Advancing the psychometrics of reverse-keyed items: Enriching cognitive theory by a logical and linguistic perspective

David Elek¹Hynek Cígler^1*David J. Grüning^2,3Stanislav Ježek¹

• ¹Psychology Research Institute, Faculty of Social Studies, Masaryk University, Brno, Czechia
• ²GESIS - Leibniz-Institut fur Sozialwissenschaften eV, Mannheim, Germany
• ³Universitat Heidelberg, Heidelberg, Germany

Inclusion of reverse-keyed items in a questionnaire usually impacts its factor structure and reliability. Therefore, their presence or absence also affects measurement validity, yet a clear consensus on their use is missing. In this paper, we provide an overview of the literature on the use of reverse-keyed items. We outline the typical arguments for and against their use, along with the cognitive explanatory framework commonly used to account for the associated issues. We further argue that while the cognitive model of responding is theoretically meaningful, it cannot, on its own, identify specific error sources from reverse-keyed item sets, and that issue lies in the implicit assumption regarding how reverse-keyed items should function. Furthermore, we note that literature on reverse-keyed items is at an impasse, with conflicting recommendations and inconclusive results. As a solution, we introduce a logical and linguistic perspective to advance our understanding of reverse-keyed items. This perspective allows researchers to understand that response inconsistencies in a statistical model do not necessarily indicate logically inconsistent answers from the respondent. Enriching the cognitive model with a linguistic perspective, which has been missing in psychometric literature, allows us to differentiate between mere statistical and actual response inconsistency. Based on this combination of cognitive and linguistic theory, we advance the historical analysis of response bias by suggesting that future research should closely draw from linguistic concepts to arrive at a promising explanatory framework that can then better inform modeling decisions. However, further empirical studies are needed to test our hypotheses and evaluate the magnitude and relevance of the proposed linguistic effects.