VOTING THEIR CONSCIENCE: POVERTY, EDUCATION, SOCIAL PRESSURE AND VOTE BUYING IN INDONESIA

Abstract

When citizens are offered cash and gifts by candidates, they can either vote for candidates offering the money, refuse the money, or take the money and vote for whichever candidate they prefer (e.g., “vote their conscience”). Which citizens “vote their conscience” versus the other options available to them? This manuscript examines the effect of poverty, education, and social pressure on how citizens in the Central Java city of Semarang, Indonesia react to offers of cash from politicians. We find that poverty and education interact to create important and interesting patterns of responses. The citizens most likely to “vote their conscience” are those with lower income yet higher levels of formal education whereas higher income, more highly educated citizens are most likely to refuse the money. We also examine factors that create social pressure for citizens to “sell” their votes by supporting the candidates offering cash.

INTRODUCTION

Vote buying—the offering of cash and gifts in exchange for voters’ support during elections—is a phenomenon that occurs in every region of the world (Schaffer 2007). When citizens are offered cash and other gifts by candidates, they essentially have three options. First, they can accept the money and vote either for the candidate who is offering the money or, if multiple candidates offer gifts, vote for the one that offers them the most money. This action fits the classic definition of “vote buying” because there is an apparent exchange of goods (often cash) for the citizen's vote. Second, citizens can refuse the money outright and vote for whichever candidate they prefer, essentially rejecting the candidate's apparent offer to “buy” their vote. Government officials and civic organizations frequently encourage citizens to engage in this action and thereby reject the practice of vote buying as immoral or unethical (see Callahan 2000 for a discussion about civic education and government efforts). Yet there is a third option available to citizens that has become a focus of efforts to undermine vote buying efforts. Because most democracies use a secret ballot, citizens can accept the money but vote for whichever candidate they prefer even if their preferred candidate did not offer similar gifts.

This article seeks to contribute to the understanding of vote buying in several ways. First, we directly examine which voters say they are more likely to “vote their conscience” compared to refusing the money or “selling” their votes to candidates offering them gifts. Despite the extensive and rapidly growing scholarly work on the micro-foundations of vote buying, few studies have directly examined variation in which citizens choose the “vote their conscience” option versus other options at the polls. Past studies provide insight for explaining this phenomenon, but the “vote your conscience” option also creates different and interesting patterns across citizens in their reactions to offers of cash and gifts from candidates. Some factors, for example, might shift responses from the “vote your conscience” option to rejecting the money while other factors might induce citizens to support candidates offering the money. Second, and relatedly, we examine the conditional relationship between poverty and education, finding that education has asymmetrical effects on responses across levels of income. Specifically, education appears to shift responses to the “vote your conscience” option among lower-income citizens while it leads higher-income citizens to reject the money outright. Third, we also examine the role of social networks and social pressure on citizens’ responses to offers of cash. We find that workers in Central Java (an area of Indonesia where many employers operate as brokers on behalf of candidates) and party loyalists are more likely to accept the money and to vote for the candidate offering it.

In the next section, we outline the factors that previous research has shown can promote or reduce the practice of vote buying. The extant literature highlights three prominent factors—poverty, education, and societal norms and social pressure—as influencing whether citizens demand money from candidates and whether they accept the money offered to them. Based on this previous research, we use these past studies to formulate theoretical expectations about which citizens will be most likely to “vote their conscience” versus rejecting the money or voting for candidates who offer the money. We also discuss social desirability bias and selection bias as possible competing explanations for our theoretical expectations. The following two sections present our methodology and the results from our multinomial logistic regression analysis, respectively. The last section provides an overview of our findings and discusses further limitations as well as possible normative and policy-related implications of our results.

WHO VOTES THEIR CONSCIENCE?

Widespread adoption of the secret ballot across the world creates an obstacle for candidates attempting to buy votes (Cox and Kousser 1981; Cox 1987). Specifically, citizens who receive money from the candidates are still free to vote for their preferred candidate and there are very few repercussions for such defection.¹ Candidates, government officials, and interest groups have attempted to capitalize on this fact to reduce and undermine the effectiveness of vote buying practices. In the Philippines, for example, vice-president Jejomar (“Jojo”) Binay reminded voters that they could pocket the money and vote for the candidate they preferred.² A variation of this message has been found in Bulgaria, Zambia, the Philippines, and elsewhere (Schaffer 2008). In Indonesia, several political leaders, government officials and outside interest groups throughout Central Java also expressed variations of this message, most commonly in the form of pamphlets reading “ambil uangnya—jangan pilih orangnya” (“take the money, but do not vote for them”). During the past several years, these appeals have become somewhat common across Central and Eastern Java and even promoted by government officials.³

Despite such appeals becoming relatively widespread across democracies around the world, there has been surprisingly little research on this phenomenon. One study on Sao Tome and Principe found that such messages lowered the perceived influence of vote buying and might have decreased electoral irregularities (Vincente 2013). Other research, however, has questioned the effectiveness of this approach. In one of the most direct tests of the “vote your conscience” message, Hicken et al. (2018) conducted a field experiment in the Philippines, finding that receiving the “vote your conscience” message actually increased vote-switching during the election, indicating that the messages had the opposite effect. Voter education efforts aimed at reducing vote buying have likewise proved to be either ineffective or counter-productive in countries such as Thailand (Callahan 2000), Taiwan (Rigger 2002), and the Philippines (Schaffer 2008). While debate continues about the effectiveness of such campaigns, less studied is variation in which citizens adopt the “vote your conscience” perspective versus other alternative reactions to cash and other gifts from candidates.

