0
$\begingroup$

I have done an experiment where I have two groups of patients that received different treatments (Group1 received treatment 1, Group 2 received treatment 2). I measured a value that should be influenced by both treatments before the treatment and after the treatment. Now, I want to test the hypothesis that treatment 1 is more effective than treatment 2. My first idea was to compute the diffence of the value before and after the treatment for each participant and then do a t-test to see wheather the means of the two groups are significantly different. However, my advisor said that a better solution would be to calculate the "interaction term of a repeated measures ANOVA" without being able to explain to me what this means. So, my question is: What does he really mean by "interaction term of a repeated measures ANOVA" and in why is this better than doing a t-test.

Thanks for your help!

$\endgroup$
2
  • $\begingroup$ I would consider using a multilevel model instead. $\endgroup$
    – Galen
    Commented Nov 30, 2021 at 15:18
  • $\begingroup$ Unless I have misunderstood your design here, a repeated measures ANOVA would not make sense due to your design only measuring the dependent variable twice per group. This page has some useful explanations of RM-ANOVA statistics.laerd.com/statistical-guides/… $\endgroup$
    – jros
    Commented Nov 30, 2021 at 15:19

1 Answer 1

1
$\begingroup$

I believe your advisor is suggesting to model both the baseline measurement and the post-treatment measurement jointly. This would allow you to examine the difference in means between pre- and post-baseline for treatment 1 and compare this to the difference in means between pre- and post-baseline for treatment 2. Unless you are conditioning on additional factors in the ANOVA model this should produce the same results as the t-test you described. With this approach you are investigating the rate of change from baseline to post-baseline in each treatment group.

Yet another approach would be to model only the post-baseline measurement in each treatment group. The comparison would be the difference in post-treatment means between the two treatment groups. This is typically the approach used in clinical drug development. You could use the baseline value as a covariate in the ANOVA. This approach is not comparing the rate of change between the two groups. It is investigating the post-baseline treatment effect while holding baseline constant.

If you do not have the luxury of randomization then you will need to make your "no unmeasured confounders" assumption clear when discussing the results. You could use propensity scores or inverse-probability-weights to balance prognostic factors related to both the treatment selection and the outcome.

$\endgroup$

Your Answer

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