Dear colleagues,
We are currenlty crunching some datasets on the parasitic load found in the fur
of several sister species of bats. Our datamatrices consist of classical counts
or observations in multilevel contingency tables (i. e. sort of...the number of
bat-flies found in each indivudal of each species in a particular habitat).
In order to find patterns, and discern causes and consequences, we are
exploring with statistical approaches based on chi-square distributions such as
Log-linear models, and correspondence analyses. However, as many of you should
know, this kind of tests "obviate" null observations (i. e. bats without bat-
flies), which is a problem when trying to interpret the dynamics involved in
parasitism.
We are currently working with canonical correspondence analysis as well, adding
an infinitesimally small value to observations with cero parasites, so they can
be included in the analyses. We assume this could be the solution to our
problem.
Many published studies of this kind use analyses based on normal distributions
such as MANOVAS and MANCOVAS, which take into account bats with cero parasites
(null observations in the case of a contingency table). However the reliability
of the parametric statistics should be affected due to the statistical
distribution and nature of the discrete variables.
Our question to you is: żIs there any kind of correction available for
statistical tests based on chi-square distributions that will allow us to
include observations of bats free of parasites (null observations)?
We appreciate your help and advise.
Sebastian Tello
Pablo Jarrin
Escuela de Ciencias Biologicas
Pontificia Universidad Catolica del Ecuador
[log in to unmask]
|