This paper was interesting, because it reported on a field experiment in New York City:
The field experiment described in this paper was conducted in the Spring and Summer of 2016 in NYC. Through a unique partnership between NYPD and MOCJ, we randomized the provision of street lights to the city's public housing developments, allowing us to avoid the potential challenges that could result due to spurious time trends as well as selection bias...
In order to select developments for the study, NYPD provided a list of 80 high-priority developments based upon their elevated crime rates and perceived need for additional lighting from among the 340 NYCHA developments in NYC. From this list, we randomized 40 developments into a treatment condition that would receive new lights and 40 developments into a control condition via paired random sampling, stratifying on each development's outdoor nighttime index crime rate and size in the two years prior to the intervention; treatment developments were then randomly assigned a lighting dosage.Most prior studies have simply compared areas with more lighting with areas with less lighting. However, there is selection bias that is not accounted for, because areas with more lighting may differ from areas with less lighting, in ways that also affect crime. This study gets around that problem because the amount of lighting increase is randomised.
Chalfin et al. looked at the impact on night-time crime, and found that:
Accounting conservatively for potential spillovers, lighting reduces outdoor nighttime index crimes by approximately 36 percent and reduces overall index crimes by approximately 4 percent in affected communities, an outcome which is likely to be cost-beneficial, should the impact of lighting persist over time.To be clear, the effects noted above are for a doubling of street lighting intensity. So, if you double streetlight intensity, you reduce outdoor night-time crime by 36 percent.
Our discussion in the EDG session highlighted a potential source of bias in the analysis, which the authors have only partially addressed. If you increase lighting in some areas, but leave other areas darker, some criminals will relocate their criminal activities from the well-lit areas to the less-well-lit areas. This will decrease measured crime in the well-lit areas, but increase measured crime in the less-well-lit areas. If the less-well-lit areas are the control areas for your analysis (and the well-lit areas are the treatment areas), then this displacement of crime will bias your comparison of treatment and control areas upwards. In other words, the impact of street lighting on crime will be overstated in this analysis.
I hadn't picked up the full implications of that point in my reading of the paper, so well done to the EDG students. Chalfin et al. did attempt to account for 'spillover effects' in their analysis:
Estimates are reported for on-campus crimes and, in order to test for spatial spillovers, for crimes that occur within a radius of 550 feet (two standard NYC blocks) from campus. While we do not detect evidence of spillovers...So, they also test for whether crime increase outside of the treatment areas (but within two city blocks), but the effects they find are tiny. However, this attempt to detect spillovers will only pick up spillovers into neighbouring city blocks. So, if criminals relocate further than a block away from the well-lit area, the analysis isn't going to pick it up.
Finally, they find that day-time crimes also reduced (by 25% net of spillovers) in the treatment areas, although the impact is pretty imprecisely measured and is not statistically distinguishable from zero. This last point should make us very wary of over-interpreting the importance of the results from this paper. So, while the field experiment is a good approach, and certainly an advance on the previous literature, this doesn't provide strong evidence for the impact of street lighting on crime.
[HT: Marginal Revolution]
No comments:
Post a Comment