I am actually looking for an efficient way to test my conceptual model. I have five product categories and each category consists of 4 brands. I want to test how brand choice is influenced, who has a high probability to buy a certain brand? For example, a person who is highly quality conscious will not buy the budget store brand, but might be more inclined to buy a national brand. However, I decided not to force respondents to choose one brand as in practice some people do not clearly prefer/choose one specific brand (which my data also clearly indicates). People could therefore have high probabilities for more than one brand.
Why am I searching for possibilities in terms of a choice model? For my research degree it is probably insufficient to perform a multivariate GLM (or SUR in stata). My lecturer indicated that it is possible to recode the variable to make MNL possible, but did not tell me exactly how. He only told a first step would be to recode my dependent variable in a binary scale. I was thinking of 4,5 --> 1 (regularly chosen) and 1,2,3 (not regularly chosen) and then deduce which brand is preferred.
Nevertheless, I still do not have clear how MNL is possible then as I still have 4 dependent variables left. Not that difficult i.e. (budget store brand=0, regular store brand=0, national brand=1, premium store brand=0), but more complicated in case of (budget store brand=0, regular store brand=0, national brand=1, premium store brand=1). Could it be possible to add an option like 'no preference'?
Hopefully it is somewhat clearer now.