This section examines three separate but related factors identified by previous research as influencing how citizens react to offers of gifts and cash by candidates. The prominent theories and empirical studies agree that education and income are important variables related to this behavior but differ substantially on the precise causal mechanisms underlying them. First, poverty creates the “demand” for money—both direct need for financial assistance and differences in policy preferences. Second, past research suggests that education can serve as a countervailing force to vote buying by creating pro-democratic values that are less tolerant of corruption. We also argue that there is likely to be a conditional relationship between these two factors where education has differential effects across levels of income. The third factor involves social pressure by brokers and political parties that can induce citizens to support the candidates offering the cash. This can occur because citizens seek guidance from community leaders (who operate as brokers), they already support for the party and its leaders, or even because there exists overt clientelistic relationships between citizens and the parties or brokers.

POVERTY AND THE DEMAND FOR MONEY

The most obvious explanation for why poor voters accept money from candidates is poverty itself—low-income voters need the money for themselves and their families. Put simply, regardless of whether they think the practice is corrupt, some citizens face a “poverty trap” where they are forced to accept money from candidates because of their economic conditions (Gonzalez Ocantos, Kiewiet de Jonge, and Nickerson 2014). The value of the cash or gifts from candidates diminishes as a citizens’ wealth increases, making it more economical for candidates to “buy” the votes of lower-income voters (Stokes 2005). This can explain why candidates (through their brokers) often target poor voters with clientelistic appeals (Stokes 2005; Nichter 2008), and why brokers often feel pressure from lower-income citizens to provide cash and gifts in exchange for their support (see Szwarcberg 2013 for an example). This “poverty trap” creates the need for some lower-income citizens to accept the money even if they view the practice of vote buying as corrupt (Tawakkal et al. 2017a) and, from the campaign's perspective, “buying” their votes is cheaper than it would be for higher-income citizens.

Additional research has demonstrated that poverty often translates into asymmetry in policy preferences. Because poor citizens have immediate (often existential) financial needs, they often value the cash and gifts more than other types of benefits. This “demand side” explanation argues that lower-income voters prefer the immediate, individual benefits such as cash and gifts compared to the more long-term, deferred benefits of policy programs such as health care or free education (Scott 1972; Desposato 2006; see also Shin 2015 for a detailed discussion). For many lower-income citizens, these individual and immediate benefits are easier to understand and recognize than more abstract and diffuse policy benefits (Geddes 1994). Shin (2015) provides one of the most direct tests of this argument about asymmetry in voter preferences. Ranking preferences for patronage versus programmatic benefits, Shin found that poorer and less educated citizens in Jakarta prefer the patronage benefits (jobs and money) while wealthier and more educated ones chose programmatic policies (free education). Beyond the immediate need for money, in other words, poverty also creates different patterns of policy demands across citizens that incentivizes candidates and their brokers to offer cash and other commodities to low-income voters. Thus, one of the prominent explanations for vote buying is that poverty creates a demand by poor voters for cash, commodities, and other immediate or individualistic benefits from candidates and their brokers. This provides an opportunity for candidates and their brokers to win support among low-income voters by meeting this demand and thus “buying” their votes.

The primary effect of poverty, then, is to create demand for financial assistance from candidates and brokers. This leads to the expectation that poor voters will be more likely to accept the gifts offered to them during a campaign as compared to wealthier voters. Much of the extant literature also points to norms of reciprocity common in rural and lower-income areas that leads these voters to support candidates offering money. These norms and cultural values (often measured using income as a proxy variable) are certainly an important explanation for vote buying behavior and are discussed below. However, we would note that they are theoretically distinct from the effect of poverty itself as creating demand for cash and gifts from voters. The main point here is that poverty itself creates demand from voters for financial assistance from campaigns, leading to the expectation that poverty will increase the likelihood of accepting money but not necessarily the likelihood that voters will support the candidate offering such gifts.

