Vote Buying With Multiple Distributive Goods

Abstract

Within the rich literature on distributive politics, models of vote buying treat the distributive logic of different particularistic incentives as theoretically similar. This article relaxes that assumption, focusing on how the nature of a good affects the political logic of its distribution, and then uses data from a new compilation of land transfers and rural investment projects from the 1958–1990 agrarian reform program in Venezuela to empirically test the resulting theoretical implications. By comparing the distribution of land and rural investment, the analysis demonstrates that a party may simultaneously target both swing and core groups of voters with particularistic goods, the choice being determined by the distributive good. Although land was primarily distributed in areas where political competition was highest, rural investment projects were targeted at parties’ core constituencies.

Keywords

clientelism distributive politics core and swing voter models land reform Latin America

How does the nature of a distributive good at an incumbent’s disposal affect the logic with which it is targeted at voters? Although there is a large, rich body of literature on clientelism and patronage politics that seeks to understand the conditions under which political parties will offer material incentives to individuals in exchange for their votes (Cox & McCubbins, 1986; Dixit & Londregan, 1996; Lindbeck & Weibull, 1987; Stokes, 2005), this question remains unanswered. The distributive logic of different particularistic rewards is comparatively understudied because political parties across time and space have made widespread use of common discretional private goods such as delivering jobs or one-time handouts of cash, food, or building materials (e.g., Nichter, 2008; Wang & Kurzman, 2007) to their electoral advantage. Yet the explicit theoretical consideration of how differing types of distributive goods affect their targeting is important, both because of the large number of new and unconsolidated developing democracies in which the rates of clientelism are high (Stokes, 2007) and because of the diversity of distributive goods that parties use in these states (Schaffer & Schedler, 2007). Even rewards such as land, which is most often associated with redistribution (e.g., Lapp, 2004), may also be used by parties as an effective tool of distributive politics.

This article argues that a party can simultaneously target swing and core groups of voters with particularistic goods, the choice being determined by the type of good distributed. Parties have an incentive to win swing voters in contexts where their support may affect election outcomes, but because the value of a swing group may vary by election and because parties have less knowledge of these voters, the likelihood of a given swing voter receiving a future payment in a particular election is less certain than that for a core voter. Swing voters will therefore often require benefits that are valuable beyond a single payment and that are difficult to renege on, the provision of which enables a party to credibly signal its commitment to future interaction with them.

Under these conditions, particularistic goods such as land with effectively finite supply that yield future payoffs to the beneficiary will be targeted at swing voters over core voters, even if a party can more effectively target these goods at a particular group over others (i.e., it is has a core constituency). These characteristics of land not only make it valuable to voters but also enable parties that distribute it to reap longer-term rewards by gaining future leverage over recipients through the activation of a new set of tools (e.g., providing credits) to buy these voters at cheaper rates in subsequent elections. Furthermore, the distributive network constructed to allocate these goods and monitor their use can be used subsequently to improve knowledge of recipients and monitor voting behavior in tight-knit rural communities, even if imperfectly.

In contrast to goods such as land, one-time election payments are more likely to accrue to core voters. A party knows these voters better and interacts with them more consistently over time to maintain their base, making promises of future payments more credible.

The history of patronage and vote buying in Venezuela provides useful ground to test this argument. From the 1958 democratic transition until 1990, the two primary political parties competed over rural votes by implementing an agrarian reform program that redistributed more than half of Venezuela’s cultivable land. These votes became important in building national electoral coalitions capable of winning the presidency. How did governing parties decide how to distribute reform benefits, and how did this contribute to winning votes?

The analysis uses two data sets to examine the spatial and temporal dynamics of the agrarian reform in Venezuela in an effort to understand which peasant groups successfully received benefits and why. Land redistribution data from 1958 to 1990 come from a new, comprehensive compilation of the universe of land transfers in Venezuela administered by the National Agrarian Institute (IAN) (Soto, 2006). The second data set used in the analysis is a compilation of statistics on IAN investment in rural agricultural inputs and infrastructure from 1960 to 1990 (Instituto Iberoamericano de Derecho Agrario y Reforma Agraria [IIDARA], 1993). I couple these data with state- and municipal-level data on agricultural production, union membership, elections, and demographic information to construct a comprehensive understanding of the dynamics of the reform process.

I find that land and IAN investment were distributed in very different ways. Both primary political parties had highly organized political machinery in the countryside during this period, which made it easier to deliver benefits to some groups over others. This is apparent in the distribution of inputs such as tractors and irrigation systems, which parties targeted at regions with high support among peasants living in established agricultural areas that had land at the time reform began. But because of the peculiar nature of land as a distributive good, in that it is effectively finite in supply, its owner reaps future rewards from it through production, and that its productive value can be further increased through the provision of credits and technical assistance, land was targeted at regions with a greater presence of swing voters. Few of those who received land were provided significant rural inputs and services.

Although the variation in the spatial and temporal dynamics of the agrarian reform in Venezuela makes it a useful context to test the argument, the theory is not limited to either Venezuela or the rural sector. Venezuela during this period shared similar characteristics with other developing and clientelism-prone countries, in which powerful political actors employ a wide range of resources across sectors to advance their electoral prospects through particularistic means (Schaffer & Schedler, 2007; Stokes, 2005). Because the theoretical framework advances predictions for the targeting of distributive benefits based on whether a good is difficult to take back and yields long-term benefits to recipients or whether it is a short-term and easily removable benefit, it can be extended to empirically different but theoretically similar types of benefits in other contexts such as the urban housing market (see, e.g., Schady, 2000).

The article proceeds as follows. The first section discusses the literature on distributive politics and presents a theory of incumbent party distributional choices. The second section details the development of political parties and land reform in Venezuela. The third section relates the legal structure of the reform to how political parties and peasant unions interacted to distribute agrarian reform benefits to rural voters. Finally, empirical analyses in the fourth section demonstrate that parties distributed land in regions where political competition and the return to particularistic benefits in votes was highest (swing districts) and agricultural inputs to regions where they had strong political machinery and voter support (core districts).

Distributive Politics and Particularistic Good Provision

Models of vote buying seek to determine the conditions under which competing political parties will offer material incentives to voters in exchange for their votes. One vibrant debate within the literature is that of whether parties will tend to distribute particularistic benefits to core (Calvo & Murillo, 2004; Cox & McCubbins, 1986; Levitt & Snyder, 1995; Nichter, 2008) or swing (Dahlberg & Johansson, 2002; Dixit & Londregan, 1996; Lindbeck & Weibull, 1987; Stokes, 2005) voters. On one hand, if a political party has an advantage at swaying certain voters because it can more accurately predict its reactions to transfers, then it will target these core voters (Cox & McCubbins, 1986; Dixit & Londregan, 1996), especially if it is difficult to monitor swing voter choices or if they require large transfers relative to the core. Stokes (2005), by contrast, argues that parties will target swing voters because those who are already predisposed to one party cannot credibly threaten to vote against them, making targeting that group a waste. This claim echoes earlier theoretical work on targeting swing voters (Lindbeck & Weibull, 1987) and has significant empirical support in the literature (Dahlberg & Johansson, 2002; Schady, 2000), although the latter is also true of core voter theories (Calvo & Murillo, 2004; Levitt & Snyder, 1995).

