Wording Effects in Assessment: Missing the Trees for the Forest

Abstract

This article examines wording effects when positive and negative worded items are included in psychological assessment. Wordings effects have been analyzed in the literature using statistical approaches based on population homogeneity assumptions (i.e. CFA, SEM), commonly adopting the bifactor model to separate trait variance and wording effects. This article presents an alternative approach by explicitly modeling population heterogeneity through a latent profile model, based on the idea that a subset of individuals exhibits wording effects. This kind of mixture model allows simultaneously to classify respondents, substantively characterize the differences in their response profiles, and report respondents' results in a comparable manner. Using the Rosenberg’s self-esteem scale data from the LISS Panel (N = 6,762) in three studies, we identify a subgroup of participants who respond differentially according to item-wording and examine the impact of its responses in the estimation of the RSES measurement model, in terms of global and individual fit, under one-factor and bifactor models. The results of these analyses support the interpretation of wording effects in terms of a theoretically-proposed differential pattern of response to positively and negatively worded items, introducing a valuable tool for examining the artifactual or substantive interpretations of such wording effects.