Using bifactor models to identify faking on Big Five questionnaires

Abstract

To identify faking, bifactor models were applied to Big Five personality data in three studies of laboratory and applicant samples using within‐subjects designs. The models were applied to homogenous data sets from separate honest, instructed faking, applicant conditions, and to simulated applicant data sets containing random individual responses from honest and faking conditions. Factor scores from the general factor in a bifactor model were found to be most highly related to response condition in both types of data sets. Domain factor scores from the faking conditions were found less affected by faking in measurement of Big Five domains than summated scale scores across studies. We conclude that bifactor models are efficacious in assessing the Big Five domains while controlling for faking.

Effect of job applicant faking and cognitive ability on self‐other agreement and criterion validity of personality assessments

“…The significant increase in mean absolute inter-scale correlations in the applicant context (r = .20 to r = .34) was consistent with previous research (Ellingson et al, 2001;Schmit & Ryan, 1993;Zickar & Robie, 1999). However, although the increase in the first unrotated factor in principal component analysis (17.0% of variance to 21.6% of variance) was consistent with previous research conducting this type of analysis (Schmit & Ryan, 1993;Ziegler & Buehner, 2009), the modest size of the increase is perhaps less suggestive of an idealemployee factor (Hendy et al, 2021;Klehe et al, 2012;Schmit & Ryan, 1993) than of a social desirability factor that increases slightly in contexts encouraging faking.…”

“…Fourth, the first unrotated principal component (Schmit & Ryan, 1993; Ziegler & Buehner, 2009) and inter‐scale correlations (Ellingson et al, 2001; Schmit & Ryan, 1993; Zickar & Robie, 1999) tend to be larger in applicants, consistent with responses being influenced not only by substantive item content, but also by the perceived desirability of responses to an employer. Some argue that this impact on responses creates an ideal employee factor in factor analysis (Hendy et al, 2021; Klehe et al, 2012; Schmit & Ryan, 1993). All of the above patterns were expected in the current study, however, in particular, the following hypothesis was proposed:…”

Comparing forced-choice and single-stimulus personality scores on a level playing field: A meta-analysis of psychometric properties and susceptibility to faking.

“…Given their reduced susceptibility to the effects of faking, FC scores are likely better suited for use in high-stakes settings than SS measures. Faking is harmful to test validity and score interpretation (Christiansen et al, 2017(Christiansen et al, , 2021Hendy et al, 2021;Holden, 2008;Jeong et al, 2017;Komar et al, 2008;Peterson et al, 2011;Schmit & Ryan, 1993), and thus, using a format with greater resistance to faking is desirable.…”

“…Personality traits assessed on these inventories explain important differences in work behavior and are useful predictors of job-related outcomes (e.g., Christiansen & Tett, 2013;Gatewood et al, 2016;Judge et al, 2013;Shaffer & Postlethwaite, 2012;Tett & Christiansen, 2007;Zimmerman, 2008). However, there have long been concerns over the use of personality tests in highstakes settings because such assessments are easily faked (e.g., Griffith & Robie, 2013), resulting in score distortion and decrements in validity (Christiansen et al, 2017(Christiansen et al, , 2021Hendy et al, 2021;Holden, 2008;Jeong et al, 2017;Komar et al, 2008;Peterson et al, 2011;Schmit & Ryan, 1993). Concerns over faking have led to exploration of measurement methods that are more resistant to the effects of faking.…”

Contribución de los Modelos Factoriales Confirmatorios a la Evaluación de Estructura Interna desde la Perspectiva de la Validez

“…Este escenario se aleja de la unidimensionalidad esencial en favor de la multidimensional, y el interés puede estar en obtener valores solventes para las subescalas (VCES y S), así como obtener puntuaciones factoriales de subescala más depuradas una vez se ha controlado la varianza común a todos los ítems (i.e., factor general). Este tipo de aplicación es más novedosa y está recibiendo atención, por ejemplo, en contextos en los que se asume la presencia de un factor general de engaño en las respuestas (Hendy et al, 2020).…”