One implicit assumption in this literature is that the type of distributive good is not a determining factor in whether it will be given to core constituents or swing voters. This is in part because many studies focus on classic discretional private goods such as one-time handouts of cash, food, or building materials that are consumed by the client (e.g., Nichter, 2008; Stokes, 2005; Wang & Kurzman, 2007). The key characteristics of such transfers are that they are discretional, private, and reversible, which may lead politicians to renege on their offer of a transfer once they have been elected, especially if their interaction with voters is short-lived or if through vote monitoring they suspect a voter has defected (Stokes, 2005). But there are other discretional private goods such as land that depart in a consequential way from reversibility. This influences the distributive logic of their provision. The consideration of land redistribution is particularly relevant given its importance in developing states (Dorner, 1992; Lapp, 2004; Tai, 1974), where rates of clientelism are higher (Stokes, 2007).

There are three distinct aspects of land that differentiate it from common forms of vote buying such as one-time cash payments. The first is that land can reward its recipient into the future without having to receive additional payments. The benefactor of a land grant can produce on that plot in future periods, effectively receiving constant payments, in contrast to a one-time election payment that is consumed immediately and cannot be renewed. At the same time, the productive value of land can be significantly increased with the provision of credits and technical assistance, leaving the patron with some future leverage over the land recipient. Finally, once distributed, land is difficult to take back and redistribute again (Tai, 1974), making its supply effectively limited.¹

To explore the effect that the nature of a distributive good has on voter targeting, I compare the distribution of land and the more typical distributive good of rural inputs, demonstrating that parties can simultaneously target swing and core voters, the choice between these groups being determined by the distributive good. This outcome is possible even if the party can more effectively target both benefits at a particular group over others (i.e., that it has a core constituency).

Land is not the only distributive good with characteristics that differ from one-time election payments. The theoretical framework here generates predictions for the targeting of distributive goods according to whether they are difficult to take back and yield future benefits to recipients or whether they are short-term and easily removable benefits. Consequently, it can be extended to empirically different but theoretically similar types of benefits in other contexts.

For instance, at the same time that the Peruvian government was engaged in a large urban settlement and titling program in the 1990s, it was also providing inputs and improvements to urban housing such as roofing, access to water, and access to electricity (Schady, 2000). Other examples such as access to education programs versus gifts of cash or food (Ortega & Penfold, 2008) similarly mirror the theoretical distinction between goods outlined here. Indeed, although the study of clientelism is multifaceted, some apparent conflicts in the literature between core voter findings focused on one-time goods such as urban services (Cox & McCubbins, 1986), food and cash (Nichter, 2008), or annual federal outlays (Levitt & Snyder, 1995) and swing voter findings focused on less reversible goods such as housing or conservation (Dahlberg & Johansson, 2002) or tariff agreements protecting workers (Dixit & Londregan, 1996) might be resolved if the data were viewed through the prism of the theory offered here.

Multiple-Good Voter Targeting

Following the definitions of core and swing groups of voters in Cox and McCubbins (1986), an incumbent party is more certain about how some groups will respond to particularistic rewards than others. Party brokers know these core voters better than unconnected swing voters, and as a result the expected return in votes is higher from a core voter than a swing voter when receiving equivalent payments. At the same time, because these core voters are more responsive to transfers, it is less costly to buy their votes. Parties can better judge when a core voter requires a payment to vote and the price of that payment.

Less knowledge of the preferences of swing voters, however, as well as their likelihood of voting increases the effective payment required to garner their vote. Furthermore, because a party does not know swing voters as well as the core, payments to these voters are harder to monitor over time and less likely consistently delivered across elections. This can be reinforced by interelection shifts in overall support for the incumbent party. If the party gains vote share in a district (e.g., through improving turnout technology for the core), former swing voters may no longer be electorally important. Even in a stable party system where continued interaction with parties is expected, that interaction for a swing voter is more likely to be stochastic in a particular election than for a core voter. Promises of future one-time election payments are therefore more credible to core voters than to a given swing voter. Consequently, swing voters will often require either greater payments or a more credible commitment that a party will not simply pay them today and return only when they need another swing vote in the future.

How can parties solve this problem in contexts where the support of swing voters may affect election outcomes, requiring that parties appeal to voters outside of their core? One solution is to deliver benefits that are valuable to these voters beyond a single payment and that are difficult to recall, thereby credibly signaling a party’s future commitment to a voter.² Land is one effective form of payment in this regard, both because through annual exploitation it pays its recipient in future periods and because, once given, land is difficult to take away for subsequent redistribution. This difficulty renders it effectively finite in supply, enabling a party to credibly commit to future interaction with a swing voter because its one-time provision to swing voters relative to core voters is costly in the present and establishes a relationship that has to endure beyond land allocation for a party to reap its rewards.

Using land as a distributive good also benefits parties in addition to swing voters. First, land distribution to rural voters may not only win their vote in the current election but also change their propensity to vote for that party in the future as they are reminded at each harvest or election by party operatives of the source of their income (Lapp, 2004). By fixing these voters to a plot of land, parties are better able to gain knowledge of these voters (e.g., through peasant unions) and return to them in future elections. This strategy yields greater returns in targeting swing voters than the core constituents about which parties already have greater knowledge and less uncertainty.³ At the same time, the provision of land to rural voters gives parties future leverage over them using tools such as the provision of credit or fertilizer, which multiply a rural owner’s income and reduce the price at which such voters can be bought in the future.

Given the differences in land and input distribution and their payoffs over time, parties are more likely to distribute land to swing voters and inputs to core voters as they value the future more, and as swing voters are more likely to vote for the incumbent if they have received land from them, even if that is not complemented by inputs. The latter is encouraged by a party’s capacity to build networks to gain knowledge of land recipients and monitor their voting behavior. Land distribution to swing voters is also more likely as a party improves its ability to deliver land to swing voters relative to core voters.

Voter Targeting in Venezuela: Why Is Vote Buying With Land Credible?

The above discussion suggests that the logic of land redistribution differs from input distribution because land yields a return to its recipient in the future and because the recipient may vote for the party in the future without receiving further land transfers. But what prevents those who receive land from reneging on their promise to vote for the incumbent?

There are three aspects of Venezuela’s land redistribution that increased the likelihood that swing voters receiving land from a party in one period would support that party in the future. First, land grants far outpaced titling, leaving more than 90% of land recipients without property title (Herman, 1986). Lack of title made it difficult for peasants to obtain private-sector loans because they could not use their land as collateral. Therefore, peasants had to either forego credit if their land was sufficiently productive or turn again to the party if they wanted to increase production output. This was a mechanism distinct from IAN investment that parties used to maintain peasant dependence at low cost.

Second, the peasant union movement, and the Peasant Federation of Venezuela (FCV) in particular, was permeated at all levels by the two major political parties, AD and COPEI.⁴ These parties used the FCV as the main land redistribution broker, which in turn oversaw every step of the redistribution process from aiding peasants in land grant applications to distributing parcels and monitoring their progress and use. The FCV developed a dense organizational network that was capable not only of dealing with peasant demands but also of monitoring their behavior after receiving grants and conditioning future benefits such as credits on that behavior. That the FCV could monitor an individual’s vote, even if imperfectly, was plausible in these rural communities where voters were often geographically immobile and lived in close proximity to family and acquaintances with whom they grew up. Some of these neighbors were active political party or FCV operatives who knew intimate details about the factors that shape an individual’s partisan preferences, and it may have been difficult to directly mislead them about one’s vote (see Stokes, 2005, on a similar environment in Argentine neighborhoods).

Finally, the two major political parties that distributed land (AD and COPEI) were long-standing parties that created stable voter expectations of their future participation in elections. Both parties were founded before 1950 and were the principal competitors throughout the period 1958–1993.

Land Reform and Political Parties in Venezuela

As in most of Latin America, Venezuela had an agriculture-based economy and repressive military rule until the early 20th century. Peasants and their local leaders represented a large base of potential political support if they could be successfully mobilized. Political entrepreneurs, foremost of whom was Rómulo Betancourt, began to mobilize a peasant union movement to form the basis of a national political party with the goal of agrarian reform. When a 1945 election between military candidates threatened a continuation of patronage to the conservative elite, a faction of younger military officers launched a successful coup and brought Betancourt’s party Acción Democrática (AD) into their junta to win popular support. AD extended the franchise to all Venezuelans older than 18 years old, and Betancourt issued a set of agrarian reform decrees that culminated in a more comprehensive, radical agrarian reform law in 1948. Meanwhile, the peasant union movement formed the FCV at the national level. But a 1948 coup joined disaffected opposition parties to a faction of military officers wary of AD rule. Supported by landed elites, the new dictator, Pérez Jiménez, cracked down on AD and reversed the land reform.

Agrarian reform did not resume until a military coup ousted Peérez Jimeénez in 1958 and set elections for later that year. The elections resulted in a victory for AD and Betancourt and ushered in a long democratic period. The pressure for reform carried over to the democratic era, with reform advocates citing the extremely lopsided distribution of land as a reason for redistribution.

Although Venezuela was more urban upon democratization than when Betancourt first began organizing a peasant movement in the 1930s, the rural vote remained critical in building national electoral coalitions capable of winning the presidency. The two most popular parties in the 1963 election, AD and COPEI, “were solidly based in rural areas, yet the Venezuelan population was highly urbanized” (Powell, 1971, p. 139). Similar to the PRI in Mexico and Peronism in Argentina as described by Gibson (1997), AD and COPEI had both urban and peripheral coalitions, the latter of which played an important role in generating electoral victories. The most popular parties in urban areas upon democratization, the URD and FDP, failed to build national coalitions, which contributed to the muted importance of the metropolitan vote to governing parties (Canache, 2004). Although urban voters later shifted toward AD and COPEI, that they were divided between these parties meant that neither party could rely solely on urban constituencies to win elections. Until the 1980s, the burgeoning urban poor were arguably the least effectively represented group by AD or COPEI (Canache, 2004). Consequently, the rural vote remained important to AD and COPEI as they evolved and rotated the presidency from the late 1950s through the 1980s.

After 1960, land was mainly distributed through the IAN according to the Agrarian Reform Law of 1960. From 1946 to 1990, 8.6 million hectares of property were distributed, roughly 10% of Venezuela’s 91 million hectares. But because most land transfers took place within the 11 million cultivable hectares, in the end more than half of all cultivable land was redistributed. The 1960 reform affected a significant portion of the electorate as well. Given the roughly 200,000 families that received land and the average electorate size from 1958 to 1988, the land reform directly affected an estimated 3% of the voting population in each election cycle during this period. This translates to an estimated 12% of the rural voting population affected by the reform per election since it was targeted at rural areas. This is a substantial figure given that two of the seven presidential elections during the period were decided by vote margins of 3% or less.

Venezuela’s land reform was well funded since, as in neighboring Colombia, large landowners played an important role in drafting the legal basis of the reform in cooperation with the major political parties (Powell, 1964, p. 84). Appeasing elites was necessary to maintain their willingness to operate within the framework of democratic institutions. Financial resources derived from the oil industry enabled both significant supplemental expenditures such as the provision of credit and infrastructure and also handsome payment to those whose land was taken. Even some landowners not affected by the reform sold their property to oil companies, becoming urban elite (Karl, 1987).

Venezuela was not the only country in the region to undergo a relatively uncontroversial land reform program or one funded largely by oil. Negotiated land reforms in which most elites collaborated also took place in Costa Rica, Colombia, Chile under Alessandri, and Ecuador, with much of the latter’s funded by oil revenues (Dorner, 1992).

The Process of Agrarian Reform and Benefit Distribution

Venezuelan democracy from 1958 to 1990 was characterized by centralized power and strong political parties that controlled candidate nominations. Simultaneous presidential and congressional elections were held every 5 years beginning in late 1958. Although electoral volatility and the number of competitive parties at the beginning of the period were higher, the principal axis of political competition was between AD and COPEI, the only parties successful in building nationwide organizations and constituencies (Molina, 2004, pp. 156-160). The president had significant policy-making influence and patronage resources, with the ability to create consultative commissions that could draft decrees and legislation, as well as agencies that gave interest groups (including the FCV) formal access to decision making (Crisp, 2000).

The Legal Basis for Agrarian Reform

The vast majority of land redistribution in Venezuela over the period 1958–1990 was managed by the IAN in accordance with the Agrarian Reform Law of 1960. The 1960 law called for redistribution of land that was not exploited according to its social function. In the case of an agrarian problem of “evident seriousness” that could not otherwise be solved, efficiently used land could be expropriated (Article 27). Landowners were to be compensated in either bonds or cash according to the market value of their property (Article 178). In addition, the law called for the provision of credit, assistance, and public services such as roads and irrigation to facilitate agricultural production (Article 112). There was no explicit ceiling set on the size of landholdings.

To acquire land, peasants had to form provisional committees (Article 94) and fill out an application indicating basic member information and a host of technical information regarding the land grant request, such as its ownership status, extent, agricultural suitability, rainfall, and distance from markets (Article 95). The application would be submitted to the local delegation of the IAN (Article 93), which would then review it and send the application to the central IAN office for a determination of whether or not to fulfill the request. If the request was fulfilled, the petitioning group would convoke an assembly at which they would choose an administrative committee to serve as the contact with the IAN, through which it would provide technical assistance and services (Articles 99, 100). Peasants who already had land could also form groups to apply for these benefits.

The Role of the FCV in the Distribution of Agrarian Reform Benefits

The peasant union movement, and the FCV in particular, played a key role in the distribution of agrarian reform benefits. Although AD had a stronger presence in the FCV in its early years, executive power in the organization was shared with COPEI starting in 1962 (Powell, 1971). COPEI built significant peasant support through the FCV and took equally partisan advantage of it while in government (Lapp, 2004). The FCV developed expertise in the law’s technical requirements that were otherwise formidable to peasant groups and typically handled these groups’ petitions. More than 90% of petitions for land submitted to the IAN under Betancourt (1959–1964) were processed by the FCV (Powell, 1964, p. 89). The IAN relied on the FCV both to increase land applications and to oversee land distribution (Lapp, 2004).

The sequence of professional activity of FCV leaders indicates that they were recruited first into political parties, which then guided them into labor union activities and finally into the peasant union movement (Powell, 1971, p. 127). These leaders therefore had an extensive network of connections with political parties and institutions that provided them with structural access to decision making, enabling them to influence the distribution of agrarian reform benefits in one of three ways. First, peasant union movement leaders at the state and national levels, and often at the local level, simultaneously held offices (e.g., agrarian secretary) in political parties. Through these appointments, leaders established party positions on agrarian reform policy and influenced appointments to government positions in agrarian reform agencies. Second, peasant leaders often gained positions in local government, state legislative assemblies, and Congress through elections. Finally, government positions were granted to FCV leaders in executive agencies, which played a large role in the reform. Parties in the governing coalition appointed officials to serve in these agencies, giving them leverage over which demands for benefits were fulfilled. Peasant movement representatives sat on the boards of directors of the IAN and the Banco Agriécola y Pecuario (BAP), on product advisory boards in the Ministry of Agriculture (MAC), and on numerous other boards and committees within the major government agencies that administered the agrarian reform program (Powell, 1971, pp. 120-125).

