Politics and social violence in developing democracies: Theory and evidence from Brazil

Political Geography 44 (2015) 29e39

Contents lists available at ScienceDirect

Political Geography journal homepage: www.elsevier.com/locate/polgeo

Politics and social violence in developing democracies: Theory and evidence from Brazil Kristian Hoelscher Department of Political Science, University of Oslo & Peace Research Institute Oslo, PO Box 9229 Grønland, NO-0134 Oslo, Norway

a r t i c l e i n f o

a b s t r a c t

Article history: Available online

How do the political institutional features of developing democracies influence how violence occurs? Building on research showing that ‘hybrid democracies’ are more prone to social violence, this article argues that elite competition for power in the context of limited institutional oversight plays an important role in explaining violence. The framework here presents possible mechanisms linking subnational political dynamics and rates of social violence in poorly institutionalised contexts. It highlights how political competition, concentrated political power, and constraints on cooperation can create opportunity structures where violence is incentivised and the rule of law is undermined. This is examined empirically using sub-national homicide data from over 5000 Brazilian municipalities between 1997 and 2010. Findings suggest violence is greater in contexts that are highly competitive e where political actors face credible challenges and have a more tenuous grip on power e and those where power is highly concentrated e where political actors have held power for longer periods or face limited credible challenges. Findings also suggest violence varies depending on whether interactions between state and municipal government are likely to be constrained or cooperative; and are consistent with literatures emphasising the importance of structural explanations of social violence. In light of on-going democratic transitions across the globe, the article highlights the value of understanding links between institutional context, contentious politics and social violence. © 2014 Elsevier Ltd. All rights reserved.

Keywords: Social violence Homicide Democratisation Hybrid democracy Political institutions Local politics Urbanisation Brazil Latin America

Introduction Third Wave democratisation sought to increase the openness of competition and citizen participation in politics. While many new democracies are more representative, democratic transitions have paradoxically coincided with increased violence in many countries, both as organised armed conflict (Hegre, Ellingsen, Gates, & Gleditsch, 2001; Mansfield & Snyder, 1995); and interpersonal forms of social violence (Fox & Hoelscher, 2012). Social violence is increasingly acknowledged as a threat to development (World Bank, 2011a, 2011b), eroding social capital; constraining access to employment, education, and health services; deterring public service provision and investment; and encouraging repressive governance or policing policies (Jütersonke, Krause, & Muggah, 2007;
ndez, & Loayza, 2002; Muggah, 2012). This has Lederman, Mene occurred particularly in Latin America (Holston, 2009b; Koonings, 2001; Zaluar, 2004), where hallmarks of formerly undemocratic

E-mail address: [email protected]. http://dx.doi.org/10.1016/j.polgeo.2014.08.002 0962-6298/© 2014 Elsevier Ltd. All rights reserved.

regimes have persisted, and factional conflicts between individuals or groups are often solved through extra-legal means. While greater competition and participation in political and economic arenas has been encouraged, recently formed democracies have often lacked the institutional capacity to prevent transgression of societal norms, and the legal frameworks to constrain the use of force (Karstedt, 2009). In understanding how political dynamics in new democracies might influence social violence, looking at the national level preclude important sub-national factors. Violence is rarely uniform within a national territory, and the degree of violence, or its causes, can differ between cities, states and regions. In Brazil, for example, state-level homicide rates ranged between 12.9 and 66.8 per 100,000 in 2010, with trends over the previous decade including declines of 70% and increases of 300%. Such intra-country variation reiterates the utility of geographically disaggregated approaches to study violence. Recent work focussing on armed conflict, for example, highlights significant spatial and geographic variation in how, when and why civil violence emerges (Buhaug & Rød, 2006; Verpoorten, 2012; Zhukov, 2012). Despite this, social violence has yet to be rigorously studied using subnational approaches; and

30

K. Hoelscher / Political Geography 44 (2015) 29e39

while the importance of local factors has been emphasised, limited empirical consideration is given to subnational political and governance contexts. Therefore, while democratising countries are generally more prone to social violence, the reasons underlying this remain unclear. Common narratives suggest the consolidation of autocratic or democratic state power and development of robust institutions are important for resolving conflicts and reducing violence (Bates, 2001, 2008; Chesnai, 1981); and violence emerges where institutions fail to consolidate the rule of law and citizens do not internalise norms of self-control (Eisner, 2001). While highlighting the disruptive nature of political transitions and the utility of political order, such explanations simply point to broad, general theories, with limited consideration of specific links between unconsolidated institutional contexts and violence. This is surprising given the emphasis on the role of institutions in structuring social, economic and political interactions and establishing a modicum of order in societies (North, 1991; North, Weingast, & Wallis, 2009). Moreover, while socio-economic and structural factors are important in explaining rates of violence (e.g. Fajnzylber, Lederman, & Loayza, 2002; LaFree & Drass, 2002), these variables are often poor at explaining variation over time. In light of this, the current article examines whether subnational political factors are related to rates of social violence in developing democratic contexts. It develops a framework suggesting political competition, concentration of power, and cooperation can affect incentives for the use of violence; and empirically tests this using longitudinal data from Brazilian municipalities. The arguments presented here draw on contentious politics literatures (Tilly, 2003; Tilly & Tarrow, 2007) which highlight how political opportunity structures produced by institutional changes can create space for social movements and protests to occur. Central here is that political contestation in weakly institutionalised contexts can create motives for social violence to occur; and weak institutions and limited institutional oversight yield the opportunity and means to do so. The study addresses three perceived knowledge gaps. First, it identifies possible relationships between political conditions and homicide rates in developing democracies, pointing to potential mechanisms by which political opportunity structures in incompletely institutionalised contexts could engender violent societies. Second, by disaggregating the study of politics and social violence, the paper seeks to understand these dynamics within a state in a novel way. Finally, by focussing on unconsolidated institutional contexts, the framework and findings offer insights into how violence may emerge in other countries undergoing democratic transitions. Findings suggest a political dimension to social violence in Brazil. Violence is greater in contexts that are highly competitive, where political actors face credible challenges and have a more tenuous grip on power; and those where power is highly concentrated, where political actors have held power for longer periods or they face limited credible challenges. Results also suggest violence may increase or decrease depending on whether state and municipal government interactions are likely to be constrained or cooperative. Moreover, results are consistent with literatures emphasising structural covariates of social violence with homicide rates greater in poorer, more unequal, more populous, and more urbanised municipalities. The paper proceeds as follows. Section 2 reviews theories on social violence and introduces a simple theoretical framework linking political opportunity structures and social violence. Section 3 briefly outlines the context of democratisation and violence in Brazil, Section 4 presents data and results of the empirical analysis, and Section 5 discusses findings and concludes.

The politics of social violence Social violence refers to pervasive violence without an overt political motive. Typically measured by rates of homicide due to greater rates of reporting, the term includes ‘acts of violence committed by individuals or groups that do not reflect an attempt to contest the authority of a state…including assault, murder, gang violence and communal violence’ (Fox & Hoelscher 2012: 433). Traditionally, research on homicide or social violence has been the mainstay of criminologists. Focussing on North America or Europe, this literature has primarily focused on macro-level structural explanations or individual and community level risk factors such as race, inequality, poverty, and the functioning of welfare systems (e.g. Tcherni, 2011; Krivo, Peterson, & Kuhl 2009; Like, 2011). A growing body of literature has also begun to study homicide and violent crime from a comparative perspective in less developed regions (e.g. Marshall, Marshall, & Ren 2009; Stamatel, 2009; Ouimet, 2010). However, comparative criminologists still tend to emphasise macro-structural factors derived from American or European contexts e such as demographic change, inequality and poverty (see Nivette, 2011 for a review). While partly data-driven, such approaches often overlook differences in political and institutional factors between developed and developing countries and whether this may be important in explaining social violence in less economically and politically developed regions. This is somewhat puzzling given links between hybrid democracy and civil conflict generally (Hegre et al., 2001; Hegre & Sambanis 2006); and democratisation and violence in Latin America (Caldeira & Holston, 1999; Pearce, 2010). Some studies have examined how institutional conditions and interpersonal violence may be related. These draw in large part on research linking rates of crime with legitimacy and reform of political, social and family institutions in the United States (LaFree, 1998). Other important contributions by Stucky (2003), and Stucky, Heimer, & Lang (2005) show how the local political context is important both in public security and criminal justice outcomes; and mediating the effect of structural conditions. In cross-national literatures, social violence has been linked with hybrid democracies (Fox & Hoelscher, 2012; Neumayer, 2003), modernisation processes (LaFree & Drass, 2002; LaFree & Tselsoni, 2006), and institutional upheaval and rapid democratic transitions (Schutte & Weidmann, 2011). Violence appears lower where there is consolidation of democratic or autocratic state power (Eisner, 2001, 2008) and where democratic norms are widely adopted (Karstedt & LaFree, 2006). While instructive, these insights are limited in identifying specific ways violence emerges during periods of social and political change. Nivette and Eisner (2013) partially address this, outlining how perceived political legitimacy relates to lower homicide rates cross-nationally. Where states cannot legitimately preserve functional social, economic, legal or political institutions, internal (personal) and external (coercive) forms of social control are weakened; alternative forms of conflict resolution and private protection emerge to fill the regulatory roles the state fails to provide; and inequalities arising from a failure of the social contract fuel grievances that promote violence. This reiterates the importance of ‘grievance as motive’, a finding which recurs throughout the homicide literature (Fajnzylber et al., 2002; Krahn, Hartnagel, & Gartrell, 1986). Particularly in Latin America, political transitions may also signal shifts from political to social violence. Rodgers highlights how the ‘shift from “political” violence to “social” violence… is frequently linked to a broader Latin American “crisis of governance”, whereby… incomplete democratisation… undermined the political authority of states and their ability to command a

K. Hoelscher / Political Geography 44 (2015) 29e39

monopoly over the use of violence’ (Rodgers, 2009: 950). This may especially apply in urban areas. While the (re)emergence of democracy has largely eliminated political violence in Latin America, ‘the contradictions of democratic governance, a tradition of impunity, weak political institutions, and rising social inequalities have provoked a recent upsurge of urban violence’ (Imbusch, Misse, &
n, 2011: 99). In Brazil, political change and violence may Carrio also go hand-in-hand. Adorno notes that ‘the progress of democracy… is not absent of conflicts’, and conflicts ‘intensify as a result of democracy… not only between institutions e old and new e but also between political leaderships using different styles and degrees of commitment to the public good’ (2013: 422). Despite these assertions, and burgeoning interest in ‘uncivil’ democratic transitions (Holston, 2009a), cross-national criminology (Nivette, 2011), changing forms of violence (Pinker, 2011), and post-conflict state weakness (Cruz, 2011), limited empirical research exists on specific ways political conditions in developing democracies may relate to social violence. This article addresses this, arguing that unconsolidated democracies have aspects of their institutional configuration that create opportunities and incentives for political actors to behave transgressively; and that this has implications for how violence displays more widely in society. Importantly, this approach does not preclude structural explanations; but rather emphasises potentially important and underresearched political contexts. In this framework, opportunistic political actors, when afforded by institutional weaknesses, may co-opt public institutions for private gain, with direct or indirect violence a potential outcome. This is argued to be more likely primarily under conditions where political power is tenuous; where it is decisively entrenched; or where cooperation is constrained. This is outlined in Fig. 1 below. The framework supposes the political context becomes important in explaining violence in institutional settings where contestation is allowed yet the rules governing contestation fail to function reliably. Here violence may occur in two ways: through the direct use of illegitimate force at the behest of political actors; or indirectly through the weakening of institutions which safeguard the rule of law and instil trust between citizens and in the security the state provides.1 Three possible mechanisms linking politics and violence are suggested: a high degree of credible political competition; a high concentration of political authority and power; and constraints on co-operation between parties at different levels of government. Political competition Highly competitive politics where checks and balances on elite behaviour are weak may encourage violence through direct

31

intimidation of political opposition or the personalised use of the state apparatus. Closely contested politics signals that organised opposition poses a credible threat to those in power. Where consolidated institutions are present, competitive democratic politics would usually remove political actors from power through the electoral process. The response to competition in functional electoral democracies therefore is greater accountability and service to constituents. Yet without functional institutions, competitive politics may see violence arise. Particularly in poorly institutionalised contexts with limited checks on executive power, responses to competition may include strategies that indirectly or directly use or create violence to maintain or attain power. In the absence of close competition, violence may be discouraged by political actors as it could undermine electoral support. However, if politics is sufficiently competitive and institutional conditions allow, it may be simpler for political actors to deploy or use state institutions to protect their electoral gains rather than risk losing legitimate political contests. This could include selectively using or encouraging violence as a means to threaten communities to vote for a particular candidate, punishing communities or individuals who are perceived to be against certain political interests, or by political actors co-opting illegal groups as de-facto paramilitaries. Several studies offer support. Homicides in Russia expanded as politics was opened to competition (Pridemore & Kim, 2008). During 2007 elections in Kenya, vote buying as incentive and violence as punishment were selectively used according to voters'
rrez-Romero, 2012). In Mexico, more party preference (Gutie competitive elections disrupted existing patronage networks and were associated with greater rates of violence (Villareal, 2002). In India, political elites manipulated use of violence between groups to further their own political ambitions in competitive circumstances (Wilkinson, 2004). A more restrained mechanism linking competition and violence is that political actors do not overtly use the threat of violence, but rather undermine state institutions by using them to their own ends rather than in the public interest. In particular, the personalised use of the state apparatus may see institutions fail to order society in a predictable manner, which if severe enough can undermine the rule of law in a society. Even where local political actors do not oversee responsibility of the formal security apparatus like the police or judiciary, local strongmen may still exert significant influence over subnational institutions. Where institutions are used in a personalised manner e or as a device to prop up power e a breakdown in the rule of law leading to increasing violence is a feasible outcome (Arias & Goldstein, 2010). All things equal, it is hypothesised that highly competitive political competition in unconsolidated institutional contexts should be related with greater rates of social violence.

Fig. 1. A stylised model linking politics and social violence in developing democracies.

32

K. Hoelscher / Political Geography 44 (2015) 29e39

Hypothesis 1. Homicide rates will be greater in municipalities where political contestation is more competitive. More directly, elections themselves are the clearest signal of elite competition. They can entail explicit use or threats of violence to preserve power (Dunning, 2011), or coincide with periods of increased criminal violence (Ghosh, 2006; Meloni, 2012). This relationship, however, is unclear, as governments may increase spending or policy implementation in election years to win votes. Levitt (1997) shows election-year interventions can include improved public security policies which reduce violence. Winning parties may also remove instances of electoral violence from official records, leading to underreporting of violence in election years. However, given that political competition should increase opportunity structures for using violence, a second hypothesis specific to elections can be formed. Hypothesis 2. Homicide rates will be greater in election years Concentration of power Similar to competition creating opportunity structures for violence, the concentration of power may do the same. In a weakly institutionalised context, concentrated power or the absence of credible opposition may provide conditions for political actors to undermine democratic norms and the rule of law. Should the rule of law be undermined by entrenched political actors neglecting the public interest, violence may be incentivised more widely in society. In particular, parties which maintain power over long periods of time e or perceive little credible opposition to them being able to do so e are feasibly more able to control the architecture of local politics to preserve their power. This can, for example, include controlling the police or judiciary through political appointments, or developing ties with criminal actors to protect political interests or hinder their opponents (Arias & Goldstein, 2010). As such, entrenched political actors or those who perceive their political position as unchallenged might reasonably have greater motive and means to exert personal influence over local institutions. Support for this position is found in theories on subnational authoritarianism, which show how political actors pursue strategies to concentrate their power and perpetuate authoritarian-style rule in subnational jurisdictions of democratic countries, particularly in Latin America (Gibson, 2010; Montero, 2010). Moreover, findings suggest second-term mayors in Brazil are more corrupt and less financially accountable than their newly elected counterparts (Ferraz & Finan, 2011). This concentration of power argument suggests incumbent political actors e or those who perceive little threat to their authority e would be more likely to take advantage of their concentrated power in a manner which could undermine public safety. This forms the following hypothesis: Hypothesis 3. Homicide rates will be greater in municipalities where power is concentrated. It is not contradictory, as is presented here, that both competition for power and concentration of power might be associated with greater social violence. Indeed these contexts might exist at opposite ends of a competition-consolidation continuum. All things equal, parties in power with very low vote shares, opinion ratings, or margins of victory face greater competition and their power is less concentrated. Conversely, very high vote shares, opinion ratings, or margins of victory suggest more limited credible competition and potentially higher concentrations of political power. As suggested in the framework, both may be conducive to violence occurring. In between these extremes, competition is present without being intense, and power is held steadily without being highly concentrated. Here, political actors are not sufficiently threatened by close

political contests; nor is their power concentrated enough that they would be willing or able to mobilise violence directly, or indirectly use state power to their own ends in a way that may undermine public security. In these situations, there may be more to gain by adhering to more ‘legitimate’ democratic rules rather than acting in ways which may undermine institutions and public security. This non-linear relationship is tested with the following hypothesis: Hypothesis 4. Homicide rates will be greater where competition for power and concentration of power are either relatively low or relatively high. While the theoretically plausible assertion made here is that violence in an unconsolidated democracy can be incentivised where there is highly competitive politics or highly concentrated power, a caveat should be made regarding the use of electoral data to infer incentive structures of political actors.2 For example, electoral victories with large margins may be due to accountable political conduct or successful policies being rewarded. Likewise, close elections may represent a race between two progressive candidates. In such circumstances, high or low margins of victory may be less reliable measures of how likely political actors might be to engage in personalistic politics, or directly or indirectly undermine public security. As such, while this article outlines how the intensity or relative absence of political competition in developing democracies might be among the best available proxies for how elite incentives can affect social violence, it should be acknowledged that electoral results may imperfectly measure how political actors internalise local institutions. Constraints on cooperation Relationships between political parties at different levels of power may also be important in understanding violence. Central and peripheral political units e such as national, state, and local authorities e coordinate on issues including budgeting, policy planning and implementation; and parties rely on one other for political and electoral support. In Brazil, for instance, municipal and state governments share responsibility for service delivery in municipal jurisdictions. While municipal governments handle functions including housing, education and health, public security is primarily the responsibility of the state government. This last point is important, as municipalities rely on states in meeting their needs for policing and law enforcement. Such institutional arrangements where decentralised political power relies on coordination across levels of government might conceivably see conflicts arise between different levels of political authority. This third aspect of the framework posits that these centreeperiphery relationships exist on a continuum between cooperative and contentious. Simplifying this, one may assume that members of the same political party would share common policy views and ideological histories, and be more likely cooperate. Conversely, opposing parties might be less cooperative, or even attempt to undermine actors from opposition parties. There are direct benefits to party congruence across administrative units in addressing public security policy. In Brazil, shared party affiliations and close ties between the mayor and governor of Rio de Janeiro were instrumental in gaining support for, and implementing, public security policies such as favela pacification programs (Lessing, 2012). Yet absent of institutions ensuring cooperation, fractious relations between opposition parties at different levels of government may undermine public security provision, negatively impacting rates of violence. Dynamics of cooperation and contention can be detrimental to policy implementation. Myers & Dietz (2002) show how ‘vertical’ political contestation hindered police reform and public security

K. Hoelscher / Political Geography 44 (2015) 29e39

provision in cases where municipal office was held by a different party to state or national office. Furthermore, political cooperation across jurisdictions may also be important when considering issues of revenue sharing and policy coordination. Conflicts between state and municipal parties, both in Latin America and elsewhere, are exacerbated by the challenges of fiscal decentralisation, particularly where fiscal transfers are disbursed selectively in order to weaken support for opposition parties (Finan, 2004; Montero, 2004; Resnick, 2011). Moreover, while broad left-centre-right categorisations might matter more than specific party affiliation regarding political cooperation and conflict, party factionalism within broader political movements seems important in explaining social violence (Fox & Hoelscher, 2012). A fifth hypothesis suggests: Hypothesis 5. Homicide rates will be lower in municipalities where state and municipal office holders are from the same political party. This framework does not suggest a simple link between unconsolidated democratic contexts and social violence. Clearly, several democratising countries or areas within them impartially administer the rule of law and curtail violence. Instead, certain political dynamics in contexts of limited oversight may encourage institutional capture and weakness, repression, clientelism, or alliance building that can undermine public security. Viewed as a form of politics of survival in a democratic context, the framework highlights how opportunities for using violence can be created. While not providing causal evidence, the framework here does provide possible causal narratives by which unconsolidated polities may be associated with greater rates of violence; and the estimation strategy supports these with empirical evidence. Democratisation and violence in Brazil Brazil is yet to fully consolidate as a democracy (e.g. see Codato, 2006; Linz & Stepan, 1996; Power, 2010). While electoral democracy is firmly established, political corruption at local levels continues to be a considerable problem; police are involved in extra-judicial killings and links with illegal actors; and elite political relationships continue to shape policy decisions. This assertion may be challenged, principally given the robust electoral aspects of its polity. Yet it is difficult to convincingly argue, quantitatively or qualitatively, that Brazil has fully consolidated its democratic institutions as have Western liberal democracies. First, Brazil is revealed as a partial democracy using quantitative measures of its polity assessing electoral competition and political participation (Goldstone et al., 2010). While Brazil's elections and executive recruitment are transparent, competitive and democratic, the openness and competitiveness of political participation is considered to be in a transitional state. Instead of a fully competitive political process where ‘ruling groups and coalitions regularly, voluntarily transfer central power to competing groups’ and ‘competition among groups seldom involves coercion or disruption’, Brazil's democracy comprises of institutions that are ‘accommodative of competing, parochial interests but have not fully linked parochial with broader, general interests… (and) sectarian and secular interest groups coexist’ (Marshall, Jaggers, & Gurr, 2010: 27). While Brazilian electoral institutions extend universal electoral rights, political actors maintain parochial interests and factionally compete for power; and other institutions curtail the legal rights of certain groups. Secondly, qualitative arguments also support the limitations of Brazil's democratic consolidation. The ‘paradox of democracy in Brazil’ (Pinheiro 2002: 113) highlights disjunctions between highly developed electoral democracy, and limits of democratic citizenship (Caldeira & Holston, 1999; Holston, 2009a, 2009b). Furthermore, federalism in Brazil has in some ways limited democratic

33

consolidation by encouraging competition for control of local resources, centralisation of power subnationally, and personalisation of political relationships between candidates and electorates (Arias, 2007; Arias & Goldstein, 2010). While the empirical literature on homicide in Brazil has rightfully emphasised inequality and economic exclusion, large youth populations, the emergence of the drug trade, small arms availability and organised crime as conduits for violence (Adorno, 2013; ~o, 2003; Goertzel Ceccato, Haining, & Kahn, 2007; Cerqueira & Loba
brega Júnior, 2010; Peres et al., 2011; & Kahn, 2009; Misse 2010; No Waiselfisz, 2011), qualitative evidence also suggests the political institutional context as important. Wacquant (2008) notes that following the end of military rule in Brazil, democratisation and institutional breakdown coincided with substantial growth in social violence, underpinned by the failure of new third wave democrats to establish a State apparatus capable of impartially applying the rule of law. Others highlight how rates of crime and violence are influenced by self-interested actions of political and economic elites and the inequalities in institutions governing law and order (Caldeira & Holston, 1999), a finding that resonates across Latin America (Jütersonke, 2007; Moser & McIlwaine, 2006). These trends can be observed in both national (Fig. 2) and subnational (Fig. 3) rates of homicide. In Fig. 2, homicide rates increased from 1979 to 1985 during the final years of the military dictatorship yet spiked in the late 1980s and early 1990s during the first years of Brazilian democracy. Fig. 3 shows homicide rates between 1980 and 2010 for 4 states in Brazil: two in the South-East ~o Paulo) and two in the North-East (Alagoas (Rio de Janeiro & Sa and Pernambuco). Importantly, during the military regime, violence rates were highly uniform in all states. Following democratisation, homicide rates diverged substantially between states, with the general upward trend slowing following Lula's election and introduction of redistributive policies. While a stylised example, it highlights how Brazil's return to democracy coincided with substantial transformations in violence with sizeable subnational variation. This underscores how focussing on subnational political institutional dynamics may be valuable to understand how democratisation and social violence co-occur. Empirical analysis Hypotheses are tested using annual time-series data for Brazil's 5564 municipalities for the years 1997e2010, where complete election data is publically available. The dependent variable measuring social violence is the natural log of the number of homicides per 100,000 population by municipality of occurrence.3 Homicide data is gathered from the Brazilian Ministry of Health's

Fig. 2. Homicide rates in Brazil, 1979e2009.

34

K. Hoelscher / Political Geography 44 (2015) 29e39

Fig. 3. Homicide rates in four Brazilian states, 1980e2010.

~o sobre Mortalidade (SIM 2012),4 and populaSistema de Informaça tion data from the Instituto Brasileiro de Geografia e Estatística (IBGE 2012), which measures actual population in census years and population estimates in inter-censal years. Political variables of interest are constructed based on data from state and municipal elections between 1994 and 2010, collected from the Tribunal Superior Eleitoral (TSE., 2012). Several variables were created. Political competition in Hypothesis 1 is operationalised in three ways: first, using the percentage share of the vote achieved by the winning party; second, the margin of victory, square root transformed to correct for normality; and third, using a dummy variable taking a value of 1 if the winning margin was less than 5% of the total vote. A lower percentage of the total vote, lower margin of victory and margin of under 5 percent are taken to indicate contexts of greater political competition over the following political cycle.5 Values are coded for the year following the election when the winning party takes office, and subsequent years until the next election.6 To test Hypothesis 2, a dummy variable is included taking the value of 1 for municipal election years. Concentration of power is measured in Hypothesis 3 using a dummy variable with a value of 1 when the party in power held office in the previous election cycle.7 To integrate hypotheses linking competition and concentration of power with social violence, Hypothesis 4 assesses non-linear effects. This is measured by including squared terms for both the share of vote to the winning party, and margin of victory variables. Finally, Hypothesis 5 measuring constraints on cooperation is operationalised using a dummy variable where a value of 1 indicates the same political party holds municipal and state office in a given year. Control variables are collected from IBGE (2012), and represent those commonly used in cross-national criminology literatures. Measures for total municipal population (logged), percentage of the municipal population living in urban areas, and youth bulge e the population aged 15e24 as a percentage of the population aged 15 and over e measure important demographic factors and are calculated based on census data and inter-censal estimates. Average monthly household-level income per person in Brazilian Reais captures motive arguments about engaging in criminal violence. Household income proxies local levels of poverty or economic hardship, and correlates highly with other measures of development such as the literacy rate and the Infant Mortality Rate.

Inequality is measured using the GINI coefficient, with data collected in decennial census years in 1990, 2000 and 2010, with inter-censal estimates linearly interpolated.8 Additionally, a lagged dependent variable is included to control for temporal dependence and inertia effects in rates of violence. Descriptive statistics are shown in Table 1. Regression analysis for panel data was performed with Stata 13, using random effects and robust standard errors clustered at the unit of analysis e the municipality. Models include dummy variables for Brazil's 26 states and 1 federal district to account for unit heterogeneity and unobserved variable biases at the state level,9 such as endogenous institutional or developmental characteristics indicative of a Federal system; and the diverse regional attributes in

Table 1 Descriptive statistics. Variable Dependent variable Homicide rate Homicide rate (ln) Political variables Share of vote to party holding local office Share of vote to party holding local office (sq) Margin of victory Margin of victory (sqrt) Margin of victory (sq) Margin of victory less than five percent Local election year Incumbent party holding local office Same party holding state and local office Control variables Total population Total population (ln) Percent urban Youth bulge Average household income per person GINI

Obs.

Mean

Std. Dev.

Min

Max

74,108 74,108

12.53 1.63

17.72 1.53

0.00 0.00

297.77 5.70

77,236

0.55

0.12

0.23

1.00

77,236

0.32

0.16

0.05

1.00

76,782 76,782 76,782 76,782

0.16 0.36 0.05 0.25

0.16 0.18 0.11 0.43

0.00 0.00 0.00 0

1.00 1.00 1.00 1

78,288 77,032

0.21 0.22

0.41 0.41

0 0

1 1

78,288

0.24

0.43

0

1

77,714 77,714 75,717 77,652 77,896

32,012 9.37 0.61 0.28 389.96

191,731 1.13 0.23 0.05 218.17

711 6.57 0.02 0.12 42.81

11,200,000 16.24 1.00 0.52 3468.20

77,896

0.53

0.06

0.28

0.88

K. Hoelscher / Political Geography 44 (2015) 29e39 Table 2 Structural correlates of social violence in Brazil, 1997e2010. Dependent variable ¼ homicide rate (ln)

Variables

(1) Homicide rate (ln) (t1) Total population (ln) Percent urban Youth bulge Household income per capita Inequality (GINI) Constant State dummies Observations Number of municipalities R-squared (Overall)

(2)

0.235*** 0.00557 0.410*** 0.00863 0.326*** 0.04 0.00241 0.00756 6.03E-05 4.90E-05 0.689*** 0.114 3.910*** 0.115 Y 67,495 5300 0.32

0.320*** 0.00593 0.365*** 0.00802 0.336*** 0.0383 0.00381 0.00698 0.000101*** 3.86E-05 1.114*** 0.117 3.159*** 0.0772 N 67,495 5300 0.28

Robust standard errors clustered on municipalities in italics. ***p < 0.01, **p < 0.05, *p < 0.1.

Share of vote to party holding local office Margin of victory (sqrt) Margin of victory less than five percent Local election year Incumbent party holding local office Share of vote to party holding local office (sq) Margin of victory

0.257*** 0.0508

(3)

Margin of victory (sq) Same party holding state and local office Constant

(4)

(5)

(6)

(7)

1.069*** 0.27 0.0645** 0.0324 0.0353*** 0.0131

0.0272*** 0.0104 0.132*** 0.013

0.0274*** 0.0104 0.124*** 0.013

0.0275*** 0.0104 0.123*** 0.013

0.0273*** 0.0104 0.133*** 0.013

0.0276*** 0.0104 0.124*** 0.0131

0.629*** 0.207

0.246*** 0.095 0.301** 0.143 0.0457*** 0.0482*** 0.0483*** 0.0451*** 0.0480*** 0.013 0.013 0.013 0.013 0.013

1.421*** 0.0711 Y 73,590 5299

State dummies Observations Number of municipalities R-squared 0.14 (Overall)

1.308*** 0.0662 Y 73,256 5298

1.277*** 0.0654 Y 73,256 5298

1.668*** 0.107 Y 73,590 5299

1.310*** 0.0659 Y 73,256 5298

0.13

0.13

0.14

0.13

Robust standard errors clustered on municipalities in italics. ***p < 0.01, **p < 0.05, *p < 0.1.

Variables

Dependent variable ¼ homicide rate (ln) (8)

(9)

(10)

(11)

(12)

Homicide rate (ln) (t1) Total population (ln) Percent urban

0.234*** 0.00557 0.409*** 0.00866 0.330*** 0.0401 0.00303 0.00755 8.25e-05* 4.94E-05

0.235*** 0.00558 0.410*** 0.00865 0.333*** 0.0401 0.0033 0.00774 8.63e-05* 4.97E-05

0.235*** 0.00558 0.410*** 0.00865 0.333*** 0.0401 0.00348 0.00778 8.73e-05* 4.98E-05

0.234*** 0.00557 0.408*** 0.00865 0.327*** 0.0401 0.00319 0.00756 8.35e-05* 4.93E-05

0.235*** 0.00558 0.411*** 0.00864 0.333*** 0.0402 0.00337 0.00774 8.74e-05* 4.97E-05

0.723*** 0.115

0.726*** 0.115

0.718*** 0.114 0.878*** 0.24

0.726*** 0.115

0.014 0.0118 0.0726*** 0.0123

0.014 0.0118 0.0683*** 0.0123

Youth bulge

Table 3 Political correlates of social violence in Brazil, 1997e2010. Dependent variable ¼ homicide rate (ln)

Demographic aspects of municipalities are also important. Greater rates of social violence occur in more populous and urbanised municipalities, while contrary to common perceptions and findings in the literature (Andrade & Lisboa, 2000), the size of youth cohorts plays little role in explaining violence. In this simple model, motive to engage in violence appears more important than opportunity in explaining rates of homicide. Findings suggest lower per capita household incomes are only significantly related to homicide rates without state controls (model 2), while inequality is robust in both models. Table 3 shows results for political variables of interest, and Table 4 shows the same models including the structural control variables from Table 2. Broadly, results indicate that several political factors are significantly related to homicide rates. First, political competition appears related to violence, with each of the 3 measures used to test hypotheses 1 significantly associated with homicide rates. Both in models including only political variables (Table 3), and in full political-structural models (Table 4), lower winning party vote shares (models 3&8), lower winning margins

Table 4 Structural and political correlates of social violence in Brazil, 1997e2010.

a large nation such as in Brazil.10 Table 2 first displays a simple model with common structural covariates of homicide rates. Results for these control variables across the different specifications appear largely consistent, and generally in line with findings in the literature. There is a clear inertia effect of social violence, with past rates of homicide strongly predicting current violence.

Variables

35

Household income per capita Inequality (GINI) 0.716*** 0.114 Share of vote to 0.134*** party holding 0.0452 local office Margin of victory (sqrt) Margin of victory less than five percent Local election 0.0139 year 0.0118 Incumbent party 0.0712*** 0.0123 holding local office Share of vote to party holding local office (sq) Margin of victory

0.0536* 0.0286 0.0301** 0.0117 0.0137 0.0118 0.0686*** 0.0123

0.0138 0.0118 0.0677*** 0.0123

0.581*** 0.185

0.211** 0.085 Margin of victory 0.263** (sq) 0.129 0.0455*** 0.0475*** 0.0475*** 0.0446*** 0.0473*** Same party 0.012 0.012 0.012 0.012 0.012 holding state and local office Constant 3.028*** 3.100*** 3.128*** 2.797*** 3.102*** 0.0925 0.0878 0.0871 0.118 0.0873 State dummies Y Y Y Y Y Observations 67,373 67,136 67,136 67,373 67,136 Number of 5294 5293 5293 5294 5293 municipalities R-squared 0.33 0.32 0.32 0.33 0.32 (Overall) Robust standard errors clustered on municipalities in italics. ***p < 0.01, **p < 0.05, *p < 0.1.

36

K. Hoelscher / Political Geography 44 (2015) 29e39

(models 4&9), and margins of under five percentage points (models 5&10) are all associated with greater violence in municipalities. This suggests that where politics is more competitive, rates of violence in proceeding years are significantly greater. Results for Hypothesis 2 are mixed. While election years are significant when control variables are absent (Table 3), the inclusion of controls in Table Four renders them insignificant. Therefore, years in which local elections are held are not associated with greater rates of social violence (Hypothesis 2) when controlling for socio-economic and demographic factors. The effect of concentration of power is assessed using an incumbency dummy in Hypothesis 3. As expected, the variable is positive and significant across models, both with (Table 4) and without (Table 3) control variables, suggesting that the concentration of political power may create conditions which can increase violence or undermine public security. Moreover, if political elites become entrenched, they may be less inclined to rescind their authority, and when legitimate competition emerges, the increased threat of the loss of embedded political power may result in greater resistance through violent means. Hypothesis 4 assesses non-linear effects of competition for power by including standard and squared terms for both the share of the vote to the winning party (models 6 & 11) and the margin of victory (models 7 & 12). Results show that for both variables, original terms are significant and negative, while squared terms are significant and positive. For both variables this indicates a significant U-shaped relationship. While closely competitive political contexts (low winner vote shares and low margins of victory) are associated with greater homicide rates, so too are contexts indicative of limited credible political competition (high winner vote shares and high margins of victory). Taken together with findings showing that close electoral margins and incumbency are related to higher homicide rates, this offers support that both competition for, and concentration of, power are both related to greater violence. Rather than being contradictory, competition for and consolidation of power may offer complementary insights as to how the political context might incentivise violence. Finally, hypothesis five assesses how constraints on cooperation between political parties may create conditions for violence to emerge. All models in Tables 3 and 4 indicate that homicide rates are significantly lower where state and municipal office holders are from the same party. This suggests subnational cooperation or coordination between political parties might discourage violence; or that limited coordination or constrained collaboration between parties can create conditions where violence is incentivised. Where the same party is in power at state and municipal levels, there may be greater cooperation around shared policy goals, or improved efficiency in coordination between state and municipal office. Examples may include prioritising resources for law enforcement or social programs to municipalities where the mayor is of the same party, or creating party-based networks of personalised support. This may plausibly enable better provisioning of security and play a role in lowering rates of municipal-level violence. Alternately, where parties differ, party politics may see governors overtly undermine opposition mayors by underfunding social or security programs, which may indirectly weaken the rule of law. Substantive impacts of variables of interest can be gauged by calculating marginal effects holding variables at their means. Effects reported here are from models including both political variables of interest and controls (Table 4). Given that the unit of analysis is the municipality, of which there are over 5000 in the dataset, marginal effects for all variables are understandably small. This should not necessarily be regarded as a problem, and presented as percentages or aggregated to the national level can reveal meaningful effects.

Presented first are marginal effects of dummies measuring concentration of power and constraints on cooperation (Model 10). For concentration of power, marginal effects show that incumbent parties holding office is associated with an 8.7% increase in homicide rates (an additional 0.36 homicides per 100,000). Measuring constraints on cooperation, state and local office being shared by the same party is associated with municipal homicide rates being 5.7% lower (0.24 per 100,000 fewer homicides).11 Descriptive statistics reveal that 22% of municipality-years have incumbent governments; and 24% show municipal parties are the same as those in state government. Aggregating to the national level and accounting for population in these municipalities, incumbent parties being in office was associated with 2015 additional homicides (144 per year) over the course of the 14 year sample between 1997 and 2010. Similarly, parties sharing state and municipal office were associated with 1351 fewer homicides (97 per year) for municipalities in the sample over the same time period. Similar observations can be made for the dummy variable assessing political competition. Electoral margins of less than 5 percent were associated with a 4% increase in homicide rates (0.16 per 100,000) compared to municipalities with winning margins over 5 percent. Aggregated to include the 25% of municipalities with sub-5% margins, this was associated with an additional 838 homicides over the sample period (60 per year). Results indicate a non-linear relationship between political competition and homicide rates. The marginal effects of share of the vote held by the winning party (Model 11) and margin of victory (Model 12) are graphically displayed in Figs. 4 and 5 respectively. Both graphs show how the relationship between logged homicide rates and respective measures of political competition vary according either to the share of the winning party's vote or the margin of victory. Using these marginal effects, examples can be constructed that relate to both the framework and findings. Municipalities where parties face greater competition (e.g. winning vote share of 35%) can be compared to those with a more solid majority (e.g. 60%). All else equal, municipalities with lower competition (60% voteshare) observe homicide rates 9.6% lower (0.43 per 100,000) than more competitive political contexts (35% voteshare). Similarly the graph indicates that passing a threshold of around 70% of the total vote sees an increased likelihood of violence. The same can be done for winning margins. Margins of victory one standard deviation below the mean (victory of <1%) are

Fig. 4. Marginal effects on ln (homicide rate) of share of vote to party holding office.

K. Hoelscher / Political Geography 44 (2015) 29e39

Fig. 5. Marginal effects on ln (homicide rate) of margin of victory.

associated with a homicide rates 3.3% higher (0.14 per 100,000) than mean winning margins (victory of 16%). It is also useful to compare municipalities with narrow margins of victory (e.g. 2%); those with comfortable yet competitive margins (e.g. 30%); and those with an absence of competition (e.g. 90%). In this example, the most competitive municipalities (2% margin) on average see homicide rates 4.3% greater (0.18 per 100,000) than those with healthy competition (30% margin); while uncompetitive municipalities (90% margin) are associated with homicide rates 8.1% greater (0.33 per 100,000) than where there is healthy competition (30% margin). Taken together, these findings illustrate findings from Hypothesis 4 that both highly competitive and insufficiently competitive political contexts are associated with greater rates of municipal violence. These substantive impacts can be compared with those of important structural variables. For example, looking at the GINIcoefficient, the marginal effect of a one standard deviation increase (0.06) from the sample mean (0.53) is associated with an 5.5% increase in rates of homicide (0.23 per 100,000). Similarly, an increase in household income per person of one standard deviation (218 Reais) from the sample mean (389 Reais) is associated with a 2.3% decrease in rates of homicide (0.1 per 100,000). Taken as a whole, municipal-level marginal effects are similar in magnitude for both political variables and important structural factors, suggesting that in addition to the importance of structural conditions, political factors have important independent effects. Several diagnostic and post-estimation techniques were performed to assess the robustness and sensitivity of results. Assumptions for normality, skewness and kurtosis are upheld except when state dummy variables are added, which are inherently nonnormally distributed. Multicollinearity is not a concern as variance inflation factor (VIF) scores are within acceptable ranges. While autocorrelation and heteroskedascity are concerns for time-series panel datasets, and several diagnostic tests assessing omitted variable bias are unavailable using a panel data setup, state-level dummies and robust municipal clustered standard errors largely account for these concerns. Discussion and conclusion This article advances a novel theoretical argument that the political institutional context is important in explaining rates of violence. Findings show empirical relationships between specific subnational political conditions and rates of social violence,

37

suggesting possible mechanisms where political opportunity structures may incentivise violence in hybrid democracies. Results also show that key national-level findings from comparative criminology literatures apply when observing within-country dynamics. Municipalities which experience greater rates of lethal violence are more likely to be unequal, poor, populous and urban, and it is useful to recognise these relationships also apply in subnational territories as they do at the national level. Some interesting puzzles emerge from the findings. First, while municipalities with more competitive local elections see greater rates of lethal violence, elections themselves are unrelated to homicides despite being obvious signals of political competition. This may be due to increased priorities on law and order during political campaigning; or greater scrutiny on the behaviour of political elites during election years. Interpreted in light of the theoretical framework, the spotlight of an election might create a more ‘legitimate’ political or electoral context, while competition in the absence of electoral-year oversights or checks on elite behaviour might instead increase opportunities for violence. Secondly, findings indicate homicide rates are greater both where electoral processes are defined by high or low levels of competition, and where political power is highly concentrated. These results may seem somewhat at odds, but competition for, and concentration of, power are not necessarily contradictory phenomena. Even if considered different concepts, they can still be incorporated into the framework as political competition and concentration of power might lead to different manifestations of violence with separate causal pathways. Fig. 1 shows that political competition might be more likely to result in direct use of institutions such as the police or judiciary to deploy violence; while consolidation of power through incumbency or the absence of competition may lead to institutional capture or dysfunction, with violence increasing due to weakened rule of law. Both pathways are consistent with theories that democratization exacerbates different forms of violence that have been embedded in perverse processes of state formation in Latin America (Pearce, 2010). Caveats must be made regarding how the statistical strategy can infer causal inference from time-series data. Results indicate robust associations between the context political actors operate in and rates of homicide, with an institutional argument providing a possible causal narrative. Despite this, alternative e though less likely e explanations may exist. Rather than political opportunity structures generating conditions for violence, more violent societies might constrain available actions of political actors, with weakened institutional outcomes the result. Politicians may respond to pervasive violence by using repressive policing in attempts to contain violence, which may again in turn see violence increase. These linkages are difficult to tease out using the current empirical strategy. Mitigating this, political variables in the study are unable or unlikely to be causally affected by rates of violence. Election years are fixed and exogenous. Neither winning margins nor state-municipal party configurations would be affected by rates of violence in a predictable way. Moreover, violence if anything may decrease likelihood of incumbency and re-election as high rates of violence may lead to officials voted from office. Given this and the theory-building nature of this article, some restraint is called for when interpreting results.12 This article emphasises three key points. First, it identifies that in addition to social, economic and demographic drivers, social violence is associated with political factors. Second, it underlines the potential importance of local political contexts within countries, and the need to go beyond rudimentary national-level classifications of polities on an autocracy-to-democracy continuum. Finally, while presenting empirical evidence about the Brazilian case, the article makes a broader statement about ways in which

38

K. Hoelscher / Political Geography 44 (2015) 29e39

semi-democratic political-institutional contexts can create conditions where violence becomes a viable tactic in solving disputes or gaining advantage in personal or political realms. Understanding the theory and evidence presented here in a broader context may be useful to move debates beyond discussions of if semi-democratic regimes are more likely to experience social violence. Instead debates may instead consider where, and under what conditions, developing democracies might experience violence and how different policy responses might address this. Future research could develop these ideas in several ways. First, by challenging and refining the possible causal processes pointed to in this article. Particularly interesting would be examining relationships between political parties at different levels of government to understand how cooperation might improve security, and how hostility or limits to political coordination may hinder it. Two cases from Brazil suggest these party linkages may be important in policy implementation, particularly in major cities. Lessing (2012) has shown how state-municipal political coalitions aided reform of Rio de Janeiro's security situation. In Recife, the capital of the North-Eastern state of Pernambuco, close party ties between state and municipal governments have seen integrated violence reduction policies emerge as key way of reducing violence in the city (Hoelscher, 2014). Second, narratives linking political conditions in hybrid democracies and rates of violence should be explored in comparative case-based research to understand how these processes may apply. Third, research should prioritise understanding and supporting the institutional conditions that lead to violence reduction. In particular, attention should be given to lessons learned from cases where political or institutional reform has led to violence reduction, and how these approaches may be scaled up and out. A fruitful future agenda would therefore seek to understand how, and under what conditions political actors and institutions contribute to insecure societies; how the political context modifies the effects of structural factors that support high rates of violence; and specific ways institutional consolidation, legitimacy and accountability can reinforce the social contract and mitigate violence. Expanding the empirical body of knowledge about how political and institutional contexts matter is important in addressing problems of social violence in current and future transitions to democracy. Endnotes 1

Despite these distinctions, it is theoretically and empirically difficult to tease apart direct violence by the state and indirect violence due to state weakness. These distinctions are not tested here but highlighted as possible alternate pathways relating political factors with social violence. 2 I thank an anonymous reviewer for highlighting this point. 3 To account for zero values, logged homicide rates take the form ln (1 þ y). 4 The category Aggressoes (X85eY09), from the 10th International Classification of Diseases (ICD-10), is used, and includes intentional and unintentional deaths from aggressive interpersonal interactions. 5 In municipalities with populations over 200,000, a second-round vote between the top two candidates occurs absent of a first round majority. In these cases data from the runoff vote is used. 6 Elections in Brazil are held every four years in October. Winners take office the following year on 1 January. State and national elections are held together, and occur 2 years before/after municipal elections. As an example, a party winning an election in October 1996 would take office on 1 January 1997, and relevant values would be coded for years until the next election, namely 1997e2000. 7 Concentration of power was also measured using a variable tallying the number of years the party in power had held office. Results are not reported here, but were virtually identical in direction and significance as the incumbency variable. 8 For some calculated variables based on different data sources, a minor number of implausible values arose, such as where urban populations exceeded total populations. Though errors were rare, values for some variables are bound at the 99th percentile to control for ‘impossible’ values. Results are virtually identical using either original or 99th percentile values. 9 There are cases for both fixed- or random-effects in panel-data analysis but choices should be theoretically driven (Clark & Linzer, 2012). A municipal-level

fixed-effects approach would unnecessarily restrict predictive power of estimations given the large number of units of analysis. Hence a random-effects approach with state-level dummies is preferred to control for subnational characteristics not captured by these models, such as the capacity of the police, investment in public security and institutional histories. 10 Controls for temporal effects are excluded due to collinearity between year dummies and several variables in the analysis. While violence may in some way be auto-correlated with certain years in the sample there is no clear theoretical reason how this may operate. Importantly, models run with yearly dummies (not shown here) are largely similar to those presented. 11 When reporting, the marginal effect on ln (homicide) was converted from the ln (1 þ y) format to actual homicide rates using the inverse transformation exp (y)1. All reported marginal effects are highly similar with or without state dummies. 12 While an instrumental-variable approach is advantageous in addressing endogeneity, it is unclear how such an approach would be employed in this context.

References Adorno, S. (2013). Democracy in progress in contemporary Brazil: corruption, organized crime, violence and new paths to the rule of law. International Journal of Criminology and Sociology, 2(1), 409e425. Andrade, M. V., & Lisboa, M. B. (2000). Desesperança de Vida: Homicídio em Minas ~o Paulo: 1981 a 1997. In R. Henriques (Ed.), DesiGerais, Rio de Janeiro e Sa gualdade e Pobreza no Brasil (pp. 347e384). Rio de Janeiro, Brazil: IPEA. Arias, E. D. (2007). Trouble en Route: Drug Trafficking and Clientelism in Rio de Janeiro Shantytowns. Qualitative Sociology, 30(1), 427e445. Arias, E. D., & Goldstein, D. (Eds.). (2010). Violent democracy in Latin America: Toward an interdisciplinary reconceptualization. Durham, NC, USA: Duke University Press. Bates, R. H. (2001). Prosperity and violence: The political economic of development. New York, NY, USA: W.W. Norton & Company. Bates, R. H. (2008). State failure. Annual Review of Political Science, 11(1), 1e12. Buhaug, H., & Rød, J. K. (2006). Local determinants of African civil wars, 1970e2001. Political Geography, 25(3), 315e335. Caldeira, T. P., & Holston, J. (1999). Democracy and violence in Brazil. Comparative Studies in Society and History, 41(4), 691e729. ~o Paulo, Ceccato, V., Haining, R., & Kahn, T. (2007). The geography of homicide in Sa Brazil. Environment and Planning A, 39(7), 1632e1653. ~o ~es socioecono ^micas, polícia e produça Cerqueira, D., & Lob~ ao, W. (2003). Condiço criminal. Rio de Janiero, Brazil: IPEA. Chesnai, J. (1981). Histoire de la Violence en Occident de 1800 a nos Jours. Paris, France: Pluriel. Clark, T. S., & Linzer, D. A. (2012). Should I use fixed or random effects? St Louis, MO, USA: Society for Political Methodology. Working Paper 1315. Codato, A. N. (Ed.). (2006). Political transition and democratic consolidation: Studies on contemporary Brazil (pp. 1e26). New York, NY, USA: Nova. Cruz, J. M. (2011). Criminal violence and democratization in Central America: the survival of the violent state. Latin American Politics and Society, 53(4), 1e33. Dunning, T. (2011). Fighting and voting: violent conflict and electoral politics. Journal of Conflict Resolution, 55(3), 327e339. Eisner, M. (2001). Modernization, self-control and lethal violence: the long-term dynamics of European homicide rates in theoretical perspective. British Journal of Criminology, 41(4), 618e638. Eisner, M. (2008). Modernity strikes back? A historical perspective on the latest increase in interpersonal violence (1960e1990). International Journal of Conflict and Violence, 2(2), 288e316. Fajnzylber, P., Lederman, D., & Loayza, N. (2002). Inequality and violent crime. Journal of Law and Economics, 45(1), 1e40. Ferraz, C., & Finan, F. (2011). Electoral accountability and corruption: evidence from the audits of local governments. American Economic Review, 101(4), 1274e1311. Finan, F. (2004). Political patronage and local development: A Brazilian case study. ARE working paper. Berkeley, CA, USA: UC Berkeley. Fox, S., & Hoelscher, K. (2012). Political order, development and social violence. Journal of Peace Research, 49(3), 431e444. Ghosh, A. (2006). Electoral cycles in crime in a developing country: Evidence from the Indian states. SSRN. Working Paper 910054. Available at SSRN: http://ssrn.com/ abstract¼910054. Gibson, E. L. (2010). Politics of the periphery: an introduction to subnational authoritarianism and democratization in Latin America. Journal of Politics in Latin America, 2(2), 3e12. Goertzel, T., & Kahn, T. (2009). The great S~ ao Paulo homicide drop. Homicide Studies, 13(4), 398e410. Goldstone, J. A., Bates, R. H., Epstein, D. L., Gurr, T. R., Lustik, M. B., Marshall, M. G., et al. (2010). A global model for forecasting political instability. American Journal of Political Science, 54(1), 190e208.
rrez-Romero, R. (2012). An inquiry into the use of illegal electoral practices and Gutie effects of political violence. CSAE Working paper WPS/2012e16. Oxford, UK: Centre for the Study of African Economies. Hegre, H., Ellingsen, T., Gates, S., & Gleditsch, N. P. (2001). Toward a democratic civil peace? Democracy, political change, and civil war, 1816e1992. American Political Science Review, 95(1), 33e48. Hegre, H., & Sambanis, N. (2006). Sensitivity analysis of empirical results on civil war onset. Journal of Conflict Resolution, 50(4), 508e535.

