How Not to Build a State: Evidence from Colombia s False Positives

How Not to Build a State: Evidence from Colombia s False Positives D. Acemoglu MIT L. Fergusson U. Andes J. Robinson Harvard D. Romero IADB J. Vargas U. Rosario 36th Meeting of the Brazilian Econometric Society December 10, 2014, Natal, Brazil Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 1 / 68

Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 2 / 68

Motivation Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 3 / 68

Motivation State capacity Many countries lack state capacity, despite large payoffs Weak states: argued to be the root of civil wars (Fearon and Laitin, 2003). Difference between economically successful and unsuccessful countries (Evans, 1995, Herbst, 2000, Besley and Persson, 2011, Acemoglu and Robinson, 2012... many others). Variation in state capacity related to many different factors: historical path dependence (Evans), ecology population density (Herbst), political economy (Besley and Persson, Acemoglu and Robinson). Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 4 / 68

Motivation But how does a state with low capacity build it? Paying for performance? Capacity is multi-dimensional: fiscal, bureaucratic, legal......and key: monopoly of violence. How should a state which lacks monopoly of violence acquire it? Imagine a political majority in favor of building capacity. Alvaro Uribe s presidential election in May 2002. To attain monopoly of violence could expand on: extensive margin (hire more soldiers), or intensive margin (make existing soldiers work harder). How to manage the intensive margin? One way is high-powered incentives. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 5 / 68

Motivation This research Paying for performance in Colombia and its consequences We investigate high-powered incentives in the Colombian army. We show these incentives: 1 Are systematically related to murder of civilians portrayed as guerrillas killed, false positives, especially: 1 In units of military officers with pressing career concerns. 2 Where the judiciary is weak. 2 Created an incentive to corrupt the judiciary, hence eroding other dimensions of institutions. High-powered incentives can have very perverse effects. Innocents killed. Stronger military, but not consensually strong. Other institutions hurt. Hard to build state in one dimension. Complementary efforts in several dimensions are needed. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 6 / 68

Context Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 7 / 68

Zona de Distensión Major peace process between FARC and President Pastrana Uribe Government Anti-FARC platform, Democratic Security Policy reelected after amending Constitution 1997 1999 2003 1960 1970 1995 AUC Umbrella Organization 2000 2002 FARC hijacked a plane 2006 2008 2010 Dismantling of AUC Disarm and peace deals Government FARC and ELN (Left-wing guerrillas) Autodefensas (Right-wing paramilitaries) Splinter paramilitary groups

Context Introducing incentives Trends and regulation FP had long existed in Colombia, but more common in 2000s. Figure Increase coincided with incentives to fight insurgents: Law 782 of 2002: fund for intelligence operations and rewards to demobilized rebels. 2003: Democratic Security document announces system of rewards for information (regulated by decrees 128 of 2003 and 2767 of 2004). Other directives and decrees: Directive 029 of 2005 (and 015 and 016 de 2007): incentive scheme for informants leading to captures or killings. ( Secret documents not so secret). Decree 1400 of 2006 (Boina or Beret). Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 9 / 68

Context Introducing incentives Features of incentives: formal and informal Formally set a reward schedule for killings and capturing insurgents, seizing weapons and sharing information: 1 Military personnel was not explicitly excluded (also not explicitly included, except in Boina: up to one year salary), 2 No authorization ex ante by a superior officer required for operation, 3 Posterior intelligence could be used to justify the killings. Informal and unregulated incentives (see, e.g. UN Special Rapporteur): Days off when holidays approached, send to platoon on Sinai (foot soldiers). Medals, and promotions (commanders). Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 10 / 68

Context Removing incentives FP fall substantially in late 2008: media scandal after killing of several men from Soacha, near Bogotá. Government claimed victims were guerrillas killed in combat. Judicial investigations revealed this was not the case, and FP were widespread. Government issued new directives changing incentive structure: 1 Explicitly exclude rewards to military personnel. 2 Prioritize rewards to successful operations that did not involve killings (demobilizations, captures). 3 Require first investigation of combat-related deaths by judiciary. 4 Require prior intelligence for operations. Also ousted high-ranked officials involved in FPs and created special unit for FPs at Office of Attorney General. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 11 / 68

