class: center, middle, inverse, title-slide # Less Policing, More Violence ## Evidence from Police Strikes in Brazil ### Marcelino Guerra ### February, 2022 --- <style type="text/css"> .pull-left2 { float: left; width: 28%; } .pull-right2 { float: right; width: 65%; } .pull-right2 ~ p { clear: both; } </style> <style type="text/css"> .pull-left3 { float: left; width: 65%; } .pull-right3 { float: right; width: 28%; } .pull-right3 ~ p { clear: both; } </style> # Motivation and Background 1. The Economics of Crime (Becker, 1968) * `\(EU_{j}=p_{j}U_{j}(Y_{j}-f_{j})+(1-p_{j})U_{j}(Y_{j})\)` 2. Common sense x Endogeneity * IV approach (Levitt, 1997; Mccrary, 2002; Levitt, 2002; Evans and Owens, 2007; Chalfin and Mccrary, 2018) * Difference-in-Differences (Di Tella and Schargrodsky, 2004; Klick and Tabarrok, 2005; Draca, Machin, and Witt, 2011; Mello, 2019) 3. Deterrence x Incapacitation 4. Uncertainty whether hiring more police officers passes in the cost-benefit test 5. Effect of police strikes on public safety * Different Brazilian law enforcement categories walked off the job on more than 700 occasions between 1997 and 2017 * During this period, the Military Police of almost all states went on strike 52 times --- # Contributions - The causal effect of the absence of Police patrolling on violence * The deterrent effects of the Police presence are large for **murder and robbery** * Motor vehicle theft x Motor vehicle robbery * In particular, the implications of Police strikes on public safety - Significant changes in police deployment in large areas - 9 and 14 million people affected in Ceará and Bahia, respectively * Two different natural experiments, very similar results - Virtually no incapacitation effect - Daily data fully identify the shock, treatment turns on and off * It is possible to see the beliefs update during and after the strike - Asymmetric effects of `\(p_{j}\)` --- # Sample Construction I .pull-left[ * I focus on the Brazilian Northeast from 2011 to 2012 * During this period, the Military Police went on a strike in six different states (mainly in the Northeast region). Sizable walkouts occurred in Maranhão (10 days), Ceará (6 days), and Bahia (12 days) * Northeastern states have similar socioeconomic and demographic levels, as well as violence patterns. Hence, I target states in that region for both control and treated groups * Among the Brazilian Northeast, Ceará, Alagoas, Sergipe, Bahia, Paraíba and Pernambuco divide their territory into **Security Areas (SAs)**. They aim to improve police work by integrating the Fire department, Civil and Military Police ] .pull-right[ <iframe src="maps/NE.html" style="width: 600px; height: 500px; border: 5px" alt=""> ] --- # Sample Construction II <div class="box"> <iframe src="maps/AIS_CE.html" frameborder="0" scrolling="no" width="50%" height="500px" align="left"> </iframe> </div> <div class="box"> <iframe src="maps/AIS_fortal.html" frameborder="0" scrolling="no" width="50%" height="500px" align="right"></iframe> </div> --- # Data * Data consists of four categories of daily crime: 1. Violent and lethal crime against a person (murder) 2. Violent crime against property (robbery) 3. Theft 4. Sexual assault * To reduce measurement error, I use vehicle robbery and vehicle theft separately * The primary source of crime information is the Police Information System (Civil Police). The Secretariat of Public Security of Ceará, Pernambuco, and Bahia provided those crimes located in each Security Area. Alagoas provided a list of all crimes between 2011-2012 * Data starts in January 2011 (except Bahia) * Other states either do not divide their territory into SAs or could not provide meaningful information * Civil *vs.