Hi everyone,
I'm using RevManager to conduct a meta-analysis from studies with continuous data.
For some of such studies, the values of mean and SD reported for the authors is 0. When I input the data into RevMan software, the software is not able to estimate the Std Mean Difference .
For these cases, which one is the better approach to conduct the meta-analysis?
I believe that RevMan is not happy with the standard deviations being zero. It sort of violates the assumptions of parametric tests (i.e., normal distribution and equal variance). If you substitute a small number for the zero standard deviations, RevMan should do the calculation and not really affect the effect size or variance. See the attached Excel file for seeing the effect on inserting a small non-zero number for the SD.
You should also check whether there is a valid reason for SD being zero.
Thank you for your answers. If I decided to follow the strategy using small values for SD, Is there any reference that I could cite?
Thanks in advance
I would recommend against imputing small values such as 0.0001 is this would give this study an unproportionally high weight, at least when using the inverse variance method, which is most commonly used in RevMan . Also see the following study which adresses the issue of missing data: Article Dealing with missing standard deviation and mean values in m...
Cheers
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis