Most practitioners believe that the best way to detect bias in a test is to develop regression equations for the majority case (usually White males in the case of pilots) and for each minority in question (i.e. White females).  Typically, test scores are regressed on some measure of performance (i.e., score on the final checkride at the end of training). The slope and the intercept of the equations are then compared.  Differences in slope indicate that the test is not equally valid for all of the groups and should be dropped.

This question is addressed more completely in the document entitled “Legal Issues in the Design of Selection System.”