Forward Search Algorithm for Robust Influence Analysis in Maximum Likelihood Factor Analysis

Mainly in regression analysis, numerous methods have been proposed historically for the analysis of the influence of single or multiple observations on the results of analysis. Such a sensitivity or stability problem is not special to the regression analysis, but is common to the other statistical methods including the multivariate methods. The authors combined the general procedure of the sensitivity analysis and the forward search method to detect the influential observations without suffering from the masking and swamping effect, and compared its performance with the other robust methods numerically. The proposed procedure can be applied to any multivariate methods with minor modification. In this paper they propose and discuss the procedure in Maximum Likelihood Factor Analysis (MLFA).