Skip to content
Prev 5607 / 20628 Next

DF in lme

On 17/03/2011 17:43, Ben Bolker wrote:
Aplogies, I didn't notice the cc had dissapeared.
I understand what you say about me actually only having in effect 10 
datapoints either way. When you say "

(although each of those data points might be fairly precise because of
all the subsampling you have done)"


Do you mean that those "in effect" 10 datapoints in the mixed model 
would be more accurate, than the averaged values and a normal 
anova/t.test/permutation test?
Or just in general my 10 datapoints are pretty accurate because of the 
subsampling - that they're both as accurate either way? I suppose that 
doing a non-parametric test on the averaged values, deals with 
inequalities of variance too, where in the mixed model I'd perhaps try 
some weighting option. T-tests and Permutation tests agree on estmates 
and p-values, Wilcocxon seemed a long way off from them. R^2 values in 
my mixed model are lower than in the anova, but then I understand R^2 is 
not the best measure, so all in all, I'm not too worried about the 
normality. Depending on what you meant about

(although each of those data points might be fairly precise because of
all the subsampling you have done)"

I'm considering the permutation test with the 10 datapoints, over the 
mixed model with many, if the points however for the mixed model are 
more accurate than the 10, I'll keep the mixed model.

Thanks,
Ben.