Measuring Extreme Response Style

Abstract

Research on extreme response style (ERS) in rating scale responses has been characterized by conflicting findings and little agreement over how to assemble and validate ERS measures. This article proposes that, when ERS is defined as a proportion of extreme responses, an ERS measure will be more accurate if the items are uncorrelated and have equal extreme response proportions. Furthermore, appropriate stochastic models should be used to assess the internal reliability and convergent validity of these measures. An ERS measure is created and validated with this method, using items from a survey administered in 1975 and 1987 to large samples of U.S. adults serving on a consumer panel. We find that ERS is stable over a lengthy survey compared to a benchmark stability for a “perfect” measure. Furthermore, the distribution of ERS over this population is stable over time. Respondents' ERS is related to their age, education level, and household income but not to their gender.