Dear all,
I am modelling selection of habitat covariates by grizzly bears.
Typically, we use logistic regression to contrast the environmental
features extracted from GIS databases at locations of monitored animals
(i.e., 1s) with random locations (i.e., 0s) that are meant to represent
the availability of habitats. The resulting beta coefficients provide a
measure of the strength of selection. I believe simple logistic
regression is inappropriate for two reasons:
1. Number of locations for each animal varies; pooling data leads to
muddied inferences as our true sample size is the number of individuals,
but one individual could contribute a large proportion of the use
locations;
2. Habitats are not equally distributed across the study area; a pooled
model of all individuals does not consider habitat heterogeneity and the
resulting influences on the strength of the beta coefficients.
I am using Stata's implementation of conditional fixed-effects logistic
regression in an attempt to control for variation in the number of
locations per individual bear and heterogeneity in the availability of
different habitats across the study area. I clustered data according to
individual bear ID and spatially explicit management units. My work is
specific to bears, but I assume researchers from the human and social
sciences deal with similar issues when modelling cohort data for people.
There are a large number of approaches to dealing with clustered data and
I am hoping that someone will comment on the usefulness of conditional
fixed-effects logistic regression in addressing issues such as I outlined
above. Any and all suggestions are appreciated.
Thanks in advance.
Chris Johnson
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
