Improving distance estimates between areal units in migration models

There are many methods of modeling migrant flows within a set of areal units, but it is common in most to incorporate some measure of distance as an explanatory variable. These distances are effectively meant to represent the typical distance between pairs of areas that would be traveled by potential migrants. They are usually calculated between population-weighted centroids derived for each zone. It is argued here that this method of calculating distance is biased and that the zonal system used will influence the Fnal model parameters that are intended to describe the underlying migration process. The distances between nearby zones will be particularly poorly specified using this approach, but other problems arise which relate to the shape of the zones and the position of the zones in relation to each other. This paper describes an alternative method of calculating these distances which reduces this bias. It is shown that the resulting models fit the data far more satisfactorily and that the residuals from models incorporating this approach are significantly different from those identified from models that use the standard method of specifying distance.