K. Hoelscher / Political Geography 44 (2015) 29e39 Hoelscher, K. (2014). Institutional reform and violence reduction in Pernambuco, Brazil. Working paper. Oslo, Norway: PRIO. Holston, J. (2009a). Dangerous spaces of citizenship: gang talk, rights talk and rule of law in Brazil. Planning Theory, 8(1), 12e31. Holston, J. (2009b). Insurgent citizenship: Disjunctions of democracy and modernity in Brazil. Princeton, USA: Princeton University Press.
ficos. Data IBGE. (2012). Instituto Brasileiro de Geografia e Estatística: Censos Demogra available at tabnet.datasus.gov.br ipeadata.gov.br.
n, F. (2011). Violence research in Latin America and Imbusch, P., Misse, M., & Carrio the Caribbean: a literature review. International Journal of Conflict and Violence, 5(1), 87e154. Jütersonke, O., Krause, K., & Muggah, R. (2007). Guns in the city: The changing landscapes of armed violence, small arms survey 2007. Cambridge, UK: Cambridge University Press. Karstedt, S. (2009). Democratization and violence: European and international perspectives. In S. Body-Gendrot, & P. Spierenburg (Eds.), Violence in Europe: Historical and contemporary perspectives (pp. 205e225). New York, NY, USA: Springer. Karstedt, S., & LaFree, G. (2006). Democracy, crime, and justice. Annals of the American Academy of Political and Social Science, 605(1), 6e23. Koonings, K. (2001). Armed actors, violence and democracy in Latin America in the 1990s: Introductory notes. Bulletin of Latin American Research, 20(4), 401e408. Krahn, H., Hartnagel, T. F., & Gartrell, J. W. (1986). Income inequality and homicide rates: cross-national data and criminological theories. Criminology, 24(2), 269e295. Krivo, L. J., Peterson, R. D., & Kuhl, D. C. (2009). Segregation, racial structure, and neighborhood violent crime. American Journal of Sociology, 114(6), 1765e1802. LaFree, G. (1998). Losing legitimacy: Street crime and the decline of social institutions in America. Oxford, UK: Westview Press. LaFree, G., & Drass, K. A. (2002). Counting crime booms among nations: evidence for homicide victimization rates, 1956 to 1998. Criminology, 40(4), 769e800. LaFree, G., & Tselsoni, A. (2006). Democracy and crime: a multilevel analysis of homicide trends in forty-four countries, 1950e2000. Annals of the American Academy of Political and Social Science, 605(1), 26e49.
ndez, A. M., & Loayza, N. (2002). Violent crime: does social Lederman, D., Mene capital matter? Economic Development and Cultural Change, 50(3), 509e539. Lessing, B. (2012). The logic of violence in criminal war: Cartel state conflict in Mexico, Colombia, and Brazil. Phd dissertation. Berkeley, CA, USA: University of California Berkeley. Levitt, S. D. (1997). Using electoral cycles in police hiring to estimate the effect on crime. The American Economic Review, 87(3), 270e290. Like, T. Z. (2011). Urban inequality and racial differences in risk for violent victimization. Crime & Delinquency, 57(3), 432e457. Linz, J., & Stepan, A. (Eds.). (1996). Problems of democratic transition and consolidation: Southern Europe, South America and post-communist Europe. Baltimore, MD, USA: Johns Hopkins Press. Mansfield, E., & Snyder, J. (1995). Democratization and war. Foreign Affairs, 74(3), 79e97. Marshall, M. G., Jaggers, K., & Gurr, T. R. (2010). Polity IV project dataset users' manual. Vienna, VA, USA: Center for Systemic Peace. Marshall, I. H., Marshall, C. E., & Ren, L. (2009). Mixed method measurement of homicide events in comparative research: an illustration of the potential of qualitative comparative analysis. International Journal of Comparative and Applied Criminal Justice, 33(2), 273e307. Meloni, O. (2012). Is there an electoral-motivated crime rate cycle? Evidence from Argentina. Munich, Germany: MPRA. Working Paper 40177. Misse, M. (2010). Social accumulation of violence in Rio de Janeiro and Brazil: some reflections. Co-Herencia, 7(13), 19e40. Montero, A. P. (2004). Competitive federalism and distributive conflict in democratic Brazil. In M. Font, & A. Spanakos (Eds.), Reforming Brazil (pp. 137e159). New York, NY, USA: Lexington. Montero, A. P. (2010). Trading spaces: The endogenous dynamics of subnational authoritarianism in Brazil. Paper presented at the Congress of the Latin American Studies Association, Toronto, Canada October 6e9, 2010, available at http:// people.carleton.edu/~amontero/Trading Spaces.pdf. Moser, C., & McIlwaine, C. (2006). Latin American urban violence as a development concern: towards a framework for violence reduction. World Development, 34(1), 89e112. Muggah, R. (2012). Researching the urban dilemma: Urbanization, poverty and violence. Ottawa, Canada: International Development Research Centre. Myers, D. J., & Dietz, H. A. (2002). Capital city politics in Latin America: Democratization and empowerment. Boulder, CO, USA: Lynne Rienner.

39

Neumayer, E. (2003). Good policy can lower violent crime: evidence from a crossnational panel of homicide rates, 1980e97. Journal of Peace Research, 40(6), 619e640. Nivette, A. E. (2011). Cross-national predictors of crime: a meta-analysis. Homicide Studies, 15(2), 103e131. Nivette, A. E., & Eisner, M. (2013). Do legitimate polities have fewer homicides? A cross-national analysis. Homicide Studies, 17(1), 3e26.
brega, J. M., Júnior (2010). Os Homicídios no Brasil, no Nordeste e em Pernambuco: No ^mica e relaço ~es causais e políticas públicas. Phd dissertation. Recife, Brazil: Dina Universidade Federal de Pernambuco. North, D. (1991). Institutions. Journal of Economic Pespectives, 5(1), 97e112. North, D., Weingast, B., & Wallis, J. (2009). Violence and social orders: A conceptual framework for interpreting recorded human history. New York, NY, USA: Cambridge University Press. Ouimet, M. (2010). A world of homicides: the effect of economic development, income inequality, and excess infant mortality on the homicide rate for 165 countries in 2010. Homicide Studies, 16(3), 238e258. Pearce, J. (2010). Perverse state formation and securitized democracy in Latin America. Democratization, 17(2), 286e306. Peres, M., Vicentin, D., Nery, M., de Lima, R., de Souza, E., Cerda, M., et al. (2011). Decline in homicide rates in Sao Paulo, Brasil: a descriptive analysis. Revista Panamericana De Salud Publica (Pan American Journal of Public Health), 29(1), 17e26. Pinheiro, P. S. (2002). The Paradox of democracy in Brazil. The Brown Journal of World Affairs, 8(1), 113e122. Pinker, S. (2011). The better angels of our nature. London, UK: Penguin. Power, T. (2010). Brazilian democracy as a late bloomer: reevaluating the Regime in the Cardoso-Lula Era. Latin American Research Review, 45(4), 218e247. Pridemore, W., & Kim, S. (2008). Testing Durkheim in Russia: democratization and political change as threats to collective sentiments. Annals of the American Academy of Political and Social Science, 65(1), 82e103. Resnick, D. (2011). In the shadow of the city: Africa's urban poor in opposition strongholds. The Journal of Modern African Studies, 49(1), 141e166. Rodgers, D. (2009). Slum wars of the 21st century: gangs, mano dura, and the new urban geography of conflict in Central America. Development and Change, 40(5), 949e976. Schutte, S., & Weidmann, N. B. (2011). Diffusion patterns of violence in civil wars. Political Geography, 30(3), 143e152. ~es sobre Mortalidade, Dados preliminares de 2010. SIM. (2012). Sistema de Informaço Data available at tabnet.datasus.gov.br ipeadata.gov.br. Stamatel, J. P. (2009). Contributions of cross-national research to criminology at the beginning of the 21st century. In M. D. Krohn, A. J. Lizotte, & G. P. Hall (Eds.), Handbook on crime and deviance (pp. 3e22). London, UK: Springer. Stucky, T. (2003). Local politics and violent crime in U.S. cities. Criminology, 41(4), 1101e1136. Stucky, T., Heimer, K., & Lang, J. (2005). Partisan politics, electoral competition and imprisonment: an analysis of states over time. Criminology, 43(1), 43e76. Tcherni, M. (2011). Structural determinants of homicide: the big three. Journal of Quantitative Criminology, 27(4), 475e496. Tilly, C. (2003). The politics of collective violence. Cambridge, UK: Cambridge University Press. Tilly, C., & Tarrow, S. (2007). Contentious politics. Boulder, CO, USA: Paradigm.
rio de Dados Eleitorais. Data available TSE. (2012). Tribunal Superior Eleitoral: Reposito at ipeadata.gov.br. Verpoorten, M. (2012). Detecting hidden violence: the spatial distribution of excess mortality in Rwanda. Political Geography, 31(1), 44e56. Villareal, A. (2002). Political competition and violence in Mexico: heirarchical social control in local patronage structures. American Sociological Review, 67(4), 477e498. Wacquant, L. (2008). The militarization of urban marginality: lessons from the Brazilian metropolis. International Political Sociology, 2(1), 56e74. ~es da Viol^ Waiselfisz, J. (2011). Mapa da Viol^ encia 2012: Os Novos Padro encia Homicida ~o Paulo, Brazil: Instituto Sangari. no Brasil. Sa Wilkinson, S. (2004). Votes and violence: Electoral competition and ethnic riots in India. Cambridge: Cambridge University Press. World Bank. (2011a). Violence in the city: Understanding and supporting community responses to urban violence. Washington DC, USA: World Bank. World Bank. (2011b). World development report 2011: Conflict, crime and violence. Washington DC, USA: World Bank. Zaluar, A. (2004). The paradoxes of democratization and violence in Brazil. Paper ^ncia Brasil e Unia ~o Europe
ia Ampliada, UFRJ, Available presented at the Confere at http://www.brasiluniaoeuropeia.ufrj.br/en/pdfs/the_paradoxes_of_democra tization_and_violence_in_brazil.pdf. Zhukov, Y. M. (2012). Roads and the diffusion of insurgent violence: the logistics of conflict in Russia's North Caucasus. Political Geography, 31(3), 144e156.