I have a dataset that consists of six different segments. I have calculated a Logit Regression Model for each of those segments (binary response variable, 30.000 observations in total, 63 variables before applying the stepwise regression algorithm).
As a next step I calculated the respective ROC (Receiver Operating Characteristic) curves and retrieved the AUC (Area under Curve) values. Now what I currently want to do is to look at the different regression models as if they are one. So when an observations belongs to segment one, its response variable is assumed to be linked to the independent variables as regression model 1 indicates and so on (essentially something like dummy variables I suppose...).
Against this background I thought of aggreggating the individual AUC values by weighing them by the number of observations in the respective segment and adding them up. In my opinion, this would represent the AUC value of the artifical aggregated regression model.
Does this thought process make sense? Also - doing six different regressions and look at them as if they are one model for the entire dataset is basically what I am supposed to do and I do not really have the choice of doing something different. So my question only really pertains the AUC part (Yet, if there is valid criticism regarding the "six different regression models" part, I would of course also be interested in it.)