Table 3.
A summary of recommendations for meta‐analysis of mean difference (MD) and standardized MD (SMD)
| Meta‐analysis of MD | ||
|---|---|---|
| Estimation | n<100 | n ≥ 100 |
| τ 2 point | All estimators are positively biased for small n | any estimator |
| CDL is the least biased | ||
| τ 2 interval | WT | any estimator other than PL |
| μ point | any estimator | any estimator |
| μ interval | HKSJ for balanced studies with τ 2<0.1 and K ≤ 10, | HKSJ |
| where SSW CDL provides conservative coverage. | ||
| SSW CDL for unbalanced studies, or when K ≥ 20 and τ 2>0.1. | ||
| Meta‐analysis of SMD | ||
| τ 2 point | MP (somewhat underestimates) for K ≤ 10, | MP, KDB, or REML |
| KDB (somewhat overestimates) for K>10 | ||
| τ 2 interval | QP | QP, PL, KDB |
| δ point | SSW, all other estimators have negative bias | any estimator |
| δ interval | HKSJ or HKSJ KDB for δ<0.5, SSW KDB for δ ≥ 0.5 | HKSJ or HKSJ KDB or SSW KDB |
Abbreviations: CDL, corrected DerSimonian‐Laird; HKSJ, Hartung‐Knapp‐Sidik‐Jonkman; KDB, Kulinskaya‐Dollinger‐Bjørkestøl; MP, Mandel‐Paule; PL, profile‐likelihood; QP, Q‐profile; REML, restricted maximum‐likelihood; WT, Welch‐type.