I have conducted independent samples t-tests to compare the pre-test metacognitive awareness scores between the control and experimental groups, and the p-value was 0.000, showing a significant difference in pre-test scores between the two groups with control group having a higher score. Now, I am wondering whether I can use a t-test to compare the post-intervention scores or if I should use an analysis of covariance (ANCOVA).
If subjects were randomly allocated to the experimental and control groups, then clinical trial folks would say there is no point in comparing them on baseline values. Do a search on to find relevant articles. But see also this blog post by Dean Eckles:
- https://statmodeling.stat.columbia.edu/2021/08/22/does-the-table-1-fallacy-apply-if-it-is-table-s1-instead/
Here is another discussion that may be of interest. See especially Frank Harrell's first post.
- https://discourse.datamethods.org/t/should-we-ignore-covariate-imbalance-and-stop-presenting-a-stratified-table-one-for-randomized-trials/547
As a caveat to Bruce Weaver 's answer, it is worth asked these "clinical trial folks" what they mean by random allocation and if there might be any bias in things like dropout. By any chance were the data clustered? And if so, how was this taken into account for your initial t-test? And how big is the sample?
I've only perused it quickly, but here is another article that looks interesting.
Article How often does random assignment fail? Estimates and recommendations
Here (https://www.amazon.com/Selection-Covariate-Imbalances-Randomized-Statistics/dp/0470863625) is a book on what can go wrong with random allocation.
- Badges
- Science topic
- Similar topics
- Correction