Campaign Effects and the Elusive Swing Voter in Modern Machine Politics

Abstract

Are vote-choice buying attempts successful? Much research across the social sciences argues that political machines expertly turn citizens into clients, undermining core aspects of democracy. Using insights from behavioral theories of vote choice, I argue that standard partisan campaigns can diminish vote-choice buying’s efficiency. Machines face a targeting problem: Local brokers identify good clients using long-term markers but then campaigns shift many citizens’ vote-relevant attitudes in ways that brokers cannot detect, leading to targeting errors. Vote-choice buying remains effective on recipients who are unmoved by the campaigns, but this group is small where campaigns are influential. Tests using panel surveys from Mexico’s 2000 and 2012 elections measure vote-buying attempts with direct questions and list experiments, employ various measures of campaign influence, and rely on new and existing estimation techniques. The findings yield a more optimistic view of the quality of elections in new democracies than current literature implies.

Keywords

vote buying clientelism elections voting behavior Mexico

Introduction

In the days following Mexico’s 2012 presidential election, the runner-up, Andrés Manuel López Obrador brought suit against the winner, Enrique Peña Nieto, in Mexico’s Federal Electoral Tribunal. His main argument for nullifying the election rested on anecdotal evidence that Peña Nieto’s campaign offered selective benefits to citizens for their votes, a practice that is illegal under Mexican law. In one event, López Obrador surrounded himself with some 3,000 grocery store gift cards that were reportedly handed over by repentant vote-sellers. Others claimed they received cash, a job, building materials, food, clothing, livestock, and telephone cards (Tribunal Electoral de la Procuraduría Judicial Federal [TEPJF], 2012). To López Obrador’s supporters, the evidence indicated a vote-buying scheme large enough to account for the 3.3 million votes or 6.77 percentage point deficit to Peña Nieto. Are vote-choice buying attempts successful? Do selective benefits and services generate votes for political machines that they would not have won otherwise?

The answer seems obvious. Analysts have documented vote buying in approximately 53 countries, implying substantial electoral returns.¹ Voters in many new democracies and developing economies are considered easy targets due to weak partisanship and grinding poverty that may elevate the value of short-term benefits over considerations like policy and government performance. Perhaps for these reasons, scholarship across the social sciences often assumes that vote-choice buying attempts are effective.

Yet a deeper look casts doubt. Every party leader, local broker, and candidate interviewed for this project said they do not know if vote-choice buying works and they lack reliable methods for assessing its efficacy (see Supplemental Appendix Table A2). Mexico’s election management bodies are also in the dark. In response to López Obrador’s challenge, the Federal Electoral Tribunal ruled that it could not ascertain the psychological effect of selective benefits on voters’ choices (TEPJF, 2012). The specialized literature argues that vote buying can stumble if machines target the wrong voters, including loyalists who already prefer the machine (Stokes et al., 2013), citizens with democratic values (Carlin & Moseley, 2015), and antimachine partisans (Calvo & Murillo, 2019).² Despite rising skepticism that vote-choice buying is effective, current arguments imply that well-organized machines can overcome targeting problems. In particular, they can expertly identify good clients, a task made easier because citizens’ vote intentions are assumed to be static during the campaign season.

In contrast, I argue that partisan competition causes unavoidable targeting errors that make vote-choice buying much less efficient than previously thought, even when practiced by well-organized political machines. Partisan competition regales voters with standard campaign influences that shift many citizens’ vote intentions. Machines cannot observe these hidden psychological effects and thus target bribes on the basis of voters’ precampaign dispositions, no matter when the bribes are actually distributed. Campaigns-season changes in voters’ attitudes thus force targeting errors because many good precampaign clients convert into bad Election-Day clients. Such errors can cumulate to make vote-choice buying inefficient where campaigns are influential.

The first section of this article discusses existing research on the subject. The second section builds on this prior literature, presents my argument that campaigns can disrupt attempted vote-choice buying, and derives hypotheses. The third section measures vote-choice buying attempts using a list experiment embedded in the Mexico 2012 Panel Study. The fourth section tests the argument using a new purpose-designed statistical methodology due to Imai, Park, and Greene (2015). Vote-buying attempts are estimated as a latent variable from the list experiment and then incorporated as a predictor in vote-choice models alongside standard behavioral influences of vote choice. The approach improves measurement and diminishes biases that may be present in existing observational research. The use of panel data and post hoc tests help mitigate potential endogeneity. The fifth section focuses on the proposed mechanism, demonstrating that targeting errors induced by campaign effects diminish vote-choice buying’s efficacy and showing that vote-choice buying attempts by the Institutional Revolutionary Party (PRI) were 99% inefficient in the 2012 elections.

The sixth section extends the analysis to other cases and provides a detailed look at Mexico’s 2000 contest when the PRI lost, ending 71 years of single-party dominance. Had vote-choice buying been perfectly efficient in this election, the PRI would have won. Data for these tests come from a four-wave panel study that permits models less prone to potential endogeneity. The analysis uses different estimation techniques and measures of vote-buying attempts and campaign influence but shows strikingly similar results to the 2012 findings. The PRI’s vote-choice buying attempts in 2000 were 90% to 94% inefficient but were effective on recipients who ignored the campaigns.

The conclusion discusses scope conditions, why machines might engage in inefficient vote-getting tactics, and the more optimistic implications of my findings for the quality of democracy. The empirical analyses are limited to Mexico, but if one of the world’s most accomplished political machines scarcely makes vote-choice buying work, then the practice may be fragile in other systems as well. Rather than vote buying undermining democracy, core features of competitive democracy can diminish vote-choice buying’s efficacy.

Citizens Into Clients: Existing Arguments About Vote-Choice Buying’s Efficacy

Vote-choice buying has commanded significant attention from social scientists across disciplines and has produced major insights. The challenge of studying illegal behavior means that analyses have not yet resolved the question of vote-choice buying’s efficacy. This section discusses the literature upon which I build in subsequent sections.

A significant portion of the research assumes that vote buying is effective. Formal theory typically models how vote buying works without asking whether it does (Cox & McCubbins, 1986; Dixit & Londregan, 1996; Lindbeck & Weibull, 1987; Stokes, 2005; Zarazaga, 2016). Ethnography rarely probes efficacy directly and seems to imply that powerful patrons can compel compliance by threatening sanctions ranging from economic exclusion to harassment and assault (Auyero, 2000; Stokes et al., 2013; Szwarcberg, 2015; Zarazaga, 2014). Most survey research focuses on which voters are targeted but notably refers to “vote buying” rather than “vote-buying attempts.”

Many studies using observational data also encounter measurement challenges. Often, direct survey questions are used, such as “Did someone offer you a good, service, or favor in exchange for your vote?.” Yet because vote buying is illegal and considered illicit in most contexts, respondents may falsify their answers, especially when the quid pro quo is made explicit (Çarkoğlu & Aytaç, 2015; Corstange, 2009; González-Ocantos et al., 2012).

As a workaround to measurement challenges and to address potential endogeneity in observational studies, recent work takes two different approaches. Kitschelt and Altamirano (2015) use expert opinion to score country cases and show that clientelist effort outpaces efficiency, especially in Latin American countries.³

Another approach randomizes some variables in the proposed causal chain. This research shows that vote-choice buying can be effective but also supports my argument that its efficiency can be diminished by standard campaigning. For instance, Cantú (2019) and Larreguy et al. (2016) show that parties win more in constituencies where candidates’ representatives have incentives to exert more effort, but these representatives’ dual roles as vote buyers and campaigners make it difficult to know whether campaign effort or selective benefits are causal. Another group of experimental studies shows that constituencies treated with anti-vote-buying messages vote differently than untreated constituencies. Interestingly, the treatments usually dissuade vote-selling by priming the same messages found in standard campaigns. For instance, Hicken et al. (2015) asked respondents to “promise to vote according to your conscience” and Banerjee et al. (2011) used report cards on incumbent performance and candidates’ qualifications (similar treatments appear in Blattman et al., 2017; Cruz et al., 2016; Fujiwara & Wantchekon, 2013; Green & Vasudevan, 2016; Kramon, 2016; Vicente, 2014). These field experiments find that even one exposure to weakly persuasive messages crafted by researchers can convince voters to defect from vote-buying arrangements. If the nonclientelist criteria are doing the work, then presumably professional campaigners blanketing the airwaves with persuasive messages on similar themes can do even better.

