A Study on Influential Subset Detection in Regression

This study is suggested tn find a methodology of influential subset detecting by using diagonal elements of the hat matrix, $h_{ij}$, and off-diagonal elements of the modified hat matrix, $h_{ij^*}$. And this study shows that elements of two matrix is very useful to detect a masking subset in various situations by analytic and emperical explanations. Also a desirable procedure to detect the significant subset and the calibration points for $h_{ij}$ and $h_{ij^*}$ are recommended.