Skip to content
Prev 4259 / 20628 Next

LRT tests in lmer

Hi,

In general I don't think transforming the fixed effect predictions by  
the inverse link  function works if you want to get the predicted  
expectation.  In this case you have to take into account the magnitude  
of the variance components.  The new predict function in MCMCglmm will  
do this for a MCMCglmm fit. By default the predictions will be on the  
data scale and all random effects marginalised, but you can also get  
predictions that include the random effects if you save their  
posterior distribution (i.e pr=TRUE)

Cheers,

Jarrod
On 11 Aug 2010, at 15:31, Ben Bolker wrote: