Psychometric properties of the Portuguese version of the Personality Assessment Inventory

Although several factor analytic studies have investigated the factor structure of the Personality Assessment Inventory (PAI), no consensus has been reached regarding the best factor solution, either considering the 11 clinical scales or the 22 full scales. Whereas some studies have found that a two-factor solution for the clinical scales and a four-factor solution for the 22 scales were the most parsimonious factor structures, other studies suggested different factor models (e.g., a three-factor solution for the 11 clinical scales, and a three- or five-factor solution for the 22 full scales). Some reasons may explain the different factor structures found in the literature, namely sample characteristics (e.g., community vs. clinical samples) and methodology (e.g., number of scales included in the factor analysis, validity criteria, factor analysis techniques, and estimators).

The present study aimed to investigate the factor structure of the Portuguese version of the PAI in a sample of 900 participants (aged 18-74 years, 57.7% female). Following the methodology proposed by other PAI factor analytic studies, we first conducted a CFA at the item level, for each individual scale, to test the unidimensionality of the 22 scales. Then, CFAs at the scale level were performed to examine the fit of the factor structure proposed by Morey (1991) for the 11 clinical scales (i.e., two-factor solution obtained in the community standardization sample) and the 22 full scales (i.e., four-factor solution) of the Portuguese version of the PAI. Additionally, we tested other competing factor structures and the three-factor solution for the 11 clinical scales found in Morey's (1991) clinical standardization sample. To our knowledge, this is the first study that examined several competing factor models through CFA, aiming to find the most parsimonious factor solution of the PAI.

CFAs at the item level suggested an adequate model fit for almost all scales. The results supported a bifactor model with two first-order factors for the 11 clinical scales (the bifactor model with three first-order factor also revealed an adequate fit) and a three-factor model for the 22 full scales. Our findings suggested that each clinical scale is simultaneously accounted for by its specific factor (internalization or externalization factors) and a broad general psychopathology factor.

Implications and directions for future research on PAI's dimensionality are discussed. In a practical manner, the present study may aid practitioners to better understand the psychological functioning of their patients resorting to the PAI. From PAI's results, technicians can adapt their intervention programs, both in clinical and forensic settings.