On the Design of Inclusive Institutions in Mitigating

On the Design of Inclusive Institutions in Mitigating Political Violence: Evidence from Basque Municipalities Georgi Boichev University of Regina Job Market Paper September 14, 2015 Abstract This paper presents micro-level evidence indicating that the presence of relatively less inclusive political institutions increases political violence in ethnolinguistically-fractionalized jurisdictions. In the context of predominant expressive voting based on citizens ethnolinguistic identity, I find evidence that the lack of inclusive political institutions relative to the size of an ethnic group in a given jurisdiction generates targeted street violence. I use municipal-level data from the Basque autonomous community and employ a negative binomial panel fixed effects model. First, I show that the reciprocal of a council size in a proportional representation electoral system acts as an electoral threshold, in which small-sized councils may exceed its legislated counterpart. I find that a higher share of Basque speakers relative to the reciprocal of council size has a negative effect in the annual number of riots, an effect that is observed only in municipal pre-election years. In addition, by interacting the key regressor with the binary indicator for hung parliament, I find that the baseline effect is driven by the legislative channel but not by the executive channel of political representation. Key words: political violence, conflict, inclusive institutions, proportional electoral system, ethnic fractionalization JEL classification: D74, P16, P48 Postdoctoral Fellow, University of Regina, Canada, email: georgi.boichev@uregina.ca. I benefited from conference participants at the Silvaplana Conference organized by the European Journal of Political Economy, held July 25-29, 2015, the CEA Conference, held May 28th - May 31st. All remaining errors are mine. 1

1 Introduction Political violence arising in ethno-linguistically fractionalized societies poses significant social and economic costs (Gardeazabal, 2003). A key determinant of political violence is the political exclusion or the underrepresentation of ethnolinguistic minorities (Besley and Persson, 2011; Reynal-Querol, 2005) even in the presence of a strong rule of law and competitive elections. As policies for maintaining post-conflict peace include the social inclusion of ethnic minorities (Collier and Hoeffler, 2005), little empirical work is dedicated to understanding how political institutions are to be reformed to address this problem. While a PR electoral system is being argued to be in relative terms more inclusive (Reynal-Querol, 2005) with the participation of ethnolinguistic-based political parties as parliamentary factions and in a subset of those instances as partners in coalition governments (Alionescu, 2004), political violence in PR-based electoral systems is not uncommon. In this paper, I address two questions: Which electoral rules in a proportional electoral system contribute to the inclusion of ethnic minorities and also impact political violence? Which channels of parliamentary representation influence political violence? The first contribution of this paper is that I pinpoint a political institution belonging to the general class of PR electoral systems the size of an ethnic group relative to the electoral threshold that has the property of being more inclusive as its value increases. In addition, I provide empirical support that the more inclusive this institutions is, the more it mitigates targeted street violence taking the form of riots. The relevance of this inclusive institution is based on two pillars: predominant expressive voting that takes the form of ethnic groups casting ballots for ethnic-based political parties; the organized nature of targeted street violence that is used to achieve political objectives. In the context of predominant expressive voting, a higher threshold for representation relative to the size of an ethnic group is suggested to increase political violence as it reduces the likelihood for legislative representation of the 2

political parties competing for the votes of an ethnic group. In this respect, my paper relates to (Hillman et al., 2015) that expressive voting in ethnically-mixed regions reduces the nature of competitive elections but I suggest that this issue is aggravated by the presence of a high electoral threshold relative to the size of an ethnic group in a given jurisdiction. The second pillar of organized political violence used in the Basque Country to seek political objectives is suggested to be the underlying reason for the targeted riots of municipalities with less inclusive institutions in municipal pre-election years. To demonstrate the main result of my paper, I first show that the reciprocal of a council size is equivalent to an electoral threshold for representation in a proportional electoral system, measured as a share of the popular vote and hereafter referred to as a full seat. A smaller council size, found to induce behavioral responses such as strategic voting and party mergers (Fina and Folke, 2014), is suggested in my paper to increase political violence. A smaller ratio of an ethnic minority per full seat decreases the likelihood that an ethnic-based political party receives legislative representation. The institutional approach of my paper relates to a large literature studying the effects of quotas of parliamentary representations for females and ethnic minorities (de Paola et al., 2010, 2014; Pande, 2011; Bhavnani, 2009). While my paper is similar to a subset of this literature, where the quota for female/ethnic minorities candidates is imposed on the set of candidates run by a political party, it differs from this literature as ethnic-based political parties legislative representation depends on their ability to obtain a vote share at least as large as the legislated electoral threshold. The second contribution of my paper is that I distinguish between two channels legislative and executive of political representation of ethnic minorities through which lack of inclusion might influence riot-based street violence. The empirical support of my findings being in favour of the legislative channel but not of the executive channel sheds light on the political determinants of political violence. The legislative channel - parliamentary representation per 3

se - provides access to legislative representation (Alionescu, 2004), immunity from prosecution (Chang, 2005; Chang et al., 2010; Wigley, 2003), which are all important forms of legitimizing the political representatives of an ethnic minority. In contrast, the executive channel proxies for the political clout of a small-sized political party in a hung parliament that has a non-trivial ability to exercise oversight over government decisions, to draft legislation in committees, to participate in a coalition governments or to provide legislative support to a minority government on an issue-by-issue basis. The policy implication of my findings is that an electoral reform in a PR electoral system aiming to improve the political representation of ethnic minorities must set the electoral threshold for representation, legislated or implied by the council size, proportionately to the size of those ethnic minorities in a given electoral jurisdiction. This jurisdiction-specific approach intends to generate inclusive institutions that also preserve the competitive nature of elections even in contexts of predominant expressive voting along ethnolinguistic lines. The absence of an effect for the executive channel suggests that such policy reforms are likely to be neutral with respect to the frequency of hung parliament formation. This consideration is important as a lower electoral threshold for representation is suggested by (Boichev, 2015; Fina and Folke, 2014) to induce the more frequent formation of hung parliaments due to mechanical and behavioural responses. To address the questions posed in this paper, I use municipal-level data on street violence collected by (de la Calle Robles, 2007). Street violence is a planned in advance activity that resembles spontaneous political protests that turn into riots. In addition to this data source, I use electoral and census municipal-level data obtained from the Statistical Institute of the Basque Country that includes information on linguistic fragmentation and voting behaviour. This dataset offers two institutional features: exogenous variation in council size based on nationwide population thresholds, which imposes high thresholds for representations in small- 4

sized municipalities and voting for ethnic-based political parties. I use a negative binomial fixed effects panel data model as the baseline identification strategy to estimate the effect of minority representation on street violence. The key regressor of interest is the ratio of Basque speakers per full seat, which measures unpicked explanatory power by the exogenous variation in council size and that of the share of ethnic Basques. To distinguish between the suggested legislative and executive channels for representation of ethnic minorities, I interact the ratio of Basque speakers per full seat with the binary indicator for a hung parliament. The main findings indicate that less political representation for ethnic minorities, measured by the share of Basque speakers per full seat, increases political violence in pre-election years. I find that an increase in the share of Basque speakers per full seat by 1 unit on average increases the number of street violence incidents by 6 %. My findings point to the legislative channel to be driving the relationship between ethnic underrepresentation and political violence. This interpretation in support of the legislative channel of underrepresentation is based on the absence of an additional effect for the interacted term between the key regressor, the share of Basque speakers per full seat, with the binary indicator for a hung parliament. A range of robustness checks provide additional credence to the baseline results. In addition, I mitigate concerns concerning the exogeneity of the key explanatory variable. First, by estimating a panel data model from two electoral cycles, I find no evidence that pre-electoral political violence changes the ethnic composition of municipalities over time. In addition, by employing regression discontinuity design I do not detect differences in the population growth rates across the subset of municipalities that lie below and above each of the population thresholds used to determine the size of the municipal council. Section 2 describes the institutional setting. Section 3 shows that the reciprocal of council size is the vote share that guarantees political representation regardless of the vote shares 5

distribution in a proportional electoral system. Section 4 describes the data and the empirical strategy. Section 5 present the baseline results and the robustness checks. Section 6 provides concluding remarks. 2 Institutional setting 2.1 Ethnic violence Political violence in the autonomous region of the Basque Country has emerged during the Franco regime and has intensified in the period immediately in the demise when a range of constitutional reforms were taking place. The organization most actively engaged in violent acts was ETA whose objective is the promotion and the establishment of an independent Basque state. Some of the more commonly addressed forms of political violence involves acts that include terrorist acts as well as the kidnappings and assassinations of targeted individuals. Considerably less attention has been paid to softer forms of political violence whose objective is the promotion of political influence in the local legislative and executive bodies. Street violence is suggested by citeprobles07 to be a softer form of political violence used in the Basque Country to increase the influence of (a subset of) Basque nationalist political parties at the municipal level. This form of violence resembles politically-based riots with respect to the damage inflicted on public facilities, private properties as well as injuries to people. Street violence, however, differs from politically-based riots in the sense that the incidents are wellorganized with pre-planned actions that purposively target a set of buildings/police officers in a particular location. Some of the examples of street violence include an organized group of individuals to launch petrol bombs against predefined targets such as public buildings or ambushes of police forces. The absence of spontaneous behaviour of street violence. 6

