Title:
Some methods for analyzing and correcting for spatial autocorrelation
Authors:
Emil O. W. Kirkegaard
Abstract
A method for examining spatial autocorrelation based on distance scores was examined and found to be lacking. Because of this an alternative measure was proposed: k nearest spatial neighbor regression. Methods were examined across simulated datasets to see if they could successfully remove SAC and distinguish between a true and a spurious cause. Furthermore, the methods were compared to a more traditional measure of spatial autocorrelation, Moran's I.
Key words: spatial autocorrelation, Moran's I, correlation of distances, k nearest spatial neighbor, knsn
PDF + files: https://osf.io/ea2bm/
Back to [Archive] Other discussions
Title:
Some methods for analyzing and correcting for spatial autocorrelation
Authors:
Emil O. W. Kirkegaard
Abstract
A method for examining spatial autocorrelation based on distance scores was examined and found to be lacking. Because of this an alternative measure was proposed: k nearest spatial neighbor regression. Methods were examined across simulated datasets to see if they could successfully remove SAC and distinguish between a true and a spurious cause. Furthermore, the methods were compared to a more traditional measure of spatial autocorrelation, Moran's I.
Key words: spatial autocorrelation, Moran's I, correlation of distances, k nearest spatial neighbor, knsn
PDF + files: https://osf.io/ea2bm/
Initially I had thought about writing a review but as I progressed, I decided to write a rebuttal. You can find the draft at the following link (code and raw files attached):
https://docs.google.com/document/d/1GxFrd20No1YksRKgw8-LkwQQPnV4uRjeD5X6didLNh0/edit?usp=sharing
Good work.
What I had in mind is that 1) I add some more stuff to the stats paper, especially spatial local regression. 2) Then we jointly write a paper on how the SNP and polygenic results perform after these new corrections for SAC and submit it to OBG.
What I had in mind is that 1) I add some more stuff to the stats paper, especially spatial local regression. 2) Then we jointly write a paper on how the SNP and polygenic results perform after these new corrections for SAC and submit it to OBG.
Good work.
What I had in mind is that 1) I add some more stuff to the stats paper, especially spatial local regression. 2) Then we jointly write a paper on how the SNP and polygenic results perform after these new corrections for SAC and submit it to OBG.
That is fine with me as long as I can be first author on the joint paper
Good work.
What I had in mind is that 1) I add some more stuff to the stats paper, especially spatial local regression. 2) Then we jointly write a paper on how the SNP and polygenic results perform after these new corrections for SAC and submit it to OBG.
That is fine with me as long as I can be first author on the joint paper
Whatever. :)
New version: https://osf.io/ea2bm/
5900 words.
Update concerns the addition of more comparisons and the new method of spatial local regression. This method can only be used to control for SAC and shows similar results to KNSNR.
5900 words.
Update concerns the addition of more comparisons and the new method of spatial local regression. This method can only be used to control for SAC and shows similar results to KNSNR.
I have asked Garry Gelade to comment on this.
He said he didn't have time.
I published this with Winnower.
---
Since then I found an error. The values for dataset 12 in the measurements comparison are wrong. I'm not sure why. There is no randomness component to dataset 12. No change in the overall pattern tho.
I also changed some default values of SAC_slr() which means that the SLR results don't match perfectly. They do if one changes the input values however. The new default summary statistic is trimmed mean, but the old was just mean. Also, the new default is to use the case itself in the clusters, while the old was not to do this.
I published this with Winnower.
---
Since then I found an error. The values for dataset 12 in the measurements comparison are wrong. I'm not sure why. There is no randomness component to dataset 12. No change in the overall pattern tho.
I also changed some default values of SAC_slr() which means that the SLR results don't match perfectly. They do if one changes the input values however. The new default summary statistic is trimmed mean, but the old was just mean. Also, the new default is to use the case itself in the clusters, while the old was not to do this.
© Copyright 2024 OpenPsych - All Rights Reserved