Assessing the extent of modifiable areal unit problem in modelling freight (trip) generation: Relationship between zone design and model estimation results

There is a growing interest in incorporating spatial indicators into freight demand model systems. The indicators are measured for different areal units (e.g., census tracts or block groups) and are often used as proxy variables or aggregation layers. Model estimation results vary according to the choice of these areal units and an analyst is thus confronted with a popular decision challenge termed as ‘modifiable areal unit problem’ (MAUP). The variability in results due to MAUP arises since areal units can be modified in theoretically infinite ways (in terms of shape, orientation and size) and magnitude of aggregation loss in information will vary for each alternative zoning system. In effect, how well the zonal (aggregated) characteristics can describe the establishment-level (disaggregated) observations is inversely related to MAUP effects. Little is known, however, about the extent of MAUP effects in freight generation (FG) models and freight trip generation (FTG) models. This study diagnoses the implications of MAUP effects in FG and FTG models by designing alternate zoning systems (by means of different zonal variables and clustering techniques) and assessing the sensitivity of model estimation results within a framework of comparative analysis (by means of hierarchical linear models). Study results assess the presence of MAUP as alternate zoning systems resulted in wide variation in the estimated coefficients for zonal characteristics (e.g., industrial area, land value, number of establishments, distance to primary arterial) in terms of magnitude, statistical significance, and even in the direction of association (sign of the coefficient). The implication of this finding is that an analyst may design different or even counterproductive policy instruments based on the way data is aggregated to capture the role of land-use, spatial effects and built-environment in influencing freight travel patterns. MAUP effects are also found to be dependent on the metric in which freight is measured (i.e., FG or FTG) and direction in which flow is measured (i.e., production or attraction). Overall, this research improves the understanding of the parameter sensitivity and performance sensitivity of freight demand model systems to alternative spatial representations of an establishment's relative location. The research findings strongly encourage analysts to acknowledge that the results of freight travel analyses with spatial indicators are sensitive to the definition of areal units.