This paper presents a study on forecasting new product penetration with flexible substitution patterns using choice models, including mixed logit models that do not exhibit the restrictive 'independence from irrelevant alternatives' (iia) property. The models are estimated on data from a stated-preference survey that elicited customers' preferences among gas, electric, methanol, and CNG vehicles. The study compares different models, including mixed logit, pure probit, and mixed logit with additional error components, to capture various substitution patterns. The mixed logit model allows for more flexible substitution patterns by incorporating error components that can induce correlation and heteroskedasticity in the unobserved portion of utility. The study finds that mixed logit models provide more realistic substitution patterns compared to standard logit models, which exhibit the iia property. For example, the mixed logit model predicts that an electric car would draw more proportionally from smaller gas cars than from larger ones, which is more realistic than the iia prediction. The study also compares mixed logit, pure probit, and mixed logit with additional error components. The mixed logit model with additional error components shows that the predicted share of households switching from large cars in response to a price increase is less than in other models, due to variation in the price coefficient over households. The results show that mixed logit models can represent various substitution patterns, making them more flexible than standard logit models. The study concludes that mixed logit models are more suitable for forecasting new product penetration with flexible substitution patterns, as they can capture the complex relationships between alternatives and their attributes. The models are estimated using data from a stated-preference survey, and the results show that the mixed logit model provides more accurate predictions of substitution patterns compared to standard logit models. The study also highlights the importance of considering the distribution of error components in the model specification to capture realistic substitution patterns.This paper presents a study on forecasting new product penetration with flexible substitution patterns using choice models, including mixed logit models that do not exhibit the restrictive 'independence from irrelevant alternatives' (iia) property. The models are estimated on data from a stated-preference survey that elicited customers' preferences among gas, electric, methanol, and CNG vehicles. The study compares different models, including mixed logit, pure probit, and mixed logit with additional error components, to capture various substitution patterns. The mixed logit model allows for more flexible substitution patterns by incorporating error components that can induce correlation and heteroskedasticity in the unobserved portion of utility. The study finds that mixed logit models provide more realistic substitution patterns compared to standard logit models, which exhibit the iia property. For example, the mixed logit model predicts that an electric car would draw more proportionally from smaller gas cars than from larger ones, which is more realistic than the iia prediction. The study also compares mixed logit, pure probit, and mixed logit with additional error components. The mixed logit model with additional error components shows that the predicted share of households switching from large cars in response to a price increase is less than in other models, due to variation in the price coefficient over households. The results show that mixed logit models can represent various substitution patterns, making them more flexible than standard logit models. The study concludes that mixed logit models are more suitable for forecasting new product penetration with flexible substitution patterns, as they can capture the complex relationships between alternatives and their attributes. The models are estimated using data from a stated-preference survey, and the results show that the mixed logit model provides more accurate predictions of substitution patterns compared to standard logit models. The study also highlights the importance of considering the distribution of error components in the model specification to capture realistic substitution patterns.