Since there are aspects of utility that the researcher does not or cannot = U = V + ε observe, V . Utility is decomposed as U , where nj nj nj nj nj ε captures the factors that affect utility but are not included in V . This nj nj decomposition is fully general, since ε is defined as simply the differ- nj ence between true utility U and the part of utility that the researcher nj captures in V . Given its definition, the characteristics of ε , such as its nj nj distribution, depend critically on the researcher’s specification of V . nj is not defined for a choice situation per se. Rather, it is In particular, ε nj defined relative to a researcher’s representation of that choice situation. This distinction becomes relevant when evaluating the appropriateness of various specific discrete choice models. ∀ j and therefore treats these terms The researcher does not know ε nj as random. The joint density of the random vector ε = ε ,...,ε n1 nJ n is denoted f (ε ). With this density, the researcher can make probabilis- n tic statements about the decision maker’s choice. The probability that decision maker n chooses alternative i is