Hello all,
I am now doing a meta-analysis of effectiveness of a intervention. And I'd like to evaluate the carry-over effects of the intervention (the lasting effects after completion of the intervention).
As the post-treatment performances are varied in each group, using the follow-up data alone may not be a good choice.
I was planning to pool the change scores (baseline to follow-up or post-treatment to follow-up). However, a large amount of studies did not report the change data (especially the SDs).
A way to get it is to assume a correlation coefficient (may be 0.8) between baseline (or post) and follow-up data, and then calculate the SDs. But we know the time to follow-up varies from study to study.
Does anyone have experience in selecting the correlation coefficient? or other ways to evaluate the carry-over effects?
Thanks and best regards,
Jack Zhang
In agriculture the carry over effect vanishes after 9 months by 90% (thumb rule). You can compare the one-year follow up after a study with the endpoint of a study for changes of a mediated behaviour pattern. I don't understand why You want to calculate a correlation between the comparison of two points.
most meta-analyses (at least those in my field) assume a .5 pre-post correlation, which effectively means the raw score effect size is equals to the change score effect size.
You can try to email the authors for the change score descriptive stats, if you can't get those stats, i think you should be able to safely use the .5 pre-post correlation.
Or alternatively, if you have a few studies which do provide change score metrics, you can compute their pre-post correlation for these studies and compute a mean of these pre-post correlations. You can then use this mean pre-post correlation value for the studies without change score metrics.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis