0

My biological question is the following: is the variance in behaviour similar in all regions studied, and if not, how does it change?

The variable 'behaviour' is a poisson-distributed count data, the experiment was reproduced in 4 different regions and the glmm I ran on the data gave a significant effect of region.

My first approach was to perform a single Fligner-killen test on the whole data set (fligner.test() in R) and then a series of post-hoc pairwise comparisons among regions with the same test. I used Bonferroni-adjusted signifificance threshold (alpha = 0.05/6 for six pairwise tests = 0.00833) but that looks really low to me, and probably too conservative.

I was also looking a bit over the coefficient of variation, but did not find any sound statistics to do with these.

Is there a better way to answer my question?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
antoinefelden
  • 11
  • 2
  • 3
    If variable is really Poisson-distributed as claimed, testing for a difference in mean is entirely equivalent to testing for a difference in variance. – Glen_b Dec 01 '16 at 04:54
  • I assumed the distribution was Poisson as it is count data that look like a poisson distribution, but I ran this test http://stats.stackexchange.com/questions/1174/how-can-i-test-if-given-samples-are-taken-from-a-poisson-distribution to figure out that it was not. I probably have a too liberal use of the "poisson" here, I should have said "quasipoisson". Does that make another answer possible? – antoinefelden Dec 05 '16 at 00:11
  • 1
    You could perhaps say something more like "it's count data that look's somewhat Poisson-like". I wouldn't worry so much about a formal test unless you really have to have it be Poisson; if it's just intended as a rough description, that should be clear. In any case, you should definitely mention that it's count data. – Glen_b Dec 05 '16 at 01:30

0 Answers0