I am trying to predict football outcome using an ordered logit model and I am using R. I used polr in R and everything works fine until the prediction part. the dependent variable is the result (home win, draw, away win) and the independent variables are mostly binary besides the ELO rating (strength index for both teams) difference, where it consist 400 different values. While using the basepredict method, I found out that the prediction contain no "draw" as a result. After changing the data I believe the problem is that the "draw" result in the training data is not enough percentage-wise because after increase the amount of draws manually it indeed start showing draw as a result. My question is if there is a way to force the program to recognise three categories regardless of the amount of data entry. Any tips toward this final solution are all appreciated.
Thanks in advance!