I am conducting an experimental study. I have 2 groups: The first group contains 3 healthy persons and the other one contains 3 patients (sick people). For each person of the sudy, we collect several cells, from which we extract 3 features. The goal is to show significant differences between the 2 groups. In other way, we have:

enter image description here

x are numerical values

Which test is the more suitable to show significant difference between the patient and the healthy group?

I saw several tests like Hotelling, ANOVA..., but in those tests the mean of each group is computed on all cells of all the people of a group without making a difference between the person within the group. Is it possible to include person variablity in such test? Is 2-way ANOVA a solution to take into consideration the people in the comparison? The two independant variables (factors) are thus the identity of the person and its situation (Healthy or patient)?

  • $\begingroup$ A test is not enough in this case. A linear model is the natural progression, to include the variability of the patients you can include a random effect. $\endgroup$ Commented Feb 6, 2020 at 9:03

1 Answer 1


You could include patient as a random effect in your model to account for correlation between cells from the same patient. Something like:

lmer(group ~ feature1 + (1 | patient), data=data)


aov(group ~ feature1 + Error(patient/feature1), data=data)

Syntax not tested, taken from this blog.


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.