Statistics problem Krustal Wallis

Question

[chantalrenauminguez]

Hi! I am trying to perform an statistical analysis with my data but I don't get any statistic significative result and I am wondering if it's because I haven't got enough samples...

I introduce you to the problem...

I have some cells without treatment (controls) and with treatment 1 and treatment 2, so 3 groups. For each group I measure 4 variables (percentage of cells that correspond to each variable). So for example for the control i have variable 1(viable) = 20, variable 2(early apoptotic) = 40, variable 3 (late apoptotic)= 30, and variable 4 (necrotic) =10... until we get the 100%. (Just for you to understand a bit the kind of data).

And of course we measured 3 times each group. So I have 3 data for each variable in each group.

I think (and checked) that my data is not following a normal distribution so I tried a Krustal-Wallis and I cannot see any significative difference between the control and the treatment 1 or the control and the treatment 2 for any of these 4 variables. So I wonder if I need more samples for this test. Do you think I am performing the right test? How would you solve this?

I appreciate any help with this matter.

Thank you!!

Answer 1

[i.sudbery]

As KS is a non-parametric test I doubt you have any power with only n=3. These sorts of non-parametric tests rely on ranks and there simply arn't enough different possible rankings with n=3 to make any particularly ranking unusual compared to chance. I'd normally want at least n=10 before I was completely comfortable using non-parametric tests, although there maybe situations where you can get away with between 5 and 10.