Clientelism as Persuasion-Buying

Abstract

In distributing clientelistic payoffs to citizens, the best strategy a party machine can pursue, we argue, is to target citizens who are opinion-leading epicenters in informal conversation networks. This persuasion-buying strategy carries the highest potential yield for the party because the payoff can create a social multiplier: The effect of the clientelistic gift can be magnified via the conversion of multiple voters within a payoff recipient’s personal networks. Using cross-sectional survey data from 22 Latin American countries and a panel survey from Mexico, we confirm that individuals who engage in frequent political persuasion and who are located in large political discussion networks are the most likely recipients of clientelistic payoffs. We also show that a finding that is key to previous theories, namely, that loyal partisans are the most likely targets of clientelism, is driven by omitted-variable and endogeneity bias.

Keywords

clientelism voting behavior networks Latin America

What type of citizen is the most likely to receive clientelistic payoffs from party machines in the developing world? A lively scholarly debate exists around the question of which citizen traits attract the attention of machine operatives as they seek to distribute enticements in search of votes. Scholars have variously asserted that, to maximize effectiveness, machines should and do purchase the votes of swing voters, buy the turnout of unmobilized supporters, reward the loyalty of past supporters, or induce the activism of their loyal partisans.

In this article, we argue that party machines direct payoffs to citizens who are opinion-leading epicenters in informal conversation networks and who thus give the party indirect access to voters who are not directly paid off. In essence, party machines are buying the persuasive services of their clients, a strategy with a greater potential yield than the purchase of swing voters or turnout because it can create a “social multiplier”: The effect of the clientelistic gift can be magnified via the conversion of multiple voters within a payoff recipient’s existing conversational networks.

Our analyses of public opinion surveys conducted in 22 Latin American countries show that parties target individuals who engage in persuasive political talk and are embedded in large political discussion networks. An important finding that is central to competing theories of clientelistic targeting, namely, that loyal partisans are the most likely targets, loses empirical support once we take account of citizens’ propensity to persuade. We also use panel data (from Mexico) to test the exogeneity of our and others’ primary causal variables. We confirm that party machines target individuals who have a past track record of being embedded in large discussion networks. In contrast, we show that the finding that loyal partisans are the most likely targets is driven by endogeneity bias: “Loyalty” emerges only after the payoff has been made. Our argument suggests that clientelism carries a higher yield for parties than previously thought, because the effects of payoffs can be multiplied through discussion networks. This helps to explain why an otherwise expensive and risky party strategy is so prevalent in the developing world.

In the first two sections, we lay out the logic of our persuasion-buying argument, contrasting it with other approaches to clientelism as we proceed. The subsequent section tests the empirical merits of our argument, first using a cross-sectional survey of 22 Latin American countries and then a panel survey of Mexicans. The closing section suggests broader implications.

Who Is Bought?

By clientelism, we refer to a system in which politicians, mostly through party machine operatives, offer goods, services, or jobs to citizens with the expectation that these clients will return the favor with some form of political support (Gans-Morse, Mazzuca, & Nichter, 2014; Stokes, Dunning, Nazareno, & Brusco, 2013). Throughout, we use “client” as shorthand for someone who has received a clientelistic enticement from a machine “operative” who distributes favors on behalf of an election-minded politician (the “patron”). As has become custom, we exclude from this definition the practice of targeting a relatively large group of people with club goods, which is better labeled as pork-barreling.

As clientelism relies on the targeting of individuals by party machine operatives, the central question in the scholarly literature has considered what kinds of voters are most likely to be granted enticements to sell their vote. Cross-national estimates from developing countries show that, in most countries, between 5% and 25% of citizens receive payoffs (Gonzalez-Ocantos, de Jonge, Melendez, Osorio, & Nickerson, 2012; Stokes et al., 2013), but which 5% to 25% are being targeted? Stokes (2005) puts the question in its most illustrative terms:

About 40 voters live in [a machine operative’s] neighborhood, and her responsibility is to get them to the polls and get them to vote for her party. But the party gives her only 10 bags of food to distribute. . . . How does she, and machine operatives like her in systems around the world, decide who among her neighbors shall and who shall not receive handouts? (p. 315)

Scholars have offered a number of potential answers. Party strategies, and thus findings, vary by context, but much evidence suggests that low-income citizens are the most frequently targeted, allegedly because they experience the highest marginal utility from clientelistic favors (Calvo & Murillo, 2004; Faughnan & Zechmeister, 2011). As a corollary of this, it also seems to be the case that clientelism is more widespread in poor countries than in rich ones (Stokes et al., 2013, Chapter 6). Some evidence suggests that dwellers of rural areas, where intimate ties with patrons and operatives are seemingly easier to forge, also have a higher propensity to be targeted than urban residents (Brusco, Nazareno, & Stokes, 2004; Gibson & Calvo, 2000; Scheiner, 2006; Stokes, 2005).

What political characteristics attract party machines? Initially, it seemed almost self-evident that a party machine would target swing voters—those most likely to be favorably swayed by the enticement—or even the subset of swing voters who are “weakly opposed” to its party (Stokes, 2005, p. 321; see also Dixit & Londregan, 1996). Under this so-called vote-buying strategy, machines view payoffs as a means of converting clients, so they target the non-supporters who are most easily moved rather than those who are already firmly in their camp or strong opponents who are immovable at the going rate. This clientelism-as-vote-buying view, however, has since been bedeviled by a stubborn empirical fact: Machines seemingly favor strong partisans of their own party when choosing clients (Nichter, 2008; Stokes et al., 2013, Chapter 2).

Does clientelism even make sense when the beneficiaries of payoffs are already in the bag? Three alternative versions of clientelism have proposed solutions to this “loyal-voter anomaly” (Stokes et al., 2013, p. 66). One sees targeted favors to already-strong supporters as maintenance buying (Cox & McCubbins, 1986) or a “rewarding loyalists” strategy (Gans-Morse et al., 2014, p. 4). Under this arrangement, machines fear that strong supporters could drift to opposing parties in the future if taken for granted, so they channel benefits to these strong supporters to maintain relationships that will keep these supporters within the party’s core constituency. Another alternative that is consistent with the targeting of strong supporters is the turnout-buying strategy (Nichter, 2008). Here, machine operatives deliver favors to demobilized or passive supporters, meaning citizens who favor their party already but who appear unlikely to vote in lieu of a clientelistic prod (Magaloni, 2006).

These innovative answers to the loyal-voter anomaly, however, have their own problems. It is hard to see how a maintenance-buying approach is an equilibrium strategy for any party. Why would machines squander benefits on strong supporters in the current election, thus putting at risk victory now, so as to maintain these voters as core constituents in a future election? (If a voter who was a strong supporter in election t – 1 finds herself on the fence in election t because the party has ignored her since t – 1, then she is no longer a strong supporter but rather a swing voter. A machine that gives her favors is thus pursuing a vote-buying, rather than a maintenance-buying, strategy.) Parties that are always focused on the future risk losing in the present. By contrast, the turnout-buying strategy is on much safer theoretical grounds; its shortcomings are largely empirical. As of yet, scholars have mustered no evidence to show that clients were more likely to abstain (before receipt) than non-clients. If anything, evidence shows the contrary: Clients are more likely than non-clients to be participatory citizens (Faughnan & Zechmeister, 2011; Stokes et al., 2013). Indeed, if clients are already supporters of a party, then (as reams of research on mass political behavior shows) they are actually more likely than the average citizen to vote (Verba, Nie, & Kim, 1978).

In response to these failings, Stokes et al. (2013, Chapter 3) offer a third solution to the loyal-voter anomaly. Machine operatives, they claim, are brokers who do not always have their patrons’ best interests at heart. Brokers end up devoting excessive payoffs to strong supporters, thus performing suboptimally for their patron and party, because of agency slack (Camp, 2014). In particular, brokers subordinate the goal of maximizing votes for their patrons to that of maintaining hired posts as brokers, posts they want because they provide benefits in the form of wages and rents accruing from the withholding of some clientelistic goods. Keeping their posts requires brokers to demonstrate effectiveness to their patrons, which they do by mobilizing as many clients as possible into their partisan network. Brokers thus seek out citizens who can be easily engaged into their network, namely, already-strong supporters who share the partisan leaning of the network and will do visible things such as attend rallies and perform campaign work (Szwarcberg, 2012b). In sum, brokers end up targeting loyalists of their party in pursuit of the maximization of network size while subordinating their patrons’ primary goals of vote- and turnout-buying. We label this strategy engagement-buying.

