In your paper you investigate the impact of two “place-based” interventions to reduce crime. The first consists of increased police patrols or “hot spots policing.” The second consists of the provision of city services such as improved lighting. Can you tell us a little bit about each of these two strategies and the crime prevention theory behind them?
Santiago Tobón: In most cities, crimes are highly concentrated. For instance, Weisburd (2015) documents that a handful of streets–between 5 and 6 percent–account for half of all reported crimes in a sample of eight American and Israeli cities. In Colombia, Mejía et al. (2015) report that 3 to 5 percent of the streets account for half of all crimes in the five largest cities. Place-based interventions are designed to target these high-crime areas.
Hot spots policing consists of a reallocation of street patrols to these high crime streets. Typically, it is difficult for patrolling officers to distribute time and resources in proportion to crime occurrence. In the case of Bogotá, for instance, our pre-intervention data suggests that streets concentrating roughly a third of all crimes received less than 5 to 8 percent of patrolling time. Hot spots policing aims at deliberately adjusting this. The theory behind this approach is that greater police presence increases the probability of arrest, hence deterring potential offenders (Becker, 1968).
Providing other city services in these high-crime streets can also signal order and more state capacity to potential offenders, who would be deterred because of a perceived increase in the probability of arrest or other state means to enforce the law. This is based on a theory of criminology known as “broken windows” (Wilson and Kelling, 1982).
The question of whether crimes are simply displaced around the corner is more nuanced. It may be that these locations are particularly profitable for criminal activity, which would imply that displacement should be minimal, at least theoretically. But since this is not generally true for all crimes, it was incumbent to empirically assess the extent and nature of spillovers.
Your study of the impact of these two interventions takes place in Bogotá, where you have partnered with the city government to randomly assign individual streets to receive one, none, or both of these interventions. Tell us about the relationship you built with the Bogotá city government. How did this relationship come about, and how did it impact the research design?
ST: This partnership followed a sequence of fortunate coincidences. Back in 2014, Daniel Mejía (then my PhD advisor), Daniel Ortega (CAF) and I set up a plan with the National Police of Colombia to conduct a series of randomized trials on hot spots policing in large Colombian cities. We ran a first one in Medellín and were planning to run at least one additional experiment in Bogotá. It was 2015, and Colombia was having local elections in October. As it turns out, someone who Daniel Mejia was advising closely was elected mayor of Bogotá and appointed him as Secretary of Security. At the same time, as part of my PhD process, I needed to spend time abroad in a research visiting role. We reached out to Chris Blattman to ask him whether he would host me, and if he was interested in partnering with us to run the experiment. We got positive answers from him quickly, and then we also asked Don Green to join the team—as he was largely interested in some of the questions we were addressing.
Then, Daniel Mejía moved to an implementation role, and helped us to navigate the bureaucracy quickly and to build up relationships to not only test hot spots policing but other place-based strategies. Also, with Chris and Don on the research team we managed to step up our ambitions, to design the experiment to answer questions about spillovers and to work on some methodological contributions around experimenting in dense networks.
Conducting large-scale studies in settings where crime is an issue can be difficult. Can you tell us a little bit about the practical implementation of your study. Were there any challenges, e.g., in terms of the safety of research staff and participants, any administrative hurdles or difficult ethical questions you had to grapple with? If so, how did you overcome these?
ST: In terms of security, I believe the risk was relatively minor. We did face some challenges with data collection and field work. For instance, some enumerators were stopped by criminals, who stole their materials and asked them about the scope of their work. But fortunately, it never went beyond that. To minimize the risks, we partnered with an experienced survey firm that had worked before in Bogotá and other contexts conducting surveys on sensitive topics. This firm had protocols to manage these situations. Also, and perhaps more importantly, we worked with Innovations for Poverty Action to conduct all research activities. IPA’s experience was also incumbent to minimize risks on all fronts.
On administrative hurdles, I believe we were fortunate to have someone close to the team leading the implementation. Still, we faced implementation challenges, especially with the provision of city services, as the agencies in charge were run independently of the Secretariat of Security and were less accountable to the commitments made with the experiment. The mayor and his team were important to overcome some of these challenges.
Finally, there was at least one point where we faced important ethical considerations: the question of whether these interventions would have adverse effects on people’s welfare. Both interventions are widely implemented worldwide—about 90% of police departments in the U.S. implement some form of hot spots policing strategy, for instance. Hence we saw our research as a way to learn whether these common interventions had adverse effects, rather than as a blind test of something new we just came up with.
What were your main findings in terms of how the two “place-based” interventions affected the streets that were assigned to receive them?
ST: We found that increasing state presence has, at best, modest and imprecise direct impacts. This is true for both hot spots policing and city services, when tested independently. Simultaneously, both interventions led to more substantial declines in crime, especially in the highest crime streets.
We also found interesting results on spillovers, to which I’ll turn in more detail below.
The focus on measuring spillover effects, i.e. ways in which “place-based” strategies may affect crime in neighboring areas that have not received the intervention, is indeed an interesting component of your study. You mention that spillovers could be positive or negative. Can you explain the nature of these different kinds of spillovers?
ST: In theory, whether or not crimes displace spatially is an ambiguous question. It could be that the criminal market is concentrated in high value areas and, if you treat those areas, criminals won’t necessarily move to the neighboring street. On the margin, some may even find crime too costly, exiting from their criminal occupation. This would imply there’s room for “beneficial” spillovers, i.e. crimes decrease both at the hot spots and nearby. Some literature in criminology suggests this is likely to be the usual outcome (Braga et al. 2014). But if the criminal market is not too concentrated, then criminals might just move one or two blocks away. These would be “adverse” spillovers.
How does your work help to identify spillovers and what were your main spillover-related findings?
ST: Our main point here, I believe, is that the empirical test to document spillovers is rather different from the test to assess direct treatment effects of place-based interventions. First, the question is no longer about average treatment effects but about aggregate treatment effects. Think about the following scenario: you have 100 hot spots, each of them had 10 crimes prior to the intervention. You treat half and crime disappears from these places. You have a very precise average treatment effect of -10 crimes. But what if each of these hot spots has 20 neighboring streets and the 10 original crimes displace proportionally to these 20 neighbors. You would likely find a largely imprecise positive spillover effect of +0.5. Temptation dictates that if the coefficient is not precise, then you found no evidence of spillovers. But this would be misleading, as the crimes are still there, just that you cannot see them precisely.
Second, in the presence of spillovers, you have a textbook violation of the independence assumption. Indeed, this is Chapter 7 of Gerber and Green (2012). If you have beneficial spillovers to control units, you underestimate direct treatment effects. If, otherwise, you have adverse spillovers to control units, you overestimate direct treatment effects.
To account for both issues, we designed the experiment to allow us to have a spread sample over the whole city and pre-specified the rules to identify the relevant spillover areas. In our main results, we also document aggregate effects in addition to the more standard average effects.
We found evidence of adverse spatial spillovers, with crimes appearing to rise in neighboring streets. In aggregate we believe we can rule out a citywide reduction of more than 2 to 3 percent in total reported crimes. More promising, however, we see that adverse spillovers are driven by property crimes. The evidence suggests that homicides and sexual assaults may have decreased by about 5 percent citywide.
What would you say are the main takeaways from this work, especially for policy-makers? What are open questions in this area that future research should address?
ST: Broadly, we add nuisance to what seems to be a policy consensus on the beneficial direct and spillover effects of hot spots policing on crime. Our results are not against the consensus, as we do observe these positive patterns for specific types of crimes. But rather suggest that it is too soon for a consensus, and that more likely, whether or not hot spots policing enhances aggregate welfare depends on the context.
More specifically, our results and the previous literature are consistent with the following propositions. If most crimes in a city do not have a sustained or pecuniary motive, then place-based interventions might have a large deterrent effect with a minimum of spillovers. If most crimes do have a sustained motive, then place-based strategies would be more successful if criminal rents in that city are concentrated and immobile, or if the supply of crimes is highly elastic to the probability of arrest at hot spots. This last case would require a criminal justice system that is swift and effective.
Becker, G. S. (1968). “Crime and Punishment: An Economic Approach.” Journal of Political Economy, 76, 169–217.
Braga, Anthony A., Andrew V. Papachristos, and David M. Hureau (2014). “The Effects of Hot Spots Policing on Crime: An Updated Systematic Review and Meta-analysis.” Justice Quarterly, 31, 633–663.
Ehrlich, I. (1973). “Participation in Illegitimate Activities: A Theoretical and Empirical
Investigation.” The Journal of Political Economy, 81, 521–565.
Gerber, Alan S. and Donald P. Green (2012). Field Experiments: Design, Analysis, and Interpretation. WW Norton.
Mejía, D., Ortega, D., Ortiz, K. (2015). Un análisis de la criminalidad urbana en Colombia. Technical report, CAF – Banco de Desarrollo de América Latina.
Weisburd, D. (2015). “The law of crime concentration and the criminology of place.” Criminology, 53(2), 133–157.
Wilson, J.Q. and G. Kelling (1982). “Broken Windows: The Police and Neighborhood Safety.” Atlantic Monthly, 249, 29–38.