Hello
We have a question concerning the resampling of data and the independence
of tests performed on these resampled data.
We use Splus2000 on windows 2000 professional.
We have done an experiment testing the influence of parasitism on
competitive ability of larvae of mosquito. In the experimental plan, we had
2 larvae per vial; either 2 infected larvae, 2 control larvae or 1 infected
and 1 control larva.
Our goal is now to analyse the influence of the infection status of the
larvae and of the infection status of the competitor on various
life-history traits.
We can not include the two individuals of each vial in the analysis because
they are not independent. We thus randomly choose one individual from each
vial to include in the analysis.
I have written a piece of Splus code that allows me to repeat the
sequence ?randomly choosing one individual per vial ? + ? doing an ANOVA on
the selected data.
I can thus have a distribution of F values and follow the evolution of
the mean F value for each factor. However, I don?t know which criterium to
use to decide if the mean F value converge and if a factor has a
significant effect.
I read the paragraph ?Combining probabilities from tests of significance?
in Sokal and Rohlf (third edition. pg 794) but it seems to me that the
method described here (-2 Sum ln (Pvalue)? to Chi? with 2k df (k=number of
tests) ) is not the right method because the half of the individuals
included in one test are also included in the subsequent one.
I think that a conservative decision criterium would be to consider a
factor as having a significant effect if more than 95% of the tests
performed are significant.
Does somebody has a better idea?
Stéphanie Bédhomme
|