ABSTRACT
This study investigates how an abrupt reduction in policing impacts upon the occurrence of homicides in a violent context in the Global South. The study utilizes a police strike in the Brazilian state of Ceará in summer 2020 as a quasi-natural experiment. Separate SARIMA and Exponential Smoothing models fitted on data on weekly homicide counts from January 2015 to the beginning of the strike are used to generate forecasts of homicides in a virtual counterfactual scenario with no police strikes. Actual homicide counts and forecasts are subsequently compared. The strike led to a statistically significant increase in homicides ranging between 110% and 250%. A difference-in-differences analysis confirms this result. The elasticity of homicides with respect to police presence is tentatively estimated at between -1.5 and -5.0. Even in a violent context, the perception of a higher risk of apprehension induced by police presence acts as a powerful deterrent against homicides.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Correction Statement
This article has been corrected with minor changes. These changes do not impact the academic content of the article.
Notes
1. Technically, in the Brazilian legal system there is no such thing as strike action by military forces; rather, it is classified as ‘mutiny’ in the Constitution of Brazil (art. 142, IV) and article 149 of the Military Penal Code [Decree-Law No. 1001, of 21 October 1969. Military Criminal Code].
2. The Military Police is a uniformed civilian police force that undertakes public safety and street policing functions. In fact, as per article 144 of the Brazilian Constitution, the Military Police is responsible for the ostensive policing and maintenance of public order, which makes it the principal crime fighting institution. The Civil Police are the other main police force in the country; the Civil Police are much less visible and numerous, and focus their activities on crime investigation. The Municipal Guards also contribute by both performing transit controls and regulating the use of public spaces.
3. The Box and Tiao (Citation1975) ‘univariate autoregressive moving-average (ARMA) analysis with the intervention approach’ provides a valid and efficient alternative to the ARIMA method. The principal difference between ARMA models and ARIMA models is the integral part of the latter (i.e. a measure of how many nonseasonal difference values are used to obtain stationarity). The preference for the ARIMA models derives from their wider adoption in recent homicide studies.
4. LOESS stands for LOcally Estimated Scatterplot Smoothing.
5. Synthetic controls, which combine data referring to multiple control units, would have been preferable to a DID comparison that used only one other state as counterfactual (see, Abadie et al., Citation2010). However, the use of synthetic controls was ruled out due to the lack of suitable data (i.e. multiple multi-annual time series of weekly – or more frequent – homicide counts at the state – or sub-state – level).
6. Alternatively, it would have been possible to produce a counterfactual scenario starting on 18 February, the day of the first episodes of insubordination. However, this would have required assigning the homicides recorded on 31 December of each year to the first week of the following year. The results of the two options are consistent, although the specifications of the preferred forecasting models are slightly different. The option with years and weeks starting on January 1 of each year was preferred for presenting the results because it more precisely captured the moment at which the police presence decreased in the streets and because it did not assign homicides registered in a year to the previous one.
7. On both 2 and 3 March 2020, which were considered as treatment days despite their being the first and second days after the strike, there were nine homicides recorded, which was less than on any other day during the strike.
8. The minimisation of the corrected Akaike Information Criterion led to the selection of the model ARIMA(4,1,0)(1,1,0) to fit the time series of homicides that occurred between 1 January 2015 and 18 February 2020. The analysis of the goodness of fit of this model indicated that the chosen model reasonably accounted for autocorrelation and can thus be used as an adequate model for forecasting lethal violence over time. Annexe 1 presents the goodness of fit analysis for the ARIMA model and its results.
9. Annexe 2 presents the specifications and the results of the STL+ETS analysis.
10. . Assuming a 50% contraction in police presence and taking the lowest estimate of homicide increase (i.e. +76.1%), the elasticity is about −1.5. Assuming a 50% contraction in police presence and taking the lowest estimate of homicide increase (i.e. +251.6%), the elasticity is about −5.0.
Additional information
Notes on contributors
Alberto Aziani
Alberto Aziani is Assistant Professor of Criminology at the Faculty of Social and Political Sciences at Università Cattolica del Sacro Cuore of Milan and Researcher at Transcrime. His main research interests are transnational crimes, organised crime, illicit markets, violence, and illicit financial flows. On these topics, he has conducted studies and developed projects for international organisations and funding institutions. He cooperates with the UN and the EC in research projects related to illicit drug trafficking and money laundering.