1 Psychometric comparison of the long and short forms of the Personality Assessment Inventory in a neuropsychiatric population

Abstract

Objective:

The Personality Assessment Inventory (PAI; Morey, 1991; 2007) is a 344 item self-report measure of personality, psychopathology, and factors affecting treatment. The PAI short form (PAI-SF) contains the first 160 items of the PAI and is often favoured as a screening tool or brief version to mitigate respondent burden and fatigue. The PAI has been psychometrically validated among numerous populations (Slavin-Mulford et al., 2012), while psychometric research on the PAI-SF is gradually emerging. The psychometric properties of the PAI-SF range from adequate to strong in psychiatric (Sinclair et al., 2009), forensic (Sinclair et al., 2010), outpatient and nonclinical (Ward et al., 2018), and stroke (Udala et al., 2020) samples. To advance research validating the PAI-SF among diverse populations, this project investigated the psychometric comparability between the PAI and the PAI-SF in a neuropsychiatric population. Based on previous literature, it was hypothesized that the PAI-SF would produce congruent results to the PAI in this sample.

Participants and Methods:

For this study, participant files (N=214) were collected retrospectively from short- and long-term residential psychiatric and substance use treatment facilities in Minnesota for patients with neurological and cognitive concerns referred for neuropsychological evaluation. The PAI-SF was scored using the first 160 items from a patient’s long-form PAI protocol. To determine the psychometric comparability of long- and short-forms, paired-samples t-tests, intraclass correlations, and percent agreement in clinical classification between forms were analyzed.

Results:

Analyses of participant data found that intra-class correlations ranged from .87 to .98 for each subscale on the PAI when compared to the PAI-SF, demonstrating good to excellent reliability between forms. Symptoms are considered clinically elevated when they exceed the clinical significance threshold for a subscale (typically a T-score of 70+). Agreement between the PAI and PAI-SF subscales in the classification of clinically elevated scores ranged from 86% to 100%. When forms did not agree, the PAI-SF was more likely to be clinically significant relative to the PAI. A comparison of subscale means between forms was examined by independent samples T-tests with a Bonferroni correction. Results revealed significant differences between the PAI and PAI-SF on one validity scale (Negative Impression Management), three clinical scales (Anxiety; Depression; Antisocial Features), and one treatment scale (Treatment Rejection).

Conclusions:

Results demonstrated that the PAI and PAI-SF have high reliability between forms in a neuropsychiatric population. Although mean scores differed on a small number of subscales between the PAI and PAI-SF, differences did not appear sufficiently large enough to shift clinical classifications, as the two forms performed similarly in their identification of clinically elevated scales. Findings align with previous literature and suggest that the PAI-SF may perform adequately in a neuropsychiatric population if brevity or participant burden is of concern. However, caution is warranted when making clinical decisions with the PAI-SF as more research is needed.