Message-ID: <loom.20120217T161249-694@post.gmane.org>
Date: 2012-02-17T15:15:15Z
From: Ben Bolker
Subject: LMER vs MLwiN
W Robert Long <longrob604 at ...> writes:
>
> Hi Federico
>
> Thank you very much.
>
> However, the model I'm trying to fit has no fixed effects. The MLwiN
> output gives me:
> beta_0_hat: 1.012(0.107) conditional log odds
> var_f0_hat: 0.031(0.066) variance between communues
> var_v0_hat: 0.760(0.141) variance between schools within commune
> var_u0_hat: 0.186(0.38) variance between classes within schools
>
> This is what I am trying to duplicate in R with the same data.
>
> BTW I am using 64 bit R
OK, based on this output you shouldn't include 'pupil' in your
random effects specification (now that I think of it, you probably
shouldn't anyway, because it's unidentifiable for a Bernoulli outcome).
If you wanted you could redo my example with your observed effects
(e.g. u.commune = rnorm(n.comm,sd=sqrt(0.031)) ...)
Note that it is a bit harder to get uncertainty estimates on the
variance parameters in lme4.
>