Thanks for the reply. Those observations came from around 50 participants. However, as I mentioned in the previous posts, the StudentT and Lognormal models do converge. My question is not about convergence issues. I just pointed out that the model does not converge when using an ExGaussian likelihood, at least as I implemented it in the previous post (maybe wrongly?). However, I decided to give the ExGaussian another try. This time with a smaller nu parameter (arbitrarily reduced to 2) and with standardised RTs. This model does converge well, but the PPCs are pretty bad:
I’ve got very similar PPCs when using a Normal distribution for the likelihood. In this sense, the models using StudentT distributions with standardised data have performed much better. And, as these models already converge with very good ESS, R_hats and BFMIs, I’m not sure how trying reduced versions of my data just to force a fit for an ExGaussian would help to solve my question. But maybe I understood your point incorrectly, for instance I don’t quite follow what do you mean by picking the wrong noise function.