A Bayesian hierarchical joint modeling approach of person and item features that contribute to response bias

Traditionally, extreme responding has been recognized as a bias in which individuals tend to select the highest or lowest response categories on self-report items, regardless of item content. This bias is believed to occur independently of the desirability of the trait or state described in the item statement. However, little effort has been made to empirically test this assumption. In this study, we propose a novel approach to control for both extreme (ERS) and socially desirable responding (SDR) in self-report questionnaires. We considered judges' ratings of item social desirability to control for SDR by employing Bayesian hierarchical and joint modeling. We also considered the effect of ERS to be multiplicative rather than additive, assuming a noncompensatory role to ERS. The advantage of this approach is mostly computational. Overall, our results indicate that modeling the impact of ER improved the model fit on measures of affect and pathological traits. However, contrary to expectations, our analysis did not indicate any substantive influence of social desirability on items’ difficulty. In conclusion, our findings favor the perspective of both ER and social desirability as true traits rather than merely nuisance factors, evidence that our approach may assist researchers in gaining a deeper understanding of response biases in self-report items.