We have a multinomial model with very different size classes, and looking into it we discovered that the variances of the effect variables do not depend on the number of samples from each class at all. In trying to understand what is going on, I boiled it down to this very simple toy example: imp…

Recovering variance from multinomial softmax models

Velochy January 25, 2025, 5:24pm 9

Nope, either set to true throws the variances for the groups to be a lot more similar than they should be.

Topic		Replies	Views
Different confidence estimates for models with(out) nuance parameters Questions	4	669	August 11, 2017
Softmax Regression with shared coefficients between classes Questions	5	1519	September 6, 2019
Issues with specification of coefficients in Hierarchichal Softmax Regression Questions	6	848	November 15, 2019
Multinomial hierarchical regression with multiple observations per group ("Bad energy issue") Questions	22	3450	March 14, 2019
MAP not working with shape>1 and VI on a pooled model Questions	11	837	November 25, 2017