TimelyScholarFree Tools › Publication Bias Selection Model
Estimate how the pooled effect would change if studies with non-significant results are less likely to be published, using a Vevea-Hedges step-function selection model.
FreeNo account neededYour data stays in your browser
| Unadjusted (random effects, ML) | 1.242τ² = 0.0204 |
|---|---|
| Adjusted for selection | 1.215 (95% CI 1.039 to 1.420)τ² = 0.0166 |
| Relative publication weight, p > 0.025 | 0.764Weight 1 = as likely to be published as significant results |
| Test for selection | LRT χ²(1) = 0.12, p = 0.732 |
| Studies per interval | ≤ 0.025: 7; 0.025–1: 30 |
Selection models are a sensitivity analysis: they show how the pooled estimate would change under a specific pattern of publication bias. Estimates match metafor's selmodel(type = "stepfun").
Copy the columns from Excel or Google Sheets, with names in the first row, or load the example.
Estimates, tests and the figure update instantly in your browser.
Download the figure (SVG or PNG) and copy a results sentence.
A selection model adds a weight function to the random-effects model: studies whose one-sided p-values fall in different intervals (for example below .025 and above) may have different probabilities of being published. The model estimates those relative probabilities and an adjusted pooled effect.
Selection models need a reasonable number of studies in each p-value interval, and their results depend on the assumed selection pattern. Treat them as a sensitivity analysis alongside funnel plots and other methods.
The one-sided p-values assume that larger (more positive) effects are favoured for publication. If your benefit is a negative effect, reverse the sign of the effect sizes first.