Struggling with ordinal data and `OrdinalProbit`

Questions
8

Thanks @bwengals!

I threw the data into the model I mentioned here and here are the results:

               mean    sd  hdi_3%  hdi_97%  ess_bulk  r_hat
mu             1.05  0.30    0.49     1.60    3737.0    1.0
sigma          2.06  0.34    1.49     2.74    2769.0    1.0
theta_star[0]  1.50  0.00    1.50     1.50    4000.0    NaN
theta_star[1]  2.32  0.17    2.02     2.66    4432.0    1.0
theta_star[2]  3.16  0.26    2.67     3.65    3590.0    1.0
theta_star[3]  4.01  0.32    3.42     4.63    3509.0    1.0
theta_star[4]  4.81  0.34    4.16     5.41    3712.0    1.0
theta_star[5]  5.64  0.29    5.10     6.16    4457.0    1.0
theta_star[6]  6.50  0.00    6.50     6.50    3662.0    1.0

Looking at the Kruschke results, the sigma mean is a little low, but then again the priors are not exactly the same. The model runs cleanly in 20-40 seconds depending on the priors used.

I haven’t looked at the ConstrainedUniform implementation in depth, but there are differences in how my model and the model above is setting the limits to the cutpoints, and that’s going to influence the mean and sd estimates. I can look further into the differences too.

1 Like