In a logistic regression the significance (p) of each covariate can be calculated.
For the case of a geographically weighted logistic regression (GWLR), I currently am working on a model for event prediction. This means I do not have one global model, but a different model (i.e. different Betas) at each geographical location.
In order to choose what covariates are included in the models I need to be able to assess each covariate for its contribution to the models. Doing so by making use of a p-value seems like the obvious choice.
I am able to calculate the p-values for each covariate for each individual model. Interapolating this to a raster can result in some locations having a high p-value (covariate is not good at explaining here) and low p-value (covariate is good in explaining here). But this only marginally helps me in deciding what covariates to include in the models. For the sake of simplicity I want all the models to include the same covariates.
How can a global p-value (or some other statistic) be calculated per covariate to determine whether a covariate should, or should not, be included in the models?
Since local P-values are calculated with the covariance matrix, and the covariance matrix is based on the logit of the model (which is geographically weighted), calculation of a global p-value (if at all possible) should include a geographic weighting as well.
