MerQuR+Statistics Glossary

Statistics Glossary → Knapp-Hartung (HKSJ) adjustment

Knapp-Hartung (HKSJ) adjustmentconcept

An adjustment in random-effects meta-analysis that rescales the standard error of the pooled effect and bases tests on the t distribution instead of the normal.

Classical random-effects meta-analysis (e.g. DerSimonian-Laird) builds the confidence interval for the pooled effect from the z distribution and ignores the uncertainty in the estimate of the heterogeneity variance τ². With few studies these intervals are too narrow, and actual coverage can fall well below 95%. The Hartung-Knapp-Sidik-Jonkman (HKSJ) method rescales the standard error according to the weighted spread of the studies around the pooled estimate and uses a t(k - p) distribution for k studies and p parameters.

Simulation studies have shown that HKSJ keeps coverage much closer to the nominal level, especially with few studies and marked heterogeneity, so many guidelines recommend it by default for random-effects models. When the studies are very homogeneous the scaling factor can fall below 1, giving an interval narrower than the classical one; some authors recommend not letting the factor drop below 1 in that case. The adjustment does not change the pooled estimate itself, only its uncertainty.

An everyday example

Only 5 studies have examined the effect of a new exercise programme, and their results differ quite a lot. The classical random-effects model gives an interval of 0.10 to 0.70 for the pooled effect, and the result looks significant. With the Knapp-Hartung adjustment and a t(4) distribution, the interval widens to -0.05 to 0.85, making it visible how hard it is to draw a firm conclusion from a few inconsistent studies.

In MerQur

Available as the 'Knapp-Hartung (HKSJ)' option in meta-analysis; it rescales the standard error and uses the t(k - p) distribution.

Analyses where this concept appears

Meta-Analysis

Related concepts

Between-study variance (τ²) and heterogeneity (I²)Fixed and random effects