[R-sig-ME] Standard errors of variances; negative variance

[Timothee]

Hi all,

For what it is worth, unfortunately it requires a license, you can 
obtain negative estimates of variance parameters in asreml-R. I agree 
with Thierry that negative estimates of variances are odd, but I have 
seen at least one case where they were practical: to identify that 
observations within a grouping level were actually very dissimilar while 
there was almost no variation across grouping level averages. That 
result was contrary to my expectation given the experimental design and 
I may have missed it otherwise. Arguably my models were misspecified, 
but that odd result helped me understand the biology of my system and 
specify more appropriate models later.

As for uncertainties of variances I usually go with Bayesian packages : 
MCMCglmm, brms, rstan, rjags... (or, again, asreml-R).

Best regards,

*************************************
Timothée Bonnet, PhD
Chargé de recherche / Researcher

Centre d’Etudes Biologiques de Chizé
UMR 7372 Université de la Rochelle-CNRS
405 route de Prissé la Charrière
79360 Villiers en Bois

timotheebonnetc using gmail.com
https://timotheenivalis.github.io//