Bad Model Performance Only When Intercept is Dropped - brms 2PL IRT

[martinmodrak]

@wgoette

Seems like I do think I need to get the no-intercept model running correctly – just curious to me that it causes such problems for Stan, especially once I start trying other types of models. I wonder whether the fact that some multidimensional models fail to converge reflects the fact that those models are just outright wrong.

There could be a lot of stuff going on - my current best advice is at Divergent transitions - a primer. I would definitely try running the no-intercept model with wider priors (at least by factor of sqrt(2)). It might also be useful to use different coding of the factor variable (e.g. as outlined at Contrast coding with brms).