Table 2. publication bias assessments in unbiased and biased simulations using the RMD, SMD or NMD in combination with an SE or sample size-based precision estimate (simulation 3).
Precision estimate SE | Precision estimate 1/√n | ||||
---|---|---|---|---|---|
Effect measure | Bias? | % of sims with Egger’s p<0.05 | Median p-value (range) | % of sims with Egger’s p<0.05 | Median p-value (range) |
RMD | No | 5.1 | 0.51 (0.001–1.0) | 5.1% | 0.50 (0.001–1.0) |
RMD | Yes | 69.1% | 0.01 (2.7*10−8 - 0.99) | 69.6% | 0.01 (1.6*10−8 - 0.97) |
SMD | No | 100% | 2.9*10−13(0–8.1*10−6) | 4.3% | 0.51 (0.001–1.0) |
SMD | Yes | 100% | 4.4*10−16(0–1.8*10−6) | 72.4% | 0.01 (5.4*10−10 - 0.99) |
NMD | No | 6.4% | 0.51 (0.001–1.0) | 6.4% | 0.50 (0.001–1.0) |
NMD | Yes | 60.5% | 0.02 (7.1*10−8 - 0.99) | 60.4% | 0.02 (8.0*10−8 - 0.98) |
Simulated meta-analyses contained 300 studies (total study n = 12–30 subjects) and the difference in normally distributed means between control and intervention group was 10. Publication bias was introduced stepwise, by removing 10% of primary studies in which the difference between the intervention and control group means was significant at p<0.05, 50% of studies where the significance level was p≥0.05 to p<0.10, and 90% of studies where the significance level was p≥0.10. SE = standard error; RMD = raw mean difference; SMD = standardized mean difference (Hedges’ g); NMD = normalized mean difference; sims = simulations.