EDUCATION AND COUNTERVAILING NORMS AND VALUES

Previous research has also identified education as a countervailing factor that can potentially reduce the likelihood that citizens demand money or, if offered, that they will accept it. While education is often used as a proxy variable to measure poverty and financial need, it is conceptually different from income.⁴ Past studies have shown that education measures a broader range of phenomena beyond poverty, including socialization into democratic values, different attitudes about government policy, and even exposure to media and political discourse (including anti-vote buying efforts from government and interest groups). In Indonesia, Mujani and colleagues (2011) demonstrate that citizens with higher levels of education are more likely to support democracy, are more tolerant of minorities, and are more likely to vote and otherwise engage in politics. Numerous studies have likewise examined the relationship between formal education and democratic values (Abramson and Inglehart 1994; Duch and Taylor 2014; Evans and Rose 2012) and education has long been promoted by “modernization theory” as an important impetus for developing pro-democratic attitudes among citizens (Lipset 1959; Inglehart and Welzel 2005).

Formal levels of education are also strongly related to lower tolerance of corruption in government, including vote buying (Sullivan and Transue 1999). In Nepal, for example, citizens with higher levels of formal education were significantly less likely to express tolerance towards thirteen different types of corruption (Truex 2010). Regarding vote buying behavior in Indonesia specifically, formal education is strongly related to rejection of gifts and money from candidates (Tawakkal et al. 2017a). In Peru, to give another example, those with higher levels of education were more likely to stigmatize or view vote buying negatively (Gonzalez Ocantos, Kiewiet de Jonge, and Nickerson 2014) while in Argentina higher education lowers the probability not just that a voter will receive gifts but also the probability that she will be influenced by those gifts on Election Day (Brusco et al. 2004). One explanation for this relationship is that less educated citizens have more difficulty understanding the “abstract” negative consequences of vote buying which are not always immediately obvious to them (Gonzalez Ocantos, Kiewiet de Jonge, and Nickerson 2014).

More generally, formal education is also related to differences in the meaning of democracy among citizens. Specifically, numerous studies have found that more highly educated citizens are more likely to hold a “procedural” conception of democracy that values individual liberty and freedoms while less well-educated citizens are more likely to hold a “substantive” conception that values economic and material benefits provided by government (see Dalton, Shin, and Jou 2007 and Canache 2012 for a discussion). In East Asia specifically, research has found that higher levels of education are strongly associated with a “procedural” conception of democracy (Chu, Huang, and Lu 2015). In the Philippines, for example, Schaffer (2007) argues that the upper- and middle-class elites viewed taking money from candidates as a corrupt “selling” of votes whereas the lower-class citizens were more likely to view such gifts and monetary offers in terms of human dignity and respect shown to common people (Schaffer 2007, pp. 164–174).⁵ Thus, underlying perceptions about vote buying are often different conceptions about the meaning of democracy itself and, particularly, a view that democracy requires positive economic and social benefits for citizens instead of purely “procedural” processes. The relationship between levels of education and acceptance of vote buying practices is not simply about poverty or short-term material gain, in other words. For some citizens with lower levels of formal education, cash and other gifts from candidates represent symbols of dignity and respect that are part of their conception of democracy instead of undermining it.

In sum, while poverty creates a demand for cash and gifts from candidates, previous research finds that education is associated with the adoption of more pro-democratic norms and values that can increase recognition of the abstract, long-term negative consequences of vote buying. Thus, citizens with higher levels of education are more likely to view vote buying as corrupt. These studies, in other words, view education as a potential countervailing factor that lead us to expect that citizens will be more likely to reject the cash offers outright instead of accepting the money.

EDUCATION AND POVERTY—A CONDITIONAL RELATIONSHIP

The strategy of “take their money, but do not vote for them” would obviously seem targeted at lower-income voters. The primary reason governments and interest groups would encourage citizens to engage in an illegal activity (even if there is little risk of prosecution) is because these actors realize low-income citizens need the money for themselves and their families.⁶ Upper-income citizens, in contrast, do not need the small cash payments or gifts and would be more likely to simply refuse them, if the campaigns even bothered to offer gifts to them in the first place. Regarding education, the previous studies have tended to view it as a countervailing force that can change people's attitudes and beliefs about the practice of vote buying, thus leading more highly educated citizens to reject cash offers outright instead of accepting the money.

Yet the effect of education is also likely to vary across income levels. Among lower-income citizens, higher education levels might increase the stigma regarding vote buying and lead to the recognition of vote buying as normatively undesirable, but the financial pressure to accept money and gifts that their families need might outweigh these normative considerations. Moreover, there is evidence that lower-income citizens are more likely to experience conflict between needing the money and viewing the practice of vote buying as corrupt or morally wrong (Tawakkal et al. 2017a). Because of their financial need, lower-income citizens with higher education levels would probably not be more likely to reject the money offered by candidates. Instead, we expect these citizens (low income, higher education) to be more likely to accept the money yet vote for other candidates (e.g., “vote their conscience”). In other words, among lower-income citizens, higher levels of education might change their attitudes about vote buying but not their behavior in terms of accepting the money. To the extent that their behavior changes regarding how they vote, the shift should be from voting for the candidates offering the money (or the most money) to the “vote your conscience” option. The same financial considerations would not exist among middle- and upper-income citizens, who would not need the small amounts of cash or gifts offered by candidates. For these citizens, education would serve to change both normative attitudes about vote buying and, because they do not need the money, would lead to behavioral changes in the form of increased likelihood of rejecting the money outright.

