Is one of your variance components essentially estimating zero?
If you check the numbers, delete the smallest one, and then do "anova"
comparing the two fits, you might find that the one you deleted was not
statistically significant.
If I'm not mistaken, "lme" estimates the logarithms of the
variance components. When one of them is zero, the logarithm wants to
go to (-Inf). In that case, "lme" will still return an answer.
However, "intervals" complains, because the observed information
matrix
for the parameter estimates is singular in the direction of log(variance
component) that wants to go to (-Inf). Doug Bates, the developer of
"lme" has taught many people about how to fix that problem using
profile
likelihood. I have not seen "lme4" yet, so the problem may already
have
been fixed. If any if this is inaccurate, I hope Doug will correct me.
hope this helps.
Spencer Graves
Bill Shipley wrote:
>I am getting an error message concerning the estimation of confidence
>intervals when fitting a mixed model and don't know what the problem is,
>or its solution.
>
>
>
>Just to provide context: the model is describing the effects of age,
>exp(age), harvest age, and climate variables on bighorn horn annular
>length.
>
>
>
>The data structure is repeated measures (between individuals, within
>individuals over time).
>
>
>
>Id is a random effect (there are between 3-11 horn measurements per ram,
>one horn measurement per age, over the 25 year period in the dataset).
>
>
>
>The mixed effect results is unable to provide confidence intervals for
>the fixed and random effects because: of an error in the
>variance-covariance structure. The error says that the intervals are
>non-positive definitive.
>
>
>
>
>
>Bill Shipley
>
>
>
>
> [[alternative HTML version deleted]]
>
>______________________________________________
>R-help at stat.math.ethz.ch mailing list
>https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html
>
>