[EDGE] A grouped calibration test for logistic regression that tolerates a few corrupted records
EDGE is a new test for logistic regression calibration that uses a reliability table to check if predicted probabilities align with observed outcomes. It tests structured miscalibration by projecting residuals onto a smooth basis, providing an inferential p-value instead of just a descriptive statistic. This makes it more powerful than existing methods for detecting specific types of systematic errors.