* Military Police * Two separate analyses - different duration and intensity of strikes --- # Time line of events .pull-left[ ### Strike in Ceará * Crime information from July `\(1^{st}\)`, 2011 to January `\(10^{th}\)`, 2012 * Comprehends 22 treated SAs (Ceará), and 50 SAs as control - 26 in Pernambuco, 24 in Alagoas * 6 days of strike, around 70% decrease in the number of Police officers on the streets  ] .pull-right[ ### Strike in Bahia * Crime information from January `\(1^{st}\)`, 2012 to February `\(18^{th}\)`, 2012 * 52 treated SAs in Bahia, and 50 SAs as control (Pernambuco and Alagoas) * 12 days of strike, around 30% decrease in the number of Police officers on the streets  ] --- # Identification Strategy I .pull-left[
] .pull-right[
] --- # Identification Strategy II .pull-left[
] .pull-right[
] --- # Estimation Methods **TWFE** `$$Crime_{ist}=\lambda_{t}+\gamma_{i}+\beta Dint_{st}+\tau_{wd,i}+\varepsilon_{ist}$$` * SA-by-day-of-week `\((\tau_{wd,i})\)` fixed effects to allow different SAs to follow different day of week trajectories. `\(\beta\)` is the difference-in-differences estimate that captures the causal effect of interest: the extent to which the treated and non-treated SAs differ in their number of crime occurrences during the strike period * Treatment leads and lags to get a sense of the time dynamics: `$$Crime_{ist}=\lambda_{t}+\gamma_{i}+\tau_{wd,i}+\sum_{j=-10}^{7}\beta_{j} Dint_{st}(t=k+j)+\varepsilon_{ist}$$` -- **Synthetic Difference-in-Differences (Arkhangelsky et al. 2019)** * SDID generalizes DID and Synthetic Control (SC): the method re-weights and matches pre-exposure trends to weaken the reliance on common trends * SDID emphasizes units that on average are similar in terms of their past to the treated units and emphasizes periods that are on average comparable to the target periods --- # Results DID Ceará I .pull-left2[ * Compared to the daily average during the pre-strike period, average murder, robbery, and car robbery raised by **147%**, **214** and **533%**, respectively * Following the numbers reported by the newspapers, there was a reduction between 70 and 90% of police officers on the streets. The roughly approximated elasticities for murder, robbery, and car robbery are at least **-1.63**, **-2.37**, and **-5.9** ] .pull-right2[ <table> <caption style="text-align:left"> Table 1 - Difference-in-Differences Estimates</caption> <thead> <tr> <td style="text-align:center;" colspan="7"> <b> The Effect of the Military Police Strike on Crime in Ceará </b> </td> </tr> <tr> <th style="text-align:left;"> Dependent variable: </th> <th style="text-align:center;"> Murder </th> <th style="text-align:center;"> Robbery </th> <th style="text-align:center;"> Motor Robbery </th> <th style="text-align:center;"> Theft </th> <th style="text-align:center;"> Motor Theft </th> <th style="text-align:center;"> Sexual Aslt </th> </tr> </thead> <tbody> <tr> <td style="text-align:left;"> <i>Dint</i> </td> <td style="text-align:left;"> 0.465*** </td> <td style="text-align:left;"> 6.891*** </td> <td style="text-align:left;"> 1.882*** </td> <td style="text-align:left;"> 0.992 </td> <td style="text-align:left;"> 0.062 </td> <td style="text-align:left;"> −0.083 </td> </tr> <tr> <td style="text-align:left;"> </td> <td style="text-align:left;"> (0.136)</td> <td style="text-align:left;"> (1.488) </td> <td style="text-align:left;"> (0.451) </td> <td style="text-align:left;"> (0.804) </td> <td