On the DSM (Clinical Psychology's 'Diagnostic Statistical Manual')

In the talk shown above about the nature of clinical psychology, Jordan Peterson spends most of his time persuasively making the case that values are an unavoidably central part of clinical psychology. That is very useful.

But I want to focus on the beginning of the talk, about how psychological diagnostic categories are “family resemblance” categories, in Wittgenstein’s terms. Jordan gives the example of a category with a list of 10 symptoms, any 5 of which suffice to declare someone to have that disorder. That means two people could share no symptoms and be diagnosed as having the same disorder.

Let me try top clear up the issues conceptually. What is fundamentally at issue is the efficacy (good effects minus bad side effects) of various types of treatment. Although enough data may not be available to do this for real, conceptually this is a matter of regressing the efficacy of a particular treatment (let’s say by a randomized trial) on a list of symptoms and other indicators. Labeling something as a particular syndrome could mean at least two different things (or could simply be incoherent). It could mean that a set of symptoms strongly co-move so that there seems to be a strong factor in the factor-analytic sense. In that case, taking an average over many symptoms (which amounts to counting if one is only assessing symptoms at the 0 or 1 level of discrimination) makes sense. The other thing labeling something as a syndrome could mean is that there are significant interaction terms in the regression, so that two or more symptoms co-occurring is more predictive than just the sum of the symptoms would predict.

The bottom-line is that if one can specify what they are for (in this case, guiding treatment), family resemblance categories can be thought of in terms of a regression with the individual characteristics as regressors.