Pooling across studies

The Synthesis tab pools evidence across your studies. Where the Evidence tab held one panel per reported thing, Synthesis groups by outcome domain and measure family, so two studies that measured the same construct through different instruments land in one grouped forest. Each pooled diamond is computed by DerSimonian–Laird random-effects, the same forest machinery the Evidence tab uses, with the DL diamond appearing once at least two studies pool (k ≥ 2). It stays a preview, not a published analysis, until you accept the studies underneath it.

The diamond is provisional until you accept it

Every pool carries a stamp. A pool built over some values the machine has not confirmed reads provisional pool; a pool whose every plotted study you have accepted reads confirmed pool. These are two different numbers: a pool over everything is not a pool over the accepted-only set, and the stamp tells you which you are looking at.

Under a provisional diamond you will see a line like 3/5 studies accepted. When the accepted-only pool can already form, it shows you that confirmed estimate and its interval right there, and invites you to accept the rest to make the pool authoritative. When fewer than two of the plotted studies are accepted, it says so plainly: accept at least two to form a confirmed pool. Accepting happens where it always does, in Reviewing evidence and the quality queue. Synthesis reflects those decisions live.

What pooled, and what didn’t

A pool never quietly drops a study. Values that would join a group but carry a comparator that does not match the pool’s reference are listed under Excluded from this pool, with the reason naming both arms, for example incompatible comparator (vs "waitlist"; pool compares vs "placebo"). The reference is the comparator most of the group’s studies share, so the pool keeps the largest compatible arm-set and shows you what it set aside. A study you rejected or flagged needs-fix appears in the same list, carrying your own reason.

Below the pool stack, a Not pooled tail catches everything that could not deterministically group at all, sorted by why:

• Non-forest family (no standardized axis) for outcomes whose measure has no forest axis to sit on.
• No outcome domain (can’t group across studies) for values with no domain to match on across studies.

And when a domain has studies but none of them carry a derivable standardized effect, the panel says the outcome was reported natively, not plotted. Nothing is ever fabricated onto the axis to fill a gap.

Reading the strip and the caveats

The overview strip across the top counts pools, studies, poolable now (groups with a live pooled diamond), and confirmed (poolable groups where every plotted study is accepted).

Each pool also surfaces honest caveats read straight off its own numbers: a pool with fewer than three studies carries no prediction interval; a heterogeneous pool names its I². For the derivation behind any single study’s effect size, open its receipt. What you confirm here is what your exports are built from: the same confirmed pool, carried out to metafor, RevMan, or a workbook.