Context False positives and career concerns The case of colonels Colombian army nearly tripled during 2000s some brigades commanded by colonels, not generals. Career concerns attached to new incentives more likely to affect colonels, who still can go up the military ladder. Captain Rozo Valbuena testimony against other officer: His only objective was to gather enough statistics to be able to be promoted to General. 27 soldiers expelled by platoon commander (a colonel) for not killing two people (dressed as civilians). Soldier description: When my colonel came in he started insulting us and scolding us, and told us that we were good for nothing, that we did not understand that a guerrilla member alive was useless for him, and that what mattered were killings because he was going to be promoted to general and that is how his performance was measured. He told us he was going to have us all expelled. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 12 / 68

Context False positives and institutional weakness The case of weak judicial institutions The difficulty in controlling abuses reflects, and promotes, institutional weakness. Of the 1,056 cases of killings by armed forces that were assigned to the Fiscaĺıa (Attorney General) through April 2009, only 16 resulted in convictions (Alston, 2010, p. 13). Example: testimony from witnesses in case against Colonel Mejía Mejía had no trouble doing it because the local director of the Attorney General Office helped him with the setup When a person disappeared, his family members went to denounce it to the Police or the Ombudsman or any other institution in charge and, after this, the next victims where those denouncing. Yesterday news example Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 13 / 68

A simple model Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 15 / 68

A simple model Incentive scheme and agent utility A principal sets a linear incentive scheme. Focus on implications for agent, who exerts: good effort: a T q T (true positives), or bad effort: a F q F (false positives). Output linear in effort and noise ε J, independent and N(0, σ 2 J ): q J = a J + ε J, for J {T, F }, CARA preferences over wage w net of effort costs Ψ(a T, a F ), [ ] E e η(w Ψ(a T,a F )). Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 16 / 68

A simple model Technological complementarity or substitutability Effort cost: Ψ(a T, a F ) = 1 2 (c T a 2 T + c F a 2 F ) + δa T a F for δ c T c A Effort substitution, δ = Ψ a T a F > 0: Extreme with specialization: δ = c T c A Ψ(a T, a F ) = 1 2 ( c T a T + c F a F ) 2. Technological complements, δ = Ψ a T a F < 0 Extreme with constant ratio: δ = c T c A Ψ(a T, a F ) = 1 2 ( c T a T c F a F ) 2. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 17 / 68

A simple model Reported killings and wage Introducing misreporting and differential incentives Misreporting: α false positives can be portrayed by true positives. α: poor quality of local institutions. Reported true positives ˆq T : ˆq T = q T + αq F. Incentives: colonels payoff depends more on output. π: relative importance of output in the agent s wage. Flat salary t plus linear incentive s based on reported killings: w = (1 π)t + πsˆq T = (1 π)t + πs(q T + αq F ). Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 18 / 68

A simple model Payoff and equilibrium effort Agent s utility: u(a T, a F ) = (1 π)t + πs(a T + αa F ) }{{} expected wage ( ) 1 2 (c T at 2 + c F af 2 ) + δa T a F ηπ2 s 2 (σt 2 2 + α2 σf 2 ). }{{}}{{} effort costs risk premium Interior solution (assume αc T δ 0 and c F δα 0), a F = πs αc T δ c T c F δ 2, (1) a T = πs c F δα c T c F δ 2. (2) Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 19 / 68

A simple model Proposition 1: Testable implications of s Technological complements, δ ( c T c F, 0] 1 false and true positives: a F s > 0, a T s > 0. 2 Larger in false and true positives where misrepresentation is more likely (high α): 2 af s α > 0, 2 at s α > 0. 3 Larger in false and true positives where output is more important in compensation (high π): 2 a F s π > 0, 2 a T s π > 0. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 20 / 68

A simple model Proposition 2: Testable implications of s Technological substitutes (interior solution), δ (0, c T c F ) 1 false and true positives: a F s > 0, a T s > 0. 2 Larger in false positives, smaller in true, with high α: 2 a F s α > 0, 2 a T s α <0. 3 Larger in false and true positives with high π: 2 a F s π > 0, 2 a T s π > 0. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 21 / 68

A simple model Proposition 3: Impact on quality of institutions Consider the agents equilibrium payoff u(at, a F ). Provided risk aversion and output volatility are not too large (bound on ησf 2 ): u(at, a F ) > 0 α 2 u(at, a F ) > 0. α π Agent s are especially interested in reducing the quality of institutions when facing stronger incentives (high π). Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 22 / 68

A simple model Summary Testable predictions 1 Increase in incentives: 1 Increase average q T and q F, 2 Especially with colonels (high π), 3 Especially with weak local institutions (high α), except if effort substitution high α attenuate effect on q T. 2 Impact on institutions: 1 Agents interested in decreasing quality of local institutions. 2 Especially where π is larger (colonels). 3 Remarks: Many other things changed when incentives were introduced. Impact on true positives indirectly test technological relation. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 23 / 68

