VOTE BUYING IN COMPARATIVE PERSPECTIVE
Vote buying is usually defined using phrases such as “the exchange of money, goods, or services for votes” (Schaffer 2008a, 1) or “the exchange of private goods for votes during electoral campaigns” (Gonzalez Ocantos, Kiewiet de Jonge, and Nickerson 2014, 197). Such definitions seem straightforward, conjuring an image of a simple economic transaction, like buying an item in a shop. However, when politicians hand out cash, goods, or favors to voters at election times, the element of exchange at the heart of such definitions is often at best implicit, and usually unenforceable. It is implicit, because vote buyers often do not explicitly request or demand a vote in exchange for their gift, especially in cultural contexts where doing such a thing would be considered coarse. It is unenforceable formally because vote buying is typically illegal (in Indonesia it is proscribed by article 86 of the Elections Law of 2012), but also because, if the secrecy of the ballot is upheld, candidates have difficulty knowing whether recipients deliver on their side of the bargain (Schaffer and Schedler 2008, 17–20). Accordingly, survey and qualitative research suggests that, often, less than 50 percent of voters who receive payments from a candidate or party end up voting for the payer (Schaffer 2008b, 185). Our findings from Indonesia confirm this general picture: in a sample of candidates presented below, the average vote total candidates received was 27 percent of the number of payments they distributed through their teams.
In response to this difficulty, scholars have pointed to different factors to explain the persistence of vote buying. For example, some argue that the effectiveness of vote buying is contingent on the ability of candidates or parties to circumvent the secrecy of the ballot. Grassroots brokers who are embedded in local communities may be able to “draw inferences about how voters voted” (Stokes et al. 2013, 35) and so punish people who do not comply with the deal. Candidates may use brokers who mobilize community groups rather than individuals, and so monitor collective rather than individual voting (Kitschelt and Wilkinson 2007, 17). Another influential interpretation, and one we examine in this article, is that parties may target payments at core voters who are already inclined to support them on ideological or other grounds, but need an incentive to come to the polling booth. Previously, many scholars assumed that parties with limited resources would target swing voters, because “[t]o reward voters who are ideologically proximate to the party or ideologically distant from it is to waste resources” (Stokes et al. 2013, 32).More recently, however, others (see especially Nichter 2008) have suggested that what is typically understood by observers as being vote buying is better understood as turnout buying, with parties or candidates distributing cash and goods to voters who are already inclined to support them. The function of the payment, in this view, is not to “buy” a vote, but to ensure that a sympathetic voter comes to the polling booth on voting day.
Another response to the problem of enforceability of the vote buying exchange stresses the cultural context in which vote buying takes place. Political scientists have long recognized that social norms of reciprocity may create a cultural scaffolding by which payments can generate a sense of moral obligation on the part of recipients to repay the buyer with their votes (Schaffer and Schedler 2008, 21–25; Finan and Schechter 2012). Accordingly, many political candidates and brokers who distribute such gifts do so without explicitly demanding a quid pro quo. Recipients might not even view the cash the politician distributes as part of a commodity exchange, as implied by the phrase “vote buying.” Instead, they might see such gifts in terms of a moral economy of gift-giving, or understand them as signaling the candidate's munificence or fitness for office. Walker (2014), writing on rural Thailand, for example, argues that cash gifts at election times are embedded in a wider system of gift exchange, pointing to “the elaborate and conspicuous ways in which money is used to signal enduring relationships at temple festivals, weddings and funerals.”
In fact, recent literature on vote buying and electoral clientelism has evinced disagreement about fundamental premises. Political scientists have tended to assume a strongly strategic logic to vote buying, and thus look for patterns of purposive targeting in vote buying strategies, and interest maximization in responses by voters. In contrast, anthropologists tend to stress symbolic logic, emphasizing the culturing moorings of vote buying in gift rather than commodity exchange and the culturally laden meanings inherent in payments to voters. Indeed, some anthropologists and other scholars are skeptical of the image they see in the political science literature of “profiteering voters wielding the abacus of rational choice” and motivated by the material benefits gained from vote buying (Piliavsky 2014, 16). In this view, the circulation of cash during election campaigns is “a (visible, material) sign of other kinds of (invisible, immaterial) resources”—such as a candidate's social authority, the strength of his or her social networks, and such like (Björkman 2014, 629).
To anticipate briefly how our findings and arguments in this article fit within these broader debates, we should stress that we find echoes of turnout buying in Indonesia, albeit in a context of party weakness that largely results from Indonesia's adoption of an open-list PR system whereby candidates compete against each other for personal rather than party votes. In explaining their vote-buying strategies, candidates repeatedly stressed they were targeting ‘base’ voters, though they categorized voters as loyalists mostly according to whether they had direct personal connections with them, or personal connections that were mediated by brokers, rather than on the basis of ideological or party ties. As a result, this categorization was often tenuous. On the one hand, candidates’ concerns to match or outbid payments offered by their rivals showed that they believed that the votes of even their so-called loyal voters were up for sale. On the other hand, many candidates adopted a strategy of “portfolio diversification” (Magaloni, Diaz-Cayeros, and Estévez 2007), focusing on a core of supposedly loyal voters, and sometimes even offering them additional payments to secure their votes, while simultaneously relying on their brokers to extend beyond that base and bring in additional voters on the strength of those brokers’ personal ties. Viewed in this way, brokerage networks were a method for scaling up a candidate's personal patron–client networks to encompass a greater number of voters. In practice, this meant that the distinction between core and uncommitted voters was often blurred.
At the same time, and in keeping with the general thrust of much of the political science literature, we show that candidates deployed a strategic logic and invoked the language of the marketplace when explaining their vote-buying efforts. For example, they frequently referred to the ‘market price’ for votes and accused voters of operating according to a ‘transactional’ logic. In their direct interactions with voters, it is true, candidates and brokers downplayed the significance of their cash gifts and emphasized their emotional bonds with recipients, in line with what we might expect from the anthropological literature. It is also noteworthy that nobody used the term ‘vote buying’ or close cognates to explain these transactions, being all too aware that many voters who received payments would not deliver their votes in return. Behind their backs, however, candidates frequently described voters as money-grubbing, self-interested and perfidious and drew on a language of market calculation when describing their strategies. In short, many candidates mixed market calculation and moral discourse about reciprocity, with their choice of language being largely determined by audience.
