Scarcity Without Leviathan - The Violent Effects - Table

200 300 400 Production (in tons) 500 Yearly homicides per 100.000 pop. 5 10 15 20 25 0 2003 2004 2005 2006 2007 2008 2009 2010 Homicide rate in Mexico (left axis) Drug−related homicide rate in Mexico (left axis) Colombian cocaine production net of seizures (right axis) Figure 1: Yearly homicide rate in Mexico (from INEGI), drug-related homicide rate in Mexico (from Mexico’s President’s Office), and cocaine production net of seizures in Colombia (UNODC) Homicide rate change in log points −2 −1 0 1 2 3 Effects of seizures on violence by distance to US −6 −5 −4 −3 −2 −1 0 1 Months relative to event 2 3 4 5 6 Estimates for all Mexican municipalities Estimates for municipalities in quintiles 1 and 2 Estimates for municipalities in the first quintile Figure 2: Response of the homicide rate in Mexican municipalities at different distances from the US to a 10% increase in cocaine seizures in Colombia. Bars present 95% confidence intervals. 3 Increase in violence in log points −1 0 1 2 −2 200 400 600 800 Distance to entry points in kilometers 1000 1200 Figure 3: Response of the homicide rate in Mexican municipalities to a 10% increase in cocaine seizures in Colombia as a function of their distance to the US border. The black lines are our estimates and the gray lines are 95% confidence intervals. The vertical lines mark the thresholds defining municipalities at the top and second quintiles. Homicide rate change in log points −2 −1 0 1 2 3 Effects of seizures on violence by Cartel presence −6 −5 −4 −3 −2 −1 0 1 Months relative to event 2 3 4 5 6 Estimates for municipalities with no cartels Estimates for municipalities with one cartel Estimates for municipalities with at least two cartels −2 Homicide rate change in log points −1 0 1 2 3 Effects of seizures on violence by PAN support −6 −5 −4 −3 −2 −1 0 1 Months relative to event 2 3 4 5 6 Estimates for municipalities with low PAN support Estimates for municipalities with high PAN support Figure 4: Response of the homicide rate in Mexican municipalities that have different numbers of cartels (top panel) and different levels of PAN support (bottom panel) to a 10% increase in cocaine seizures in Colombia. Bars present 95% confidence intervals. −15 −10 −5 Estimates 0 5 10 15 Relationship between seizures in Colombia and other policies Colombian seizures of liquid inputs Colombian seizures of solid inputs Colombian destruction of cocaine labs Mexican seizures of cocaine Mexican seizure rate of cocaine Mexican seizures of heroin Figure 5: High-frequency relationship between a 10% increase in cocaine seizures in Colombia and other anti-drug policies in Colombia and Mexico. The bars around each estimate represent its 90% confidence interval; the capped lines represent its 95% confidence interval. Table 1: Summary statistics. 