A new way of determining distance decay parameters in spatial interaction models with application to job accessibility analysis in Sweden
2016 (English)In: European Journal of Transport and Infrastructure Research, ISSN 1567-7133, E-ISSN 1567-7141, Vol. 16, no 2, 344-362 p.Article in journal (Refereed) Published
In this paper we explore and compare various techniques for the calculation of distance decay parameters which are estimated using statistical methods with half-life decay parameters which are derived mathematically. Half-life models appear to be a valid alternative to traditional spatial interaction models, especially in the presence of spatially highly disaggregate data. Our results indicate that Half-life models are more accurate for the construction of decay parameters than are unconstrained spatial interaction models in 'medium' sized datasets but not as accurate as doubly-constrained models. However, using highly detailed and disaggregate datasets Half-life models may be viable alternatives to doubly-constrained spatial interaction models as the latter will be difficult to estimate when the number of origins and destinations increase. In addition, Half-life models rise in accuracy with increasing degrees of disaggregation due to reductions of systematic errors between observed individual level commuting distance and modelled distances between origins and destinations.
In sum, our findings are as follows. First, since unconstrained and doubly-constrained spatial interaction models become increasingly difficult to estimate and/or less accurate to use compared to Half-life models as the spatial disaggregation increases choice of decay parameter estimation model should be considered in relation to level of disaggregation. Secondly, Half-life models are not affected by the systematic errors observed in the statistically derived models. Finally, using Half-life models for the estimation of decay parameters is simple which may make it easy to employ among practitioners lacking skills or computer means for the estimation of more complex statistically derived models.
Place, publisher, year, edition, pages
2016. Vol. 16, no 2, 344-362 p.
IdentifiersURN: urn:nbn:se:uu:diva-281078ISI: 000372328000004OAI: oai:DiVA.org:uu-281078DiVA: diva2:912729
FunderSwedish Research Council, 2012-5509