Why programmatic parties reduce criminal violence: Theory and evidence from Brazil

Abstract

Extensive research suggests that electoral competition and power alternations increase violence in weakly institutionalized democracies. Yet, little is known about how political parties affect violence and security. We theorize that the type of party strengthened in elections shapes security outcomes and argue that the rise of programmatic parties, at the expense of clientelistic parties, can significantly reduce violence. In contexts of large-scale criminal violence, programmatic parties are less likely to establish alliances with coercive actors because they possess fewer incentives and greater coordination capacity. Focusing on Brazil, we use a regression discontinuity design that leverages the as-if random assignment of election winners across three rounds of mayoral races. We find that violent crime decreased in municipalities where programmatic parties won coin-flip elections, while it increased in those where clientelistic parties triumphed. Our findings suggest that whether electoral competition increases violence depends on the type of party that wins elections.

Introduction

One of the most surprising findings of the last decades is that democracy can have unintended consequences for peace and stability: coercion and democracy can be “enduringly compatible” (Staniland, 2014). Some studies suggest that close victories (Hafner-Burton et al., 2014), ethnic divisions (Wilkinson, 2006), and power balances (Chacón et al., 2011) may turn elections into violent contests. Countries transiting to peace or democracy may also result in outbreaks of large-scale criminal violence if they fail to implement major security sector reforms or transitional justice mechanisms (Trejo et al., 2018).

More recent work has shifted its attention to the role of political parties—their organizational dynamics and political incentives—in sparking conflict. As extensively documented, politicians and criminal actors collude with specialists in violence in pursuit of common goals, which can ultimately lead to outbreaks of violence (Albarracín, 2018; Fjelde, 2020; Mares and Zhu, 2015; Siddiqui, 2022). Yet, despite the ubiquity of criminal ties in some contexts, not all politicians or parties establish relationships with specialists of violence and affect public security to the same extent.

In this article, we study how the type of political party that rises out of power alternations impacts lethal violence. We draw on a key distinction between programmatic and clientelistic parties to understand variation in violence. Programmatic parties implement policy with clear rules of distribution, while also enjoying more solid organizational devices to control their members. In contrast, politicians from clientelistic parties, in addition to conditioning the distribution of goods to political support, tend to actively engage powerful actors capable of mobilizing voters and manipulating elections through fraud (Kitschelt et al., 2010; Mares and Young, 2016).

Building on this distinction, we argue that the rise of clientelistic parties is associated with greater levels of violence: because their victory does not depend on programmatic appeals, clientelistic organizations are more likely to attract opportunistic politicians, cultivate stronger connections to specialists in violence, and selectively enforce the law based on political considerations (e.g., Barnes, 2017; Arias, 2006; Novaes, 2021). In contrast, the rise of programmatic parties, at the expense of their clientelistic counterparts, can prevent greater levels of violence: both fewer incentives and greater coordination capacity make programmatic parties less likely to establish alliances with coercive actors.

We investigate the impact of political parties on violence in Brazil—a country with large-scale criminal violence that transitioned to democracy without substantial security sector reforms. Assessing the relationship between the type of party strengthened in elections and crime rates is challenging because politicians with ties to coercive actors are likely to win in already violent municipalities. If so, it would be hard to tell whether some parties increase violence or select themselves into attractive municipalities.

To overcome this challenge, we leverage the as-if random variation in the assignment of election winners of mayoral elections, allowing us to compare violent conflict in municipalities where programmatic versus non-programmatic won “coin-flip” elections. Specifically, we implement a regression discontinuity design that compares municipalities where programmatic and clientelistic parties barely won or lost elections, helping us construct comparable units governed by either type of party.

Our findings indicate that the exogenous rise of programmatic parties significantly reduced violence in Brazilian municipalities. We document how the victory of programmatic parties led to a decrease in the murder rate. In contrast, the election of clientelistic parties in Brazilian municipalities had the opposite effect and produced a substantial and long-lasting increase in violent crime.

This article makes several contributions. First, we emphasize the importance of political parties, along the programmatic-clientelistic dimension, to explain variation in violence. In doing so, we hope to contribute to an agenda that underscores parties’ incentives and capacity to influence public security outcomes. While there is a growing literature examining how political parties’ organizational dynamics impact conflict, most of it has focused on electoral settings and civil conflicts (Fjelde, 2020; Siddiqui, 2022; Staniland, 2014). This body of work offers crucial insights for understanding how different types of political parties shape criminal violence. Second, our discussion has implications for understanding the connections between political and criminal violence. While civil conflicts and criminal wars have been typically conceptualized as distinct phenomena, scholars have begun to acknowledge the political dimension of criminal actors as well as the pervasiveness of state-crime collusion during civil conflicts. As Barnes (2017) contends, the neat separation between political and criminal violence is “neither conceptually nor theoretically justified” given that “organized crime and its violence is very much a political phenomenon” (967–968). Our study suggests that the politics of alliances between politicians and criminal actors have far-reaching implications for local security outcomes. Lastly, we also add nuance to the argument according to which subnational competition and alternation in new democracies breed violence. We show that power rotation is particularly likely to increase homicide rates when clientelistic parties become empowered, but not when programmatic ones are strengthened.

