This paper, authored by David Brownstone and Kenneth Train, published in 1999, addresses the limitations of traditional logit and nested logit models in forecasting new product demand, particularly their restrictive "independence from irrelevant alternatives" (iiA) property. The authors propose and apply mixed logits, which do not exhibit iiA and can approximate various substitution patterns. These models are estimated using data from a stated-preference survey on households' preferences among gas, electric, methanol, and CNG vehicles. The paper introduces mixed logits, which combine a non-stochastic linear-in-parameters part with stochastic parts that can be correlated and heteroskedastic. The choice probability is derived from a logit formula integrated over the distribution of a random term, $\eta$, which can have various distributions. The authors also discuss pure probits, where the stochastic part is normally distributed, and compare the simulation methods for mixed logits and probits. The estimated models are applied to data from a survey on households' preferences for alternative-fuel vehicles. The results show that mixed logits and probits can capture more realistic substitution patterns compared to standard logit models, such as disproportionate switching from larger gas cars to smaller gas cars when a new electric car is introduced. The flexibility of mixed logits is demonstrated through different specifications, including error-components structures and random-parameters specifications. The paper concludes by highlighting the advantages of mixed logits in representing complex substitution patterns and their potential for policy analysis, particularly in scenarios where the iiA property leads to unrealistic predictions.This paper, authored by David Brownstone and Kenneth Train, published in 1999, addresses the limitations of traditional logit and nested logit models in forecasting new product demand, particularly their restrictive "independence from irrelevant alternatives" (iiA) property. The authors propose and apply mixed logits, which do not exhibit iiA and can approximate various substitution patterns. These models are estimated using data from a stated-preference survey on households' preferences among gas, electric, methanol, and CNG vehicles. The paper introduces mixed logits, which combine a non-stochastic linear-in-parameters part with stochastic parts that can be correlated and heteroskedastic. The choice probability is derived from a logit formula integrated over the distribution of a random term, $\eta$, which can have various distributions. The authors also discuss pure probits, where the stochastic part is normally distributed, and compare the simulation methods for mixed logits and probits. The estimated models are applied to data from a survey on households' preferences for alternative-fuel vehicles. The results show that mixed logits and probits can capture more realistic substitution patterns compared to standard logit models, such as disproportionate switching from larger gas cars to smaller gas cars when a new electric car is introduced. The flexibility of mixed logits is demonstrated through different specifications, including error-components structures and random-parameters specifications. The paper concludes by highlighting the advantages of mixed logits in representing complex substitution patterns and their potential for policy analysis, particularly in scenarios where the iiA property leads to unrealistic predictions.