I'm running a mixed effect logistic regression model in RStudio with two random intercepts ad three different predictors as fixed effects. All the indipendent variables are categorical. However, although the model does not show any convergence problem, there is something strange. One of the categorical predictors (Age) has three different layers (7, 8, 20). When the level of this categorical predictor is 20, all the values of the dependent variable are 1 (and theoretically that's okay, this is exactly what I expected). Those participants with Age=20 are a control group, but I'm interesting in the comparison between them and the others. I've compared contrasts, and while there is significant difference between 7 and 8, no difference appear between 7 and 20, 8 and 20 and 7+8 vs 20. This sounds strange, since when Age=7 there are both 0 and 1 outputs (due to the other predictors), as well as when Age=8. So, the difference between 7 and 20 and 8 and 20 or 7+8 vs 20 should be bigger than the one between 7 vs 8. Is there a command that I can add to my model to solve the problem?
I've read something about "separation of the data", but I'm not sure about this (I am very new to the world of statistics). I'm just using the optimizer control = glmerControl(optimizer = "bobyqa").