Data and Empirical Strategy Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 24 / 68

Data and Empirical Strategy Data Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 25 / 68

Data and Empirical Strategy Data Measuring FP Source: Colombian Human Rights NGO CINEP. Compiles list of events of arbitrary executions of alleged rebels. Information on: date and place of recruitment and execution; victim presented as guerrilla or paramilitary; perpetrators from Army, Police, or Navy; battalion or brigade responsible. Alternative datasets are likely to be less accurate: Official counts based on investigations: underreporting or geographic bias related to state capacity. Counts from victims associations: criticized as overstating FP. Our data: 925 cases of FP involving 1,513 victims from 1988 to 2011. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 26 / 68

Data and Empirical Strategy Data Identifying army ranks Reconstructed historical organizational structure of the army: Current structure (number, position, jurisdiction and commanders of Divisions, Battalions and Brigades) available from the army website. For previous: Expired versions of the website (available since 2000 from the Way Back Machine ). Other online sources (notably news stories in media archives, especially El Tiempo, Colombia s main newspaper) Match with CINEP data on brigade involved in FP cases to identify rank of commanders of alleged criminals. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 27 / 68

Data and Empirical Strategy Data Measuring judicial inefficiency Inspector General (Procuraduría): disciplinary oversight of public servants. Event-based dataset with all processes, by municipality, 1995-2008. Jud. Inefficiency m,0 = 1999 t=1995 Judicial functionaries cases m,t 1999 t=1995 All cases m,t Jud. Inefficiency m,t = Judicial functionaries cases m,t, All cases t {2000,...,2008} m,t Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 28 / 68

Zona de Distensión Uribe Government 1997 1999 2003 1960 1970 1995 AUC Umbrella Organization 2000 2002 FARC hijacked a plane 2006 2008 2010 Baseline Judicial Ine ciency Dismantling of AUC Yearly Judicial Ineficiency

Data and Empirical Strategy Data Other data Large set of municipal-specific characteristics to control for differential trends associated with the following sets of variables: 1 Geographical characteristics (7 variables) 2 Socioeconomic characteristics (42 variables). 1 Population 2 Security conditions in terms of conflict and crime (4 variables) 3 Educational outcomes (4 variables) 4 Municipal income and rents (6 variables) 5 Presence of natural resources (5 variables) 6 State presence and institutional capacity (22 variables) Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 30 / 68

Data and Empirical Strategy Empirical Strategy Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 31 / 68

Data and Empirical Strategy Empirical Strategy False positives, judicial inefficiency, and army ranks F. Positive m,t = α + δ m + δ t + β 0 Colonel m,t + β 1 (Colonel m,t Postȳ ) ( ) + β 2 Judicial Inefficiencym,0 Postȳ + Φ x x Postȳ + ε m,t, x X m t {2000 2008}. F. positive m,t = {Dummy, Count, Killings}. 1, if t 2003, 1, if t 2004, Postȳ = 1, if t 2005, 1, if t 2005 Excl. 2003, 2004. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 32 / 68

Data and Empirical Strategy Empirical Strategy False positives, judicial inefficiency, and army ranks F. Positive m,t = α + δ m + δ t + β 0 Colonel m,t + β 1 (Colonel m,t Postȳ ) ( ) + β 2 Judicial Inefficiencym,0 Postȳ + Φ x x Postȳ + ε m,t, x X m t {2000 2008}. F. positive m,t = {Dummy, Count, Killings}. 1, if t 2003, 1, if t 2004, Postȳ = 1, if t 2005, 1, if t 2005 Excl. 2003, 2004. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 33 / 68

Data and Empirical Strategy Empirical Strategy False positives, judicial inefficiency, and army ranks F. Positive m,t = α + δ m + δ t + β 0 Colonel m,t + β 1 (Colonel m,t Postȳ ) ( ) + β 2 Judicial Inefficiencym,0 Postȳ + Φ x x Postȳ + ε m,t, x X m t {2000 2008}. F. positive m,t = {Dummy, Count, Killings}. 1, if t 2003, 1, if t 2004, Postȳ = 1, if t 2005, 1, if t 2005 Excl. 2003, 2004. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 34 / 68

Data and Empirical Strategy Descriptive Statistics Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 35 / 68