Political Parties and the Distribution of Agrarian Reform Benefits

The reform was designed in a manner that enabled politicians to garner support based on the success of their land redistribution efforts. Yet the system for capturing votes was complex given the difference in the distributive logic of land, credits, and agricultural inputs induced by the nature of these goods. Land was targeted to regions with a greater presence of swing voters, as were credits, albeit on a more selective basis that depended on local factors such as land quality. But swing groups were not the primary beneficiaries of rural inputs: A 1994 survey of land reform recipients demonstrates that although a sizeable majority was satisfied with the way land was assigned, only 9% approved of the provision of technical assistance and 11% of the delivery of public services (MAC-IAN-IICA, 1995). Instead, parties targeted their core constituencies with investment in agricultural inputs such as tractors and irrigation systems.

Figure 1a demonstrates that the timing of land reform corresponds closely with the election cycle. Most land transfers, as Figure 1b shows, took place in an arc from southern Lake Maracaibo in the northwest to Sucre on the northeast coast and in the municipalities bordering the Orinoco River, which runs through the center of the country from west to east. The greatest intensity of reform was in the same northern arc from Lake Maracaibo to Sucre, where land is more fertile and productive and population density is higher.

Figure 1.

Land reform in Venezuela, 1958–1990: (a) Aggregate land transfers and elections; (b) Log land transfer area by municipality

Once land was received from the IAN, its productivity could be enhanced by agricultural credits, which primarily had to be obtained through the government (in particular, the BAP) since peasants rarely received land titles and therefore lacked collateral to obtain loans. The ability of parties to create both a future stream of benefits and a dependency, in addition to their success in tapping the FCV’s dense organizational network to monitor the future behavior of peasant beneficiaries, led them to distribute land to swing voters. This was a highly effective electoral strategy given the limited ideological differences between AD and COPEI and swing voters in sufficiently large number to play a deciding role in elections between these parties (Coppedge, 1994; Myers & O’Connor, 1983).

Political parties, however, cannot strictly cater to swing voters in a country where voters do not strongly condition their vote on ideological differences between parties because support from their core constituency may erode over time. At the same time incumbent parties distributed land through the IAN to swing voters, they used the same agency to invest in agricultural inputs such as tractors and irrigation systems that would benefit their core constituencies. Both AD and COPEI gained peasant constituencies that extended beyond those who received land through the FCV. Each had an advantage distributing goods in certain areas. These constituencies came largely from rural families who worked on the 250,000 small farms that existed at the time of the reform, a sizeable number in comparison to the roughly 200,000 beneficiaries of the reform. They targeted core constituents in these areas with agricultural inputs, which if delivered to other areas would have yielded a less certain return in votes. Indeed, Powell (1971, pp. 166-171) finds that although FCV membership played a strong role in land redistribution, its effect was less pronounced for obtaining credits. Furthermore, FCV membership had little effect on overall IAN expenditures, which were generally targeted at established agricultural areas (Powell, 1971, pp. 179-180).

Empirical Analysis of the Reform Results: Targeting Voters

The analysis makes use of two geographically and temporally disaggregated data sets to empirically explore the distribution of agrarian reform benefits. The first data set is composed of the universe of land transfers in Venezuela from 1946 to 1990 (Soto, 2006). These data are used to analyze land redistribution through the IAN. Because the causal mechanisms of interest apply to the democratic era of political competition, the analysis restricts attention to the period 1958 to 1990. The second data set is a compilation of statistics on infrastructure investment by the IAN (IIDARA, 1993). These data are used to assess the role of government financing in improvements on rural property and the provision of agricultural inputs. The data include investment on works actually executed rather than proposed or planned investment. The period of analysis is 1960–1990 because of data availability.

For each of these data sets, the analysis follows an empirical strategy similar to that in Dahlberg and Johansson (2002). That is, I first estimate a set of regression models including political variables that tap whether benefits were provided to swing voters, as argued by Lindbeck and Weibull (1987) and Dixit and Londregan (1996). I then estimate models that test the Cox and McCubbins (1986) theory that parties will target core constituencies with particularistic benefits. Finally, a set of encompassing models are estimated to test both theories in an effort to determine which has more support for each dependent variable. The results are also subjected to robustness tests to examine their sensitivity to both model specification and the level of data aggregation.

There are two primary strategies used to empirically identify swing and core groups in the literature. One strategy uses surveys (e.g., Dahlberg & Johansson, 2002; Stokes, 2005). The second strategy, which this analysis follows, uses aggregate vote returns (e.g., Calvo & Murillo, 2004; Dahlberg & Johansson, 2002; Magaloni, 2006; Schady, 2000). Building from Lindbeck and Weibull (1987) and Dixit and Londregan (1996), who derive conditions under which district electoral returns are linked to the presence of swing and core groups of voters, Dahlberg and Johansson (2002, p. 30) indicate that “under some assumptions about the distribution functions (i.e., symmetry and single peakedness) and parties’ objective functions, there will be a one-to-one correspondence between the density at the cutpoint and the closeness of the last election.” The presence of swing voters is therefore associated with the margin of victory in a district. Nonetheless, there are methodological difficulties with both empirical strategies. Surveys suffer a potential endogeneity problem in that self-reported party inclinations may cause an individual to receive handouts or be a result of previous handouts (Stokes, 2007). Using district vote returns is potentially problematic because parties can concentrate benefits to core voters within swing districts and vice versa (Cox, 2006). The extent of the problem with using vote returns depends on the size and heterogeneity of the unit of analysis. Some studies have improved identification of voter groups by focusing on benefit allocation across municipalities in multiple electoral districts (e.g., Dahlberg & Johansson, 2002).

A significant portion of the empirical analysis is conducted at the statewide district level at which congressional legislative candidates were elected.⁵ However, the analysis is also extended to the municipal level for land reform and further focuses on rural municipalities that were the chief potential beneficiaries of reform.⁶ An analysis of these smaller localities within state electoral districts enables better identification of the groups of voters targeted with benefits. Furthermore, similarities between the municipal- and state-level analyses enhance confidence that the statewide district results are indicative of underlying voter group targeting rather than strictly district targeting.

Dependent Variables

There are two dependent variables used in the analysis. The first is the physical area of land transferred in a given state-year, measured in log hectares.⁷ Land transfers are then further disaggregated for analysis at the municipal level. This enables a fine-grained analysis that can assess the marginal effect of the independent variables on each hectare transferred by the government. There were 1,403 land transfers from 1958 to 1990. Given the roughly 200,000 beneficiaries and their families who received land in this period, each family received 42 hectares of land on average. At the state level, land area redistributed has a mean of 10,083 hectares, a standard deviation of 50,197 hectares, a minimum of 0 hectares, and a maximum of 955,600 hectares. At the municipal level, land area redistributed has a mean of 1,289 hectares, a standard deviation of 13,861 hectares, a minimum of 0 hectares, and a maximum of 697,681 hectares.

