Predicting the distribution of households and employment: a seemingly unrelated regression model with two spatial processes

Household and employment counts (by type) are key inputs to models of travel demand and air quality. For a variety of reasons, spatial dependence is very likely present in and across these counts. In order to identify the nature of these unobserved relationships, this study provides the first application of a feasible generalized spatial 3SLS estimation procedure for a seemingly unrelated regression (SUR) model with two spatial processes. Statistical tests reveal that this more generalized model is superior to its constrained versions (e.g., SUR models without spatial components or with just a spatial lag or spatial error process).In the resulting model of Austin, Texas data, local land-use conditions offer substantial predictive power of household and job densities, and transportation access plays a role, as anticipated. The work demonstrates that SUR estimation of land-use intensities from parcel-level data with two types of spatial dependence is feasible and meaningful. Coupled with an upstream model of land-use type, this work offers key inputs for travel demand analyses, with transportation system performance feedback.