2006 2007 2008 2009 2010 All Mexico: 2,457 municipalities. Average distance to the US: 763 km. Homicide rate (annualized) Drug-related homicide rate Share of municipalities with cartels Population (in thousands) Historical PAN vote share (time invariant) Homicide rate (annualized) Drug-related homicide rate Share of municipalities with cartels Population (in thousands) Historical PAN vote share (time invariant) Homicide rate (annualized) Drug-related homicide rate Share of municipalities with cartels Population (in thousands) Historical PAN vote share (time invariant) 9.74 8.15 12.68 17.59 22.53 [22.55] [19.40] [28.92] [38.17] [50.52] 0.69 2.59 6.30 8.69 13.57 [ 8.63] [18.42] [26.26] [34.99] [57.78] 0.12 0.16 0.24 0.27 0.29 [ 0.33] [ 0.37] [ 0.43] [ 0.44] [ 0.45] 42.72 43.42 44.15 44.95 45.80 [127.90] [129.22] [130.47] [131.68] [132.85] 0.25 0.25 0.25 0.25 0.25 [0.13] [0.13] [0.13] [0.13] [0.13] Quintiles 1 and 2: 983 municipalities. Average distance to the US: 507 km. 8.76 [16.83] 0.41 [5.07] 0.17 [0.38] 64.35 [167.97] 0.28 [0.12] First quintile: 7.79 13.98 19.76 27.81 [14.55] [29.49] [40.99] [59.12] 2.76 8.42 10.86 17.32 [21.26] [30.89] [40.53] [70.11] 0.24 0.35 0.37 0.40 [0.43] [0.48] [0.48] [0.49] 65.32 66.35 67.45 68.62 [169.61] [171.21] [172.76] [174.27] 0.28 0.28 0.28 0.28 [0.12] [0.12] [0.12] [0.12] 492 municipalities. average distance to the US: 333 km. 9.37 [21.14] 0.77 [7.79] 0.24 [0.43] 53.10 [146.47] 0.30 [0.13] 8.62 [18.14] 3.64 [31.64] 0.34 [0.47] 54.00 [148.71] 0.30 [0.12] 19.97 [41.17] 14.03 [42.64] 0.42 [0.49] 54.93 [150.80] 0.30 [0.12] 28.67 [56.50] 18.90 [57.47] 0.43 [0.50] 55.89 [152.74] 0.30 [0.12] 44.46 [83.54] 31.40 [103.90] 0.49 [0.50] 56.88 [154.53] 0.30 [0.12] Note.- The table presents summary statistics for the main variables used in our empirical exercises. The data are presented separately for municipalities in all Mexico, in the first two distance quintiles to entry points in the US, and in the first distance quintile to entry points in the US. Standard deviations for each variable are reported below the corresponding mean in square brackets. Table 2: Time series effects of cocaine seizures in Colombia on violence in Mexico. All of Mexico (1) (2) Quintiles 1 and 2 (3) (4) (5) (6) First quintile (7) (8) (9) Dependent variable: homicide rate. log of seizures in Colombia Observations R-squared log of seizures in Colombia Observations R-squared log of seizures in Colombia Observations R-squared log of seizures in Colombia Observations R-squared log of seizures in Colombia Observations R-squared Time series controls: Unemployment rate Economic activity Seasonality ∗∗ ∗∗ 0.059 0.056 0.050 0.082∗∗ 0.079∗∗ 0.073∗∗ 0.119∗∗∗ 0.120∗∗∗ (0.027) (0.025) (0.026) (0.033) (0.032) (0.034) (0.033) (0.034) 49 49 49 49 49 49 49 49 0.950 0.957 0.961 0.947 0.953 0.955 0.950 0.952 Dependent variable: drug-related homicide rate. 0.109∗∗∗ (0.037) 49 0.953 0.120∗∗ 0.123∗∗ 0.117∗ 0.126∗∗ 0.126∗∗ 0.125∗ 0.134∗∗ 0.137∗∗ (0.055) (0.055) (0.060) (0.057) (0.057) (0.063) (0.060) (0.061) 49 49 49 49 49 49 49 49 0.952 0.955 0.956 0.961 0.964 0.964 0.959 0.961 Dependent variable: assassinations of cartel members. 0.141∗∗ (0.066) 49 0.961 0.123∗∗ 0.131∗∗ 0.126∗∗ 0.124∗∗ 0.128∗∗ 0.127∗ 0.122∗ (0.057) (0.056) (0.060) (0.060) (0.059) (0.064) (0.065) 49 49 49 49 49 49 49 0.951 0.955 0.956 0.957 0.961 0.961 0.956 Dependent variable: confrontations deaths. 0.130∗ (0.066) 49 0.958 0.132∗ (0.072) 49 0.958 0.140 0.096 0.107 0.095 0.048 0.050 0.266 0.222 (0.221) (0.223) (0.228) (0.191) (0.196) (0.205) (0.184) (0.187) 48 48 48 47 47 47 46 46 0.720 0.731 0.734 0.733 0.747 0.747 0.720 0.739 Dependent variable: assassinations of government officials. 