Problems of Vote Buying During Partisan Campaigns

Analysts recognize that the secret ballot and a shift from old-style clientelism to thinner relations between patrons and clients in many contexts challenge machine politics (Mares & Young, 2016). I argue that another aspect of modern democracy, partisan campaigns, also diminishes vote-choice buying’s efficacy because it forces machines to make targeting errors.

Scholars have debated which voters machines should rationally target. The dominant view is that when machines seek to buy vote choices, bribes are unnecessary to generate support from Election Day that the literature calls “loyalists” and who like the machine for nonclientelist reasons. Bribes are insufficient for opposition voters who will not support the machine even if they receive a selective benefit at the going rate; however, bribes are necessary and sufficient to win weakly opposed voters that the literature labels “swing.” I follow this labeling convention and present my argument with respect to the dominant view, but my central claims hold even if machines seek to buy vote choices from loyal or opposition voters.

Current work suggests that swing voters might not receive bribes (and thus vote-choice buying founders) because machines face organizational challenges. Machines cannot control local brokers who may target loyalists to cheaply grow their personal networks (Stokes et al., 2013). Machines may also purposely overtarget loyalists (Dixit & Londregan, 1996) or network-proximate core voters (Cox & McCubbins, 1986) because they are inefficient at distributing benefits to swing voters who they know less well. Even so, all these models show that machines target swing voters as they gain competence.⁴ The idea that targeting errors challenges machine politics is thus well established in literature.

I present a different argument for targeting errors, one based in the influence of partisan campaigning that is fundamental to competitive democracy. Even if brokers act as perfect agents for the machine by attempting to target swing voters, standard campaigns can diminish the efficiency of vote-choice buying attempts. Machines seek to pay Election Day swing voters (i.e., voters who would choose the machine if given a selective benefit before voting but choose a challenger if not). Yet, brokers can only identify precampaign swing voters no matter when they distribute goods. This mismatch causes targeting errors because effective campaigns convert some precampaign swing voters into postcampaign loyalists for whom bribes are unnecessary and some into postcampaign opposition for whom bribes are insufficient. Machines cannot escape these effects by distributing bribes close to Election Day or by offering citizens a flat rate, as I show below.

Consistent with my argument, existing work argues that brokers use voters’ precampaign dispositions to make targeting decisions. The clearest statement comes from Stokes (2005), who writes

Certain party–organizational structures allow parties to discern individual voters’ types . . . familiar neighbors work as operatives for political parties. They therefore know much about an individual that shapes his partisan attachments: his job, associational membership, parents’ ideological inclinations, and public statements about parties and policies . . . Information about individual voters’ partisan pre-dispositions helps the machine make inferences about how individuals vote and whether they are good candidates for vote buying. (p. 317) [emphasis added]

Ethnography in Argentina (Szwarcberg, 2015; Zarazaga, 2014), India (Schneider, 2019), and Mexico (author interviews, see Supplemental Appendix Table A2) as well as quantitative evidence from Tukey (Çarkoğlu & Aytaç, 2015), Lebanon (Corstange, 2012), Nicaragua (González-Ocantos et al., 2012), Brazil (Nichter, 2018), Argentina, and Chile (Calvo & Murillo, 2019) are consistent with Stokes’s view.

Brokers thus identify good clients using voters’ precampaign dispositions, but decades of research show that campaigns can deeply affect vote choices by priming vote-relevant variables (Petrocik, 1996), persuading voters to change their attitudes on key issues (Bartels, 1993), or informing voters about the competitors’ platforms (Lenz, 2009). Even in the United States where strong partisan identities limit campaign effects, about 20% of voters choose a presidential candidate that is inconsistent with their precampaign dispositions (Finkel, 1993). In new democracies, weaker partisanship makes voters more susceptible to campaign influences (Greene, 2011). Studies in these countries, including in Mexico, show that vote-relevant variables change due to mass media exposure (Lawson & McCann, 2005), advertising (Beltrán, 2007), reception of campaign messages (Hart, 2016), control over media content (Boas & Hidalgo, 2011), and discussion among peers (Baker et al., 2006), among others.

Campaign effects are hidden from brokers for three reasons. First, unlike demographics and long-term partisanship that embedded brokers can discover, campaigns typically prime less stable attitudes such as retrospective performance evaluations and the best way to tackle pressing public policy issues like crime.

Second, parties do not know all the markers that govern campaign-induced attitude change. In addition to mean effects that raise or lower all voters’ utility for the machine, campaigns have stochastic effects that pull some voters toward the machine and repel others. Stochastic effects underlie probabilistic (Enelow & Hinich, 1989) and valence theories of vote choice (Adams et al., 2005). Voters’ reactions to the same campaigns differ due to so many personal traits (Erikson & Romero, 1990) that analysts treat stochastic effects as unknown to competitors, even in advanced countries where consultants use sophisticated measurement tools (Adams & Merrill, 2009). For instance, many analysts in the United States were surprised by the effects of the 2016 presidential campaigns.

Finally, even if machines attempt to gather information on how campaigns affect individual clients, a claim that is not supported by in-depth studies or my fieldwork in Mexico, they could not improve targeting. Campaign effects typically crystalize in voters’ minds close to Election Day (Fournier et al., 2004), so that machines—often characterized as lumbering networks of personal contacts (Calvo & Murillo, 2019)—could not retarget bribes to postcampaign swing voters. These three challenges mean that brokers can only identify clients by their precampaign dispositions. There is no opportunity for machines to wait until campaign influence manifests and then make targeting decisions. Thus, as campaigns become more effective, targeting errors rise and vote-choice buying becomes less efficient.

To contrast my argument with the standard model that does not include campaign effects and thus predicts efficient targeting, I present three stylized spatial scenarios in Figure 1. The bottom row represents Stokes’ (2005) predictions where the machine targets pre-campaign swing voters. Voters with preferences to the left of x^* and thus closer to the machine (x_m) are loyalists who do not receive bribes, whereas those to the right are closer to the opposition party (x_o) and are nonloyalists. The machine replaces some nonloyalists disutility with a bribe b. Provided the selective benefit is large enough to overcome initial disutility, these nonloyalists become swing voters whose support the machine wins. The machine does not pay (or pays too little) to opposition voters that deeply dislike the machine and lie beyond x^* + b^*.

Figure 1.

Stylized effects of campaigns on vote-buying yield.

Adding the impact of the campaigns to this standard model introduces targeting errors that diminish the machine’s yield from vote-choice buying attempts. The top row shows what happens when the campaigns decrease voter utility for the machine relative to the opposition. The striking prediction is that all precampaign swing voters become converted opposition voters who abandon the machine. This seemingly severe reaction is actually provoked by the machine’s proficiency. The most competent machines tailor selective benefits that are the right size to just overcome swing voters’ initial disutility, making them prefer the machine by a hair over the opposition. Embedded brokers who know their clients’ needs can set the right (precampaign) price, thus permitting the machine to stretch its budget and (attempt to) maximize vote-buying yield (Szwarcberg, 2015; Zarazaga, 2014). Yet precampaign swing voters who receive a bribe and thus incline slightly for the machine are extremely sensitive to any reduction in their utility. If these voters dislike the machine more after the campaigns, then their cost rises above the precampaign rate they were paid. Consequently, they convert into postcampaign opposition.

The middle row in Figure 1 shows effects if the campaigns increase voter utility for the machine relative to the opposition. Precampaign swing voters who like the machine more become cheaper to buy. Some will like it enough to vote for it without a bribe, becoming postcampaign converted loyalists. The machine still wins these votes, but not due to vote buying. It also wastes all benefits paid to them because their postcampaign cost falls to zero.

The only votes the machine wins due to vote buying come from precampaign swing voters who are unaffected by the campaigns or shift utility within a sufficiently narrow band that their cost does not fall to zero or rise above the value of the bribe they received. For these continuing swing voters, the selective benefit they received is still necessary and sufficient to support the machine.

My key claim is that where campaigns affect voter utility, continuing swing voters represent a reduced portion of the precampaign swing voters that machines attempt to buy. Although Figure 1 represents separate scenarios as if campaigns only have mean effects, one can easily imagine that stochastic effects cause some voters to react negatively, as in the top row, and some to react positively, as in the middle row. (The Supplemental Appendix shows other graphical depictions of the argument.) These polarizing reactions cumulate to cause targeting errors that diminish vote-choice buying’s efficiency. In some circumstances, standard campaigns may be ineffective, moving few voters and thus maintaining high vote-buying efficiency. In others, campaigns may significantly shift vote-relevant attitudes, making machine politics stutter.