The second dependent variable is the amount of IAN investment in agricultural inputs in a given state-year, measured in log constant 1970 bolívares. The IAN invested in a range of projects to improve rural properties, such as preparing land to be farmed and providing agricultural equipment such as tractors, installing irrigation and drainage systems, building rural housing structures, and constructing local roads. IAN investment (in millions of bolívares) has a mean of 1.8, a standard deviation of 11.95, a minimum of 0, and a maximum of 259.7.

Independent Variables

The key independent variables in the analysis measure different aspects of political competition to determine whether land and agricultural inputs were allocated to swing or core groups. State-level districts are the primary unit of analysis for three reasons. First, state-level support for legislative candidates was important because presidents who held majorities in Congress were much more effective than those with minority support (Carey, 1996, p. 53).⁸ Second, party control over candidate nominations enabled parties to use congressional nominations strategically by granting seats as both rewards and cooptation devices to forge inroads into organizations that could aid in implementing policy and expanding their constituency (Coppedge, 1994, p. 29). For example, a number of FCV officials served in Congress and as political party leaders during periods of AD rule (Martz & Myers, 1986, p. 336). Because legislators were controlled by the national party leadership rather than state constituencies (Carey, 1996, p. 58), they were responsive to party demands, and the party therefore used them to transmit these demands to the state and local levels through their organizational affiliations. A final reason for using the state as the unit of analysis is that some key variables such as agricultural unions, IAN investment, and agricultural output are not available at further levels of disaggregation. Nonetheless, land redistribution, demographic information, and election returns are available at the municipal level, and because of potential concerns about the identification of different targeted voter groups at the state level as discussed above, I also conduct an analysis of land redistribution at the municipal level.

Political competitiveness is measured in two ways to capture the presence of swing voters. The first measure, electoral win margin, is calculated as the difference between AD and COPEI presidential (large-card) vote shares.⁹ A smaller win margin indicates a more competitive election and a greater presence of swing voters (see, e.g., Dahlberg & Johansson, 2002). The second measure, the effective number of competing parties (ENCP), is the same as that used by Calvo and Murillo (2004) to measure provincial competition levels. A greater number of competitive political parties in a state indicates a more competitive election in which the major parties will compete for the vote share of third-party challengers. From the perspective of the principal parties, these voters act analogously to valuable swing voters. The measure of ENCP is taken from Laakso and Taagepera (1979), who define it as $E N C P = 1 / \sum_{i = 1}^{n} v_{i}^{2},$ where ν_i is the vote share of each political party in agiven election and n is the number of parties competing. As more parties compete and the likelihood of gaining an outright majority declines, smaller parties must be considered as coalition partners and competition increases, leading to an increase in the index.

To test whether benefits are allocated to a party’s core, the analysis employs a standard variable that measures the vote share for the winning party. This variable is intended to capture the strength of political support for the incumbent in each state. If benefits are allocated to the core constituency, then regions with more support for the incumbent should receive more reform benefits.

A number of control variables are included in the analysis.¹⁰ The first is log agricultural production (in thousands of constant 1970 bolívares) since an underperforming agricultural sector is often an impetus for land reform (Dorner, 1992). The Agrarian Reform Law of 1960 specifically targeted potentially productive land not being exploited according to its “social function” to be redistributed to peasants. These data are taken from various years of the Anuario estadístico agropecuario, published by MAC.

Parties targeted the reform at peasants in large part because poverty was concentrated in more rural areas (Oficina Central de Estadística e Informática, 1993), which made rural votes easier and cheaper for parties to buy given diminishing marginal utility of income (see Dixit & Londregan, 1996). Land grants and investment should therefore be greater where the rural population is larger. I include a logged measure of rural population size to capture this, taken from census data. Total population, however, is an indicator of urban areas where agricultural production is less important because of both lack of space and a greater focus on industry. These areas should be less likely to receive land that might lead to further agricultural development, but this is not necessarily the case with investment, which might be equally targeted at these areas because of their proximity to urban food markets. I include log population size as another control variable.

The rural population, as discussed above, was often organized by the peasant union FCV, which lobbied for land grants on behalf of peasant groups and acted as an intermediary between peasants and the government (Powell, 1971). Union strength should therefore play a significant role in receiving land. The FCV did not play as large a role in obtaining investment from the IAN, however, and following Powell (1971) it is not expected to have a strong effect on investment levels. Data on the number of agricultural unions formed yearly by state are taken from annual publications of Memoria y cuenta from the Ministry of Labor. Because there may be a delay between the formation of an agricultural union and government response in the form of a land transfer, the number of agricultural unions in a given year is taken as the sum of the unions formed that year with the unions formed in the previous 2 years.

The analysis also includes a variable for log land area redistributed in models where investment is the dependent variable. Investment is more likely to occur in areas where land is being actively redistributed because of technical demands such as irrigation systems and local road construction, even if these were not the swing regions associated with the heaviest land redistribution efforts. The few regions that received little or no land transfers because of their location and geography are also expected to receive little IAN investment.

Finally, exogenous trends and contemporaneous shocks in land redistribution are controlled for by including decade dummies. Summary statistics of the variables are in an online appendix (available at http://home.uchicago.edu/~albertus/index_files/Page345_files/Appendix_VoteBuying.pdf).

Statistical Methodology: Continuous Models With Censored Data

The main analysis uses tobit models to explore land redistribution and IAN investment, which take into account the censored nature of the data. There were a significant number of state- and municipality-years in which there were no land transfers or IAN investment. In this case, using ordinary least squares models will yield biased coefficients. Robustness tests show the results of models with alternative distributional assumptions.

Because the analysis uses panel data, it is desirable to allow states to systematically differ in the value of the dependent variable for unobservable reasons. Temporal dependence between observations is also a concern. As a result, the models control for regional effects and account for dynamics through the inclusion of a lagged dependent variable for both land transfers and investment.¹¹

Regression Results: Land Redistribution

Table 1 presents the results of a set of tobit regressions on redistributed land area.¹² The models capture significant variation, with a pseudo-R² of about .41. Model 1 provides a baseline, to which political competitiveness variables are added in Models 2–6. Political competitiveness in Models 2 and 3 is associated with more land redistribution. As votes are spread between fewer parties in a state, or the vote share gap between AD and COPEI increases, less land is redistributed. The substantive effect is large: a 1-point increase in ENCP is associated with a roughly 150% increase in land redistribution. There is little evidence, however, that districts with strong incumbent support receive more land. Winning party support in Model 4 is negative and significant. The Model 5 and 6 robustness tests suggest a similar picture. Win margin remains negative in Model 5, although it is no longer statistically significant, and winning party support remains strongly negative. ENCP is positive and strongly significant in Model 6, and winning party support retains its statistical significance. In sum, there is significant evidence that land was targeted at swing districts with a greater presence of swing groups, but no evidence suggesting parties targeted land to core constituents.

Table 1.