0.224 (0.194) 46 0.740 -0.047 -0.043 0.001 -0.061 -0.070 -0.026 (0.168) (0.175) (0.191) (0.151) (0.166) (0.178) 43 43 43 41 41 41 0.588 0.595 0.620 0.426 0.434 0.476 -0.018 (0.256) 34 0.393 0.026 (0.264) 34 0.425 X X X X X X X ∗ X X X X X X X X 0.035 (0.280) 34 0.348 Note.- The table presents estimates of the effect of cocaine seizures in Colombia on violence in different groups of municipalities in Mexico. The dependent variable is the log of the homicide rate indicated in each panel, aggregated at the monthly level. The explanatory variable is the log of the monthly cocaine seizures in Colombia. We estimate the model for all 49 months from December, 2006 to December, 2010. All models include a cubic polynomial in time (months) and year dummies, so that only high frequency variation is exploited. Columns 1 to 3 present estimates for Mexico as a whole; columns 4 to 6 for quintiles 1 and 2; and columns 7 to 9 for the first quintile. Columns 2, 5 and 8 add the monthly unemployment rate and a measure of economic activity as controls. Columns 3, 6 and 9 control for the rainy and hurricane season in Mexico. Standard errors robust against heteroskedasticity are reported in parenthesis. Coefficients with ∗∗∗ are significant at the 1% level, with ∗∗ at only the 5% level and with ∗ at only the 10% level. Table 3: The differential effect of cocaine seizures in Colombia on violence in Mexico by distance from the nearest US entry point. All of Mexico Estimation method: Poisson model (1) (2) Quintiles 1 and 2 Dep. var. ln(p + x) (3) (4) Poisson model (5) (6) Dep. var. ln(p + x) (7) (8) Dependent variable: homicide rate. Seizures in Colombia × Proximity to U.S. Observations Municipalities Seizures in Colombia × Proximity to U.S. Observations Municipalities Seizures in Colombia × Proximity to U.S. Observations Municipalities Seizures in Colombia × Proximity to U.S. Observations Municipalities Seizures in Colombia × Proximity to U.S. Observations Municipalities Covariates: Differential trend by distance to US Differential trend by PAN support 0.031∗∗∗ (0.011) 98,245 2,005 0.023∗∗ 0.014∗∗∗ 0.013∗∗∗ 0.043∗∗ 0.045∗∗ 0.022∗∗∗ (0.011) (0.004) (0.004) (0.019) (0.019) (0.008) 85,309 120,295 97,902 42,973 42,189 48,167 1,741 2,455 1,998 877 861 983 Dependent variable: drug-related homicides rate. 0.023∗∗∗ (0.008) 47,089 961 0.067∗∗∗ (0.024) 56,203 1,147 0.062∗∗ 0.026∗∗∗ 0.026∗∗∗ 0.128∗∗∗ (0.025) (0.008) (0.008) (0.044) 53,214 120,295 97,902 27,342 1,086 2,455 1,998 558 Dependent variable: assassination of 0.127∗∗∗ 0.053∗∗∗ (0.044) (0.017) 27,097 48,167 553 983 cartel members. 0.054∗∗∗ (0.017) 47,089 961 0.037∗ (0.021) 53,900 1,100 0.029 0.016∗∗∗ 0.016∗∗ 0.073∗ 0.073∗ 0.031∗∗ (0.021) (0.006) (0.007) (0.039) (0.039) (0.014) 50,911 120,295 97,902 25,970 25,725 48,167 1,039 2,455 1,998 530 525 983 Dependent variable: confrontation deaths. 0.032∗∗ (0.014) 47,089 961 0.124∗ (0.070) 20,090 410 0.124∗ 0.023∗∗∗ 0.026∗∗∗ 0.198∗ 0.197∗ (0.070) (0.009) (0.010) (0.108) (0.107) 19,943 120,295 97,902 12,348 12,299 407 2,455 1,998 252 251 Dependent variable: assassination of government 0.053∗∗ (0.021) 48,167 983 officials. 