Date of Award

Spring 2023

Document Type

Open Access Dissertation



First Advisor

Alberto Maydeu-Olivares


Self-reports (SRs) of typical behavior are often the only existing feasible method to gather data on important drivers of human performance. In applications such as personnel selection, SRs are vulnerable to intentional distortions, often referred to as faking. A review of the literature suggests that so far, the methods proposed to address faking are unsatisfactory. In a recent breakthrough, Pavlov et al. (2019) showed that high-stakes scale scores are best modeled as a function of a) propensity to fake, b) honest scores, and c) the interaction of these two terms. Pavlov et al. did not, however, propose any method to extract honest scores in assessment settings, when only high stakes data is available. In this dissertation I investigate using factor mixture modeling (FMM) with class specific intercepts and factor loadings to this aim. I assume that responses to high stakes items are a function of the respondents’ “honest” factor scores on the attribute being measured, an unobserved categorical “tendency to fake” latent class, and their interaction. I perform simulations using parameter estimates based on Pavlov et al.’s data to determine the extent to which factor scores estimated using a two class factor analysis model outperform the current standard, a single class model. Results suggest that only under specific conditions FMM scores provide higher correlations with true factor scores compared to single class models. Moreover, empirical findings indicate that the theoretical potential of FMM to detect faking is not realized in practice, where class separations are not as defined.


© 2023, Raul Corrêa Ferraz