I'm looking at the effect of a gene mutation on the litter size in mice. As litter size is discrete data, I'm not sure whether Student's t-test is appropriate. Sample size is n=20 litters, and some of the litters had no pups.
Thanks in advance!
60
Naresh is wrong. Though he links a really excellent paper.
Litter size is counted data, as you point out. And it has another horrible property: it cannot be zero. Animals who do not become pregnant are not included in the analysis. Data like this are called truncated.
I would have a look at a truncated Poisson regression with zero as your truncation point, and use robust variance estimation.
That said, I would have a quick peep at the data using our old friend the Wilcoxon Mann-Whitney test, and if that's significant I probably wouldn't bother with the Poisson regression. The Poisson fits the data generation process well, but the WM-W is a pretty good all-rounder. How many animals are we talking about?
Paired t-test or the two-way ANOVA can be used.
Pls check this link
http://www.3rs-reduction.co.uk/assets/applets/Festing_Altman.pdf
60
Naresh is wrong. Though he links a really excellent paper.
Litter size is counted data, as you point out. And it has another horrible property: it cannot be zero. Animals who do not become pregnant are not included in the analysis. Data like this are called truncated.
I would have a look at a truncated Poisson regression with zero as your truncation point, and use robust variance estimation.
That said, I would have a quick peep at the data using our old friend the Wilcoxon Mann-Whitney test, and if that's significant I probably wouldn't bother with the Poisson regression. The Poisson fits the data generation process well, but the WM-W is a pretty good all-rounder. How many animals are we talking about?
Hi Naresh
Ronan is almost right. You do say that some litters have no pups, so zero is a valid value. Probably Poisson regression is the most suitable analysis.