:-(
But could I still compare predictors A, B, and C, across data sets, as long as they are all standardized within their respective data set? Or would the different amount of cutpoints estimated in the models screw this up?
I think unfortunately the latter… The predictors are, in principle, trying to predict the outcome, and the outcome has changed…
You can combine the datasets and specify in the model formula that one level of D (or whatever indicates dataset) has 5 thresholds and the other one 4: Set up a model formula for use in brms — brmsformula • brms (find the section on thres() aterm.)
Aah, yes, @Matti is completely correct. I’ve never used that myself!
Thanks for the hint, that sounds very useful. I will look into that.
I actually just realized that people had the option to answer from 1 to 6 in one data set, but no one answered 6. Does this save me in some way? Both models indeed have the same number of cutpoints now. Or do I need to indicate this in the formula (number of thresholds, as you described) when fitting the model?
I’d say that the fact that no one chose to respond 6 is important information and you shouldn’t discard it by pretending it was a 1-5 scale. There might be some issues estimating the model because one of the response cells is empty, depending on what exactly the model is, but you can cross that bridge when you get there.