Skip to content
Prev 11072 / 20628 Next
Original Raw

reference for standard error of estimated variance components

Ben Bolker 
Another option for diagnosing problems with the fit is to
compute the Hessian of the variance-covariance parameters, which
are fitted with a Cholesky-factor parameterization ... this
doesn't actually get you the uncertainty of the variance-covariance
matrices, but it gives something that is arguably more useful.
https://github.com/lme4/lme4/issues/120 has some useful machinery
written by Rune Haubo which we eventually hope to integrate into
the package.

  This should be a lot cheaper than likelihood profiling too
(takes only O(p^2) function evaluations where p is the number
of model parameters)

library(lme4)
fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
dd <- update(fm1,devFunOnly=TRUE)
library(numDeriv)
h <- hessian(dd,getME(fm1,"theta"))

Various diagnostics:

rcond(h)
diag(solve(h))
On 13-11-15 08:10 PM, John Maindonald wrote:

Thread (12 messages)

Szymek Drobniak Define prior in MCMCglmm Nov 13 Maria Paola Bissiri Define prior in MCMCglmm Nov 14 Jarrod Hadfield Define prior in MCMCglmm Nov 14 Szymek Drobniak Define prior in MCMCglmm Nov 15 Jake Westfall reference for standard error of estimated variance components Nov 15 John Maindonald reference for standard error of estimated variance components Nov 15 Ben Bolker reference for standard error of estimated variance components Nov 15 Jarrod Hadfield Define prior in MCMCglmm Nov 15 Maria Paola Bissiri Define prior in MCMCglmm Nov 20 Szymek Drobniak Define prior in MCMCglmm Nov 20 Maria Paola Bissiri Define prior in MCMCglmm Nov 21 Szymek Drobniak Define prior in MCMCglmm Nov 21