[R-sig-ME] heterogeneous variance models

[Aaron Mackey]

Thanks for this. What's confusing to me in the example is that the residual
std. dev is still 1.3100 whether the dummy term is included or not; I would
have thought that the residual sd without the dummy term would be
intermediate between the two sex-specific sd's seen in the full model
(reported as 1.0521 and 1.311).  The reported residuals also don't seem to
change between the two models.

thanks for any clarification you can provide,
-Aaron

On Sun, May 10, 2015 at 4:22 PM, Ben Bolker <bbolker at gmail.com> wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
>
>    Someone sent me an e-mail recently asking how/whether heterogeneous
> variances can be implemented in lme4.  I can't find it any more, so
> I'm answering here in hopes that they'll see it. The short answer is
> that they can, at least for heterogeneity among categorical groups
> (similar to what varIdent does in the nlme package), although it's a
> bit of a nuisance for categorical variables with lots of levels, as
> the variance terms all need to be specified separately.
>
>    The trick is to set up a dummy variable for each level: there is a
> helper function dummy() for this purpose, and example("dummy") gives
> an example showing how to estimate heterogeneous residual variances
> for females vs. males in the standard sleep study example.
>
>   Ben Bolker
>
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.11 (GNU/Linux)
>
> iQEcBAEBAgAGBQJVT73oAAoJEOCV5YRblxUHlzQIAMuoYv9K06NGG+3mEcvHcmPW
> Xp7rLyxRfiTtRkmr/UieR52WpFDapxNgMurLT1aESD5TWeSBTXmeev2ltajPM5It
> A5Mz1cNob1Tl67zsBCUrKMVVclRicO/Ymbpab1QJYcXxGAVZ03gkaguPWJpw3aHQ
> kt2GZW6SGidbI17PwZXTyVb69cJ8p/mq459LwMAFNusD4MCiIt1/K2HTL8NnekVn
> G7XPMDBg/JoyHmnqWpJbhzccjxZetNMtK6GyXaamlnyPJaJI0i9j6t/X0jKiAqO0
> k1BdwNkLo03sneHodyhdRqAS5KMzeNSo8FOinDIUCLtVikz5k0IbQhHO6JWChz4=
> =KPfA
> -----END PGP SIGNATURE-----
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>

	[[alternative HTML version deleted]]