Very slow ordinal model with unequal variances and category specific effects

Hi everyone,

I am fitting an ordinal regression that accounts for unequal variances and category specific effects of the variables. Before running the model with the whole dataset, I have fitted the model with only one variable with category specific effects. To my surprising the model is too slow and is taking days to complete.
Given that I only used one variable, the whole dataset consists of only 926 observations, and the model ran fast when I did not consider category specific effects, I am wondering (1) if there is something wrong in the model specification,(2) what to do to increase the speed. I have attached the code and the data. Thanks for your help in advance.

note: X1 is a categorical variable

fit_1<- brm(bf(Y~cs(X1))+lf(disc~0+cs(X1),cmc = FALSE),data =wwd, 
                          family = acat("probit"), 
                          chains=2,inits=0,warmup = 1000,iter = 2000,algorithm = "sampling", seed = 125,
                          control = list(adapt_delta = 0.9,max_treedepth = 20))

fit_1_data.csv (7.6 KB)

It looks your question fell through a bit, sorry.

My first suspicion is that having category-specific effects for the disc parameter seems fishy - I’ve always thought disc is meant to be constant across categories. I would guess it gives the model too much flexibility and the data cannot inform both the category specific effects for main outcome and for disc at the same time (i.e. there is a non-identifiability/degeneracy). What is the reason to change disc between categoreis?

I guess you were having fitting issues which led you to increase adapt_delta and max_treedepth and I would guess that if you resolve the degeneracy, you won’t need those high settings and your model will fit much faster.