The research context and our methods
Existing literature on Indonesian politics often mentions the importance of clientelism—the exchange of material benefits for political support—in the country's political fabric, but few studies explore it in detail. Numerous works on elections in regional Indonesia, and local politics more broadly, mention vote buying—often glossed as ‘money politics’—while making other points about the nature of local political coalitions and class power (see for example Hadiz 2010, 2012; Tans 2012). Others explore the interaction of clientelism with other phenomena such as the party system (Allen 2014, Tomsa 2014), the personal vote (Allen 2015) or ethnic politics (Van Klinken 2008). Works that examine electoral clientelism itself focus on topics such as the role of personal networks (Buehler 2009), brokerage by community leaders (Clark and Palmer 2008) or campaign teams (Aspinall 2014b), and the distribution of club goods, or collective benefits (Aspinall and As'ad 2015). So far, however, we lack studies that detail the mechanisms and strategies employed by candidates who practice individualized vote buying.
To investigate systematically the role of clientelist exchange in interactions between candidates and voters, a large multi-sited research project was organized in the lead up to, and immediately following, Indonesia's legislative election of April 9, 2014 (see our acknowledgement note for details). Fifty researchers interviewed candidates and campaign workers, observed campaign events, and shadowed candidates. Forty of the researchers were concentrated in ten national electoral districts in the provinces of Aceh, South Sumatra, Banten, Jakarta, West Java, Central Java, East Java, Central Kalimantan, North Sulawesi and East Nusa Tenggara. These regions were selected to capture variation in social and cultural settings across Indonesia: urban, peri-urban and rural; ethnically heterogeneous versus relatively homogeneous; majority Muslim versus majority non-Muslim; relatively high versus low income; resource rich versus resource poor, and so on. Ten researchers were placed in additional locations across the country.
The researchers in this project found evidence of vote buying in the form of the “dawn attack” in most of the locations we studied. However, the practice was apparently most widespread, and certainly most socially accepted, in two constituencies in Java that form the focus in this article. In both places, most candidates and brokers discussed vote buying with us, describing their methods and strategies in detail, and we were able to observe many stages in the vote buying process directly. In these constituencies virtually all competitive candidates engaged in vote buying: we believe that, of the 70 candidates interviewed for this article, only two did not distribute cash to voters and still won seats. We also interviewed a handful of non-competitive candidates—individuals whose function was to fill out party lists but who never had the resources or intention to campaign seriously—who did not distribute cash. But the large majority of candidates did so.
In other locations covered by the research, most—though not all—candidates were reluctant to discuss their vote-buying efforts, and they often denied they engaged in the practice. We cannot determine conclusively whether vote buying was more prevalent in the areas of Java that are the focus of this article, or merely more socially legitimate here; we suspect elements of both factors were at play. Certainly, some survey data suggest that voters in Central and East Java—the homelands of the ethnic Javanese—are more likely to view vote buying as being an acceptable practice (Muhtadi 2015). On the other hand, we know from the wider research project that the broad contours of the vote-buying system we describe in this article can be found in most locations throughout Indonesia. In other parts of the country, too, candidates construct tiered vote brokerage networks that identify voters and distribute cash payments to them in ways similar to those described in this article, and they use similar terms to describe their practices. Even so, we must be careful about generalizing our findings beyond rural Java. It is likely that both the intensity of candidates’ vote buying efforts, and the degree of overlap between them, are less in most parts of the country.
As for why vote buying might be especially prevalent in rural Java, we can speculate on several possible explanations, though we cannot pronounce definitively on them here. An historical institutionalist argument would point to the long history of vote buying in Javanese village head elections during the New Order regime (1966–1998) and to how this provided a repertoire of techniques that could readily be scaled up for higher elections after democratization (this is a topic two of us have examined separately: Aspinall and Rohman forthcoming). Complementary explanations would point to social-structural factors that make rural Java distinctive. In urban parts of Indonesia, the dense social networks and widespread poverty that are generally seen as conducive to vote buying are less prevalent (in our wider research project we found least evidence for individualized vote buying in Jakarta: Dewi, Harjanto and Purba 2016). In rural areas in many (though not all) other parts of Indonesia, social networks are less egalitarian than in Java, with the result that politicians often target formal and informal community leaders—village heads, clan chiefs, adat (customary law) leaders and the like—with patronage, rather than distributing payments to individual voters (the most extreme example is highland Papua where clan chiefs cast votes on behalf of their communities through a system of collective voting, often in exchange for bribes: Nolan 2016).
With regard to the institutional context, as noted above, Indonesia has adopted an electoral system that has generated a system of complex multi-partyism and that motivates candidates to campaign for personal votes. Competing in the April 9, 2014 election were candidates for seats in the national People's Representative Council (Dewan Perwakilan Rakyat, DPR) and for Regional People's Representative Councils (Dewan Perwakilan Rakyat Daerah, DPRD) at provincial and district levels.
Candidates nominated by 12 registered national parties competed in multi-member constituencies; each constituency had between three and ten seats at the national level and three and 12 in districts and provinces. Under Indonesia's open-list PR electoral system, voters could choose between voting for a party or voting for an individual candidate on a party list. In each electoral district, the number of seats a party won was determined by adding the total number of votes for the party and for its individual candidates. For each party winning one or more seats, those seats were allocated to the individuals with the highest personal votes. As we shall see, this system generated strong incentives for candidates to build personal campaigns, and greatly affected the nature of vote buying.
The two electoral constituencies (daerah pemilihan or dapil) that are our focus in this article are Central Java III, consisting of the four rural districts (kabupaten) of Grobogan, Pati, Blora, and Rembang, and East Java VIII, consisting of the districts of Jombang, Nganjuk, Madiun, Mojokerto, and the urban districts (kota) of Mojokerto and Madiun (see map). The former constituency had about 3.3 million registered voters, the latter 3.5 million. As with other constituencies throughout Java, the boundaries of the national and provincial electoral constituencies in both locations coincided. Constituencies at the district level were much smaller, incorporating from 30,000 to 300,000 voters. Thus, candidates running for seats in the national and provincial legislatures had to win tens or even hundreds of thousands of personal votes; candidates running for district seats could win with between 3,000 and 10,000 personal votes, and sometimes even fewer.
