ApolloChoiceModelling forum

Hello,

I have a basic question about defining utility functions in Apollo. When we want to fit a multinomial logit regression, should we consider one of the alternatives as a base category like this?

V[["A"]]= b_cov*cov_A + b_acc1*acc1_A + b_acc2*acc2_A + b_bid * bid_A
V[["B"]]= b_cov*cov_B + b_acc1*acc1_B + b_acc2*acc2_B + b_bid * bid_B
V[["C"]]= 0

Or should we define them like below? (as in most of the examples in apollo "http://www.apollochoicemodelling.com/examples.html")

V[["A"]]= b_cov*cov_A + b_acc1*acc1_A + b_acc2*acc2_A + b_bid * bid_A
V[["B"]]= b_cov*cov_B + b_acc1*acc1_B + b_acc2*acc2_B + b_bid * bid_B
V[["C"]]= b_cov*cov_C + b_acc1*acc1_C + b_acc2*acc2_C + b_bid * bid_C

Thanks,

Hi

this depends on your data. If attributes are defined for all alternatives, then you need to include them in the utility. Setting one utility to zero is only acceptable if this is a base alternative without attributes or if you subtracted the values for the attributes for that alternative from those for both other alternatives. In both cases, you would still want to include constants too

Stephane