While the engagement-buying solution to the loyal-voter anomaly is a major theoretical breakthrough, it, too, rests on thin empirical grounds. Scholars have yet to demonstrate that clientelistic payoffs go to those who are swayable into becoming engaged, visible members of a broker’s partisan network. More broadly, the loyal-voter anomaly is also largely derived from Argentina data and thus may be specific to that context. In addition, the jury is still out on whether the loyal-voter anomaly results from endogeneity bias. Virtually all findings in support of this claim are post-receipt measures, meaning survey questions in which respondents report their current partisanship and whether they previously received clientelistic benefits at the same time. This cross-sectional approach is particularly problematic when deployed in countries where partisan loyalties can be fickle (Baker, Sokhey, Ames, & Renno, 2014; Roberts, 2013). While some scholars (Stokes et al., 2013, pp. 54-66) have been attentive to the specter of “endogenous loyalty” raised by this methodological approach (i.e., that gift recipients became loyal partisans only after receiving the benefit), they have yet to convincingly address the issue by comparing partisanship before and after receipt.

Persuasion-Buying and the Social Multiplier

To preview our argument about who is likely to be bought by party machines, we return to Stokes’ illustration. A party operative with payoffs for just 10 of 40 persons would be wise to divvy them out not to 10 swing voters, not to 10 potential non-voters, not to 10 potential rally attenders, but rather to the 10 voters who would most likely be able to reach the other 30 with the machine’s partisan message. In other words, machine operatives should target the opinion leaders of the many small, informal discussion networks that exist in their bailiwicks (Lazarsfeld, Berelson, & Gaudet, 1944).

Figure 1 illustrates the logic of our argument, which we label the persuasion-buying strategy, and shows how it fits with the underlying logic that is common to the standard vote-, turnout-, and engagement-buying models. In these standard models, the logic of which is indicated by the dashed arrow, machines make payoffs with an eye merely toward directly influencing the voting behavior of the client. Operatives see clients as (in the language of network theory) terminal nodes, meaning clientelistic influence hits a dead end upon reaching the client. In the persuasion-buying strategy, the returns from which are shown by the thick dark arrows, machine operatives choose clients who they expect will engage in horizontal persuasion—that is, proselytize the machine’s partisan message to their non-client peers—upon internalizing the goodwill toward the candidate from the payoff. The figure is clear that the two sets of strategies are not mutually exclusive: Operatives pursuing a persuasion-buying strategy will surely hope to achieve some direct influence on the client. For operatives seeking to resolve the trade-offs inherent to targeting, however, the persuasion-buying model posits that they will choose clients who are able to persuade a relatively large number of non-clients over potential clients who are, at most, persuadable themselves. In the language of network theory, operatives seek clients who have high outdegree (i.e., are persuaders) and who are high-degree nodes (i.e., are embedded in large networks).

Figure 1.

Network ties in two different conceptualizations of clientelism.

In targeting influential individuals, persuasion-buying party machines seek to magnify the effect of a single payoff. Horizontal social networks can create a “social multiplier” (Becker & Murphy, 2000; Glaeser, Scheinkman, & Sacerdote, 2003) or “cascading effect” (Baker, Ames, & Renno, 2006; Bikhchandani, Hirshleifer, & Welch, 1992) whereby the impact of the payoff disseminates out to non-clients like ripples in a pond via persuasive political discussion: “ . . . by working through social networks, political leaders need not provide selective incentives themselves, need not coax, cajole, and persuade people . . . Social networks do it for them” (Rosenstone & Hansen 1993, p. 29; see also Huckfeldt, Johnson, & Sprague, 2004). Preference change occurs among those who were never direct beneficiaries of a clientelist enticement, and parties can forge indirect linkages with voters via direct targeting of relatively few individuals. Therefore, for parties, the purchase of social influence potentially yields higher returns than the mere purchase of individual votes or turnout, making it more profitable than strategies that treat clients as terminal nodes.

To multiply the effect of a single payoff, machines attempt to capitalize on pre-existing, informal micronetworks in which discussions about politics and candidates occur. Operatives seed these networks by paying off and thus energizing and currying the good favor of individuals who wield influence within them. They thereby attempt to shape the political color of the day-to-day, impromptu conversations that are otherwise occurring among friends, families, and acquaintances. Informal political talk between and, especially, during political campaigns are voluminous, with some individuals seemingly wired to engage in it more than others (Hatemi & McDermott, 2012; Mondak, 2010), and a rich research tradition in developed and developing democracies shows it to be influential in voters’ decision making (Baker, 2009; Finkel & Smith, 2011; Huckfeldt & Sprague, 1995; Lazarsfeld et al., 1944). Technically, persuasive talk is a subset of political talk, but, in practice, rare is the political conversation that is lacking in bias, so most political discussions hold the potential to persuade. If operatives target individuals with a long-standing propensity to talk and persuade, they need not incur extra costs to create conversations or persuasive opportunities that would not otherwise exist.

These informal opinion leaders are relatively easy for operatives to identify and target, as research on clientelism shows machine operatives to be deeply embedded in and knowledgeable of their local communities: “Brokers are engaged in sustained and frequent interactions with voters, observing their individual behavior and gaining knowledge of their inclinations and preferences” (Stokes et al., 2013, p. 75; see also Camp, 2014). Indeed, that is precisely why patrons and machines hire or draft the ones they do—Hirable operatives have effectively gathered information in their neighborhoods and thus resolve the machine’s own information problem about voters. Scholars have shown that operatives know their potential clients’ material needs (Auyero 1999, 2000; Szwarcberg, 2012a), their partisan leanings (Stokes et al., 2013), their propensity to turn out (Nichter, 2008), and even their likelihood of reciprocating the gift with their vote (Finan & Schechter, 2012). These facts are “basic craft knowledge” (Stokes et al., 2013, p. 102). Operatives thus know who wields social influence, even in informal conversational networks.

Our persuasion-buying model is more grounded in informal, unorganized, and horizontal networks than are existing models of clientelistic targeting. Previous models stress “partisan networks”¹ (Calvo & Murillo, 2013, p. 855), which are (as depicted in Figure 1) vertical relationships between operatives and clients. In this formulation, clientelism is a strictly hierarchical, socially atomizing phenomenon of one-on-one exchange between patron and voter (Anderson, 2010; Kitschelt, 2000; Weyland, 1996). Horizontal networking and persuasion, if alleged to occur at all, is “outsourcing,” whereby operatives try to create with their payoffs new miniactivists who will canvass and turn out non-client voters (Stokes et al., 2013, pp. 72-73; see also Calvo & Murillo, 2013, p. 873).² In contrast, our persuasion-buying model posits operatives who seek to take advantage of “personal” networks and informal political conversations in largely unorganized contexts.

Our persuasion-buying argument entails a reinterpretation of the loyal-voter anomaly (Stokes et al., 2013). Targets, to reiterate our argument, are chosen not for their partisanship but for their persuasive propinquities and their access to relatively large pools of downstream voters. In other words, it can be worthwhile for a party operative to target even minimally swayable outpartisans if they provide the machine with indirect access to numerous voters. In this case, the average vote yield to the machine may be higher than that of targeting a socially isolated swing voter or copartisan.³ That previous work shows partisans to be more likely to receive gifts than non-partisans is due, we argue, to omitted-variable and endogeneity bias. Empirically, micronetwork opinion leaders are more likely to be strong partisans, so failure to control for the respondent’s propensity to persuade or for network size creates omitted-variable bias in any estimation of partisanship’s effect (Huckfeldt & Sprague, 1995). Moreover, as our persuasion-buying approach relies to some extent on internalization by the recipient of the goodwill from the payoff, we suspect the loyal-voter anomaly is also partially due to endogeneity bias, whereby it is the payoff that boosts the probability of strong partisanship, not vice versa.

Our argument also reinterprets the machine’s instrumental vision of network linkages. Previous studies see social networks as providing a means for operatives to monitor whether clients vote in accordance with the machine’s wishes. An operative, the argument holds, can rely on a client’s community contacts to convey whether the client reciprocated the payoff, thereby resolving the “compliance puzzle”—the question of why clients vote for the party that paid them off when they can use the secret ballot to vote their conscience (Cruz, 2013; Stokes, 2005). On its face, however, it is not clear why a client’s social contacts would betray the client’s lack of reciprocity to an operative. Moreover, the compliance puzzle lies in contradiction with the finding that clients are loyalists (either pre- or post-receipt). Finally, the compliance puzzle is only a true puzzle in a world where voters are strictly egocentric, yet research shows that arational norms of reciprocity in clientelistic relationships are widespread (Finan & Schechter, 2012; Lawson & Greene, 2014). In the end, clients’ horizontal networks are far more useful to machines as social multipliers than they are as compliance enforcers.