Both constituencies are located in Java's rural rice-growing heartland, though they also contained numerous small and medium-sized towns, as well as upland, timber, and horticultural regions, and other scattered industries. In sociocultural terms, both constituencies have mixed populations, largely consisting of abangan or non-devout Muslims who adhere to a heterodox or syncretist form of the Islamic faith, as well as more pious, or santri, Muslims affiliated with the mass-based Islamic institution Nahdlatul Ulama (NU). Historically, the abangan have mostly supported nationalist-pluralist political parties, most recently the PDI-P (Indonesia Democracy Party-Struggle), while the traditionalist santri have supported Islamic parties, especially those affiliated to NU, which in 2014 meant the PKB (National Awakening Party). These divisions have started to break down over recent electoral cycles as new “catch-all” parties have arisen (in both constituencies candidates with backgrounds in NU and PDI-P ran for multiple parties) and as clientelist strategies have allowed candidates to reach across community lines (Rubaidi 2016). Moreover, we did not find significant differences in vote-buying strategies between santri and abangan candidates, except that candidates from Islamic parties were more likely to rely upon Islamic networks in constructing their brokerage networks. They were neither more nor less likely to rely upon vote buying than other candidates. Indeed, we should stress that in these two electoral districts, candidates from all parties engaged in vote buying, using patterns that were fundamentally similar.
During one month leading to the April 9 election, we carried out fieldwork in these constituencies, between us interviewing 70 candidates running at all levels and from all parties, as well as more than 100 campaign workers. We also made return visits of varying durations to the research locations over the subsequent three months. Where possible, we selected candidates who were judged by local actors to have a strong chance of winning their seats. Most of the candidates we interviewed were running for seats in district parliaments, with about 20 of the total being provincial and national candidates. During the period leading to the election, we were able to observe campaign events, and frequently accompanied candidates and brokers as they interacted with voters and, sometimes, as they planned or engaged in vote buying. In order to protect the anonymity of our informants, in this article we do not use candidates’ names and have obscured other potentially identifying material. We should also emphasize that our ethnographic methods impose limits on our findings: for instance, because we do not rely on surveys of voters we cannot here be definitive about the accuracy of candidates’ assessments of the importance of cash payments for voters.
We should also emphasize that the vote buying we discuss in this article is part of a broader spectrum of patronage politics associated with elections in Indonesia, which is elaborated upon in other publications arising out of the broader project (see Aspinall and Sukmajati 2016). For example, many of the politicians we interviewed also distributed gifts that conferred a collective benefit (“club goods”) when campaigning: paying for repairs to community infrastructure (e.g., a village road), or donating goods to social organizations (e.g., sports equipment to a youth club). This article does not focus on this broader universe of gift-giving but on a specific practice: the “dawn attack,” defined as the distribution of cash (and sometimes, goods) in the few days leading to an election, or on ballot day itself, to a targeted list of recipients who have previously been identified by brokers. Accordingly, we adopt an understanding of vote buying at the narrow end of the range of meanings used in the wider literature (Nichter 2014). The dawn attack is the least socially, legally, and politically acceptable patronage practice at election time in Indonesia. In many places, candidates who openly admit to other forms of gifting will say that they abjure the “dawn attack.” In our two constituencies it was all but ubiquitous, however.
Vote brokerage networks
As noted above, much recent literature on electoral clientelism and vote buying assumes the key role is played by party machines. In Indonesia, the electoral system undermines the role of parties in grassroots election campaigning. The emphasis on the personal vote under open-list PR means that candidates typically see rival candidates from their own party as their main competitors, because in most cases only one or two individuals from each party will win a seat in any given electoral district. Accordingly, candidates devote their energy to promoting their individual prospects, producing a pattern of “ground war” campaigns in which candidates directly interact with voters and try to build personal relationships with them (Aspinall 2014a). Rather than well-informed party machines running vote-buying efforts and carefully deciding where to invest resources on the basis of past experience and careful analysis, we encountered a political world made up of individual candidates designing campaigns on the basis of intimate and informal knowledge about local communities and, sometimes, poor information and unrealistic expectations.
In other countries, ethnographic studies have described both party-based and personalized campaign structures that candidates use when vote buying.
These structures typically involve large numbers of community-level vote brokers. Such brokers are critical for candidates who need to gain detailed knowledge about voters’ political preferences and material needs. They are also useful for increasing the effectiveness of vote buying, given that recipients will usually feel greater obligation to deliver a vote when they receive a payment or gift from someone with whom they already enjoy intimate social relations.
In our two constituencies, vote brokerage organizations were generally known as “success teams” (tim sukses), as in other parts of Indonesia. Their structure varied, but most shared a pyramidal shape that had at its apex the candidate and his or her core team of advisers, and proceeded through a series of regional coordinators. For national or provincial candidates, these typically consisted of district coordinators, sub-district coordinators, and village coordinators. Finally, at the base were ground-level brokers who were charged with recruiting and influencing voters. The base-level brokers were known by a variety of terms. The most common were kader (or cadre) in parts of East Java VIII, and korlap (koordinator lapangan or field coordinator) and sabet or gapit (Javanese terms meaning “whip”) in Central Java III. A single candidate could recruit between a handful and more than 100 brokers in a single village; the total number of brokers could vary from under a dozen in a team (for poorly resourced district candidates) up to over 3,000 for the best-resourced DPR candidates.
Brokers were distributed with great density through the community. To take one example, we collected broker and voter lists from one candidate who was running for a seat in the DPRD of Blora district. In some villages where he placed brokers, he had one broker for more than 100 registered voters. But in villages which he considered to be part of his base, he had one broker for every five voters. In all the villages where we have data for this candidate, there was a total of 122,000 registered voters. He had 490 brokers scattered through these villages, a ratio of one broker to 249 voters. There were 83 district-level DPRD candidates running in this constituency. Assuming that they all had the same ratio of brokers to voters across the constituency, that means that potentially a third of the voters would have been brokers. If we further assume that the DPR and provincial DPRD candidates had similarly large teams, then the number of brokers and voters in the electoral district would have been evenly matched: every voter would have been a broker. Of course, this was not literally true: this candidate was relatively well organized, and he was one of only 12 candidates who won a seat in this constituency, so we can assume that his ratio of brokers to voters was higher than the average. Moreover, many national and provincial candidates did not establish success teams of their own but simply borrowed brokers from district candidates with whom they arranged personal deals. Even so, the example helps illustrate the enormous competition for grassroots brokers.
The initial function of the brokers was to identify persons willing to vote for the candidate. Typically, brokers were instructed to recruit people in their immediate household, and their close neighbors, friends, and relatives in the village. In almost every case we encountered, the brokers were required to draw up lists, known by the English phrase, “by name, by address.” These lists contained the names and addresses of individual voters and, sometimes, their signatures and additional information, such as mobile telephone and identity card numbers. These lists would be handed upward through the chain of coordinators to the candidate. The lists could be huge. One incumbent running for a DPR seat in Central Java had a list that contained 320,000 names. Serious district-level candidates, who needed to secure a much smaller total vote, typically compiled lists of between 10,000 and 30,000 names. The number of voters each broker “recruited” varied from a low of about five or six up to 30, or even a hundred or more in some cases. Experienced candidates mostly said they were skeptical when a broker furnished a list with over 30 people, doubting it was possible to be sure of the voting intentions of so many persons.
Candidates then used these lists for distribution of payments, generally within the final few days leading to the vote. Typically, the broker who had entered the person's name on the list would deliver the cash, almost always in an envelope, typically with the candidate's campaign card pinned to it. Some candidates, suspecting that their brokers would steal all or part of this money, ensured that core team members were present when the money was handed out. By the time the payment was made, voters typically knew it was coming and little had to be said about it. If a broker did mention the payment when it was handed over, he or she would typically describe it as pocket money (sangu), or as a small gift with which to buy a snack or a drink (uang es—ice money, was one widely used expression), and ask the voter to “remember” the candidate at the ballot box. The standard conceit, meanwhile, was that candidates paid voters an amount equivalent to a day's income, compensating them for work they lost by going to cast their ballot. Many candidates described their payments in these terms, as a sign of their gratitude to voters for putting themselves through the inconvenience of voting. In fact, payments often exceeded a standard day's agricultural wage, and went to all voters, not only those in paid work.
Candidates recruited brokers through a variety of networks. Party organizations were certainly one source of brokers, especially for the lead candidates on party lists who were often chairpersons or secretaries of the local party branch. Such persons often dominated the local party structure, drawing heavily on it for their brokers, and effectively subsuming the local party branch structure into their personal team. In other cases, several competing candidates would raid the party structure for brokers, drawing individual party leaders and cadres as well as their clients into their personal teams. Most candidates—in part because the party structure was dominated by a small number of candidates—looked beyond the party for brokers. They relied first and foremost on relatives, friends, and business associates, but candidates also drew upon almost any social network in which they or core supporters had influence. The range of such networks used by candidates is so vast as to defy easy categorization (to appreciate fully, see chapters in Aspinall and Sukmajati 2016), but among those most used were religious networks and organizations, such as Islamic study groups, as well as sports clubs, youth organizations, women's groups, business networks, and village government. Many candidates enjoyed strong links with such networks, and had made long-term investments in them, acting as patrons for low-level leaders and financing their activities. Incumbent legislators had likewise typically courted long-term support by directing pork-barrel projects and “social assistance” funds from the government toward target communities or organizations (Mahsun 2016). All candidates believed the most reliable success teams were founded on the basis of such long-term clientelistic ties, rather than being hurriedly thrown together in the weeks leading to an election. However, candidates varied greatly in their capacity to build longer-lasting networks. Those who were most advantaged were incumbents, who had access to state patronage resources, as well as persons who occupied nodal points in networks upon which persons depended for economic opportunities and livelihood support—the wealthy trader who had a network of suppliers or distributors, for instance, or the leader of a branch of an Islamic women's organization who had a say in determining which members would receive livelihood projects.
Most candidates lacked direct personal connections with a sufficient number of individuals in a sufficient number of villages to be able to form a success team entirely out of personal acquaintances. Accordingly, they usually relied on their sub-district coordinators to recruit village coordinators, and the village coordinators to recruit base-level brokers. Often, recruitment occurred within the networks described above, but would quickly stray once mid-level brokers started recruiting their own personal friends, relatives, business associates, and so on. As a result, a highly attenuated chain of connections often linked the candidate to the base-level brokers. Candidates sometimes used language that was uncannily reminiscent of the social science literature on clientelism to describe the results. One female Golkar member of the Pati district parliament explained:
Not all of my coordinators were recruited through primary ties; a lot were through secondary ties … Primary ties are [with] people I've helped in the past. But secondary ties are when, let's say, I've helped Pak Slamet, and then I ask him to look for people for me. My hold [over them] is weaker.
Motivations for participating in a success team thus varied. Some brokers had long-standing relationships with, or clientelistic dependence on, a candidate, and were prepared to make personal sacrifices to ensure his or her victory. Others were motivated simply by material rewards, either the modest payments they could expect for their work or the sums they could pilfer from the envelopes they were supposed to distribute. Such ‘opportunist brokers’ (Aspinall 2014b; see also Stokes et al. 2013) were a major concern of candidates.
The sheer number of candidates competing meant that many candidates found it difficult to find village-level brokers with the skills, experience, or social influence necessary to deliver votes effectively. This competition created a market for brokers who could readily exploit the credulity and desperation of candidates. But it also meant that candidate brokerage networks were often interlocking, indeed overlapping, with some brokers even working for more than one candidate simultaneously (Triantini 2016). If some of the literature on party-based clientelism assumes rather neat zoning, “where each machine controls distinct geographic territories” (Gans-Morse, Mazzuca and Nichter 2014, 417), the situation in rural Java was more of a free-for-all, where the machines not only of multiple rival parties, but also of rival candidates within the same party, often competed against each other in the same neighborhoods, streets, and households.
Who to target?
Many candidates used terminology superficially suggestive of turnout buying (Nichter 2008, Larreguy, Marshall and Querebín 2016). When asked about their targeting strategies, most candidates replied that they prioritized voters they considered to be their basis (another Indonesianized English language term frequently used in this context), believing this would yield greater return on their efforts. As one candidate put it: “Some people think you should throw your net into the sea, but I'm not like that. I throw my net into the fish pond where we know there will be fish.”
Candidates differentiated such base voters from uncommitted or undecided voters, using a variety of terms for this second group, the most common of which was ‘massa mengambang,’ or floating mass (a term that was first coined in the 1970s to describe the depoliticized population, delinked from party identities, which was the goal of the authoritarian New Order regime's policies of rural depoliticization).