My argument’s central claims hold under alternative targeting scenarios. If machines target precampaign loyalists as an insurance policy, they would still lose support among converted opposition voters. If machines target precampaign opposition voters, they would overpay those who like the machine more after the campaigns and fail to win those who dislike it enough to continue as postcampaign opposition voters.

Machines cannot escape the inefficiency induced by the campaigns by anticipating their impact. (I again focus on swing voters, but the argument applies if machines target others.) First, if the campaigns cause lost support due to shifts in voters’ utility, then it stands to reason that they also create new opportunities. The campaigns will make some precampaign loyalists and opposition into postcampaign swing voters whose support could be bought. But because brokers can only identify precampaign types, they will not have paid these voters and they thus represent missed opportunities.

Second, even if machines cannot anticipate stochastic campaign effects, maybe savvy organizations could forecast mean effects. Knowing whether the candidate will have a generally positive or negative impact on voters could ameliorate some targeting errors, but machines are unlikely to act on such knowledge even if they had it. Literature on vote buying often refers to local brokers as independent agents that are not easily controlled (Stokes et al., 2013). None of the relevant literature or my interviews document pipelines of information flowing from war rooms in central party headquarters to brokers on the front lines (Schneider, 2019; Szwarcberg, 2015; Zarazaga, 2014). Even if brokers receive such information, they may not change their targeting strategy. Distributing resources according to the machine’s counterfactual theory of campaign effects requires brokers to ignore concrete contemporaneous evidence. The machine would have to convince brokers that, anticipating a good campaign, they should divert resources to precampaign opposition voters, hoping they will convert into postcampaign swing. Alternatively, anticipating a bad campaign, brokers should target pre-campaign loyalists, fearing they will turn against the machine. Brokers are often paid by the votes they produce, making it difficult to imagine they would bet on a late-season conversion they cannot observe.

Third, machines could pay clients a flat rate rather than an individually tailored amount (Zarazaga, 2016). Yet, overpaying some clients means buying fewer votes from the outset and wasting even more resources on converted loyalists and continuing swing voters who react positively to the campaigns. Paying a flat rate also means underpaying some voters who thus continue as postcampaign opposition voters. A similar logic means that machines cannot avoid campaign effects by targeting groups instead of individuals except under specialized circumstances where brokers can credibly threaten entire constituencies (Rueda, 2015).

Finally, in many contexts, machines cannot overcome targeting problems by jointly determining the level of investment in campaigns versus vote buying. Campaign resources are typically audited, whereas resources for vote buying are from illicit provenance, making coordination of investment difficult. Furthermore, in Mexico, resources for vote buying are raised and spent at multiple levels of the political system so that any one constituency might be targeted by local-, state-, and federal-level candidates from the same party, each of whom organizes her own vote-buying operation (author interviews, see Supplemental Appendix Table A2). This is a far cry from a party dictator sitting on one pot of money that can be strategically divided between campaigning and vote buying. Such coordination difficulties do not rule out a corner solution. But if machines invest solely in vote buying, they would suffer negative campaign effects as the opposition pummels the machine in the media. Notably, existing formal models do not even contemplate pure machine politics. They add bribes to the utility derived from the partisan variables that campaigns seek to manipulate. Empirically, the best known machines invest heavily in campaigning (Kitschelt & Wilkinson, 2007), and only 14 of the 318 parties that use clientelism in an expert survey were judged as solely clientelist (Singer & Kitschelt, 2011, pp. 21–22).

My argument that core aspects of modern democracy can diminish the efficiency of vote-choice buying attempts generates three main hypotheses:

Hypothesis 1: Campaign Effects. Campaigns convert some precampaign swing voters into postcampaign loyalists or postcampaign opposition voters, on whom bribes are ineffective. Campaigns thus reduce the proportion of continuing swing voters, on whom vote-choice buying remains effective.

Hypothesis 2: Targeting Errors. Campaign effects cause machines to overtarget postcampaign loyalists and opposition, even if brokers channel benefits to precampaign swing voters as the machine wishes.

Hypothesis 3: Vote-Choice Buying Efficiency. The influence of vote-choice buying attempts on vote choices is low where vote-relevant attitudes change during the campaign season.

Testing these hypotheses requires determining whether vote buying is effective when controlling for the effects of the campaigns on voters’ choices. I first focus on Mexico’s 2012 elections, measuring vote-choice buying attempts with a list experiment and campaign effects with election-season changes in performance assessments. I then extend the analysis to Mexico’s 2000 elections, using a direct measure of the provision of selective benefits and media consumption habits to tap exposure to the campaigns.

Campaigns and Vote Buying During Mexico’s 2012 Presidential Election

This section contextualizes the main measures I use to analyze vote-choice buying’s influence in Mexico’s 2012 presidential elections. In the final tally, Enrique Peña Nieto of the formerly dominant PRI won with 39.2% of the vote, besting López Obrador of a leftist coalition led by his Party of the Democratic Revolution (PRD) with 32.4%, and Josefina Vázquez Mota of the incumbent National Action Party (PAN) with 26%. Gabriel Quadri of New Alliance (PANAL) won 2.3%.

Campaigns for the presidency were legally restricted to 90 days, but many voters’ choices changed in this brief period. Only 52.4% of respondents to the Mexico 2012 Panel Study voted in the July 1 election for the candidate they supported in April, 28.2% switched, and 19.6% went from undecided to a vote choice on Election Day (Supplemental Appendix Table A3). During the campaign season, Peña Nieto attracted 18.2% of voters and repelled 11.6%. The campaigns and/or vote-choice buying could account for these significant and polarizing shifts in voters’ choices.

Campaigns exerted substantial influence on voters’ choices in prior presidential contests, with nearly half of the voters eventually choosing a candidate different from their precampaign dispositions (Greene, 2011). In 2012, the campaigns generated 5,466 hours of television advertising for an average of 62 broadcast hours each day.⁵ The candidates focused on economic well-being and incumbent performance, but their messages diverged. In contrast to the incumbent’s aggressive use of the military to combat drug trafficking, Peña Nieto reminded voters that personal security was better when the PRI held power before 2000 and promised to diminish homicides and kidnappings by 50%. He also touted his performance in creating jobs and infrastructure as Governor of the State of Mexico. López Obrador focused on ameliorating his radical outsider image. He promised to create jobs for young people and reduce violence by making Mexico into a “republic of love.” Vázquez Mota strained to differentiate her platform from her copartisan incumbent’s ill-fated war on drugs that saw a massive spike in homicides. Her lackluster campaign gave the two frontrunners opportunities to poach her initial supporters.

Citizens’ vote-relevant attitudes changed substantially during the campaign season. Presidential approval and evaluations of economic well-being changed by more than one standard deviation each. On both variables, attitudes polarized, with about 26% of voters shifting in a direction that favored Peña Nieto, about 35% shifting against, and the remainder remaining the same. This polarization is inconsistent with changing objective economic and security conditions during the campaign season (see Supplemental Appendix). Attitude change is also not due to vote-buying attempts, a theme I address in the following section.

The observed attitude changes are consistent with exposure to campaign messages. Supplemental Appendix Table A6 shows that shifting television viewing from “every once in a while” to “daily” moves sociotropic economic evaluations 22% of one standard deviation and presidential approval 15% of one standard deviation, ceteris paribus. Hypothesis 1 postulates that these substantial and polarizing campaign effects should convert some precampaign swing voters into postcampaign loyalists and even more into postcampaign opposition.

Mexico also has a rich history of vote buying that could have affected vote choices. Historically, the PRI was one of the world’s most proficient machines. In 2012, it dominated among partisan poll watchers who also often work as brokers, covering 97.3% of precincts compared with the PAN’s 80.8% and the PRD’s 60.3%.⁶ Alianza Cívica’s (2012) 500 election observers estimated that 71% of polling place violations benefited Peña Nieto. Evidence from the direct question about vote buying in the Mexico 2012 Panel Study is remarkably similar: 72% of respondents who received a selective benefit for their vote said they were paid by the PRI, 22% by the PAN, and 6.3% by the PRD.

I gauge vote-buying attempts with a list experiment embedded in the Mexico 2012 Panel Study.⁷ The study interviewed a nationally representative sample of citizens with a valid voter registration card in their homes in April near the start of the presidential campaigns and recontacted the same respondents after the July 1 election (N = 952). To measure vote buying, the sample was randomly divided into treatment and control groups. Enumerators said, “I am going to read you a list of [3/4] activities that appear on this card and I would like you to tell me how many of these activities you have done in recent weeks. Please do not tell me which activities, just how many.” The activities appear in Table 1. Supplemental Appendix Table A7 shows balance on demographic and political variables.

Table 1.

Treatment and Control Items in the List Experiment.

List A (control)	List B (treatment)
• Watch TV news that mentions a candidate	• Watch TV news that mentions a candidate
• Go to a campaign event	• Go to a campaign event
• Talk about politics with other people	• Receive a gift, favor, or access to a service in exchange for your vote
	• Talk about politics with other people

The difference in the mean number of reported activities across treatment and control groups indicates the incidence of vote-buying attempts. Table 2 shows that 6.7% of respondents experienced a vote-choice buying attempt near the start of the campaigns, but this value does not reach statistical significance. During the campaigns, 21.4% experienced a vote-choice buying attempt, and this value is statistically significant at the 99% level. (p = .0002).

Table 2.

Estimated Percent of the Electorate Targeted for Vote Buying.

Item count	List experiment				Direct measure
	April 2012		July 2012		April 2012	July 2012
	Control	Treatment	Control	Treatment	April 2012	July 2012
0	12.1%	14.1%	6.3%	6.3%
1	52.2	47.4	45.9	38.2
2	26.7	27.8	33.9	33.6
3	9.0	6.8	13.9	16.3
4		4.0		5.6
N	446	454	460	461	915	513
Vote-buying estimate	6.7% (p = .12)		21.4%*** (p = .0002)		2.5%	5.2%

Panel respondents only.

p < .1. **p < .05. ***p < .01.

Published work shows that list experiments outperform direct questions of sensitive behavior (Blair & Imai, 2012; Corstange, 2009; Kiewiet de Jonge & Nickerson, 2014; Rosenfeld et al., 2016). Indeed, a direct question that copied the sensitive item verbatim estimated that only 5.2% of respondents experienced a vote-buying attempt during the campaigns. Still, the list experiment could have underestimated vote-buying attempts. Reading the items aloud could have sparked social desirability bias, but there was no increase in the list count when another adult was present. Clever respondents in the treatment group might have realized that affirming all items would identify them as recipients; yet the data pass Blair and Imai’s (2012) test for design effects. Conversely, the list experiment could have overestimated vote-buying attempts, but asking if respondents “exchanged” their vote for a benefit encouraged ignoring benefits deemed not worth a vote. The question wording also makes it less likely that respondents interpreted the item to imply a turnout-buying attempt instead of a vote-choice buying attempt.

Table 3 uses the list experiment to show basic targeting patterns. These results reinforce the conclusion that the PRI was the main vote buyer and show that targeting was strategic rather than random. The patterns imply that the PRI focused on vote-choice buying rather than turnout buying because the best clients for the latter—consistent Peña Nieto supporters—were relatively less likely to receive an offer. Attempting to expand its electoral coalition seems sensible, given that the PRI was in the opposition in 2012. In addition, the patterns show that the PRI targeted Vázquez Mota supporters and undecided voters at much higher rates than initial López Obrador supporters. What these patterns do not show is whether and to what extent vote-buying attempts may have succeeded in winning votes the PRI would not have won otherwise.

Table 3.

Probability of Vote-Buying Attempt by Group.

April vote intention	July vote choice				All April supporters
April vote intention	Peña Nieto	López Obrador	Vázquez Mota	No response/did not vote	All April supporters
Peña Nieto	17.8%*	43.2%*	73.1%***	—	25.3%***
López Obrador	0	—	—	—	2.0
Vázquez Mota	82.1***	—	—	77.8**	25.1**
Other	—	—	—	—	—
Undecided	13.7	36.0	—	58.8**	30.0***
PRI ID only	95.0**	—	—	—	52.7*

Cells report unpaired t tests between treatment and control groups. Cells with 10 or fewer respondents in the treatment group excluded. Vote buying measured in July; party identification measured in April. No response/did not vote and undecided include spoiled and blank ballots and nonvotes. PRI ID = Institutional Revolutionary Party identification.

p < .1. **p < .05. ***p < .01.

The Efficiency of Vote-Choice Buying Attempts in Mexico’s 2012 Elections

In this section, I test whether the vote-choice buying attempts uncovered by the list experiment influenced vote choices in 2012. My argument is that vote buying stumbles because campaigns force brokers to target many bad Election-Day clients. This claim runs counter to the assumption that vote buying works consistently. It also differs from arguments that vote buying founders because machines cannot credibly threaten to monitor voters’ choices, because brokers purposefully target loyalists against the machine’s wishes, or because machines avoid targeting swing voters who they know less well.

My approach compares the results from two models of final July vote choice. The models only differ in the included explanatory variables; they use the same dependent variable. The precampaign dispositions model includes the variables commonly used to predict vote choice in Mexico and elsewhere, measured in April, plus the likelihood of vote-buying attempts between April and July. This model tests the existing argument represented by the bottom row in Figure 1 that voters begin the campaigns with preexisting attitudes and then some are targeted for vote buying, plausibly leading to support for the machine. The postcampaign dispositions model includes the same predictors but adds voters’ July presidential approval and sociotropic economic evaluations. This model assesses whether vote-buying attempts still influence choices when campaign-season changes in vote-relevant attitudes are included. Comparing precampaign and postcampaign disposition models is a mainstay of the behaviorist literature on campaign effects that uses panel survey data (e.g., Finkel, 1993). If Hypothesis 3 holds, vote-buying attempts will appear to influence choices in the precampaign but not in the postcampaign disposition model. By July, the campaigns will have made some initially good clients into bad ones.

The models employ a new purpose-designed statistical methodology to estimate the influence of vote-buying attempts on vote choices using the list experiment. The approach develops a maximum likelihood estimator based on an expectation-maximization (EM) algorithm. The M-step maximizes the conditional expectation of the complete-data log-likelihood function. The E-step predicts an affirmative response to the vote-buying item in the list experiment using information from the control items, outcome, and the included covariates. Standard errors are calculated analytically.⁸ For a full explanation of this methodology, which is beyond the scope of this article, see Imai, Park, and Greene (2015).

My approach mitigates three core challenges in current observational designs. First, the nonrandom targeting of electoral bribes shown in Table 3 suggests that many of the same variables that influence vote choices also probably influence which voters machines target. The models in this section remove observable potential confounds that come from the machine’s targeting criteria by estimating the probability of being targeted as a latent variable (see Supplemental Appendix Table A8). The probability of receiving a selective benefit is then incorporated as a predictor alongside the standard variables included in well-worn behaviorist models of vote choice.⁹ Second, the approach improves measurement because it is the first study that uses a list experiment to measure vote-choice buying as an explanatory variable in vote-choice models. Finally, pairing the methodology with panel data controls for prebribe vote intentions and thus ameliorates some of the endogeneity concerns that arise in cross-sectional observational data. Together, these elements help provide a plausible test of the influence of vote-buying attempts on vote choices.

Table 4 presents the outcome vote-choice models.¹⁰ Negative coefficients indicate support for Peña Nieto. In the precampaign disposition model, vote-choice buying attempts appear to sway voters from Vázquez Mota to Peña Nieto. In fact, bribes seem to overwhelm voters’ other precampaign attitudes. Few of these standard influences reach statistical significance in the model below but many do when vote-buying attempts are excluded (Supplemental Appendix Table A12). Support for the standard view that vote-choice buying is effective seems strong.

Table 4.

Vote-Choice Models for Mexico’s 2012 Elections.

