[R] Perhaps Off-topic lme question

From: Berton Gunter <gunter.berton_at_gene.com>
Date: Wed 13 Apr 2005 - 08:43:31 EST


A question on lme() :

details: nlme() in R 2.1.0 beta or 2.0.1

The data,y, consisted of 82 data value in 5 groups of sizes 3 9 8 28 34 .

I fit a simple one level random effects model by:

myfit <- lme( y~1, rand = ~1|Group)

The REML estimates of between and within Group effects are .0032 and .53, respectively; the between group component is essentially zero as is clearly evident from a plot of the data. So, thus far, no problem.

However, the confidence interval for the between Groups sd that I get from intervals(myfit) goes from essentially 0 to infinity (6 x 10^13, actually). I assume that this is because the between component estimate is too close to the boundary of 0 so that the likelihood approximations with default control values fail, but I would appreciate a more definitive comment from someone who knows what they're talking about.

If anyone cares to try this, the data in group order are below (i.e., first 3 from Group 1, next 9 from Group 2, etc.).

Cheers to all,
Bert Gunter

14.7
15.8
14.6
15.18
15.04
15.23
15.37
15.3
15.13
16.1
15.5
15.53
15.83
15.53
15.61
15.69
16.17
15.27
15.34
15.3
15.48
15.8
15.1
15.15
15.12
15.3
14.88
14.36
15.62
14.61
15.27
16.05
14.93
15.19
15.23
15.62
15.9
15.21
15.01
14.86
16.55
15.75
15.04
15.55
15.54
14.66
15.9
15.24
15.25
15.18
14.83
15.17
15.05
15.37
14.87
15.33
15.55
15.5
15.89
15.67
15.55
15.65
15.8
15.84
16.25
14.71
15.39
16.11
15.64
15.99
16.54
14.46
15.47
15.11
15.59
15.49
15.08
14.6
14.32
13.73
16.61
14.19

______________________________________________
R-help@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Wed Apr 13 08:52:29 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:31:07 EST