Closed famuvie closed 9 years ago
It turns out that for REML estimation of Linear Mixed Models, is not clear in general how to compute p-values for the variance components.
The same arguments for which R package lme4
does not report p-values apply to breedR. Specially when individual-level random effects are used (e.g. additive-genetic, or spatial effects).
Of course I understand the need to answer the question to whether a variance component is significant or not. I recommend fitting the model with and without the random effect in question, and using Model Selection tools.
The easiest one is looking at the AIC, and select the model with a lower value. Alternatively, it is possible to perform a Likelihood Ratio Test (LRT). However, this p-value will tend to be conservative, since the parameter value being tested is in the boundary of the parameter space (Bates 2010, Sec. 2.2.4).
I agree that breedR should provide means to perform these assessments more comfortably.
On behalf of Bruno FADY, PhD, Directeur de recherches, INRA.