[R-sig-ME] unreplicated repeated measures

[Ryan Hope]

This would be your model:

m1<-lmer(lncon~Treatment+(1|month),data=dataset)

you did not manipulate the "month" variable thus it should not be in
the fixed effects side of the formula.

Now is treatment a coded variable for multiple factors? If so why are
you not using the actual factors like:

m2<-lmer(lncon~FactorA+FactorB+(1|month/(FactorA+FactorB)),data=dataset)


On Tue, Jun 23, 2009 at 1:29 PM, Christine
Griffiths<Christine.Griffiths at bristol.ac.uk> wrote:
> Dear R users,
>
> Sorry if this question is not applicable to this site. I am having problems
> analysing unreplicated repeated measures. I calculated food web properties
> for three treatments over time (10 months). It is unreplicated in that I
> only have one observation per month per treatment. My problem is that I am
> interested in how a food web property varies between Treatments and over
> time.
>
> Originally I had tried using lmer:
> m4<-lmer(lncon~Treatment*month+(1|month),data=dataset)
> but this provides the following error for which I have not found an
> explanation to on the R site.
> Error in mer_finalize(ans) : Calculated PWRSS for a LMM is negative
>
> I suspect it is because I had treated month as a factor and consequently due
> to the lack of replication and trying to interact these two categorical
> variables it fails. I can overcome this problem by treating month as a
> continuous variable from which I calculated confidence intervals using MCMC
> method. However I am not sure how these are being calculated and if this is
> accurate, given I have no mean per Treatment as such. Is it acceptable to
> use month as a continuous variable?
>
> Alternatively, I tried using repeated measures ANOVA, aov, to model the
> data. I am cautious to use this method as it indicates significant
> differences which are not apparent from the plotted raw data. Given that my
> data lacks replication, I am wary of this method.
>
> I have investigated time series analysis, but I am reluctant to venture down
> this route.
>
> Any reassurance or advice as to the best technique given my data would be
> greatly appreciated.
>
> Many thanks,
> Christine
>
>
> ----------------------
> Christine
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>