Robust exploratory factor analysis

Abstract

In classical factor analysis, a few outliers can bias the factor structure extracted from the relationship between manifest variables. As in least-squares regression analysis there is no protection against deviant observations. This paper discusses estimation methods which aim to extract the “true” factor structure reflecting the relationships within the bulk of the data. Such estimation methods constitute the core of robust factor analysis. By means of a simulation study, we illustrate that an implementation of robust estimation methods can lend considerable improvement to the validity of a factor analysis.