Confidence Region Estimation

Confidence region estimation studies how to construct sets that contain an unknown parameter or future outcome with a specified probability, making decisions more reliable than point estimates alone. Approaches include conformal prediction and other distribution-free methods, bootstrap and asymptotic approximations, and self-normalized or robust procedures for dependent, private, or heavy-tailed data. Methods are increasingly tailored to settings such as reinforcement learning, selective classification, and inference with predicted or missing labels, where data collection and dependence affect coverage. A central tension is obtaining valid coverage under weak assumptions while keeping regions small and computation practical.