Choice Models

Abstract Choice models are used to explain and predict how consumers choose among multiple alternatives, usually brands within a product category (such as ketchup and laundry detergent). The most popular one is the multinomial logit (MNL) model, which has been used in numerous marketing research applications. The article starts with a discussion of the MNL, explicitly discussing least squares and maximum likelihood techniques that are used to calibrate the model, as well as hypothesis tests and measures of model fit. The IIA problem associated with the MNL model is discussed. This leads to a discussion of the multinomial probit (MNP) model, the second most popularly used choice model in Marketing, which alleviates some of the concerns arising out of the IIA problem. Estimating the MNP model, which does not have an analytical closed form (as does the MNL), requires simulation methods. Two popularly used simulation techniques, the frequency‐based simulator and the smooth recursive conditioning (SRC) simulator, are discussed in detail. Lastly, the article briefly discusses some alternative choice models (generalized extreme value (GEV), nested logit, Batsell and Polking) that also relax the IIA restriction of the MNL model, while also yielding closed‐form choice probabilities.