) (0.416) ×Largest vote share 9.852*** 2.842 (3.061) (2.855) ×Largest party vote share (2006) squared −9.963*** −7.456*** (2.872) (2.565) ×PAN largest party (2006) 2.772** 2.510** (1.134) (1.140) ×Largest party vote share (2006) −13.153*** −11.714**  × PAN largest party (2006) (4.654) (4.654) ×Largest party vote share (2006) squared 13.967*** 12.382***  × PAN largest party (2006) (4.507) (4.458) Panel B: PRD vote share PRD advertising share 0.702 0.603 1.592*** 1.266*** −0.689 3.444*** (0.424) (0.462) (0.561) (0.362) (0.452) (0.848) ×Basic development (factor) −0.139** −0.099** (0.053) (0.047) ×Effective number of political parties (2006) −0.369** −0.748*** (0.144) (0.187) ×2012 presidential election −0.845 −0.757* (0.560) (0.439) ×Largest party vote share (2006) 5.030*** 0.729 (1.130) (1.041) ×Largest party vote share (2006) squared −4.492*** −3.269*** (0.973) (0.824) ×PRD largest party (2006) −0.110 −0.133 (1.026) (1.015) ×Largest party vote share (2006) 0.191 0.291  × PRD largest party (2006) (4.484) (4.460) ×Largest party vote share (2006) squared −0.718 −0.795  × PRD largest party (2006) (4.698) (4.689) Panel C: PRI vote share PRI advertising share −0.257 −0.247 −0.292 −0.549 −0.041 −0.467 (0.295) (0.297) (0.318) (0.651) (0.356) (0.716) ×Basic development (factor) −0.030 −0.064 (0.042) (0.039) ×Effective number of political parties (2006) 0.015 0.044 (0.055) (0.070) ×2012 presidential election 0.516 0.390 (0.697) (0.728) ×Largest party vote share (2006) −0.190 −0.089 (1.059) (1.044) ×Largest party vote share (2006) squared 0.467 0.533 (1.061) (1.066) ×PRI largest party (2006) −0.215 −0.255 (0.362) (0.365) ×Largest party vote share (2006) 1.049 1.210  × PRI largest party (2006) (1.569) (1.576) ×Largest party vote share (2006) squared −1.751 −1.925  × PRI largest party (2006) (1.628) (1.626) Notes: All specifications include neighbor group-year fixed effects, all neighboring precincts within 1 km of a coverage boundary, and weight by the inverse of the number of precincts per neighbor–year grouping. Columns (2)–(6) include interaction terms (denoted by “× ...”) between the party’s advertising share and covariates; lower-order interaction terms are omitted. The basic development factor variable (see Online Appendix for construction) has mean zero and a standard deviation of one, whereas the effective number of political parties (calculated using 2006 vote share) ranges from 1 to 4.6, and largest party vote share (2006) ranges from 0.13 to 0.99. Further summary statistics are in Table 3. All specifications include 146,140 observations. Standard errors are clustered by state; our sample contains 30 clusters. *p < 0.1; **p < 0.05; ***p < 0.01. View Large Table 4. Effect of AM radio campaign advertising on PAN, PRD, and PRI vote share. Party vote share (1) (2) (3) (4) (5) (6) Panel A: PAN vote share PAN advertising share 1.224*** 1.076** 5.006*** 1.542*** −1.314 4.654*** (0.346) (0.400) (0.944) (0.496) (0.791) (1.451) ×Basic development (factor) −0.249** −0.136 (0.116) (0.098) ×Effective number of political parties (2006) −1.548*** −1.145*** (0.396) (0.347) ×2012 presidential election −0.581 −0.614 (0.513) (0.416) ×Largest vote share 9.852*** 2.842 (3.061) (2.855) ×Largest party vote share (2006) squared −9.963*** −7.456*** (2.872) (2.565) ×PAN largest party (2006) 2.772** 2.510** (1.134) (1.140) ×Largest party vote share (2006) −13.153*** −11.714**  × PAN largest party (2006) (4.654) (4.654) ×Largest party vote share (2006) squared 13.967*** 12.382***  × PAN largest party (2006) (4.507) (4.458) Panel B: PRD vote share PRD advertising share 0.702 0.603 1.592*** 1.266*** −0.689 3.444*** (0.424) (0.462) (0.561) (0.362) (0.452) (0.848) ×Basic development (factor) −0.139** −0.099** (0.053) (0.047) ×Effective number of political parties (2006) −0.369** −0.748*** (0.144) (0.187) ×2012 presidential election −0.845 −0.757* (0.560) (0.439) ×Largest party vote share (2006) 5.030*** 0.729 (1.130) (1.041) ×Largest party vote share (2006) squared −4.492*** −3.269*** (0.973) (0.824) ×PRD largest party (2006) −0.110 −0.133 (1.026) (1.015) ×Largest party vote share (2006) 0.191 0.291  × PRD largest party (2006) (4.484) (4.460) ×Largest party vote share (2006) squared −0.718 −0.795  × PRD largest party (2006) (4.698) (4.689) Panel C: PRI vote share PRI advertising share −0.257 −0.247 −0.292 −0.549 −0.041 −0.467 (0.295) (0.297) (0.318) (0.651) (0.356) (0.716) ×Basic development (factor) −0.030 −0.064 (0.042) (0.039) ×Effective number of political parties (2006) 0.015 0.044 (0.055) (0.070) ×2012 presidential election 0.516 0.390 (0.697) (0.728) ×Largest party vote share (2006) −0.190 −0.089 (1.059) (1.044) ×Largest party vote share (2006) squared 0.467 0.533 (1.061) (1.066) ×PRI largest party (2006) −0.215 −0.255 (0.362) (0.365) ×Largest party vote share (2006) 1.049 1.210  × PRI largest party (2006) (1.569) (1.576) ×Largest party vote share (2006) squared −1.751 −1.925  × PRI largest party (2006) (1.628) (1.626) Party vote share (1) (2) (3) (4) (5) (6) Panel A: PAN vote share PAN advertising share 1.224*** 1.076** 5.006*** 1.542*** −1.314 4.654*** (0.346) (0.400) (0.944) (0.496) (0.791) (1.451) ×Basic development (factor) −0.249** −0.136 (0.116) (0.098) ×Effective number of political parties (2006) −1.548*** −1.145*** (0.396) (0.347) ×2012 presidential election −0.581 −0.614 (0.513) (0.416) ×Largest vote share 9.852*** 2.842 (3.061) (2.855) ×Largest party vote share (2006) squared −9.963*** −7.456*** (2.872) (2.565) ×PAN largest party (2006) 2.772** 2.510** (1.134) (1.140) ×Largest party vote share (2006) −13.153*** −11.714**  × PAN largest party (2006) (4.654) (4.654) ×Largest party vote share (2006) squared 13.967*** 12.382***  × PAN largest party (2006) (4.507) (4.458) Panel B: PRD vote share PRD advertising share 0.702 0.603 1.592*** 1.266*** −0.689 3.444*** (0.424) (0.462) (0.561) (0.362) (0.452) (0.848) ×Basic development (factor) −0.139** −0.099** (0.053) (0.047) ×Effective number of political parties (2006) −0.369** −0.748*** (0.144) (0.187) ×2012 presidential election −0.845 −0.757* (0.560) (0.439) ×Largest party vote share (2006) 5.030*** 0.729 (1.130) (1.041) ×Largest party vote share (2006) squared −4.492*** −3.269*** (0.973) (0.824) ×PRD largest party (2006) −0.110 −0.133 (1.026) (1.015) ×Largest party vote share (2006) 0.191 0.291  × PRD largest party (2006) (4.484) (4.460) ×Largest party vote share (2006) squared −0.718 −0.795  × PRD largest party (2006) (4.698) (4.689) Panel C: PRI vote share PRI advertising share −0.257 −0.247 −0.292 −0.549 −0.041 −0.467 (0.295) (0.297) (0.318) (0.651) (0.356) (0.716) ×Basic development (factor) −0.030 −0.064 (0.042) (0.039) ×Effective number of political parties (2006) 0.015 0.044 (0.055) (0.070) ×2012 presidential election 0.516 0.390 (0.697) (0.728) ×Largest party vote share (2006) −0.190 −0.089 (1.059) (1.044) ×Largest party vote share (2006) squared 0.467 0.533 (1.061) (1.066) ×PRI largest party (2006) −0.215 −0.255 (0.362) (0.365) ×Largest party vote share (2006) 1.049 1.210  × PRI largest party (2006) (1.569) (1.576) ×Largest party vote share (2006) squared −1.751 −1.925  × PRI largest party (2006) (1.628) (1.626) Notes: All specifications include neighbor group-year fixed effects, all neighboring precincts within 1 km of a coverage boundary, and weight by the inverse of the number of precincts per neighbor–year grouping. Columns (2)–(6) include interaction terms (denoted by “× ...”) between the party’s advertising share and covariates; lower-order interaction terms are omitted. The basic development factor variable (see Online Appendix for construction) has mean zero and a standard deviation of one, whereas the effective number of political parties (calculated using 2006 vote share) ranges from 1 to 4.6, and largest party vote share (2006) ranges from 0.13 to 0.99. Further summary statistics are in Table 3. All specifications include 146,140 observations. Standard errors are clustered by state; our sample contains 30 clusters. *p < 0.1; **p < 0.05; ***p < 0.01. View Large Column (1) reports the average effect of campaign advertising, showing significant variation by political party across panels. In panel A, we find that the share of PAN campaign advertising significantly increased the PAN’s vote share. Specifically, a percentage point increase in the PAN’s advertising share increased their vote share by 1.2 percentage points. At least in the context of Mexico’s relatively unconcentrated political ad markets, where no party receives more than 35% of advertising slots in any precinct, this implies a substantial persuasion rate.46 Although no such counterfactual can be approximated, we anticipate that such large effects would diminish at substantially high party ad concentration levels. Alternatively put, a standard deviation increase in campaign advertising corresponded to a 3 percentage point increase in the PAN’s vote share, or a 11% increase in their vote share. For the PAN, we therefore find support for Hypothesis 1—that campaign advertising was effective on average. In panel B, the PRD’s campaign advertising also substantially increased the party’s vote share, but this is less precisely estimated. The positive coefficient indicates that a percentage point increase in advertising translated into a 0.7 percentage point increase in vote share, whereas a standard deviation increase in advertising corresponded to a 2.3 percentage point and 14% increase in their vote share. The relative imprecision reflects the ineffectiveness of PRD ads in 2012: column (4) shows that the effect of PRD ads in 2009 was statistically significant and similar in magnitude to the average effect of PAN ads. These estimates further suggest that political ads can be highly effective in our relatively rural sample, especially from the starting point where no precinct receives more than 20% of their ads from the PRD. There is no evidence in panel C, however, that PRI campaign advertising influenced their vote share. Our estimate of the effect of the PRI’s advertising share is both negative and far from being statistically significant. This suggests that voters held relatively strong priors about the PRI after seven decades in power, especially in the relatively rural sample that we examine here, and may thus have been relatively unaffected by PRI advertising. Our interviews with political strategists also suggested that voter opinions of the PRI were highly polarized. During Chile’s 1988 plebiscite, Boas (2015) similarly finds that opposition advertising was effective whereas pro-Pinochet advertising was not. Finally, Table A.4 in the Online Appendix shows that no party’s campaign advertising significantly affected electoral turnout on average. This implies that changes in party ad shares could have persuaded those that turn out to switch parties, opposition voters to reach a point where they became indifferent and did not vote, and indifferent voters to support the party, or could have demobilized opposition supporters and mobilized own supporters in equal measure. Without individual-level data, we cannot differentiate between these explanations. 5.2. Heterogeneous Effects of AM Radio Campaign Advertising on Party Vote Share We next turn to our interactive specifications in columns (2)–(6) to examine Hypotheses 2–5. Column (6) includes all heterogeneous effects simultaneously, to demonstrate that the individual interaction estimates are not driven by correlations among our interaction variables. Column (2) shows that, consistent with Hypothesis 2, PAN and PRD campaign advertising was significantly more effective at winning votes in the less developed electoral precincts where voters were least politically informed. Specifically, our estimates indicate that a standard deviation increase in the development factor variable reduced the increase in vote share due to every percentage point increase in campaign advertising by 0.25 percentage points for the PAN and 0.14 percentage points for the PRD. In the least developed precincts (with a standardized development score of −4.7), the effects of campaign advertising were substantial, increasing the PAN’s and the PRD’s vote share by 2.2 and 1.3 percentage points, respectively, for each additional percentage point of advertising share. These estimate decline somewhat—especially for the PAN—in column (6), when controlling for our other heterogeneous effects. The PRI’s campaign advertising appears to have been equally ineffective across more and less developed electoral precincts. The results in columns (3) and (4) show that campaign advertising’s weakest effects were also in competitive precincts and elections, where voters likely developed more precise prior beliefs due to other simultaneous form campaigning activity. First, and supporting Hypothesis 3, the large and statistically significant interaction with the ENPV shows that PAN and PRD campaign advertising was most effective in precincts where a small number of parties garnered most of the votes in 2006. The differential is particularly large for PAN advertising, where a percentage point increase in their advertising share increased their vote share by 3.5 percentage points in the least competitive precinct in our sample, and only reached zero in the 20% of precincts with at least 3.2 effective parties. The effect of PRD advertising on the PRD’s vote share, which is 0.2 percentage points lower after a standard deviation increase in political competition, declined 4 times slower with ENPV, but similarly hit zero in the less than 1% of precincts with at least 4.4 effective parties. These effects are robust to the simultaneous inclusion of our other interactions with campaign advertising in column (6), where the PAN’s and the PRD’s coefficients converge to more similar magnitudes. Consistent with the lack of an average effect, we find no difference in the effectiveness of PRI advertising in panel C. Second, providing some support for Hypothesis 4, column (4) shows that AM radio advertising was less effective during the 2012 presidential election than the 2009 legislative election. Neither differential is quite statistically significant. Nevertheless, consistent with the crowding out previous