Data and Empirical Strategy Descriptive Statistics Table 1 : Descriptive Statistics for Variables, 2000-2008 VARIABLES Mean Std. Dev. Min Max False positives dummy 0.0498 0.2175 0.00 1.00 False positives (cases) 0.0782 0.4716 0.00 15.00 False positives (killed) 0.1229 0.7747 0.00 20.00 True positives dummy 0.1838 0.3873 0.00 1.00 True positives (cases) 0.3608 1.0813 0.00 24.00 True positives (killed) 0.8385 3.7624 0.00 260.00 Colonel 0.2042 0.3924 0.00 1.00 Judicial Inefficiency 0.0594 0.1202 0.00 1.00 Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 36 / 68

Main results: Incentives and false positives Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 37 / 68

Table 2 : False Positives, Judicial Inefficiency, and Colonels Dependent variable: Dummy Number (killings) (1) (2) (3) (4) (5) (6) Panel A: Post 2003 Judicial Inefficiency x Post 0.1622 0.1868 0.0916 0.4807 0.6087 0.3981 (0.0608) (0.0647) (0.0617) (0.2141) (0.2537) (0.2612) Colonel x Post 0.0510 0.0317 0.0261 0.1845 0.1516 0.1617 (0.0139) (0.0145) (0.0143) (0.0515) (0.0575) (0.0609) R-squared 0.048 0.066 0.087 0.028 0.038 0.052 Panel D: Post 2005 Excl. 2003-4 Judicial Inefficiency x Post 0.1697 0.2128 0.1002 0.6713 0.8831 0.5257 (0.0761) (0.0826) (0.0860) (0.3044) (0.3612) (0.3809) Colonel x Post 0.0751 0.0502 0.0477 0.2465 0.1930 0.2032 (0.0177) (0.0179) (0.0179) (0.0650) (0.0713) (0.0769) R-squared 0.056 0.084 0.119 0.032 0.050 0.071 Geography (7) Socioeconomic (42) Observations 7524 6282 5698 7524 6282 5698 Number of municipalities 1078 898 814 1078 898 814

Table 3 : False Positives, Judicial Inefficiency, and Colonels Dependent variable: Dummy Number (killings) (1) (2) (3) (4) (5) (6) Panel B: Post 2004 Judicial Inefficiency x Post 0.1593 0.2181 0.1216 0.5865 0.7906 0.5546 (0.0674) (0.0773) (0.0802) (0.2577) (0.3070) (0.3244) Colonel x Post 0.0644 0.0451 0.0378 0.2195 0.1810 0.1829 (0.0164) (0.0169) (0.0170) (0.0598) (0.0657) (0.0684) R-squared 0.049 0.071 0.097 0.029 0.043 0.061 Panel C: Post 2005 Judicial Inefficiency x Post 0.1035 0.1540 0.0661 0.6116 0.8480 0.4598 (0.0713) (0.0800) (0.0890) (0.3084) (0.3702) (0.3987) Colonel x Post 0.0675 0.0469 0.0473 0.2276 0.1779 0.1846 (0.0170) (0.0173) (0.0177) (0.0615) (0.0669) (0.0745) R-squared 0.049 0.073 0.099 0.029 0.047 0.065 Geography (7) Socioeconomic (42) Observations 7524 6282 5698 7524 6282 5698 Number of municipalities 1078 898 814 1078 898 814

Table 4 : Pretrends. False Positives, Judicial Inefficiency, and Colonels, 1992-2002 Dependent variable: Post 1994 Post 1996 Post 1998 Post 2000 (1) (2) (3) (4) Panel A: Dummy Judicial Inefficiency x Post 0.0591 0.0680 0.0502 0.0072 (0.0550) (0.0593) (0.0429) (0.0531) Colonel x Post 0.0138 0.0052 0.0042 0.0000 (0.0055) (0.0040) (0.0033) (0.0039) R-squared 0.004 0.004 0.004 0.004 Panel B: Number (cases) Judicial Inefficiency x Post 0.0282 0.0375 0.0287 0.0200 (0.0359) (0.0330) (0.0237) (0.0254) Colonel x Post 0.0106 0.0040 0.0039 0.0013 (0.0045) (0.0034) (0.0027) (0.0029) R-squared 0.004 0.004 0.004 0.004 Panel C: Number (killed) Judicial Inefficiency x Post 0.0172 0.0550 0.0217 0.0173 (0.1413) (0.1034) (0.0722) (0.1008) Colonel x Post 0.0217 0.0078 0.0042 0.0022 (0.0183) (0.0102) (0.0080) (0.0088) R-squared 0.005 0.004 0.004 0.004 Observations 11636 11636 11636 11636 Number of municipalities 1078 1078 1078 1078

The impact on true positives and institutions Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 41 / 68

The impact on true positives and institutions True positives Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 42 / 68

Table 5 : True Positives, Judicial Inefficiency, and Colonels Dependent variable: Dummy Number (killed) (1) (2) (3) (4) (5) (6) Panel A: Post 2003 Judicial Inefficiency x Post 0.0356 0.0393 0.0392 0.3063 0.3725 0.7434 (0.0985) (0.1104) (0.1267) (1.2366) (1.4918) (1.9117) Colonel x Post 0.0504 0.0514 0.0206 0.4885 0.6217 0.5870 (0.0264) (0.0300) (0.0334) (0.1878) (0.2297) (0.2959) R-squared 0.007 0.011 0.019 0.009 0.013 0.023 Panel D: Post 2005 Excl. 2003-4 Judicial Inefficiency x Post 0.0528 0.0391 0.1260 0.2675 0.0831 0.6264 (0.1069) (0.1186) (0.1309) (1.2756) (1.5490) (1.9801) Colonel x Post 0.0597 0.0639 0.0290 0.5971 0.7847 0.6818 (0.0283) (0.0325) (0.0360) (0.2004) (0.2538) (0.3301) R-squared 0.007 0.012 0.025 0.009 0.016 0.028 Geography (7) Socioeconomic (42) Observations 7524 6282 5698 7524 6282 5698 Number of municipalities 1078 898 814 1078 898 814

Table 6 : True Positives, Judicial Inefficiency, and Colonels Dependent variable: Dummy Number (killed) (1) (2) (3) (4) (5) (6) Panel B: Post 2004 Judicial Inefficiency x Post 0.0215 0.0175 0.0829 0.1274 0.0599 0.3275 (0.0944) (0.0995) (0.1144) (1.0266) (1.2053) (1.4960) Colonel x Post 0.0671 0.0760 0.0418 0.5942 0.7551 0.5582 (0.0264) (0.0310) (0.0336) (0.1733) (0.2222) (0.2672) R-squared 0.007 0.012 0.022 0.009 0.015 0.027 Panel C: Post 2005 Judicial Inefficiency x Post 0.0540 0.0248 0.1772 0.1175 0.2723 0.2245 (0.0892) (0.1010) (0.1094) (0.8888) (1.0875) (1.3508) Colonel x Post 0.0683 0.0744 0.0536 0.5817 0.7711 0.6684 (0.0264) (0.0309) (0.0336) (0.1692) (0.2147) (0.2545) R-squared 0.007 0.011 0.021 0.009 0.015 0.025 Geography (7) Socioeconomic (42) Observations 7524 6282 5698 7524 6282 5698 Number of municipalities 1078 898 814 1078 898 814

The impact on true positives and institutions Impact on institutions Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 45 / 68

Table 7 : Impact on institutions: relative frequency of judicial disciplinary cases: Baseline results Dependent variable: percent judicial cases Post 2003 Post 2004 Post 2005 Post 2005 Excl. 2003-4 (1) (2) (3) (4) Panel A: Percent judicial cases Colonel x Post 0.0227 0.0172 0.0145 0.0221 (0.0098) (0.0093) (0.0091) (0.0103) R-squared 0.011 0.011 0.010 0.014 Observations 7326 7326 7326 5698 Number of municipalities 814 814 814 814 Geography (7) Socio-economic (42)

Table 8 : Impact on institutions: relative frequency of judicial disciplinary cases Verifying robustness to zero cases Dependent variable: Post 2003 Post 2004 Post 2005 Post 2005 Excl. 2003-4 (1) (2) (3) (4) Panel B: Percent judicial cases Without 0/0 Colonel x Post 0.0243 0.0180 0.0160 0.0251 (0.0103) (0.0097) (0.0095) (0.0108) R-squared 0.012 0.012 0.011 0.015 Observations 6917 6917 6917 5333 Number of municipalities 807 807 807 807 Geography (7) Socio-economic (42)

Table 9 : Impact on institutions: relative frequency of judicial disciplinary cases Verifying not driven by fall in general-led brigades Dependent variable: Post 2003 Post 2004 Post 2005 Post 2005 Excl. 2003-4 (1) (2) (3) (4) Panel C: With Post Dummy Colonel x Post 0.0205 0.0154 0.0119 0.0208 (0.0093) (0.0086) (0.0086) (0.0100) Post 0.2282 0.4190 0.2020 0.2626 (0.1990) (0.1847) (0.1849) (0.2179) R-squared 0.010 0.010 0.009 0.013 Observations 7326 7326 7326 5698 Number of municipalities 814 814 814 814 Geography (7) Socio-economic (42)

Table 10 : Impact on institutions: relative frequency of judicial disciplinary cases Verifying not driven by trends in total cases Dependent variable: Post 2003 Post 2004 Post 2005 Post 2005 Excl. 2003-4 (1) (2) (3) (4) Panel D: Constant Denominator Colonel x Post 0.0594 0.0484 0.0451 0.0569 (0.0210) (0.0212) (0.0201) (0.0225) R-squared 0.039 0.041 0.036 0.047 Observations 7325 7325 7325 5697 Number of municipalities 814 814 814 814 Geography (7) Socio-economic (42)

The impact on true positives and institutions Impact on institutions Size of the effects Taking 2003 as the post year A one-standard deviation increase in Judicial Inefficiency implies an increase, relative to the pre-period,... about 1-2 percentage points higher probability of a FP occurring (FP dummy: average increase was from 0.7% to 7%)... about 0.05-0.07 more false guerrilas killed (FP killed: average increase was from 0.02 to 0.13). A colonel rather than a general implies an increase, relative to the pre-period,... about 3 to 5 percentage points higher FP dummy... about 0.16 more FP killed... about 0.02 higher judicial inefficiency (Judicial inefficiency constant on average at 0.06). Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 50 / 68

Conclusions Contents 1 Motivation 2 Context 3 A simple model 4 Data and Empirical Strategy Data Empirical Strategy Descriptive Statistics 5 Main results: Incentives and false positives 6 The impact on true positives and institutions True positives Impact on institutions 7 Conclusions Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 51 / 68

Conclusions Conclusion: How Not to Build a State How does a State lacking the monopoly of violence acquire it? High-powered incentives to army members in the fight against the insurgency in Colombia: Are systematically related to false positives. Specially for military officers with career concerns & where state judicial institutions are weak. Created an incentive to corrupt other institutions. What do we learn from this? Building state capacity in one dimension is difficult, even counterproductive, when state is generally weak. High-powered incentives in this context can have very perverse effects. Complementary efforts in several dimensions at the same time are required. Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 52 / 68

Conclusions Thank you! Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 53 / 68

Back 6

SECRETO REPUBLICA DE COLOMBIA MINISTERIO DE DEFENS.A NACIONAL COPIA No [2- DE I j, COPIP,s MINISTERIO DE DEFENSA NACIONAL BOGOTA, D.C. 1 7 NOV..2005 DIRECTIVA MINISTERI.AL PERMANENTE ASUNTO : Politica ministerial que desarrolla criterios para el pago de recompensas por la captura 0 abatimiento en combale de cabecillas de las organizaciones armadas al margen de la ley, material de guerra, intendencia 0 comunicaciones e informacion sobre actividades relacionadas con el narcotrafico y pago de informacion que sirva de fundamento para la de labores de. inteligencia y el posterior planeamiento de operaciones. AL

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ack +%*3.4"/# > 6 7#?+"/%# (!# "+# -/)'8'.+.'9)7# ;)! F#,-.-')'-/,! 0#2$N! #5#$9)$*%! O)*5)! 0#$! (,! K)&#$! %M(-K)&%,5%! )! 2#'%! S>@T! )*-9,)'-#,%*! FN*-')*! 4%,*()&%*! M(%! 0%$'-F)! %&! 4-%4F$#! 2%! &)! G(%$H)! "IF&-')! #! %&!.(,'-#,)$-#! 2%&! J%0)$5)4%,5#! D24-,-*5$)5-K#! 2%! L%9($-2)2! 7! JDLA! )! M(-%,! *%! &%! $%'#,#H')! %&! 2%$%'O#!)!0%$'-F-$!&)!4-*4)P!

