Dear colleagues,
A year ago I made an inquiry regarding the use of multivariate models
based on Poisson and Log-linear distributions when one needs to analyze
the effects of various parameters in the counts of bat ectoparasites.
The fact is that some approaches like Log-linear models, and other GLMs
based on discrete theoretical distributions do not include in the
analysis those cells including cero counts in the contingency table.
Thus, any results from these kinds of models will lack information
regarding the meaning of absent ectoparasites in some individuals, which
from an ecological point of view are of great value for understanding
the dynamics involved in parasitism. Although some of you kindly
answered my previous post on this subject, some of the suggestions,
claiming that Log-linear models and other multivariate approaches for
the analysis of frequency counts on complex contingency tables do
include cells with the cero counts, were wrong. You just need to run
the same analysis in a matrix containing cells with cero counts, and the
same matrix with those cells erased, to realize that Log-linear models
do not account for all observation in the original data matrix. I will
appreciate any suggestions for analysis similar to Log-linear models
whose design will allow me to include information with cero counts. Is
there any available software including applications with the negative
binomial distribution that you could point towards?
Thank you very much.
Pablo Jarrin
Grad. Student
Dept. of Biology
Boston University
|