In this paper, we consider the contextual variant of MNL-Bandit problem. More specifically, a dynamic set optimization problem, where decision-maker offers subset (assortment) products to consumer and observes response in every round. Consumers purchase maximize their utility. We assume that attributes describe products, mean utility product is linear values these attributes. model choice behav...

We develop and evaluate a test of choice set misspecification for a multinomial logit choice model. This test determines whether the choice set designated by the researcher mistakenly assigns relevant substitutes to the numeraire good. We develop this test by generalizing the traditional McFadden-type conditional logit model to evaluate whether the traditional model is conditioned on an overly ...

There is a known connection between the multinomial and the Poisson likelihoods. This, in turn, means that a Poisson regression may be transformed into a logit model and vice versa. In this paper, I show the data transformations required to implement this transformation. Several examples are used as illustrations.