This report considers the issues of collinearity arising from the use of Geographical Weighted Regression (GWR) to describe and/or explain spatial variations in car ownership and use. GWR is a modelling approach which aims to explore possible patterns of spatial heterogeneity in a regression model of spatially-located data points, by considering that the independent variables in a standard global regression model may, in fact, vary in their impact across space. There is an extensive literature on the use of GWR in spatial analysis and a number of statistical programs have been developed to construct GWR models. In this work, two examples – the average distance travelled by car in each Data Zone in Scotland; and the average age of cars in each Medium Super Output Area in England and Wales – were examined. In both cases, GWR models were successfully developed and ‘explained’ a significant amount of additional variation over and above that explained by the standard global regression model. However, in both cases, serious issues of collinearity also arose. Attempts to control for collinearity were only partially successful, but were sufficient to indicate that the basic spatial patterns were probably part of the dataset. The implication is that the impacts of socio-economic variables on transport variables are unlikely to be spatial invariant, thereby requiring further investigation and understanding.

Want to know more about this project?