Hi Jake,

Indeed, we are two Jake's studying psychology in NC! 

I am very interested in working on research that explores dimensional models vs. categorical ones. I recently became very interested in the HiToP framework and wanted to add something substantive to it, if at all possible. 

I was thinking of using an archival sample, so I am limited in regards to how much control over what I am working with, but I do know that success in therapy is available (in ordinal format), and there are many different DSM diagnoses in the sample as it is a community outpatient mental health center. 

In terms of measurement, I am using the MMPI-2-RF, as it was identified in the main HiToP paper as one of several dimensional measures of pathology to use.

What I was thinking of doing is going for higher-order domains vs. specific mental disorders, for example, the Thought Dysfunction domain as a predictor of treatment outcomes vs. schizophrenia, schizophreniform, schizotypal etc. Internalizing Dysfunction vs. MDD, GAD, PTSD etc.

What do you think? I am unable to use symptom severity and am left solely with the diagnoses themselves to use as predictors. I could use variables from the MSE and group them together based off of DSM criteria, but I am slightly uncomfortable with that approach. 

The main problem is definition.

DSM is only an expert's consensus of low strenth.

A good definition must be stablealong time  and precise specificity and sensitivity i.e. what is depression and what is not depression.

Nowadays, no scale fits this problematic and, especially, successive DSM.

So, what I am hearing is maybe to scratch the DSM constructs as variables and instead predict treatment outcomes using just transdiagnostic constructs?

For example, use each domain (somatic, internalizing, thought dysfunction, externalizing, etc.) to see which one has the "best" vs "worst" treatment outcomes?

Jacob - from a statistical side, be careful using predictors that overlap or you will run into problems with multicollinearity in which your predictors share variance and look less predicting than they would be on their own. That is, I would recommend that you do not include dimensional variables of symptomatology as well as the categorical diagnosis, for example, to predict your outcome because they would be so highly overlapping.

I hope that helps!

Whichever method you ultimately choose will have its imperfections, so a step forward might be to acknowledge this dilemma and justify the method you've chosen and how it suits your research purposes but not others'.