Using IRLS estimators to detect faking on personality inventories

Personality instruments have been widely used in both academic and commercial fields for general self-evaluation, job selection, and educational assessment. Although the stakes vary with these purposes, faking remains a serious concern in many decision contexts. The goals of this dissertation were to: (a) introduce a new method to detect faking using indices of atypical responses identified by a new procedure based on Iteratively Reweighted Least Squares (IRLS) factor model estimators, and (b) evaluate the effectiveness of this procedure in detecting either faking good or faking bad in comparison to other fake detection techniques that included scores from validity scales from a separate instrument, total scores from the targeted personality measure itself, and item response theory (IRT) based person fit indices.

A sample of 1270 respondents completed subscales from the Big Five Inventory (BFI; John, Donahue, & Kentle, 1991) and/or Bidimensional Impression Management Index (BIMI; Blasberg, Rogers, & Paulhus, 2014) under honest, fake-bad, or fake-good conditions. Results revealed that faking had marked effects on distributional characteristics of personality scores (means, standard deviations, skewness, kurtosis, intercorrelations, and factor structure). When considered collectively, fake detection scores achieved classification accuracy rates as high as 95.5% for detecting faking good and 98% for detecting faking bad. Validity scale scores, total personality measure scores, and indices derived from IRLS estimators were much more accurate in detecting faking than were IRT person fit indices with each performing best in particular contexts. Recommendations for future use of IRLS estimators for fake detection and other applications are discussed.