Effect versus Reference coding

I understand the difference between these, but I had a question. It appears to me that for the parameters themself SAS compares each dummy (aka design) variable to the mean of the means of its parent categorical variable (the variable it was created from) for test of signficance.

But for the Odds Ratio it uses reference coding, comparing the dummies to the level that is not directly measured (that is the reference level) in determining statistical signficance.

Which seems strange.