The Insufficiency of Statistics for Detecting Racial Discrimination by Police

Abstract Benchmark tests are employed when testing for racial discrimination by police. Neil and Winship (2019) emphasize that such tests are threatened by Simpson’s paradox, but avoid analyzing the paradox causally. They consequently cannot elucidate the link between statistical quantities and discrimination hypotheses. Simpson’s paradox reveals that the statistics given by benchmark tests are not invariant to conditioning on additional variables. On this basis, I argue that benchmark statistics should not by themselves be taken to provide any evidence regarding discrimination, absent additional assumptions. Causal models can represent these assumptions.