2.2 Electoral rules and ethnic-based representation The autonomous community of the Basque Country enjoys a considerable degree of autonomy over political, cultural and fiscal matters granted in the Spanish Constitution of 1978. Municipalities are also self-governing bodies with an elected municipal council. The elected by the council municipal government and major enjoy a considerable degree of fiscal autonomy on the allocation of government expenditure and the use of fiscal tools to raise revenue. The municipal council is the legislative body in all Basque municipalities with population of at least 100 residents, which is elected for a fixed-term of 4 years. 1 The municipal council s responsibilities include passing legislation, voting the budget as well as electing a municipal government including a mayor. Spanish municipalities, including those in the autonomous community of the Basque Country, use a pure proportional electoral system determined by a nationwide legislation. The allocation of votes casts for a party lists into council seats is governed by the d Hondt method as well as the legislated electoral threshold and the total number of municipal council seats. While each political party must meet the legislated electoral threshold of 3 % of the cast ballots to be eligible for council seats, meeting the requirement does not guarantee political representation in a small-sized council. In section 3, I demonstrate that the electoral threshold that guarantees representation regardless of the distribution of the popular vote across political party equals the reciprocal of the council size. The political polarization in the Basque Country, according to Gardeazabal (2011), arises on the Right/Left and the Nationalist/Non-nationalist dimensions, where the Nationalist/Nonnationalist dimension generates division along ethnic lines. The major political parties cover all Right/Left and Nationalist/Non-nationalist combinations. 1 Municipalities with less than 100 residents directly elect a mayor and have no municipal council. 7

3 Methodology In this section, I formally show that the reciprocal of a council size is equivalent to an electoral threshold in a proportional electoral system that uses the d Hondt method of allocates vote share council seats. Proposition 1 A share of the popular vote equal to 1 N in a riding of size N guarantees political representation for a political party regardless of the vote shares distribution in a proportional electoral system that uses the d Hondt method of allocating vote shares into council seats. The proof of Proposition 1 is relegated Appendix B, whose argument is based on the allocation of vote shares into council seats in a proportional electoral system. The rules governing the allocation of vote shares into parliamentary seats is presented in Appendix A. Appendix C provides a numerical example that illustrates the arguments of Appendices A and B. The reciprocal of a council size acts as a binding threshold for small-sized municipal councils. Unlike the legislated threshold, the reciprocal of a council seat is to be viewed as the highest vote share that grants any political party representation regardless of the vote shares distribution. 4 Data 4.1 Data and descriptive statistics In this paper, I use three data sources that include data on street violence, electoral as well as census data. The first source is data on street violence collected by (de la Calle Robles, 2007) for the period 1996-2000 as well as the years 1990 and 1967. This variable measures the number of incidents of street violence in a given municipality in a given year. The reported incidents, which are collected from news reports in several Basque Country newspapers as well as yearbooks, include attacks that inflict damage on public buildings and property in public areas. 8

The other two data sources of data include electoral data obtained from the Department of Security of the Basque Country as well as census data obtained from the Statistical Institute of the Basque Country. The electoral data encompasses 3 municipal electoral cycles (1987-1991, 1995 1999, and 1999-2003) for the 251 Spanish municipalities of the autonomous community of the Basque Country. The use census data, collected once every 5 years, includes information on municipal-level demographic and socioeconomic characteristics such as population, ethnolinguistic affiliation, labour force characteristics and GDP. The data on street violence from 1990 is matched with the 1986 census survey, while that from the years 1996-2000 with the 1996 census survey. To construct the key institutional variable, the share of (native) Basque speakers per full seat, I use the ratio of the share of native Basque speakers in a census year relative to the reciprocal of the council size of each municipality in a given electoral cycle. Table 1 provides descriptive statistics of the key variables of interest: the number of incidents of street violence, the share of native Basque speakers, the relative frequency of hung councils as well as the share of native Basque speakers per full seat. Figure 1 reports a skewed to the left distribution of incidents of street violence across Basque municipalities with a sizable number of observations with no incidents of street violence. Table 1 reveals that the Basque Country municipalities exhibit considerable variation with respect to their ethnolinguistic variation, variation across council size as well as the share of majority parliaments. The first two sets of variation provide for exploring what forms of inclusive institutions are determinants of political violence the magnitude of the electoral threshold per se or its magnitude relative to the size of the Basque minority. In addition the variation in type of parliament hung versus majority provides suitable conditions for investigating whether it is the legislative representation of ethnic-based political parties or their ability to influence public policy that triggers political violence. 9

