[R-sig-ME] mixed modelling with random effects with different correlation structures? GLMM/GEE mix

Ben Bolker bbolker at gmail.com
Tue Oct 1 16:28:07 CEST 2013


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 13-10-01 10:04 AM, R.Klink, van wrote:
> Dear Prof. Bolker
> 
> I write you on behalf of my colleagues working in the same
> project. We have a complicated study design, and as far as I have
> been able to tell, there are no published statistical packages that
> can deal with this design correctly.
> 
> We have a blocked design, of 3 blocks, with each 5 plots with each
> a different treatment. Each of these 15 plots has been sampled for 
> plant, insect and bird species richness during 4 consecutive years.
>  We would like to analyze all this in a mixed model where we
> include plot nested in block, and also the year effect as random
> factors. But this seems not to be possible, since the nested design
> has an "exchangeable correlation structure" (in GEE terms), or
> compound symmetry, whereas the time series is auto-regressive.
> Hence, we would like to construct a model that combines different
> correlation structures for the different random factors. My
> question to you is of course whether something like this is
> currently possible, if yes how? and if no, what would your advise
> be to analyze these data.
> 
> Feel free to post this question on one of the GLMM-fora
> 
> We hope you can give us some advice.
> 
> On behalf of me and my colleagues`

  Short answer: not that I know of (maybe AS-REML?)  You will probably
have to 'roll your own' with WinBUGS/JAGS, or AD Model Builder.  I
would also be quite concerned about fitting a model of this complexity
to a data set of this size -- it's going to be hard to estimate a year
effect with n=4, let alone an autoregressive correlation term.  I
would be inclined to fit a simpler model: year as a fixed effect --
and then test/examine the residuals for autocorrelation within plots.

  Hopefully others will come forward with different ideas.

 [cc'ing to r-sig-mixed-models]
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJSStv3AAoJEOCV5YRblxUHzhAIALS/9jKqgjuJHFmBG3uibMyB
q+4W44WXR6tOSQYy0CSuLZHNdy+FfM2nXF7VDmyrwEZOfsx6s3S4jTEbxUauFxnG
oUacy1ZsXJyp59FWN8m0LKTUutsYPebZRRZ9I3S6WXLRHSG3uv1iL8/qRlEvG2ZA
oh/NK+1GP0UXgIYIdBgBwUmjc7hqCA4PGHxW7iDZ69dqpbZ306H8r2OuonzMvC8A
uVzjiAHXF789njVAiNN154wCOLHIHn/6IfcHLx04wRLEBMqGMMIsBs3UfhDBMbZL
NXIAq6ce5rvhpBWhQH8n94IwTTkc74WMmIZhaefkPbil97IzAEyeIgYYiP4DJV4=
=N+gU
-----END PGP SIGNATURE-----



More information about the R-sig-mixed-models mailing list