Hypotheses

Thus far, we have argued that party operatives prefer clients who have the capacity and proclivity to create indirect returns for the machine through informal, interpersonal persuasion. In the following section, we evaluate empirically whether this profile of clients-as-persuaders explains patterns of clientelistic targeting. To set the stage for this analysis, we start by laying out our primary hypotheses, comparing them with those offered by the vote-, turnout-, and engagement-buying alternatives.

We expect to find that citizens who frequently attempt to shape the political choices of others and who are embedded in large political discussion networks are more likely to be targeted by machines than those less prone or less well-positioned to engage in interpersonal persuasion. Moreover, machines should prefer potential clients who are in conversational networks with a low degree of insularity, meaning networks with linkages to the community and not just immediate family. Individuals with a large number of bridging ties to non-family contacts wield greater influence than those with few non-relative ties, holding the potential for larger social multiplier effects (Granovetter, 1973). Among highly insular individuals—those whose political discussants are all family members—persuasive efforts are subject to a dead end, staying within their isolated familial bonds.

We also expect that, in selecting their targets, party machines are more focused on potential clients’ persuasive capacities than their partisan leanings. This contrasts with the turnout- and engagement-buying alternatives, which identify partisans of the operative’s party as the most likely targets. We hypothesize that this loyal partisan finding is partly spurious and partly endogenous, meaning it will weaken upon controlling for measures of persuasive tendency and upon looking at pre-receipt measures of partisanship.

Finally, we expect a similar pattern to obtain with regard to alternative hypotheses that focus on organized political and apolitical networks (Calvo & Murillo, 2013; Levitsky, 2003; Szwarcberg, 2012a). Individuals who tend to be joiners of formal civil society organizations (including political parties) are more likely to be targets, we suspect, partly because they are “movers and shakers” in their communities—people who wield some degree of informal interpersonal influence. If so, the relationship between clientelistic targeting and organizational memberships should also weaken when controlling for persuasion tendencies. Similarly, our statistical tests distinguish between the purchase of informal persuasive talk and formal outsourcing (i.e., the creation of miniactivism).

Data and Results

To test these hypotheses, we use the 2010 Americas Barometer by the Latin American Public Opinion Project’s (LAPOP) and the 2006 Mexican Panel Study.

LAPOP 2010 had a question on whether respondents were offered an election-year clientelistic enticement as well as measures of respondents’ propensity to persuade. Because of the cross-sectionality of the LAPOP data, a finding that there is a positive correlation between the propensity to persuade and to be a target of clientelistic giving would fail to distinguish between two different causal scenarios. Do machines seek out pre-existing opinion leaders, nudging them toward persuading on the machine’s behalf? Or do they create opinion leaders with the payoff, enticing otherwise politically reticent citizens into becoming persuaders? As alluded to above, we suspect the former is at work for two reasons. First, research on the propensity to discuss politics and be communally involved shows a high degree of temporal persistence within individuals (Hatemi & McDermott, 2012; Mondak, 2010). This means that a single campaign gift is unlikely to be the thing that turns a politically mute person into a talkative one. Second, efficiency-minded machines would do well to seek out individuals with proven records as frequent persuaders and high-degree nodes rather than trying to create new social relations with a nominal payoff.

Still, to provide a more concrete foundation for this claim, we follow up our cross-sectional analysis of LAPOP with an analysis of the 2006 Mexico panel data. The longitudinal structure—repeated interviews before and after the July election—allows us to distinguish between these two logics of persuasion-buying and to explore whether the loyal-voter anomaly is simply due to endogenous partisanship. Also, the 2006 Mexican panel study is the only survey data set (from any context) that contains both a political discussant name generator battery, which affords us an alternative measure of horizontal social influence, and a self-report of clientelistic targeting.

For both data sets, the dependent variable is a dichotomous measure, which we label target of clientelism, of whether the respondent reported an experience with clientelism. The exact wording of the question in LAPOP was “In recent years and thinking about election campaigns, has a candidate or someone from a political party offered you something, like a favor, food, or any other benefit or object in return for your vote or support?” For the Mexico panel study, in which interviews took place in the context of a campaign, the question was “Over the last few weeks, has a representative of a political party or candidate given you a gift, money, meals, groceries, or any other type of help?”

Admittedly, both of these are obtrusive measures of clientelistic payoffs, meaning they require respondents to openly admit to receiving a benefit, so they are subject to underreporting due to social desirability bias (Gonzalez-Ocantos et al., 2012). Unfortunately, unobtrusive measures like list experiments that protect respondents from having to confess to the interviewer are unavailable for this many countries. That said, we defend our unobtrusive measures on three grounds. First, in LAPOP, the percentages of “yes” answers to these questions in Argentina (18%), Mexico (17%), and several other countries are close to the 24% that Gonzalez-Ocantos et al. (2012) found and the 20% that Imai, Park, and Greene (2015) found using unobtrusive measures in Nicaragua and Mexico, respectively. Second, with our obtrusive measures, respondents are asked only to admit to being approached and paid, not to actually selling their vote (i.e., complying), so the social undesirability of admission is somewhat mitigated. Third, if dissimulation in the name of social desirability is randomly distributed across respondents, it attenuates observed correlations toward zero, thereby creating a higher bar for achieving statistically significant results (Corstange, 2009). In our case, the bar is even higher than this, because it is precisely the political socialites that we expect to be clientelistic targets who have a greater exposure and sensitivity to social pressures (Bernstein, Chadha, & Montjoy, 2001; Iñiguez, Govezensky, Dunbar, Kaski, & Barrio, 2014). In other words, if social desirability bias is present, the pool of respondents we tally as having been targets of clientelism slightly underrepresents the socially engaged, pushing our observed correlations downward and indicating that they are lower bound estimates of the impact of persuasion and informal network involvement.

Argentine Clients as Frequent Persuaders

We start with the 1,401-respondent Argentina sample from LAPOP 2010. We begin with analyses of an Argentina-only sample, doing so for two reasons. First, many of the literature’s primary findings on who gets targeted come from Argentine samples (Nichter, 2008; Stokes, 2005; Stokes et al., 2013), so we replicate some of these previous models and see how their findings hold up when controlling for measures of social influence. Second, in focusing on one country with a well-defined machine party (the Partido Justicialista, often called the Peronist party), we can more precisely define loyalty with the party that is giving payoffs (Calvo & Murillo, 2004). Regardless, we turn to a regionwide analysis using all countries in the LAPOP dataset in the next subsection.

To test our main hypothesis, we used persuasion frequency, a 4-point scale of the frequency with which respondents tried “to convince others to vote for a party or candidate” during election times. The variable is scaled from 0 to 3, with 0 indicating never and 3 indicating frequently. (Question wordings and descriptive statistics for all variables used are in the Supplemental Information Appendix.) We expect to find that Argentine party machines targeted those who reported a high frequency of persuasive attempts. Our preliminary analysis in Figure 2 shows this to be the case, and strongly so. Benefit recipients were almost twice as likely to be frequent persuaders as non-recipients, and non-recipients were twice as likely to never engage in persuasion.

Figure 2.

Persuasion frequency among clients and non-clients in Argentina.

We test the major alternative visions of clientelistic targeting with the inclusion of some control variables. Most importantly, we include a dummy variable measure of whether the respondent was a partisan identifier. If the vote-buying (i.e., swing-voter) hypothesis is correct, the coefficient on this variable should be negatively signed. If engagement-buying is taking place, then the coefficient for this variable will have a positive sign. In contrast, if persuasion-buying is driving client selection, then we expect partisan identifier to be uncorrelated with the dependent variable once controlling for persuasion frequency. We also entertain the prospect of heterogeneous effects by party, parsing partisan identification into Peronist identifier, radical identifiers, and other party identifier. We also consider presidential approval—of the Peronist incumbent Cristina Fernández de Kirchner—as a proxy for loyalty. Kirchner approval is coded so that higher values mean more positive evaluations.

As another alternative, if the turnout-buying model is correct, then the dummy variable for partisan identifier should be a strongly positive correlate of whether the individual reported being targeted and a variable called political participation index should be negatively signed. Since the turnout model argues that parties are attempting to get unmobilized supporters to the polls, those who are disinclined to participate are expected to be targeted for clientelism.⁴ We also want to control for these forms of participation because our measure of persuasion frequency could be charged with merely proxying for participatory tendencies. As those who participate may be more likely to be rewarded with benefits (Martin, 2003), a positive correlation between persuasion frequency and receipt of payoffs could be, in lieu of this control, spurious.

