search

jakobdambon / varycoef

The R package varycoef implements Gaussian processes spatially varying coefficient models.
9 stars 2 forks source link

test significance of nugget variance #10

Open jakobdambon opened 2 years ago

jakobdambon commented 2 years ago

Wald test for nugget variance in summary output of an SVC_mle object: Is this sensible?

reinhardfurrer commented 2 years ago

You used 'W' for Wald, I see. I presumed it was a likelihood ratio test framework. For Wald, the p-value is (somewhat) equivalent to the standard error or CI. The question is rather, Wald test for the other variance parameters sensible? I presume (here to be discussed), that it would be more sensible to look at estimated likelihoods and derive CI for these. That means, your likelihood function is built up, plug-in all estimates and get the bounds. With this approach, you could also provide the CI for the range parameters. Computationally, it is quite straightforward and fast.

I think I brought this up some time ago, I wonder if an individual or all SVCs could be tested with generalized LRT (Held and Sabanes Boves, 5.5). Much should cancel...

fabsig commented 2 years ago

Are there any studies that compare the finite sample properties of Wald and LR tests for mixed effects / GP models?

reinhardfurrer commented 2 years ago

Not that I know of... but I have not checked in years... There are some permutation tests for overall dependency. Refs in https://doi.org/10.1007/978-3-030-57306-5_45 are not much of a help either.