I think you're getting things the wrong way round. Just to clarify, you mean those 30 questions about how a person feels about sport should predict how many times they did sport at 3 points in time?
If so, then those 30 questions should act as dependent variables (or predictor variables to name them correctly) with the amount they did sport being the independent variable (or outcome variable).
As far as I know, MR only allows for one outcome variable, so you may have to just make three seperate models. I'm not sure how your type I error would look though. I assume there are corrections you can make to compensate for this.