As a test of the impact of formal networks, we also looked at respondents’ organizational involvement (Boulding, 2014). We used four questions that asked about respondents’ attendance in the meetings of four different kinds of organizations: parents associations (at schools), community improvement associations, trade or business associations, and political party meetings. The variables are scored on ordered scales from 0 to 3, with 0 indicating never and 3 indicating once a week.⁵ We also include a measure titled worked in a campaign to distinguish the brokers themselves and miniactivists who may have been recruited via an outsourcing strategy from the informal persuaders that are key to our argument.

Finally, we also include measures of socioeconomic status (income and education) and geographical location, all of which, given the balance of evidence in the literature, we expect to be negatively correlated with our dependent variable. Another finding in the literature relates to the size of the city or town in which the respondent lives. To control for this, we used an ordinal measure (place of residence) of the size of each respondent’s location. We expect to find that each of these variables is negatively signed. Controls for age and gender (female) were also included.

Results of 10 different binary logit models are reported in Table 1.⁶ Model 1 is a rough replication of Stokes (2005) and Nichter (2008), with partisan identification and political participation as the main variables of interest. The results support neither the traditional vote-buying model nor the turnout-buying model. It is partisans, not independents or swing voters, who are most likely to get payoffs. Whereas the average predicted probability of a party identifier being targeted is .28, that of a nonpartisan being targeted is .14.⁷ While this finding lies in partial support of the turnout-buying hypothesis, this hypothesis is undermined by the positive and statistically insignificant coefficient on political participation. Party machines do not target an unengaged citizenry. If anything, their clients are more likely than non-clients to participate.

Table 1.

The Correlates of Being a Client in Argentina.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10
Persuasion frequency		0.596** (0.133)		0.608** (0.134)		0.598** (0.135)		0.595** (0.137)	0.571** (0.135)	0.450** (0.127)
Partisan identifier	0.667** (0.247)	0.325 (0.232)					0.529* (0.223)	0.247 (0.217)	0.144 (0.230)	−0.013 (0.221)
Peronist identifier			0.489 (0.303)	0.123 (0.279)
Radical identifier			0.457 (0.293)	0.208 (0.347)
Other party identifier			0.711* (0.347)	0.316 (0.356)
Kirchner approval					0.254** (0.078)	0.196* (0.083)			0.181* (0.088)	0.178* (0.084)
Political participation index	0.207 (0.175)	0.195 (0.180)	0.237 (0.173)	0.222 (0.178)	0.315* (0.150)	0.253 (0.164)	0.091 (0.180)	0.134 (0.183)	0.146 (0.187)	0.130 (0.184)
Parents associations							0.215 (0.142)	0.260 (0.144)	0.252 (0.140)	0.230 (0.141)
Community improvement associations							0.318** (0.108)	0.349** (0.111)	0.331** (0.109)	0.333** (0.108)
Trade or business associations							0.449** (0.112)	0.377** (0.113)	0.372** (0.114)	0.383** (0.127)
Political party meetings							−0.001 (0.150)	−0.200 (0.166)	−0.163 (0.166)	−0.241 (0.173)
Worked for a campaign										1.172** (0.251)
Income	−0.055 (0.054)	−0.055 (0.053)	−0.047 (0.053)	−0.053 (0.053)	−0.043 (0.056)	−0.052 (0.055)	−0.064 (0.052)	−0.066 (0.053)	−0.068 (0.053)	−0.073 (0.052)
Education	−0.055* (0.022)	−0.061** (0.023)	−0.056* (0.023)	−0.062** (0.023)	−0.053* (0.023)	−0.061** (0.024)	−0.068** (0.022)	−0.072** (0.023)	−0.072** (0.024)	−0.070** (0.024)
Age	0.002 (0.004)	0.002 (0.005)	0.003 (0.004)	0.002 (0.005)	0.004 (0.004)	0.002 (0.005)	0.001 (0.005)	0.001 (0.005)	−0.000 (0.005)	0.001 (0.005)
Female	−0.024 (0.184)	0.042 (0.184)	−0.025 (0.185)	0.040 (0.186)	−0.040 (0.187)	0.039 (0.185)	−0.07 (0.184)	−0.02 (0.179)	−0.011 (0.179)	−0.057 (0.164)
Place of residence	0.204 (0.336)	0.098 (0.304)	0.203 (0.336)	0.094 (0.301)	0.204 (0.336)	0.093 (0.300)	0.127 (0.322)	0.024 (0.292)	0.019 (0.288)	−0.012 (0.266)
Intercept	−1.451 (0.533)	−1.684 (0.551)	−1.481 (0.538)	−1.709 (0.555)	−1.941 (0.567)	−2.016 (0.570)	−1.253 (0.551)	−1.562 (0.568)	−1.815 (0.586)	−1.946** (0.615)

Source. Latin American Public Opinion Project (LAPOP) 2010, Argentina sample.

N = 1,355. Dependent variable is target of clientelism, whether respondent had received a gift (1) or not (0). Entries are logit coefficients with standard errors in parentheses. Results are from estimates performed on five multiply imputed data sets. The Wald χ² statistic for all models is statistically significant at p < .01.

p < .05, two-tailed. **p < .01, two-tailed.

Model 2 introduces the measure of persuasion frequency, a move that dispels the possibility that the positive finding on the partisan identifier variable is due to engagement-buying. In support of our main hypothesis, persuasion frequency is positive and statistically significant. Moving from those who said they never engage in persuasion to those who reported frequently attempting to persuade others corresponds to a .25 increase in the predicted probability of reporting a clientelism experience. Moreover, as expected, the introduction of the persuasion frequency variable reduces the coefficient on partisan identifier to statistical insignificance and its magnitude by nearly half. In other words, machines target persuaders, and they are largely indifferent to their targets’ pre-existing partisan leanings. Partisanship shows up as relevant in the more restricted model (Model 1) only because of its correlation with persuasion frequency.⁸

One thing that could artificially weaken the impact of partisan identifier, and thus the empirical support for the turnout-buying and engagement-buying arguments, is the fact that Models 1 and 2 do not separate this variable out by party. As one party in Argentina—the Peronist Partido Justicialista—makes most clientelist payoffs (Calvo & Murillo, 2004), the partial correlation between partisan identifier and being targeted would be weaker than that between Peronist identifier and being targeted. We address this in Models 3 and 4, which soundly dismiss this concern by showing that variables for Peronist and Radical partisan identifiers are statistically insignificant, even when not controlling (Model 3) for persuasion frequency. More importantly for our purposes, the coefficients on these two variables, as well as that on a measure of identification with other parties, experience a dramatic drop in size of 55% to 75% upon controlling for persuasion frequency. Models 5 and 6 repeat the exercise using approval of Kirchner as the measure of loyalty. Presidential approval has a bit more staying power than partisanship—the drop in the coefficient’s size once persuasion frequency is included is about 25%—but the impact of persuasion frequency is invariant to the inclusion of this variable.

Models 7, 8, 9, and 10 in Table 1 introduce the formal community involvement variables. Two of these four variables are positive and statistically significant. Among Argentine respondents, those who are involved in community improvement associations and professionally oriented associations are more likely to be the targets of clientelism than those who are not involved in such groups. Again, however, the effect size of persuasion frequency is invariant to the exclusion (Models 2, 4, and 6) or inclusion (Models 8 and 9) of these controls. At the very least, we can thus conclude that a potential client’s propensity to engage in persuasive political talk has huge drawing power for party machines, above and beyond their membership in formal organizations. Moreover, formal party membership is not associated with the dependent variable in any of the models, suggesting that informal, personal persuasion and apolitical formal organizations are more important than partisan networks, which runs counter to arguments that emphasize rally attendance and formal partisan engagement (Szwarcberg, 2013, 2014). Having worked in a campaign (added in Model 10) is associated with targeting, and its introduction does slightly reduce the coefficient (by about 20%) on persuasion frequency, but the latter remains highly statistically significant. Machines’ attraction to persuaders does not lie solely in creating formal, miniactivist campaign workers but also in harnessing networks of informal persuasion.

The models support conventional findings about socioeconomic status in Argentina. As expected, the results indicate that clientelism is primarily deployed on the poor and uneducated. Although the coefficient on income is statistically insignificant, the education coefficient is significant in all specifications and both measures are negatively signed. The reported results do not confirm a clear rural/urban divide: Clientelism is equally likely in both settings. Finally, the results suggest that machines do not prefer clients to be of a particular age or gender.

