This project studies the assignment of cases to immigration judges in the U.S. Over the last twenty years, immigration courts have faced rapid growth in the number of cases to be heard, creating time pressure that might affect the allocation of cases to judges and the decisions of judges on a given case. This draft takes a first step in examining these issues by testing the exogeneity of case assignment. We calculate the stringency of a judge in a given case using this judge’s decisions in other cases, and we find a highly statistically significant relationship between this measure and baseline characteristics such as the non-citizen’s language, suggesting bias in current estimates of judge effects which is a widely used identification strategy. The paper highlights the need for methodological tools capable of detecting, modeling, or correcting for non-random assignment in courts. We provide guidance for identifying quasi-random subsets of cases for future causal analysis.