[R-sig-ME] year and site

[Will Morris]

Gelman suggests that specifying a weakly informative half-Cauchy
distribution for the group level standard deviation can overcome the
difficulty in estimation experienced when there are few groups and
your using a uniform prior (and I presume using Maximum likelihood
too). See Gelman & Hill, 2007.

Will Morris
Masters of Philosophy candidate
Vesk Plant Ecology Lab
The School of Botany
The University of Melbourne
Australia
Phone: +61 3 8344 0120
http://www.botany.unimelb.edu.au/vesk/

On 26/09/2009, at 23:29, Andrew J Tyre <atyre2 at unlnotes.unl.edu> wrote:

> Hi all,
>
> in ecology year a study is done in, and the site it is done at, are
> obvious random effects. However, for most student projects, and
> indeed,
> even larger projects, having more than 2-3 years of data is unusual.
> Having more than 2-3 sites is more common, but even then the number is
> very limited. This means that estimating a random effect of year
> crossed
> with site is difficult. The solution that I've tried a couple times
> is to
> create a "siteyear" random effect, where each year at each site is a
> different level. This gives many more levels with which to estimate
> variances, and seems to work well. The only downside that I've been
> able
> to come up with is that if there are "site" effects that are
> consistent
> through time, or year effects consistent across space, then this
> approach
> misses separating them, at a minimum. What I'm concerned about is
> that it
> might lead to bias as well - I don't think it will, but I'd rather be
> certain!
>
> I have used the "fallback" of fitting year and/or site as a fixed
> effect,
> but I'm hearing some concern from reviewers who've been sold on the
> random
> effects idea convincingly enough to reject a paper that doesn't
> treat year
> or site as random without recognizing that it doesn't help when the
> number
> of levels is low. Also, if there is a continuous covariate whose
> effect
> varies by siteyear, the fixed effect interaction term gets ugly, fast.
>
> Any discussion or insights appreciated!
>
> thanks,
>
>
>
> Drew Tyre
>
> School of Natural Resources
> University of Nebraska-Lincoln
> 416 Hardin Hall, East Campus
> 3310 Holdrege Street
> Lincoln, NE 68583-0974
>
> phone: +1 402 472 4054
> fax: +1 402 472 2946
> email: atyre2 at unl.edu
> http://snr.unl.edu/tyre
> http://aminpractice.blogspot.com
>    [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models