statistical help

To: statsoc-l@mhc.mtholyoke.edu, lrouse@asu.edu
Subject: statistical help
From: Julian Goulet <jgoulet@mola.gso.uri.edu>
Date: Tue, 13 May 1997 11:36:31 -0400 (EDT)

Leigh,

  Lets see.  You have a sampling unit of spring_area-survey, with one survey
per season per year.  You have four dependant variables which are not
mutually independant (the number of breeding birds and the number of
non-breeding birds is probably correlated).  You have four independant
variables, which also are not mutually independant (the vegetation volume is
probably correlated with the riparian patch size, etc.).

  Your dependant measure,  'average number of breeding birds' is averaged
over what?  Did you do several surveys per season and average that?  Did you
count birds by species and average across species?  Do you mean 'total
number of breeding birds'?

  No matter.  Four separate regression models is not a good idea, because
the dependant variables are not mutually independant.  Thus a 'fancy
regression model' is probably what you want.  Consult again with your
statistical consultant to clearly define your objective.  Bear in mind that
he/she doesn't speak your language as well as you do not speak his/hers. 
Stepwise regression works for univariate (not multi-variate, such as your
four possibly related dependant variables).  The solution may be in a
sequence of 'fancy regression models', each formulated slightly differently
as you both gain understanding of the real relations among your observations
and the statistical relations among your data.

  Good Luck

  Julien Goulet
  jgoulet@mola.gso.uri.edu