Tools for Those Who Summarize the Evidence Base
I am wondering if anyone can point me to literature regarding approaches for estimating pre-post correlations for pre-post studies in which this statistic is not reported. It is my understanding that effect size pre-post correlations may have a considerable influence on effect size calculation for pre-post studies, and I would like to make sure that my estimates of these correlations are convincing.
Might any of the following data, which are available to me, be useful for estimating pre-post correlations:
Pre- and post-means for treatment and control groups
Pre- and post-SD for treatment and control groups
N for treatment and control
Effect size direction
I have also considered using previously published estimates of test-retest reliability for the outcome instrument in question, although I do not have estimates of test-retest reliability in the specific types of individuals included in the studies in my meta-analysis or at the same timing as the assessments in my meta-analysis.
A third approach I have considered involves a combination of the aforementioned use of previously published estimates of test-retest reliability in conjunction with sensitivity analyses across a range of plausible test-retest correlations.
Thank you for your consideration.
I have a similar problem of not knowing of how to estimate the pre posttest score correlation, which is needed to calculate the pooled standard deviation for gain scores when only sd of gain score is reported.
(Lipsey & Wilson. Practical Meta-Analysis)
It seems to be a common problem, because many papers report only gain scores and their sd but not the sd of just the post score or the pre post test score correlation. The more I am surprised how little answers I find in my google search.
Would be great if you could tell me how you handeled the problem.
Any advice is appreaciated, also from the community.
Thanks a lot
I used previously published estimates of the pre-post correlations from studies with other populations. I also conducted sensitivity analyses in which I substituted a wide range of alternate pre-post correlations.
Hope this helps.