Factor Analysis of Ordinal Variables: Polychoric and Pearson correlations in the exploratory approach

Abstract: This thesis considers the problem of applying exploratory factor analysis to data obtained though the Likert scale. It is often the case that this sort of data is treated as an interval level of measurement and used in analyses that require continuous variables, although given the categorization and nature of the scale, it should rather be treated as of ordinal level. Exploratory factor analysis is a prevailing technique for studying the construct validity of data, a method that relies on the correlation matrix of the data to obtain factor solutions. Previous research recommend the use of polychoric correlations as opposed to the common Pearson correlation method when attempting to apply factor analysis to ordinal data. Preceding research is complemented with further consideration of asymmetry in the data. The results, obtained through simulation studies, show that the polychoric correlations provide a more accurate reproduction of the theoretical model used to generate the data in three cases of item skewness.