I'm running a binomial GWR on MGWR software with a synthetic database. On the global statistics and in GLM results, the ratio of the standard deviation over the mean is not high. However, on the GWLR model that ratio is over one or even bigger. Any clues on what might be causing it?
Aside from that, the local model's AICc, percent of deviance explained all improved compared to the global model, and the monte carlo spatial variability test gave a p-value of 0 for all variables.
This might be a result of skewness in your data. Another problem could be the presence of outliers that were not accounted for. Maybe you should try trimmed mean to see if there will be an improvement.
Gbenga Olayanju but that would be a post model treatment, not a data treatment, if I understood you correctly.
Since it's a synthetic data, I could correct the skewness problem in it's root, or treat it to minimize the effects. Anything you suggest in this regard?
My suggestion presumes that you are at the data treatment stage, however, you might need to look at your metadata and see if errors have not been indirectly introduced. Application of advanced regression analysis of the data set might also assist.
Am also of the opinion that you use the advanced regression analysis and see the result
Gbenga Olayanju hello, sorry for the late reply
In my case of study, this problem mostly arose from the variables with low correlation to the dependent variable. Plus, it might be due to the local modeling allowing heterogeneity to be explicitly modeled, since there are varied regions inside the sample.
However, I don't see how a trimmed mean would work since most of my independent variables are binary or low count variables. Plus, what tools do you mean by advanced regression analysis? Could you cite the papers you have in mind?
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude