[taking the liberty of cc'ing to r-sig-mixed-models] I've just put up a document on rpubs <http://rpubs.com/bbolker/varwald> that shows how to get the asymptotic (Wald) variance-covariance matrix for the *ML* estimates, using the internal 'devfun2' function. It will take a bit more work to do the same thing for the REML estimates, but maybe this document will provide enough hints that someone else can do it ... cheers Ben Bolker
On 14-05-23 12:26 AM, Douglas Hawkins wrote:
Hi, Ben, I saw your thread from a couple years back on the the standard errors of variance component estimates. I see lots of nested random effects data sets where we estimate the variance components by REML and then want to get asymptotic confidence intervals for partial sums of them. Traditionally, I have done this in SAS using the asymptotic covariance matrix of the estimates. But I would much rather do it in R. It sounds like you were writing code that would do this -- have there been any developments along these lines? Best, Doug Hawkins -- Douglas M. Hawkins Professor, School of Statistics University of Minnesota 313 Ford Hall Minneapolis, MN 55455-0493 +1 612 624 4166Hi,