	DV = July vote choices						DV = July vote choices
	Precampaign dispositions						Postcampaign dispositions
	Vázquez Mota vs. Peña Nieto			López Obrador vs. Peña Nieto			Vázquez Mota vs. Peña Nieto			López Obrador vs. Peña Nieto
Variable	Coef	Sig	SE	Coef	Sig	SE	Coef	Sig	SE	Coef	Sig	SE
Vote-buying attempt, Apr to July	−2.063	**	0.855	0.126		0.632	−0.986		0.642	0.799		0.671
Peña vote intention, Apr	−0.408		0.460	−0.591	*	0.360	−0.579		0.398	−0.451		0.399
Vázquez vote intention, Apr	1.514	**	0.583	−0.289		0.552	0.932	**	0.435	−0.663		0.631
López vote intention, Apr	−0.757		0.578	0.604		0.454	−0.773		0.546	0.919	*	0.500
Female	−0.101		0.345	−0.507	**	0.261	0.014		0.295	−0.581	**	0.269
Age	0.080		0.070	−0.061		0.048	0.031		0.055	−0.048		0.051
Education	−0.062		0.080	0.008		0.059	−0.044		0.068	0.025		0.068
PRI ID, Apr	0.022		0.557	−0.564		0.413	−0.176		0.471	−0.776	*	0.487
PAN ID, Apr	0.305		0.429	0.171		0.419	0.633	*	0.365	0.434		0.467
PRD ID, Apr	0.069		0.610	1.082	***	0.419	0.331		0.561	1.016	**	0.422
Peña competence, Apr	−0.372	**	0.145	−0.231	**	0.096	−0.181	*	0.105	−0.236	**	0.101
Vázquez competence, Apr	0.139		0.141	−0.053		0.112	0.146		0.120	0.014		0.118
López competence, Apr	0.310	*	0.167	0.274	***	0.098	0.133		0.129	0.284	***	0.105
Ballot secrecy	0.077		0.185	0.059		0.141	−0.101		0.164	0.093		0.156
No non-Peña poll watchers	0.740		0.631	0.167		0.491	0.758		0.563	0.089		0.437
Sociotropic econ evals, Apr	0.249		0.175	−0.032		0.131	0.100		0.149	0.019		0.140
Presidential approval, Apr	0.022		0.182	−0.197	*	0.108	0.076		0.128	−0.181	*	0.112
Sociotropic econ evals, July							0.309	**	0.149	−0.342	**	0.155
Presidential approval, July							0.395	**	0.176	0.063		0.121
Intercept	−1.027		1.022	0.505		0.864	−2.406	**	0.981	0.101		1.095

The dependent variable for all models is final July vote choice. Peña Nieto is the excluded category. N = 498 for Vázquez Mota vs. Peña Nieto and N = 532 for López Obrador vs. Peña Nieto. The 10 Quadri voters were omitted. Estimates are from five multiply imputed data sets using Amelia. Average missingness was 1.4%. Cases with missing vote-choice or vote-buying data were dropped. Rubin’s (1987) “rules” combine estimates. Vote-choice models (logistic regression) are estimated jointly with control items and targeting models (item count regression) in Supplemental Appendix Table A8. DV = dependent variable; PAN ID = National Action Party identification; PRD ID = Party of the Democratic Revolution identification ; PRI ID = Institutional Revolutionary Party identification.

p < .1. **p < .05. ***p < .01.

To ease interpretation, Figure 2 shows the partial effect of selective benefits on vote choices using the precampaign disposition model. Targeted citizens appear a whopping 40% more likely to vote for Peña Nieto over Vázquez Mota in July (p < .05).

Figure 2.

Effects of vote buying on vote choices using precampaign dispositions.

Yet, as Hypothesis 3 suggests, the apparent impact of vote buying in these models may be illusory. When voters’ postcampaign attitudes are included, the coefficient on vote-buying attempts in the Peña Nieto vs. Vázquez Mota model is less than half the size it was in the precampaign model and is no longer statistically significant. The associated bump in the predicted probability of supporting Peña Nieto due to vote buying falls to just 11.7 percentage points, fully 75% below what it was in the precampaign model, and loses statistical significance. These findings are represented in Figure 3. The impact of bribes in swaying voters away from López Obrador remains nil, a finding that is consistent with Peña Nieto’s targeting strategy discussed above.

Figure 3.

Effects of vote buying on vote choices using postcampaign dispositions.

This is not a classic “null” finding because the strong effect of vote-buying attempts in the precampaign model is explained away by the addition of postcampaign attitudes. In the rest of this section, I show that the findings are not due to potentially worrying methodological artifacts.

First, the logic of the test rules out rival explanations associated with variables that do not change across the models. Thus, the results are not due to measurement or the number of observations, both of which remain constant across the models.¹¹ The results are also not due to the potentially confounding effects of the machine’s strategy. For instance, if the PRI focused on turnout buying rather than vote-choice buying (a strategy that fits poorly with the findings in Table 3), both the precampaign and postcampaign models would register null effects on vote choices. To further ensure that turnout-buying attempts are not causing the null results in the postcampaign models, I reran them excluding Peña Nieto precampaign loyalists (as identified in Table 5 below) who reported that they were anything less than certain to turnout. The findings, shown in Supplemental Appendix Table A9, mirror the main models in Table 4.

Table 5.

Predicted Voter Types Before and After of the Campaigns.

Precampaign dispositions	Postcampaign dispositions
Precampaign dispositions	Loyal	Swing	Opposition	Total
Loyal	30.4	0.8	10.3	41.4
Swing	6.5	1.4	8.9	16.8
Opposition	7.1	0.3	34.3	41.8
Total	44.0	2.5	53.5	100.0

Entries are percent of respondents. Threshold of loyalty = 50%. See Supplemental Appendix Tables A19–A21 for results using other thresholds.

Second, the results are probably not due to the most pernicious form of endogeneity for my argument. Hypothetically, bribes could generate promachine shifts on vote-relevant attitudes. Yet models in the Supplemental Appendix show that this is not the case. Table A14 mirrors the postcampaigns models in Table 4 but regress the July postcampaign attitudes on vote-buying attempts between April and July, as well as the other covariates featured above. Vote buying is measured using the list experiment. The correlations between July attitudes and vote buying between April and July are very weak and are remote from statistical significance. Table A15 deepens the case against endogeneity by running the same model but with vote-buying attempts measured in April, before the campaigns went into full swing. Vote-buying attempts are very weakly associated with subsequent attitudes. To make sure that these findings are not due to the use of the list experiment, Tables A16 and A17 show analogous models but instead use the direct measure of vote-buying attempts. The results show no empirical support for the claim that July attitudes are endogenous to vote buying in these data.

The absence of attitude endogeneity is consistent with existing literature. Formal theory on vote buying models attitudes and bribes as separate quantities. Theories of “endogenous loyalty” argue that support comes from repeated bribes, not by shifting vote-relevant attitudes (Diaz-Cayeros et al., 2016). Kramon’s (2016) work shows that recipients make inferences about machine candidates, but he argues that inferences concern electoral viability, not “programmatic competence.”

How Campaigns Diminish the Efficiency of Vote-Choice Buying

The results demonstrate that a massive vote-buying scheme by one of the world’s most accomplished machines in a high-stakes election did not affect vote choices in 2012. Why did vote-choice buying stumble?

I argue that brokers are forced to use flawed information to determine which voters to target with selective benefits. Brokers identify precampaign swing voters and cannot detect campaign influences that convert some into postcampaign loyalists for whom bribes are unnecessary and others into postcampaign opposition for whom bribes are insufficient to support the machine. This section tests Hypothesis 2 that links campaign-season changes in vote-relevant variables to vote-buying inefficiency through brokers’ targeting errors.

My empirical strategy involves two steps. First, I determine how brokers would classify voters under two counterfactual conditions. Brokers could naïvely predict voter types using precampaign dispositions only, as the existing literature and I argue they do (Stokes, 2005; Zarazaga, 2014, 2016). Alternatively, if brokers were omniscient, they could forecast voter types using postcampaign dispositions. I model which voters would be classified as precampaign and postcampaign loyal, swing, or opposition absent receiving a bribe. Second, I compare the likelihood that voters categorized using precampaign and postcampaign criteria actually receive bribes. To underscore, these are the same voters, classified using naïve precampaign criteria or omniscient postcampaign criteria, absent the influence of vote-buying attempts.

If my argument holds, brokers target many voters that count as precampaign swing voters, but inadvertently target many who become postcampaign loyalists and opposition. Alternatively, if Stokes et al.’s (2013) mechanism for vote buying’s failure is correct, brokers are bad agents who purposefully overtarget precampaign loyalists from the start.

