Suppose I have a dataset as follows:
Product      Week    Type             Num times ad shown      Purchases
1         2/2/13    young,old,hipster        100                    5
and I want understand what are characteristics of people that drive people to purchase a particular product(2 types). I have one column 'type' that shows the types of people who bought the product in that week. What would be a good modeling approach for this problem? Here is my approach. Transform the data set so that each level of the type variable gets its own row.
 Product      Week       Type      Num times ad shown     Purchases
    1         2/2/13    young       100                    5
    1         2/2/13    old         100                    5
    1         2/2/13   hipster      100                    5
The purchases are rare so most weeks the number of purchases is 0, and the range of values for the purchase variable is 0-200. I could either model the raw number of purchases, the proportion of purchases to the number of ads shown, or transform purchases into a binary 0/1 variable and model the probability of a purchase. Respectively, I could either run a Poisson regression for counts, a CART model for the proportions, or a logistic regression for the binary outcome. I was just wondering if all three of these approaches are viable. Also is making the levels for the 'type' variable into its own row a viable option to capture what type of users buy the product and is type of transformation only good for logistic regression?