Figure 1: Distribution of the Street Violence Across the Basque Country Municipalities Notes: 1. The histogram displays the share of observations that lie within 10 intervals of equal length of 1 %. Variable Table 1: Descriptive Statistics Variable Street violence 1.516*** Basque speakers 47.524*** (nr. of incidents per year) (0.149) (% of total population) (1.668) Basque speakers per full seat 4.306*** Unemployment rate 17.998*** (0.121) (0.291) Hung parliament 0.398*** Population 8,618.802*** (binary indicator) (0.022) (1,401.906) Nr. of observations 1,486 Nr. of observations 1,486 Notes: 1. Levels of significance: *** p < 0.01, ** p < 0.05, * p < 0.10. Standard errors are reported in parentheses. 10

!ht Table 2: Descriptive Statistics Municipalities with Street Violence Incidents Variable Pre-election Year Non-pre-election Year Panel A: Both majority and hung councils Basque speakers per full seat 2.459*** 5.159*** (0.215) (0.319) Nr. of observations 125 265 Panel B: Hung councils Basque speakers per full seat 2.114*** 5.102*** (nr. of incidents per year) (0.344) (0.345) Nr. of observations 97 174 Panel C: Majority councils Basque speaker per full seat 3.655*** 5.266*** (nr. of incidents per year) (0.753) (0.351) Nr. of observations 28 91 11

4.2 Empirical strategy I use a conditional negative binomial fixed effect overdispersion model as the baseline identification strategy to estimate: SV it = β 1 + β 2 SBF S it + β 3 P Y t + β 4 SBF S it P Y t + X itδ + λ i + µ t + u it (1) where the dependent variable SV it denotes the number of street violence incidents in municipality i in year t. The key regressor SBF S it denotes the council size and the share of Basque speakers per full seat respectively of municipality i in year t. In order to demonstrate the presence of an effect in municipal pre-election years, I include the binary indicator for a pre-election year P Y t as well as an interactive terms of the key regressor SBF S it with this binary indicator, i.e. SBF S it P Y t. In a non-linear model, the coefficient of an interacted regressor does not equal the corresponding marginal effect and, for this reason, the reported coefficients are the incidence-rate ratios. For instance, the coefficient β 3 measures the likelihood of a given SBF S on street violence in a pre-election relative to the corresponding effect of a non-pre-election year. In addition to the key regressors of the baseline model, I include a set of control variables denoted by the vector X i. λ i +µ t are the municipality-specific and the time fixed effects respectively. The error term is modelled to allow for overdispersion. In a non-linear model, the coefficient of an interacted continuous regressor with a binary indicator does not equal the corresponding marginal effect and, for this reason, the reported coefficients are the incidence rate ratios. The incidence-rate ratio for the interacted term is computed using the formula exp(β 3 + β 4 SBF S), where SBF S takes on a particular value. The standard errors of incidence rate ratios for the interacted terms are computed using the formula: var( ˆβ 3 ) + SBF S 2 var( ˆβ 4 ) + 2SBF Scov( ˆβ 3, ˆβ 4 ). To distinguish between the legislative and the executive channels of representation on political violence I augment the baseline model with the vector with Z = (HP it, HP it 12

P Y t, SBF S it HP it, SBF S it HP it P Y t ), which includes two differential intercept coefficients (HP it and HP it P Y t and two differential slope coefficients (SBF S it HP it and SBF S it HP it P Y t ). The augmented model is given by: SV it = β 1 + β 2 SBF S it + β 3 P Y t + β 4 SBF S it P Y t + Z itγ + X itδ + λ i + µ t + u it (2) The coefficient of the interacted term SBF S it HP it P Y t measures whether hung parliaments in pre-election years exhibit an effect different from that of majority parliaments in pre-election years. 5 Results 5.1 Baseline results Table 3 presents the baseline results based on the panel data negative binomial model. In columns (1) (4), I present evidence that a lower ratio of native Basque speakers per full seat on average increases the number of street violence incidents in pre-election years but not during other years of the electoral cycle. Due to the inclusion of an interacted term for SPFS in pre-election years in a non-linear model, all reported coefficients in Table?? are the incidence ratios. In column (1), I find that a higher ratio of native Basque speakers per full seat in pre-election years, relative to that in non-pre-election years, on average increases the number of street violence, a result that is statistically significant at the 1 % level. The baseline result is robust to the inclusion of fixed effects in column (2) and of fixed effects and covariates in column (3). These results are also robust to the exclusion of the provincial capitals from the sample. I report the incidence rate ratios interactive term of SPFS in pre-election years in Table 4 at steps of 0.25, all of which are statistically significant at the 1 % level. SFPS has a negative (in a statistical sense) effect on street violence as SFPS increases, the effect becomes larger. 13