The naïve and omniscient broker models appear in Supplemental Appendix Tables A11 and A12. They mirror the models in Table 4 above but exclude vote-buying attempts. Note that the naïve broker model inclines against my hypotheses by including more variables than just the demographics and partisanship that existing work says brokers’ use to categorize voters. Operationally, I count voters as loyal to Peña Nieto if they have greater than 50% probability of voting for him absent bribes and nonloyal otherwise. For a subsequent test, I divide nonloyalists into opposition voters who have a 50% or greater chance of voting against Peña Nieto and swing voters who have less than a 50% chance of supporting any candidate, also absent bribes.¹² This classification approach helps illustrate the mechanism that I argue underlies vote-choice buying’s inefficiency, but the inefficiency finding itself comes from Table 4 above and does not rely on classifying voters.

The left side of Figure 4 shows that precampaign loyalists had a 26.2% probability of receiving bribes, whereas nonloyalists (swing and opposition combined) were somewhat more likely at 36.1%. If the null hypothesis is that brokers should ignore loyalists, then this represents “over-targeting.” However, in support of their argument about overtargeting, Stokes et al. (2013) show that Argentina’s Peronist Party targeted twice as many loyalists as nonloyalists. By comparison, Peña Nieto’s brokers targeted rationally, paying more nonloyalists. Mexico’s brokers are either well controlled by the machine or they spontaneously target the voters that they believe will make vote-buying work.

Figure 4.

Probability that voters classified absent bribes actually receive a bribe.

Although brokers correctly target more precampaign swing voters, I argue that the campaigns convert many of these voters into postcampaign loyalists or opposition. The right side of Figure 4 supports this claim: Postcampaign loyalists for whom bribes are unnecessary had a greater probability (30.4%) of receiving bribes than nonloyalists (27.8%). Dividing nonloyalists into swing and opposition makes the findings starker. Machines should buy the support of postcampaign swing voters, yet this key group is the least likely to have received a bribe at just 11.4% probability, and this quantity is not statistically significant.

I argue that this pattern of bribes occurs because the campaigns push many precampaign swing voters to become loyalists or opposition by Election Day. To bring this point home, Table 5 shows the transitions from precampaign to postcampaign types, using the models above. Most voters that naïve brokers would identify as precampaign loyalists or opposition continue as their respective types by the campaigns’ end; however, most precampaign swing voters convert: 6.5% of all respondents converted from swing to loyalist, 8.9% converted to opposition, and just 1.4% continued as swing voters. The vast majority of paid voters converted into bad clients during the campaign season. Moreover, conversion dynamics correspond to my predictions: The campaigns converted more voters to opposition than loyalists.

Did vote buying work on anyone in the 2012 election? One estimate comes from multiplying the proportion of postcampaign swing and opposition voters in Table 5 by their respective probabilities of having received a bribe in Figure 4 and then multiplying these quotients by the proportion of voters that switched to Peña Nieto during the campaign season. This exercise suggests that Peña Nieto could have successfully bought up to 0.6% of voters. Having attempted to buy 21.4%, this estimate implies 99% inefficiency in vote-choice buying attempts during the 2012 contest.

Extending the Analysis to Other Elections and Measurement Techniques

Evaluating the impact of vote-choice buying attempts is fraught with challenges. The approach adopted above mitigates some of the key challenges, but readers may still harbor reservations. To further address concerns about measurement and potential endogeneity, I provide new tests using data on Mexico’s 2000 elections. Empirically, this race marked a turning point because the PRI lost the presidency after 71 years in power, despite engaging in the lion’s share of vote-buying attempts. If vote-choice buying were successful, as much of the literature reviewed above argues or assumes, the PRI would have won the 2000 elections and dominant party rule would have been prolonged.

Analytically, these tests add to the prior analysis in three important ways. First, I measure vote-buying attempts by each party, thus mitigating the potentially offsetting effects of competitive vote buying in the prior section’s models that do not distinguish which party made the vote-buying offers. Second, I measure campaign influence using mass media exposure rather than campaign-season changes in performance assessments. I view this test as a useful addition rather than an improvement. The advantage of the approach in the prior section is that the list experiment improves measurement, and changes in vote-relevant variables register all channels through which campaigns influence attitudes (e.g., mass media exposure, discussion, canvassing, rally attendance). The advantage of the narrower conception of campaign effects in this section is that it shows that vote-choice buying attempts are successful on recipients that do not imbibe mass media but founder on those that do. Finally, I use the Mexico 2000 Panel Study that interviewed a nationally representative sample of citizens three times before the elections (February, April, and June) and once after the July 2 elections (Lawson et al., 2000). These data permit models that mitigate potential endogeneity due to reverse causality.

To measure vote-buying attempts, I rely on the following direct question that appeared in all but the February wave: “In recent weeks, have you received gifts or assistance from a political party? Which one?” Direct questions can underestimate vote-buying attempts, as noted above. However, Beltrán and Castro Cornejo (2020) argue that they are easy to understand and thus may yield better estimates than list experiments for less sophisticated respondents. The approach revealed that the PRI made offers to 11.8% of panel respondents to support Francisco Labastida who finished a distant second with 36.1% of the vote. The PAN made offers to 4% to support Vicente Fox who won the presidency with 42.5% of the vote. The PRD made offers to 2.3% to support Cuauhtémoc Cárdenas who won 16.6% of the vote. Hypothetically, benefits could have been used to buy turnout; however, Stokes et al. (2013, p. 72) analyze the same data and show that probable abstainers were less rather than more likely to receive a selective benefit.

I follow prior work on Mexico (Lawson & McCann, 2005) by calculating the number of days per week that respondents report consuming news from television, radio, and print sources.¹³ This variable ranges from zero to seven where the maximum score reflects exposure to all three news sources every day. Mean exposure is 2.2 with a standard deviation of 1. I measure news exposure in February (wave 1) to diminish the likelihood that news consumption was caused by later benefits provision. Unsurprisingly, news consumption scarcely varied across survey waves, implying that it was a habit acquired before the campaigns began. Table A18 in the Supplemental Appendix further suggests that the provision of benefits in June did not affect news consumption the prior February.

To test Hypothesis 3, I examine the influence of vote-buying attempts on self-reported final vote choice in July. I interact February news consumption habits with campaign-season benefits provision. I control for February vote intentions so that the model focuses on campaign-season changes in vote choices. To mitigate potential endogeneity due to confounds that come from the machine’s targeting practices, I use coarsened-exact matching on February vote intention, media exposure, and partisanship, as well as age, gender, socioeconomic status, and education (Iacus et al., 2012). By balancing on observables among recipients of the machine’s largesse with those that the machine decided not to bribe, matching can improve the estimation of causal effects using observational data and reduce dependence on model specification.

The results in Table 6 provide further support for Hypothesis 3. Models 1 and 2 show that there were no main effects of the PRI’s campaign-season vote-buying attempts on July vote choices between its candidate, Labastida, and either Fox or Cárdenas. However, the picture changes when exposure to news media is introduced as a moderator. Although the influence of PRI benefits remains null in Model 4 against Cárdenas (much like against the PRD’s 2012 candidate, López Obrador, in the prior section), Model 3 shows two large effects against Fox. The negative sign on the unmoderated PRI benefits variable implies that selective benefits encouraged recipients that were not exposed to mass media news to vote for Labastida. The positive sign on the interaction term indicates that the influence of benefits falls as news exposure climbs. As in the analysis of the 2012 data, the differences across the models cannot be due to measurement, sample size, or the machine’s particular mix of turnout-buying and vote-choice buying attempts.

Table 6.

Vote-Choice Models for Mexico’s 2000 Elections.

	Model 1			Model 2			Model 3			Model 4
	Fox vs. Labastida			Cárdenas vs. Labastida			Fox vs. Labastida			Cárdenas vs. Labastida
Variable	Coef	Sig	SE	Coef	Sig	SE	Coef	Sig	SE	Coef	Sig	SE
PRI vote-buying attempt × Media exposure							0.667	**	0.331	0.902		0.603
PRI vote-buying attempt	−0.096		0.364	0.078		0.570	−1.557	*	0.849	−1.690		1.413
Media exposure							−0.370	**	0.165	−0.909	***	0.312
Labastida vote intention, Feb	−1.202	**	0.530	−1.464		0.929	−1.176	**	0.534	−1.602	*	0.949
Fox vote intention, Feb	2.443	***	0.634	1.619	*	0.986	2.637	***	0.648	1.926	*	1.005
Cárdenas vote intention, Feb	1.264		0.952	3.403	***	1.131	1.370		0.946	3.571	***	1.150
Constant	0.054		0.504	−1.539	*	0.849	0.786		0.610	0.215		1.026

Dependent variable is the final vote choice in July. Labastida is the omitted category. N = 357 matched individuals. Multinomial logistic regressions. PRI ID = Institutional Revolutionary Party identification.

p < .1. **p < .05. ***p < .01.

To underscore the moderating influence of campaign-related information, Figure 5 shows the average marginal effect of PRI benefits on the probability of voting for Labastida over Fox across the range of exposure to mass media news. Benefits recipients that ignored the news were an estimated 24 percentage points more likely to vote for Labastida than those who did not receive an offer (p = .044). If we mildly relax standard levels of statistical significance, then positive effects for the PRI candidate continue until just shy of the 25th percentile on news exposure when selective benefits still incline recipients toward Labastida by 13.1 percentage points (p = .108). The effect on those who pay more attention to the news is indistinguishable from zero, save those above the 95th percentile who react negatively to offers from the PRI by favoring Fox.

Figure 5.

Average marginal effect of PRI vote-buying attempt on presidential vote choice, 2000 (with 95% confidence intervals).

To estimate the yield from vote buying, I sum the product of the nonzero estimates represented in Figure 5 and the corresponding density among respondents who received an offer from the PRI. This procedure yields the overall estimate that offers may have netted up to 1.3 percentage points for Labastida if negative effects are not taken into account, and just 0.43 percentage points if they are. Having made offers to 11.8% of the respondents in the panel survey, this yield implies that 90% to 94% of the party’s offers did not alter vote choices.

Conclusion: Clients Into Citizens

Vote buying is considered a bane of democracy. Where political machines thrive, delegation of power from voters to politicians is inverted and mandates are invalidated, gutting core notions that sustain representative democracy. Effective machines can also monopolize public power for long periods of time, discourage clean politicians from seeking office, and diminish public goods provision. Existing literature suggests that these normatively bad outcomes occur because machines expertly turn citizens into clients.

But modern democracy presents challenges to political machines that go beyond ballot secrecy. Unlike in old-style rural clientelism where citizens had little access to information on competing parties, competitive democracy exposes more citizens to persuasive campaigns that can tug against clientelist outcomes. I show that these influences can cause machines to make unavoidable targeting errors. Brokers pay voters they predict will be good clients based on information they can gather, including voters’ demographics, partisanship, and early vote intentions. But brokers cannot forecast the psychological effects that campaigns exert. Such effects often crystalize close to or even on Election Day and can change vote-relevant attitudes enough to make good clients into bad ones. Machines are especially prone to targeting errors when voters dislike their candidates, meaning that vote buying is least effective when machines need it most.

These conclusions spring from consistent findings across three methods, including ethnography with political bosses and brokers, panel survey data using a direct question to measure vote-choice buying attempts and tried-and-true regression models, and panel survey data using a list experiment and a new statistical methodology designed to uncover the impact of vote-choice buying attempts on voting behavior. The demands of empirical rigor mean that the main tests are limited to Mexico, but if one of the world’s most experienced political machines can scarcely make vote-choice buying effective, it implies that other machines will also struggle.

One naturally wonders why political machines engage in vote-choice buying attempts if their efficiency is so limited. A full theory is beyond the scope of this article, but two considerations may be useful. First, even in the United States where candidates benefit from sophisticated technologies for assessing the efficacy of vote-getting tactics, Kalla and Broockman (2018) show that canvassing and advertising do not affect vote choices. (NB: findings on Mexico and other developing countries show that standard campaigns are highly effective, as detailed above.) If uncovering the efficacy of legal tactics is infeasible for well-heeled candidates in the United States, correctly separating the influence of illegal vote-choice buying attempts from all other potential influences in less developed contexts is presumably far less plausible. Indeed, broker interviews indicate that the main litmus for efficacy is the precinct vote share. Second, brokers have a keen interest in the continued flow of illicit resources for their own livelihood, regardless of their efficiency in winning votes.

If vote-buying attempts persist but are scarcely effective, then the practice becomes significantly less concerning for the quality of electoral democracy. Voters that receive bribes but are unmoved by them can still emit electoral mandates, engage in performance and issue-oriented voting, and hold their electoral representatives accountable. Despite its extensive use, vote buying may not be ruinous for electoral democracy after all.

Supplemental Material

CPS-19-0264_final_online_appendix – Supplemental material for Campaign Effects and the Elusive Swing Voter in Modern Machine Politics

Supplemental material, CPS-19-0264_final_online_appendix for Campaign Effects and the Elusive Swing Voter in Modern Machine Politics by Kenneth F. Greene in Comparative Political Studies

Footnotes

Author’s Note

This article had a long gestation, during which time I incurred more debts to colleagues than I can list here. Exceptionally generous among them were Andy Baker, Mark Williams, Angelo Cioffi, Gerry Munck, Alberto Simpser, Mariano Sánchez-Talanquer, Luis Fernando Medina, Rodrigo Zarazaga, Andreas Schedler, Kosuke Imai, Gilles Serra, Noam Lupu, John Gerring, Ruth Collier, Thad Dunning, Zeynep Somer-Topcu, Mariela Szwarcberg, Milan Svolik, Ernesto Calvo, Ignacio Sánchez-Cuenca, Zachary Elkins, Horacio Larreguy, Jim Adams, Cesi Cruz, Ryan Carlin, anonymous reviewers, and the editors of CPS. All errors are my own. Financial support for this research came from the Mellon Foundation, The Lozano Long Institute of Latin American Studies, The Juan March Institute, and Carlos III University.

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Kenneth F. Greene

Supplemental Material

Supplemental material for this article is available online at the CPS website

Notes

Author Biography

Kenneth F. Greene studies political regimes, political parties, and voting behavior. His first book, Why Dominant Parties Lose: Mexico�s Democratization in Comparative Perspective (Cambridge University Press, 2007) won the Best Book Award from the Comparative Democratization Section of the American Political Science Association. His recent work focuses on the efficacy of electoral bribes in influencing voter attitudes and behavior.

References

Adams

Merrill

(2009). Policy-seeking parties in a parliamentary democracy with proportional representation: A valence-uncertainty model. British Journal of Political Science, 39(3), 539–558.

Adams

Merrill

Grofman

(2005). A unified theory of party competition. Cambridge University Press.

Alianza Cívica. (2012). Boletín de Prensa, 3 julio. www.alianzacivica.org.mx/archivos/pub/9557Informe%203%20de%20julio%20.pdf

Auyero

(2000). The logic of clientelism in Argentina: An ethnographic account. Latin American Research Review, 36(1), 55–81.

Baker

Ames

Renno

(2006). Social context and campaign volatility in new democracies: Networks and neighborhoods in Brazil’s 2002 elections. American Journal of Political Science, 50(2), 382–399.

Banerjee

Kumar

Pande

(2011). Do informed voters make better choices? Experimental evidence from urban India. http://scholar.harvard.edu/files/rpande/files/doinformedvoters_nov11_0.pdf

Bartels

(1993). Messages received: The political impact of media exposure. American Political Science Review, 87(2), 267–285.

Begg

Gray

(1984). Calculation of polychotomous logistic regression parameters using individualized regressions. Biometrika, 71(1), 11–18.

Beltrán

(2007). The combined effect of advertising and news coverage in the Mexican presidential campaign of 2000. Political Communication, 24, 1, 37–63.

10.

Beltrán

Castro Cornejo

(2020). Who receives electoral gifts? sensitive behavior & question wording effects. Political Behavior, forthcoming.

11.

Blair

Imai

(2012). Statistical analysis of list experiments. Political Analysis, 20(1), 47–77.

12.

Blattman

Larreguy

Marx

Reid

(2017). A market equilibrium approach to reduce the incidence of vote-buying: Evidence from Uganda. https://sites.google.com/site/hlarreguy/research

13.

Boas

Hidalgo

F. D.

(2011). Controlling the airwaves: Incumbency advantage and community radio in Brazil. American Journal of Political Science, 55(4), 869–885.

14.

