I have run a meta-analysis comparing the change score of a continuous outcome between females and males using the Comprehensive Meta-analysis (CMA) Software. I entered the pre and post values, and the N of each sex. The software then gives me the mean difference between the change score of females and males. I wanted to see if the intervention is better for one sex over the other one.
I haven't found a sex difference. The p value is 0.7. Some studies favour females, other studies favour males. The ns of females in the studies are quite small.
The I2 statistic that CMA gives me is 0... is this an error if there seems to be some variation across studies?
Thanks for your help @research community
Hello,
I2 = between-study variance / (between-study variance + within-study variance)
I2 can only be zero when there is no significant between-study variance. That occurs only when all squares (i.e., representing the study effect sizes) in a forest plot are all lined up over each other (or nearly all lined up over each other). See the attached example. If the 95% CI for all studies in a meta-analysis overlap each other, I2 will equal zero.
Gordon Warren