Dependent variable from ratio counts. That's discrete? If yes, which test for low Ns?

[solar]

Hi all,
I have a rather specific question, I somehow don't seem to be able to find the answer to.
So I work on biological tissue, where I count the number of cells that have been 'touched' by a particular axon. Meaning that my counts are ratios-percentages: dependent variable: 'percentages of touched cells' (= number of touched cells/total number of counted cells).
For me that dependent variable is discrete, because it cannot exceed 100%. Is that correct?
If yes, do I understand correctly that ANOVA is not appropriate, because it assumes the dependent variable to be continuous?
My 2 independent variables are categorical and have 2 and 7 levels respectively. I also have low and unequal N's (4-10 biological samples per group and I cannot increase these numbers).
I also have slight normality problems, but that I could see to via transformation if necessary, so that is not my primary concern right now.
I am looking for appropriate tests suggestions. I have come up with non-para Kruskal Wallis. Is that appropriate? Do you know of any other tests (even not very 'traditional' ones) that I could look into?

I am grateful for any suggestions. Thanks in advance!!!

 

[GretaGarbo]

Re: Dependent variable from ratio counts. That's discrete? If yes, which test for low

Maybe you could try to use a Poisson distribution as the dependent variable (or a negative binomial) for the counts.

An other possibility is a binomial distribution in a logit or probit model since:

“I count the number of cells that have been 'touched' by a particular axon ...= number of touched cells/total number of counted cells ”

This can be estimated in a generalized linear model (glm).

 

[solar]

Re: Dependent variable from ratio counts. That's discrete? If yes, which test for low

Thanks for the reply! I will try to read myself into Poisson distribution then