[R-sig-ME] prediction intervals and lmer

[Muldoon, Ariel]

Ah, I think I see.  

>confidence intervals, incorporating either (1) only the uncertainty on the fixed-effect parameters (beta) or (2) >uncertainty on beta plus variation due to random effects [this would be a sort-of confidence interval for a population->level prediction, i.e. the expected variation (conditional on the random-effect parameter effects) of the mean of a large >number of samples from a _single_ previously unobserved block].

Rather than making a confidence interval for (1) inference about the "typical" or "average" subject (if subject is your random effect) where we only account for the uncertainty around the estimated mean, we might want to (2) make inference to some unobserved subject.  We don't know where that new subject falls in the distribution of all subjects, so we account both for the uncertainty in the estimated mean and the additional uncertainty of the new subject when building the confidence interval.

Thanks for the clarification! 



_______________________________________________
R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models