SE for predictions in lme

Hi Marta,

My understanding of this is that the SE*2 gives 95% confidence  
intervals on predictions. You can add a bonferroni type adjustment by  
varying the *2 (which is really 1.96)

The SE2 gives prediction intervals. Confidence intervals of  
predictions tell you how well you have predicted the mean. Prediction  
intervals tell where you could expect the next to see the next value  
sampled. The key point is that the prediction interval tells you about  
the distribution of values, not the uncertainty in determining the  
population mean.

Therefore as far as I can tell, the CIs are the most useful for making  
inferences about differences in the conditional means/BLUPs

You can also use the function predictSE.lme from the package  
AICcmodavg to produce prediction and SEs from an lme object and then  
calculate CIs accordingly. I like this method as it gives a reference  
to the method:

"predictSE.lme? computes predicted values based on fixed effects and  
associated standard errors.
Standard errors are approximated using the delta method (Oehlert 1992)"

These two methods result in identical predictions and almost identical  
SEs for my dataset

Some confusion arise for me because the predict.lme function calls the  
prediction Best linear Unbiased Predictions (BLUPs) which combine  
random and fixed effects, and the predictSE.lme function states that  
these predictions do not incorporate the random effects. Yet they give  
identical predictions for me.

Hope this helps

Phil

Quoting marta rufino <marta.m.rufino at gmail.com>: