I am conducting a meta-analysis on plant abundance in restored vs. natural habitats. I am using a response ratio measure of effect size ln(restored)-ln(natural). I want to calculate effect size variance, but in some situations I find that the variance in one of the treatments is 0 because all sample values are equal to the mean (either 100% or 0% cover). Must I omit these trials if I want to include variance in the analysis? I know sometimes authors will omit studies that do not properly report variance, but is seems like there might be a bias introduced if I omit studies because the data itself did not allow for the calculation of effect size variance.
Yes, you can calculate the variance for a study if one of the standard deviations is zero. The variance that you are interested in is that of the natural log of the response ratio. See the attached PowerPoint file for how the calculation is done.
I agree. Because the calculation has SD in the numerator, it's okay if SD = 0
Variance 0 is an absurd case. Variance 0 means, all data are same i.e. there is no variation in data which can never happen in usual situation. Moreover if all data are same then the data do not represent a variable.
Thus I think that some error has been committed.
One should also not use a ratio as an effect size measure when either the numerator or denominator would be equal to zero
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis