STRONGLY (DIS)AGREE: DISTORTING EFFECTS OF EXTREME RESPONSE STYLE IN SURVEY-BASED EXTREMISM RESEARCH

Abstract

This dissertation investigates whether a common but often overlooked source of bias, extreme response style (ERS; a tendency for survey responders to consistently select endpoint options on Likert-type scales), distorts findings in the survey-based empirical research at the social identity–extremism (encompassing conspiracy theories, mis/disinformation, and violent extremism) nexus using a three-study design. The first study uses bibliometric coupling analysis to map the intellectual structure of core empirical publications at this nexus, identifying three thematically distinct but structurally interconnected research clusters organized by intellectual orientations. The second study evaluates the psychometric properties of three focal studies in this bibliometric network—Bertin et al. (2022), Sternisko et al. (2023), and Ozer et al. (2020)—identified through a stratified influence-based sampling procedure. Tau-equivalence assumption was rejected in all three datasets, convergent validity failures appeared in scales central to each study's substantive claims, and discriminant validity concerns were identified for construct pairs that occupy adjacent positions in the original models. The final study replicates and extends these three studies by applying bifactor latent class factor analysis (LCFA) to separate response style variance from substantive trait variance. The results show that ERS correction simultaneously reverses, amplifies, and attenuates key statistical effects within the same study while in some cases completely reorganizing correlation matrices, altering how variables central to studies in this area relate to one another. These findings carry direct implications for communication theory and practice. Communication research is starkly absent among the most influential studies in this literature despite the communicative nature of these phenomena. These findings also demonstrate that the communicative processes central to extreme belief formation—conspiracy narratives, mis/disinformation, violent extremism—have been studied using measurement tools whose psychometric foundations are more fragile than previously recognized, and whose substantive conclusions are, in some cases, artifacts of unmodeled response style variance. The project discusses broader questions about the validity of psychometric assumptions underlying standard survey-based research practices in the social sciences, appropriate psychometric approaches to account for the systematic variance transmitted through measurement instruments, and the reliability of statistical conclusions that drive theoretical models and policy decisions.