Gaussian Process estimate interpretation

Interfaces brms
1

Please also provide the following information in addition to your question:

  • Operating System:Windows7
  • brms Version:2.5.0

Hello!

I was using the function gp with longitude and latitude in decimal degree and the parameter gr set to TRUE. My question is now what the estimates for sdgp and lscale tell me. Could I interpret the estimate as a distance? If so on what unit is it? Particularly what if the estimate of lscale is zero, does it mean, that there is no spatial autocorrelation? And given that is it possible to drop the gp-term then from the model?

Thanks for your help!

2

The case studies of Michael may be helpful. Part 1 here: https://betanalpha.github.io/assets/case_studies/gp_part1/part1.html

3

Thank you for the hint!

4

Hi Paul,
the more i get into it, i would like to ask, whether the use of the gp function is the recommended way to control for spatial autocorrelation within brms?
Or are there perhaps different/better option i am not aware of?
Thank you!

5

For discrete spatial locations, CAR-autocorrelation may be a good solution (see cor_car). For continuous spatial data, you may consider a two-dimensional spline of the form s(x, z), which will be considerable for efficient than exact GPS.

Approximate GPs that are considerable faster than exact GPs are coming to brms as well, but we still have to run more experiments before I would recommend it.

6

Ok! Thank you!