Calvo

Murillo

M. V.

(2019). Non-policy politics: Richer voters, poorer voters, and the diversification of electoral strategies. Cambridge University Press.

15.

Cantú

(2019). Groceries for votes: The electoral returns of vote buying. The Journal of Politics, 81(3), 790–804.

16.

Çarkoğlu

Aytaç

(2015). Who gets targeted for vote-buying? Evidence from an augmented list experiment in Turkey. European Political Science Review, 7(4), 547–566.

17.

Carlin

Moseley

(2015). Good democrats, bad targets: Democratic values and clientelistic vote buying. The Journal of Politics, 77(1), 14–26.

18.

Coppedge

Gerring

Knutsen

C. H.

Lindberg

S. I.

Teorell

Altman

. . . Ziblatt

(2019). V-dem dataset v9 [Data set]. Varieties of Democracy (V-Dem) Project. https://www.v-dem.net/en/data/data-version-9/

19.

Corstange

(2009). Sensitive questions, truthful answers? Modeling the list experiment with LISTIT. Political Analysis, 17(1), 45–63.

20.

Corstange

(2012). Vote trafficking in Lebanon. International Journal of Middle East Studies, 44(3), 483–505.

21.

Cox

McCubbins

(1986). Electoral politics as a redistributive game. The Journal of Politics, 48(2), 370–389.

22.

Cruz

Keefer

Labonne

(2016, July 2). Incumbent advantage, voter information and vote buying (IDB Working Paper No. IDB-WP-711). Inter-American Development Bank.

23.

Diaz-Cayeros

Estevez

Magaloni

(2016). Strategies of vote buying: Democracy, clientelism, and poverty relief in Mexico. Cambridge University Press.

24.

Dixit

Londregan

(1996). The determinants of success of special interests in redistributive politics. The Journal of Politics, 58(4), 1132–1155.

25.

Dow

Endersby

(2004). Multinomial probit and multinomial logit: A comparison of choice models for voting research. Electoral Studies, 23, 107–122.

26.

Enelow

Hinich

(1989). A general probabilistic spatial theory of elections. Public Choice, 61(2), 101–113.

27.

Erikson

Romero

(1990). Candidate equilibrium and the behavioral model of the vote. American Political Science Review, 84(4), 1103–1126.

28.

Finkel

(1993). Reexamining the “minimal effects” model in recent presidential campaigns. The Journal of Politics, 55(1), 1–21.

29.

Fournier

Nadeau

Blais

Gidengil

Nevitte

(2004). Time-of-voting decision and susceptibility to campaign effects. Electoral Studies, 23(4), 661–681.

30.

Fujiwara

Wantchekon

(2013). Can informed public deliberation overcome clientelism? Experimental evidence from Benin. American Economic Journal: Applied Economics, 5(4), 241–255.

31.

Gonzalez-Ocantos

Kiewiet de Jonge

Meléndez

Osorio

Nickerson

(2012). Vote buying and social desirability bias: Experimental evidence from Nicaragua. American Journal of Political Science, 56(1), 202–217.

32.

Green

Vasudevan

(2016). Diminishing the effectiveness of vote buying: Experimental evidence from a persuasive radio campaign in India. https://www.povertyactionlab.org/sites/default/files/publications/votebuying_paper.pdf

33.

Greene

(2011). Campaign persuasion and nascent partisanship in Mexico’s new democracy. American Journal of Political Science, 55(2), 398–416.

34.

Hart

(2016). Economic voting: A campaign-centered theory. Cambridge University Press.

35.

Hicken

Leider

Ravanilla

Yang

(2015). Measuring vote-selling: Field evidence from the Philippines. The American Economic Review, 105(5), 352–356.

36.

Iacus

King

Porro

(2012). Causal inference without balance checking: Coarsened exact matching. Political Analysis, 20(1), 1–24.

37.

Imai

Park

Greene

(2015). Using the predicted responses from list experiments as explanatory variables in regression models. Political Analysis, 23(2), 180–196.

38.

Kalla

Broockman

(2018). The minimal persuasive effects of campaign contact in general elections: Evidence from 49 field experiments. American Political Science Review, 112(1), 148–166.

39.

Kiewiet de Jonge

Nickerson

(2014). Artificial inflation or deflation? Assessing the item count technique in comparative surveys. Political Behavior, 36(3), 659–682.

40.

Kitschelt

Altamirano

(2015). Clientelism in Latin America: Effort and effectiveness In Carlin

Singer

Zechmeister

(Eds.), The Latin American voter (pp. 246–273). University of Michigan Press.

41.

Kitschelt

Wilkinson

(2007). Patrons, clients and policies: Patterns of democratic accountability and political competition. Cambridge University Press.

42.

Kramon

(2016). Electoral handouts as information: Explaining unmonitored vote buying. World Politics, 68(3), 454–498.

43.

Larreguy

Marshall

Querubín

(2016). Parties, brokers and voter mobilization: How turnout buying depends upon the party’s capacity to monitor brokers. American Political Science Review, 110(1), 160–179.

44.

Lawson

Basanez

Moreno

Poire

Paras

Camp

McCann

Mexican Election Panel Study. (2000) [Data set]. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 2008-01-28. https://doi.org/10.3886/ICPSR03380.v1

45.

Lawson

McCann

(2005). Television coverage, media effects, and Mexico’s 2000 elections. British Journal of Political Science, 35(1), 1–30.

46.

Lenz

(2009). Learning and opinion change, not priming: Reconsidering the evidence for the priming hypothesis. American Journal of Political Science, 53(4), 821–838.

47.

Lindbeck

Weibull

(1987). Balanced-budget redistribution as the outcome of political competition. Public Choice, 52, 273–297.

48.

Mares

Young

(2016). Buying, expropriating, and stealing votes. Annual Review of Political Science, 19(1), 267–288.

49.

Nichter

(2018). Votes for survival: Relational clientelism in Latin America. Cambridge University Press.

50.

Petrocik

(1996). Issue ownership in presidential elections, with a 1980 case study. American Journal of Political Science, 40(3), 825–850.

51.

Price

Zaller

(1993). Who gets the news? Alternative measures of news reception and their implications for research. Public Opinion Quarterly, 57(3), 133–164.

52.

Rosenfeld

Imai

Shapiro

(2016). An empirical validation study of popular survey methodologies for sensitive questions. American Journal of Political Science, 60(3), 783–802.

53.

Rubin

(1987). Multiple imputation for nonresponse in surveys. John Wiley.

54.

Rueda

(2015). Buying votes with imperfect local knowledge and a secret ballot. Journal of Theoretical Politics, 27(3), 428–456.

55.

Schneider

(2019). Do local leaders know their voters? A test of guessability in India. Electoral Studies, 61, 1–12.

56.

Singer

Kitschelt

(2011). “Do Everything” (DoE) parties: When can politicians combine clientelistic and programmatic appeals? https://sites.duke.edu/democracylinkage/files/2014/12/3.3.Kitschelt_Singer.pdf

57.

Stokes

(2005). Perverse accountability: A formal model of machine politics with evidence from Argentina. American Political Science Review, 99(3), 315–325.

58.

Stokes

Dunning

Nazareno

Brusco

(2013). Brokers, voters, and clientelism: The puzzle of distributive politics. Cambridge University Press.

59.

Szwarcberg

(2015). Mobilizing poor voters. Cambridge University Press.

60.

Tribunal Electoral de la Procuraduría Judicial Federal [Electoral Tribunal of the Federal Judiciary]. (2012). Juicio de Inconformidad SUP-JIN-0359-2012. https://docs.mexico.justia.com/federales/sentencias/tribunal-electoral/2012-08-15/sup-jin-0359-2012-inc3.pdf

61.

Vicente

(2014). Is vote-buying effective? Evidence from a field experiment in West Africa. The Economic Journal, 124(574), 356–387.

62.

Weghorst

Lindberg

(2013). What drives the swing voter in Africa? American Journal of Political Science, 57(3), 717–734.

63.

Weitz-Shapiro

(2014). Curbing clientelism in Argentina. Cambridge University Press.

64.

Zarazaga

(2014). Brokers beyond clientelism: A new perspective through the Argentine case. Latin American Politics and Society, 56(3), 23–45.

65.

Zarazaga

(2016). Party machines and voter-customized rewards strategies. Journal of Theoretical Politics, 28(4), 678–701.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.91 MB