argument of PAN advertising was lower in 2012, although it continued to significantly increase their vote share on average. PRD ads had a large positive effect in 2009, almost on a par with PAN advertising. However, the negative interaction between campaign advertising and the presidential election year indicates that PRD advertising, on average, was ineffective in 2012. This difference becomes statistically significant once we control for the other interactions in column (6). The estimates in panel C show that in neither election was the effect of PRI advertising positive. Although the 2009 and 2012 elections potentially differed in other important respects—including the content of the ads, turnout rates, and the presence of presidential candidates—the difference across elections provides suggestive evidence consistent with our theory. The estimates in column (5) show that campaign advertising was most effective for non-dominant parties and where the dominant party had intermediate strength. For both the PAN and the PRD, the coefficients in the second and third rows show that the marginal effect of campaign advertising was initially increasing in the vote share of the locally dominant party, but started to decrease once that dominant party’s vote share reached around 50% of the vote. The final two coefficients in these specifications show that the marginal effect, for any level of the locally dominant party’s vote share, was both lower and its gradient flatter with respect to local dominance when either party was themselves dominant. In the case of the PAN, the coefficients in Table 4 indicate that these differentials are statistically significant. Figure 4 illustrates these non-linear marginal effects graphically, providing support for Hypothesis 5 by demonstrating that PAN and PRD advertising were more effective in precincts dominated by other political parties until the locally dominant party became too strong. To demonstrate that these findings are not driven by imposing a quadratic form, Table A.15 in the Online Appendix reports similar results using a less parametric approach, where indicators are used for each quartile of the dominant party’s vote share. Again, PRI advertising is estimated to have been equally ineffective across all types of precincts. Finally, while clearly an out-of-sample extrapolation, these heterogeneous effects can be used to impute the predicted marginal effects for every precinct in the country. We can thus estimate the average nationwide marginal effect of advertising in 2009 and 2012 for each party. Consistent with the claim that the effects of ads on AM radio estimated in our rural sample were larger than those that we would expect nationwide, the results imply an average marginal effect of 0.96 for a unit increase in PAN advertising in 2009, and 0.34 for 2012. For the PRD, these estimates are 0.86 and 0.10 for 2009 and 2012, respectively. For the PRI, these estimates are −0.34 and 0.05 for 2009 and 2012 respectively. These estimates suggest that campaign advertising could have altered electoral outcomes in districts where the race was close and voters received more or less PAN and PRD advertising because of the 2007 reform. 5.3. Robustness Checks Given that our identification strategy leverages cross-state media spillovers and only exploits variation between comparable neighboring precincts, there are good reasons to be confident in our estimates. Nevertheless, we conduct a variety of checks to ensure that the estimates are robust to potential violations of our identification assumptions and generalize to FM and television advertising. The results of these checks are presented in the Online Appendix. We first employ a permutation test to examine the likelihood that spillovers from other hypothetical state advertising distributions could have produced our results. Since the regulation that determines the distribution of political ads within a state does not vary across the states that are not holding local elections, we only randomly reassign the state-level advertising distribution to each of the AM radio stations in states holding local elections. Based on 100 random reassignments, Table A.7 in the Online Appendix shows the average effects aggregating across these placebo assignments (see Online Appendix for more details). The results consistently reveal smaller and less precise estimates. For the average effects of both PAN and PRD advertising, our actual estimate is larger than any of the 100 placebo estimates. In contrast, our estimate for the PRI falls in the 25th percentile of the distribution of placebo estimates. These results suggest that our findings do not reflect idiosyncrasies in the data that the random reassignment of advertising shares at the state level could have produced. Measurement error in AM radio coverage is another potential concern. Such error occurs where changes in the probability of coverage around the commercial quality boundary are smaller than the IFE maps suggest, and likely results in underestimating the effects of campaign advertising. To check that our findings are not driven by such measurement error, we restrict attention to boundaries originating from lower-powered AM radio signals—for whom coverage is less variable and more accurately measured—by excluding antennae with high-powered outputs: wattages above 10,000.47 Table A.8 of the Online Appendix shows that our point estimates are similar, and the average effect of PRD advertising becomes statistically significant at the 10% level. An alternative check in Table A.9 of the Online Appendix shows that controlling for the interaction between campaign advertising and precinct area—in order to partial out differences in our heterogeneous effects that could simply reflect differential measurement error in signal coverage—similarly does not affect our results. Furthermore, to ensure that our results are not driven by precincts covered by different numbers of media stations, Table A.10 of the Online Appendix demonstrates that the results are robust to the inclusion of fixed effects for the total number of AM radio stations covering an electoral precinct. These fixed effects also address the potential concern that precincts subject to cross-state spillovers could be covered by more AM radio stations, and thus provide voters with more consumption options that generate greater exposure to campaign ads. More generally, we examined the sensitivity of our results to different specification choices. First, Table A.11 of the Online Appendix shows that our average effects are substantively similar when we include the 29 variables used for our balance tests, although the point estimates decline somewhat. Second, we control for the interaction between campaign advertising and each variable in separate regressions. The results, available in our replication code to save space, also show that our main findings are not substantially affected. Third, we examined the sensitivity of our estimates to the choice of maximum distance from the coverage boundary. Tables A.12 and A.13 of the Online Appendix