Abstract
Freight demand models typically employ a priori classification systems for dividing establishments into hypothetically complementary groups with homogeneous patterns in freight production (FP) and freight trip production (FTP). Although an attractive and popular notion, the assumption of homogeneity within these a priori industrial classes is reductive in nature and is not yet tested in literature. This research examines this hypothesis and explores the possibility of a data-driven segmentation by examining the relationships between FP/FTP patterns and prevalent a priori classes; subsequently, it creates homogeneous ensembles of a posteriori segments through aggregation. This research labels, explains, and interprets these novel segments using commodity value density of industrial classes. The alternate segmentation schemes are compared in their ability to predict FP and FTP and it is found that: (i) industrial classification systems (NAICS, ISIC) perform significantly better than product classification systems (ASICC); (ii) a considerable portion of variability in FTP does not depend on employment predictor due to the underlying influence of shipment size; (iii) an a posteriori segmentation scheme considering shipment size may represent an effective middle ground for developing both FP and FTP models in freight demand model systems. Adoption of these novel segments of the freight travel market has the potential to reduce the sample size requirements of freight demand model systems and minimize the financial necessities for future freight surveys.
Get full access to this article
View all access options for this article.