Testing between competing lme models should be done via LRTs and the anova.lme() function. However, take care of the fitting procedure (REML vs ML), especially in case you also change the fixed-effects. The latter has been recently discussed on the list.
I hope it helps.
Best, Dimitris
---- Dimitris Rizopoulos Ph.D. Student Biostatistical Centre School of Public Health Catholic University of Leuven
Address: Kapucijnenvoer 35, Leuven, Belgium Tel: +32/16/336899 Fax: +32/16/337015 Web: http://www.med.kuleuven.ac.be/biostat http://www.student.kuleuven.ac.be/~m0390867/dimitris.htm
----- Original Message ----- From: "Christoph Scherber" <[EMAIL PROTECTED]>
To: "Dieter Menne" <[EMAIL PROTECTED]>
Cc: <r-help@stat.math.ethz.ch>
Sent: Friday, February 04, 2005 10:09 AM
Subject: Re: [R] publishing random effects from lme
Hi Dieter,
Yes, Iīve tried both options. The anova(lme(...)) gives me good results for the fixed effects part, but what Iīm specifically interested in is what to do with the random effects.
I have tried glmmPQL (generalized linear mixed-effects models), which did in fact greatly help account for heteroscedasticity, but I canīt do model simplification with these models (and theyīre still heavily debated, as I read from previous postings to "R Help".
How would you deal with the random effects part of the models when publishing results from lme?
Thanks for your help! Christoph
### Here are my original questions once again (with an example below):
1) What is the total variance of the random effects at each level?
(2) How can I test the significance of the variance components?
(3) Is there something like an "r squared" for the whole model which I can state? ##it seems, there isnīt (as I learned from a previous posting
The data come from an experiment on plant performance with and without insecticide, with and without grasses present, and across different levels of plant diversity ("div").
Thanks for your help! Christoph.
lme(asin(sqrt(response)) ~ treatment + logb(div + 1, 2) + grass,
random = ~ 1 | plotcode/treatment, na.action = na.exclude, method = "ML")
Linear mixed-effects model fit by maximum likelihood
Data: NULL AIC BIC logLik -290.4181 -268.719 152.209
Random effects: Formula: ~ 1 | plotcode (Intercept) StdDev: 0.04176364
Formula: ~ 1 | treatment %in% plotcode (Intercept) Residual StdDev: 0.08660458 0.00833387
Fixed effects: asin(sqrt(response)) ~ treatment + logb(div + 1, 2) + grass
Value Std.Error DF t-value p-value
(Intercept) 0.1858065 0.01858581 81 9.997225 <.0001
treatment 0.0201384 0.00687832 81 2.927803 0.0044
logb(div + 1, 2) -0.0203301 0.00690074 79 -2.946073 0.0042
grass 0.0428934 0.01802506 79 2.379656 0.0197
Standardized Within-Group Residuals: Min Q1 Med Q3 Max -0.2033155 -0.05739679 -0.00943737 0.04045958 0.3637217
Number of Observations: 164 Number of Groups: plotcode ansatz %in% plotcode 82 164
Dieter Menne wrote:
Suppose I have a linear mixed-effects model (from the package nlme) with nested random effects (see below); how would I present the results fromthe random effects part in a publication?
Have you tried anova(lme(....))?
Your asin(sqrt()) looks a bit like these are percentages of counts. The method is still quoted in old books, but has fallen a bit out of favor. Have you thought of some glm model instead (http://www.stats.ox.ac.uk/pub/MASS4/)?
Dieter Menne
______________________________________________
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
______________________________________________
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
______________________________________________ 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
