David Villegas R?os <chirleu <at> gmail.com> writes:
> Hi,
> For a number of individuals, I have measured several behavioral traits in
> the wild. Those traits (e.g. home range) can be estimated on different
> temporal scales, for example daily, weekly or monthly. I want to estimate
> repeatability of those traits, assuming that the daily/weekly/monthly
> measurements represent replicates. I have 3 months (90 days) of data for
> each trait. Two questions:
>
> 1) How can assess if there is temporal autocorrelation in my model? I guess
> that if I consider daily measurements as replicates (90 replicates), I will
> have some autocorrelation, but if I use just monthly measurements (3
> replicates) maybe I avoid it.
>
> 2) How can account for temporal autocorrelation in MCMCglmm?
>
> Sorry for this pretty basic questions but I haven't found an answer so
far.
You'll probably be better off asking this question at r-sig-mixed-models
(at) r-project.org.
As a first pass, you might be able take the residuals from your fit and use
acf() to compute the autocorrelation function. Actually, though, you'll
probably be better off fitting a 'null' lme() model (fixed=resid~1,
random=~1|individual) and then using the ACF() method (not the same
thing as acf()) on the resulting model fit.
Ben Bolker