Can factor analysis carve human nature at the joints?

Sophisticated statistical analysis of PTSD and the false belief that human trauma reactions can be finitely understood.

CATEGORY: CONTROL OF LANGUAGE AND IDEAS

Joan M. Cook, Ph.D., Yale University

Source: International Psychogeriatrics. The paper is free open access.
Read time: 2.3 minutes
 
This Happened
In January 2025, a study was published claiming that an eight-factor model of posttraumatic stress disorder (PTSD) symptoms was the best statistical fit. This is representative of the problem inherent in factor analysis studies that have promoted a false view of human nature that the post-trauma responses can be distilled into a single structure.
 
Who Did This?
The first author was psychologist Joan M. Cook, Ph.D., Professor in the Yale School of Medicine. She has over 150 peer-reviewed publications in the areas of traumatic stress, geriatric mental health and implementation science. She frequently writes op-eds in popular media outlets bashing Trump, pathologizing Trump supporters, and advocating for gun control, the myth that trauma damages brains, climate catastrophe, the fallacy of ACEs, Black Lives Matter, and the view that America is systemically racist.
 
The Claim
Believing that the assessment of PTSD in older adults was under-studied (it’s not), the researchers’ aim was to examine how older adults’ reported their PTSD symptoms. They recruited military veterans 60 years or older, of whom 3,001 self-reported their PTSD symptoms and other problems. One of the study’s main tests was to conduct a confirmatory factor analysis to compare models restricted to one, four, six, seven, and eight factors.  They concluded that the eight-factor model provided the best statistical fit.
The authors concluded that their findings have paved the road for future research that can lead to better diagnosing of PTSD and more targeted treatments despite providing no evidence themselves of real-life benefits.
 
Analysis
This study is one of over six hundred studies over the past four decades using some version of the obtuse factor analysis statistical technique with PTSD symptoms. More background on this useless exercise can be found here and here.
The many problems with the factor analysis technique include that it possesses an unearned aura of importance. Despite massively inconsistent and un-replicated results in the past (see here), the body of factor analysis literature was used in the 2013 DSM-5 revision to change the diagnostic algorithm needlessly from three to four clusters of symptoms.
The premise of factor analysis is the theory that there exists a latent model of PTSD that cannot be observed by the efforts of the human mind. Being a latent, i.e., unobservable model, it is inherently unprovable and whether the latent model exists is a theoretical question not a verifiable proposition.
Of most immediate concern for the useless Cook et al. study is the undisputed fact that whenever models are compared to each other, the model with the most factors obtains the best statistical fit nearly 90% of the time due simply to how mathematical variance is accounted for. A ridiculous eight-factor model is the most any researcher has been willing to conjure so far, but inevitably when someone creates a nine-factor model, that will be best-fitting.
 
Why Did This Happen?
It can be difficult to see in a brief blog post, but this type of psychiatric research is one small part of the liberal progressive false worldview that human nature can be finitely understood by abstract reasoning and controlled by revolutions of cherry-picked “science.” While PTSD is a well-validated discrete syndrome, it is also true that the manifestations of PTSD symptoms are enormously heterogenous between individuals. It is ludicrous to believe that one factor structure can or should be superior to others and would possess unique explanatory powers.

This sort of publication is the real travesty of the peer review system, not the so-called replicability crisis or researchers faking data. (I nevertheless strongly support the current peer review system because all alternatives are worse.) Studies like this get published because peer reviewers have a vested interest in keeping the factor analysis grift going for themselves, or they are ideological academic researchers who have a disinterest in studies that benefit real humans in clinical practice.