Rafael Ch et al. of 12%(9%)of a standard deviation of the munic- attacks)over the period 1997-2002 from the median ipal land informality rate over the period 2003-2006 to the ninetieth percentile of the distribution is asso- (column 8).33 ciated with an average drop of six percentage points :000100006/L0LoL (increase of five percentage points)in the probability Mechanisms:Electoral Outcomes that the Uribe coalition wins the mayoral election in either 2003 or 200%Similarly,an increase of the same The asymmetric relationship between levels of insur- magnitude of cumulative past guerrilla(paramilitary) gent and paramilitary violence and outcomes related violence over the period 2003-2010 is associated with to both property tax revenue and land formality sup- an average drop (increase)of seven (five)percentage port a mechanism of institutional capture.Given that points in the probability that Uribe's coalition wins the local authorities set property tax rates,armed groups mayor's office in 2011 (column 6). might try to exercise influence in two ways:indirectly Panels B and C of Table 4 show the correlation be- through intimidation,or directly by getting preferred tween past cumulative violence of both guerrillas and candidates elected.Specifically,given the authority that paramilitaries and the vote share of the parties forming municipal councils have to set property tax rates,we Uribe's coalition in mayoral elections and city council would expect armed groups to try to influence council elections,respectively.The results are qualitatively the members and the local council elections toward their same as those described for Panel A,specifically for favored candidates.In addition,the mayor has the re- the third period:while guerrilla violence is associated sponsibility to update the cadaster,so mayoral elec- with a smaller vote share of Uribe's coalition parties(in tions could be similarly vulnerable to armed groups the period of 2011),paramilitary violence is associated influence.We test these implications with data on elec- with a larger share toral results for election years 1997,2000,2003,2007 By and large,the evidence of this section is consis- and2011. tent with a mechanism in which armed groups capture First,we study whether the probability that a candi- political institutions through electoral outcomes. date from President Uribe's right-wing political party But what is the relative contribution of the electoral coalition wins a mayor's election is greater in places capture mechanism?There are two ways to assess this. with higher past cumulative paramilitary violence,and First,we can informally control for electoral outcomes lower in places with more guerrilla violence.Results immediately after each period of violence (and at the are reported in Table 4,Panel A.34 Because Uribe was start of each revenue measurement period)and see first elected President in 2002,we aggregate the 1997 how much the coefficients on violence change.If con and 2000 election results (when no Uribe coalition ex- trolling for post-treatment electoral outcomes signifi- isted).The null results for these elections (columns 1 cantly attenuates the coefficients of interest,then we and 2)are thus expected,and can be interpreted as know some of the apparent treatment effect is work- a falsification test.35 We also aggregate the third and ing through elections.Second,causal mediation analy- fourth periods described in the subsection Civil War sis is a more formal tool to test mechanisms that under- Dynamics and Capture in Colombia,given that the lie the relationship between a treatment variable-in 2003 and 2007 local elections occurred in the middle this case cumulative violence per group-and an out- of Uribe's administration(2002-2010).36 Both in the come variable-property tax revenue-by measuring S5.501g case of the 2003-2007 elections (columns 3 and 4).as how much of that relationship works through a third well as in the 2011 elections (columns 5 and 6),places intermediate variable,the mediator.Not only does me with higher guerrilla past cumulative violence experi- diation analysis point to the main mechanism under- ence a decrease in the probability that a candidate from lying the observed relationship of interest,but it also Uribe's coalition wins a mayoral election.Instead,in provides a way of clarifying the nature of the main re- places with higher levels of past paramilitary violence, lationship of interest(Imai,Jo,and Stuart(2011)). that probability is higher. These analyses are presented in Tables C1 and C2 Using the specification of column 4,an increase in of Appendix C.Interestingly,both approaches suggest cumulative per capita guerrilla attacks (paramilitary that little of the effect of violence on tax performance works through elections.First,with the first approach, 33 Figure 6reports marginal-effect plots for these four outcomes.The we would expect that if the main effect worked through horizontal axes include values of our violence measures ranging from elections,then controlling for Uribe's coalition victo- the median to the 90 percentile.In all cases,the asymmetry of ries should attenuate our estimates,particularly during the correlation of cumulative past guerrilla violence and that per- the Uribe administration from 2002 to 2010.Table Cl shows the effect of cumulative violence on property in mayoral elections.This is indeed the approach followed for the city tax revenue varies little controlling for posttreatment council elections (Table 6,Panel C),as the existence of several coun- electoral outcomes.37 Second,Table C2 presents the cil seats makes the winning dummy approach inappropriate for this causal mediation analysis on Uribe's coalition electoral context.For completeness,we show the correlation between cumula- victory dummy variable.The average causal mediation tive past violence and the vote share of Uribe's coalition in mayoral elections in Panel B. 35 These null results also hold if we estimate the same model for both 37 Columns 1,3,5,and 7 show the same baseline specification as the election years separately (results available upon request). one presented in Table 2,while columns 2,4,6,and 8 include the post- Again,the results are robust to treating both election vears treatment Uribe's coalition victory dummy.Coefficients vary little in separately. terms of the magnitude. 