Bifactor Model Is Not the Best-Fitting Model for Self-Esteem: Investigation With a Novel Technique.

Abstract

In the measurement of self-esteem, previous research assumes that all respondents are qualitatively similar. The assumption has not been adequately tested. The current study examines its validity using factor mixture modeling. Results reveal two qualitatively distinct classes: the first provides more consistent responses to positive self-esteem items than the second. The correlations between positive and negative self-esteem suggest that self-esteem is essentially unidimensional in the first class but bidimensional in the second. Furthermore, those with high self-esteem are more likely to belong to the first class; those with low self-esteem are more likely to belong to the second class. The observed dimensionality of self-esteem depends on a person's level on the trait. Finally, we found that the two-class solution fits the data much better than a simple one-class, two-factor solution or a bifactor solution. Psychometric researchers should no longer ignore the possible existence of qualitatively distinct groups in an underlying population. We include Mplus syntax together with a detailed explanation for researchers to conduct similar investigations on constructs of interest.