Partisan affinity was not irrelevant to these designations. Some parties had blocs of support in particular areas that could be traced back over several electoral cycles. PDI-P, in particular, identified ‘red’ (the party's color) villages where it had “traditional voters.” Some PDI-P candidates targeted their campaigns almost exclusively in these villages. Socially, too, there were patterns; for instance, PDI-P candidates mostly distributed cash in abangan villages, while PKB candidates mostly targeted santri villages or neighborhoods. Remembering also that party leaders tended to use party structures for their campaigns, it is not surprising that people more closely linked to parties were more likely to be targeted by vote buying. Accordingly, Burhanuddin Muhtadi (2015) has shown that national survey respondents who stated they felt ‘very close’ to a party were more than twice as likely to be offered gifts than non-partisan respondents (the figures were 48 to 22 percent). Overall, however, the proportion of partisan voters in the electorate has declined significantly since the early post-Suharto period, such that in the lead up to the 2014 legislative election only 15 percent of survey respondents nationally said they felt somewhat, moderately or strongly attached to any party, while 85 percent were not aligned (Muhtadi 2015). The result was that most vote-buying efforts targeted these non-party aligned voters. Moreover, in our Java case study areas, no individual candidate who was focusing his or her campaign within what was seen as a party base area could be confident that those voters were not being targeted by rivals: candidates from the same party list after all shared the same partisan voters, while candidates from rival parties could always target voters in any such village by accessing them through non-party social networks of the sort described above.
Overall, when candidates used terms suggesting core versus swing voters, they thus typically understood such terms in personal terms, rather than in terms of voters’ identification with a party or program. They were talking about a clientele rather than a party core. The word “base” typically referred to one of four circumstances. First, most candidates viewed their home village (and, if they were running for a national or provincial seat, their home sub-district and district), along with other areas where they had strong personal connections (e.g., a spouse's home village) as their core area. Second, candidates viewed as their bases places where they had a history of providing personal patronage. This mostly meant incumbents targeting villages where they had delivered pork-barrel projects, but it could also apply to wealthy candidates who had a reputation for charitable giving in a particular area. Third, candidates could think of core voters in network rather than geographic terms. For example, a candidate who was closely associated with, say, pencak silat martial arts groups might view the groups’ members as his or her core constituency. Fourth, each and any of these categories could be extended by way of the clientelistic ties that candidates used to construct their success teams. For instance, a candidate who had recruited a leader of a group of fish-farmers’ cooperatives in several villages into the success team might view these villages as part of his or her base.
In sum, a long and tightly stretched chain of links might tie the candidate to his or her allegedly ‘loyal’ voters. Though candidates might claim that voters in a particular locale or network were particularly loyal, the reality was that each site was vulnerable to penetration by rival social networks and, hence, raiding by rival vote buyers. The obvious example was home village or region: candidates typically defined their place of origin as a ‘base,’ but there were usually multiple candidates from the same region doing the same, each drawing upon his or her local networks to enlist voters and provide them with payments.
No matter how much candidates stressed the personal loyalty of such voters, they almost never believed that they could rely on their votes without making cash payments to them. Most candidates explained that payments had become part of the local culture (sudah membudaya) and were thus unavoidable. They felt trapped in the vote-buying exchange, supporting analyses which see clientelist practices as, at least in part, demand driven (Kitschelt and Wilkinson 2007, 24–26; Shin 2015). Thus, one Golkar incumbent in Madiun was assiduous about delivering infrastructure, education, and agricultural assistance programs to certain villages in his constituency and then targeting recipient villages with his re-election campaign efforts. Even so, he had to distribute cash in these villages: “If not, I'll be crushed.”
Incumbents like this man frequently stressed the ingratitude and unreliability of voters who had benefited from their previous private gifts, pork-barreling or constituency service, often believing that voters were ready to sell their votes to competitors who offered higher payments. After the elections, too, candidates and brokers had many stories explaining poor results in particular villages by reference to the voters’ greed and their readiness to be swayed by cash payments. Typically, these stories stressed outbidding by more cashed up rivals and the ingratitude of voters with regard to the losing candidate's prior service or personal connections.
Even candidates and brokers who were less bitter rarely took their core voters’ loyalty for granted. Indeed, we found that about a dozen of the candidates we interviewed made slightly higher payments to their core voters than to others. Thus, one female Golkar candidate in Pati constructed her voter list largely by drawing on female members of Islamic prayer groups, viewing them as personal loyalists who knew her through the religious activities she regularly attended and sponsored. As a result, she viewed these payments as merely an “encouragement” (penguat—lit. “strengthener”) to these voters. Even so, she paid those closest to her 20,000 rupiah (approx. US$1.70) each, while giving other voters 15,000 rupiah. As she put it, “we should think first of those who are close to us.”
More common was devotion of greater organizational resources to core areas. Thus, candidates for district seats would sometimes have over 100 success team members in their home village, but no more than a dozen in any other location. Though candidates often talked in affective terms about repaying the loyalty and “closeness” of these voters, they equally frequently referred to the strategic logic of shoring up a core base that might otherwise be vulnerable to defection. Overall, they spoke in a language that betrayed great anxiety about, even suspicion of, the allegiance of their supposed loyalists.
Many candidates thus effectively hedged their bets, focusing first on a core group of voters, but extending their efforts to voters whose personal connections to them were more tenuous, trusting that their brokers would be able to mobilize their own affective ties with these voters, or that cash payments alone would be enough to secure their votes. Appropriately, candidates sometimes used the word ‘gambling’ (another English loanword) to refer to such efforts to distribute cash into areas where they lacked a personal base.
More generally, candidates varied considerably in their assessments of how important cash payments were in motivating voter choices. A few candidates were very cynical and believed that, in their constituency, voters were basically “transactional” (transaksional). As one candidate explained: “Yeah, it's all about money here, brother, in the town of Mojokerto; it's 97 percent money politics and only three percent who vote with real awareness.”
But this was not a universally shared view. Another candidate running in the same constituency thought very differently:
Although many candidates use money politics, when it comes to making a final decision the voters still assess the figur [personality, character and reputation] of the candidate. If there is a group of voters to whom three candidates distribute money, it's the candidate who is closest to them who will be elected, even though s/he may have given smaller sums than the others.
Most candidates came somewhere between these poles and made varying estimates of the share of voters who were influenced “only” by cash payments versus those influenced by other factors, such as the candidate's personality or record of community service: some put the ratio at 60:40; others estimated it to be 80:20, and so on. One PDI-P candidate in Madiun believed that candidates who made low payments (10,000 rupiah) were bound to lose; medium payments (20,000 rupiah) were sufficient to bind the votes of loyalists and ensure they came to the polling booth; large payments (50,000) would be enough even to raid the voter bases of competitors and get people to shift their choices.
