Thread by @EikoFried

Here's a simple thought experiment showing that if psychopathology were categorical, it would still show up as dimensional using current taxometric methods. 🧵

https://twitter.com/awaisaftab/status/1456990574951665668

Taxometric analyses happen at the between subjects (group) level, usually using one measurement point.

You usually get distributions on the left side (often with skew), not distributions on the right side—again, at the population level.

But psychopathology is not a between-subjects process, it's a within-subjects process.

So let's now simulate within-subjects data for 100 people who all have depression, as a categorical, not dimensional process. I'll show you processes for 6 people.

As you can see, people get categorical depression over time, there is no slow transition—they jump into a second, stable state. Crucially, people differ in several parameters. How large is the jump; where is their starting point; where is their endpoint; and so on.

Now you do that for 100 people, & let people differ sufficiently in their bi-stable systems. Guess what you get when you take a cross-section of this data, and apply group-level taxometric analysis to such data? A roughly dimensional distribution of scores.

These ideas aren't new (see e.g. journals.cambridge.org/abstract_S0033…), but seem worth reiterating so I'll publish a small simulation on this in the next few months.

Core message: we need to grapple more with issues of between vs within subjects data, processes, and analyses.

End 🧵