Modeling commercial establishments’ freight trip generation: a two-step approach to predict weekly deliveries in total of vehicles

Assuming freight trip generation as the weekly total of vehicles arriving for loading/unloading purposes, at commercial establishments, the authors experiment and compare alternative modeling methodologies. The motivation is to achieve better freight trip generation models, a contribution to increased chances of predicting correctly, for example, the necessary parking infrastructure to accommodate demand, or to calculate the freight traffic impacts at a micro level. The main source of data is an Establishment-based Freight Survey, used to collect data about 604 retail establishments in the city of Lisbon, Portugal. The selected independent variables were establishments’ industry category, number of employees and sales/consumer area. The first methodology uses Poisson Log-linear Generalized Linear Models (GLM) to predict weekly deliveries. The second consists on a novel methodology, starting by the prediction of delivery ranges, followed by prediction of the precise number of deliveries inside each range. In this step other Generalized Linear Models, Ordinal Logit and Multinomial Logistic models are used. Both the prediction of delivery ranges, albeit with lower resolution, and the follow up models delivered comparable or superior predictions to those of GLMs. The analysis allowed concluding that there is considerably variability in the quality of predictions depending on the selected model, which requires a careful selection by researchers or practitioners.