0.054∗∗ (0.021) 47,089 961 0.041 (0.124) 7,448 152 0.042 (0.124) 7,448 152 0.001 (0.002) 120,295 2,455 0.001 (0.002) 97,902 1,998 -0.017 (0.221) 4,655 95 -0.017 (0.220) 4,655 95 0.002 (0.004) 48,167 983 0.002 (0.004) 47,089 961 X X X X X X X X X X X X Note.- The table presents estimates of the interaction between cocaine seizures in Colombia (measured at the monthly level in logs) on violence in Mexican municipalities located at different proximities to US entry points (measured in 100km). The dependent variable is the log of the homicide rate indicated in each panel. All estimates include a full set of municipality and time fixed effects, as well as a differential cubic time trend by distance from US entry points. Additional covariates are indicated at the bottom of the table. Columns 1 to 4 present estimates for all of Mexico. Columns 5 to 8 present estimates for the first two quintiles of distance to the US border. Columns 1,2,5 and 6 use a Poisson fixed effects model (the sample is the set of municipalities with a positive number of homicides in at least one month); while columns 3,4,7 and 8, use a monotone transformation of the dependent variable, ln(p + x), with p the 90th percentile of x. Standard errors robust to heteroskedasticity and serial correlation within municipalities are reported below each estimate in parenthesis. Coefficients with ∗∗∗ are significant at the 1% level, ∗∗ only at the 5% level, and ∗ only at the 10% level. Table 4: Additional time series estimates of the effect of cocaine seizures in Colombia on the homicide rate in different subsamples of Mexican municipalities. Panel A: Heterogeneity by Cartel presence No cartels (1) (2) 0.021 (0.023) 49 0.809 -0.209 (0.134) -0.187∗ (0.104) -0.287∗ (0.144) -0.001 (0.025) 49 0.847 log homicide rate t − 1 log homicide rate t − 2 log homicide rate t − 3 log of seizures in Colombia t Observations R-squared One cartel Panel B: Heterogeneity by PAN support Multiple cartels (3) (4) 0.053 (0.039) 49 0.913 -0.013 (0.127) -0.242∗ (0.125) -0.351 (0.221) 0.038 (0.037) 49 0.935 (5) (6) 0.084∗∗ (0.032) 49 0.943 0.044 (0.150) -0.016 (0.161) 0.008 (0.208) 0.082∗∗ (0.033) 49 0.943 Low PAN support (7) (8) 0.028 (0.028) 49 0.926 -0.135 (0.128) -0.187 (0.143) -0.351∗∗ (0.142) 0.009 (0.023) 49 0.943 High PAN support (9) (10) 0.102∗∗∗ (0.034) 49 0.943 0.181 (0.156) 0.065 (0.199) -0.046 (0.222) 0.103∗∗ (0.038) 49 0.946 Note.- The table presents estimates of the effect of cocaine seizures in Colombia on the homicide rate for different groups of municipalities in Mexico. The dependent variable is the log of the homicide rate. The explanatory variable is the log of the monthly cocaine seizures in Colombia. We estimate the model for all 49 months from December, 2006 to December, 2010. All models include a cubic polynomial in time (months) and year dummies, so that only high frequency variation is exploited. Each model is estimated for the subsample of municipalities in the group indicated on the top rows. Standard errors robust against heteroskedasticity are reported in parenthesis. Coefficients with ∗∗∗ are significant at the 1% level, with ∗∗ at only the 5% level and with ∗ at only the 10% level. Table 5: Differential effects of cocaine