1
$\begingroup$

I'm planning to study a variable from a group of patients before and after a particular operation. I do not want to, or plan to, compare measurements across the "before" and "after" data (that is, I wasn't planning on using methods for serial data - i.e. mixed effects. GLM). I only want to compare data at the preoperative time point and then also compare data at the postoperative time point.

I was planning on using one-way ANOVA for the preoperative data and then a separate one-way ANOVA for the postoperative data (and potentially using nonparametric methods if data was not normally distributed). However, a colleague of mine is suggesting that I use longitudinal methods instead (even though I don't want to compare across time) to compare all the data.

It never occurred to me to do that since I'm not interested in the "before" and "after" comparisons, but is that okay?

Does putting that all into one model give me an advantage compared to doing do separate ANOVA? If so, how does that change the power?

Improve this question
$\endgroup$
3
  • $\begingroup$ I'm not sure that I understand what you are doing. Do you have 2 or more treatments on each group (pre and post-op)? If you have the same people in both the pre and post-op groups, I would do a repeated measures analysis. If you include a time by treatment interaction, you could test whether the treatment effect pre-op is different from the treatment effect post-op, as well as looking at whether the operation had an effect. If you could clarify what your treatments are, I might be able to give a better answer. $\endgroup$
    – Placidia
    Commented Nov 20, 2015 at 23:17
  • $\begingroup$ I don't want to study any time effects in this design. I have three groups of people that are going to be studied "before" an intervention and "after" an intervention. We are only interested in hormonal measurements before and after the intervention. We don't care about the across time comparisons - i.e. before vs. after (they are meaningless). We just want to compare these three groups "before" and the same groups/patients "after". Would an ANOVA on the "before" data and a separate ANOVA on the "after" data be inferior to all that in a mixed effect model/GLM? That's what I'm struggling with. $\endgroup$
    – Vance L Albaugh
    Commented Nov 20, 2015 at 23:51
  • $\begingroup$ @Placidia - sorry forgot to tag you in last post... hope that helps for clarification $\endgroup$
    – Vance L Albaugh
    Commented Nov 20, 2015 at 23:58

1 Answer 1

Reset to default
2
$\begingroup$

You probably won't be led to a false conclusion by doing two anova's, especially if your sample size is large. But I can't give it two thumbs up (maybe one thumb).

The difficulty here is that your sample size is the number of patients. If you analyse the pre time separately from the post time, and report p-values for each, you are analysing the data as if you had done two independent studies with a total of $2n$ patients, when you actually have $n$ patients seen twice. So let's say that Mr Smith from group 1 is an outlier with weird hormones. Then Mr Smith's weird hormones will show up again post-op, only your proposed analysis does not take that possibility into account. It assumes that Mr Smith post-op is "not the man he was".

Data should be analysed in a way that reflects the randomization and the study design (I'm a frequentist). That said, the practical consequences will probably not be large. But you are setting yourself up for negative comments from whatever snarky methodologist reviews your paper if you treat the data as if they came from two different studies when it actually was longitudinal. And if your study is small, like 3 patients in each group, you really need to go longitudinal.

Improve this answer
$\endgroup$

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.