From: Risk of bias: a simulation study of power to detect study-level moderator effects in meta-analysis
 | Power for the identified heterogeneity (as present in the empirical dataset) | Power in the presence of reduced heterogeneity (if other moderators could be identified and heterogeneity could be reduced) | Power in the presence of no residual heterogeneity (if all other moderators could be identified) |
---|---|---|---|
Dataset 1 | Â | Â | Â |
216 trials, mean sample size 80 | |||
Observed heterogeneity: I2 = 72.4%, τ = 0.305 | |||
Modeled residual heterogeneity | 14% | 0.25% | 0% |
Moderator effect = 0.1 | 38% | 50% | 85% |
Moderator effect = 0.2 | 91% | 100% | 100% |
Dataset 2 | Â | Â | Â |
165 trials, mean sample size 286 | |||
Observed heterogeneity: I2 = 97.5%, τ = 0.345 | |||
Modeled residual heterogeneity | 70% | 35% | 0% |
Moderator effect = 0.1 | 12% | 20% | 100% |
Moderator effect = 0.2 | 37% | 60% | 100% |
Dataset 3 | Â | Â | Â |
100 trials, mean sample size 119 | |||
Observed heterogeneity I2 = 59.6% τ = 0.131 | |||
Modeled residual heterogeneity | 5% | 0.25% | 0% |
Moderator effect = 0.1 | 42% | 58% | 73% |
Moderator effect = 0.2 | 92% | 99% | 100% |
Dataset 4 | Â | Â | Â |
149 trials, mean sample size 342 | |||
Observed heterogeneity I2 = %, τ = 0.03 |