I'm looking for a test to statistically compare different groups that have multiple data points for each subject within a group. Without getting into too much detail about my specific measurements, an analogous situation would be the following:

If you were looking at the difference in freckle size between 3 groups, (say, redheads, blonds and brunettes). The measurement is the size of the freckle. Each individual has a different number of freckles, and so there will be a varying number of data points for each individual subject. The problem with taking the mean size value for each subject is that these size values are not normally distributed (say there are a few very big freckles and many small freckles for each individual, but it can vary between individuals).

I'm assuming that the size of each individual freckle is independent (although some people tend to have more large freckles and others small, but it's not related to an experimental manipulation), so I don't think a repeated measures ANOVA is correct? Or might it be?

I don't have much experience with statistics at all, so any help or suggestions would be greatly appreciated!


1 Answer 1


I agree that rmANOVA would not be correct: what is repeated here? If each consequtively measured freckle was dependent on the one before (within the same individual), you could have used this approach. I'd suggest using individual as a random factor in your design; and a subsample-size unbalanced design is ok.

  • $\begingroup$ Thanks for your reply, katya. So using individual as a random factor in my design, would you suggest just using a one-way ANOVA then? What I'm stuck on then is how to enter my data, since my data is not normally distributed (so I can't really use means), and there are multiple measurements for each subject. By subsample-size unbalanced design, do you mean just enter the values for each individual data point (freckle size)? I'm afraid I'm fairly new to data analysis in general, sorry for all the questions! Thanks again. $\endgroup$
    – Sarah
    Commented May 9, 2015 at 3:16
  • $\begingroup$ how you structure the dataset would depend on the program you use. As an example: size/person/haircolor (say 1.2mm/subject1/blonde - sorry can't get formatting in comments) with multiple entries under the same person (random factor), yes. It is important, however, that your data comply with anova assumptions at the level of your treatment, otherwise, a different method must be selected. The exact definition of the model would depend on what you use, here are some examples conjugateprior.org/2013/01/formulae-in-r-anova $\endgroup$
    – katya
    Commented May 9, 2015 at 3:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.