demonstrate that restricting attention to precincts within 0.5 or 5 km of the nearest coverage boundary produced similar results. Finally, our results also generalize to other media formats. Although the smaller FM and television samples differ markedly from our main AM sample, the heterogeneous effects—which are similar to the AM results and generally remain statistically significant in spite of more than halving the sample size—in Tables A.18 and A.19 of the Online Appendix further indicate that campaign advertising was most effective where voters were less informed, political competition was low, and a party was not locally dominant. The only difference is that FM and television ads were not more effective in 2009 than 2012. Consistent with our theory, changes in sample composition ensure that the average effects of campaign advertising were lower in the better informed and more competitive precincts that constitute the FM and television samples. Moreover, we again find that neither FM nor television campaign advertising won votes for the PRI. 5.4. Alternative Interpretations An important consideration is the possibility that our results reflect underlying differences in media content across states, rather than the effects of campaign advertising. For example, AM stations in states with larger distributions of PAN advertising, and thus higher PAN vote shares, may also have more favorably or more frequently covered the PAN in the news. To address such concerns, we employ the 2006 election as a placebo. Using the allocation formula specified by the 2007 reform, we compute the advertising share that each party would have received in 2006 had the reform already been passed. Using the same identification strategy, we compare neighboring precincts that differ in their predicted 2006 advertising distribution.48 Supporting our claim that it is campaign advertising—rather than biases in media content—that affected vote choice, Table 5 shows that the predicted 2006 campaign advertising shares did not systematically affect the 2006 vote share of any party. Column (1) shows that the predicted advertising share did not significantly affect the vote share of any party on average. Columns (2) and (3) further indicate that there is little evidence that the predicted campaign advertising share produced heterogeneous effects akin to those in Table 4. In the case of local dominance, the estimates in column (4) report more similar interactions to our main results. However, closer inspection of the coefficients indicates that the overall point estimate for campaign advertising when the PAN was not locally dominant would never have been positive for any campaign advertising share with support in our sample. Although the placebo slope estimates for the PRD are significant in the same direction, the magnitudes in Table 4 are considerably larger. Table 5. Effect of the AM radio 2006 placebo on PAN, PRD, and PRI vote share. Party vote share (1) (2) (3) (4) (5) Panel A: PAN vote share PAN advertising share 0.074 −0.027 −1.524 2.220 5.860* (0.747) (0.728) (1.423) (3.094) (2.879) ×Basic development (factor) 0.163 −0.230 (0.320) (0.232) ×Effective number of political parties (2003) 0.603 −0.734 (0.515) (0.546) ×Largest party vote share (2003) −10.055 −13.992 (12.208) (10.513) ×Largest party vote share (2003) squared 11.208 12.282 (12.199) (11.650) ×PAN largest party (2003) 0.849 1.264 (2.295) (2.301) ×Largest party vote share (2003) −5.513 −7.122  × PAN largest party (2003) (10.279) (10.304) ×Largest party vote share (2003) squared 3.955 5.432  × PAN largest party (2003) (10.154) (10.297) Panel B: PRD vote share PRD advertising share 0.567 0.529 0.311 −1.075 −1.107 (0.349) (0.344) (0.546) (0.640) (1.418) ×Basic development (factor) 0.086 0.188** (0.089) (0.078) ×Effective number of political parties (2003) 0.137 −0.104 (0.212) (0.268) ×Largest party vote share (2003) 6.996** 8.103*** (2.554) (2.880) ×Largest party vote share (2003) squared −6.915** −8.214*** (2.607) (2.528) ×PRD largest party (2003) 3.878** 3.949** (1.682) (1.710) ×Largest party vote share (2003) −15.173** −15.151**  × PRD largest party (2003) (7.209) (7.271) ×Largest party vote share (2003) squared 10.782 10.569  × PRD largest party (2003) (7.605) (7.582) Panel C: PRI vote share PRI advertising share −0.023 0.136 0.406 −2.910** −4.625*** (0.332) (0.334) (1.382) (1.244) (1.245) ×Basic development (factor) −0.223 −0.034 (0.151) (0.093) ×Effective number of political parties (2003) −0.155 0.230 (0.525) (0.189) ×Largest party vote share (2003) 11.270* 14.569*** (5.901) (5.047) ×Largest party vote share (2003) squared −10.788* −12.849** (5.940) (5.191) ×PRI largest party (2003) 2.299 2.046 (1.623) (1.698) ×Largest party vote share (2003) −9.770 −9.106  × PRI largest party (2003) (6.996) (7.112) ×Largest party vote share (2003) squared 9.658 9.510  × PRI largest party (2003) (6.571) (6.496) Party vote share (1) (2) (3) (4) (5) Panel A: PAN vote share PAN advertising share 0.074 −0.027 −1.524 2.220 5.860* (0.747) (0.728) (1.423) (3.094) (2.879) ×Basic development (factor) 0.163 −0.230 (0.320) (0.232) ×Effective number of political parties (2003) 0.603 −0.734 (0.515) (0.546) ×Largest party vote share (2003) −10.055 −13.992 (12.208) (10.513) ×Largest party vote share (2003) squared 11.208 12.282 (12.199) (11.650) ×PAN largest party (2003) 0.849 1.264 (2.295) (2.301) ×Largest party vote share (2003) −5.513 −7.122  × PAN largest party (2003) (10.279) (10.304) ×Largest party vote share (2003) squared 3.955 5.432  × PAN largest party (2003) (10.154) (10.297) Panel B: PRD vote share PRD advertising share 0.567 0.529 0.311 −1.075 −1.107 (0.349) (0.344) (0.546) (0.640) (1.418) ×Basic development (factor) 0.086 0.188** (0.089) (0.078) ×Effective number of political parties (2003) 0.137 −0.104 (0.212) (0.268) ×Largest party vote share (2003) 6.996** 8.103*** (2.554) (2.880) ×Largest party vote share (2003) squared −6.915** −8.214*** (2.607) (2.528) ×PRD largest party (2003) 3.878** 3.949** (1.682) (1.710) ×Largest party vote share (2003) −15.173** −15.151**  × PRD largest party (2003) (7.209) (7.271) ×Largest party vote share (2003) squared 10.782 10.569  × PRD largest party (2003) (7.605) (7.582) Panel C: PRI vote share PRI advertising share −0.023 0.136 0.406 −2.910** −4.625*** (0.332) (0.334) (1.382) (1.244) (1.245) ×Basic development (factor) −0.223 −0.034 (0.151) (0.093) ×Effective number of political parties (2003) −0.155 0.230 (0.525) (0.189) ×Largest party vote share (2003) 11.270* 14.569*** (5.901) (5.047) ×Largest party vote share (2003) squared −10.788* −12.849** (5.940) (5.191) ×PRI largest party (2003) 2.299 2.046 (1.623) (1.698) ×Largest party vote share (2003) −9.770 −9.106  × PRI largest party (2003) (6.996) (7.112) ×Largest party vote share (2003) squared 9.658 9.510  × PRI largest party (2003) (6.571) (6.496) Notes: All specifications include neighbor group-year fixed effects, all neighboring precincts within 1 km of a coverage boundary, and weight by the inverse of the number of precincts per neighbor-year grouping. Columns (2)–(6) include interaction terms (denoted by “× ...”) between the party’s advertising share and covariates; lower order interaction terms are omitted. The basic development factor variable (see Online Appendix for construction) has mean zero and a standard deviation of one, whereas the effective number of political parties (calculated using 2003 vote share) ranges from 1 to 6.1, and largest party vote share (2003) ranges from 0.13 to 0.99. All specifications include 66,677 observations. Standard errors are clustered by state. *p < 0.1; **p < 0.05; ***p < 0.01. View Large Table 5. Effect of the AM radio 2006 placebo on PAN, PRD, and PRI vote share. Party vote share (1) (2) (3) (4) (5) Panel A: PAN vote share PAN advertising share 0.074 −0.027 −1.524 2.220 5.860* (0.747) (0.728) (1.423) (3.094) (2.879) ×Basic development (factor) 0.163 −0.230 (0.320) (0.232) ×Effective number of political parties (2003) 0.603 −0.734 (0.515) (0.546) ×Largest party vote share (2003) −10.055 −13.992 (12.208) (10.513) ×Largest party vote share (2003) squared 11.208 12.282 (12.199) (11.650) ×PAN largest party (2003) 0.849 1.264 (2.295) (2.301) ×Largest party vote share (2003) −5.513 −7.122  × PAN largest party (2003) (10.279) (10.304) ×Largest party vote share (2003) squared 3.955 5.432  × PAN largest party (2003) (10.154) (10.297) Panel B: PRD vote share PRD advertising share 0.567 0.529 0.311 −1.075 −1.107 (0.349) (0.344) (0.546) (0.640) (1.418) ×Basic development (factor) 0.086 0.188** (0.089) (0.078) ×Effective number of political parties (2003) 0.137 −0.104 (0.212) (0.268) ×Largest party vote share (2003) 6.996** 8.103*** (2.554) (2.880) ×Largest party vote share (2003) squared −6.915** −8.214*** (2.607) (2.528) ×PRD largest party (2003) 3.878** 3.949** (1.682) (1.710) ×Largest party vote share (2003) −15.173** −15.151**  × PRD largest party (2003) (7.209) (7.271) ×Largest party vote share (2003) squared 10.782 10.569  × PRD largest party (2003) (7.605) (7.582) Panel C: PRI vote share PRI advertising share −0.023 0.136 0.406 −2.910** −4.625*** (0.332) (0.334) (1.382) (1.244) (1.245) ×Basic development (factor) −0.223 −0.034 (0.151) (0.093) ×Effective number of political parties (2003) −0.155 0.230 (0.525) (0.189) ×Largest party vote share (2003) 11.270* 14.569*** (5.901) (5.047) ×Largest party vote share (2003) squared −10.788* −12.849** (5.940) (5.191) ×PRI largest party (2003) 2.299 2.046 (1.623) (1.698) ×Largest party vote share (2003) −9.770 −9.106  × PRI largest party (2003) (6.996) (7.112) ×Largest party vote share (2003) squared 9.658 9.510  × PRI largest party (2003) (6.571) (6.496) Party vote share (1) (2) (3) (4) (5) Panel A: PAN vote share PAN advertising share 0.074 −0.027 −1.524 2.220 5.860* (0.747) (0.728) (1.423) (3.094) (2.879) ×Basic development (factor) 0.163 −0.230 (0.320) (0.232) ×Effective number of political parties (2003) 0.603 −0.734 (0.515) (0.546) ×Largest party vote share (2003) −10.055 −13.992 (12.208) (10.513) ×Largest party vote share (2003) squared 11.208 12.282 (12.199) (11.650) ×PAN largest party (2003) 0.849 1.264 (2.295) (2.301) ×Largest party vote share (2003) −5.513 −7.122  × PAN largest party (2003) (10.279) (10.304) ×Largest party vote share (2003) squared 3.955 5.432  × PAN largest party (2003) (10.154) (10.297) Panel B: PRD vote share PRD advertising share 0.567 0.529 0.311 −1.075 −1.107 (0.349) (0.344) (0.546) (0.640) (1.418) ×Basic development (factor) 0.086 0.188** (0.089) (0.078) ×Effective number of political parties (2003) 0.137 −0.104 (0.212) (0.268) ×Largest party vote share (2003) 6.996** 8.103*** (2.554) (2.880) ×Largest party vote share (2003) squared −6.915** −8.214*** (2.607) (2.528) ×PRD largest party (2003) 3.878** 3.949** (1.682) (1.710) ×Largest party vote share (2003) −15.173** −15.151**  × PRD largest party (2003) (7.209) (7.271) ×Largest party vote share (2003) squared 10.782 10.569  × PRD largest party (2003) (7.605) (7.582) Panel C: PRI vote share PRI advertising share −0.023 0.136 0.406 −2.910** −4.625*** (0.332) (0.334) (1.382) (1.244) (1.245) ×Basic development (factor) −0.223 −0.034 (0.151) (0.093) ×Effective number of political parties (2003) −0.155 0.230 (0.525) (0.189) ×Largest party vote share (2003) 11.270* 14.569*** (5.901) (5.047) ×Largest party vote share (2003) squared −10.788* −12.849** (5.940) (5.191) ×PRI largest party (2003) 2.299 2.046 (1.623) (1.698) ×Largest party vote share (2003) −9.770 −9.106  × PRI largest party (2003) (6.996) (7.112) ×Largest party vote share (2003) squared 9.658 9.510  × PRI largest party (2003) (6.571) (6.496) Notes: All specifications include neighbor group-year fixed effects, all neighboring precincts within 1 km of a coverage boundary, and weight by the inverse of the number of precincts per neighbor-year grouping. Columns (2)–(6) include interaction terms (denoted by “× ...”) between the party’s advertising share and covariates; lower order interaction terms are omitted. The basic development factor variable (see Online Appendix for construction) has mean zero and a standard deviation of one, whereas the effective number of political parties (calculated using 2003 vote share) ranges from 1 to 6.1, and largest party vote share (2003) ranges from 0.13 to 0.99. All specifications include 66,677 observations. Standard errors are clustered by state. *p < 0.1; **p < 0.05; ***p < 0.01. View Large A further potential issue with interpreting our findings is that the estimates could also capture the response of political parties to media coverage. However, conversations with a prominent political consultant in Table 4 suggest that parties are either unaware of the cross-state signal spillovers that we exploit, or do not take these spillovers into account when designing their campaign advertising strategies. As highlighted in Figure 2, spillovers in AM radio signals across states are also not straightforward to detect, and are likely to be second-order in determining party strategies. Nevertheless, we ultimately regard the overall effect of access to advertising—which combines the equilibrium behavior of both parties and voters—as the primary estimate of interest for both institutional reformers and parties themselves. Figure 2. View largeDownload slide Commercial quality signal coverage of all AM radio stations (source: IFE). Figure 2. View largeDownload slide Commercial quality signal coverage of all AM radio stations (source: IFE). Figure 3. View largeDownload slide Neighboring electoral precincts that differ in their commercial quality radio signal coverage from out-of-state AM radio stations. Figure 3. View largeDownload slide Neighboring electoral precincts that differ in their commercial quality radio signal coverage from out-of-state AM radio stations. Figure 4. View largeDownload slide Effects of AM campaign advertising by vote share of largest party and local dominance. The figures plot the estimated marginal effect of AM campaign advertising, based on the estimates in Table 2. The figures show that campaign advertising is only effective for non-dominant parties, and particularly so when facing a locally dominant party of intermediate strength. The density of the data is shown in gray along the x axis; less than 1% of our sample lies outside the range depicted on the x axis. The insignificant relationships for the PRI are omitted. Figure 4. View largeDownload slide Effects of AM campaign advertising by vote share of largest party and local dominance. The figures plot the estimated marginal effect of AM campaign advertising, based on the estimates in Table 2. The figures show that campaign advertising is only effective for non-dominant parties, and particularly so when facing a locally dominant party of intermediate strength. The density of the data is shown in gray along the x axis; less than 1% of our sample lies outside the range depicted on the x axis. The insignificant relationships for the PRI are omitted. 