Comparison of Distance Estimates for Commodity Flow Survey: Great Circle Distances Versus Network-Based Distances

A study was conducted to compare distance estimates derived from great circle distances (GCD) with distance estimates derived from a network-based model. The study used a sample of shipments from the 1993 commodity flow survey (CFS). For each shipment in the sample, the distance from the zip code of origin to the zip code of destination was calculated by using the Oak Ridge National Laboratories National Highway Network and assuming that the minimum impedance path was utilized. For each of these origin–destination pairs, the GCD and several variations of the GCD also were estimated. Finally, the network-based estimates and the GCD-based estimates were statistically compared. As expected, distance estimates based on GCD were found to be different from network-based estimates. However, applying a constant circuity factor of 1.22 or using variable circuity factors based on distance category did not result in a statistical bias in these distance estimates. Examination of distance estimates at the level of origin–destination pair revealed that distance estimates could vary as much as 75%. A comparison of published values for the 1997 CFS with values derived from GCD-based distance estimates shows that approximately 5% to 35% of the GCD-based values for the 1997 CFS would fall outside a two standard error interval. Although GCD-based estimates, under some conditions, may produce unbiased estimates of the mean distance, this does not eliminate the need for network-based estimates.