I have been working on the geographically-weighted regression tool, exploring the spatial distribution of the various explanatory variables in the 'Devon rents' paper. One variable is distance to the nearest market town of the parish. When I was doing 'ordinary' regression this particular variable caused me some grief because it wasn't always significant. It kept changing its sign depending on the size of the dataset. The theory holds that the sign should be negative. Greater distance to market should lower the rent. Below is a map of the signs and coefficients of the market distances for the parishes. Some areas, notably to the west and the northeast (Devon and Herefordshire) are respectably negative, but Somerset and Dorset are positive. How can being further from the market increase your rent?