<p>The rate of fatal shootings by officers is highly variable across American police agencies, yet the reasons for this variability remain unclear. This study uses Bayesian multilevel models to partition the variation in fatal police shooting rates across 2,727 police agencies between 2015 and 2020 into 3 sources of variability: randomness, differences in social context, and outlier agencies that have exceptionally higher or lower rates than is predicted by observed factors. The results indicate that variation in police shooting rates across agencies is far from random and is primarily associated with differences in social contexts, especially differences in crime rates, 911 call rates, officers per capita and the prevalence of guns. However, even after accounting for extensive differences in contexts, several police agencies consistently have lower or higher police shooting rates than predicted.</p>