Get a Quote

Free Tools › Leave-One-Out Sensitivity Analysis

Leave-One-Out Sensitivity Analysis for Meta-Analysis

Find out whether a single study drives your meta-analysis. Re-run the pooled analysis with each study left out in turn and see which ones change the conclusion.

FreeNo account neededYour data stays in your browser

Effect measure
Data entry
Model
StudyEvents (treatment)No event (treatment)Events (control)No event (control)
StudyRisk ratio (95% CI)Estimate [95% CI]Omitting Aronson 19480.49 [0.34, 0.71]Omitting Ferguson & Simes 19490.52 [0.36, 0.74]Omitting Rosenthal 19600.50 [0.35, 0.72]Omitting Hart & Sutherland 19770.54 [0.39, 0.74]Omitting Frimodt-Moller 19730.47 [0.32, 0.68]Omitting Stein & Aronson 19530.49 [0.33, 0.73]Omitting Vandiviere 19730.52 [0.36, 0.74]Omitting TPT Madras 19800.45 [0.33, 0.62]Omitting Coetzee & Berjak 19680.48 [0.33, 0.69]Omitting Rosenthal 19610.52 [0.36, 0.75]Omitting Comstock 19740.47 [0.31, 0.70]Omitting Comstock & Webster 19690.47 [0.33, 0.67]Omitting Comstock 19760.46 [0.32, 0.66]All studies0.49 [0.34, 0.70]0.51Orange rows change the conclusion or fall outside the overall 95% CI

Leave-one-out results (random effects)

Study omittedRisk ratio [95% CI]pI²τ²
Aronson 19480.49 [0.34, 0.71]< 0.00193%0.312
Ferguson & Simes 19490.52 [0.36, 0.74]< 0.00192%0.300
Rosenthal 19600.50 [0.35, 0.72]< 0.00193%0.309
Hart & Sutherland 19770.54 [0.39, 0.74]< 0.00189%0.217
Frimodt-Moller 19730.47 [0.32, 0.68]< 0.00193%0.326
Stein & Aronson 19530.49 [0.33, 0.73]< 0.00191%0.383
Vandiviere 19730.52 [0.36, 0.74]< 0.00192%0.301
TPT Madras 19800.45 [0.33, 0.62]< 0.00184%0.211
Coetzee & Berjak 19680.48 [0.33, 0.69]< 0.00193%0.327
Rosenthal 19610.52 [0.36, 0.75]< 0.00192%0.295
Comstock 19740.47 [0.31, 0.70]< 0.00193%0.388
Comstock & Webster 19690.47 [0.33, 0.67]< 0.00193%0.309
Comstock 19760.46 [0.32, 0.66]< 0.00193%0.318
None (all 13 studies)0.49 [0.34, 0.70]< 0.00192%0.309

The pooled estimate stays within the overall 95% CI and keeps the same statistical significance whichever study is removed, so no single study drives the result.

Get new free research tools firstOccasional emails when we add tools and guides. No spam.

We only use your email to send these updates, never share it, and you can unsubscribe by replying to any email or writing to info@timelyscholar.com. Privacy Policy

How to Use the Leave-One-Out Sensitivity Analysis

  1. 1

    Enter your studies

    2×2 counts or estimates with 95% CIs.

  2. 2

    Choose the model

    Random effects or fixed effect.

  3. 3

    Check flagged studies

    Rows in orange change significance or move outside the overall CI.

What is a leave-one-out analysis?

A leave-one-out (influence) analysis repeats the meta-analysis k times, omitting one study each time. If the pooled estimate and its significance stay similar, the result is robust; if removing one study changes the conclusion, that study is influential and should be discussed.

Sensitivity analyses like this are recommended by the Cochrane Handbook to check whether findings depend on arbitrary or unclear decisions, such as including a study at high risk of bias.

How it is calculated

Sources

Leave-One-Out Sensitivity Analysis: FAQ

How do I report a leave-one-out analysis?

Give the range of pooled estimates across the omissions and name any study whose removal changes the conclusion.

What should I do with an influential study?

Look for reasons such as risk of bias, population or dose, and present results with and without it. Do not simply drop it.

Related Free Tools

Forest Plot GeneratorFunnel Plot GeneratorI² Heterogeneity CalculatorAll free tools →
Get a Free Quote