Introduction

Mexico’s transition from single-party dominance to democracy focused on leveling the playing field so that opposition candidates could compete on equal footing with incumbents. The 2018 elections provide strong evidence that these efforts were successful. Andrés Manuel López Obrador, leading the upstart Morena party, swept into office with the most decisive win since the 1982 presidential election. Thanks to world-leading electoral management institutions, gone are the days when outcome-changing electoral fraud could deny opposition candidates victories they earned at the polls. But despite these advances, problems in the conduct of elections lurk below the surface. Not only have systematic schemes to purchase citizens’ electoral support survived the transition to fully competitive democracy, the use of electoral clientelism may have increased over time.¹ This article provides a detailed description of the practice of electoral clientelism in the 2018 general elections based on original survey data.

Overall, we find that Mexico’s 2018 general elections were awash in electoral handouts. Over 42 per cent of the eligible voters in our panel survey data report that they were offered some good or service by a political party during the campaign season, excluding small gifts that could be interpreted as campaign advertising. (Including all offers reaches 52.9%).² A whopping 83.7 per cent of these citizens were asked, in exchange, to vote for a particular candidate, to turn out, or to stay home on election day.³ All political parties distributed handouts, though the PRI-led coalition did so the most, followed by the PAN-led coalition. Morena’s coalition partook in the attempt to buy votes too, albeit as a minor player.

Our survey data, collected before and after the 2018 general elections, show that, despite manifold attempts to buy votes, such attempts were likely unsuccessful. Building on recent specialized literature that questions the efficacy of vote buying (^{Stokes et al., 2013}; ^{Schneider, 2019}; ^{Greene, 2018}), we show that many recipients had a muddled understanding of what they were asked to do in exchange, that most recipients ascribed limited value to the handouts they received, and that out of those recipients who understood what was asked of them, few feared sanctions for non-compliance. Several elements that analysts have argued are essential for vote buying to alter recipients’ behavior (^{Stokes, 2005}; ^{Kitschelt and Wilkinson, 2007}; ^{Dixit and Londregan, 1996}) were thus missing or in short supply.

The first section of this article discusses measurement challenges and presents estimates of the extent of vote-buying attempts in Mexico since 2000. The second section reports a detailed inventory of the handouts that citizens were offered during the 2018 election season. Our empirical findings are largely based on the Mexico Elections and Quality of Democracy Survey (EQD), an original survey administered to a nationally representative group of eligible voters in May and June of 2018, and then again after the July 1^st elections. The conclusion draws out the implications of our findings for the quality of elections in Mexico’s new democracy.

What do we know about vote buying? detecting electoral handouts

Virtually all observers of Mexico’s politics know that electoral handouts are a routine part of the campaigns, even after the transition to fully competitive democracy (^{Aparicio, 2017}; ^{Becerra, 2012}; ^{Beltrán and Castro Cornejo, 2015}, ²⁰¹⁹; ^{Buendía and Somuano, 2003}; ^{Casar and Ugalde, 2018}; ^{Cornelius, 2003}; ^{De la O, 2015}; ^{Díaz- Cayeros et al., 2016}; ^{Greene, 2018}; ^{Hilgers, 2008}; ^{Larreguy et al., 2016}; ^{Nichter and Palmer-Rubin, 2015}; ^{Schedler, 2004}; ^{Serra, 2016}; ^{Simpser, 2012}, ²⁰¹³; ^{Szwarcberg, 2015}). Nevertheless, many basic questions about the vote buying enterprise remain unanswered for lack of systematic information. For instance:

It is admittedly challenging to elicit this kind of information from citizens. After all, the explicit exchange of material rewards for electoral support is illegal in Mexico and many implicit exchanges are viewed as illicit, and socially stigmatized, by the participants (^{Schedler, 2004}). Consequently, recipients of electoral handouts may be reluctant to divulge their participation in the practice (^{Beltrán and Castro Cornejo, 2020}). Just as importantly, surveys seldom collect information with enough detail to answer many of the questions listed above.

Ethnography by talented fieldworkers who genuinely gain the trust of their subjects may be the best technique for eliciting truthful answers to sensitive questions, but this approach cannot easily provide information that is representative of an electorate (^{Auyero, 2000}; ^{Hilgers, 2009}; ^{Lomnitz, 1982}; ^{Rizzo, 2015}; ^{Schedler, 2004}; ^{Stokes et al., 2013}; ^{Szwarcberg, 2015}; ^{Zarazaga, 2014}). Field experiments that attempt to suppress vote-buying or citizens’ compliance with political machines’ wishes are able to shed light on the overall electoral effects at the constituency level, but research utilizing this approach has so far focused on estimating reduced- form causal effects of specific interventions on vote totals, not on describing the extent and characteristics of vote-buying efforts (^{Banerjee et al., 2011}; ^{Blattman et al., 2017}; ^{Cruz et al., 2016}; ^{Green and Vasudevan, 2016}; ^{Fujiwara and Wantchekon, 2013}; ^{Hicken et al., 2017}; ^{Kramon, 2016}; ^{Vicente, 2014}).

In recent years, researchers have employed sample surveys as a means of collecting nuanced data on individuals that can be representative at the national level. However, analysts have not reached consensus about the best questioning technique for detecting electoral handouts. Questionnaires have employed two types of direct questions. In one approach, respondents are asked some version of: “Have you received a good, service, or favor from a candidate or political party?” Questions such as this that do not mention an explicit quid pro quo have appeared in the Mexico 2000 and 2006 Panel Studies (^{Lawson et al., 2000}, ²⁰⁰⁶; ^{Cornelius, 2003}) and ^{the Comparative National Elections Project (CNEP) survey for Mexico in 2012} (^{Moreno, 2012}). This approach may cast too wide a net, including affirmative responses when citizens receive a policy-based benefit or when they receive campaign advertisements of little material value such as a pen or hat.

In another use of direct questions, respondents are asked some version of: “Have you received a good, service, or favor in exchange for your vote?” Items that include this kind of explicit quid pro quo have appeared in the Comparative Study of Electoral Systems survey for Mexico in 2000 ( ^{used in Buendía and Somuano, 2003}), the Americas Barometer (LAPOP) survey in 2010 (^{used in Faughan and Zechmeister, 2011}), the Mexico 2012 Panel Study (^{Greene et al., 2012}), and the ^{CNEP for Mexico in 2018} (^{Moreno, 2018}).⁴ Yet this approach often yields estimates of the rate of vote buying that observers believe are too low. Moreover, research suggests that respondents that are more knowledgeable about the law are less likely to answer direct questions truthfully (^{Kiewiet and Nickerson, 2014}), potentially leading to inferential bias when using vote buying either as an outcome or an explanatory variable.

Indirect questioning techniques seek to overcome the social desirability bias that is likely embedded in responses to direct questions (especially those that mention an explicit exchange) by allowing respondents to communicate information about their behaviors without openly admitting to them. One such approach is the list experiment where randomly selected members of the control group receive a list of J non-sensitive behaviors and members of the treatment group receive a list of J+1 where the extra item is some version of “I received a good, service, or favor in exchange for my vote”. Under specific assumptions, the difference in the mean number of items checked off in one versus the other list reveals the aggregate rate of respondent involvement in vote buying exchanges (^{Blair and Imai, 2012}). This approach typically leads to higher estimates of vote buying than direct questions (^{Çarkoğlu and Aytaç, 2015}, ^{Corstange, 2009}; ^{González-Ocantos et al., 2012}; ^{Nichter and Palmer-Rubin, 2015}). Recent advances permit researchers to diagnose the degree of remaining social desirability bias (^{Simpser, 2017}) and to use list-experiment data as a dependent variable (^{Blair and Imai, 2012}) or an independent variable in outcome regression models (^{Imai et al., 2015}).

Nevertheless, any gain in respondent privacy comes at the cost of the information available to the researcher. During the interview, enumerators cannot know whether any one particular respondent did or did not engage in vote buying. Consequently, researchers cannot ask detailed follow-up questions that would illuminate which parties give handouts, what items citizens receive from them, and what citizens are asked to do in return. Moreover, for reasons of time and cost researchers rarely ask about handouts from multiple parties, obscuring the degree of competitive clientelism. In addition, indirect questioning techniques are cognitively more demanding than direct questions and therefore cause their own measurement problems. Enumerators or respondents can become confused, for example, leading to poor administration and, at times, nonsensical results (i.e., negative estimates of the prevalence of vote buying). High cognitive demands may also cause greater measurement problems among poor and less-educated respondents —precisely those that some literature predicts should be the prime targets of vote-buying efforts (^{Beltrán and Castro Cornejo, 2020}, ^{Holbrook and Krosnick, 2010a}, ^2010b).

We point to three additional limitations of existing approaches to measuring vote buying that cut across questioning techniques. As noted above, all of the indirect questions and many of the direct questions used in Mexico (and elsewhere) have included the explicit quid pro quo “in exchange for your vote”. This may spark response bias to a greater degree than questions that do not specify the quid pro quo.⁵ Second, whether or not a quid pro quo is mentioned, this kind of question is too blunt in light of basic distinctions that have become standard in the literature —for example, between vote-choice buying, turnout buying, and abstention buying (^{Gans-Morse et al., 2014}). Third, questions used in Mexico have asked respondents whether they “received” something, thus filtering out those respondents who were offered a payoff but refused it as well as those who were promised a benefit after election day. These latter categories could account for a significant portion of the parties’ handout activity during campaign periods. Moreover, receiving, refusing, or being promised a handout could conceivably have different effects on a host of outcomes of interest, including turnout behavior and vote choice.

As an improvement on existing approaches, we deployed a modified direct questioning technique in our Mexico EQD Survey that is easy to understand for enumera- tors and respondents, maintains the descriptive detail of standard direct questioning, and aims to diminish certain forms of social desirability bias. Our approach yields a full inventory of the material benefits and services that citizens received, were offered but rejected, or were promised from all the parties during the campaign season. We fielded these questions to a nationally-representative sample of ordinary Mexican citizens eligible to vote in May and June of 2018 (N=1 310), and then re-interviewed the same sample of citizens after the July 1st elections (N=583).⁶

Enumerators began by reading the following preamble:

Now I would like to ask you about your own experiences in 2018. Sometimes, the political parties give groceries, cash, gift cards, construction materials, water cisterns, medicine, or they may help get access to government programs like Prospera or Seguro Popular, give educational subsidies, or medical attention. They also may offer jobs or legal services. In the following questions, I would like to know about your experience with these sorts of things in 2018. To guarantee your privacy, I am going to give you my phablet so you can answer the questions in private. Like before, push the green button after each response. When you are finished, hand me the phablet. I want to remind you that all your answers are confidential.

Our procedure was designed to diminish social desirability bias that may arise due to the personal interaction between enumerator and respondent. The preamble communicated that survey researchers share with the respondent knowledge about the existence of electoral handouts. In addition, respondents were handed the phablet so that they could answer the questions by themselves out of the view of the enumerator. As they advanced from the practice screen, respondents could observe that each prior response would disappear from the screen, effectively hiding their answer from the enumerator and therefore offering a sense of privacy.

We represent the flow logic of our set of questions about electoral handouts in Figure 1 and refer to other follow-up questions in the text. The full wording appears in the Appendix. We believe that these simple and straightforward questions improve survey administration, diminish response bias and increase measurement validity.

Source:^{Greene and Simper (2018)}.

Figure 1 Electoral handout questions in the Mexico 2018 EQD survey  

Before describing our results we briefly review prior findings on vote buying in Mexico. There are no clear rules for comparing unlike surveys that use different questioning techniques, draw samples at different times with respect to election day, and interview respondents once versus more than once. Nevertheless, to provide a sense of the prior findings on vote buying in Mexico, Table 1 shows estimates for the last four general elections from the sample survey datasets we can access. Direct questions with an explicit quid pro quo (in italics in Table 1) registered levels of vote-buying attempts as low as 5.9 per cent in 2012 and as high as 17.9 per cent in 2018. Direct questions without an explicit quid pro quo generally showed higher levels, ranging from 16.7 per cent in 2010 off the election cycle to 26 per cent in 2000 to 51 per cent in 2015.⁷The most straightforward comparisons between items with and without a quid pro quo come from the 2000 and 2012 election cycles. Both items were asked, albeit in different surveys and at different points in their respective electoral cycles. In these contests, the questions without a quid pro quo registered a much higher rate than those with a quid pro quo. Interestingly, the surveys in 2012 also employed a list experiment. Even though this indirect question contained an explicit quid pro quo that typically diminishes vote-buying estimates, it yielded a higher estimate than any of the other approaches fielded in that election year at 21.2 per cent. Presumably, a question without a quid pro quo that also maintained confidentiality would have yielded even higher estimates.

We used such an approach in the Mexico ^{2018 EQD}. Our “modified direct question” yielded the highest levels of vote buying recorded in a national election in Mexico using any technique. Our question does not include a quid pro quo clause and it strives to maintain the respondents’ sense of privacy at the moment of response. At the same time, it avoids the challenges of indirect questioning techniques. Using this approach, we find that 52.9 per cent of panel respondents were offered a benefit, whether they accepted it or not. For much of the analysis below, we exclude respondents who were only offered a small gift or did not specify the type of handout they were offered, referring to the remaining group as “targeted citizens,” as mentioned previously. Limiting our estimate to targeted citizens here brings our estimate for panel respondents down to 42.4 per cent (see footnote 2 for details).

Direct, no quid pro quo

Direct, quid pro quo

Indirect, quid pro quo

Modified direct, no quid pro quo

We next describe the variety of benefits offered to citizens and what they were asked to do in exchange for these, giving the fullest picture to date of the competing parties’ use of material incentives during elections.

Handouts in the 2018 election: evidence from the Mexico 2018 EQD panel survey

Mexico’s recent electoral history is dotted with stories about spectacular handouts and systematic schemes to buy electoral support. In 2000, the governor of Yucatán, Víctor Cervera Pacheco, was accused of distributing washing machines in exchange for votes ^{(Proceso, 2003)}. Following the 2012 elections, Andrés Manuel López Obrador brought suit in the Federal Electoral Tribunal (TEPJF), alleging that the victorious Peña Nieto campaign had bought votes with everything from goats to tens of thousands of cash cards redeemable at Soriana grocery stores (^{TEPJF SUP- JIN-0359-2012}; ^{Cantú, 2019}). In the 2018 elections, reports to Acción Ciudadana Contra la Pobreza included claims of receiving up to 5000 pesos as an electoral handout (^{ACP, 2018}).

These dramatic anecdotes notwithstanding, most electoral handouts are more mundane. Figure 2 reports the types of handouts that targeted citizens reported being offered in either wave of the Mexico EQD survey.

Source:^{Greene and Simper (2018)}. Notes: N = 220 “targeted citizens”. Of the 583 panel respondents, 10 did not respond to the questions about electoral handouts, 270 (47.1%) did not receive any offer, 29 (5.1%) were only offered a small gift (and are excluded from this figure), and 54 did not specify what they were offered, yielding 220 targeted citizens. Bars do not sum to 100 per cent because respondents could report multiple offers. Because we excluded those respondents who only reported receiving a small gift, the “Small gift” bar in the figure represents the percentage of targeted citizens who received a small gift in addition to another larger benefit. Mexico ^{2018 EQD} Panel Survey.

Figure 2 Inventory of benefits received, refused, and promised in 2018  

Groceries appear to be the main currency of the campaigns. Over 66 per cent of targeted citizens were offered groceries. Over 52 per cent of targeted citizens were offered cash. On the basis of other sources, it appears that the modal cash offer in 2018 was 500 pesos (about USD 26) (^{ACP, 2018}). Cash cards are reportedly of similar value but, compared to 2012, they may have played a smaller role in 2018 (when our data indicate that “only” 10.9 per cent of targeted citizens were offered an active cash card and 10.5 per cent an empty one with the promise it would be activated after the election).

Other benefits likely require more selectivity and planning by the parties and involve active participation by clients in the weeks and months before election day. These include home construction materials like cement, brick, plastics, water cisterns, paint, and electrodomestic items (offered to 26.4% of targeted citizens); school supplies or minor scholarships (11.4%); work that often involves short-term employment in canvassing and get-out-the-vote efforts (12.7%); and medical attention, such as medicines and eyeglasses (5.9%).

Which political parties distributed handouts?

Research on vote buying often argues that one political machine —generally the incumbent’s— dominates the provision of electoral handouts (^{Kitcshelt and Wilkinson, 2007}; ^{Stokes, 2005}; ^{Dixit and Londregan, 1996}; ^{Cox and McCubbins, 1986}; ^{Gans- Morse et al., 2014}). Other literature, however, has considered the possibility that multiple parties attempt to buy votes (^{Simpser, 2013}) and documented cases of competitive vote buying (^{Wurfel, 1963}). Whether one or more parties offer handouts is an empirical question, but existing surveys seldom ask which party or parties supplied them.