Back

True Positives by Quarter Guerrilla Kills 1988-2009 Kills 0 200 400 600 800 1990q1 1995q1 2000q1 2005q1 2010q1 Quarter

Back 6

Conclusions Table 11 : False Positives, 1988-2011. Alleged group of the victim and organization of the perpetrator Cases Executions Panel A: Alleged group of the victim Guerrilla 693 (74.9%) 1,162 (76.8%) Paramilitary 36 (4.9%) 67 (4.4%) Other 196 (21.2%) 284 (18.8%) Panel B: Organization of the perpetrator Army 853 (92.2%) 1,422 (94%) Police 37 (4%) 37 (2.4%) Other 35 (3.8%) 54 (3.6%) Total 925 1,513 Brazilian Econometric Society (12-10-2014) How Not to Build a State Colombia s False Positives 60 / 68

Table 12 : Descriptive Statistics for Variables, 2000-2008 VARIABLES Mean Std. Dev. Min Max Panel A: 2000-2003 False positives dummy 0.0498 0.2175 0.00 1.00 False positives kills 0.1229 0.7747 0.00 20.00 True positives dummy 0.1838 0.3873 0.00 1.00 True positives kills 0.8385 3.7624 0.00 260.00 Colonel 0.2042 0.3924 0.00 1.00 Judicial Inefficiency 0.0594 0.1202 0.00 1.00 Judicial Inefficiency 1995 1999 0.0779 0.0805 0.00 0.53 Panel B: Pre 2003 False positives dummy 0.0072 0.0844 0.00 1.00 False positives kills 0.0197 0.2725 0.00 7.00 True positives dummy 0.1920 0.3939 0.00 1.00 True positives kills 1.0493 5.4375 0.00 260.00 Colonel 0.1013 0.2999 0.00 1.00 Judicial Inefficiency 0.0616 0.1231 0.00 1.00 Panel C: Post 2003 False positives dummy 0.0711 0.2569 0.00 1.00 False positives kills 0.1744 0.9245 0.00 20.00 True positives dummy 0.1797 0.3839 0.00 1.00 True positives kills 0.7334 2.5373 0.00 51.00 Colonel 0.2557 0.4220 0.00 1.00 Judicial Inefficiency 0.0582 0.1187 0.00 1.00

Table 13 : False positives by commander rank, 2000-2008 Full Sample General Coronel Mean Std. Dev. Min Max N Mean Std. Dev. N Mean Std. Dev. N Diff Panel A: Dummy All years 0.0498 0.2175 0 1 10062 0.0386 0.1926 7622 0.0923 0.2894 2168 0.0537 Before year...... 2003 0.0072 0.0844 0 1 3349 0.0075 0.0864 2929 0.0030 0.0544 338 0.0046 After year...... 2003 0.0711 0.2569 0 1 6713 0.0580 0.2337 4693 0.1087 0.3114 1830 0.0508 Panel B: Number (cases) All years 0.0782 0.4716 0 15 10062 0.0559 0.3505 7622 0.1628 0.7660 2168 0.1069 Before year...... 2003 0.0093 0.1324 0 5 3349 0.0099 0.1392 2929 0.0030 0.0544 338 0.0069 After year...... 2003 0.1126 0.5666 0 15 6713 0.0846 0.4305 4693 0.1923 0.8301 1830 0.1078 Panel C: Number (Killed) All years 0.1229 0.7747 0 20 10062 0.0896 0.6039 7622 0.2500 1.2111 2168 0.1604 Before year...... 2003 0.0197 0.2725 0 7 3349 0.0205 0.2758 2929 0.0030 0.0544 338 0.0175 After year...... 2003 0.1744 0.9245 0 20 6713 0.1328 0.7349 4693 0.2956 1.3130 1830 0.1629

Table 14 : False positives by Judicial Inefficiency, 2000-2008 Full Sample Low Inefficiency High Inefficiency Mean Std. Dev. Min Max N Mean Std. Dev. N Mean Std. Dev. N Diff Panel A: Dummy All years 0.0246 0.1549 0 1 25076 0.0203 0.1410 12670 0.0292 0.1684 12255 0.0089 Before year...... 2003 0.0077 0.0874 0 1 16119 0.0065 0.0805 8126 0.0089 0.0940 7959 0.0024 After year...... 2003 0.0550 0.2281 0 1 8957 0.0449 0.2071 4544 0.0668 0.2497 4296 0.0219 Panel B: Number (cases) All years 0.0366 0.3119 0 15 25076 0.0298 0.2934 12670 0.0440 0.3313 12255 0.0141 Before year...... 2003 0.0091 0.1157 0 5 16119 0.0078 0.1090 8126 0.0106 0.1223 7959 0.0028 After year...... 2003 0.0862 0.4944 0 15 8957 0.0693 0.4652 4544 0.1059 0.5288 4296 0.0366 Panel C: Number (killed) All years 0.0601 0.5378 0 20 25076 0.0476 0.4683 12670 0.0735 0.6036 12255 0.0259 Before year...... 2003 0.0197 0.2911 0 13 16119 0.0162 0.2681 8126 0.0234 0.3134 7959 0.0071 After year...... 2003 0.1327 0.8057 0 20 8957 0.1037 0.6915 4544 0.1664 0.9188 4296 0.0628

