2023 Volume 50 Issue 2 Pages 61-79
Estimation, statistical testing, and model selection are the main focusarea in statistics. This study focuses on the relationship betweenestimation precision (Fisher information) and model selection(Kullback—Leibler or KL information). This relationship is importantbecause researchers often conduct estimations or statistical tests aftermodel selection. Additionally, this study examines how ComputerizedAdaptive Testing (CAT), a stimulus selection method that maximizesFisher information, affects model selection performance. A simulationstudy demonstrates the relationship between the difference in Fisherinformation between two models and the degree of asymmetry in KLinformation. Furthermore, we confirm that controlling Fisher informationusing stimulus selection can influence model selection performance. Asimulation study suggests that increasing the Fisher information of afalse model reduces model selection performance.
