I have a dataset with transactions for five types of goods (which I coded as binary variables). I'm trying to test the effect of demographic characteristics on an individual's spending choices.
What I've been doing is using five separate models, where the left hand side is an interaction term. So for a model on spending on good 1, I'd have the independent variable being the interaction between the continuous variable for amount spent and the dummy for good 1(1=good 1, 0=not).
I'm starting to wonder if this approach makes sense at all. Is it appropriate (especially for interpretation)? Should I instead have put the dummy interactions with the predictors?