[R-sig-ME] Specifying correlation structure

[Gavin Simpson]

On Tue, 2012-05-15 at 06:27 -0400, Gang Chen wrote:
> I have some time series data (with 100 time points) collected from one
> subject. At each time point three measures were taken under three
> conditions, and there is an explanatory variable for each condition.

The temporal correlation is, in this sense, *nested* within the three
different conditions. You were close with time:condition, but what I
think you need is

corARMA(c(0.02, 0.03), form = ~ time | condition)

HTH

G

> Below is just some fake data to demonstrate the data structure:
> 
> # 100 time points per condition
> tp <- 100                             # time points
> trials <- seq(1, tp, 1)          # time counter
> condition <- c('a', 'b', 'c')   # three conditions
> 
> # fake data
> set.seed(5)
> Dat <- data.frame(time=rep(trials, 3), cond=rep(condition, each=tp),
> res=rnorm(3*tp))
> 
> reg1 <- c(rnorm(tp), rep(0, 2*tp))            # explanatory variable
> for condition 1
> reg2 <- c(rep(0, tp), rnorm(tp), rep(0, tp))   # explanatory variable
> for condition 2
> reg3 <- c(rep(0, 2*tp), rnorm(tp))             # explanatory variable
> for condition 3
> 
> First I thought that I'd start with gls in nlme package with an ARMA
> model for the correlation structure:
> 
> (fm <- gls(res ~ 1+reg1+reg2+reg3, correlation=corARMA(c(0.02, 0.03),
> form=~time, p=1,q=1), data=Dat))
> 
> The above model does not work because of the following error:
> 
> Error in Initialize.corARMA(X[[1L]], ...) :
>   Covariate must have unique values within groups for corARMA objects
> 
> Moreover, I would like to account for the fact that the ARMA structure
> should be similar or the same across the three conditions. My
> questions are:
> 
> 1) how to impose a same ARMA structure across the three conditions?
> 2) alternatively I'd like to have something like:
> 
> (fm <- gls(res ~ 1+reg1+reg2+reg3, correlation=corARMA(c(0.02, 0.03),
> form=~time:condition, p=1,q=1), data=Dat))
> 
> but this would not work either, and seems to crash R!
> 
> 3) maybe construct a multivariate gls model, but how to do that? what package?
> 
> Thanks,
> Gang
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> 

-- 
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
 Dr. 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
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%