An increase of SFPS from 1 to 2 decreases street violence by 5.4 %. Which channel(s) the legislative or the executive are driving this baseline effect? Column (5) (8) of Table 3 indicate that the baseline effect of the ratio of native Basque speakers per full seat on the number of street violence incidents in pre-election years is detected only for the subset of observations with hung parliaments. This effect is captured by the interacted term of the share of Basque speakers per full seat in a pre-election year with the binary indicator for a hung parliament. The estimated effect in the fully-specified model with fixed effects and covariates, reported in column (7), measures 5.3 % and is statistically significant at the 5 % level. In contrast, the effect for majority parliaments is considerably smaller, measuring 1.4 %, and not statistically significant at the 10 % level. Table 3: Baseline Results Street violence (number of incidents per year) (1) (2) (3) (4) (5) (6) (7) (8) Share of Basques per full seat (SBFS) 1.044 0.960 0.959 0.942 0.995 0.934 0.872 0.872 (0.036) (0.037) (0.117) (0.125) (0.039) (0.041) (0.125) (0.125) SBFS * Pre-election year (PY) 0.909*** 0.911*** 0.938*** 0.940*** 0.869*** 0.881*** 0.900*** 0.900*** (0.011) (0.012) (0.017) (0.019) (0.027) (0.027) (0.030) (0.030) SBFS * Hung parliament (HP) 1.107*** 1.052 1.048 1.048 (0.030) (0.030) (0.031) (0.031) SBFS * HP * PY 1.046 1.038 1.050 1.050 (0.035) (0.035) (0.036) (0.036) Fixed effects No Yes Yes Yes No Yes Yes Yes Covariates No No Yes Yes No No Yes Yes Provincial capitals No No No Yes No No No Yes Nr. of observations 1,486 1,486 1,486 1,463 1,486 1,486 1,486 1,463 Notes: 1. Levels of significance: *** p < 0.01, ** p < 0.05, * p < 0.10. Standard errors are reported in parentheses. 5.2 Robustness checks The baseline results are robust to the inclusion of additional controls that proxy for different channels suggested in the literature to be influencing political violence. The controls, included 14

Table 4: Baseline Results: Incidence Rate Ratios SBFS Incidence Rate Ratio Standard error SBFS Incidence Rate Ratio Standard error 1 0.666*** 0.110 3 0.612*** 0.189 1.25 0.659*** 0.123 3.25 0.606*** 0.196 1.5 0.653*** 0.134 3.5 0.600*** 0.204 1.75 0.646*** 0.145 3.75 0.593*** 0.211 2 0.639*** 0.154 4 0.587*** 0.217 2.25 0.632*** 0.164 4.25 0.581*** 0.224 2.25 0.632*** 0.164 4.5 0.575*** 0.230 2.5 0.625*** 0.172 4.75 0.569*** 0.237 2.75 0.619*** 0.181 5 0.563*** 0.243 Notes: 1. Levels of significance: *** p < 0.01, ** p < 0.05, * p < 0.10. Standard errors are reported in parentheses. in Table??, include economic factors such as the municipal unemployment rate that proxies both for economic inequality and social exclusion. The inclusion of socio-demographic factors capturing the population size of the municipality and the share of Basque speakers also leave the baseline results unchanged. The inclusion of the council size alleviates concerns that another institution, the magnitude of the electoral threshold, may influence the results. The inclusion of an interactive term of council size with pre-election years does not alter the results, thus suggesting that it is not the magnitude of the electoral threshold per se but its magnitude relative to the size of the Basque minority. To mitigate concerns that the inference in the fixed effects negative binomial model is not influenced by the large number of observations in the dependent variable that take on the value of zero, I report in Table 5 the results from the zero-inflated negative binomial model. In this model the standard errors are clustered at the year level. The reported results in this table conform with the baseline results from the fixed effects negative binomial model in Table??. I also perform a sensitivity analysis as to whether the tails of the council size distribution drive the baseline results. At the lower tail, one-step increase generate a substantial increase in the guaranteed threshold for representation, while, at the upper tail, such increase is relative small. In addition, the municipalities with small-sized councils are also likely to be subject to 15

an underreporting bias (de la Calle Robles, 2007). The baseline results remain unchanged if the sample is restricted at each tail of the distribution. Table 5: Robustness checks: ZINB model Street violence (number of incidents per year) (1) (2) (3) (4) (5) (6) Share of Basques per full seat (SBFS) 0.970* 1.072** 1.137*** 0.898*** 1.050 1.106** (0.017) (0.037) (0.044) (0.020) (0.044) (0.049) SBFS * Pre-election year (PY) 0.936*** 0.945*** 0.947*** 0.907** 0.942* 0.942* (0.017) (0.019) (0.020) (0.036) (0.032) (0.032) SBFS * Hung parliament (HP) 1.117*** 1.015 1.019 (0.027) (0.020) (0.020) SBFS * HP * PY 1.021 1.003 1.004 (0.045) (0.035) (0.035) Covariates No Yes Yes No Yes Yes Provincial capitals No No Yes No No Yes Notes: 1. Levels of significance: *** p < 0.01, ** p < 0.05, * p < 0.10. Standard errors are reported in parentheses. 5.3 Endogeneity concerns The key regressor, the share of ethnic Basques per full seat, is potentially endogenous for two reasons. First, political violence may change the ethnic composition of the population of a given municipality over time through political and/or economic channels. Secondly, political violence might be targeted across municipalities so that it influences the population size of the municipalities in proximity to the population thresholds used to determine council size. I address the first concern by employing a fixed effect panel data model that estimates the effect of political violence in pre-election years on the population growth rate. For this estimation, I use two unbalanced panels of data on the number of riots in 1990 and 1998 as well as the corresponding 5-year population growth rate over the 1986-1991 and 1996-2001. The estimated model is: 16

P opgrowth it = δ 1 + δ 2 SV it + µ i + ν t + ε it (3) where the dependent variable P opgrowth it is the population growth rate over a 5 year period. The key explanatory variable of interest is SV it. For the 1990 pre-election year, I use the 1986-1991 period to measure the population growth rate and for the 1998 pre-election year the 1996-2001 period. The model also includes municipal and time fixed effects, µ i + ν t respectively. The error term ε it is modelled as heteroskedastic-robust. Table 6 reports the results from the panel data estimation, which do not indicate the presence of a statistically significant relationship. These results are reported in the absence of fixed effects or covariates (Column 1), in the presence of fixed effects but no covariates (Column 2), as well as in the presence of fixed and covariates. The included covariates include the logarithm of initial population size and the initial share of native Basque speakers. The results are also robust to the logarithmic transformation of the street violence or the exclusion of the provincial capitals from the sample. Table 6: OLS results Five-year growth rate of SBFS (1) (2) (3) Street violence (number of incidents per year) -0.627-0.245-0.565 (0.297) (0.266) (0.724) Fixed effects No Yes Yes Covariates No No Yes Notes: 1. Levels of significance: *** p < 0.01, ** p < 0.05, * p < 0.10. Standard errors are reported in parentheses. I address the second concern by using RD design to estimate whether there are differences in the population growth rate on each side of the population thresholds used to determine the size of the municipal councils. To construct a unique assignment variable that is comparable 17

across all population thresholds, I define the distance from each population threshold as a percentage change relative to the magnitude of each population threshold. I use a sharp RD design due to the perfect compliance of all observations with respect to the treatment assignment status implied by the assignment variable. I estimate the model using the following regression equation: Basquesharegrowth it = γ 1 + γ 2 Assignvar it + X δ + ε it (4) y i denotes the population growth rate of municipality i over the period 1996-2001. D i is the treatment status variable that takes on a value of 1 if an observation lies above a population threshold and 0 otherwise. ε it is the error term. I use flexible polynomial smoothing as estimation. As support, I include those observations within ± 10 % of the threshold of the assignment variable Assignvar it. I do not detect the presence of a statistically significant effect. The absence of an effect is robust to the change of the window or the bandwidth. In addition, the densities of other variables, such as the number of riots, are also insignificant at the population threshold. 6 Conclusion My paper investigated how the design of more inclusive political institutions in ethnolinguisticallymixed regions mitigates street violence. To address this question, I used the setting of Basque municipalities, which exhibit variation with respect to the inclusiveness of political institutions and also with respect to the number of incidents of street violence. I found that the key institution that influences street violence is the size of an electoral threshold relative to the size of an ethnic group. First, I formally showed that the reciprocal of a council size, referred to as a full seat, is equivalent to an electoral threshold in a proportional electoral system. Then, I used a negative binomial fixed effects model as the baseline empirical strategy to estimate the 18

