I am currently trying to model a discrete choice problem using multinomial logit. The issue is that each individual faces a "different" choice set.

In Stage 1 an individual is told that he can select from a set of alternatives (the number of alternatives and the alternatives themselves may vary between individuals).

In Stage 2 the individual can select any number of the alternatives offered i.e. not mutually exclusive. So individual $i$ is allowed to select from a set of alternatives $\mathbb{A}_i$ that has $J_i$ elements and so can choose from $2^{J_i}$ alternatives.

I am modeling stage 2.

I was thinking of specifying latent utility as follows: $$U_{ij} = \alpha + \beta X_i + \gamma Z_j + \epsilon_{ij}$$ where $\epsilon_{ij}$ is T1EV, iid and $j$ denotes one of the $2^{J_i}$ alternatives.

Then I would specify the choice probabilities by each individual and perform an MLE to recover the coefficients. I am thinking that I should be able to estimate coefficients since they are not varying by individuals and choice.

Will such a procedure work? I cannot include choice specific constants because individuals have different choice sets.

Is there any paper or book that discusses this issue? I have not been able to find any.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.