I want to estimate a regression model for skewed cost data with a Gamma distribution and Log Link using PROC Genmod.
For the training sample, I receive the predicted values as well as the coefficient estimates and goodnes of fit measures by specifying the appropriate options in Proc Genmod.
However, if I want to test the estimated model with a test data set, I get stucked.
I know there is the possibility to append my test data set to the training data set and set the dependent variable of the test data to missing in order to obtain the predicted values.
However, there are 2 problems with this solution: First, I don't know how to receive the Goodness of Fit meaures for the test data. Second, and even more crucial - In general, the test data is not available yet at the time I am estimating the model. Since the involved data base is very large and the estimation therefore takes really long, the goal is not reestimate the model each time I get new data, but to estimate it once and then apply the resulting coefficients on the new data set.