Mexico’s history of single-party dominance makes it a likely case of a single machine environment. Indeed, before 2000, the PRI ran what ^{Cornelius and Craig (1991)} called “a nationwide reward and punishment system”, and as late as 2012, data from the direct question in the Mexico 2012 Panel Study showed that 78 per cent of those who received a benefit in exchange for the vote were provided the benefit by the PRI (^{Greene, 2018}). But fieldworkers have increasingly noted that other parties are distributing more electoral handouts (^{Casar and Ugalde, 2018}; ^{Hilgers, 2008}). This development may be driven both by increasing access to resources as parties win more elected offices and by increasing voter demand. In our 2018 survey, nearly 25 per cent of respondents said that they asked for the benefits they received ( ^{see also Nichter, 2018 on Brazil}).

Our data document what might be termed a relative “democratization” of hand-out provision in the 2018 elections. Figure 3 shows the proportion of targeted citizens offered a benefit by each party. The PRI made offers to 69.4 per cent of this group. Adding offers made by its coalition allies —the Partido Verde Ecologista de México (PVEM), Nueva Alianza (Panal), and an unspecified member of the Todos por México (TPM) coalition— the figure reaches 88.9 per cent. (Note that some targeted citizens received offers from multiple members of each coalition.)

Source:^{Greene and Simper (2018)}. Notes: Based on 206 targeted citizens. 14 targeted citizens did not specify which party made them the offer. Mentions of different parties by the same respondent were counted separately. The coalition totals count targeted citizens only once per coalition so that offers from more than one party in the same coalition are not reflected in the figure. Mexico ^{2018 EQD} Panel Survey.

Figure 3 Handouts offered, by political party and coalition  

Other parties distributed notable amounts of benefits as well. The pan and PRD made offers to 35 per cent and 30.6 per cent of targeted citizens, respectively. Taken together, these parties and their coalition allies —the mc and an unspecified member of the Por México al Frente (PMF) coalition— made offers to 76.7 per cent of targeted citizens. Also notable is that Morena, its coalition partner the PT, and an unspecified member of the Juntos Haremos Historia (JHH) coalition made offers to “just” 15.5 per cent of targeted citizens. (No respondents mentioned the pes.) The winning coalition’s lower level of handout offers could reflect a moral and/or strategic decision, or it could have resulted from the simple fact that its component parties occupied fewer government offices that could provide resources for handouts.

What did recipients believe they were asked to do in return?

What counts as an illegal electoral handout is not straightforward under Mexico’s law. By itself, it is not illegal to distribute benefits or services of any monetary value to citizens, as long as the resources that fund them are of legal origin. However, Article 7 of the Law on Electoral Crimes establishes that it is illegal for anyone to: “Request votes for payment, promise of money or other consideration, or by violence or threat, to press anyone to attend proselytizing events, or to vote or abstain from voting for a candidate, political party or coalition” and Articles 7 and 11 make it illegal to condition voting behavior on the provision or suspension of benefits from social programs or any other “public service, compliance with government programs, granting concessions, permits, licenses, authorizations, franchises, or exemptions, or the creation of public works”. The law does not define what counts as soliciting or conditioning votes on the provision of benefits.

In practice, recipients’ perceptions of the rationale behind the offer of benefits probably matters most for voting behavior. That is, if recipients are compelled to comply with the machines’ wishes, presumably they would comply with what they believe they were asked to do. Figure 4 shows the activities that targeted citizens thought were asked of them in exchange for the benefit offered. Multiple responses were permitted, so the data represent the proportion of recipients asked to do each activity; however, we only solicited information about the behavior requested by the party that offered what the recipient believes was the most valuable benefit. Thus, requests by other parties and coalitions are not reflected in the survey responses or in Figure 4.

Source:^{Greene and Simper (2018)}. Notes: Based on 215 targeted citizens (5 targeted citizens did not specify which activities were requested of them). Multiple response options permitted. Mexico ^{2018 EQD} Panel Survey. White bars denote activities that may be legal to buy, depending on the circumstances and origin of the funds. Solid and striped bars denote activities that are illegal to buy under current electoral law. Striped bars denote contradictory behavior on the part of the buying party.