Latin American Clients as Frequent Persuaders

We now turn to the full LAPOP sample of 22 countries to consider whether our argument holds up in other Latin American countries. As in the Argentina case, the conditional marginal distributions show that persuasion frequency seemingly had a huge impact on who became clients. Nearly three-quarters of non-clients said they never engaged in political persuasion, whereas less than 50% of clients were non-persuaders. The share of frequent persuaders was also twice as high among clients. But does this relationship stand up when controlling for potential confounds? To provide a more rigorous statistical test, we estimated five different models using the full LAPOP sample.⁹ Results are reported in Table 2.

Table 2.

The Correlates of Being a Client in Latin America.

	Model 1	Model 2	Model 3	Model 4	Model 5
Persuasion frequency		0.517** (0.021)		0.485** (0.021)	0.446** (0.022)
Partisan identifier	0.284** (0.047)	0.095* (0.048)	0.203** (0.047)	0.059 (0.048)	0.014 (0.048)
Parents associations			0.111** (0.020)	0.106** (0.020)	0.104** (0.020)
Community improvement associations			0.111** (0.024)	0.097** (0.025)	0.088** (0.026)
Trade or business associations			0.201** (0.030)	0.175** (0.031)	0.177** (0.031)
Political party meetings			0.247** (0.028)	0.119** (0.030)	0.062* (0.031)
Political participation index	0.368** (0.036)	0.273** (0.036)	0.218** (0.037)	0.173** (0.037)	0.143** (0.035)
Worked for a campaign					0.472** (0.058)
Income	−0.016 (0.010)	−0.022* (0.010)	−0.016 (0.010)	−0.022* (0.010)	−0.022* (0.010
Education	−0.006 (0.006)	−0.014* (0.006)	−0.010 (0.006)	−0.016* (0.006)	0.017** (0.005)
Age	−0.009** (0.001)	−0.011** (0.001)	−0.010** (0.001)	−0.011** (0.001)	−0.011** (0.001)
Female	−0.135** (0.035)	−0.079* (0.036)	−0.127** (0.037)	−0.082* (0.037)	−0.074* (0.037)
Place of residence	−0.016 (0.022)	−0.023 (0.022)	0.006 (0.022)	−0.004 (0.022)	−0.002 (0.022)
Intercept	−1.115 (0.135)	1.138 (0.134)	−1.325 (0.135)	−1.303 (0.134)	−1.431 (0.135)

Source. Latin American Public Opinion Project (LAPOP) 2010.

N = 36,963. Dependent variable is target of clientelism, whether respondent had received a gift (1) or not (0). Entries are logit coefficients with standard errors in parentheses. Country fixed effect coefficients were included but not reported. Results are from estimates performed on five multiply imputed data sets. The Wald χ² statistic for all models is statistically significant at p < .01.

p < .05, two-tailed. **p < .01, two-tailed.

The first model confirms that the traditional vote-buying and turnout-buying explanations are inaccurate. Both partisans and the highly participatory are more likely to report receipt of a campaign gift, suggesting that it is neither swing voters nor unmobilized supporters who are targeted. Rather, Model 1 would seem to indicate that machine operatives target loyal, participatory voters, lending preliminary evidence to the engagement-buying argument.¹⁰ Yet the inclusion of persuasion frequency, introduced in Model 2, changes this conclusion. The coefficient on persuasion frequency is positive and highly statistically significant, and (moving from “never” to “frequently”) it yields a massive .20 increase in the probability of receiving a campaign gift. By way of comparison, the Model-2 predicted probabilities increase just .04 and .04 when moving from the top to the bottom of the education and income scales, respectively. Just as important is the fact that the addition of persuasion frequency precipitates a whopping decline (of 67%) in the size of the coefficient for partisan identifier, mirroring the dynamics in the Argentina models. Again, this is a strong indication that Latin American party machines are seeking persuaders when distributing clientelistic benefits and are less focused on the strength and direction of potential recipients’ partisanship.

Figure 3 shows the effects of persuasion frequency and partisan identifier by country. We ran two regressions per country. (Full results are not shown but can be viewed upon request.) The list of independent variables for each was equivalent to those in Model 3 and Model 4, respectively, of Table 2. Figure 3 refers to the latter model (which includes persuasion frequency, partisan identifier, and a host of other control variables) as the “full model.” The other model only differs in that it excludes persuasion frequency; we are seeking to convey the effect that controlling for persuasion frequency has on the estimated effect of partisan identifier.

Figure 3.

The effect of persuasion frequency and partisan identification on being a client in 22 countries: Logit coefficients and 95% confidence intervals.

In the full models, persuasion frequency has a statistically significant effect in 20 of the 22 countries, an incredibly robust and near-ubiquitous effect. In contrast, the effect of partisan identification is minimal in the full models—only statistically significant for one country. When not controlling for persuasion frequency, partisan identifier is statistically significant in six countries (as in the Argentina example discussed above), and the size of the coefficient on partisan identifier is almost always smaller in the full model. All told, partisan identifier is statistically insignificant far more frequently than it is significant, suggesting that the loyal partisan effect that is key to various theories of clientelistic targeting may be borne of the context (largely Argentine) from which it stems.

The coefficients on the apolitical organization variables (Models 3 and 4) are similar in direction and magnitude to the Argentina results: operatives target community joiners. In contrast to the Argentina results, those who attend political party meetings are more likely to be targeted than those who do not. This suggests that political networks matter, but it is also telling that the size of this coefficient drops by just over 50% when controlling for persuasion frequency (Models 3 and 4), whereas the coefficient on persuasion frequency changes little when introducing this control (Models 2 and 4). In other words, a fair portion of the effect of party networks is due to the fact that party joiners are informal persuaders. The impact of persuasion frequency is also largely invariant (dropping in size by just 8%) to the inclusion of the worked in a campaign variable (Model 5), an indication that persuasion-buying is far more than formal outsourcing. In sum, we can draw virtually the same major conclusions from the Argentina and regionwide analyses. Party machines seek to make clients out of those who are local opinion leaders in informal personal networks. They are not seeking copartisans, swing voters, or unmobilized supporters.

Mexican Clients as High-Degree Nodes

To tease out direction-of-causality questions and assess a completely different measure of respondents’ propensity to attempt horizontal influence, we turn to the Mexico 2006 panel data and its alternative measure of our central concept (Domínguez, Lawson, & Moreno, 2009). Our primary expectation is that Mexican machine operatives sought to make clients out of high-degree, low-insularity nodes, meaning individuals with a relatively large number of non-familial political discussants. We also expect that clients were high-degree, low-insularity nodes before they became targets, whereas they were not necessarily strong partisans before becoming a target.

In the three-wave panel study, a question about receipt of campaign favors from a party in the preceding weeks was asked in Wave 2 (May) and Wave 3 (July, 2 weeks after the July 2 election). The percentage of the population that reported receiving an enticement in the preceding weeks was 4% in May and 5% in July. Across the two waves, a total of 8.5% of respondents reported receiving a gift at one point. (This number is lower than that reported for Mexico in LAPOP [17%] due to the “recent years” time frame in the LAPOP question and the “last few weeks” time frame in the Mexico panel question.)

Waves 2 and 3 also contained a traditional political discussant name generator (Sokhey & Djupe, 2014). In each wave, respondents were asked to name the three people with whom they most discussed politics and then to subsequently report their relationship (e.g., spouse, work friend) with each named discussant (Baker, 2009). (The second battery was wholly independent of the first, meaning respondents could mention completely new discussants in Wave 3 if they so desired.) For our first set of analyses, we use the Wave 3 measure of target of clientelism as the dependent variable and the Wave 2 measures of network characteristics as the primary independent variables. Unlike the analysis of the LAPOP cross-section above (not to mention most previous analyses of clientelistic targeting in the developing world), this puts our hypothesis test in its proper temporal order: Operatives learn who wields social influence and then target them.

We generated two variables to describe our respondents’ political networks: number of non-familial discussants and number of familial discussants. As a primary test of the persuasion-buying theory, we expect those with a large number of non-familial political discussants in May (Wave 2) to have drawn the attention of political machines in the subsequent weeks leading up to the July election. In some models, we also include a lagged dependent variable, meaning a measure of whether the respondent reported (in Wave 2) having received a gift before the Wave 2 interview. The inclusion of this lag rules out the charge that any positive results we have on the network variables are driven by a form of endogeneity in which pre–Wave 2 campaign gifts that are misreported as pre–Wave 3 campaign gifts (given the vagueness of the “previous weeks” time frame posed to respondents in the clientelism question) caused a boost in network size. That said, we also report models without this lagged dependent variable.

