Examining the Effects of Local Sample Sizes on Spatial Transferability of Freight Production Models

Abstract

Recent research in freight transportation planning has been exploring the spatial transferability of freight demand models, which can help the planning agencies of developing economies save the cost and time incurred in freight surveys. As the demand models are time-, cost-, and data-intensive, it is prudent to analyze the effects of sample size on the transferred model in a region. The findings and inferences from such analysis will save resources in freight data collection programs. Earlier, conventional models like ordinary least squares (OLS) regression were assessed for transferability. However, the predictive ability and transferability of such non-conventional models are not well studied. It is necessary to understand whether the extent of transferability of non-conventional models is greater than that of conventional models so that planning agencies can adopt more reliable modeling approaches. This paper investigates the spatial transferability of freight production models using OLS, robust regression, and multiple classification analysis (MCA). The results of the transferability assessment show that MCA models have better transferability using the naïve transfer method. In addition, transferability is assessed for different sample sizes to examine the variation in the extent of transferability. The MCA models have shown the least deviation, indicating that these models are preferred for transferability when the sample size is small.