Convergence issues for ordinal probit when including subject dependent effect for a categorical predictor

[Solomon]

Hmm, it looks like you’ve already covered some of my go-to suggestions. The best I can offer is I’ve been working on a series of models like this with a collaborator. We wanted to compare them by LOO, and we ran into high Pareto k values, too. In our case, it seemed like we overparameterized the disc model, and when we removed some of the parameters, the k values dropped by quite a bit. In our case, the dropped disc parameters were kinda cool from a substantive standpoint, but we didn’t really need them; they were more like flashy nuisance parameters. I don’t know if you’re in the same position, but it’s something you might consider.