6. Conclusion Despite the prevalence of political ads on broadcast media across the world, little is known about the effectiveness of campaign advertising. This is especially true outside of the United States and other developed democracies, and is particularly relevant in contexts where ads may be most effective because one party is dominant. Given that informational advantages are a key feature of dominance, we theorize that campaign advertising is especially effective for non-dominant parties. Our empirical design exploits within-neighboring precinct differences in campaign ad distributions originating from cross-state media coverage spillovers to test the implications of our theoretical argument in the aftermath of a major media regulation reform in Mexico. We find that campaign ads significantly benefited the PAN and the PRD, but had no discernible effect on the PRI’s vote share. Consistent with our model, campaign ads were most effective in less informed electoral precincts with lower levels of competition and intermediate levels of local party dominance. An intriguing implication of our findings is that equalizing campaign advertising opportunities across political parties may be able to support democratic consolidation in two ways. First, greater equality in campaign advertising has the potential to enhance political representation by better matching voter preferences with like-minded parties. In the long term, this could increase support for democracy (e.g., Mattes and Bratton 2007). Second, by increasing the vote share of non-dominant parties in less competitive precincts, greater equality in campaign advertising opportunities can promote multi-party competition and incentives for politicians to cater to the electorate’s preferences in context of initial hegemony. Conversely, as Boas and Hidalgo (2011) show, when increased media access is concentrated among incumbent politicians, cycles of political dominance can instead be perpetuated. Our results thus suggest that recent reforms providing equitable access to election advertising could deepen democracy in parts of the world where electoral competition remains weak. Nevertheless, further work is required to understand exactly how campaign advertising wins votes among the least knowledgeable, and how parties strategically allocate their ads as a consequence. Notes The editor in charge of this paper was Paola Giuliano. Acknowledgments: We thank the editor, 4 anonymous referees, Scott Ashworth, Andy Baker, Taylor Boas, Ernesto Dal Bó, Aditya Dasgupta, Jorge Domínguez, Ruben Enikolopov, Leopoldo Fergusson, Andy Hall, Brian Knight, Chappell Lawson, Devra Moehler, Jonathan Phillips, Maxim Pinkovskiy, Gilles Serra, Edoardo Teso, and participants at the LACEA Annual Meeting 2014, Second Annual Formal Theory and Comparative Politics Conference 2014, APSA Annual Meeting 2014, and Harvard Political Economy Workshop for comments on earlier drafts. We thank Michelle Kuroda, Rohan Pidaparti, Mayaram Quintero, and Rodrigo Salido Moulinié for excellent research assistance. All errors are our own. Footnotes 1 Washington Post, “Mad Money: TV ads in the 2012 presidential campaign”. http://www.washingtonpost.com/wp-srv/special/politics/track-presidential-campaign-ads-2012/. Accessed 21 January 2018. 2 The IFE has since become the National Electoral Institute (INE). 3 Unfortunately, in the absence of extensive ad consumption data, we cannot credibly estimate persuasion rates (see DellaVigna and Gentzkow 2010). 4 For example, see the Ace Project’s map detailing free broadcast allocations across the world here. 5 A constitutional reform in 2014 permitted re-election up to three times for deputies and once for senators elected from the 2018 election onward. 6 Mexico’s major parties often form coalitions for both local and national elections with smaller parties. In 2009, the PRI formed a coalition with PVEM, whereas in 2012 the PRD formed a coalition with the Workers Party (PT) and Citizen’s Movement (MC) for the national legislative elections. 7 The 15 in 2012, shown in Figure 1, were: Campeche, Chiapas, Colima, Distrito Federal, Guanajuato, Guerrero, Jalisco, México, Morelos, Nueva León, Querétaro, San Luis Potosí, Sonora, Tabasco, and Yucatan. Chiapas, Guerrero, Tabasco, and Yucatán did not hold concurrent elections in 2009. 8 These ads are publicly available at http://pautas.ife.org.mx/transparencia/camp. State-level ads were not systematically collected. 9 Campaign advertising could also convey information such as attractiveness, which may be uncorrelated with political attributes, although our empirical analysis focuses primarily on radio rather than television advertising. 10 With the exception of one case (see in what follows), all results apply where F″ < 0 is sufficiently small. 11 In our empirical application, no party’s campaign advertising significantly affects average turnout. An interesting extension could develop a model to also explain heterogeneous effects of campaign advertising on turnout. 12 This constant absolute risk aversion utility function is chosen because of its convenient mathematical properties when taking expectations over normally distributed lotteries. For simplicity, we set the coefficient of risk aversion to unity. 13 Since D’s position is known with certainty, we ignore any signals sent by D. 14 At the cost of mathematical complexity, the model could be extended to include voters updating negatively about N’s policy outcome. However, our main results hold provided that this share is relatively small. 15 Morgenstern and Zechmeister (2001) have shown that risk-aversion was a significant factor in explaining continuing support for the PRI at the 2000 presidential elections. 16 After the end of our sample period, the National Regeneration Movement (MORENA) also became an important electoral player in the 2015 elections. 17 For each party, we define indicators for whether the respondent both knows a given party’s candidate and has an opinion about that candidate. 18 McCann and Lawson (2006) find similar correlations before 2006. 19 Confirming this correlation, panel A of Table A.2 in the Online Appendix shows that our measure of basic local development—defined in what follows—is positively and significantly correlated with the respective probabilities that respondents know of, and have an opinion on, the PAN’s, the PRD’s, and the PRI’s presidential candidates, as well as an index of political knowledge probing a respondent’s knowledge of political institutions. 