Figure 4 Perceived rationale for benefits offers (multiple responses permitted)  

White bars indicate activities that could be legal for providers to request of recipients, depending on the origin of the benefits offered. Interestingly, 14 per cent of recipients believed that their benefactors requested nothing in return, making this sort of transaction legal but a presumably inefficient form of machine politics. Other requests could also be construed as a legal use of resources, including monitoring whether others voted (6.0%) and canvassing (20.9%). Indeed, the PRI’S argument following the 2012 elections was that it distributed Soriana gift cards to its activists who were employed as canvassers and poll watchers.

In the main, however, those who received offers believed they were being asked to engage in activities that political parties are not legally permitted to buy, represented by solid bars. Some were asked for their vote choice only (24.7%) and many for both their participation at the polls and their vote choice (33.5%). Presumably, these requests would be made of initial opponents, implying that the parties mostly sought to win votes away from their competitors. A much smaller proportion —8.4 per cent— were asked to participate in the election, implying less emphasis on turning out loyalists. A smaller still but notable group —2.3 percent— were asked to abstain on election day.

The final group of activities, represented in striped bars, remain illegal for political parties to buy but are irrational from the perspective of machine politics. Some 1.4 per cent were asked to turn out and to abstain, 4.7 per cent were asked for their vote choice but also to abstain, and 8.8 per cent were asked to complete all three activities. These responses could be chalked up to classic survey measurement error, but they could also indicate a confused clientele.

Do handouts improve electoral performance? circumstantial evidence

Mexico’s elections are clearly awash in attempts to influence electoral behavior through the provision of selective benefits, but four elements should temper the concern that the national-level election outcomes are routinely altered by vote buying attempts.

First, the findings above fail to distinguish whether citizens accepted, refused, or were promised benefits that had not been delivered by the time they were surveyed. Figure 5 shows that 42.7 per cent of targeted citizens said they rejected what they were offered. Future research might investigate the degree to which social desirability bias influences these responses. Taken at face value, however, it is notable that just 36.8 per cent of those who were offered something said they accepted it.

Source:^{Greene and Simper (2018)}. Notes: Based on all targeted citizens (N=220). Responses recorded in each survey wave. Only targeted citizens that rejected all offers they received are counted as “Rejected all offers”. Mexico ^{2018 EQD} Panel Survey.

Figure 5 Reception, acceptance, and rejection of handout offers  

It is also useful to note that the lion’s share of handouts is offered before election day rather than after the fact. Promises for later distribution may not be viewed as credible. For instance, following the 2012 election, numerous recipients of the Soriana cards reported that they either contained no money or substantially less than they had been promised. In our 2018 survey, 20.5 per cent of those who received an offer were promised that they would benefit after the election, yet just one person reported that they eventually received the promised benefit.

Second, even among those who reported having accepted the handout that was offered to them, the benefit they received may not have been sufficient to alter their electoral behavior. Figure 6 describes the subjective value of the most valuable electoral handout that targeted citizens were offered. Perhaps surprisingly, nearly half of recipients estimated that it would be somewhat (29.1%) or very (19.2%) easy to raise resources equivalent to the handout they were offered. Another 35.5 per cent said that it would be somewhat difficult and 16.3 per cent said it would be very difficult or impossible.

Source:^{Greene and Simper (2018}). Notes: Based on 203 targeted citizens. 17 targeted citizens did not answer the relevant questions. Respondents were asked how difficult it would be to raise resources equivalent to the most valuable gift they received. Mexico ^{2018 EQD} Panel Survey.

Figure 6 Subjective value of electoral handouts  

Figure 7 probes further by showing what percentage of those who were offered a specific item also reported that buying the item for themselves would be difficult or impossible. For example, 42.5 per cent of those who were offered groceries reported that it would be difficult or impossible for them to raise equivalent resources, while almost 82 per cent of those who received medical attention, medicine, or eyeglasses deemed it difficult or impossible to raise equivalent resources. Overall, the benefits that were infrequently cited as offers (Figure 2) such as enrollment in a government-sponsored social program, household materials and electrodomestic goods, free medical attention, and help with a bureaucratic task, were considered much more valuable by more respondents. The larger value of these services could reflect their market value, but may also reflect the relative difficulty of obtaining them where recipients live.

