[R-sig-ME] F test for two mixed-effects models

Christina Bogner christina.bogner at uni-bayreuth.de
Wed May 5 15:58:21 CEST 2010


Dear list,

I am stuck with an F test in a mixed-effects model. Some time ago Dr. 
Bates explained that if we had factors for fixed-effects, the model 
always spaned an intercept, even if we excluded it in the model formula. 
Now, I fit two models:

model1 = lme(response ~ factor1 + factor2 + x, random= ~1|plot, data=mydata)

model2 = lme(response ~ 0  + factor1 + factor2 + x, random= ~1|plot, 
data=mydata)

response and x are continuous, factor1, factor2 and plot are factors.
And I would like to test the hypothesis, that all model coefficients 
except the intercept are zero. For model2, instead of the intercept, I 
use the reference level (the same as the estimate of the intercept in 
model1, of course). For the test, I use the function wald.test in the 
package aod. To my surprise, the F value in the test is not the same for 
model1 and model2.
I certainly miss something important, but I can’t figure out what it is. 
Of course, the fixed-effects estimates of the factors are different 
(except for the reference level) and the var-cov-matrices differ as 
well. But, I thought, that the two models were equivalent and that I was 
testing the same hypothesis.

Any help would be greatly appreciated.

Christina




More information about the R-sig-mixed-models mailing list