20 Since impoverished voters are typically also the most susceptible to vote buying (e.g., Stokes 2005), which may reduce the effectiveness of campaign advertising (see Hypothesis 5), which effect dominates is an empirical question. Our empirical analysis also seeks to distinguish these effects empirically by using different proxies and showing that both interactive effects hold simultaneously. 21 Theoretically, campaign advertising could complement other activities. However, it is not clear why complementarities with one party’s advertising should overcome both advertising and non-advertising countervailing forces emanating from other political parties. Furthermore, strategies like vote buying are unlikely to serve as complements since they are designed to overcome political preferences. Ultimately, this is an empirical question. 22 Panel B of Table A.2 in the Online Appendix shows that the effective number of political parties is positively correlated with knowledge of candidates and political institutions. 23 Although this logic follows from the model, we do not provide a formal statement because local dominance is multidimensional in our model. 24 5% of votes were null or not registered, whereas 15% of votes were cast between six small parties. Table A.4 in the Online Appendix shows that turnout is unaffected by campaign advertising. 25 Although we focus on Congressional elections, which allow us to pool results across two elections, the correlation between PAN, PRI, and PRD legislative and presidential vote shares always exceeds 0.91. Table A.15 in the Online Appendix reports similar results for the 2012 presidential election. 26 This data was obtained from IFE using a freedom of information request. 27 AM radio coverage was typically calculated using the Kirke (or equivalent distance) method, which adjusts for local terrain disrupting ground conductivity. Strömberg (2004) shows that ground conductivity is a good predictor of the number of households with radios in the United States in the 1930s. Coverage of FM radio and television stations was calculated similarly. 28 In the United States, it “is recognized as the area in which a reliable signal can be received using an ordinary radio receiver and antenna” (NTIA link). 29 Although we were unable to obtain data for 2012, the number of radio and television stations did not change in any year between 2003 and 2010. 30 Since the uncovered precincts differ systematically, we focus on comparing differences in party campaign advertising shares among precincts receiving AM coverage from at least one radio station. Balance across covariates declines when comparing precincts with and without AM coverage. 31 The first variable was computed from electoral and spatial data from the IFE, and the final 4 variables come from the 2010 Census. 32 In our main sample (see in what follows), the first factor has an eigenvalue of 1.72, whereas the second factor’s eigenvalue is only 0.56. 33 Although most elections are two-party races, smaller parties remain sufficiently large that they should not be ignored. We thus prefer ENPV to measures based on the two largest parties. 34 In our main sample, the correlation between 2006 ENPV and (endogenous) contemporaneous ENPV is 0.50. 35 We obtain essentially identical results when splitting the sample. 36 See also U.S. studies exploiting differences in media market boundaries (e.g., Ansolabehere et al. 2006; Huber and Arceneaux 2007; Snyder and Strömberg 2010); see Enikolopov et al. (2011) for a non-U.S. study adopting a similar approach. 37 Our design differs from geographic regression discontinuity designs in two further respects. First, differences in the number of commercial quality local media signals between neighbors are non-binary because neighbors can differ by more than one media station. Second, the multidimensionality of these differences determining the distribution of campaign advertising does not naturally translate into a continuous forcing variable. 38 Ideally, we could also identify the electoral effect of receiving or consuming an additional media station using instrumental variable techniques. However, in the absence of detailed individual-level variables measuring which radio or television stations voters have access to or actually consume, we cannot estimate an appropriate first stage. 39 Unfortunately, no such data was available for radio stations. However, studies from other contexts also suggest that the volume and breadth of media access translate into the consumption of political information (Barabas and Jerit 2009; Prat and Strömberg 2005). 40 To examine whether television produces larger effects than radio, as previous studies in Mexico comparing FM radio and television have suggested (Larreguy et al. 2017a), we could in principal compare the effects of campaign advertising among neighboring precincts that receive different advertising shares through both radio and television. Unfortunately, the intersection of these 3 samples is too small to allow a meaningful comparison: the AM sample drops by around 91%. 41 The power output in watts for the AM radio stations in our sample are almost exclusively round thousands and divisible by 5,000. 42 Table A.6 in the Online Appendix shows similar results when we also control for the share of ads allocated to other parties on the left, center and right. The controls allow us to examine vote substitutions, and suggest that the PRD benefited from centrist advertising that likely loosened the ties of voters supporting other leftists parties, whereas PAN advertising harmed the PRI. 43 The results are robust to further weighting by the number of registered voters per precinct (see Table A.14 in the Online Appendix). 44 We have 30 clusters because, as Figure 1 shows, no precinct in Durango differed in its ad share from that of its neighbors. 45 Even when the treatment is indeed uncorrelated with 87 independent outcomes, finding 9 or more relationships that are statistically significant at the 10% level occurs around 51% of the time. 46 We do not estimate the persuasion rates proposed by DellaVigna and Gentzkow (2010) because we cannot credibly measure media consumption and because our results primarily reflect intensity of exposure—which may be non-linear—rather than binary exposure. 