Two-armed bandit hierarchical reinforcement learning model - interpreting conflicting loo and posterior predictive check results

Modeling
, , , ,
8

Hi @avehtari ,

thanks for your feedback!

Yes, I also exclude them when computing the likelihood in the model, as the trial loop

for (t in 1:Tsubj[s,v])

only runs up until the individual numper of non -999 trials per subject and condition. -999 trials are always at the end of a trial sequence (e.g. trials, 48, 49, 50).

I did this and openend a new topic here. Any insights would be appreciated.

Best,
Milena