How much to pay?
As part of our research, we gathered extensive information on the cash payments several dozen candidates made to voters, in terms of both the numbers of voters they targeted and the nominal sums they distributed. Our findings on how candidates set the price of a vote helps clarify the market logic governing vote buying.
Scholars often argue that, as constituency size increases, vote buying becomes more difficult and, therefore, less likely. As Hicken explains, “[a]ll else being equal, the larger the district, the larger the amount of money required to buy a winning number of votes and the weaker or less dense the social networks needed to distribute money and monitor compliance.”
In these Java constituencies, the number of voters was typically 20 times greater in the national/provincial constituencies than in the district constituencies. We did not find that the likelihood of a candidate engaging in vote buying decreased as one moved up in scale, but what did change was the scale and magnitude of the payments.
Typically, DPR and provincial candidates had to win many times the number of individual votes to ensure election than did district candidates. Depending on the size of the electoral constituency, district candidates could often win a seat with just three or four thousand votes; national or provincial candidates needed at least 15 or 20 times that number. Accordingly, district candidates rarely distributed payments to more than 30,000 voters and usually to far fewer. National and provincial candidates frequently distributed to one or two hundred thousand persons—or even more. But district candidates consistently paid a higher rate per head than candidates who were running in the larger provincial and national constituencies. Most competitive district candidates in Central Java III paid 15,000 to 30,000 rupiah (i.e., US$1.30 to US$2.60) per voter, though the figure sometimes reached 50,000 rupiah; while serious provincial/national candidates paid on average 10,000 rupiah. In East Java VIII, the figures were 20,000 to 30,000 rupiah at the district level, compared to 5,000 to 10,000 rupiah for provincial or national candidates.
Because the higher-level candidates were distributing envelopes in greater numbers, they still ended up spending several times more on their vote-buying efforts. Most successful candidates at the district level spent between 400 million and 700 million rupiah (i.e., approximately US$35,000–60,000) on vote buying; most serious DPR candidates spent into the billions (in other words, in excess of US$90,000 and sometimes up to $500,000). Moreover, the source of the payments was often obscured in constituents’ minds by the fact that candidates at the different levels often cooperated in arrangements known as “tandems,” under which they would share costs and distribute a single envelope to each voter with a single payment jointly funded by three candidates.
Although we have so far largely talked in terms of average payments, we should stress that, in practice, payments varied significantly: in a single village as voting day approached, multiple envelopes from different candidates might circulate containing 10,000, 15,000, 20,000, 25,000, 30,000, and up to 50,000 rupiah (i.e., between US$0.90 and US$4.50). Some candidates liked to make a one-off payment; others liked to make it in two tranches (a “DP” or down payment when someone was signed up, and a second payment close to voting day). In the district of Pati, one candidate—a major rice wholesaler—was famous for distributing packets of rice alongside cash.
A major determinant of such variation was simply the candidate's wealth and ambition. A few candidates were well resourced and ready to do whatever it took to outbid competitors; rivals spoke with awe about these candidates’ spending power. Others, standing simply to fill out party lists, did not harbor hopes of victory. Some such weak candidates did not engage in vote buying at all. Others made token payments (say, 10,000 rupiah per voter when the going rate was 30,000) for reasons of personal pride: such candidates explained to us that they would find it shameful to seek support from voters without distributing money, even though they knew they had little chance of winning.
Despite this variation, and the obvious cultural pressures to participate in vote buying, we should stress that an underlying market logic was operating. Candidates spoke as if they were in a sort of marketplace, using language replete with trading and economic metaphors when describing their strategies. They assiduously monitored what their competitors were paying and tried to adjust their prices if they could afford to do so. As one Gerindra candidate in Madiun put it, “if my competitor puts out 20,000, I have to go to 25,000.”
They often talked explicitly about the “market price” (harga pasar) of a vote in a particular district, electoral constituency, or village. Some varied payments slightly between locations; others feared that if they did so they would alienate voters in the areas where payments were lower. They were also aware of how a particularly wealthy buyer could affect average prices: in one constituency in Rembang, a few candidates who were particularly wealthy “wrecked the market,” in the words of one candidate, and pushed up the price to 50,000 rupiah per vote (almost double that of neighboring constituencies); in another Rembang constituency, the main contestants got together and reached an agreement to cap the price at 30,000 rupiah.
Market panic set in for many candidates in the final days leading to the vote when they received reports from brokers about the extent of vote buying by rivals, and the sums being paid. Brokers warned candidates that they were being trumped and pleaded with them to release additional payments. Many did so, raising their prices by 5,000 to 15,000 rupiah (or, in rare cases, even more). This could be problematic logistically, since it was no mean feat to collect sufficient banknotes and divide them up in time. Other candidates lacked sufficient funds to increase their payments and instead allowed their brokers to adjust payments, for instance, taking envelopes with 15,000 rupiah each and combining them to make payments of 30,000 rupiah to only half the people on their lists. Some candidates rejected last-minute requests for additional payments from their brokers because they suspected their brokers were trying to cheat them, but later came to regret their parsimony. In a few cases, we learned of village brokers returning envelopes to candidates because they believed the payments were too small and the money would be wasted. We even heard reports of wealthy brokers investing their own money to increase the sums they distributed (in such cases, the brokers were typically tied to candidates by strong clientelistic ties, such as being distributors in a candidate's fertilizer business, or being alumni from his religious school). It is also likely that many brokers who believed the payments they were distributing were too small simply pocketed the cash. The breadth of this last-minute panic suggests that, no matter how much they claimed to be targeting “loyalists,” many candidates were far from confident that voters’ loyalty would equip them to resist higher payments paid by rivals. Candidates behaved as if they were in a competitive marketplace.
How effective was vote buying?
It should already be obvious already that distributing cash for votes in these Java electorates—as in other settings—was a risky and uncertain process. We met numerous candidates prior to the election who were convinced that their calculations were foolproof and their systems infallible; not a few of them failed to win seats. Most experienced a “margin error” far exceeding what they had expected. This term (yet another borrowing from English) referred to the gap between the number of voters to whom a candidate distributed payments and the total vote he or she received on ballot day. All candidates expected that many of their payments would go to waste, and they responded by distributing a number that exceeded the vote total they believed they needed. Candidates who failed to be elected later admitted they had miscalculated in this process, underestimating either voter turnout or the margin they needed to build into their vote-buying effort.
