michal.cudzilo

[GWR] Missing predictions when number of neighbours is given

Discussion created by michal.cudzilo on Sep 8, 2011
Latest reply on Sep 27, 2011 by lrosenshein-esristaff
Hi everyone,

I have a set of 500 observation, dispearsed in the space, which distribution varies in space (mainly concentrated in the center). I used adaptive kernel to make a model. When AIC or CV criterion is set, everything is ok (besides hetereskedastity of residuals). I tried to define number of neighbours, when gwr is calculating and... yes, there is some strange results. Operation is successful, but I don't know why, some observation are missed (about 1/3 - those, which were on the outskirts, not in the centre). They are in attribute table, but all fields connected with estimation are fulfilled with <<Null>> value.

Any ideas why it happened? As far as I read GWR theory, all observation should be estimated in this case. Only phisically radius should increase, 'cos we take into consideration constant number of neighbours (not distance). So..?

Thx in advance and have a nice evening,
Michal

Outcomes