effect of the share of Basque speakers per full seat on the number of street violence incidents. I found that as this institution becomes more inclusive, it has a negative (in a statistical sense) effect on street violence in municipal pre-election years. I demonstrated that this result is robust to a series of robustness checks. In addition, I mitigated endogeneity concerns: first, by using a panel data least-squares model indicating that street violence in pre-election years has no effect on the growth rate of the share of Basque speakers; secondly, by employing an RD design model indicating that there are no differences in the population growth rates between those municipalities below and those above the population thresholds used to determine a municipality s council size. I also investigated whether the baseline result of this paper is driven by the legislative and/or the executive channel of political representation. The inclusion of an interacted term between the key political institution, the share of Basque speakers per full seat, and the binary indicator for hung parliament provided evidence indicating that the baseline effect is driven only by the legislative channel irrespective of the prevailing type of parliament. This channel, which encompasses access to legislative representation (Alionescu, 2004), immunity from prosecution (Chang, 2005; Chang et al., 2010; Wigley, 2003), is neutral to the frequency in the formation of hung parliaments. The results of this paper have important implications on the design of inclusive institutions in proportional electoral systems. The first policy implication of my results was that the adoption of jurisdiction-specific electoral thresholds that take into account the size of the ethnic groups in each jurisdiction. Such policy would alleviate concerns by ethnic-based political parties that they may end up not being representation in the legislature and thus reduce their incentive to engage in street violence. Such policy is also to improve upon the competitive nature of elections in settings with predominant expressive voting along ethnolinguistic lines, where the size of an ethnic group serves as a vote ceiling. Secondly, the baseline results suggest 19

that a lower electoral threshold has no indirect effect on political violence by inducing the more frequent formation of hung parliaments. This is an important consideration as a lower electoral threshold for representation is generally found to induce the more frequent formation of hung parliaments due to mechanical and behavioural effects. The results of my paper could spur future research in two directions. First, the mechanisms that drive the legislative channel, but not the executive channel, that are suggested to be influencing political violence are presently under-researched. Some of these mechanisms include political recognition as a parliamentary force, immunity from prosecution, as well as access to state funds. Secondly, my paper indicates the importance of designing inclusive institutions for representative bodies with a relatively small size or such institutions in settings that conform to the election system of proportional representation. 20

Appendix A Seat allocation in PR electoral systems The d Hondt method is defined as a sequence of divisors, 1, 2,..., N, whose members equal the size, N, of a multi-seat constituency. This sequence of divisors is used to obtain the socalled scores from the vote shares of all political parties, where the ranking of these scores determines the seat allocation across the participating political parties. There are P scores for each political party as its vote share is divided by each of the divisors in the sequence. In a constituency with P number of participating political parties, the total number of scores equals N P. The highest ranked N scores grant the corresponding political parties council seats, while the remaining N(1 P ) scores do not. The allocation mechanism governing the translation of vote shares into council seats depends on the electoral rules pertaining to a proportional electoral system, which include the allocation method of seat allocation, the size of the multi-seat riding (N), the legislated electoral threshold ( v). Let (v 1, v 2,..., v P p v p = 100, 0 v p 100) be the distribution of vote shares of political parties (α 1, α 2,..., α P ) in a multi-seat constituency of size N, where N is a positive integer. For notational convenience, let Ω be the short-hand notation for (v 1, v 2,..., v P p v p = 100, 0 v p 100). W.l.o.g., I assume that v 1 v 2... v P and in addition that N is odd. To be eligible for seats, the vote share each political party α p p must meet the electoral threshold, i.e. v p v. Appendix B Proof of Proposition 1: Consider an initial distribution of the form: ( 1 N + b 1, a 2 1 N + b 2,...., a P 1 N + b P ), where each a p {0, 1, 2,..., N} p and Σ p a p = N. Let b p = 0 for all political parties. Transfers of the form (δa 1 1 N +δb 1, δa 2 1 N +δb 2,...., δa P 1 N +δb P ), where Σ p δa p = 0, and Σ p δb p = 0, where δb p (0, 1 N ) 21

