Comparing means with Poisson distributed data?

#1
Hi everyone!

How can I perform ANOVA, so to compare means, when the data of my groups are clearly Poisson distributed (towards 0 values)?

Thanks a lot!
Cheers
 
#3
I'll explain a little bit more of my actual data:

I have a pool of 15 values for each group (7 groups: 1 control + 6 treatment) that are the number of aborted eggs produced by a certain aquatic animal in response to different water samples.

In control replicates I have, for example:
animal 1) 0
animal 2) 0
animal 3) 0
animal 4) 0
animal 5) 0
animal 6) 1
animal 7) 2
animal 8) 0
animal 9) 0
animal 10) 1
animal 11) 2
animal 12) 3
animal 13) 0
animal 14) 0
animal 15) 0

Wherease in treatment groups:

animal 1) 10
animal 2) 2
animal 3) 34
animal 4) 3
animal 5) 6
animal 6) 12
animal 7) 16
animal 8) 18
animal 9) 9
animal 10) 11
animal 11) 21
animal 12) 30
animal 13) 10
animal 14) 14
animal 15) 18

The distribution in the control group is positively skewed (most values are near 0), for this reason the first assumption for ANOVA (normality) is not fulfilled.
Moreover, the homoscedasticity of viarances between groups is not fulfilled either.

How can I tell (with the aid of a statistical tool) that the mean number of aborted eggs in control replicates is different from the one I have in treatments?

Thanks again
 

hlsmith

Not a robit
#4
Side questions, should you look at the number of eggs aborted/total number of eggs (e.g., abort + not aborted)?

Pairwise t-test comparisons for unequal variance, controlling for false discovery rate (multiplicity). Or maybe Poisson regression controlling for group?
 
#5
Thanks hlsmith for your reply!

I want to compare just the number of aborted eggs between groups.
The ratio aborted/(aborted+not aborted) is an interesting data too, but I've decided to express this as a percentage.
 
#6
What about using Generalized Linear Models with Negative Binomial to compare this data (that are counts variable, therefore are not suitable for ANOVA)?

Anybody know how can I set a reference group (control group in my case) so that I can read my post-hoc output (Wald Chi-square test)?