I'm using a discrete choice experiment (DCE), and I've estimated the answers with a nested logit model (using mlogit package of r software), which gives quite good results.
However, I've coded attribute levels using numeric coding for the estimation
For example, if an attribute has three levels, the levels were coded as:
level 1 = 0
Level 2 = 1
Level 3 = 2
This coding seems tricky when it comes to predictions. For exemple, if I want to predict the impact of 5 unit increase in one attribute, the level coding will stays the same:
Level 1 + 5 = 0 (since it stills the lower level so it takes 0)
Level 2 + 5 = 1
Level 3 + 5 = 2
Using (predict) function, the predicted probabilities remain the same regardless the level of the attribute change, since once coded, the levels are unchanged
So How to predict probabilities with new levels when those are coded ?