Argument: How political parties shape violence

Recent research has sought to understand the role played by political parties in the prevention and production of violence. A key lesson from this scholarship is that politicians and criminal actors—including drug cartels, militias, paramilitaries, and gangs—may be willing to cultivate ties in pursuit of common interests. By establishing ties with corrupt and violent actors, politicians are able to increase their leverage over voters and eliminate political opponents (Mares and Zhu, 2015; Albarracín, 2018). While some political parties play a central role in sparking violence, some parties do not.

We emphasize how the type of political party—whether programmatic or clientelistic—has important consequences for violence and provision of public security. Concretely, we suggest that the rise of programmatic parties, relative to clientelistic ones, is likely to favor the reduction of violence. Programmatic parties tend to establish solid linkages with voters and coordinate between members of the political party. Parties are programmatic when they are able to coordinate their members’ behavior around a common project, such that it overrides all the singular, idiosyncratic, personal preferences (Kitschelt et al., 2010). In contrast, clientelistic parties seek victory by targeting material inducements and manipulating elections through vote-buying and fraud. Once in power, clientelistic parties compensate their supporters by delivering subsidies, jobs, and other types of resources (Mares and Young, 2016).

How do programmatic parties, relative to their clientelistic counterparts, affect subnational levels of violence? First, clientelistic parties are more likely to attract lower-quality and opportunistic politicians, less interested in the implementation of programmatic policies, and with stronger ties to specialists of violence (Mares and Young, 2016). Parties that rely on non-programmatic exchanges to secure victory may have incentives to recruit politicians with pre-existing clientelistic resources even if they have ties to coercive actors. Certainly, recruiting politicians with resources and connections to specialists of violence may constitute a comparative advantage in elections and in the selective provision of security for some subsets of the population (Daly, 2021). However, their participation in party politics can stimulate the use of violence as a mechanism of dispute resolution during elections and during office. Indeed, evidence has shown that opportunistic and criminal politicians—some involved in drugs or organized violence—have mostly entered political life through weak and non-programmatic parties. From Indian politicians involved with local gangs (Vaishnav, 2017) to Colombian senators (López, 2010) and Brazilian mayors (Albarracín, 2018) colluded with criminal actors, clientelistic parties have been the platform for such connections.

Second, relative to their clientelistic counterparts, programmatic parties have additional incentives to prevent politicians from establishing alliances with violent actors. Because their probability of victory depends on how well they perform in office, relying on corruption and coercion are risky strategies for programmatic parties. For instance, in case of irregularities, programmatic parties may face higher reputational costs and fears of legal prosecution relative to clientelistic organizations. Of course, all parties are susceptible of having unreliable politicians in their ranks (Novaes, 2018). However, research shows that programmatic parties can in fact constrain politicians’ actions, partly due to anticipated electoral sanctions (Johannessen, 2020): violence against voters and political contenders, a commonly used strategy by weak clientelistic parties, may alienate observers and supporters alike (Asunka et al., 2019).

Finally, while programmatic parties may be reticent to cultivate ties with specialists of violence due to reputational costs, it is also their capacity—their set of organizational devices—what prevents their members from engaging in such practices. This is because programmatic parties are better equipped to control their members’ behavior. Politics becomes programmatic when parties “coordinate their politicians’ pursuit of policies around a collective preference function that overrides all the diverse idiosyncratic personal preference functions held by each individual politician” (Kitschelt et al., 2010, 22). Programmatic coordination implies that parties have internal mechanisms for monitoring and disciplining the behavior of their members.

In contrast, members of clientelistic parties are usually weakly monitored and enjoy higher levels of individual autonomy, which is favorable to the establishment of ties with corrupt and potentially violent actors. Indeed, evidence suggests that members of personalistic and clientelistic parties face fewer constraints to seek alliances with private and violent actors to garner electoral support and eliminate their competitors (Acemoglu et al., 2013; Albarracín, 2018). Controlling opportunism (Mares and Young, 2016), party-switching (Novaes, 2018), and unholy alliances is thus essential for violence reduction because it minimizes the possibility that politicians rely on coercive actors to eliminate competitors and settle disputes.

