[R-sig-ME] Specifying random effects in lme()

[Gavin Simpson]

On Wed, 2007-05-09 at 16:05 +0100, Mike Dunbar wrote:
> Gav
> 
> I don't know the answer to your question, but it may not be possible.
> If it is possible to specify it, then you may still run into problems
> fitting it. You're trying to fit quite a complex model here with two
> random slopes and an AR term, are you really sure your model needs to
> be that complicated? 

Hi Mike,

That's one of the things we want to test.

The random effects just take up an extra df or eleven. I have plenty to
spare in the model, with many years data. For the extra 22 df used up,
does the model fit significantly better using AIC/BIC and friends?

I can fit the model in lmer very easily, converges nicely and quickly.
But I wanted to have the AR1 term in there to mop up some of the
residual autocorrelation in the error structure. There is some
indication of autocorrelation in the residuals, and testing models
without the AR term and with the AR show that the model with the AR term
is a big improvement as measured by AIC, BIC and the likelihood ratio
test.

There is some evidence that the effects of the two covariates does
differ amongst sites based on analysis of the model output from the
model with only random intercepts. We'd like to test if the more complex
model fits the data better.

Cheers Mike,

G

> 
> Can you fit the lmer model you mention? If so, is there evidence for
> autocorrelation of residuals?
> 
> cheers
> 
> Mike
> 
> 
> 
> 
> >>> Gavin Simpson <gavin.simpson at ucl.ac.uk> 09/05/2007 14:23 >>>
> Dear List,
> 
> I am modelling repeated measures data from 11 sites in the UK using
> lme() from the nlme package. I have the following call for a random
> intercept model with an AR1 error structure:
> 
> mod <- lme(doc ~ temp + log.so4 + log.cl, random = ~ 1 | site,
>            correlation = corAR1(), data = hydro)
> 
> I would like to fit a model that has random effects for log.so4 and
> log.cl within site, so that I am fitting a random slop and intercept
> model --- we want the effect of log.so4 and log.cl to vary between sites
> and to compare the fits of the two models.
> 
> If I am following lmer() in lme4 correctly, I believe I could fit this
> model as:
> 
> hydro.lmer <- lmer(doc ~ temp + log.so4 + log.cl +
>                    (log.so4 | site) + (log.cl | site),
>                    data = hydro)
> 
> but without the AR1 correlation.
> 
> I am struggling to get the syntax correct for random in lme() to add
> random effects to log.so4 and log.cl. I have only managed to get a model
> where one of the coefficients (log.cl) also has a random effect.
> 
> I would be most grateful if someone could explain how to fit the model
> in lme() by showing me the correct form for the random argument.
> 
> Many thanks,
> 
> Gav
> 
> -- 
> %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
>  Gavin Simpson                 [t] +44 (0)20 7679 0522
>  ECRC, UCL Geography,          [f] +44 (0)20 7679 0565
>  Pearson Building,             [e] gavin.simpsonATNOSPAMucl.ac.uk
>  Gower Street, London          [w] http://www.ucl.ac.uk/~ucfagls/ 
>  UK. WC1E 6BT.                 [w] http://www.freshwaters.org.uk 
> %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> 
> 
-- 
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
 Gavin Simpson                 [t] +44 (0)20 7679 0522
 ECRC, UCL Geography,          [f] +44 (0)20 7679 0565
 Pearson Building,             [e] gavin.simpsonATNOSPAMucl.ac.uk
 Gower Street, London          [w] http://www.ucl.ac.uk/~ucfagls/
 UK. WC1E 6BT.                 [w] http://www.freshwaters.org.uk
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%