I will be using both self report and clinician observation (the MMPI-2-RF for self report and the PANSS, which is a structured interview with a rating scale, for the observer report)
With these two measures, what statistical test would be best to test out a dimensional model of psychosis? The best model I can tell from current research is a five factor model (positive, negative, manic, depressive, and disorganized factors).
One issue I am coming across is that there is research out there (a lot from Jim Van Os) suggesting that underlying this dimension of psychosis is a categorical structure. This leaves me scratching my head at what tests I can run using my measures to see if (a) this dimensional model maps out, and (b) how to tell if there is an underlying categorical structure.
The lit review I have done is essentially suggesting that along the psychosis continuum, there are individuals who are either at risk for functional impairment (at risk for diagnosable psychotic disorders) or are not functionally impaired (yet still experience "symptoms" without the distress and loss of functioning).
Thank you.
Edit: Also, a former mentor had suggested taxometrics (L-Mode, MAMBAC, etc.) How would this work? I cannot find anything aside from some articles by Paul Meehl on the subject.
Dear Jacob,
if you really want to focus on the question whether psychosis is dimensional or a taxon (i.e. categorical), I would also suggest a taxometric analysis.
Here are 2 papers from the domain of psychopathy. They also refer to Meehl et al., but maybe you find some ideas for implementation.
https://www.ncbi.nlm.nih.gov/pubmed/16492104
https://www.ncbi.nlm.nih.gov/pubmed/18020717
However, I am not sure how one would test this or also the dimensional structure itself (e.g., through confirmatory FA), when the used measures are also operationalizations of several other latent constructs/disorders. In an (exploratory) factor analysis you would probably get several factors that are not related to psychosis, but other disorders. So maybe one needs to (first) see what measures or latent variables predict psychosis (e.g., https://www.ncbi.nlm.nih.gov/pubmed/21756212)
Best,
Stefan
Instead of CFA you could perhaps better apply Exploratory Structural Equation Modeling (Asparouhov and Muthén, 2009), because your study is still in a very exploratory stage. This approach was intended to overcome the limitations of CFA. In this method, in addition to a CFA measurement model,an EFA measurement model with rotations can be used within a structural equation model. There is no longer a need to fix the non-indicators per factor on 0. They can be freely estimated, and they are reproduced in the output. So there are secondary factor loadings.
The superiority of ESEM to CFA in cases of more complex measurement models has already been demonstrated in studies like Marsh, Ludtke, Muthèn, Asparouhov,Morin, Trautwein, et al. (2010), Furnham, Guenole, Levine, & Chamorro-Premuzic (2013), Booth and Hughes (2014), and Guay, Morin, Litalien, and Vallerand (2014). Also, ESEM could further be used for a data-driven model-modification with more realistic and informative results than CFA.
For these references, and limitations of CFA, see my publication:
https://www.researchgate.net/publication/280077456_CFA_as_a_tool_in_research_using_questionnaires_a_critique
Good luck!
Article CFA as a tool in research using questionnaires: a critique
Another idea, to help decide whether the psychotic disorders are essentially categorical, is to apply cluster analysis on the subjects, for instance by means of CLUSTAN. In that case, your use your variables to form subgroup of subjects within your sample. If psychotic disorders are categorical indeed, then there should emerge one or more subgroups with no or only mild symptoms, that differ qualitatively from the psychotic subgroups, and insofar they have features in common with the psychotic subgroups, there should be a clear quantitive difference on these features.
Hi everyone, sorry for the late reply! Thanks for the thoughtful and engaging responses. I will definitely have some follow-up questions when I am a little less busy.
Hi.
Although I am far less well informed on statistics than the rest of the discussants, what drew my attention is the lack of research question. What I read is, "what statistical test would be best to test out a dimensional model of psychosis". Does this mean that you already have a dimensional model of psychosis in mind? Was this reached in the literature (e.g., in different populations)? What measures and what analyses yielded this model? Or does this mean that you want to test if psychosis is a uni-dimensional or a multi-dimensional construct?
As a clinician, I would love to know more about what "dimensional model" means in practice. Also, this brings to mind that a "diagnosis" is a representation of different dimensions - symptoms. So, I wonder if your question has a cyclical character: A diagnosis is multidimensional as a construct, I am administering multidimensional measures, I want to check if the construct is multidimensional. I am not sure why such a question would make sense.
Hope this helps.
Geogios, yes there is a specific model in mind, so the first question would be "does my sample fit this dimensional model?"
And yes this has been found all over the world in different populations using many different research frameworks.
In clinical practice what this means is partly what has already been done for many years; we treat people, not disorders. Research for years has used DSM diagnoses as variable and this is problematic for many reasons, for example, PTSD as a diagnosis has over 636,000 combinations of symptoms. This means you could have 636,000+ people "with" PTSD and all different presentations.
Using dimensional models with categorical ones seems to have yielded the best results in terms of criterion utility, treatment outcomes, and disorder etiology, essentially meaning this:
Disorders are categorical in clinical settings and dimensional in nature. The problem for many years is we have been studying nature through a clinical, artificial, lens.
Does this make sense Georgios? If not let me know. Great questions.
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India