Data and Empirical Strategy

Brazil offers an ideal opportunity to study the links between political parties and violence. With its high levels of electoral volatility, weak linkages between voters and politicians, and over thirty political parties, Brazil has one of the most fragmented party systems worldwide. Moreover, despite its transition to democracy in the 1980s, the country attained one of the highest homicide rates in South America. Indeed, Brazilians have perceived violent crime as the principal problem they face in their daily lives since the mid-1980s (Caldeira and Holston, 1999). While violence was uniform across all states during the military regime, after democratization homicide rates took divergent subnational trajectories (Hoelscher, 2015), with increasing evidence overwhelmingly pointing to the coexistence of electoral politics and criminal actors as a key driver of violence (e.g., Albarracín, 2018; Arias, 2006; Lessing, 2017).

Major Brazilian parties in the mid-1990s constituted “loose federations of free-wheeling politicians” (Mainwaring et al., 2018), with weak territorial organization and high levels of party indiscipline. Especially in the Brazilian Northeast, conservative machines have dominated local and state politics (Van Dyck, 2014). Since the late 1990s, Brazil’s party system began a gradual transformation. From personalistic appeals and party-switching, the party system gave way to increasing stabilization and reduced electoral volatility (Santos, 2008). Parties heightened their territorial penetration and strengthened organizationally, thus becoming more important players. Despite these changes, most parties continue to be territorially weak and clientelistic, with the major exception of the Workers’ Party (PT), which has crafted mass attachments and developed strong organizational roots through alliances between unions, Catholic groups, social movements, and leftist intellectuals.

Against this backdrop, Brazil has experienced violence across most regions and both urban and rural areas. In urban regions, politicians have allied with powerful drug gangs and pro-government militias seeking to establish territorial control and leverage gains from illicit markets (Arias, 2006). In other regions, particularly small, rural areas, political machines have established clientelistic relationships with paramilitary groups and rural oligarchs, oftentimes leading to violence and impunity in the countryside. In these contexts, local dynasties of families dispute control over municipal politics through alliances with coercive actors and traditional forms of oligarchic rule associated with multiple forms of violence. Indeed, homicide rates increased to a greater extent in rural municipalities, relative to urban ones, during the first decade of the twentieth century. The growth of powerful criminal groups in the peripheries of metropolitan areas implied that many public safety policies targeted urban areas, despite the fact that criminal actors have also proliferated in small-sized peripheries and that “rural Brazil is still a lawless territory where the feeling of impunity feeds crime” (Justus et al., 2016, 218).

Mayoral close elections

To study how the type of party shapes security outcomes in Brazil, we leverage three rounds of mayoral close elections and implement a regression discontinuity design. In particular, we assess whether the as-if random victory of programmatic versus clientelistic parties contributed to changes in violence. This design focuses on a sample of races where a relatively programmatic party (i.e., PT; the Brazilian Social Democracy Party—PSDB; the Democratic Labor Party—PDT; and the Popular Socialist Party—PPS) either lost or won between 2000 and 2008. By the same token, because the victory of clientelistic organizations may cause a spike in violence, we compare towns where such parties (i.e., the Brazilian Democratic Movement Party—PMDB; the Brazilian Socialist Party—PSB; the Liberal Front Party—PFL; and the Brazilian Progressive Party—PPB) won or lost by a narrow margin. ¹

The main underlying assumption of this empirical strategy is that municipalities where close elections were disputed are fairly similar to each other on all observed and unobserved covariates, except for the fact that a type of political party (i.e., programmatic vs. clientelistic) won by a narrow margin in some of them. We provide evidence supporting this assumption in the appendix (Supplemental Figure C1), showing that treated and untreated municipalities were similar across dozens of variables. Our empirical analysis then estimates the following equation

$Y_{i,t}=\alpha +{\beta }_{1}Programmati{c}_{i,t}+{\beta }_{2}Margi{n}_{i,t}+{\beta }_{3}Programmati{c}_{i,t}\times Margi{n}_{i,t}+{\gamma }_{i,t}$

(1)

where Y_i,t is homicide rates. Programmatic_i,t is a dummy variable indicating whether a programmatic party won the mayoral election in year t and municipality i and M_i,t is a polynomial of the programmatic party margin of victory, the difference between the winner and the runner-up.

The resulting estimates are based on local polynomial regressions with robust biased-corrected confidence intervals. To ensure that the results are not driven by chance or confounded by other factors, we report a number of tests in the Appendix. We show that municipalities barely below and above the cutoff are similar on a number of political, economic, and social factors. We then conduct a density analysis of the running variable and find no evidence of sorting around the cutoff. To determine the size of the window within which observations are included, we use a data-driven procedure based on Cattaneo et al. (2016), but we also perform placebo tests with various bandwidth sizes and find consistent results.

Empirical results

Programmatic parties and the reduction of violence

The empirical analysis that follows analyzes how the rise of programmatic parties decreased lethal violence. Our main results, reported in Figure 1, suggest that municipalities governed by programmatic political parties experienced lower levels of homicide rates. We focus on the PT, PSDB, PDT, and PPS, four parties with a programmatic base, and show the results in plot (a). Municipalities right of the vertical line, where programmatic parties barely won, show a fall in violence, relative to municipalities left of the vertical line, where programmatic parties lost the elections. Substantively, the victory of a programmatic party is associated with around 2 fewer homicides per a thousand population and equivalent to 14% of a standard deviation. When using an alternative coding of programmatic party (i.e., PT, PSDB), we obtain consistent and significant results. However, an exclusive focus on the PT yields null results, suggesting that the effect is not simply driven by the left-wing orientation of this particular party, but by its joint effect along with center-right PSDB (Supplemental Appendix B.3).

Figure 1.

Effect of Victories on Homicide Rates by Type of Party (a) Programmatic Parties (b) Non-programmatic Parties. Note: The forcing variable is the margin of victory of the main programmatic parties (PT, PSDB, PDT, and PPS in Panel A) and non-programmatic parties (PMDB, PSB, PFL and PPB in Panel B). The analysis includes the mayoral elections of 2000, 2004, and 2008.

Crucially, our results also suggest that this is not just a story about the beneficial effects of programmatic parties, but also about the pernicious consequences of clientelistic parties for public security. As we show in plot (b) of Figure 1, municipalities where clientelistic parties barely won experienced an important increase in homicides relative to other localities. The victory of such a party leads to 3.6 more homicides per 100,000 population, equivalent to 26% of one standard deviation.

We now disaggregate the previous results and examine the effects of political parties on violence over time in Figure 2. As shown in plot (a), the rise of programmatic parties is, on average, associated with fewer homicide rates although the effect over time is modest and only significant at the 0.1 level in two particular years. Moreover, and consistent with the previous analysis, the results in plot (b) show that the victory of clientelistic parties has deleterious consequences for public security. Once a party with a non-programmatic linkage to voters is elected, municipalities experience a consistent and lasting increase in homicide rates over time. These post-treatment effects are persistent over time and generally significant at the 0.10 or 0.05 level, although they are not detectable anymore four years after the election.

Figure 2.

Effect of Programmatic Parties on Homicide Rates—Yearly Estimates (a) Programmatic (b) Non-programmatic. Note: The outcome variable is homicide rates for individual years, where t is the election year.

Conclusion

The type of party that wins competitive elections has important consequences for subsequent levels of violence. Drawing on the case of Brazil, we have shown that programmatic parties may be better equipped at reducing violence, whereas clientelistic machines have negative consequences for public security.

Our findings contribute to the literature on institutional mechanisms to reduce criminal violence. In many contexts, including Latin America and elsewhere, the most popular response to rising levels of violence has been the adoption of militarized iron fist policies—and Brazil has been no exception. However, this approach has been shown to be counterproductive, as it typically increases violence and undermines human rights (Flores-Macías and Zarkin, 2021; Magaloni and Rodriguez, 2020; Calderón et al., 2015). Consequently, scholars have recently investigated alternatives to tough-on-crime policies (Trejo and Nieto-Matiz, 2022). We add to this growing scholarship by suggesting that empowering programmatic parties at the expense of clientelistic ones can be a powerful alternative to the widely-used iron fist approach.

We hope that future research advances this emerging agenda. First, it is important to empirically document the causal mechanisms through which political parties improve or worsen security. This novel agenda would benefit from investigating how politicians’ connections to a plethora of coercive actors—including militias, insurgencies, and criminal organizations—affect subsequent levels of violence. Crucially, this implies challenging artificial and rigid distinctions, such as that between political and criminal violence, and instead investigating the dynamics and consequences of criminal politics. Lastly, future research should explore how our findings generalize beyond Brazil. Insofar as party politics is present in countries experiencing both criminal wars and “classic” civil conflicts, it is not unreasonable to believe that empowering programmatic organizations, at the expense of clientelistic ones, could be a viable way to lessen lethal violence. Investigating how this effect unfolds and varies across these contexts is of utmost importance.

Supplemental Material

Supplemental Material - Why programmatic parties reduce criminal violence: Theory and evidence from Brazil

Supplemental Material for Why programmatic parties reduce criminal violence: Theory and evidence from Brazil by Camilo Nieto-Matiz, Natán Skigin in Research & Politics