I've been asked to compute a sequential logit model with threshold variables. The response v is a categorical variable with three levels. There are two covariates, both are categorical variables. w has two levels and x has three levels. We are also given the frequency of each possible combination is also given. The data looks similar to this:
After manipulating the data into a wide format, it looks similar to this:
For the regression we know that w acts as a shift variable with global weight. The x and the interaction wx are thresholds with categoric-specific weights.
How do I include the information that w is a shift variable with global weight and that x and wx are threshold variables with categoric-specific weights into the regression? Normally, I would simply use the vglm()-function like this:
summary(vglm(cbind(Small, Normal, Big) ~ w*x,
family = sratio(parallel = FALSE ~ cbind(Small, Normal, Big)),
data = data))
Unfortunately, this does not give the result that I want. I would like to have one estimate for
x1:1, x1:2, x2:1 and x2:2 as well as for w1* x1:1, w1* x1:2, w1* x2:1 and w1* x2:2 each.