Disentangling substantive responses from faking: Statistical and practical performance of the Bayesian Retrieve-Edit-Select model
Bo Zhang · Illinois Digital Environment for Access to Learning and Scholarship (University of Illinois at Urbana-Champaign) · 2020
Faking has been a major concern for the use of personality tests in high-stakes conditions because it poses a threat to the validity of selection decisions. Despite decades of efforts, researchers have neither reached a consensual understanding of the nature of faking, nor come up with effective ways to identify fakers or correct for faking. This dissertation added to the faking literature a theory-based psychometric approach – the Retrieve-Edit-Select (RES; Böckenholt, 2014) model – that can provide faking-free estimates of personality. Specifically, I first derived four variants of the original RES model that may be better suited for personality tests and developed a Bayesian algorithm to estimate these models. Then I conducted a Monte Carlo simulation study (Study 1) and an empirical study (Study 2) to examine the statistical and practical performance of the RES models. Study 1 showed that the RES models, especially the faking-free personality scores, can be estimated accurately in most conditions. Study 2 employed a within-subjects design and two different samples (real job applicants and Mturk respondents). However, I did not find unequivocal support for the empirical utility of the RES model in the two empirical datasets. Potential reasons for the lack of support, limitations and future directions are discussed.