We also control for the two variables that capture the main alternative theories. Strength of partisanship is an ordinal trichotomy (0 for independents, 1 for weak partisans, 2 for strong partisans), and political interest is our proxy for participatory tendencies. As above, we also include measures of community involvement, gauging whether the respondent is a parents association member, neighborhood improvement association member, professional association member, new social movement member, sports association member, or a political party member. We also control for age, gender, wealth, education, and place of residence. When available, we use Wave 1 or Wave 2 measures of all of these control variables to gauge respondents’ traits before the clientelistic benefits were given. Community involvement measures were only available, however, for Wave 3.

The results (reported in Table 3) conform closely to the expectations generated by the persuasion-buying model. Most importantly, the number of non-familial discussants, as measured before the delivery of clientelistic benefits, is a statistically significant predictor of which Mexicans were targeted. Focusing first on the Model 1 results, among respondents with no non-familial discussants, the average predicted probability of being targeted was .027, while among those with 3 non-familial discussants, the probability was .064, a more than doubling of the propensity to receive a payoff.¹¹ As expected, the impact of family network size is much smaller, about one third the size and not statistically significant. Mexican machines seek targets with bridging ties, and lots of them. Moreover, the effect of non-familial political discussants is paramount, especially relative to that of community involvement. In other words, party machines in Mexico specifically target individuals who have a large number of political conversation partners, as opposed to individuals who are community activists and formal organization members.

Table 3.

The Correlates of Being a Client in Mexico.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Number of non-familial discussants_{t – 1} (Wave 2)	0.264* (0.131)	0.263* (0.130)	0.256* (0.130)	0.259* (0.127)	0.257* (0.127)	0.252* (0.126)
Number of familial discussants_{t – 1} (Wave 2)	0.082 (0.144)	0.082 (0.144)	0.050 (0.146)	0.019 (0.140)	0.019 (0.140)	−0.017 (0.141)
Strength of partisanship_{t – 2} (Wave 1)	−0.089 (0.174)			−0.060 (0.169)
Strength of partisanship_{t – 1} (Wave 2)		−0.085 (0.178)			−0.049 (0.172)
Strength of partisanship_t (Wave 3)			0.393* (0.179)			0.400* (0.173)
Political interest_{t – 1} (Wave 2)	−0.129 (0.158)	−0.126 (0.159)	−0.175 (0.158)	−0.069 (0.156)	−0.068 (0.157)	−0.114 (0.156)
Parents associations member_t (Wave 3)	0.016 (0.210)	0.015 (0.210)	0.009 (0.210)	0.083 (0.204)	0.083 (0.203)	0.071 (0.203)
Neighborhood improvement association member_t (Wave 3)	0.347 (0.236)	0.347 (0.235)	0.351 (0.236)	0.210 (0.229)	0.211 (0.229)	0.225 (0.229)
Professional association member_t (Wave 3)	0.041 (0.239)	0.035 (0.239)	0.024 (0.240)	0.084 (0.235)	0.080 (0.235)	0.070 (0.236)
New social movement member_t (Wave 3)	−0.467 (0.458)	−0.474 (0.457)	−0.373 (0.446)	−0.412 (0.425)	−0.414 (0.424)	−0.331 (0.417)
Sports association member_t (Wave 3)	0.214 (0.214)	0.215 (0.214)	0.208 (0.214)	0.195 (0.211)	0.197 (0.211)	0.186 (0.212)
Political party member_t (Wave 3)	−0.008 (0.249)	0.001 (0.251)	−0.153 (0.254)	0.042 (0.238)	0.045 (0.240)	−0.093 (0.242)
Wealth_t-2 (Wave 1)	−0.092 (0.705)	−0.075 (0.704)	−0.023 (0.703)	−0.138 (0.701)	−0.129 (0.700)	−0.065 (0.072)
Education_{t – 1} (Wave 2)	0.070 (0.074)	0.072 (0.074)	0.095 (0.075)	0.041 (0.072)	0.043 (0.072)	0.062 (0.072)
Age	0.013 (0.009)	0.013 (0.009)	0.012 (0.009)	0.014 (0.009)	0.014 (0.009)	0.013 (0.009)
Female	0.454 (0.274)	0.457 (0.275)	0.402 (0.275)	0.444 (0.267)	0.445 (0.267)	0.406 (0.266)
Urban residence	0.325 (0.182)	0.323 (0.182)	0.344 (0.184)	0.357* (0.181)	0.356* (0.182)	0.373 (0.183)
Lagged dependent variable: Target of clientelism_t – ₁ (Wave 2)	2.291** (0.336)	2.291** (0.336)	2.276** (0.338)
Intercept	−4.829 (0.756)	−4.861 (0.743)	−5.334 (0.782)	−4.531 (0.715)	−4.559 (0.703)	−4.988 (0.710)
N	1,358	1,358	1,358	1,358	1,358	1,358

Source. Mexico 2006 Panel Study.

Dependent variable is target of clientelism_t, whether respondent had received a gift (1) or not (0). Entries are logit coefficients with standard errors in parentheses. Results are from estimates performed on five multiply imputed data sets. The Wald χ² statistic for all models is statistically significant at p < .01.

p < .05, two-tailed. **p < .01, two-tailed.

The results also cast serious doubt on the claim that machines seek already-loyal partisans. Instead, in line with the persuasion-buying approach, they show evidence that clients internalize support for the party only after receiving the payoff. Strong partisanship, as measured in Wave 1 (before the campaign began) and Wave 2 (before the payoff occurred), was actually negatively associated with campaign gift receipt (Models 1, 2, 4, and 5), although the coefficients are nowhere near conventional statistical significance levels. In other words, in Waves 1 and 2, those who later became clients were not more partisan than those who remained non-clients. Perhaps even more damning of the loyal-voter claim is that strength of partisanship has a positive and statistically significant effect only when it is measured in Wave 3, after the payoffs were made (Models 3 and 6). Stated differently, the observed correlation upon which turnout-buying and engagement-buying claims are based requires partisan loyalty to be measured after targeting has occurred. The standard gap in partisan loyalties between clients and non-clients emerged only after campaign benefits had been delivered. Collectively, these two model results are more supportive of the endogenous loyalty phenomenon, whereby clientelistic targeting creates partisans rather than vice versa, than the loyal-voter anomaly that is fundamental to these alternative perspectives.

We can address this issue more directly by depicting trends in partisanship, as well as network size, through time. Using the panel data, we can assess whether receiving payoffs between Waves 2 and 3 resulted in a corresponding increase in partisanship or network size relative to partisanship and network size among those who did not receive payoffs. Findings of this sort would cast doubt on viewing these variables as exogenous drivers of clientelistic targeting.

Figure 4 confirms the suspicions raised in Table 3: Partisanship is largely endogenous to having received a payoff, while network size is not. Panel A of the figure depicts means (with 95% confidence intervals) of strength of partisanship across the three panel waves for two groups: those who received a clientelistic payoff between Waves 2 and 3 (“clients,” black line) and those who did not (“non-clients,” gray line). To be clear, those who are tallied as “clients” did not become so until after Wave 2, and the set of individuals defined as clients and non-clients is equivalent across all three waves. In Waves 1 and 2, the two groups were statistically indistinguishable in their degree of partisanship, and those who were eventually paid off were actually a bit less partisan (as in Table 3) than those who were never targeted. In other words, those who received a payoff after Wave 2 were not more partisan before Wave 2 than those who did not receive one. It is only after the payoff is made that a gap emerges, because recipients became more partisan while non-recipients remained stable. The simplest and most plausible explanation for this is that partisan loyalty is endogenous to clientelistic targeting. Merely looking at post-targeting cross-sectional measures yields misleading results of the impact of partisanship on payoff receipt.

Figure 4.

Trends in average strength of partisanship and average number of discussants for clients and non-clients, Mexico 2006: Means and 95% Confidence Intervals.

By contrast, network size shows all indications of not being endogenous to clientelistic targeting. Panel B of Figure 4 reports the means and confidence intervals on non-familial discussants for clients and non-clients in Waves 2 and 3. Already in Wave 2, those who later received a payoff (black line) had more discussants than those who did not later receive a payoff (gray line). This gap remained constant in size across Waves 2 and 3, meaning that clients did not grow their networks in specific response to being paid off.¹² Rather, both sets of individuals grew their networks at an equivalent rate, an indication—one that is standard in research on political discussion networks—that individuals tend to talk to more people about politics as elections draw near (Huckfeldt et al., 2004). To sum, in Mexico, loyalty to a party is endogenous to being paid off, whereas discussion network size is not.

Implications and Conclusion

In this article, we argued that vote-maximizing political machines seek to target their payoffs to citizens who are politically verbose and well-networked. Such individuals hold the highest potential yield for parties because they are epicenters of persuasion within horizontal networks. Clients with these characteristics have the ability to deliver multiple votes—with just a single payoff—to the patron, creating a social multiplier effect of the payoff. Our empirical demonstration of this argument also showed that the propensity to persuade is the lurking variable behind the previous literature’s observed correlation between partisan loyalty and clientelistic targeting and that, when not spurious, the alleged impact of partisan loyalty on payoff receipts is endogenous.

Our theory and findings allow us to resolve what we see as an unanswered puzzle in the literature on clientelism in the developing world: Why do so many politicians around the world consider it to be a cost-effective way of winning votes? As currently conceptualized by political scientists, clientelism is an expensive and inefficient party strategy. The granting of particularistic benefits to individuals in exchange for their political support carries more costs than just the price of the payoffs themselves. Patrons must maintain an army of machine operatives, who in turn must devote time and effort to choosing which citizens to pay off, delivering the benefits, and then monitoring their beneficiaries’ subsequent voting behavior (Kitschelt & Wilkinson, 2007). All the while, patrons and operatives run the risks of legal sanction, non-compliance by ungrateful clients, waste via the targeting of voters who are already supporters, and resentment among non-recipients (Ocantos, de Jonge, & Nickerson, 2014; Weitz-Shapiro, 2012). Moreover, all of this occurs, according to the prevailing scholarly framework, in search of votes that are merely “picked up one at a time,” putting clientelism “among the least-efficient strategies of manipulation” (Schaffer, 2007, p. 191). On a per-vote basis, it would seem to be cheaper for patrons to reach voters with pork-barrel projects or programmatic stances that benefit entire communities or socioeconomic strata (Desposato, 2007; Greene, 2007), not to mention direct media-based appeals that reach broad swathes of the population. Auyero (Auyero, 1999) summarizes this puzzle most effectively:

The brokers’ . . . capacity to deliver . . . is limited because the broker can get jobs, deliver medicine, do an essential (or founding) favor, and assist someone as if he or she were part of her family, for a restricted number of people. . . . The size of the brokers’ inner circle can hardly account for the “conquest of the vote” and “building of electoral consensus” that is usually attributed to clientelism. (p. 326)

Our answer to this puzzle is that scholars have underestimated the benefits of clientelism. Scholars have overlooked the fact of clients’ persuasive influence, by which clients magnify the effect of a payoff by convincing other community members of the merits of casting a vote for the patron. The paradigm of methodological individualism, it seems, has exerted a lasting pull on scholarship on clientelism, leading a literature that is otherwise quite attentive to networks to treat the client as socially isolated once he or she has been paid off. Clientelism is not a narrow strategy of targeting a relatively small number of presumed high yield voters. Rather, it is an encompassing strategy of pollinating the electorate, enticing a minority directly while attempting to reach the non-client majority indirectly.

Our persuasion-buying argument also carries important macro-level implications, namely for understanding the strong negative correlation between GDP per capita and the prevalence of clientelism (Stokes et al., 2013, Chapter 6). The long-standing claim has been that clientelistic gifts, because of the diminishing marginal utility of income, produce a bigger “bang for the buck” among the poor than among the rich (Dixit & Londregan, 1996). The problem with this logic when it is scaled up to the cross-national question is that virtually every country has its relatively poor who party machines could afford to buy off. Politicians in developed countries can afford to pay off the relatively poor, even if the absolute cost of doing so is higher than for politicians in developing countries, because they themselves have greater absolute financial resources upon which to draw.

Our persuasion-buying argument suggests a different causal mechanism: the availability of mass communications (Stokes et al., 2013). Clientelism thrives in media-poor environments because horizontal persuasion via informal discussion is virtually the only means which parties have of reaching the many potential voters they cannot directly contact.¹³ Where discussion among horizontal social relations is the primary source of political information and persuasion, as is necessarily the case in media-poor countries, the benefits of engaging in clientelism are large. Once citizens are able to access mass-mediated sources of political information, however, interpersonal persuasion is no longer the sole means of reaching voters indirectly (Boas, 2005; Lawson, 2002), although these benefits never completely disappear because horizontal social influence is present even in media-rich contexts (Druckman & Nelson, 2003; Huckfeldt et al., 2004).

Finally, our findings suggest that scholars have tended to overemphasize the extent to which clientelism is a hierarchical process that occurs across a strictly vertical division of power (Anderson, 2010; Weyland, 1996). Clientelism has reverberations within horizontal social relations, creating meaningful political conversations and connections as well as indirect linkages between citizens and parties.

Footnotes

Acknowledgements

We thank those who provided valuable comments on previous drafts: Ernesto Calvo, Eddie Camp, Cesi Cruz, Ken Greene, Richard Jessor, Vicky Murillo, Simeon Nichter, Ezequiel Gonzalez-Ocantos, Brian Richter, Doug Schuler, Anand Sokhey, Rebecca Weitz-Shapiro, and Jennifer Wolak, and the Institutions group of the University of Colorado at Boulder Institute for Behavioral Sciences (Lee Alston, Jennifer Bair, Carew Boulding, Edward Greenberg, Joseph Jupille, Nelson Montenegro, Isaac Reed, and James Scarritt) for comments. We also thank the Latin American Public Opinion Project (LAPOP) and its major supporters (the United States Agency for International Development, the United Nations Development Program, the Inter-American Development Bank, and Vanderbilt University) for making the data available (www.LapopSurveys.org). Senior Project Personnel for the Mexico 2006 Panel Study include (in alphabetical order) Andy Baker, Kathleen Bruhn, Roderic Camp, Wayne Cornelius, Jorge Domínguez, Kenneth Greene, Joseph Klesner, Chappell Lawson (Principal Investigator), Beatriz Magaloni, James McCann, Alejandro Moreno, Alejandro Poiré, and David Shirk. Fieldwork for the Mexico 2006 Panel Study was conducted by Reforma newspaper’s Polling and Research Team, under the direction of Alejandro Moreno. ()

Authors’ Note

Both authors are equal and primary coauthors. All computer code and the Supplementary Information Appendix can be accessed at .

Declaration of Conflicting Interests

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

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Funding for the data used in this study was provided by the National Science Foundation (SES-0517971), Reforma newspaper, the United States Agency for International Development, the United Nations Development Program, the Inter-American Development Bank, and Vanderbilt University.

Notes

Author Biographies

Joby Schaffer is a PhD candidate in the Political Science Department at the University of Colorado at Boulder. He is currently writing a dissertation on the informal sector in China.

Andy Baker is Associate Professor in the Department of Political Science at the University of Colorado at Boulder. He conducts research on mass political behavior and political economy, mostly in the Latin American context. He is the author of two books, The Market and the Masses in Latin America and Shaping the Developing World, and he has written articles for the American Political Science Review, American Journal of Political Science, Latin American Research Review, World Politics, and several other journals.

References

Anderson

L. E.

(2010). Social capital in developing democracies: Nicaragua and Argentina compared. New York, NY: Cambridge University Press.

Auyero

(1999). From the client’s point(s) of view: How poor people perceive and evaluate political clientelism. Theory and Society, 28, 297-334.

Auyero

(2000). The logic of clientelism in Argentina: An ethnographic account. Latin American Research Review, 35(3), 55-81.

Baker

(2009). Regionalized voting behavior and political discussion in Mexico. In Dominguez

Lawson

Moreno

(Eds.), Consolidating Mexico’s democracy: The 2006 presidential campaign in comparative perspective (pp. 71-88). Baltimore, MD: Johns Hopkins University Press.

Baker

Ames

Renno

L. R.

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

Baker

Sokhey

Ames

Renno

L. R.

(2014, August). The dynamics of mass partisan identification when party brands change: The case of the workers party in Brazil. Paper presented at the 2014 Annual Meeting of the American Political Science Association, Washington, DC.

Becker

G. S.

Murphy

K. M.

(2000). Social economics: Market behavior in a social environment. Cambridge, MA: Harvard University Press.

Bernstein

Chadha

Montjoy

(2001). Overreporting voting: Why it happens and why it matters. Public Opinion Quarterly, 65, 22-44.

Bikhchandani

Hirshleifer

Welch

(1992). A theory of fads, fashion, custom, and cultural change as informational cascades. Journal of Political Economy, 100, 992-1026.

10.

Boas

T. C.

(2005). Television and neopopulism in Latin America: Media effects in Brazil and Peru. Latin American Research Review, 40(2), 27-49.

