How to compare 2 groups where 1 has repeated measures


I have three groups: healthy controls, pre-surgery, and post-surgery. The pre and post-surgery are the same patients who participated in the study before and 10 months after their operations. The healthy controls were matched to the patient group based on anthropometrics. My issue is what test to select considering 1 of the 3 testable groups are independent while the other 2 are dependent. I have considered the following options:

1) Repeated measures ANOVA: Even though the healthy controls were matched, I still am not confident that they are matched well enough to be considered the 'same' as the injured groups.

2) One-way ANOVA with t-tests as post-hocs. Is it appropriate to use independent t-tests fro healthy vs pre op and healthy vs post-op but a paired samples t-test for pre vs post-op? Would I have to correct my alpha from 0.05 to (.05/3) 0.0167 because of the three tests even though I'm doing 2 independent and 1 paired?

3) Treating each comparison as a separate research question: Q1) Independent t-test between healthy and pre-op. Q2) Paired samples t-test between pre and post-op. Q3) Independent t-tests between post-op and healthy. I still have the issue of what to correct my alpha to because 0.0167 appears to be much too conservative when visualizing the data in the figures.

Any advice on this matter will be greatly appreciated!



Well-Known Member
Hi kbs
This is an interesting one. The repeated measures anova is doubtful because the between subject variability is different between the groups, as you have pointed out. The one way anova lacks power because it ignores the common subjects.
I think 3 is the way to go, but there is no reason why you can't use paired tests for all three comparisons. Even if the marching isn't perfect, it all helps to remove uncertainty. This is effectively an honest version of 1.
There isn't much you can do about the 0.0167, except to say that Bonferroni is known to be conservative, and use 0.02 instead. It depends on who you have to convince.
Cheers, kat
Hi Kat,

Thanks for your input! It's nice to hear that others see the reason behind the approach and thank you for suggesting 3 dependent tests.

In regards to the correction, is it reasonable to correct by 2 instead of three since each group each only being tested twice throughout the 3 tests?




Less is more. Stay pure. Stay poor.
I believe what is being proposed is comparing controls to pre, and also controls to posts. If so, that seems reasonable and you would only need to correct for 2 tests.

The one thing that is a little unsavoury is that it appears you already know the measures for your outcome and you are trying to maintain significance. This is why a priori planning is so important and adherence to protocols. Your personal biases are potentially coming into play.