[R-sig-ME] Extracting random effect variance from a MuMIn-averaged object of several glmer models

Phillip Alday me @end|ng |rom ph||||p@|d@y@com
Sun Jul 17 18:58:41 CEST 2022

A naive approach would be to simply compute the pooled variance for each
random effect. For example, take the between-animal variance value for
the intercept from each model and use the usual formulas for computing
pooled variance. (The formulas for pooled standard deviation involve
squaring things back to the variance scale so might as well start from
the variances.)

I haven't thought in detail about the properties of this estimator and
it won't capture the original covariance structure, but it might be
"good enough" for the task at hand.

Hope that helps

On 13/7/22 1:28 pm, Nathan Jero wrote:
> Hi everyone,
> I'm working on analyzing the results from a trial of a new animal
> management technology, with a random intercept to capture variance
> between animals (I have multiple measurements of each animal).  No
> single model from my candidate set is obviously the best (AIC weights
> < 0.95, several models with delta AIC < 10), so I am planning on using
> MuMIn::model.avg to obtain an averaged model for inference.
> However, the random effect variance represents differences in how
> animals respond to the technology and is therefore of interest to me.
> Unfortunately, while obtaining an averaged model object and
> coefficients for the fixed effects has been easy, the random effect
> variance is not reported in summary(averaged.mod) and I haven't found
> another way to extract it from an object of class 'averaging' created
> by model.avg.  Is there a way to do this?
> All candidate models were fit using the same random effect structure
> and dataset in glmer() with the binomial family and logit link.
> I'll freely admit my lack of statistical knowledge, so if there are
> theoretical issues with this idea, I would love to learn.
> Thanks!
> Nathan

More information about the R-sig-mixed-models mailing list