This question is related to the question here: Speeding up a hierarchical multinomial logit model

I am using the Stan code from that post to run a hierarchical multinomial logistic regression to analyze “MaxDiff” data (also known as “object case best-worst scaling,” see reference here).

The post above says it was taking 340 seconds to run 100 iterations, but for my data (350 respondents, 13 items displayed across 13 blocks with 4 alternatives per block) it is taking multiple hours to run the default 2000 iterations, and then I get lots of problems that suggest to change adapt delta, max tree depth, and up the number of iterations.

Something tells me that I am likely doing something wrong with either the model (but it is essentially entirely copy-pasted over from the post above) or preparing the data in R and feeding it to `rstan`

.

I have attached the data as a .csv below, along with the .stan file and the .R code that prepares the data and feeds it to Stan via `rstan`

.

Could someone here help me figure out how to make this model run in a number of minutes, instead of a number of hours? It doesn’t have to be blazing fast, it just has to be usable. (And the slowness and various warning messages makes me think something is specified incorrectly.)

Thank you!

mnl_q.csv (302.5 KB) mnl_q.R (1.9 KB) mnl_q.stan (1.9 KB)