Why does varcomp() give discrepant results compared with raov() in this case?
The example is from MASS 6.11 -- data are in Table 6.2. Two measurements of
concentration were taken on three batches of some substance by six
laboratories. Replic is a factor I added which identifies the replicated
measurements within labs (redundantly in this case). The expected results are
given by raov() whether or not Replic is included in the model, but varcomp()
somehow comes up with an additional component when Replic is included.
anova(raov(Conc~Lab/Batch,tem))
Df Sum of Sq Mean Sq Est. Var.
Lab 5 1.890214 0.3780428 0.06016824
Batch %in% Lab 12 0.204400 0.0170333 0.00536806
Residuals 18 0.113350 0.0062972 0.00629722
anova(raov(Conc~Lab/Batch/Replic,tem))
Df Sum of Sq Mean Sq Est. Var.
Lab 5 1.890214 0.3780428 0.06016824
Batch %in% Lab 12 0.204400 0.0170333 0.00536806
Replic %in% (Lab/Batch) 18 0.113350 0.0062972 0.00629722
varcomp(Conc~Lab/Batch,temp)
Variances:
Lab Batch %in% Lab Residuals
0.06016824 0.005368056 0.006297222
varcomp(Conc~Lab/Batch/Replic,temp)
Variances:
Lab Batch %in% Lab Replic %in% (Lab/Batch) Residuals
0.06016824 0.005368056 0.04569358 -0.03939636
varcomp(Conc~Lab/Batch/Replic,temp,method='reml')
Variances:
Lab Batch %in% Lab Replic %in% (Lab/Batch) Residuals
0.06016829 0.005368059 0.00002054951 0.00627666
Thanks.
David Siev
USDA Center for Veterinary Biologics
510 South 17th Street
Ames, IA 50010
(515) 232-5785 ext. 153
david.siev@usda.gov
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
|