[R-sig-ME] FW: [Lme4-authors] sandwich estimators in lme4?

[DeCesare, Nicholas]

Hi All,
I’m wondering if you would ever consider incorporating Huber-White sandwhich estimators as an option for glmer in lme4.  Some google searching has revealed various posts about this, as well as the recent ‘robustlmm’ package (Gaussian data only).

My own interest stems from this being a recommended approach in the ecology literature.  A common application of glmms in ecology is the use of logistic regression to model animal selection of resources.  A random intercept per animal is often used to account for correlation of data within each animal, but in the response data the 1’s (autocorrelated animal locations) and 0’s (random locations in a polygon) can have different correlation structures.  Koper and Manseau (2009 - attached) recommended using a glmm with a random intercept per animal in combination with a Huber White sandwhich estimator to get proper variance estimates.    I can’t speak to the statistical validity of the approach, but it has some appeal for avoiding underestimates of variance of coefficients.

Thanks for all of your time and efforts –Nick

------------------------------------
Nick DeCesare
Research Wildlife Biologist | Montana Fish, Wildlife & Parks
406-542-5558 (office) | 406-370-3403 (cell)
ndecesare at mt.gov