I have irregular gridded data and I wish to resample from this data in such a way that the spatial correlation in the data is preserved. I will assume that grids that are within 300Kms of one another are correlated. I have 153 grids in all and some grids have 2 neighbours (i.e. 2 grids that lie within 300 kilometres) and some grids have as many as 11 neighbours. I wish to know how to draw a bootstrap sample from this grid. My ultimate objective is to run a regression on the gridded data that I have to get at the distribution of the standard error of the regression coefficient. Many thanks!

  • $\begingroup$ After bootstrapping, some will even have "neighbours" at 0 km. $\endgroup$ – cbeleites Jan 23 '15 at 10:38
  • $\begingroup$ All grids have neighbours some just have too many and some little. $\endgroup$ – Ridhima Jan 23 '15 at 11:04
  • $\begingroup$ For this you typically have blocks of data removed at once (aptly called block bootstrapping), so there are sufficient other observations with nearby neighbors to reliably calculate whatever spatial statistics. (Another common approach is to do leave-one-out resampling.) $\endgroup$ – Andy W Jan 23 '15 at 13:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.