11.

Boulding

(2014). NGOs, political protest, and civil society. New York, NY: Cambridge University Press.

12.

Brusco

Nazareno

Stokes

(2004). Vote buying in Argentina. Latin American Research Review, 39(2), 66-88.

13.

Calvo

Murillo

M. V.

(2004). Who delivers? Partisan clients in the Argentine electoral market. American Journal of Political Science, 48, 742-757.

14.

Calvo

Murillo

M. V.

(2013). When parties meet voters: Assessing political linkages through partisan networks and distributive expectations in Argentina and Chile. Comparative Political Studies, 46, 851-882.

15.

Camp

(2014, September). Cultivating effective brokers: A party leader’s dilemma. Vanderbilt University, Nashville, TN.

16.

Corstange

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

17.

Cox

G. W.

McCubbins

M. D.

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

18.

Cruz

(2013, August). Social networks and the targeting of illegal electoral strategies. Presented at the Annual Meeting of the American Political Science Association. Retrieved from http://ssrn.com/abstract=2299356

19.

Desposato

(2007). How does vote buying shape the legislative arena? In Schaffer

F. C.

(Ed.), Elections for sale: The causes and consequences of vote buying (pp. 101-122). Boulder, CO: Lynne Rienner.

20.

Díaz-Cayeros

Estévez

Magaloni

(2009). Buying-off the poor: Effects of targeted benefits in the 2006 presidential race. In Dominguez

J. I.

Lawson

Moreno

(Eds.), Consolidating Mexico’s democracy: The 2006 presidential campaign in comparative perspective (pp. 229-245). Baltimore, MD: Johns Hopkins University Press.

21.

Dixit

Londregan

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

22.

Domínguez

J. I.

Lawson

Moreno

(Eds.). (2009). Consolidating Mexico’s democracy: The 2006 presidential campaign in comparative perspective. Baltimore, MD: Johns Hopkins University Press.

23.

Druckman

Nelson

K. R.

(2003). Framing and deliberation: How citizens’ conversations limit elite influence. American Journal of Political Science, 47, 729-745.

24.

Faughnan

B. M.

Zechmeister

E. J.

(2011). AmericasBarometer insights: 2011 (AmericasBarometer Insights Series 57). Vanderbilt University, Nashville, TN.

25.

Finan

Schechter

(2012). Vote-buying and reciprocity. Econometrica, 80, 863-881.

26.

Finkel

S. E.

Smith

A. E.

(2011). Civic education, political discussion, and the social transmission of democratic knowledge and values in a new democracy: Kenya 2002. American Journal of Political Science, 55, 417-435.

27.

Gans-Morse

Mazzuca

Nichter

(2014). Varieties of clientelism: Machine politics during elections. American Journal of Political Science, 58, 415-432.

28.

Gibson

E. L.

Calvo

(2000). Federalism and low-maintenance constituencies: Territorial dimensions of economic reform in Argentina. Studies in Comparative International Development, 35(3), 32-55.

29.

Glaeser

E. L.

Scheinkman

J. A.

Sacerdote

B. L.

(2003). The social multiplier. Journal of the European Economic Association, 1, 345-353.

30.

Gonzalez-Ocantos

de Jonge

C. K.

Melendez

Osorio

Nickerson

D. W.

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

31.

Granovetter

M. S.

(1973). The strength of weak ties. American Journal of Sociology, 78, 1360-1380.

32.

Greene

K. F.

(2007). Why dominant parties lose: Mexico’s democratization in comparative perspective. New York, NY: Cambridge University Press.

33.

Hatemi

McDermott

(2012). The genetics of politics: Discovery, challenges, and progress. Trends in Genetics, 28, 525-533.

34.

Hicken

(2011). Clientelism. Annual Review of Political Science, 14, 289-310.

35.

Huckfeldt

Johnson

P. E.

Sprague

(2004). Political disagreement: The survival of diverse opinions within communication networks. New York, NY: Cambridge University Press.

36.

Huckfeldt

Sprague

(1995). Citizens, politics and social communication: Information and influence in an election campaign. New York, NY: Cambridge University Press.

37.

Imai

Park

Greene

K. F.

(2015). Using the predicted responses from list experiments as explanatory variables in regression models. Political Analysis.

38.

Iñiguez

Govezensky

Dunbar

Kaski

Barrio

R. A.

(2014). Effects of deception in social networks. Proceedings of the Royal Society B, 281, 20141195.

39.

King

Honaker

Joseph

Scheve

(2001). Analyzing incomplete political science data: An alternative algorithm for multiple imputation. American Political Science Review, 95, 49-69.

40.

Kitschelt

(2000). Linkages between citizens and politicians in democratic politics. Comparative Political Studies, 33, 846-879.

41.

Kitschelt

Wilkinson

(2007). Citizen-politician linkages: An introduction. In Kitschelt

Wilkinson

(Eds.), Patrons, clients, and policies: Patterns of democratic accountability and political competition (pp. 1-49). New York, NY: Cambridge University Press.

42.

Lawson

(2002). Building the fourth estate: Democratization and the rise of a free press in Mexico. Berkeley: University of California Press.

43.

Lawson

Greene

K. F.

(2014). Making clientelism work: How norms of reciprocity increase voter compliance. Comparative Politics, 47, 61-85.

44.

Lazarsfeld

P. F.

Berelson

Gaudet

(1944). The people’s choice: How the voter makes up his mind in a presidential campaign. New York, NY: Columbia University Press.

45.

Levitsky

(2003). Transforming labor-based parties in Latin America: Argentine Peronism in comparative perspective. New York, NY: Cambridge University Press.

46.

Magaloni

(2006). Voting for autocracy: Hegemonic party survival and its demise in Mexico. New York, NY: Columbia University Press.

47.

Martin

P. S.

(2003). Voting’s rewards: Voter turnout, attentive publics, and congressional allocation of federal money. American Journal of Political Science, 47, 110-127.

48.

Mondak

(2010). Personality and the foundations of political behavior. New York, NY: Cambridge University Press.

49.

Nichter

(2008). Vote buying or turnout buying? Machine politics and the secret ballot. American Political Science Review, 102, 19-32.

50.

Nichter

Peress

(2014, August). Request fulfilling: When citizens ask for clientelist benefits. University of California San Diego.

51.

Ocantos

E. G.

de Jonge

C. K.

Nickerson

D. W.

(2014). The conditionality of vote-buying norms: Experimental evidence from Latin America. American Journal of Political Science, 58, 197-211.

52.

Roberts

(2013). Market reform, programmatic (De)alignment, and party system stability in Latin America. Comparative Political Studies, 46, 1422-1452.

53.

Rosenstone

S. J.

Hansen

J. M.

(1993). Mobilization, participation, and democracy in America. New York, NY: Macmillan.

54.

Schaffer

F. C.

(2007). Lessons learned. In Schaffer

F. C.

(Ed.), Elections for sale: The causes and consequences of vote buying (pp. 183-200). Boulder, CO: Lynne Rienner.

55.

Scheiner

(2006). Democracy without competition in Japan: Opposition failure in a one-party dominant state. New York, NY: Cambridge University Press.

56.

Sokhey

A. E.

Djupe

P. A.

(2014). Name generation in ego-centric network data: Results from a series of experiments. Social Networks, 36, 147-161.

57.

Stokes

S. C.

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

58.

Stokes

S. C.

Dunning

Nazareno

Brusco

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

59.

Szwarcberg

(2012a). Revisiting clientelism: A network analysis of problem-solving networks in Argentina. Social Networks, 34, 230-240.

60.

Szwarcberg

(2012b). Uncertainty, political clientelism, and voter turnout in Latin America: Why parties conduct rallies in Argentina. Comparative Politics, 45, 88-106.

61.

Szwarcberg

(2013). The microfoundations of political clientelism: Lessons from the Argentine case. Latin American Research Review, 48(2), 32-54.

62.

Szwarcberg

(2014). Political parties and rallies in Latin America. Party Politics, 20, 456-466.

63.

Verba

Nie

N. H.

Kim

(1978). Participation and political equality: A Seven-Nation Study. New York, NY: Cambridge University Press.

64.

Weitz-Shapiro

(2012). What wins votes: Why some politicians opt out of clientelism. American Journal of Political Science, 56, 568-583.

65.

Weyland

(1996). Democracy without equity: Failures of reform in Brazil. Pittsburgh: University of Pittsburgh Press.

66.

Zarazaga

Ronconi

(2013, February). The tragedy of clientelism: Opting children out. Working paper, Centro de Investigación y Acción Social, Buenos Aires, Argentina.