As far as I understand, spatial smoothing convert the post code (which is 1.7 million in the UK -> a lot) to a post code factor, where the level of factor reduce significantly, to improve the predictive of the model. So how are the smoothing do that? What is the differences between distance based and adjacency smoothing? Thanks very much,
Spatial Smoothing is discussed in Section 2.2 of Chapter 17. The idea is that you reduce some of the error in your model in a given rating area (be that postcode or whatever level you have decided to rate your business on) by bringing in information from neighbouring rating areas. With distance-based spatial smoothing it simply depends on the distance between neighbouring rating areas as the crow flies (so it is typically used for weather-related perils), whereas adjacency-based spatial smoothing, assumes each rating area is influenced by its neighbouring ones, and those ones are influenced by their neighbouring ones and so on. Have another read of that section now and see if the above helps.