因子分析における感度分析 : uniterated PFAとiterated PFAの比較

概要

論文の詳細を見る
The present paper discusses sensitivity analysis in principal factor analysis (PFA). Numerical procedures studied are: (1) noniterative PFA based on correlation matrix with uities as the communalities; (2) noniterative PFA based on correlation matrix with SMC's as the communalities; and (3) iterative PFA. As a continuation of Tanaka, and Odaka (1988a) in which the corresponding influence functions were derived with respect to the iterative PFA, we propose theoretical influence functions I(x; LL^T) and I(x; Δ) for the common variance matrix T=LL^T and the unique variance matrix Δ, respectively, in the common factor decomposition Σ=LL^T+Δ with respect to the noniterative procedures (1) and (2) (cf. Tanaka and Odaka, 1988b). Through application to some numerical examples we discuss the validity or accuracy of the proposed influence functions and study the similarity and dissimilarity of the three procedures from the viewpoint of sensitivity.
日本計算機統計学会の論文
1988-05-19