Message-ID: <54206EB8.4090903@gmail.com>
Date: 2014-09-22T18:47:20Z
From: Ben Bolker
Subject: Within group estimate of autocorrelation
In-Reply-To: <89A06A313CF21543AC6BE0F5AACCE68734F46D8D@C18760.chelt.local>
On 14-09-22 02:39 PM, PATRICK, Samantha wrote:
> Hi
> I have a model which fits an corAR1 autocorrelation structure and
> the code tells the model that time is nested within individual:
> fit0<-lme(Respone~1,
> random=~1|indiv,
> correlation=corAR1(form=~time|indiv),
> data=Data2, method = "REML")
>>From this I get a Phi single estimate of the autocorrelation.
However I want to have an estimate of the autocorrelation for each
individual. I have checked but can not find any code to extract
this value.
The model assumes that autocorrelation structure is homogeneous
across individuals.
> So my first question is: Is it possible to extract an estimate of
> autocorrelation per individual or does the model not save/calculate
> this?
So, the answer is: no/correct.
> And second, if it isn't possible, I wondered if there is any way to
use the weights function to group individuals.
> If I add the code: weights=varIdent(form=~1|group), I can fit
> multiple residual variance terms so I wondered if this could be used
> to estimate a Phi value per group, by somehow structuring the model
> so it fits an autocorrelation for each residual variance group?
> The solution does not need to be lme necessarily - I?m open to any
suggestions!
Perhaps just use acf() along with your favourite by-group tool
in R (plyr, aggregate, dplyr, data.table, for loop ...)
to compute/extract the first-order autocorrelation parameter for each
individual? Or you could use
gls(Response~1,correlation=corAR1(form~time),...)
for each individual and similarly return the autocorrelation parameter
per individual.
I hope you have a reasonably large number of observations per
individual ...