generate the continuum of vote shares distributions. First, I show that, for the initial vote shares distribution, the lowest electable score is 1. N Lowest electable score is 1. For the political parties with a score a N i > 0, there exists a score that score = 1. This is true if a N i = N i since score = a i v i /N i. There are Σ p a p scores at least as large as score = 1 N. Since Σ pa p = N, score = 1 N is the lowest electable seat. Next, consider a transfer in vote shares only in δa p, while holding b p = 0 constant. Be redefining â i = a i + δa i and invoking the property Σ p a p = N, the proof is analogous to that for the initial vote share distribution. Finally, consider a transfer in vote share only in δb p, while holding a p constant. In this case I show that score < 1 N is the lowest electable seat. In the absence of a change in the seat allocation, each score for the subset of political parties with b p < 0 strictly decreases in value. The lowest electable score decreases is, therefore, now lower than 1 N. In the presence of a change in the seat allocation, extra seats are gain by a subset of the parties that experience an increase in their vote share at the expense of those that see a decrease in their vote share. Suppose that the transfers are such that generates score = (v i+b i ) (N i +1) to be electable. I show that: 1 N = v i N i > (v i+b i ) (N i +1). The equality 1 N = v i N i was shown to be true for b p = 0. The inequality v i N i > (v i+b i ) (N i +1) is true because simple algebraic steps imply that v i N i assumed). QED > b i (which is Appendix C The following example illustrate the allocation of vote shares into parliamentary seats that is consistent with the general notation presented in Appendix A. Panels A, B and C illustrate the proof presented in Appendix B. 22

Table 7: Council Size Reduction: Mechanical and Behavioral Effects Panel A: All vote shares are multiples of 1 N 100% Party Vote Score Score Score Score Score Seats name share n = 1 n = 2 n = 3 n = 4 n = 5 A 40 % 40 20 13.33 10 8 2 B 20 % 20 10 6.67 5 4 1 C 20 % 20 10 6.67 5 4 1 D 20 % 20 10 6.67 5 4 1 Panel B: All vote shares are multiples of 1 N 100% Party Vote Score Score Score Score Score Seats name share n = 1 n = 2 n = 3 n = 4 n = 5 A 60 % 60 30 20 15 12 3 B 40 % 40 20 13.33 10 8 2 C 0 % 0 0 0 0 0 0 D 0 % 0 0 0 0 0 0 Panel C: Some vote shares are not multiples of 1 N Party Vote Score Score Score Score Score Seats name share n = 1 n = 2 n = 3 n = 4 n = 5 A 40 % 40 20 13.33 10 8 2 B 33 % 33 16.5 11 8.25 6.6 1 C 11 % 11 5.5 3.67 2.75 2.2 0 D 7 % 7 3.5 2.33 1.75 1.4 0 Notes: 1. The scores in blue color in Panel A are the five highest-ranked corresponding to a legislature of size N = 5 electing a seat for the corresponding political party. 23

References Bibliography C.-C. Alionescu. Parliamentary Representation of Minorities in Romania. Southeast European Politics, 5(1):60 75, 2004. T. Besley and T. Persson. The Logic of Political Violence. The Quarterly Journal of Economics, 10:1 35, 2011. R. R. Bhavnani. Do electoral quotas work after they are withdrawn? Evidence from a natural experiment in India. American Political Science Review, 103:23 35, 2009. G. Boichev. Combating Voter Manipulation Strategies by Inducing Hung Parliaments in a Proportional Electoral System. Unpublished manuscript, 2015. E. C. C. Chang. Electoral Incentives for Political Corruption under Open-List Proportional Representation. The Journal of Politics, 49(4):625 633, 2005. E. C. C. Chang, M. A. Golden, and S. J. Hill. Legislative Malfeasance and Political Accountability. World Politics, 62(2):177 220, 2010. P. Collier and A. Hoeffler. Resource Rents, Governance, and Conflict. The Journal of Conflict Resolution, 49(4):625 633, 2005. L. de la Calle Robles. Fighting for the Local Control: The Street Violence in the Basque Country. Unpublished manuscript, 2007. M. de Paola, V. Scoppa, and R. Lombardo. Can gender quotas break down negative stereotypes? Evidence from changes in electoral rules. Journal of Public Economics, 94:344 353, 2010. 24

M. de Paola, V. Scoppa, and M. A. de Benedetto. The impact of gender quotas on electoral participation: Evidence from italian municipalities. European Journal of Political Economy, 35:141 157, 2014. J. H. Fina and O. Folke. Mechanical and Psychological Effects of Electoral Reform. British Journal of Political Science, pages 1 15, 2014. J. Gardeazabal. The Economic Costs of Conflict: A Case Study of the Basque Country. American Economic Review, 93(1):113 132, 2003. A. L. Hillman, K. Metsuyanim, and N. Potrafke. Democracy with group identity. European Journal of Political Economy, Available online, 2015. R. Pande. Can Mandated Political Representation Increase Policy Influence for Disadvantaged Minorities? Theory and Evidence from India. American Economic Review, 10:1132 1151, 2011. M. Reynal-Querol. Does democracy preempt civil wars?. European Journal of Political Economy, 21:445 465, 2005. S. Wigley. Parliamentary Immunity: Protecting Democracy or Protecting Corruption? Journal of Political Philosophy, 11(1):23 40, 2003. 25