I am attempting to run a Geographically Weighted Regression for the final part of a thesis project, but I have so far been unsuccessful. I receive the error 040038, which occurs whenever there is redundancy in the explanatory variables. However, when I run an Ordinary Least Squares to check for redundancy in the explanatory variables, the VIF values are all low (<7.5) and the probabilities are significant. The factors I see that are present in the OLS (high AIC, 10E6; low R^2, 10E-3; residuals not normally distributed) are not among the factors listed for the description of this error: 040038: Results cannot be computed because of severe model design problems.—Help | ArcGIS Desktop
The attached set of radars is what I am using to predict out to the full extent of Illinois, and I am aware that GWR does not work very well with point features. However, extracting to multiple point features has been the only way I have been able to append explanatory variable data to be in the same feature as the dependent variables. The features do not reassemble properly when I attempt to convert to polyline and then back to polygon. Could formatting issues or processing environments be contributing to the 040038 error?
I have discussed this problem with faculty and plan on meeting again this week to try and resolve these issues. I am also attempting to clean up the data by building a model. All explanatory variable rasters were set to the same cell size and processing extent in the model, but I am still unsure whether this will resolve the GWR when the explanatory variables are appended to the dependent variable features to run the model. I have so far only run GWR outside the model rather than with a full model.
Would there also be an alternative to GWR that can accomplish the same goal and have the dependent variable predicted out to the full extent as it is currently formatted? OLS only predicts to the extent of the merged radar file.
Any suggestions are greatly appreciated!