Recovery and Replication of Internalizing and Externalizing Dimensions Within the Personality Assessment Inventory

Abstract

In this study, we examined the internal structure of 13 Personality Assessment Inventory (PAI; Morey, 1991, 2007) scales in a corrections sample (N = 1,099). Previous findings regarding the PAI internal structure have been somewhat inconsistent. We investigated the utility of a 2-dimensional model comprised of internalization and externalization to organize the 11 PAI clinical scales and 2 additional scales, Suicidal Ideation and Aggression. We randomly divided the sample, and a factor analysis revealed a 2-dimensional model representing internalization and externalization. Confirmatory factor analyses conducted with an independent subsample revealed acceptable fit when the model was revised to include correlated error terms between mood and anxiety disorder scales. The revised model exhibited acceptable fit when cross-validated, had better fit than a 1-dimension model, and demonstrated preliminary construct validity in relation to extratest variables.

Notes

¹ The sample from the forensic inpatient unit was not included in the normative sample reported in the PAI–CS (Edens & Ruiz, 2005) but was used for some reliability and validity analyses in the professional manual that accompanies the report.

² Bartlett's Test of Sphericity for the correlation matrix = 5592(78), p < .001.

³ The AIC is a goodness-of-fit index that takes into account model complexity. The AIC can be used to compare alternative models, with lower AIC scores indicating better fitting models.

⁴ We were concerned that excluding individuals with elevated NIM or PIM scores may have affected the results because validity scales can capture important variance related to psychopathology. Therefore, CFAs were also calculated with subsamples including individuals with elevated NIM or PIM scores who had been removed in the other analyses. Although the fit of the revised two-dimensional model (see model in Figure 2) was still acceptable for the majority of indexes, RMSEA values for the CFA in Subsample 1 were questionable (RMSEA = .104; 90% CI = .095–.113).

⁵ Factor scores obtained using regression methods have the advantage of combining information from all of the scales using empirically derived weights. However, PAI dimensional scores can be obtained by summing the T scores of the scales within each dimension. For example, an EXT score can be obtained by summing the T scores of the following scales: BOR, Aggression (AGG), Drug Problems, ANT, MAN, ALC, and PAR. Scale score sums were strongly correlated with the factor scores in the validation subsample (n = 610): Internalization r = .99, and Externalization r = .96. Differences in correlations between the two types of scores and external criteria were less than .01 (median [Mdn] = |.00|) for Internalization and ranged from |.01| to |.05| (Mdn = |.04|) for Externalization.

⁶ The base rates for the dichotomous variables were as follows: 93% positive for a history of drug use, 32% with a history of violence, 7% with a history of DUI, 8% with a history of health problems, 9% with a history of internalizing disorder, and 20% with a least one incident of institutional misconduct.

⁷ The magnitude of an AUC can be measured by its distance from .50 (see Swets et al., 2000). In a process similar to reflecting a correlation, the magnitude of AUC less than .50 (i.e., AUC = .30) can be used to transform the observed value to a value greater than .50 (i.e., AUC = .70). This transformation can ease the interpretation of the magnitude of the finding and does not alter the associated significance level. However, the substantive interpretation of the AUC should account for the original direction of the result.

^a Unstandardized residual scores controlling for EXT

^b Unstandardized residual scores controlling for INT

^c These AUCs are reflected; the original AUC values were .39 (EXT and Hx of internalizing disorder), .37 (INT and DUI), and .43 (INT and institutional misconduct)

*p < .05

** p < .01