If you have a discrete valued predictor in reg, you can recode it to a binary or series of binary predictors (coded as numeric 0-1) and include them in the model. Reg will not 'lump' them together for the inclusion or exclusion in the various stepwise models; if you want to do that, you will need to manually implement the steps in GLM.
ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.