Forecasting jobs location choices by Discrete Choice Models: A sensitivity analysis to scale and implications for LUTI models

This paper proposes an empirical analysis of the sensitivity of Discrete Choice Model (DCM) to the size of the spatial units used as choice set (which relates to the well-known Modifiable Areal Unit Problem). Job's location choices in Brussels (Belgium) are used as the case study. DCMs are implemented within different Land Use and Transport Interactions (LUTI) models (UrbanSim, ILUTE) to forecast jobs or household location choices. Nevertheless, no studies have assessed their sensitivity to the size of the Basic Spatial Units (BSU) in an urban context. The results show significant differences in parameter estimates between BSUs. Assuming that new jobs are distributed among the study area proportionally to the utility level predicted by the DCM for each BSU (as in a LUTI model), it is also demonstrated that the spatial distribution of these new jobs varies with the size of the BSUs. These findings mean that the scale of the BSU used in the model can influence the output of a LUTI model relying on DCM to forecast location choices of agents and, therefore, have important operational implications for land-use planning.