seizures in Colombia on the homicide rate in Mexican municipalities with different cartel presence and PAN support. Poisson model (1) (2) Dep. var. ln(p + x) (3) (4) (5) (6) 0.047∗∗∗ (0.011) -0.034 (0.081) -0.004 (0.010) 0.242∗∗∗ (0.093) 0.033∗∗∗ (0.012) 0.164∗∗ (0.076) 0.019∗∗ (0.010) 97,902 1,998 0.045∗∗∗ (0.012) -0.037 (0.085) -0.005 (0.011) 0.241∗∗∗ (0.092) 0.032∗∗∗ (0.012) 0.158∗∗ (0.079) 0.019∗ (0.010) 97,902 1,998 Panel A: All Mexico. Seizures in Colombia × ... Proximity to U.S. Cartel presence 0.011 (0.011) -0.025 (0.055) Proximity to U.S. × Cartel presence Multiple cartels 0.096 (0.061) Proximity to U.S. × Multiple cartels High PAN support 0.167∗∗ (0.072) Proximity to U.S. × High PAN support Observations Municipalities Additional covariates: Seizures in Colombia× (Proximity to the US)2 Seizures in Colombia × ... Proximity to U.S. Cartel presence 85,309 1,741 0.026 (0.018) 0.007 (0.084) 0.152∗ (0.086) Proximity to U.S. × Multiple cartels High PAN support 0.165∗∗ (0.083) Proximity to U.S. × High PAN support Observations Municipalities Additional covariates: Seizures in Colombia× (Proximity to the US)2 0.061∗∗ (0.024) -0.032 (0.175) -0.004 (0.023) 0.389∗∗ (0.173) 0.056∗∗ (0.023) 0.301∗ (0.161) 0.030 (0.025) 85,309 1,741 0.011∗∗ (0.005) -0.012 (0.027) 0.047 (0.033) 0.036∗ (0.021) 97,902 1,998 X Panel B: Quintiles 1 and 2. Proximity to U.S. × Cartel presence Multiple cartels 0.073∗∗∗ (0.021) -0.026 (0.173) -0.003 (0.023) 0.400∗∗ (0.172) 0.057∗∗ (0.023) 0.330∗∗ (0.161) 0.035 (0.024) 85,309 1,741 42,189 861 0.099∗∗∗ (0.029) -0.104 (0.216) -0.028 (0.040) 0.490∗∗ (0.208) 0.094∗∗ (0.039) 0.371∗∗ (0.188) 0.055 (0.035) 42,189 861 0.088 (0.066) -0.106 (0.217) -0.029 (0.040) 0.487∗∗ (0.209) 0.093∗∗ (0.039) 0.365∗∗ (0.186) 0.054 (0.035) 42,189 861 X 0.018∗∗ (0.009) -0.004 (0.042) 0.088∗ (0.047) 0.039 (0.032) 47,089 961 X 0.060∗∗∗ (0.017) -0.073 (0.103) -0.018 (0.017) 0.290∗∗∗ (0.110) 0.049∗∗ (0.019) 0.217∗∗ (0.102) 0.035∗∗ (0.016) 47,089 961 0.086∗∗ (0.040) -0.061 (0.109) -0.015 (0.018) 0.296∗∗∗ (0.110) 0.050∗∗∗ (0.019) 0.235∗∗ (0.104) 0.038∗∗ (0.017) 47,089 961 X Note.- The table presents estimates of the interaction between cocaine seizures in Colombia (measured at the monthly level in logs), proximity to US entry points (measured in 100km), cartel presence, and PAN support on violence in Mexican municipalities. The dependent variable is the log of the homicide rate. All estimates include a full set of municipality and time fixed effects, as well as a differential cubic time trend by distance from US entry points, by the number of cartels and by PAN support. The top panel present estimates for all of Mexico. The bottom panel presents estimates for quintiles 1 and 2. Columns 1 to 3 use a Poisson fixed effects model (the sample is the set of municipalities with a positive number of homicides in at least one month); while columns 4 to 6, use a monotone transformation of the dependent variable, ln(p + x), with p the 90th percentile of x. Standard errors robust to heteroskedasticity and serial correlation within municipalities are reported below each estimate in parenthesis. Coefficients with ∗∗∗ are significant at the 1% level, ∗∗ only at the 5% level, and ∗ only at the 10% level.