Table 15 : True Positives by Commander Rank, 2000-2008 Full Sample General Coronel Mean Std. Dev. Min Max N Mean Std. Dev. N Mean Std. Dev. N Diff Panel A: Dummy All years 0.1838 0.3873 0 1 10062 0.1807 0.3848 7622 0.1970 0.3978 2168 0.0163 Before year...... 2003 0.1920 0.3939 0 1 3349 0.1919 0.3938 2929 0.2160 0.4121 338 0.0241 After year...... 2003 0.1797 0.3839 0 1 6713 0.1737 0.3789 4693 0.1934 0.3951 1830 0.0198 Panel B: Number (cases) All years 0.3608 1.0813 0 24 10062 0.3474 1.0643 7622 0.4004 1.1134 2168 0.0530 Before year...... 2003 0.3398 0.9414 0 15 3349 0.3404 0.9488 2929 0.3698 0.9289 338 0.0294 After year...... 2003 0.3712 1.1447 0 24 6713 0.3518 1.1305 4693 0.4060 1.1444 1830 0.0542 Panel C: Number (killed) All years 0.8385 3.7624 0 260 10062 0.8502 4.0621 7622 0.7869 2.4532 2168 0.0633 Before year...... 2003 1.0493 5.4375 0 260 3349 1.0775 5.7416 2929 0.9231 2.4507 338 0.1544 After year...... 2003 0.7334 2.5373 0 51 6713 0.7083 2.4852 4693 0.7617 2.4535 1830 0.0535

Table 16 : True Positives by Judicial Inefficiency, 2000-2008 Full Sample Low Inefficiency High Inefficiency Mean Std. Dev. Min Max N Mean Std. Dev. N Mean Std. Dev. N Diff Panel A: Dummy All years 0.1514 0.3585 0 1 25076 0.1491 0.3562 12670 0.1551 0.3620 12255 0.0060 Before year...... 2003 0.1562 0.3631 0 1 16119 0.1506 0.3577 8126 0.1623 0.3688 7959 0.0117 After year...... 2003 0.1428 0.3499 0 1 8957 0.1463 0.3535 4544 0.1418 0.3488 4296 0.0046 Panel B: Number (cases) All years 0.2757 0.9133 0 24 25076 0.2649 0.8617 12670 0.2890 0.9669 12255 0.0241 Before year...... 2003 0.2676 0.8535 0 17 16119 0.2458 0.7554 8126 0.2905 0.9432 7959 0.0447 After year...... 2003 0.2902 1.0119 0 24 8957 0.2991 1.0239 4544 0.2863 1.0094 4296 0.0128 Panel C: Number (killed) All years 0.6656 2.9667 0 260 25076 0.6588 3.3481 12670 0.6789 2.5317 12255 0.0201 Before year...... 2003 0.7090 3.2590 0 260 16119 0.6878 3.7821 8126 0.7325 2.6270 7959 0.0447 After year...... 2003 0.5875 2.3489 0 51 8957 0.6070 2.3822 4544 0.5796 2.3419 4296 0.0273

Table 17 : Judicial Inefficiency by Colonel, 2000-2008 Full Sample General Colonel Mean Std. Dev. Min Max N Mean Std. Dev. N Mean Std. Dev. N Diff All years 0.0594 0.1202 0 1 10062 0.0603 0.1199 7622 0.0592 0.1255 2168 0.0011 Before year...... 2003 0.0616 0.1231 0 1 3349 0.0637 0.1259 2929 0.0487 0.1036 338 0.0151 After year...... 2003 0.0582 0.1187 0 1 6713 0.0581 0.1160 4693 0.0611 0.1291 1830 0.0030

Figure 1 : False positives (dummy) by judicial inefficiency FP 0.05.1.15.2.25 2000 2001 2002 2003 2004 2005 2006 2007 2008 Year Low Judicial Inefficiency High Judicial Inefficiency

Figure 2 : False positives (executions) by judicial inefficiency Executions 0.1.2.3.4 2000 2001 2002 2003 2004 2005 2006 2007 2008 Year Low Judicial Inefficiency High Judicial Inefficiency