We therefore expect that education will have varying effects depending on a citizens’ economic or financial circumstances. The citizens most likely to “vote their conscience” are expected to be those with lower incomes and less education. The citizens most likely to reject the money are expected to be those with higher levels of income and education.

SOCIAL PRESSURE—COMMUNITY LEADERS AND POLITICAL PARTIES

Poverty can explain why some citizens demand (and accept) money from candidates, and education can help explain why some citizens (e.g., low income) will “vote their conscience” while others (e.g., upper income) will reject the money. But there are also factors that might induce citizens to support the candidates offering them cash. A rapidly growing body of research can provide insight into this phenomenon by examining the role of brokers and the often-social nature of offering cash and gifts to the voters.⁷ The use of brokers to distribute goods to voters is common across the world and the prominent theories of brokerage activity underscore their central importance to the practice of vote buying (Stokes 2005; Stokes, Dunning, and Nazareno 2013; Aspinall 2014; Aspinall and Sukmajati 2016). Campaigns do not simply hire brokers as couriers to deliver these goods, however (Tawakkal et al. 2017b). Instead, most candidates seek out brokers because they are community leaders (religious leaders, business owners, village elders, etc.) with high social status and thus have social influence over voters in their respective communities (for a partial list, see: Callahan 2000; Finan and Schechter 2012; Aspinall and Sukmajati 2016; Aspinall et al. 2017).

There is evidence that voters look to such community leaders for information and guidance on which candidates to support. Callahan, for example, noted that, “some rural voters will readily agree to support candidates who are known to have the backing of leaders such as the district or village headman, the schoolteacher, or the abbot of the local temple” (Callahan 2000, 19). Norms of reciprocity are common across the world (Finan and Schechter 2012; Ocantos, de Jonge, and Nickerson 2014) and are particularly prominent in Javanese society. Our own interviews with candidates, brokers, and citizens across Indonesia (including Central and Eastern Java, Kalimantan, Bali, Lombok, and Manado) reinforce the findings of previous research that local community leaders are aggressively sought out by candidates to work as brokers, and that citizens (especially in rural Central Java) often look to such community leaders for counsel and advice during elections. Moreover, the money offered by brokers, especially community leaders, is rarely couched in transactional terms. Instead, it is common for them to present the money to voters as gifts of friendship and respect on behalf of the candidates (Schaffer 2007; Aspinall et al. 2017), making the rejection of such gifts more difficult for voters operating within a norm of reciprocity.

As discussed above, poor and rural voters are often the target of such gifts and therefore income as a proxy variable is likely to measure some of this phenomenon. Yet we consider two additional factors that might induce citizens to support the candidate offering them cash. The first involves social pressure from community leaders that sometimes involve outright clientelistic relationships. In Central Java, and particularly the city of Semarang, there is a large industrial sector that includes several manufacturing businesses. Workers at these businesses thus comprise a sizable portion of the citizenry in Central Java and the city of Semarang. As discussed above, candidates often seek out prominent community leaders, including business owners, to work as brokers in their campaigns (for additional examples of business owners as brokers, see Scott 1972; Brusco, Nazareno, and Stokes 2004; Gingerich and Medina 2013; and Hellmann 2014). This practice is especially common throughout Central Java, where candidates recruit business owners precisely because of their influence over the workers in their businesses (Aspinall and Sukmajati 2016; Tawakkal et al. 2017b). Moreover, many of the business owners who operate as “brokers” in the campaign hired workers and managers from their businesses to oversee the distribution of money to voters.⁸ Because business owners often work directly for campaigns and distribute money to their own workers, this creates pressure on the workers not only to accept the money but also to support the candidate endorsed by their employer.⁹

The second factor involves a sense of loyalty that many voters feel to their political parties and brokers often appeal to such loyalty during elections.¹⁰ Stokes, Dunning, and Nazareno (2013), for example, find that brokers frequently target both loyal supporters and potential “swing” voters for cash offers during the election.¹¹ Indeed, one of the primary components of the “broker-mediated theory” proposed by Stokes and colleagues is that parties use brokers because they have better information about the intentions of voters in their respective communities. Other research has found that party loyalists are often targeted for “turnout buying” because monitoring voters at the ballot is difficult and party loyalists as known supporters are more likely to support the candidate offering money (Nichter 2008; Carkoglu and Aytac 2015; Larreguy, Marshall, and Querubin 2016). When parties frequently engage in vote buying behavior, they can socialize voters into viewing the practice not only as normal but also “as a perverse form of ‘welfarism,’” (Gonzalez Ocantos, Kiewiet de Jonge, and Nickerson 2014, 200). Thus, party loyalists are not only more likely to be offered cash by brokers, but they also might have different normative or ethical views about the practice that make them more likely to accept the money. Because they are offered money by brokers precisely because of their preexisting support for their party's candidate, party loyalists are obviously more likely to support the candidate offering the money.