After the election, it was possible to collect data that showed that the range of wastage was very wide—from a low of about 35 percent to around 85 percent. One of the most successful candidates we interviewed was a provincial candidate in East Java who distributed 109,000 envelopes and obtained about 67,000 votes. But we also met candidates running in district elections who distributed 20,000 payments and received fewer than 4,000 votes.
In explaining such discrepancies, one of the first issues to consider is to what extent candidates’ varying rates of success were a function of the magnitude of the payments they made to voters. Other literature on vote buying rarely addresses this question directly, because the illegality of vote buying makes candidates reluctant to disclose details. We were mostly able to overcome these problems and collected much relevant detail during the course of our pre-election research, but our material was patchy. To address the issue more systematically, one of us (Noor Rohman) returned to the field and interviewed nine district-level candidates who won seats and ten who came close to winning seats in two neighboring electoral constituencies in one of the Central Java districts (we excluded candidates who won insignificant vote totals because such individuals often did not engage in vote buying at all). The candidates were interviewed on their electoral strategies, especially the payments they made to voters (a twentieth candidate provided incomplete data and was excluded from the analysis).
To summarize our findings, these 19 candidates distributed cash payments ranging from 10,000 to 30,000 rupiah to voters, though two candidates also distributed rice in addition to cash, making the value of their gifts the greatest.
Eight of the 19 candidates made payments of 20,000 rupiah per voter, reflecting the tendency of such payments to clump around the mean, referred to above. Meanwhile, the reach of their efforts varied considerably: the candidate with the largest vote-buying network distributed 32,000 envelopes containing cash; the candidate with the smallest such network distributed a mere 5,000 envelopes. Using these data, we calculated these candidates’ total expenditure on vote buying. They ranged from a maximum of 1.2 billion rupiah (around US$100,000) to a mere fifty million rupiah (US$4300), with an average expenditure of 420 million rupiah (US$ 36,000) (these figures, we should stress, excluded spending on other campaign items, such as payments to brokers and club goods).
Using these data it is possible to demonstrate the centrality of vote buying to electoral success in these constituencies. Our first three charts plot different aspects of the vote buying efforts of these candidates against the personal vote totals they obtained. These vote totals ranged from 970 to 16,800 votes, with all but one of the candidates who received over 4,000 votes being elected. The first three charts all show a consistent story. Figure 1 shows a positive correlation between the magnitude of the individual payments made to voters and candidates’ vote totals. The more candidates paid, the higher their votes. On average the nine candidates who won seats made higher individual payments to voters (23,700 rupiah) than the ten who lost (17,250 rupiah). Figure 2 shows a similarly positive relationship between the total number of payments candidates made and their vote totals. The greater the extent of these candidates’ vote buying efforts, the more votes they achieved. Candidates who won seats distributed payments to an average of 24,700 voters; those who missed out distributed to, on average, 14,130 voters. Figure 3 simply combines these two factors, demonstrating a strong relationship between candidates’ total expenditure on vote buying (i.e. the sums paid to individuals multiplied by the number of payments made) and their personal votes. Candidates who won seats expended on average 612 million rupiah on vote buying; those who lost spent on average 253 million rupiah. Taken together, these three charts demonstrate the central importance of vote buying, and of material wealth, to candidate success. Even if they do not show a causal relationship, strictly speaking, they demonstrate that candidates from poor backgrounds who lacked the family or personal resources, business or bureaucratic networks with which to mobilize funds for vote buying were not elected to parliamentary seats in these electoral districts.
Figure 1 Magnitude of payments
Figure 2 Number of payments
Figure 3 Total expenditure
The most revealing chart, however, is Figure 4. We calculated what candidates called their margin error by expressing the gap between the number of voters to whom they distributed envelopes and the personal vote they recorded on polling day as a percentage.
For example, the two candidates represented by the points at the far left of the scatterplot each made the same nominal payments to voters (10,000 rupiah per head). But they experienced different margin errors: one distributed 10,000 envelopes but received a little more than 1,400 votes, making a margin error of 85.7 percent, while the other distributed 5,000 envelopes and received just under 1,000 votes, making a margin error of 80.5 percent. These rates of wasted payments varied widely, from a high of 85.7 percent to a low of 47.3 percent, with an average of 72.7 percent. In Figure 4, we plot candidates’ margin error rates against the value of the payments they made to each voter. In one respect, the result confirms the findings of the previous charts, showing that there was a correlation between payment size and margin error. In general, the higher the payments, the lower the margin error. The nine candidates who won seats recorded an average margin error of 58.2 percent, as opposed to the 76.9 percent for those who did not gain seats.
However, the relationship between payment magnitude and margin error was far from being straightforward. For example, the candidate who made the highest payment was not the most successful by this measure: this candidate won fewer than 8,000 votes after distributing 32,000 envelopes, making a margin error of just over 75 percent—close to the average. He still won comfortably, but he expended significantly more resources compared to several candidates who distributed far fewer payments, at much lower rates, and yet still managed to be elected. This candidate spent over 1 billion rupiah on vote buying; some winning candidates spent just 400 million rupiah. Similarly, the chart also draws attention to the greatly varying effectiveness of the candidates who distributed the standard payment of 20,000 rupiah in these constituencies; the margin error of the least successful of these candidates was 82.4 percent compared to 61.6 percent for the most successful.
Such variation lends support to what might be called an “entry ticket” view of vote buying, which suggests that payments to voters may be akin to the price of entry paid by candidates, who are then assessed by voters on other grounds.
Indeed, as already noted, many candidates felt that though they were obliged to make payments to voters, these payments provided little guarantee of success in themselves. Taken with what we know about the backgrounds of the various candidates, they also support a personalized version of the turnout-buying argument, underlining the relative strength of candidates who had strong networks of personal loyalists to mobilize the vote. It is striking that the two candidates on our chart whose margin errors were lowest—who also, it will be noted, distributed sums of very different magnitudes—were both very influential local political patrons, with district-wide reputations and access to especially strong patronage networks. Their success teams were based, respectively, on a strong local business network and on NU. Behind almost every one of the points that lie above the trend line in Figure 4 is a similar story of a candidate possessing particularly strong networks, often centered around his or her home village, and whose vote-buying effort was backed by other strategies involving cultivation of a base area through long-term distribution of club goods and projects.