Tobit Analyses of Land Redistribution by State, 1958–1990.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Log population	−0.693*	−0.948**	−1.057***	−1.084***	−1.088***	−1.122***
	(0.382)	(0.385)	(0.386)	(0.385)	(0.386)	(0.385)
Log rural population	0.247	0.647	0.770*	0.676	0.688	0.792*
	(0.432)	(0.441)	(0.437)	(0.435)	(0.440)	(0.436)
Log agricultural prod.	2.778***	2.600***	2.618***	2.952***	2.933***	2.777***
	(0.456)	(0.443)	(0.440)	(0.450)	(0.462)	(0.451)
Agricultural unions	0.020***	0.027***	0.026***	0.024***	0.024***	0.026***
	(0.008)	(0.008)	(0.008)	(0.008)	(0.008)	(0.008)
Win margin		−6.117***			−0.426
		(1.670)			(2.361)
Effective number of competitive parties			1.467***			0.874**
Effective number of competitive parties			(0.290)			(0.425)
Winning party support				−8.447***	−8.137***	−4.690*
				(1.697)	(2.414)	(2.481)
Lag redistributed land	0.548***	0.505***	0.447***	0.468***	0.468***	0.444***
	(0.061)	(0.062)	(0.062)	(0.062)	(0.062)	(0.062)
Period dummies	Yes	Yes	Yes	Yes	Yes	Yes
Regional fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Pseudo-R²	.393	.412	.421	.407	.408	.417
Observations	759	759	759	759	759	759

The dependent variable is the log of land area redistributed. Standard errors are in parentheses. Constants, period dummies, and regional effects are not shown.

p < .10, **p < .05, ***p < .01, (two-tailed).

As expected, peasant unions have a consistently strong, positive impact on land redistribution. If the number of unions in a state increases by 35, the standard deviation of this variable, land redistribution is predicted to nearly double. A larger rural population, although significant in only some specifications, is associated with greater vote-buying efforts through land redistribution. A 10% gain in a state’s rural population increases land redistribution by approximately 8%.

The remaining coefficients are highly significant in the expected direction. Greater population is tied to less land redistribution, but redistribution is greater where agricultural production is higher. Lagged land reform is strongly positive in all models but is not a sufficient statistic for states attracting land distribution, which could otherwise cause the coefficients for the political variables to be picking up a residual effect. These variables remain significant, and become stronger, excluding the lagged dependent variable.

Municipal-Level Land Redistribution in Rural Municipalities

As discussed above, there are some drawbacks to an analysis using state-level electoral districts, in particular because of potential difficulties in attributing the targeting of benefits in politically competitive or incumbent-dominated states to swing and core voter groups, respectively. This section disaggregates the data to the municipal level to better identify voting groups within districts. Furthermore, it restricts attention to more rural municipalities, thereby focusing more explicitly on the populations eligible for land reform benefits.¹³

There were 201 municipalities in Venezuela during the period; 22 municipalities with larger cities had populations greater than 100,000 on average over the period. In addition, the federal capital district, the island of Nueva Esparta, and the remote states of Amazonas and Delta Amacuro (all shaded lightly in Figure 1b) all had particularly low levels of agricultural production and rural populations and were less likely to be included in the reform. After excluding urban municipalities and those without significant agricultural production, the average municipality remaining had about 33,000 individuals, or roughly 5,000 families when taking into account average rural family size. Additional summary statistics for these municipalities are found in the online appendix.

Table 2 displays the results, which are largely consistent with those in Table 1. All Table 2 models include a lagged dependent variable, and the models account for unit-specific unobserved heterogeneity using state fixed effects for the tobit models and municipal fixed effects for the conditional logit and negative binomial models.¹⁴ Data on rural population, agricultural production, and agricultural unions were unavailable at the municipal level and are therefore included with state values.

Table 2.

Analyses of Land Redistribution by Municipality in Rural Municipalities, 1958–1988.

	Tobit					Logit		Negative binomial
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9
Log population	1.522***	1.392***	1.549***	1.541***	1.449***	−0.968***	−1.034***	−0.352	−0.343
	(0.411)	(0.411)	(0.409)	(0.410)	(0.412)	(0.338)	(0.338)	(0.242)	(0.293)
Log rural population	−0.562	−0.568	−0.595	−0.596	−0.591	−0.162	−0.152	−0.099	−0.085
	(0.683)	(0.679)	(0.680)	(0.680)	(0.679)	(0.136)	(0.136)	(0.124)	(0.122)
Log agricultural prod.	2.393*	2.402*	2.310*	2.267*	2.236*	0.705**	0.814***	0.026	0.036
	(1.350)	(1.340)	(1.341)	(1.350)	(1.343)	(0.287)	(0.284)	(0.238)	(0.206)
Agricultural unions	0.0305***	0.0325***	0.0296***	0.0301***	0.0321***	0.0091***	0.0089***	0.0066***	0.0062***
	(0.008)	(0.008)	(0.008)	(0.008)	(0.008)	(0.002)	(0.002)	(0.001)	(0.001)
Win margin	−4.994***			−0.548		−1.705***		−1.043***
	(1.623)			(2.021)		(0.510)		(0.370)
Effective number of competitive parties		1.646***			0.913*		0.470***		0.241**
		(0.357)			(0.518)		(0.119)		(0.100)
Winning party support			−6.787***	−6.489***	−4.087*	−1.162***	−0.665	−1.094***	−0.909*
			(1.456)	(1.823)	(2.115)	(0.388)	(0.447)	(0.356)	(0.543)
Lag DV^a	0.891***	0.864***	0.867***	0.867***	0.860***	0.542***	0.516***	0.135***	0.133***
	(0.077)	(0.077)	(0.077)	(0.077)	(0.077)	(0.106)	(0.107)	(0.030)	(0.027)
Period dummies	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes	Yes	No	No	No	No
Municipal fixed effects	No	No	No	No	No	Yes	Yes	Yes	Yes
Observations	4,645	4,645	4,645	4,645	4,645	4,015	4,015	4,015	4,015

The dependent variable is the log of land area redistributed for Models 1–5, a dummy for land redistribution in Models 6–7, and a count of properties redistributed in Models 8–9. Standard errors are in parentheses, and are bootstrapped in Models 8–9. Constants, period dummies, and state and municipal effects are not shown.

This is the lagged log of redistributed area in Models 1–5, the lagged dummy for land redistribution in Models 6–7, and the lagged count of properties redistributed in Models 8–9.

p < .10, **p < .05, ***p < .01 (two-tailed)

Models 1–5 are estimated with a tobit model and mirror the specifications of Models 2–6 in Table 1. The political variables are similar in magnitude and statistical significance to those in Table 1. Municipalities with higher levels of political competitiveness, indicative of a greater presence of swing voters, were more likely to be targeted with land redistribution. Winning party support is again negative and statistically significant across the tobit models, indicating little support for the core voter model for land redistribution.

Focusing on the municipality rather than the state as the unit of analysis, however, introduces significantly more zeros in the data. Models 6–9 address this feature of the data by treating land reform first as a binary outcome and then as an event count. Models 6 and 7 present conditional municipal fixed effects logit estimations of the likelihood that a municipality experiences land reform in a given year. As in the tobit models, political competitiveness is associated with increased odds of land redistribution whether measured using the win margin or ENCP. A one standard deviation increase in win margin reduces the odds of land redistribution in a municipality by roughly 30%. By contrast, winning party support is negative, indicating that incumbent parties did not target their core constituencies with land.

Models 8 and 9 use a negative binomial specification with the number of properties redistributed as the dependent variable.¹⁵ This specification models cases of land redistribution as event counts. The results of the encompassing Models 8 and 9 closely mirror those of other models in Tables 1 and 2: politically competitive municipalities with a greater presence of swing voters were more likely to be targeted with land redistribution over municipalities where core constituencies boosted incumbent support.

