Modeling Faking in High-Stakes Personality Assessments

Self-report personality tests are widely used in high-stakes assessments such as personnel selection. Usually, test-takers are instructed to make self-judgments about their personality, which holds the risk that test-takers deliberately distort responses to present themselves in an overly favorable manner (i.e., engage in faking). To model heterogeneity in the degree to which test-takers fake, the multidimensional nominal response model (MNRM) in the context of item response theory (IRT) can be used. This faking modeling approach accounts for a continuous faking dimension reflecting test-takers’ tendency to edit responses according to item-specific effect patterns of faking. However, it assumes that there is variation in desirability characteristics across items and that all test-takers consider such item-specific desirability characteristics in their responses. In this research, consequences of homogeneous item desirability characteristics are examined with regard to person parameter recovery in different conditions. Also, a mixture modeling extension is presented to allow for qualitatively different faking tendencies between test-takers.