Intermittent Faking of Personality Profiles in High-Stakes Assessments: A Grade of Membership Analysis

In high stakes assessments of personality and similar attributes, test takers may engage in impression management (aka faking). This paper proposes to consider responses of every test taker as a potential mixture of ‘real’ (or retrieved) answers to questions, and ‘ideal’ answers intended to create a desired impression, with each type of response characterized by its own distribution and factor structure. Depending on the particular mix of response types in the test taker profile, grades of membership in the ‘real’ and ‘ideal’ profiles are defined. This approach overcomes the limitation of existing psychometric models that assume faking behavior to be consistent across test items. To estimate the proposed Faking-as-Grade-of-Membership (F-GoM) model, two-level factor mixture analysis is used, with two latent classes at the response (within) level, allowing grade of membership in ‘real’ and ‘ideal’ profiles, each underpinned by its own factor structure, at the person (between) level. For collected data, units of analysis can be item or scale scores, with the latter enabling analysis of questionnaires with many measured scales. The performance of the F-GoM model is evaluated in a simulation study, and compared against existing methods for statistical control of faking in an empirical application using archival recruitment data, which supported the validity of latent factors and classes assumed by the model using multiple control variables. The proposed approach is particularly useful for high-stakes assessment data and can be implemented with standard software packages.