The spatial transferability of parameters in a gravity model of commuting flows

This paper studies whether gravity model parameters estimated in one geographic area can give reasonable predictions of commuting flows in another. To do this, three sets of parameters are estimated for geographically proximate yet separate regions in south-west Norway. All possible combinations of data and parameters are considered, giving a total of nine cases. Of particular importance is the distinction between statistical equality of parameters and ‘practical’ equality, i.e. are the differences in predictions big enough to matter. A new type test based on the standardised root mean square error (SRMSE) and Monte Carlo simulation is proposed and utilised.

Research highlights
► Classical statistical tests are ill-suited to studying transferability.
► A new testing procedure based on bootstrapping is proposed.
► Transferability is found to be sensible only in certain cases.
► Meta-Analysis is the way to develop comprehensive guidelines on transferability.