Understanding when mixture models effectively identify careless responding: Simulation and validity evidence

Abstract Careless and insufficient effort responding (C/IER) in self-report questionnaires refers to responses provided without adequate attention to item content. Latent variable mixture models have recently been proposed as a promising approach for detecting and accounting for C/IER. This study systematically evaluates the performance of mixture modeling for C/IER detection. Using an exemplar mixture model that captures core features of existing approaches, we (a) investigated, via simulations, conditions that facilitate or impede separation between response patterns indicative of attentive versus careless responding and (b) examined the validity of interpreting the latent class variable as an indicator of C/IER in an empirical application. The simulation study assessed how questionnaire and item characteristics affect latent-class identification under different C/IER response patterns. Results indicate that reliable separation is achieved when scales comprise at least ten items, exhibit substantial threshold heterogeneity, and include both positively and negatively worded items. Empirical support for the C/IER interpretation of the latent class variable was obtained by reanalyzing a publicly available Big Five inventory data set collected across multiple platforms. Model-based classifications closely corresponded to alternative C/IER indicators that do not rely on response patterns, exhibited consistency across multiple scales, and were sensitive to platform-specific differences in C/IER prevalence. Overall, the results support mixture modeling as a viable approach for addressing C/IER under many conditions commonly encountered in applied research.