Regression Results: IAN Investment

Although Tables 1 and 2 provide evidence that the IAN targeted land at swing over core groups, Table 3 suggests that the opposite was true of IAN investment in agricultural inputs. Model 1 provides a baseline. Agricultural unions and log population are now statistically insignificant, as expected. Political competitiveness variables are added in Models 2–6, as in Table 1. Win margin and ENCP in Models 2–3 are both statistically insignificant, suggesting that IAN investment was not targeted at swing districts as was land. By contrast, the winning party support variable in Model 4 is positive and statistically significant. Incumbent parties invested more in agricultural inputs in regions where they had greater core support: a 0.15 increase in the winning party’s vote share, the standard deviation for this variable, leads to an estimated 23% increase in agricultural investment in that state. The effect is even greater in the encompassing models (Models 5–6) that account for election competitiveness. Win margin is again statistically insignificant in Model 5, and ENCP is significant in Model 6, but with the wrong sign.¹⁶ Overall, and in contrast to the Table 1 and 2 results for land redistribution, Table 3 presents evidence that IAN investment in agricultural inputs was directed at core districts, whereas there is little evidence it was targeted at regions with more swing voters.

Table 3.

Tobit Analyses of National Agrarian Institute (IAN) Investment by State, 1960–1990.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Log population	0.178	0.191	0.169	0.211	0.199	0.174
	(0.152)	(0.154)	(0.154)	(0.153)	(0.153)	(0.153)
Log rural population	0.393**	0.377**	0.406**	0.340*	0.350*	0.386**
	(0.177)	(0.179)	(0.180)	(0.179)	(0.179)	(0.179)
Log agricultural prod.	0.315**	0.314**	0.312**	0.310**	0.310**	0.283**
	(0.126)	(0.126)	(0.126)	(0.125)	(0.125)	(0.125)
Agricultural unions	−0.0002	−0.0008	0.0001	−0.0013	−0.0005	−0.0001
	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)
Log land redistributed	0.108***	0.111***	0.105***	0.120***	0.120***	0.108***
	(0.031)	(0.031)	(0.032)	(0.031)	(0.031)	(0.032)
Win margin		0.478			−0.938
		(0.805)			(1.072)
Effective number of competitive parties			0.057			0.518**
Effective number of competitive parties			(0.141)			(0.202)
Winning party support				1.561*	2.202**	3.769***
				(0.828)	(1.106)	(1.191)
Lag IAN investment	0.320***	0.320***	0.320***	0.313***	0.311***	0.304***
	(0.039)	(0.039)	(0.039)	(0.039)	(0.039)	(0.039)
Period dummies	Yes	Yes	Yes	Yes	Yes	Yes
Regional fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Pseudo-R²	.319	.319	.319	.322	.323	.329
Observations	690	690	690	690	690	690

The dependent variable is the log of IAN investment. Standard errors are in parentheses. Constants, period dummies, and regional effects are not shown.

p < .10, **p < .05, ***p < .01, (two-tailed).

All of the remaining coefficients have the expected sign. Log land redistributed is positive and significant, indicating more investment in areas of active reform, even after accounting for political competition. Rural population and agricultural production are both positive and strongly significant.

Additional Robustness Tests of Land Redistribution and IAN Investment

The results in Tables 1 –3 present overall a consistent picture of how voters were targeted with both land and agricultural inputs. This section further probes the robustness of these results to alternative model specifications. In particular, both land redistribution and IAN investment at the state level are analyzed using generalized least squares (GLS) estimators to model the effect that the presence of swing and core voters has on the targeting of land and agricultural inputs to different states.¹⁷ Land redistribution is also analyzed using a negative binomial specification as in Models 8-9 of Table 2. The GLS and negative binomial models include state fixed effects and a lagged dependent variable. Encompassing models are presented for each dependent variable and each estimation technique.

Models 1–2 in Table 4 present GLS estimates of redistributed land area. These models, as with Models 5–6 of Table 4, specify a heteroscedastic panel error structure with cross-sectional correlation and an AR(1) autocorrelation process within panels. Models 3 and 4 are negative binomial estimates of the count of properties redistributed. As in Tables 1 and 2, Models 1–4 indicate that a greater presence of swing voters is positively linked to land redistribution. This is true using both the win margin and the ENCP measures in the GLS and negative binomial specifications. Winning party support, by contrast, is mostly strongly negatively linked to land redistribution in these models, providing little evidence that parties targeted their core constituencies with land.

Table 4.

Robustness Tests of Land Redistribution and National Agrarian Institute (IAN) Investment by State, 1958–1990.

	Land redistribution				IAN investment
	Generalized least squares		Negative binomial		Generalized least squares
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Log population	−0.390*	−0.503***	−0.731***	−0.729***	0.791***	0.629***
	(0.223)	(0.180)	(0.229)	(0.209)	(0.206)	(0.187)
Log rural population	0.283	0.391**	0.372*	0.458*	−0.149	0.007
	(0.243)	(0.196)	(0.196)	(0.255)	(0.183)	(0.165)
Log agricultural prod.	−0.139*	−0.186***	0.460***	0.447**	−0.726***	−0.690***
	(0.078)	(0.065)	(0.153)	(0.175)	(0.097)	(0.096)
Agricultural unions	0.0172***	0.0184***	0.0063***	0.0063***	−0.0008	0.0003
	(0.002)	(0.002)	(0.001)	(0.002)	(0.001)	(0.001)
Log land redistributed					0.089***	0.078***
					(0.006)	(0.006)
Win margin	−1.540***		−0.751**		0.059
	(0.529)		(0.349)		(0.316)
Effective number of competitive parties		0.921***		0.242***		0.509***
Effective number of competitive parties		(0.089)		(0.072)		(0.053)
Winning party support	−2.451***	−0.008	−1.539***	−0.931*	0.750**	2.973***
	(0.616)	(0.541)	(0.411)	(0.522)	(0.328)	(0.298)
Lag DV^a	0.311***	0.273***	0.050***	0.046***	0.139***	0.119***
	(0.031)	(0.030)	(0.009)	(0.008)	(0.026)	(0.026)
Period dummies	Yes	Yes	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Observations	759	759	693	693	690	690

The dependent variable is the log of land area redistributed for Models 1–2, the count of properties redistributed in Models 3–4, and the log of IAN investment in Models 5–6. Standard errors are in parentheses. Models 1–2 and 5–6 specify a heteroscedastic panel error structure with cross-sectional correlation and an AR(1) autocorrelation process within panels; Models 3–4 have bootstrapped standard errors. Constants, period dummies, and state effects are not shown.

This is the lagged log of redistributed area in Models 1–2, the lagged count of properties redistributed in Models 3–4, and the lagged log of IAN investment in Models 5–6.

p < .10, **p < .05, ***p < .01 (two-tailed).

Models 5–6 provide GLS estimates of IAN investment. Unlike land redistribution, incumbent parties invested more heavily in agricultural inputs where their core support was higher. This is indicated by the positive and statistically significant coefficients for winning party support. Win margin, by contrast, is statistically insignificant in Model 5, although ENCP is positive and significant.

Conclusion

This article has sought to answer the question of how the provision of particularistic rewards varies according to the distributive goods at an incumbent’s disposal. How did governing parties in Venezuela decide to distribute the agrarian reform benefits that resulted in redistribution of more than half of the cultivable land in the state and large rural investment projects, and how did their allocation decisions contribute to their ability to win political support?