47 Stations with high wattage (high power) have larger total coverage areas and tend to have wider zones where signal strength is between 50 and 60 dBμ, in which coverage may be spotty or poor but often not zero. 48 Since there is a significant imbalance on the 2003 PAN vote share, we control for this imbalance in all specifications. However, as noted previously, our main results do not suffer from this imbalance and are robust to controlling for pre-treatment vote shares. References Ackerberg Daniel A. ( 2001 ). “Empirically Distinguishing Informative and Prestige Effects of Advertising.” RAND Journal of Economics , 32 , 316 – 333 . Google Scholar Crossref Search ADS Ansolabehere Stephen , Snowberg Erik C. , Snyder, Jr James M. . ( 2006 ). “Television and the Incumbency Advantage in US Elections.” Legislative Studies Quarterly , 31 , 469 – 490 . Google Scholar Crossref Search ADS Barabas Jason , Jerit Jennifer ( 2009 ). “Estimating the Causal Effects of Media Coverage on Policy-Specific Knowledge.” American Journal of Political Science , 53 , 73 – 89 . Google Scholar Crossref Search ADS Baron David P. ( 1994 ). “Electoral Competition with Informed and Uninformed Voters.” American Political Science Review , 88 , 33 – 47 . Google Scholar Crossref Search ADS Boas Taylor ( 2015 ). “Voting for Democracy: Campaign Effects in Chile’s Democratic Transition.” Latin American Politics and Society , 57 , 67 – 90 . Google Scholar Crossref Search ADS Boas Taylor C. , Hidalgo F. Daniel ( 2011 ). “Controlling the Airwaves: Incumbency Advantage and Community Radio in Brazil.” American Journal of Political Science , 55 , 869 – 885 . Google Scholar Crossref Search ADS Chiang Chun-Fang , Knight Brian ( 2011 ). “Media Bias and Influence: Evidence from Newspaper Endorsements.” Review of Economic Studies , 78 , 795 – 820 . Google Scholar Crossref Search ADS Cornelius Wayne A. ( 2004 ). “Mobilized Voting in the 2000 Elections: The Changing Efficacy of Vote Buying and Coercion in Mexican Electoral Politics.” In Mexico’s Pivotal Democratic Elections: Candidates, Voters, and the Presidential Campaign of 2000 , edited by Domínguez Jorge I. , Lawson Chappell . Stanford University Press , pp. 47 – 65 . Da Silveira Bernardo S. , De Mello João M.P. ( 2011 ). “Campaign Advertising and Election Outcomes: Quasi-Natural Experiment Evidence from Gubernatorial Elections in Brazil.” Review of Economic Studies , 78 , 590 – 612 . Google Scholar Crossref Search ADS DellaVigna Stefano , Gentzkow Matthew ( 2010 ). “Persuasion: Empirical Evidence.” Annual Review of Economics , 2 , 643 – 669 . Google Scholar Crossref Search ADS DellaVigna Stefano , Kaplan Ethan ( 2007 ). “The Fox News Effect: Media Bias and Voting.” Quarterly Journal of Economics , 122 , 1187 – 1234 . Google Scholar Crossref Search ADS Djankov Simeon , McLiesh Caralee , Nenova Tatiana , Shleifer Andrei ( 2003 ). “Who Owns the Media?” Journal of Law and Economics , 46 , 341 – 382 . Google Scholar Crossref Search ADS Downs Anthony ( 1957 ). An Economic Theory of Democracy . Addison Wesley, Boston, MA . Durante Ruben , Gutierrez Emilio ( 2014 ). “Political Advertising and Voting Intentions: Evidence from Exogenous Variation in Ads Viewership.” Working paper, Science Po, Paris . Durante Ruben , Knight Brian ( 2012 ). “Partisan Control, Media Bias, and Viewer Responses: Evidence from Berlusconi’s Italy.” Journal of the European Economic Association , 10 , 451 – 481 . Google Scholar Crossref Search ADS Enikolopov Ruben , Petrova Maria , Zhuravskaya Ekaterina ( 2011 ). “Media and Political Persuasion: Evidence from Russia.” American Economic Review , 101 ( 7 ), 3253 – 3285 . Google Scholar Crossref Search ADS Fergusson Leopoldo , Vargas Juan F. , Vela Mauricio ( 2014 ). “Sunlight Disinfects? Free Media in Weak Democracies.” Cede Working Paper 2013-14, Universidad de los Andes, Bogotá . Ferraz Claudio , Finan Frederico ( 2008 ). “Exposing Corrupt Politicians: The Effects of Brazil’s Publicly Released Audits on Electoral Outcomes.” Quarterly Journal of Economics , 123 , 703 – 745 . Google Scholar Crossref Search ADS Fox Jonathan ( 1994 ). “The Difficult Transition from Clientelism to Citizenship: Lessons from Mexico.” World Politics , 46 , 151 – 184 . Google Scholar Crossref Search ADS Gentzkow Matthew , Shapiro Jesse M. , Sinkinson Michael ( 2011 ). “The Effect of Newspaper Entry and Exit on Electoral Politics.” American Economic Review , 101 ( 7 ), 2980 – 3018 . Google Scholar Crossref Search ADS Gerber Alan S. , Gimpel James G. , Green Donald P. , Shaw Daron R. ( 2011 ). “How Large and Long-Lasting are the Persuasive Effects of Televised Campaign Ads? Results from a Randomized Field Experiment.” American Political Science Review , 105 , 135 – 150 . Google Scholar Crossref Search ADS Greene Kenneth F. ( 2007 ). Why Dominant Parties Lose: Mexico’s Democratization in Comparative Perspective . Cambridge University Press . Greene Kenneth F. ( 2011 ). “Campaign Persuasion and Nascent Partisanship in Mexico’s New Democracy.” American Journal of Political Science , 55 , 398 – 416 . Google Scholar Crossref Search ADS Grossman Gene M. , Helpman Elhanan ( 1996 ). “Electoral Competition and Special Interest Politics.” Review of Economic Studies , 63 , 265 – 286 . Google Scholar Crossref Search ADS Hallin Daniel ( 2000 ). “Media, Political Power, and Democratization in Mexico.” In De-Westernizing Media Studies , edited by Currant James , Park Myung-Jin . Routledge , London , pp. 97 – 110 . Huber Gregory A. , Arceneaux Kevin ( 2007 ). “Identifying the Persuasive Effects of Presidential Advertising.” American Journal of Political Science , 51 , 957 – 977 . Google Scholar Crossref Search ADS King Gary , Pan Jennifer , Roberts Margaret E. ( 2014 ). “Reverse-Engineering Censorship in China: Randomized Experimentation and Participant Observation.” Science , 345 , 1 – 10 . Google Scholar Crossref Search ADS Krasno Jonathan S. , Green Donald P. ( 2008 ). “Do Televised Presidential Ads Increase Voter Turnout? Evidence from a Natural Experiment.” Journal of Politics , 70 , 245 – 261 . Google Scholar Crossref Search ADS Langston Joy ( 2003 ). “Rising from the Ashes? Reorganizing and Unifying the PRI’s State Party Organizations after Electoral Defeat.” Comparative Political Studies , 36 , 293 – 318 . Google Scholar Crossref Search ADS Langston Joy ( 2006 ). “The Changing Party of the Institutional Revolution Electoral Competition and Decentralized Candidate Selection.” Party Politics , 12 , 395 – 413 . Google Scholar Crossref Search ADS Larreguy Horacio ( 2013 ). “Monitoring Political Brokers: Evidence from Clientelistic Networks in Mexico.” Working paper, Harvard University, Cambridge . Larreguy Horacio , Marshall John , Querubín Pablo ( 2016 ). “Parties, Brokers and Voter Mobilization: How Turnout Buying Depends Upon the Party’s Capacity to Monitor Brokers.” American Political Science Review , 110 , 160 – 179 . Google Scholar Crossref Search ADS Larreguy Horacio , Marshall John , Snyder, Jr. James M. ( 2017a ). “Publicizing Malfeasance: How Local Media Facilitates Electoral Sanctioning of Mayors in Mexico.” Working paper, Harvard University, Cambridge . Larreguy Horacio , Marshall John , Trucco Laura ( 2017b ). “Breaking Clientelism or Rewarding Incumbents? Evidence from an Urban Titling Program in Mexico.” Working paper, Harvard University, Cambridge . Lawson Chappell H. ( 2002 ). Building the Fourth Estate: Democratization and the Rise of a Free Press in Mexico . University of California Press , Berkeley . Lawson Chappell H. ( 2004 ). “Television Coverage, Vote Choice, and the 2000 Campaign.” In Mexico’s Pivotal Democratic Election: Candidates, Voters, and the Presidential Campaign of 2000 , edi