I am running 2 models on a yes/no response variable,. One uses the logit link (for odds ratios and stats analysis) and one uses the identity link (for deriving adjusted proportions), but otherwise they are identical. The model with the identity link is not converging, but the other model is.
The main effects included in the model are treatment group (3 levels) and strata (6 levels), no interactions are included. There are 3 (I think) empty cells. Can anyone explain why one model is not converging?