[R-sig-ME] Are prediction interval values ungettable?
pmilin at ff.uns.ac.rs
Sun Nov 21 18:11:54 CET 2010
This is right, but I have compared values I get with the below procedure
explained by Andrew and the end points of whiskers on the caterpillar
plot (I tried to approximate them the best I could), and they do match
for five different data sets and models.
On 21/11/10 18:07, Dimitris Rizopoulos wrote:
> On 11/20/2010 2:21 PM, Petar Milin wrote:
>> This is great! Many thanks!
>> Now, practically, I can build: +/- 1.96*my.se
>> Am I right?
> maybe one thing that I think needs to be kept in mind is that the
> posterior variances that ranef(..., postVar = TRUE) returns condition
> on the MLEs and do not take their variability into account.
>> On 20/11/10 12:13, Andrew Robinson wrote:
>>> last I tried this, the estimated variance of the random effects is
>>> (optionally) stored as an attribute. So, something like this should
>>> rfg<- ranef(my.lmer, postVar=TRUE)
>>> my.se<http://my.se> = sqrt(as.numeric(attributes(rfg$group)$postVar))
>>> On Sat, Nov 20, 2010 at 8:46 PM, Petar Milin wrote:
>>> How can one get upper and lower limits of a prediction interval
>>> for random-effect levels; the exact exact values, numbers? They
>>> are shown on caterpillar plot using ranef() with
>>> argument postVar=TRUE, but I would like to know them. A while ago,
>>> discussions were opened on "Confidence Intervals for Random Effect
>>> BLUP's", but the answer was never clear:
>> [[alternative HTML version deleted]]
>> R-sig-mixed-models at r-project.org mailing list
More information about the R-sig-mixed-models