Determining the precise causal mechanism underlying the effect of party loyalty is difficult. In one sense, such behavior could actually represent a form of “voting their conscience,” because these voters might support their party for reasons beyond the money offered by candidates, which could include party programs, ideology, or candidate personality. Such voters would merely be receiving money to vote for a candidate they already support, which would arguably represent them voting their “conscience.” On the other hand, in Indonesia and abroad party loyalty is also sometimes based on overt clientelistic relationships. Some party loyalists might feel pressure to support their party out of fear of losing their jobs or other benefits accrued from participation with the party. Thus, it could be that party loyalists are voting their sincere beliefs and only receive the money because of a strategic “turnout buying” calculation by brokers or, alternatively, that party loyalists might feel pressured to support the party because of clientelistic relationships. Regardless of the reason, however, we expect that party loyalists will be more likely than other citizens to accept the money and to support the candidate offering it to them.

SOCIAL DESIRABILITY AND SELECTION BIAS

Finally, we would note two methodological concerns that might explain the hypothesized patterns of responses discussed above. The first involves the well-documented phenomenon of “social desirability” bias where respondents feel pressure to provide the “correct” or socially desirable response. This effect is especially strong regarding survey questions about sensitive topics, including illegal behavior such as vote buying. Gonzalez Ocantos, Kiewiet de Jonge, and Nickerson (2012) and Kiewiet de Jong (2015), for example, use list experiments to estimate the extent to which citizens misrepresent receiving gifts from candidates in Latin America, finding that citizens with higher levels of education and middle incomes are more likely to “lie” about receiving such gifts. Kiewiet de Jong (2015), in particular, finds significant variation across Latin American countries in the propensity of voters to “lie” about receiving gifts from candidates, with the highest rates occurring in Nicaragua. This research raises a serious concern that the hypothesized patterns of response detailed in previous sections might be due to social desirability bias—more highly educated and middle-income citizens being more likely to give the “correct” or socially desirable answer—instead of the theoretical factors identified by past studies.

While social desirability bias is indeed a concern for almost all survey-based research on sensitive topics or those relating to illegal activity, we believe the effect is less likely to bias the findings presented below for several reasons. First, Central Java has become a focus of vote-buying studies precisely because of the prevalence of the practice and the openness with which various actors discuss it. This is especially true compared to Nicaragua, where the previous studies found that only 2.4 percent of respondents were willing to report receiving a gift or favor from candidates (Gonzalez Ocantos, Kiewiet de Jonge, and Nickerson 2012, 210). By contrast, our survey in Central Java includes almost 60 percent of respondents admitting that they would accept the gifts offered by candidates, including over 45 percent of respondents with high levels of income. In addition, almost 40 percent of respondents in our survey also found it acceptable for candidates to distribute cash to voters. Thus, there is a greater openness and willingness among respondents in our sample to declare their willingness to accept the money offered, suggesting that the pressure to give socially desirable answers is lessened. Moreover, Amick (2018) employed a list experiment in the Tegal regency (located near Semarang, Central Java) to determine the scope and extent of social desirability bias among Indonesian citizens. Only approximately 4 percent of respondents in Amick's study underreported receiving cash or in-kind gifts from candidates, demonstrating that social desirability bias is less of a concern for studies of Central Java (and perhaps Indonesia generally) than is the case for Latin America and other regions.

In addition, even if social desirability bias might explain some of the separate effects of income and education, it is much less likely to explain the interaction between these two factors. If the “correct” or socially desirable response is to reject the money offered by candidates, then more highly educated citizens should be more likely to give this response regardless of their income levels. We hypothesize, however, that the effect of education will vary systematically across citizens’ income levels, with education having little or no effect on the likelihood that low-income citizens will refuse the money. Instead, our expectation is that education will shift responses among lower-income citizens from accepting the money and voting for that candidate to the “vote their conscience” response. Only among the higher levels of income should education induce citizens to refuse the money. Thus, for social desirability bias to explain this interactive effect, the socially desirable or “correct” answer would have to systematically differ across citizens’ income levels. We find it improbable that citizens would systematically differ across levels of income in their perceptions of the “correct” answer and we are unaware of a theoretical framework that would lead to this expectation.

Lastly, we also acknowledge that our survey-based approach is limited by the inability to independently verify how citizens voted in the election. Our survey asks people how they would respond to a hypothetical instance of candidates offering money, not whether they actually received the money nor how they actually voted in a specific election. Thus, we measure only how citizens say they would react to offers of cash by candidates and not actual voting behavior by these citizens. The studies discussed above show that some citizens will lie in a survey about receiving the money and, presumably, about how they voted in the election. This is a limitation inherent in all survey-based approaches to voting behavior.¹² Without minimizing this concern, we would note that the mechanism that allows citizens to “vote their conscience”—the secret ballot—also makes a direct measure of which candidate the voters supports virtually impossible.

Despite this limitation, we also believe that our approach has some additional strengths over alternative methodologies. Many of the studies cited above demonstrate a selection bias where higher-income (and possibly more highly educated) citizens often do not receive the money from candidates. Asking respondents if they received offers of cash or gifts from candidates would result in many respondents (especially high-income ones) saying they had not. Presumably, these responses would be excluded from subsequent questions asking whether or not they voted for the candidate offering the money, creating a sample selection bias where a non-random segment of the sample would be excluded from the analysis. Thus, even survey-based approaches (e.g., asking citizens how they voted and whether they received money) would be prone to this sample selection effect if the probability of receiving money were significantly higher for poorer citizens than wealthier ones. Our survey avoids this possible selection bias because the results do not depend on voters actually receiving the money as a precondition for asking how they would vote in response. We are thus able to measure the responses from citizens (especially those with higher incomes) who ordinarily do not receive cash and other gifts from candidates because they would either refuse the offer or the small amount of money would have little impact on them.

DATA AND METHODS

The data are drawn from a stratified probability sample of citizens in the area around Semarang, Indonesia. Semarang has a population around 1.2 million people and is located in the Central Java province, one of the largest provinces in the country. Moreover, vote buying is prominent in Central Java, including the areas surrounding Semarang (see Aspinall and Sukmajati 2016 for research on vote buying in Central Java). Our sample consists of randomly selected individuals from 19 kecamatan, which are political subunits of the larger regency. The sample size varies across kecamatan from a low of 25 respondents to a high of 72 respondents. There are a total of 1,005 respondents in the survey, although response rates varied across questions.

The dependent variable is created from responses to a question asking how respondents would react if offered money by the candidates. Respondents who said that they would accept the money and either vote for that candidate or vote for the candidate who offered the most money (“Accept Money—Vote Candidate”) were coded 1. Those who said that they would accept the money but vote for the candidate they preferred (“Accept Money—Vote Conscience”) were coded 0. And respondents who said that they would simply refuse the money (“Refuse Money”) were coded -1. Because the dependent variable is dichotomous and categorical, the models were estimated using multinomial logistic regression. The “Refuse Money” category was left as the baseline. The model was also estimated using clustered standard errors to account for the fact that the respondents are clustered by kecamatan, which could violate the “independently and identically distributed” assumption for multinomial logistic regression and could artificially inflate the statistical significance of the results (Primo, Jacobsmeier, and Milyo 2007).

The independent variables include income, education, employment as a worker, gender, age, and influence of party on voting decision. Income is measured as a five-point scale indicating monthly income that ranges from less than 1 million rupiah per month at the low end of the scale and greater than 5 million rupiah a month at the upper end. Education is measured as a six-point scale ranging from “0” for respondents who did not complete primary school to “5” for respondents who had a college or graduate level degree. Age is also a six-point scale with respondents who are 17–22 years old coded as “0” and those over 60 years old coded as “5.” Gender and employment as a business worker are measured as dummy variables where “1” indicates that the respondent is male or is a business worker, respectively, while “0” indicates that the respondent is female or has another employment status. Finally, the survey did not include a direct question about respondents’ loyalty or attachment to a political party. However, the survey did include a question asking which among several factors was most important for their voting decision in the local election. Respondents who chose “party” as the most influential were coded “1” while those choosing other factors (family, spouse, village, religious leaders, etc.) were coded “0.” While the operationalization of worker status and party loyalty are somewhat indirect, we believe that these variables capture enough of the variation due to party loyalty and employment in businesses involved with vote buying to serve as a conservative test of the theoretical expectations. We also conducted a robustness test to ensure that the measurement strategy did not significantly alter the results in the models presented below. We included each of the categories for the two variables (leaving one category each as a baseline) and found that the results remained consistent with the theoretical expectations discussed above.¹³

In addition, religion might seem to be an obvious control variable to include in the model. However, this area of Central Java is overwhelmingly Muslim, and thus less than 3 percent of the respondents in the sample indicated belonging to another religion. This lack of variation in the sample (which is broadly representative of Central Java) prevented us from including religion as an independent variable in the models. Moreover, as discussed in the previous section, the effect of religion and influence by religious leaders is not theoretically straightforward. Our own interviews reveal significant variation across mosques in Indonesia regarding support and opposition to the practice of vote buying. The literature discussed above demonstrates that religious leaders are often chosen as brokers, but often these religious leaders do not distribute the money to the voters. In other circumstances, such as one instance we observed in Banjarmasin (Kalimantan, Indonesia), a religious leader told followers during service that accepting money was haram (forbidden) yet personally gave money to his followers the night before the election. Our survey is unable to distinguish between mosques (and their respective imams) who condemn vote buying versus those that promote or engage in the practice and we therefore excluded religion and influence by religious leaders from the model.¹⁴

RESULTS

Table 1 presents the results for the multinomial logistic regression model, with the “Refuse Money” category left as the base outcome in the latter model while Table 2 presents the substantive effects of each variable in the model. The last column also includes a likelihood ratio test for difference of coefficients for each variable between the “Vote Candidate” and “Vote Conscience” models. The p-values for these likelihood ratio tests indicate the probability that there is no difference between the coefficient in the “Vote Candidate” and “Vote Conscience” model. For example, the coefficients for education in the first two columns are statistically significant (p < .10), but there is also a statistically significant difference between the two education variables between the two models.

Table 1 Multinomial Logistic Regression Models for Vote Buying

Note: Cell entries are unstandardized logistic regression coefficients (Vote-Buying Acceptable) and unstandardized multinomial logistic regression coefficients (Accept Money) with robust standard errors in parentheses. “Reject Money” is left as the baseline category for the multinomial logistic regression model.

Significance Level: *** p < 0.01, ** p < 0.05, * p < 0.1.

Table 2 Predicted Percentages for Vote Buying Responses, by Independent Variable

The coefficients for each variable in Table 1 are interpreted in relation to the base category (“Refuse Money”). Thus, higher levels of education reduce the probability that citizens will accept money from candidates and increases the probability that they will refuse the money. The effect is stronger for the “Vote Candidate” category than for the “Vote Conscience” category, meaning that the largest shifts from the lowest to the highest educated respondents is from the “Vote Candidate” response to the “Refuse Money” response. Both of the education coefficients are strongly significant at the .01 and .05 significance levels, respectively. Likewise, both coefficients for income are negative and statistically significant at the .05 significance level, indicating that higher-income respondents are more likely to refuse the money than to accept it.¹⁵ The same is true for the gender variable, which is negative and statistically significant for both columns. The coefficient for age is only significant in the “Vote Conscience” column and is negative, indicating that older respondents shift from accepting the money and voting their conscience to refusing the money offered by candidates. There is not a statistically significant effect of age on the “Vote Candidate” category.

Turning to the worker variable, the coefficients are positive and statistically significant (.05 level for “Vote Candidate” and .10 level for “Vote Conscience”), meaning that those who are employed as workers are more likely to accept the money, and much more likely to vote for the candidate who offers them money, than are respondents who work in other sectors. While the size of the effect between the two models is large, the difference in the coefficients, is not statistically significant (p < .201). It should be noted that this effect is independent of both education and income (as well as age and gender), meaning that there is something unique about workers that cannot be explained by lower levels of education or income. This is consistent with our interviews and observations regarding the role of business owners in political campaigns, often acting as “brokers” who are hired because they have influence over their workers.¹⁶ While there is a statistically significant effort for worker status, however, the substantive effect is not overwhelming. For workers, there is about an 18.9 percent chance that they will accept the money and vote for the candidate who offered it (or who offered the most), compared to a 13.3 percent chance for those working in other areas. Workers have about a 49 percent change of voting their conscience compared to a 44 percent change for others. And workers are only about 32 percent likely to refuse the money compared to 42.8 percent of those outside the labor sector.

Regarding party influence, the coefficient is positive and statistically significant (p < .05) in the “Vote Candidate” model while the coefficient is negative but not statistically significant in the “Vote Conscience” model. However, the difference in the coefficients between the two models is strongly significant (p < .001) and rather large, meaning that party influence increases the likelihood that respondents say that they will vote for the candidate offering the money and decreases the likelihood that they will vote their conscience, but has little to no impact on the probability of refusing the money. In substantive terms, those most influenced by the party have a 42.3 percent chance of voting for the candidate offering the money compared to only 13.8 percent of those influenced by other factors. Similarly, only 17.9 percent of respondents most influenced by the party will “vote their conscience,” compared to about 46 percent of respondents most influenced by other factors. But the difference between the two groups in their likelihood of refusing the money is virtually zero (39.9 percent of those most influenced by the party compared to 40.1 percent of those influenced by other factors). In summary, party influence does not change people's likelihood of refusing the money, but rather shifts their behavior from voting their conscience to voting for the candidate offering the money.

Finally, to test the interaction between income and education, we used split-sample regression to estimate the multinomial logistic regression model separately for low, medium, and high levels of income.¹⁷ As expected, education has a statistically significant effect only for respondents with middle and higher levels of income, while the effect is small and statistically insignificant for those with lower incomes. Interestingly, however, the likelihood ratio test for the low-income model is statistically significant (p < .049), meaning that education does not change the likelihood that lower-income voters will reject the money, but rather shifts their responses from voting for the candidate to the “vote their conscience” category. Table 3 presents the predicted percentages by income and education.

