← allmeta

Heterogeneity Explorer

Paste study-level effect sizes. Reports Q, I² with its 95% CI, τ² by eight estimators (PM, REML, ML, EB, SJ, HE, HS, DL) with its 95% CI, the pooled estimate with z and HKSJ intervals, the 95% prediction interval, a Baujat plot and leave-one-out influence. DerSimonian-Laird needs at least 10 studies.

Studies

Data
Blank/comment lines and one recognised estimate/SE header are ignored. Every study row must be valid; malformed rows block pooling. Uploaded CSV requires exact study,yi,vi headers.
Options

The normal CI uses test="z"; the modified Hartung–Knapp CI uses a variance floor of 1 (test="adhoc"). The prediction interval uses the unadjusted model SE and tk−1 (test="t"). These are distinct approximations. Q-based I² and model-based I² are labelled separately; Q-profile intervals refer to model-based I².

Baujat plot

Leave-one-out

τ² estimators per DerSimonian & Laird (1986), Paule & Mandel (1982), REML (Viechtbauer 2005). HKSJ correction applied to RE pooled CI. I² via Q-statistic with 95% Q-profile CI (Viechtbauer 2007; matches metafor::confint); 95% PI uses tk-1 (Cochrane Handbook v6.5). Baujat after Baujat et al. (2002).