Performance Evaluation of Exploratory Graph Analysis and Traditional Methods for Determining Optimal Number of Factors

• In Event: Machine Learning

Abstract

The current study evaluated the performances of traditional methods and an alternative method “Exploratory Graph Analysis (EGA)” to determine the optimal number of factors to retain using Monte Carlo simulation. Sample sizes, number of factors, number of variables per factor, factor loadings, and factor correlations were manipulated, resulting in 72 conditions in total. The accuracy of all methods tends to be negatively affected as the factor correlations increase. With large factor correlations, parallel analysis (PA) with principal axis factoring performs the best, followed by EGA; while with smaller factor correlations, EGA performs as well as PA with principal component analysis. Minimum average partial and its variant have lower accuracy than other methods.

Authors

• Wen-Lan Chen, National Yang Ming Chiao Tung University
  Presenting Author

• Meng-Ting Lo, National Yang Ming Chiao Tung University
  Presenting Author