Accounting for response styles in rating data using IRT and IRTree models: A tutorial in R with applications to self-reported metacognitive competence.

Response styles refer to individual differences in the use of the item response format irrespective of item content. They include tendencies to favor or avoid midpoint or extreme response categories, which can distort the measurement of latent constructs. This tutorial provides an overview of IRT and IRTree models that accommodate response style factors, allowing researchers to analyze response-style effects and adjust latent-trait estimates for potential bias. In particular, we give a practical guide to using IRT and IRTree models for response styles and illustrate the workflow of model implementation, estimation, and interpretation with an empirical data set on self-reported metacognitive competence (Double, 2025). Throughout the tutorial, IRT and IRTree models are specified with standard R tools, and parameter estimation is accomplished in the common framework of multidimensional IRT with the R package mirt (Chalmers, 2012; Chalmers et al., 2025). The empirical results demonstrate that IRT and IRTree models can disentangle substantive traits and response styles, and that model-based estimates of substantive traits are adjusted accordingly. More specifically, both model classes reveal pronounced correlations among latent trait dimensions and among response style factors, but only small correlations between traits and response styles. In addition, the model-based trait estimates show the expected patterns of amplification and shrinkage after adjusting for midscale and extreme response style. The tutorial concludes with a discussion of recent developments in psychometric modeling.