But despite the variation, the overall positive correlation between payment magnitude and success rates does suggest as one possibility that, at least for some voters, the size of the payment did matter, implying that just as different candidates have different strategies, different voters had different preferences. In this interpretation, a majority might support a particular candidate independently of the payments they receive. For them, payment acts primarily as a means of confirming that choice or ensuring that they make their way to the polling booth on voting day. A smaller number were more sensitive to the size of the payments. This interpretation matches what most candidates believed—as already explained, many thought a proportion of votes were simply up for sale—but we should stress that it is not the only possible explanation for the pattern revealed in Figure 4.
Our study of two electoral constituencies in Java during Indonesia's 2014 general election reveals a pattern of vote buying that was highly fragmented, even atomized. To be sure, there was order within the chaos, in the presence of highly structured and systematic brokerage networks, and in the relatively open way in which payments were discussed and delivered at the community level. But these elements of organization and legitimacy were embedded in a context of electoral races that were highly competitive and unpredictable, with multiple candidates from multiple parties competing against each other to recruit brokers, identify and defend their voter bases, keep their brokers honest, and adjust to the strategies and payments of rivals.
In part, of course, the uncertainty that was such a feature of these campaigns is typical of vote buying everywhere, given that in most circumstances “vote buyers have no guarantees that voters who accept their material offers shall dutifully reciprocate on election day” (Schaffer and Schedler 2008, 19).We have demonstrated that this aspect of uncertainty was evident in our two electoral constituencies, with the majority of candidates gaining a vote total that was much lower than the number of individual payments they distributed.
Candidates responded to this uncertainty in ways familiar from other country settings. Above all, in order to try to induce voter compliance with the vote-buying exchange, candidates in these Javanese constituencies did what Wang and Kurzman (2008, 69) observed of the Kuomintang party in Taiwan: they drew upon pre-existing “strong social ties to build a sufficient network of brokers for vote buying.” One of the most robust findings of research on vote buying virtually everywhere is that would-be vote buyers lean heavily upon existing social networks and on the authority of locally influential brokers to make their efforts effective (e.g. Schaffer and Schedler 2008, 22); the Indonesian experience confirms this general picture.
Similarities between our observations and findings elsewhere also extended to technical aspects of vote buying. For example, candidates in Java relied upon written lists of voters, drawn up by their brokers, in order to distribute cash (as in, for example, Taiwan and Thailand: Wang and Kurzman 2008, 70; Chattharakul 2010, 84–86). They also mostly delayed the distribution of cash payments to very late in the election cycle, trying to ensure that their gifts remained fresh in the voters’ minds (while in Indonesia the term “dawn attack” is widely used, in Thailand the night before elections is often called “the night of the barking dogs”—because brokers supposedly disturb the local dogs as they move through villages (Fuller 2007); in the Solomon Islands it is known as “devil's night”: Wood 2014b, 12). Even the targeting of supposedly “loyal” voters was reminiscent of the so-called “turnout buying” observed elsewhere, with our informants suggesting that, for a significant proportion of voters, payments were simply an added encouragement to cast a vote for a candidate they already preferred. Overall, one of the fascinating features of vote buying is the striking similarity of the techniques and strategies used to prosecute it across diverse country settings, despite it being a phenomenon where we can be reasonably sure that little or no transnational learning takes place, given the illegality and covertness of the practice. Instead, parties and candidates adopt similar strategies because they are responding to similar political incentives and social structures.
In conclusion, however, we would again emphasize the notably fragmented and free-wheeling pattern of vote buying in Indonesia, and the institutional origins of this pattern. With regard to the basic form, it should be stressed that the similarities to “turnout buying” were in some ways superficial: as a result of the largely personalized and clientelistic—rather than partisan—nature of the loyalties concerned, many candidates greatly over-estimated their individual pull, discovering on polling day that many of their supposedly loyal voters had deserted them. Indeed, candidates’ behavior—especially the way that many tried frantically to adjust their payments upwards when they realized that their competitors might outbid them—suggested that candidates believed that even their putatively loyal voters, or at least a significant proportion of them, were price-sensitive rather than being true loyalists. With regard to the institutional origins of this pattern of free-wheeling clientelism, Indonesia's adoption of an open-list electoral system was especially critical. This system has encouraged candidates from even the same parties to compete against each other and to emphasize their personal appeal, providing incentives to vote buying and adding greatly to the complexity and overlapping nature of vote-buying efforts. The open-list system has also contributed to the weakening of Indonesian parties, and helped give rise to a situation in which only a small proportion of the population positively identify with any party (though other factors have also contributed to this outcome). Indeed, many of the candidates we encountered complained strongly about the open list system, looking back nostalgically to the days of closed-list proportional representation when candidates were elected according to their place on a party list rather than their personal vote totals, and when there was much less individualized vote buying, and less pressure on candidates to fund and organize their own campaigns.
These observations in turn point toward the variety of vote-buying systems. One implication of our analysis is that there is a critical distinction between systems, such as that found in Indonesia, which are relatively free-wheeling and candidate-centered, and where vote buying is organized by ‘social network machines,’ and systems where relatively well institutionalized political parties play the lead role. Descriptions of vote buying and clientelist exchange in settings where political parties are relatively weak such as Thailand (at least prior to its 1997 party-strengthening constitutional reform; see for example Callahan and McCargo 1996) or the Solomon Islands (Wood 2014a), are reminiscent of the picture we have painted here. In both of these contexts, observers have described candidates competing to recruit non-party brokers and bidding for the support of the same voters through fluid and overlapping brokerage structures. In contrast, much of the recent literature on clientelism has drawn examples from political systems, especially in Latin America, which are relatively party-focused, giving rise to assumptions that parties are generally the actors that count in vote buying, and that brokers are agents of parties (though here, too, it has also been pointed out that parties recruit non-party brokers: Holland and Palmer-Rubin 2015). Much of the recent discussion of “turnout buying,” in particular, arose from discussion of clientelist politics in Argentina (see especially Nichter 2008), where elections used closed-list proportional representation, where levels of party identification in the population have been relatively high, and where the Peronist party in particular maintained strong base areas in the country's poor communities, making targeting a party core a feasible strategy. Such conditions are less likely to apply in countries with candidate-focused electoral systems, or where political parties are weakly institutionalized, giving rise to more fluid and fragmented patterns of vote buying such as those described in this article.