Statistical test for counts

Question

[elb]

Hi guys,

I would like to compare measures that are reported as percentages (with respect to a total) of cellular counts in 7 patients treated versus 7 patients untreated with a stimulating growth factor. I basically would like to compare the two distributions and assess, in a statistically significant way if the two distributions are equal or not. I already performed some test on normality and I verified if my data have equal variance or not and they have often a non normal distribution with 0s and they have unequal variance. Which statistical test should I use to compare the two populations? Thank you in advance.

E.

Answer 1

[Devon Ryan]

I would have been amazed if your data were normally distributed, there should be ceiling and floor effects that prevent that. You essentially have two routes that you can pursue. Firstly, you could perform a Wilcoxon test, which can be thought of as a non-parametric T-test. Alternatively, if you're really truly are interested in differences in distribution, then you could use a Kolmogorov-Smirnov (KS) test. This is literally testing for differences in distribution, so if one group has higher variance than the other then you can pick this up too. I'm not sure how powerful the KS test is with N=7 per group.

In general, I would suggest that tests like the Wilcoxon are preferable whenever possible. In most cases you really want to know if the expected value of the groups are significantly different, not whether they have different variance.

[elb]

Thank you very much Devon! Your answer helped me a lot to clarify some key points. Thank you again.