Previous work has examined PAI response patterns on those with mild traumatic brain injury (mTBI), but these research efforts have faced a number of notable challenges. Accordingly, findings between studies have often contradicted. There are a variety of reasons for these contradictions including, but not limited to, that prior research has: (1) not excluded individuals from analysis with failed validity testing, meaning that data analyzed is likely not fully valid/reliable, (2) the inappropriate use of item-level grouping analyses (factor analysis) to identify groups of participants, which is better suited to cluster/profile analysis, (3) insufficient sample sizes to conduct any of the analyses undertaken (i.e., far fewer participants per observed variable analyzed than required; see Brown, 2015), (4) interpretation of scale means that are entirely normative (i.e., T-score mean of 50, corresponding to the normative sample’s mean) as indicating a clinical pattern, and (5) use of analyses without fit statistics, making comparison between identified cluster solutions tentative at best. We aimed to address these limitations through our study, now in press at Archives in Clinical Neuropsychology. CLICK ME TO DOWNLOAD A PREPRINT.
Taking these challenges into account, as well as the unique need to focus on military populations who face higher head injury rates than others, our recent paper used latent profile analysis (LPA) to explore potential groups of mTBI diagnosed respondents on the Personality Assessment Inventory (PAI). Although prior and related mTBI work (see above) had identified a variety of cluster solutions (2-4 classes), we hypothesized that we would not find meaningful classes. Rather, class extraction would represent a continuous underlying pattern of symptom severity – not functionally distinct groups. We grounded this finding in some prior work we conducted on PTSD groups in the PAI (see Ingram et al., 2022; Click for PDF) as well as the first- factor problem of the PAI.
I pulled a few sentences that I think sum up everything particularly well from the discussion below to summarize these results.
The findings here have a few distinct implications: (1) prior group identification efforts are not replicated in AD personnel and may, instead, represent broader issues with analysis discussed above rather than meaningful findings, (2) the first factor problem (general elevation of substantive clinical scales due to distress, not specific pathology) may also play a role in the observed patterns and should be addressed to aid in the future of the PAI (see Morey, 1996 for discussion of the first factor problem).
As an aside, Tristan (post-bac RA) was the second author on this paper and absolutely killed it with his work on the paper. Not only did he handle about half of the analyses (everything not doing latent clustering), he also learned about how these cluster method work so that he could write some of the discussion and handled 95% of all edits needed for the revise and resubmit. Really awesome work and hats off to him!