Scoring the Big Five for longitudinally assessed academic achievement predictiveness: Manifest, correlated-factors model, and bifactor modeling across multiple contexts

Based on a large (N = 612) longitudinal sample in a teacher education program, we compared how three methods of personality scoring—manifest mean scores, correlated-factors model scores, and bifactor model scores—predict academic achievement assessed by grade point averages. Furthermore, we compared predictiveness across honest responses, applicants' responses and responses collected under laboratory faking-good instructions. To this end, a real-life selection setting was part of our study (i.e., applicants to initial teacher education selected, among other things on their personality). We found the expected pattern of manifest mean scores (honest responses were the lowest, applicants' responses higher and faking-good responses highest) and could demonstrate that applicant faking does not reduce personality assessment's predictiveness. Overall, correlated-factors model scoring increased the predictiveness of honest and applicants' responses, and scoring via bifactor model even more so. No method of scoring could retrieve the predictiveness in the faking-good response condition. Regarding the practical application within selection processes, bifactor model scores only slightly outperformed mean scores, and this only occurred in the case of small selection ratios. Nevertheless, we showed that there is criterion-related and systematic variance within applicants' personality scores above and beyond their personality traits that can be extracted when modeled with bifactor models.