Combining Model Scores

Question for the regression modeling experts:

I am looking for an efficient and correct way to combine model scores taking into account individual model performance. I have a dataset that has been scored by four (logit) models. So each observation has four different probability scores. I'd like to create a sort of combined score for each of the models, taking into account empirical model performance (True Positives/False Positives identified, etc). Is there any information about how to correctly create a combined score, taking individual model performance into consideration?

Any help would be appreciated!