The spatial association between drugs and urban violence: an analysis for the Metropolitan Region of Recife, Brazil

ABSTRACT

The objective of this article is to provide evidence about the association between illegal drugs and urban crimes in Brazil at the neighbourhood level, specifically considering the case of the Recife Metropolitan Region. We apply spatial econometric models to estimate a reliable relationship between drug trafficking and possession and homicide and violent property crimes. The main results indicate strong and robust associations between drug trafficking and both kinds of crimes, but not between drug possession and violent crimes. The set of evidence is obtained even after controlling for the influence of a large set of crime determinants, including not only traditional local socioeconomic conditions affecting violence, but also the presence of slums, employment access, and the presence of bars and restaurants in neighbourhoods.

KEYWORDS:

• Drug trafficking
• drug possession
• urban violence
• neighbourhoods
• spatial dependence

JEL:

• C31
• R23
• K14

ACKNOWLEDGEMENTS

Support from the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq, productivity grant) and the Fundação de Amparo a Ciência e Tecnologia de Pernambuco (FACEPE) is gratefully acknowledged.

CONTRIBUTION STATEMENT

All authors contributed to the study’s conception and design. All authors read and approved the final manuscript.

DATA AVAILABILITY STATEMENT

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the author(s).

Notes

1 The five municipalities outside the sample are Ipojuca, São Lourenço da Mata, Abreu e Lima, Moreno and Itapissuma.

2 We avoid zero using the log(X + 1) transformation. Reflecting the spread of crime in different parts of the RMR, this happens only in 14.5% and 11.5% of the cases for homicide and violent property crime rates, respectively. As for drug-related activities, these percentages correspond to 26.8% and 30.6%, respectively, for drug trafficking and possession rates. The situations with zero occurrences of homicide occur mainly in the richest neighbourhoods; for violent property crime, these cases occur mainly in neighbourhoods located on the outskirts of cities. We obtain similar results when using the log(X + 0.5) transformation.

3 In Brazil, the difference between trafficking and consumption is defined by the Drug Law (11,343/2006). Drug trafficking occurs when the individual has conduct related to the entire structure of production, packaging, distribution, logistics, sale and financing. Two kinds of measurement errors may arise in this context: the incorrect distinction between drug trafficking and possession/consumption and differences between drug arrest data and effective drug-related activities. The first one is more problematic in making comparisons across states or cities of different states, since it involves different police jurisdictions. Here, we deal with neighbourhoods subject to a single police jurisdiction and the most important measurement error is probably the second one.

4 Define the measurement error as $e=d-d^{\ast }$, where $d^{\ast }$ and $d$ are, respectively, the effective (non-observable) and the proxy of the explicative variable of interest (in our case, a drug-related variable), and assume that $Cov(d,e)=0$. Thus, we can have in a traditional linear specification $y=\alpha +{\beta }_1{x}_1+{\beta }_2{d}^{\ast }+\epsilon r$ $y=\alpha +{\beta }_1{x}_1+{\beta }_{2}d+(\epsilon -{\beta }_{2}e)$, where $y$is the dependent variable (crime), $x_1$ is a control variable, and $\epsilon$ an error term. Because of $Cov(d,e)=0$ the OLS estimate of ${\beta }_2$ using the second specification remains unbiased and consistent. Note also that even in the less probable case when the measurement error is not associated with the effective explicative variable of interest ($Cov(d^{\ast },e)=0$ the introduced bias in the OLS estimator would probably underestimate the positive effect of drug-related activities on crime (since the coefficient is positive) (Wooldridge, 2010).

5 We use the official definition of favelas (slums) used by the Brazilian Institute of Geography and Statistics (IBGE, 2011). According to it, a household is in a slum if it is in a census tract that has at least 51 dwellings located in disorderly and dense manner, on land owned by others (public or private), and that do not have access to basic public services.

6 In Appendix A in the online supplemental data, we present eigenvalues, eigenvectors, and the expression used for obtaining this index.

7 Many municipalities, especially larger ones, have an unarmed municipal guard service, mainly for traffic/parking control, crowd control at events and enforcement of zoning regulations.

8 The neighbourhood matrices considered are a binary contiguity matrix, an inverse distance matrix, and an n-nearest neighbours matrix for n = 3, 4, … , 10 (we use Euclidian distance). The results of log marginal likelihood and posterior probabilities are reported in Table C1 in the online supplemental data.

9 Tables C2 and C3 in Appendix C also present the estimated coefficient for SLX and SDM specifications, respectively, for homicide and violent property crimes.

10 For each regression we used the same initial strategy to select the appropriate spatial model.