OP XII – 5 Short-term effects of daily weather characteristics on violent crimes in the boston area

Background/aim Recently, attention has been given to the impact of climate change on crimes through weather variations. Associations between weather variables and crimes have been reported but causality still needs to be addressed. Our goal is to investigate whether there exists a causal relationship between temperature and violent crimes as well as between precipitations and violent crime. Methods Since our study is based on data that were not collected in a randomised experiment, it is necessary to incorporate a design stage before any causal analysis stage. Our approach is to design observational data in a way that approximates a randomised experiment, using matched-sampling strategies. The framework considered in this paper is often denoted as the ‘Rubin Causal Model’ and sees causal inference as a missing data problem. Propensity score matching enables us to reconstruct four hypothetical randomised experiments before estimating the average causal effect (ACE). The ACE can be interpreted as the average number of daily violent crimes that are assumed to be caused by a high exposure level as compared to a lower exposure level. Results In this manuscript, we observe that changes in heat index (apparent temperature) or the occurrence of rainfalls were followed by changes in daily violent crimes in Boston between July 2012 and February 2017. For instance, while we found increased daily crimes on temperate days compared to cold days (2.33, 95% CI=[1.56; 3.09]), it was not the case when comparing extremely hot days to hot days (0.59, 95% CI=[−1.36; 2.68]), suggesting a ‘plateau’ in the dose-response of heat index and violent crimes. As expected, the occurrence of rainfalls tended to decrease crimes (−1.45, 95% CI: −2 to −0.91). Conclusion These results suggest that: meteorology should be used to prevent future crimes in Boston, and the weather–crime relationship should be taken into account when quantifying the impact of climate change. We want to encourage researchers to use matching strategies when only observational data is available and no randomised experiment can be conducted but causal estimates need to be quantified.