The analysis demonstrates that incumbents’ distributive decisions depended on the nature of the particularistic goods they used to court rural voters. Although the two major political parties each had core constituencies to which they directed rural investment projects funded by the IAN, they used the same agency to target land to swing districts where political competition was highest. This strategic choice was the result of the nature of land as a distributive good, in that it is effectively finite in supply, its owner reaps future rewards from it through production, and its value can be increased with agricultural credits that the government can monopolize by withholding property titles from land transfers.

There are at least two reasons why these findings from Venezuela are applicable to other settings. First, Venezuela in this period shared similar characteristics with other developing and clientelism-prone countries. Parties in these contexts do not typically limit their distributive repertoire to one good, but rather avail a range of resources across sectors to reach a wider constituency and advance their electoral prospects, as indicated in recent literature (see, e.g., Schaffer & Schedler, 2007; Stokes, 2005). In developing contexts where rural voters are valuable and land ownership is an important determinant of rural well-being, a market for rural votes can be generated using a set of land policies targeted at different aspects of rural livelihood. There is strong evidence of this in a range of other cases from Latin America (see, e.g., Lapp, 2004; on Mexico, see Magaloni, 2006) to Africa (Boone, 2007) and beyond.

Second, because the theory anticipates that benefit targeting is a function of whether a good is difficult to recall and yields long-term benefits to recipients or whether it is short term and reversible, it can be extended to theoretically similar benefits in different contexts. Fujimori’s simultaneous urban land titling program and investment in housing improvements in Peru (Schady, 2000) as well as Chaévez’s misiones programs in both adult education and subsidized food in Venezuela (Ortega & Penfold, 2008) provide two examples. Future research might fruitfully examine whether vote buying in such contexts is conditioned on the type of good provided. If so, existing models of voter targeting would do well to incorporate how the nature of a good affects whether it is targeted at swing or core voter groups.

Footnotes

Acknowledgements

The author would like to thank Jim Fearon, Steve Haber, Karen Jusko, Terry Karl, David Laitin, Jonathan Rodden, Dan Slater, and seminar participants at the Berkeley–Stanford Comparative Politics Conference for comments on previous versions of this article.

Declaration of Conflicting Interests

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

Funding

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

Notes

Author Biography

Michael Albertus is an assistant professor of political science at the University of Chicago. His research interests include distributive politics, the relationship between political regimes and redistribution, the political economy of dictatorship, and democratization.

References

Boone

Catherine

. (2007). Property and constitutional order: Land tenure reform and the future of the African state. African Affairs, 106, 557-586.

Calvo

Ernest

Murillo

Maria

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

Canache

Damarys

. (2004). Urban poor and political order. In McCoy

Jennifer

Myers

David

(Eds.), The unraveling of representative democracy in Venezuela (pp. 33-49). Baltimore, MD: Johns Hopkins University Press.

Carey

John

. (1996). Term limits and legislative representation. Cambridge, UK: Cambridge University Press.

Coppedge

Michael

. (1994). Strong parties and lame ducks. Stanford, CA: Stanford University Press.

Cox

Gary

. (2006). Swing voters, core voters and distributive politics. Unpublished manuscript.

Cox

Gary

McCubbins

Mathew

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

Crisp

Brian

. (2000). Democratic institutional design. Stanford, CA: Stanford University Press.

Dahlberg

Matz

Johansson

Eva

. (2002). On the vote-purchasing behavior of incumbent governments. American Political Science Review, 96, 27-40.

10.

Dixit

Avinash

Londregan

John

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

11.

Dorner

Peter

. (1992). Latin American land reforms in theory and practice. Madison: University of Wisconsin Press.

12.

Gibson

Edward

. (1997). The populist road to market reform. World Politics, 49, 339-370.

13.

Herman

Donald

. (1986). Agriculture. In Martz

John

Myers

David

(Eds.), Venezuela: The democratic experience (pp. 329-363). New York, NY: Praeger.

14.

Instituto Iberoamericano de Derecho Agrario y Reforma Agraria. (1993). Banco de Datos de la Reforma Agraria [Database of the agrarian reform]. Mérida, Mexico: Universidad de Los Andes.

15.

Karl

Terry

. (1987). Petroleum and political pacts: The transition to democracy in Venezuela. Latin American Research Review, 22, 63-94.

16.

Laakso

Markku

Taagepera

Rein

. (1979). The “effective” number of parties: A measure with application to West Europe. Comparative Political Studies, 12(1), 3-27.

17.

Lapp

Nancy

. (2004). Landing votes. New York, NY: Palgrave Macmillan.

18.

Levitt

Steven

Snyder

James

. (1995). Political parties and the distribution of federal outlays. American Journal of Political Science, 39, 958-980.

19.

Lindbeck

Weibull

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

20.

MAC-IAN-IICA. (1995). Evaluación de la Reforma Agraria [Evaluation of the agrarian reform]. Caracas, Venezuela: MAC.

21.

Magaloni

Beatriz

. (2006). Voting for autocracy. Cambridge, UK: Cambridge University Press.

22.

Molina

José

. (2004). The unraveling of Venezuela’s party system. In McCoy

Jennifer

Myers

David

(Eds.), The unraveling of representative democracy in Venezuela (pp. 152-178). Baltimore, MD: Johns Hopkins University Press.

23.

Myers

David

O’Connor

Robert

. (1983). The undecided respondent in mandatory voting settings. Political Research Quarterly, 36, 420-433.

24.

Nichter

Simeon

. (2008). Vote buying or turnout buying? American Political Science Review, 102, 19-31.

25.

Oficina Central de Estadística e Informática. (1993). Mapa de la Pobreza [Map of poverty]. Caracas, Venezuela: Author.

26.

Ortega

Daniel

Penfold

Michael

. (2008). Does clientelism work? Electoral returns of excludable and non-excludable goods in Chávez’s misiones programs. Unpublished manuscript.

27.

Powell

John

. (1964). A brief political history of agrarian reform in Venezuela. Unpublished manuscript.

28.

Powell

John

. (1971). Political mobilization of the Venezuelan peasant. Cambridge, MA: Harvard University Press.

29.

Schady

Norbert

. (2000). The political economy of expenditures by the Peruvian Social Fund (FONCODES), 1991–95. American Political Science Review, 94, 289-304.

30.

Schaffer

Frederic

Schedler

Andreas

. (2007). What is vote buying? In Schaffer

Frederic

(Ed.), Elections for sale (pp. 1-17). Boulder, CO: Lynne Rienner.

31.

Soto

Oscar David

. (2006). La Cuestión Agraria en Venezuela [The agrarian question in Venezuela]. Mérida, Mexico: Universidad de Los Andes.

32.

Stokes

Susan

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

33.

Stokes

Susan

. (2007). Political clientelism. In Boix

Carles

Stokes

Susan

(Eds.), Handbook of comparative politics (pp. 604-627). Oxford, UK: Oxford University Press.

34.

Tai

Hung-Chao

. (1974). Land reform and politics. Berkeley: University of California Press.

35.

Wang

Chin-Shou

Kurzman

Charles

. (2007). The logistics: How to buy votes. In Schaffer

Frederic

(Ed.), Elections for sale (pp. 61-78). Boulder, CO: Lynne Rienner.