style="text-align:left;"> (0.075) </td> <td style="text-align:left;"> (0.043) </td> </tr> <tr> <td style="text-align:left;"> <b>Fixed-Effects:</b> </td> <td style="text-align:left;"> </td> <td style="text-align:left;"> </td> <td style="text-align:left;"> </td> <td style="text-align:left;"> </td> <td style="text-align:left;"> </td> <td style="text-align:left;"> </td> </tr> <tr> <td style="text-align:left;"> SA </td> <td style="text-align:center;" colspan="6" > Yes </td> </tr> <tr> <td style="text-align:left;"> Day </td> <td style="text-align:center;" colspan="6" > Yes </td> </tr> <tr> <td style="text-align:left;"> SA-by-day-of-week </td> <td style="text-align:center;" colspan="6" > Yes </td> </tr> <tr> <td style="text-align:left;"> <b>Mean of Outcome</b> </td> <td style="text-align:center;"> 0.316 </td> <td style="text-align:center;"> 3.216 </td> <td style="text-align:center;"> 0.353 </td> <td style="text-align:center;"> 4.306 </td> <td style="text-align:center;"> 0.200 </td> <td style="text-align:center;"> 0.250 </td> </tr> <tr> <td style="text-align:left;"> <b>Observations</b> </td> <td style="text-align:center;" colspan="6" > 13,968 </td> </tr> <tfoot><tr><td colspan="7">Standard-errors in parentheses and clustered at the SA level.</td></tr></tfoot> <tfoot><tr><td colspan="7"> <b>Note:</b> ***: 0.001, **: 0.01, *: 0.05 </td></tr></tfoot> </tbody> </table> ] --- # Results DID Bahia I .pull-left2[* Compared to the daily average during the pre-strike period, average murder, robbery, and car robbery raised by **77%**, **42%**, and **94%**, respectively * Following the numbers reported by the newspapers, there was a reduction between 30 and 50% of police officers on the streets. The roughly approximated elasticities for murder, robbery, and car robbery are at least **-1.54**, **-0.84**, and **-1.88** ] .pull-right2[ <table> <caption style="text-align:left"> Table 2 - Difference-in-Differences Estimates</caption> <thead> <tr> <td style="text-align:center;" colspan="7"> <b> The Effect of the Military Police Strike on Crime in Bahia </b> </td> </tr> <tr> <th style="text-align:left;"> Dependent variable: </th> <th style="text-align:center;"> Murder </th> <th style="text-align:center;"> Robbery </th> <th style="text-align:center;"> Motor Robbery </th> <th style="text-align:center;"> Theft </th> <th style="text-align:center;"> Motor Theft </th> <th style="text-align:center;"> Sexual Aslt </th> </tr> </thead> <tbody> <tr> <td style="text-align:left;"> <i>Dint</i> </td> <td style="text-align:left;"> 0.238*** </td> <td style="text-align:left;"> 1.589*** </td> <td style="text-align:left;"> 0.593*** </td> <td style="text-align:left;"> −0.812* </td> <td style="text-align:left;"> −0.014 </td> <td style="text-align:left;"> 0.008 </td> </tr> <tr> <td style="text-align:left;"> </td> <td style="text-align:left;"> (0.057)</td> <td style="text-align:left;"> (0.355) </td> <td style="text-align:left;"> (0.137) </td> <td style="text-align:left;"> (0.391) </td> <td style="text-align:left;"> (0.037) </td> <td style="text-align:left;"> (0.029) </td> </tr> <tr> <td style="text-align:left;"> <b>Fixed-Effects:</b> </td> <td style="text-align:left;"> </td> <td style="text-align:left;"> </td> <td style="text-align:left;"> </td> <td style="text-align:left;"> </td> <td style="text-align:left;"> </td> <td style="text-align:left;"> </td> </tr> <tr> <td style="text-align:left;"> SA </td> <td style="text-align:center;" colspan="6" > Yes </td> </tr> <tr> <td style="text-align:left;"> Day </td> <td style="text-align:center;" colspan="6" > Yes </td> </tr> <tr> <td style="text-align:left;"> SA-by-day-of-week </td> <td style="text-align:center;" colspan="6" > Yes </td> </tr> <tr> <td style="text-align:left;"> <b>Mean of Outcome</b> </td> <td style="text-align:center;"> 0.310 </td> <td style="text-align:center;"> 3.748 </td> <td style="text-align:center;"> 0.628 </td> <td style="text-align:center;"> 4.991 </td> <td style="text-align:center;"> 0.171 </td> <td style="text-align:center;"> 0.178 </td> </tr> <tr> <td style="text-align:left;"> <b>Observations</b> </td> <td style="text-align:center;" colspan="6" > 4,998 </td> </tr> <tfoot><tr><td colspan="7">Standard-errors in parentheses and clustered at the SA level.</td></tr></tfoot> <tfoot><tr><td colspan="7"> <b>Note:</b> ***: 0.001, **: 0.01, *: 0.05 </td></tr></tfoot> </tbody> </table> ] --- # Results DID Ceará II <div class="box"> <iframe src="figs/plotly1.html" frameborder="0" scrolling="no" width="200%" height="520px" align="right"></iframe> </div> --- # Results DID Bahia II <div class="box"> <iframe src="figs/plotly2.html" frameborder="0" scrolling="no" width="200%" height="520px" align="right"></iframe> </div> --- # Results SDID I <table> <caption style="text-align:left"> Table 3 - Synthetic Difference-in-Differences Estimates</caption> <tr> <td style="text-align:center;" colspan="7"> <b> Panel A: The Effect of the Military Police Strike on Crime in Ceará </b> </td> </tr> <tr> <th style="text-align:left;"> Dependent variable: </th> <th style="text-align:center;"> Murder </th> <th style="text-align:center;"> Robbery </th> <th style="text-align:center;"> Motor Robbery </th> <th style="text-align:center;"> Theft </th> <th style="text-align:center;"> Motor Theft </th> <th style="text-align:center;"> Sexual Aslt </th> </tr> <tbody> <tr> <td style="text-align:left;"> Point Estimate </td> <td style="text-align:center;"> <b> 0.487 </b> </td> <td style="text-align:center;"> <b> 6.763 </b> </td> <td style="text-align:center;"> <b> 1.699 </b> </td> <td style="text-align:center;"> 1.420 </td> <td style="text-align:center;"> 0.066 </td> <td style="text-align:center;"> -0.051 </td> </tr> <tr> <td style="text-align:left;"> 95% CI bootstrap </td> <td style="text-align:center;"> (0.21, 0.76)</td> <td style="text-align:center;">(3.60, 9.93) </td> <td style="text-align:center;">(0.86, 2.54) </td> <td style="text-align:center;"> (-0.15, 2.99) </td> <td style="text-align:center;"> (-0.11, 0.24) </td> <td style="text-align:center;"> (-0.16, 0.05) </td> </tr> <tr> <td style="text-align:left;"> 95% CI jackknife </td> <td style="text-align:center;"> (0.23, 0.74) </td> <td style="text-align:center;"> (3.69, 9.84) </td> <td style="text-align:center;"> (0.81, 2.59) </td> <td style="text-align:center;"> (-0.19, 3.03) </td> <td style="text-align:center;"> (-0.10, 0.23) </td> <td style="text-align:center;"> (-0.16, 0.06) </td> </tr> <tr> <td style="text-align:left;" > <b>Observations</b> </td> <td style="text-align:center;" colspan="6" > 13,464</td> </tr> </tbody> <tr> <td style="text-align:center;" colspan="7"> <b> Panel B: The Effect of the Military Police Strike on Crime in Bahia </b> </td> </tr> <tr> <th style="text-align:left;"> Dependent variable: </th> <th style="text-align:center;"> Murder </th> <th style="text-align:center;"> Robbery </th> <th style="text-align:center;"> Motor Robbery </th> <th style="text-align:center;"> Theft </th> <th style="text-align:center;"> Motor Theft </th> <th style="text-align:center;"> Sexual Aslt </th> </tr> <tbody> <tr> <td style="text-align:left;"> Point Estimate </td> <td style="text-align:center;"> <b> 0.197 </b> </td> <td