Abstract
Response styles, such as the tendency to prefer extreme or midpoint response options, are a systematic source of variance in self-report data. Individual differences in response styles are largely consistent across traits and over time, but it is unknown whether they generalize across Likert-type scales and visual analogue scales (VAS). In a preregistered within-subjects study, N = 501 participants completed the HEXACO-PI-R, with two content-balanced item subsets presented on a 7-point Likert-type scale and on a VAS. We modeled extreme and midpoint responding as opposite poles of a single dimension, referred to as extreme response style (ERS), using conceptually aligned models: a partial credit model for the Likert-type items and a beta item response model for the VAS items, each extended by a person-specific ERS parameter. Estimating both models jointly in a Bayesian framework, we found that ERS on Likert-type scales and on VAS was substantially correlated (ρERS = .76, 95% CrI [.68, .83]). Individual differences in ERS thus generalized to a considerable extent across response formats. Choosing VAS instead of Likert-type scales therefore does not avoid the influence of response styles, which should be modeled in VAS data as well.