On Thu, 13 Jan 2000 10:32:09 +0100, christophe tourenq
<[EMAIL PROTECTED]> wrote:
> I am a biologist (definitely not a statistician) in France with very
> limited access to statistical resources and I wonder if someone can give
> me some advice. I have been conducting weekly bird counts in wetland
> areas for a period of 1 year. The intention is to compare numbers of
> birds (of each species) among different wetland types. I was intending
> on using a generalized linear model framework where counts are treated
> as a Poisson random variable. The problem is this:
< snip, about lack of independence >
You have counts by week; compute some averages by month or by season,
where you select the interval so that the numbers are pretty similar -
that way, the average does not throw away much information.
The average is making the best use of the information that is shared
across weeks, and that you might pursue in a much more *difficult*
sort of modeling (such as, the repeated measures that were suggested).
That is, the average *is* the important effect from the repeated
measures, sometimes.
If you are looking at just a few seasonal averages, then (as you say)
there is not so much correlation.
> ... Someone else suggested to me that there
> might be a way to use a bootstrapping approach to resample the data and
> eliminate some of the independence problem.
- "eliminate dependence"? That sounds glib, and wrong. I think that
correlational structure is a really big Unknown, which prevents an
easy use of bootstrap.
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