1010Rafael Ch et al. of 12% (9%) of a standard deviation of the munic￾ipal land informality rate over the period 2003–2006 (column 8).33 Mechanisms: Electoral Outcomes The asymmetric relationship between levels of insur￾gent and paramilitary violence and outcomes related to both property tax revenue and land formality sup￾port a mechanism of institutional capture. Given that local authorities set property tax rates, armed groups might try to exercise influence in two ways: indirectly through intimidation, or directly by getting preferred candidates elected. Specifically, given the authority that municipal councils have to set property tax rates, we would expect armed groups to try to influence council members and the local council elections toward their favored candidates. In addition, the mayor has the re￾sponsibility to update the cadaster, so mayoral elec￾tions could be similarly vulnerable to armed groups’ influence. We test these implications with data on elec￾toral results for election years 1997, 2000, 2003, 2007, and 2011. First, we study whether the probability that a candi￾date from President Uribe’s right-wing political party coalition wins a mayor’s election is greater in places with higher past cumulative paramilitary violence, and lower in places with more guerrilla violence. Results are reported in Table 4, Panel A.34 Because Uribe was first elected President in 2002, we aggregate the 1997 and 2000 election results (when no Uribe coalition ex￾isted). The null results for these elections (columns 1 and 2) are thus expected, and can be interpreted as a falsification test.35 We also aggregate the third and fourth periods described in the subsection Civil War Dynamics and Capture in Colombia, given that the 2003 and 2007 local elections occurred in the middle of Uribe’s administration (2002–2010).36 Both in the case of the 2003–2007 elections (columns 3 and 4), as well as in the 2011 elections (columns 5 and 6), places with higher guerrilla past cumulative violence experi￾ence a decrease in the probability that a candidate from Uribe’s coalition wins a mayoral election. Instead, in places with higher levels of past paramilitary violence, that probability is higher. Using the specification of column 4, an increase in cumulative per capita guerrilla attacks (paramilitary 33 Figure 6 reports marginal-effect plots for these four outcomes.The horizontal axes include values of our violence measures ranging from the median to the 90th percentile. In all cases, the asymmetry of the correlation of cumulative past guerrilla violence and that per￾petrated by paramilitaries is evident. 34 Alternatively, we can look at the vote share of the Uribe coalition in mayoral elections.This is indeed the approach followed for the city council elections (Table 6, Panel C), as the existence of several coun￾cil seats makes the winning dummy approach inappropriate for this context. For completeness, we show the correlation between cumula￾tive past violence and the vote share of Uribe’s coalition in mayoral elections in Panel B. 35 These null results also hold if we estimate the same model for both election years separately (results available upon request). 36 Again, the results are robust to treating both election years separately. attacks) over the period 1997–2002 from the median to the ninetieth percentile of the distribution is asso￾ciated with an average drop of six percentage points (increase of five percentage points) in the probability that the Uribe coalition wins the mayoral election in either 2003 or 2007. Similarly, an increase of the same magnitude of cumulative past guerrilla (paramilitary) violence over the period 2003–2010 is associated with an average drop (increase) of seven (five) percentage points in the probability that Uribe’s coalition wins the mayor’s office in 2011 (column 6). Panels B and C of Table 4 show the correlation be￾tween past cumulative violence of both guerrillas and paramilitaries and the vote share of the parties forming Uribe’s coalition in mayoral elections and city council elections, respectively. The results are qualitatively the same as those described for Panel A, specifically for the third period: while guerrilla violence is associated with a smaller vote share of Uribe’s coalition parties (in the period of 2011), paramilitary violence is associated with a larger share. By and large, the evidence of this section is consis￾tent with a mechanism in which armed groups capture political institutions through electoral outcomes. But what is the relative contribution of the electoral capture mechanism? There are two ways to assess this. First, we can informally control for electoral outcomes immediately after each period of violence (and at the start of each revenue measurement period) and see how much the coefficients on violence change. If con￾trolling for post-treatment electoral outcomes signifi￾cantly attenuates the coefficients of interest, then we know some of the apparent treatment effect is work￾ing through elections. Second, causal mediation analy￾sis is a more formal tool to test mechanisms that under￾lie the relationship between a treatment variable—in this case cumulative violence per group—and an out￾come variable—property tax revenue—by measuring how much of that relationship works through a third intermediate variable, the mediator. Not only does me￾diation analysis point to the main mechanism under￾lying the observed relationship of interest, but it also provides a way of clarifying the nature of the main re￾lationship of interest (Imai, Jo, and Stuart (2011)). These analyses are presented in Tables C1 and C2 of Appendix C. Interestingly, both approaches suggest that little of the effect of violence on tax performance works through elections. First, with the first approach, we would expect that if the main effect worked through elections, then controlling for Uribe’s coalition victo￾ries should attenuate our estimates, particularly during the Uribe administration from 2002 to 2010. Table C1 shows the effect of cumulative violence on property tax revenue varies little controlling for posttreatment electoral outcomes.37 Second, Table C2 presents the causal mediation analysis on Uribe’s coalition electoral victory dummy variable. The average causal mediation 37 Columns 1, 3, 5, and 7 show the same baseline specification as the one presented in Table 2, while columns 2, 4, 6, and 8 include the post￾treatment Uribe’s coalition victory dummy. Coefficients vary little in terms of the magnitude. 1010 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:05, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0003055418000333