Hi,
I have a GWR model with good r2 and AICC. I've tried as far as possible to remove co-linearity from the model, the explanatories for which were selected first by longitudinal MLR of the aggregated data (to avoid spatial auto-correllation), and afew more weeded out through OLS. I know there is still some spatial co-linearity there but as the explanatories vary in significance over space none of the remaining parameters may be considered redundant to the global model.
I would like now to predict the dependent variable's map for new scenarious by plugging in different parameter values to one or more of the co-efficients, just as one might in MLR (I appreciate that for GWR this relies on the substantial assumption that the co-efficients will be stable over time accross space).
But I'm concerned that local co-linearity may be inflating the effect in some locations for some parameters.
Any suggests as to how to model this? Is it legitmate in GWR to attempt to look at the effect of idividual variables in this way?
Thanks,
Neil
p.s. The data is already normalised to a percentage change from each variables mean (i.e. the regression is built on the variance not the absolute values).
