Dear R-experts,
I am working with generalized linear mixed models in lmer and have a
question on the interpretation of variance components. Specifically,
I would like to know how do we estimate the residual variance for random
effects when the value reported for the
residual error represents a scale parameter.
In lme4, The summary statement for a random intercept model gives me
something like:
Random effects:
Groups Name Variance Std.Dev.
Transect (Intercept) 0.074105 0.27222
number of obs: 70, groups: Transect, 10
Estimated scale (compare to 1 ) 1.053643
What I'd like to know is the % of total variation explained by the random
variation in intercepts; i.e., var(intercept) / [var(intercept) +
var(residual)].
A search on the previous posts in R -sig led me to read that
the residual variance of a glm or glmm with logarithm link is defined as
(pi^2)/3. As such, if the output from model run in lmer gives 'Estimated
scale (compare to
1)', how would I estimate the residual variance? . Do I have to multiply
the reported scale parameter (but not the variance components of
the random terms) by (pi^2)/3 to arrive at the actual residual variance.
And if so, do I have to make any changes to the variance components of the
random terms for the intercept (Transect).
Thanks for any helpful suggestions !
Byju
Graduate Student, FNR, Purdue University
[[alternative HTML version deleted]]