Hello everybody,
I'm carrying out an analysis using the gwr command for geographically weighted regression. The estimated local coefficients (resulting from the gwr) can change depending on the choice of the spatial kernel and bandwidth, and for my study I specifically need to specify the bandwidth (in a way that the estimation takes into account only the closer observations) before running the command gwr.
At the moment, I just reduced the percentage of observations sampled by stata in the default bandwidth calibration process that the software does by default when running the gwr command, but I know that it's absolutely not the right way to do it and that it would be much better to specify the bandwidth.
gwr yvar xvar, east(_CX) north(_CY) sample(25) saving(modelN_gwrN) mcsave(mc_se_modelN_gwrN) replace
My question would come as very naive for some of you, but I'm having some troubles in finding a way to calibrate the bandwidth using stata. In the literature I found that the optimal bandwidth for a fixed spatial kernel with a Gaussian weighting function can be chosen using different methods - like the Akaike Information Criterion (AIC). Is there a way to find this optimal bandwidth in stata, so that in the following line of command I can specify it being sure that I'm selecting the more adequate one?
gwr yvar xvar, east(_CX) north(_CY) bandwidth(#) saving(modelN_gwrN) mcsave(mc_se_modelN_gwrN) replace
I know I may not have been super clear, so I remain available for further clarifications.
Thank you for your time!
Dario
I'm carrying out an analysis using the gwr command for geographically weighted regression. The estimated local coefficients (resulting from the gwr) can change depending on the choice of the spatial kernel and bandwidth, and for my study I specifically need to specify the bandwidth (in a way that the estimation takes into account only the closer observations) before running the command gwr.
At the moment, I just reduced the percentage of observations sampled by stata in the default bandwidth calibration process that the software does by default when running the gwr command, but I know that it's absolutely not the right way to do it and that it would be much better to specify the bandwidth.
gwr yvar xvar, east(_CX) north(_CY) sample(25) saving(modelN_gwrN) mcsave(mc_se_modelN_gwrN) replace
My question would come as very naive for some of you, but I'm having some troubles in finding a way to calibrate the bandwidth using stata. In the literature I found that the optimal bandwidth for a fixed spatial kernel with a Gaussian weighting function can be chosen using different methods - like the Akaike Information Criterion (AIC). Is there a way to find this optimal bandwidth in stata, so that in the following line of command I can specify it being sure that I'm selecting the more adequate one?
gwr yvar xvar, east(_CX) north(_CY) bandwidth(#) saving(modelN_gwrN) mcsave(mc_se_modelN_gwrN) replace
I know I may not have been super clear, so I remain available for further clarifications.
Thank you for your time!
Dario
Comment