Evaluating the use of alternative distance metrics in spatial regression analysis of health data: A spatio-temporal comparison.

A method is discussed to enhance the reliability of multivariate spatial regression analysis: alternative values of the Minkowski distance metric are used in the spatial weight matrix. The method is tested on an analysis of the association between heart disease incidence and a pool of socio-economic variables in Calgary over two consecutive census surveys. The method provides a reliable model, which can guide locational decisions to mitigate present and future disease incidence. The model is underpinned by a quantitative definition of neighbourhood connectivity throughout the city. Such connectivity, usually described by Euclidean distance, can be more effectively described by a specifically calibrated distance metric. The analytical results are meaningful, robust to neighbourhood size, and relatively constant over time. Owing to its effectiveness and simplicity, the procedure is generalizable to other health and socioeconomic analysis. An automatic implementation is suggested, to assist in the definition of reliable spatial regression models.