Table 3 Predicted Percentages for Vote Buying Response, by Income and Education

As the table shows, there is a modest (and statistically significant) shift from those who say they would accept the money and vote for the candidate versus those that would accept the money yet vote their conscience. There was little change among the lower-income respondents in the probability that they would refuse the money and, again, the effect is not statistically significant at conventional significance levels. Regardless of education level, among these lower-income respondents the strong majority were predicted to accept the money (more than 65 percent) versus refusing it (between 32–35 percent).

The effect of education is much larger, and is statistically significant at conventional significance levels among those with medium and high levels of income. In both cases, the largest shifts were away from accepting the money (either voting for the candidate or voting their conscience) and towards the “Reject Money” category.¹⁸ Among those with the highest level of income, for example, a large majority of respondents with lower levels of education were predicted to accept the money (16.5 percent voting for the candidate and 54.4 percent voting their conscience), with just less than 30 percent of respondents predicted to refuse the money. Among those with higher levels of education, on the other hand, about 67.5 percent were predicted to refuse the money while almost 30 percent were predicted to accept the money and vote their conscience. Only three percent of respondents with high levels of both income and education were predicted to take the money and vote for the candidate who offered it.

CONCLUSIONS

When candidates or parties offer money to citizens in efforts to “buy” their votes, citizens have three basic options—they can accept the money but “vote their conscience,” they can accept the money and vote for the candidate offering it, or they can reject the money. This article has sought to contribute to the growing scholarly research on vote buying by directly examining which of these three responses citizens in the city of Semarang, Indonesia, choose, including the effect of three related factors (poverty, education, and social networks or social pressure) on those responses. Our findings suggest that the citizens most likely to “vote their conscience” are low income, more highly educated voters. For these citizens, poverty creates a need to accept the money, while education leads them to vote for their preferred candidate. In contrast, the citizens most likely to accept the money and vote for the candidates offering it are party loyalists, employees, and low-income/less educated voters. And finally, the citizens most likely to reject the money are middle- and upper-income citizens who also have higher levels of formal education. These patterns of responses suggest complex and subtle interactions among the various factors that deserve greater attention in future studies of voting buying in Indonesia and elsewhere.

In addition to the social desirability issue discussed above, there is another limitation with these findings that should be noted. The survey was taken in the surrounding areas of one city in Central Java and may not be representative of such a large and diverse country as Indonesia. In addition to differences across the many islands of Indonesia, attitudes about vote buying (as well as how vote buying is practiced by candidates) might differ across the island of Java itself. While we do not attempt to generalize about attitudes regarding vote buying across all of Java, let alone the entire country, we would note that Central Java is one of the largest and most politically influential provinces in Indonesian politics. Moreover, Central Java has received significant recent scholarly attention, because vote buying is very prominent in this region (see Aspinall and Sukmajata 2016). Thus, despite our results being limited in terms of generalizability to the entire country, we also believe that our findings can contribute to the scholarly literature and help better understand which citizens are more likely to “vote their conscience” versus other options at the ballot box.

Our findings also have interesting and important implications for studies of vote buying and government efforts to reduce the practice. The asymmetrical results for education across levels of income, for example, suggest that education alone is not sufficient to stop citizens from accepting the money. In one sense, this indicates that political socialization efforts by government in the schools will only have limited impact on citizens. The effect of poverty is simply too strong to convince more than a plurality of lower-income citizens to reject the money. On the other hand, even among lower-income voters, education does appear to impact whether citizens “sell” their votes to candidates. For citizens of all levels of income, in other words, education appears to make them more independent-minded and more likely to vote for their true preferences during the election. Our results suggest that this effect of education simply manifests in different ways depending on the person's financial situation. For low-income citizens who need the money, the effect of education manifests as the “vote your conscience” option. However, because higher-income citizens do not need the money, the greater independence from education manifests as the “reject money” option. In both cases, however, more highly educated citizens appear more likely to pursue their preferences and make electoral choices independent of the influence of money.

Finally, the results for employment status and party loyalty also provide an intriguing opportunity for future research. As discussed above, the precise causal mechanism underlying these results—overt clientelistic relationships versus “turnout buying” for those who already support the party leaders and policies—is difficult to determine from our data. If the results we report are the consequence of clientelism and pressure from employers or parties, then this suggests a major obstacle to reducing and undermining vote buying practices through the “vote your conscience” option. In other words, to the extent that our results are driven by clientelism, this group of citizens would appear to be “selling” their votes in exchange for patronage benefits or maintaining their employment status.¹⁹ On the other hand, citizens may look to their employers for guidance on political matters as they do to other community leaders or they may vote for a candidate because they support the party leaders and its policy agenda. In this scenario, these citizens would make choices that would be somewhat similar even if they did not receive the money and this would not necessarily represent them “selling” their votes in exchange for cash.²⁰ We believe that some combination of these competing causal mechanisms is most likely, and our interviews with citizens, moreover, provide some tentative evidence to support both mechanisms. However, the normative implications of the alternative explanations for this behavior make uncovering the causal mechanisms and possible variation across voters regarding this behavior an important and interesting avenue for future research.