Hello,
I have 3 one-arm pretest-posttest quasi-experiments.
Is it possible to do meta-analysis to their findings?
If so, how can I analyse them using Revman?
For me, I considered pretest data as control group's data, and posttest data as experimental group's data. Is that correct?
Also, I have other 3 two-arm RCTs. These studies have the same outcome in the former three studies. my question is do you recommend me to make two meta-analyses: the first to meta-analyse the former study, and the second to meta-analyse the latter studies? Or Conduct one meta-analysis using data of pre&posttest for the experimental group in all 6 studies.
Thanks in advance.
You can do a meta-analysis on anything - the question is more whether or not you should. I think probably not. Firstly, single arm pre-test post-test desings are more properly classified as 'pre-experimental' rather than quasi-experimental according to Campbell and Stanley's original formulation. Statistically combining them to get and average treatment effect risks masking the inherent bias in the designs while achieving very little. Nothing to stop you commenting on and discussing the apparent treatment effects but the similarity or difference of the change is likely to be an important part of the interpretation of what might be going on.... but the proble is really i) things change over time and ii) things change in response to lots of inputs and in some cases ANT (Hawthorne effect).
There may be some very well controlled lab studies where the pre-test post test is fine because all other sources of change are controlled for - in which case just treat it like any other controlled trial in a meta analysis...
Reginald Quansah
makes a good point about considering the effect for trial design within the set of results - bu given the low numbers of studies and the fact that 3/5 are uncontrolled before and after studies means it is probably better to do this qualitatively and not include a statistical meta-analysis . The “garbage in garbage out“ rule applies here as much as anywhere...
I agree with Peter Peter Donald Griffiths .
Down below may be a fuller discussion on this issue:
Cuijpers, P.; Weitz, E.; Cristea, I. A.; Twisk, J. (2017): Pre-post effect sizes should be avoided in meta-analyses. In Epidemiology and psychiatric sciences 26 (4), pp. 364–368. DOI: 10.1017/S2045796016000809.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis