Introduction

The Rosenberg Self-Esteem Scale (RSES) is designed to provide an evaluation of one's self-worth. Investigations on the dimensionality of the RSES do not result in unifactorial solutions. The direction of the item wording has been suggested as a reason for the contamination of its factorial structure. These methodological effects appear to be systematic response style tendencies, with predictable associations with individual difference variables.

Objectives

The aims of the current study are to (a) evaluate the presence of method effects due to wording in a German version of the RSES, (b) examine measurement invariance of the best fitting model for males and females, (c) compare latent factor mean scores of males and females and (d) examine the relationships of method factors with personality characteristics.

Methodology

Data from 346 adults participating in a longitudinal survey in Germany were used. Latent variable analysis procedures were employed to compare ten alternative model specifications for the RSES; multi-group invariance analysis for gender and structural equation modeling with personality variables as predictors of RSES latent factors were also carried out.

Results

A bifactor solution with one substantive and two specific factors related to positive and negative wording performed best out of the alternative models. Partial scalar invariance of the best model for gender was supported. Despite a non-significant gender difference in observed self-esteem mean scores, latent mean comparisons revealed a small advantage in favor of males, consistent with the literature, as well as a difference on the specific factor for positive wording. When personality variables are used as predictors of the latent factors, emotional stability was the strongest predictor of self-esteem and of both specific factors.

Conclusions

The results highlight the substantive nature of method effects and the need to utilize statistical remedies to control for them.