style="text-align:center;"> <b> 1.784 </b> </td> <td style="text-align:center;"> <b> 0.604 </b> </td> <td style="text-align:center;"> <b> -0.977 </b> </td> <td style="text-align:center;"> -0.031 </td> <td style="text-align:center;"> -0.013 </td> </tr> <tr> <td style="text-align:left;"> 95% CI bootstrap </td> <td style="text-align:center;"> (0.07, 0.33)</td> <td style="text-align:center;"> (1.12, 2.45) </td> <td style="text-align:center;"> (0.36, 0.85) </td> <td style="text-align:center;"> (-1.77, -0.18) </td> <td style="text-align:center;"> (-0.12, 0.06) </td> <td style="text-align:center;"> (-0.08, 0.05) </td> </tr> <tr> <td style="text-align:left;"> 95% CI jackknife </td> <td style="text-align:center;"> (0.08, 0.31) </td> <td style="text-align:center;"> (1.11, 2.46) </td> <td style="text-align:center;"> (0.36, 0.85) </td> <td style="text-align:center;"> (-1.68, -0.28) </td> <td style="text-align:center;"> (-0.10, 0.04) </td> <td style="text-align:center;"> (-0.07, 0.04) </td> </tr> <tr> <td style="text-align:left;" > <b>Observations</b> </td> <td style="text-align:center;" colspan="6" > 4,284</td> </tr> </tbody> </table> --- # Results SDID II (Ceará) <div class="box"> <iframe src="figs/plotly3.html" frameborder="0" scrolling="no" width="200%" height="520px" align="right"></iframe> </div> --- # Results SDID III (Bahia) <div class="box"> <iframe src="figs/plotly4.html" frameborder="0" scrolling="no" width="200%" height="520px" align="right"></iframe> </div> --- # Final Remarks .pull-left3[ * Robust and meaningful estimates of the causal effect of the absence of police patrolling on violence * Motor vehicle theft x motor vehicle robbery * My findings are much higher than the previously estimated elasticities related to violent crime * Increase in policing levels does not generate symmetric effects as a decrease in policing levels (**asymmetric effects of `\(p_{j}\)`**) * The time dynamics of treatment effects show that the perceived risk of arrest matched the reality as time passed * Drawbacks: the measure of policing levels reduction is imprecise, and results represent a short-run effect ] .pull-right3[ <table class=" lightable-paper table" style='font-family: "Arial Narrow", arial, helvetica, sans-serif; width: auto !important; margin-left: auto; margin-right: auto; margin-left: auto; margin-right: auto;'> <thead> <tr> <th style="text-align:left;"> Paper </th> <th style="text-align:left;"> Murder </th> <th style="text-align:left;"> Robbery </th> </tr> </thead> <tbody> <tr> <td style="text-align:left;"> Levitt (AER 1997) </td> <td style="text-align:left;"> -0.91 </td> <td style="text-align:left;"> -0.45 </td> </tr> <tr> <td style="text-align:left;"> Evans and Owens (JPUB 2007) </td> <td style="text-align:left;"> -0.84 </td> <td style="text-align:left;"> -1.34 </td> </tr> <tr> <td style="text-align:left;"> Chalfin and McCrary (RESTAT 2018) </td> <td style="text-align:left;"> -0.61 </td> <td style="text-align:left;"> -0.57 </td> </tr> <tr> <td style="text-align:left;"> Mello (JPUB 2019) </td> <td style="text-align:left;"> -0.11 </td> <td style="text-align:left;"> -1.9 </td> </tr> <tr> <td style="text-align:left;"> This paper (Ceará) </td> <td style="text-align:left;"> -1.63 </td> <td style="text-align:left;"> -2.37 </td> </tr> <tr> <td style="text-align:left;"> This paper (Bahia) </td> <td style="text-align:left;"> -1.54 </td> <td style="text-align:left;"> -0.84 </td> </tr> </tbody> <tfoot> <tr> <td style = 'padding: 0; border:0;' colspan='100%'><sup></sup> Note: murder and robbery elasticities from selected papers.</td> </tr> </tfoot> </table> ]