Residential location choice is a key component of the models for predicting land-use and transport demand in urban planning. In general, it requires to consider correlation between spatial alternatives. The approach of nested alternatives of the nested logit model has proved highly efficient in this context. This approach incorporates into the nested logit model both spatial and non-spatial correlations due to unobserved variables. The approach of metric extensions to the spatially correlated logit model specifies models for capturing spatial correlations between alternatives without having to design a nested structure. A model combining both approaches is proposed in this research. The spatially correlated nested logit model proposed herein models the correlation between alternatives of the nests of a nested logit model using a metric of spatial correlation between pairs of alternatives. The proposed model improves the properties of the nested logit model without the need of increasing the number of unknown parameters. Our model also improves the properties of a spatially correlated model with the same spatial metric. When needing to incorporate preference heterogeneity into the model, the proposed model is compatible with a mixed specification with random coefficients. The spatially correlated nested logit model was empirically applied to the real case of residential location choice in the city of Santander in Spain. In this empirical context, this model improved the explanatory and predictive power of the models that it combines.
Read full abstract