Advantages of latent class models over continuous mixture models in capturing heterogeneity

摘要: Latent class model structures present a flexible extension of standard choice modelling approaches in the context representation taste heterogeneity. In this study evidence is provided flexibility latent approach and potential advantages over other structures. As first contribution, formulae are derived for correlation between individual coefficients show how function socio-demographic attributes used class-allocation model. From this, an analyst can example conclude that specific subgroups population, time cost negatively correlated, while others, may be positive. This crucial advantage Mixed Logit model, which only produces fixed measure two randomly distributed coefficients. next step, it shown same principle applies to elasticities models, where these again expressed as applied part study, use made stated data departure travel mode collected Dutch National The results showed obtains significant gains basic MNL log-likelihood comparable those obtained by with fully specified covariance structure random However, real come at interpretation stage. Here, analysis illustrates relationship indicators information not useful heterogeneity but also provide forecasting. Additionally, differences models. example, gives 0.41 car train, ranges from 0.43 0.9 depending on characteristics. For covering abstract see ITRD E145999