Dear Doug & Frank:
Thanks for the reply, Doug. First a comment, then another
question for you, Doug, if you have time:
1. COMMENT: Here's an ugly hack that would seem to answer
Frank's question using 'lmer':
(vt <- with(dtf, tapply(x, trth, sd)))
(vr <- vt[2]/vt[1])
mod1b <
On 10/15/06, Douglas Bates <[EMAIL PROTECTED]> wrote:
> On 10/14/06, Spencer Graves <[EMAIL PROTECTED]> wrote:
> > You want to estimate a different 'cs' variance for each level of
> > 'trth', as indicated by the following summary from your 'fake data set':
> >
> > > tapply(dtf$x, dtf$trth, s
On 10/14/06, Spencer Graves <[EMAIL PROTECTED]> wrote:
> You want to estimate a different 'cs' variance for each level of
> 'trth', as indicated by the following summary from your 'fake data set':
>
> > tapply(dtf$x, dtf$trth, sd)
>FALSE TRUE
> 1.532251 8.378206
>
> Since var(x
You want to estimate a different 'cs' variance for each level of
'trth', as indicated by the following summary from your 'fake data set':
> tapply(dtf$x, dtf$trth, sd)
FALSE TRUE
1.532251 8.378206
Since var(x[trth]) > var(x[!trth]), I thought that the following
should prod
I have a model:
mod1<-lmer( x ~ (1|rtr)+ trth/(1|cs) , data=dtf) #
Here, cs and rtr are crossed random effects.
cs 1-5 are of type TRUE, cs 6-10 are of type FALSE,
so cs is nested in trth, which is fixed.
So for cs I should get a fit for 1-5 and 6-10.
This appears to be the case from the random