1 Factor Analysis Using PCA and GLS Methods for Dichotomous Data

2 This article compares two methods, Principle Component Analysis (PCA) and Generalized Least Square (GLS), using in dichotomous variables. This study uses R to simulate data, Tetcorr to compute tetrachoric correlation coefficient, and SPSS to do PCA and GLS, and hope to find which method can estimate factor loadings more accurate in different sample sizes and different numbers of factor. This article gives some advises and suggestions in dichotomous variate analyses for further studies. The conclusions of this study are as follow: 1. In one-factor models, GLS method is more fitted to the original data than PCA. 2. In two-factors models, when the loadings are greater than 0.5, GLS method and PCA methods are very closed. When the loadings are less than 0.5, the loadings estimation of GLS is more fitted than PCA