ABSTRACT
We present results of our winning solution to the National Institute of Justice recidivism forecasting challenge. Our team, "MCHawks," placed highly in both terms of accuracy (as measured via the Brier score), as well as the fairness criteria (weighted by differences in false positive rates between White and Black parolees). We used a non-linear machine learning model, XGBoost, although we detail our search of different model specifications, as many different models' predictive performance is very similar. Our solution to balancing false positive rates is trivial; we bias predictions to always be "low risk" so false positive rates for each racial group are zero. We discuss changes to the fairness metric to promote non-trivial solutions. By providing open-source replication materials, it is within the capabilities of others to build just as accurate models without extensive statistical expertise or computational resources.
ABSTRACT
INTRODUCTION: Among the factors related to survival among individuals with gun shot wounds (GSW) is distance to trauma care. Relatively little is known about neighbourhood-level patterns of GSW mortality and distance to hospitals with trauma centres. This study focuses on distance to the nearest trauma centre as a correlate of survival among GSW victims. METHODS: Fatal and non-fatal shooting incident data for 9,205 victimisation in Detroit, Michigan between 2011 and 2017 were collected. A Bayesian conditional autoregressive model was utilised to estimate block-group levels of GSW mortality. Clustering techniques were used to identify spatially proximate neighbourhoods with higher or lower than expected rates of GSW mortality. RESULTS: Distance to the nearest trauma centre was associated with a 22% increase in fatal outcomes, per-mile (OR 1.22, 95% CI, 1.06 to 1.40) after adjusting for block-group level covariates. A Getis-Ord Gi* analysis identified 91 block groups with lower than expected rates of GSW mortality and 12 block-groups with higher than expected rates. CONCLUSION: Distance to the nearest trauma centre is associated with GSW victim survival. Clusters of block-groups with below-average GSW mortality were observed within close proximity of major trauma centres in Detroit. Improving speed and access to trauma care may play a role in reducing GSW mortality.