Source:^{Greene and Simper (2018)}. Notes: For each bar, the denominator is the set of targeted citizens who reported receiving the corresponding benefit, and the numerator is the subset of these who reported that it would be difficult or impossible for them to buy the benefit themselves. Mexico ^{2018 EQD} Panel Survey.

Figure 7 Offered items that would be difficult to obtain from other sources  

The value ascribed to two of the items listed in Figure 7 is particularly telling. On the one hand, the fact that more than 40 per cent of those who received groceries or cash deemed these difficult or impossible to buy suggests that these recipients comprised very-low-income populations. On the other hand, the fact that only 45.7 per cent of those who received a cash offer similarly rated it as difficult or impossible to raise equivalent resources implies that clients are less economically dependent than frequently assumed. Interviews and observation by one of the authors with local brokers and voters, news accounts, and 707 citizen reports to a Mexico-based election watchdog organization suggest that 500 pesos was the modal offer among those who were offered cash in 2018 (^{Telemundo, 2018}; ^{Solís, 2018}; ^{ACP, 2018}).⁸

This amount represents 5.6 days of work at the 2018 minimum wage and almost a day and half of work at the average daily wage. Although one might expect this amount to represent significant value for poor recipients, survey responses imply that citizens are much less dependent on political machines than the image conjured by historical studies of 19th or early 20th century agrarian clientelism when recipients’ life chances were strongly conditioned by their patrons’ largesse (^{Baland and Robinson, 2008}). More research is needed to determine how much is enough to alter electoral behavior (^{Becerra, 2012}).

Third, the proportion of recipients who alter their behavior in response to a material offer is likely further reduced by limits on the mechanisms that machines use to ensure that clients comply with their wishes. Analysts have argued that compliance with vote-buying transactions relies on the threat of a cost to be exacted against defecting voters (^{Stokes, 2005}; ^{Kitschelt and Wilkinson, 2007}). Our survey asked: “If a citizen accepts a gift from a political party in exchange for their vote but they don’t vote for that party, do you think it will cause them a big problem, a little problem, or no problem at all?” [ “Si un ciudadano acepta un regalo de un partido político a cambio de su voto, pero no vota por ese partido, ¿cree usted que le causaría problemas serios, problemas menores, o no le causaría ningún problema?” ]. In the post-election wave of our survey, 10.6 per cent of targeted citizens said they would experience a serious problem, 11.5 per cent said they would have a minor problem, and a full 77.9 per cent said they would not experience any problem at all. These figures cast doubt on the threat of retaliation as a useful means for ensuring broad compliance of handout recipients with their end of the bargain, at least in contemporary Mexico.

Our data also suggest that attempts to directly monitor vote choices may be much less frequent than often assumed. Figure 4 reports that just 4.2 per cent of targeted citizens said they were asked to photograph their marked ballot and only one respondent reported that they actually did so. In addition, just one person reported that they were asked to vote with a pre-marked ballot, presumably part of the oftreported technique of “carrousel voting”. Indeed, this one person reported having complied. If we take these findings at face value, then direct monitoring of vote choices through the means explored in the survey was nearly non-existent in 2018.

It is possible that parties instead rely on the goodwill of recipients for compliance. Indeed, there is evidence that political machines may attempt to target recipients who demonstrate high levels of reciprocity (Finan and Schechter, 2012) or even to activate these feelings as a low-cost compliance mechanism (^{Lawson and Greene, 2014}). If so, our primary measure of reciprocity as an attitude scarcely captured it: only 14.1 per cent of targeted citizens said they strongly agreed with the phrase “One should always return favors” [ “Siempre hay que regresar los favores que alguien nos hace”. ] in the post-election survey wave. This figure is 3.6 percentage points higher than the proportion of non-targeted citizens that held the same attitude, implying that recipients of handout offers either are initially more reciprocal, or the offer itself renders them more reciprocal (it is also possible that the vote buying attempt influences the way they respond to the survey question, although we see no obvious reason why that would be the case). However, the figure for targeted citizens is sufficiently low that reciprocity is unlikely to serve as a main compliance mechanism.

Finally, electoral returns imply that vote-buying attempts did not determine which candidate won the presidency. In the final tally, López Obrador won 53.19 per cent of the vote, nearly 31 percentage points beyond that of his closest competitor, Ricrado Anaya of the Forward Mexico [Por México al Frente] coalition at 22.28 per cent and nearly 37 percentage points above José Antonio Meade of the Everyone for Mexico [Todos por México] coalition. Yet Figure 3 shows that López Obrador Together We Will Make History [Juntos Haremos Historia] coalition accounted for a relatively small proportion of electoral handouts. If our post-election panel wave is representative of the electorate as a whole, then parties supporting López Obrador made non-trivial offers to slightly less than 5 per cent of voters. Even if all of these attempts were effective —a very unlikely outcome given the findings documented in this section of the article— and none of the other coalition’s attempts were effective, vote buying would account for less than one-third of López Obrador’s winning margin. By the same token, Anaya’s vote total was scarcely higher than the percentage of voters that received offers from his coalition and Meade’s vote total was nearly 10 percentage points below the proportion that his coalition attempted to buy. It stretches credulity to believe that all of their votes were won through the selective provision of goods and services.⁹

It seems obvious that vote-buying attempts have limited power to win votes, but more work is needed to find those limits. Existing research argues that vote buying falters when political machines target the wrong voters (Carlin and Moseley, 2015; ^{Greene, 2018}; ^{Schaffer and Baker, 2015}; ^{Stokes et al., 2013}; ^{Weitz-Shapiro, 2014}), when voters become too rich to buy at an affordable rate (^{Kitschelt and Wilkinson 2007}), or when institutional reforms diminish political machines’ access to public resources and illicit private donations that can be used as electoral handouts (^{Shefter, 1977}). It is unlikely that these conditions held in Mexico in 2018.

Vote-buying attempts could have influenced the competing parties’ vote shares and possibly the outcomes of down-ballot races —themes that we do not explore here— but the evidence presented strongly implies that the provision of selective benefits did not determine which candidate won. It is even possible, opposite much of the literature, that vote buying attempts turned recipients away from the offering parties, particularly in an election where fighting corruption was one of the main campaign issues and the flagship issue of the eventual winner, Morena.

Conclusion

Mexico’s transition to fully competitive democracy has focused on creating fair elections so that political parties representing a variety of ideologies and social interests could compete on a level playing field with the PRI and with each other. One pillar of support for electoral fairness involved the control of illicit money in politics, a process that culminated in nearly complete public financing of electoral campaigns. Another pillar involved the construction of independent, large-scale, professional electoral management institutions capable of rooting out fraud, imposing sanctions on the malfeasant, and creating public trust in the outcome of elections where little existed previously. By most accounts, Mexico was extraordinarily successful in these pursuits, allowing for the peaceful handover of power between rival political groups, running elections that domestic and international observers hail as free and fair, and building world-leading institutions to carry the substantial load of doing so. In many ways, 2018 signaled these institutions’ crowning achievement by overseeing the runaway victory of a candidate from the left, which had never held the presidency despite having greatly contributed to Mexico’s democratization.

Notwithstanding these genuine achievements, potential problems bubble beneath the surface. As the analyses we review in the first section show, attempts to buy electoral support never faded from Mexico’s politics. Differences in survey timing, questioning approach, and wording make comparisons across surveys imperfect, but electoral clientelism appears to trend upward. Even as parties establish broadly recognized name brands, gain reputations in government, and invest heavily in standard campaigns, their use of electoral handouts —many of them illegal— appears to have increased since 2000. It is even plausible that the very regulations designed to level the playing field through aggressive control of legitimate campaign finances have encouraged the parties to seek advantage by spending illicit money on street-level brokered politics. Worse still, the difficulty of proving vote buying in a court of law renders routine vote brokerage nearly untouchable (^{TEPJF, 2012}).

One reading of the 2018 elections is that the front-runner won handily despite prevalent vote buying by rival parties. But even though electoral clientelism failed to change who won the presidency, it could influence down-ballot races as has been argued for electoral clientelism in Brazil (^{Nichter, 2018}), it may increase tolerance of political corruption (^{De la O, 2015}), and it could erode public trust in electoral outcomes —a theme that we explore in depth elsewhere. Mexico already displays one of the lowest levels of support for democracy in Latin America (LAPOP), making it more vulnerable than one might expect given its history of investment in highquality election management institutions.