ABSTRACT
The economic question of interest lends itself well to estimation with the conditional multinomial logit (MNL) model, developed by McFadden (1973), since the choice of a prescription drug in a class with multiple alternatives is a discrete one. Utility derived by the ith patient in the jth plan from choosing the kth drug, Uijk, is latent and only the choice is observed. If Yijk represents the choice of the kth alternative we have
The MNL estimates the impact of exogenous variables on the probabilities19 of using different drug alternatives. If X represents the set of measured attributes of agents with an individual drawn at random from the sample population having some attribute vector xij with m elements, then the conditional probability that the ith consumer in the jth plan chooses the kth drug can be written as
where Y represents the universe of drugs from which the consumer may choose a particular alternative.
