Contents
Academy · Research Data, Analytics & AI · Effect Size Calculation
Effect Size Calculation in Meta-Analysis: Metrics, Conversions and Weighting
Effect size calculation converts the result of each included study into a common, comparable metric before pooling. Selection of an inappropriate metric for the outcome type can distort a meta-analysis as much as the inclusion of inappropriate studies.
Effect size calculation is the step in a meta-analysis where each included study's result is converted into a standardised, comparable numeric metric, such as an odds ratio, a risk ratio, a mean difference, or a standardised mean difference, so that studies measuring the same underlying construct in different ways can be combined on a common scale. The choice of metric is determined by outcome type (binary outcomes typically use odds or risk ratios; continuous outcomes use mean differences or standardised mean differences), and that choice shapes every subsequent step of the synthesis.
Learning objectives
- The main effect size metrics used in meta-analysis and which outcome types each one fits
- Why standardised mean difference exists specifically to handle studies using different measurement instruments
- How effect sizes are weighted when pooled, and why larger studies do not automatically dominate the result
Common misconceptions
- “Any numeric result from a study can be pooled directly with any other study's numeric result.” Studies that measure the same underlying outcome with different instruments or scales cannot be pooled as raw numbers; they require conversion to a common effect size metric, such as the standardised mean difference, specifically so differences in measurement scale are not mistaken for differences in effect.
Purpose of effect size calculation
Effect size calculation takes each included study's raw results (event counts, means and standard deviations, correlation coefficients) and transforms them into a standardised quantity that expresses the magnitude of an association or difference on a scale comparable across studies1. This step exists because individual studies rarely report results in an identical, directly combinable form (sample sizes, measurement instruments and sometimes the summary statistics themselves differ), so meta-analysis cannot simply average reported numbers without first converting them to a common metric2.
Each individual study's effect size is calculated with its own confidence interval or standard error, which carries forward into the pooling stage as the basis for that study's weight; a study's effect size calculation is therefore never just a single number but a point estimate plus a measure of its own precision, and both components must be extracted and carried through the analysis together.
Common effect size metrics and their applications
For binary outcomes, the odds ratio and risk ratio (relative risk) are the standard metrics, with odds ratios more common in case-control designs and logistic regression output, and risk ratios often preferred for cohort and trial data because they are more directly interpretable as a change in probability3. For continuous outcomes measured on the same scale across studies, the mean difference is used directly; when studies measure the same construct using different instruments or scales, the standardised mean difference, commonly Cohen's d or Hedges' g, rescales each study's result by its own standard deviation so results become comparable despite using different measurement tools4. For associations between two continuous variables, correlation coefficients, often Fisher's z-transformed for pooling, serve as the effect size metric5.
A fifth, increasingly common situation arises when included studies report time-to-event outcomes, such as time to disease progression or death, where the hazard ratio from a survival analysis serves as the effect size metric instead; hazard ratios cannot be meaningfully combined with odds or risk ratios for the same outcome without an explicit conversion, since they are estimated from a different statistical model and carry a different interpretation, a distinction reviewers sometimes overlook when a body of evidence mixes trials that reported binary outcomes with trials that reported full time-to-event data.
Importance of small-sample corrections and conversions
Cohen's d, the original standardised mean difference formula, is known to overestimate the true effect size slightly when calculated from small samples, which is why Hedges' g, which applies a specific correction factor based on the total sample size, has become the more commonly recommended default in meta-analysis software and guidance, particularly when a review includes several studies with modest sample sizes. The difference between the two is usually small for large studies and more consequential for small ones, so a reviewer pooling a mix of small and large trials should confirm which formula a given software package or prior published estimate used before treating two reported “standardised mean differences” as directly comparable.
Converting between metrics is sometimes unavoidable when included studies report results in genuinely incompatible forms, for instance one trial reporting a mean difference with standard deviations and another reporting only a p-value and sample size; published conversion formulas exist for many such situations, but each conversion introduces some additional uncertainty beyond what the original study's own reporting would have carried, which is a reasonable basis for a sensitivity analysis excluding converted studies to check whether they are driving the pooled result.
Selection of the metric according to outcome type
The Cochrane Handbook's chapter on choosing effect measures stresses that the decision should be made based on the outcome's nature and how the result will be interpreted clinically, not on which metric produces the most favourable pooled result6. Choosing an odds ratio for a common outcome can exaggerate the apparent size of an effect relative to a risk ratio, because odds and probability diverge further from each other as the outcome becomes more frequent, which is a well-documented interpretive error rather than a computational one3. Reviewers and readers should check which metric was used before comparing effect sizes across different meta-analyses, since the same underlying data can look different depending on the metric chosen.
For the standardised mean difference specifically, interpreting the pooled number clinically requires translating it back into units a clinician or patient would recognise, since a standardised mean difference of, for example, 0.5 standard deviations conveys little on its own to a reader accustomed to the original scale's points or units. Good practice is to report the pooled standardised mean difference alongside an illustrative back-conversion using a representative study's standard deviation, so that readers obtain both the statistically necessary common metric and a clinically interpretable translation of what it represents.
Weighting of effect sizes in pooled analysis
Once each study's effect size is calculated, pooling combines them using a weighted average, where weight is generally driven by each study's precision, approximated by the inverse of its variance, rather than by sample size alone, so a smaller study with an unusually precise estimate can carry more weight than a larger, less precise one2. The DerSimonian and Laird random-effects method, still the most widely used approach, adds a between-study variance component to this weighting specifically to account for genuine differences between studies rather than assuming every study estimates one single true effect7. Software packages, including the widely used metafor package in R, implement these weighting schemes so that the mechanics of pooling are standardised across analyses rather than computed by hand8.
Because the between-study variance component in a random-effects model is itself an estimate, calculated from the observed spread of effect sizes across the included studies, it carries its own uncertainty, particularly when a review includes only a small number of studies. Reviewers working with fewer than roughly ten included studies should treat the random-effects pooled confidence interval with some additional caution for this reason, and reporting the number of included studies prominently alongside the pooled estimate gives readers the context they need to judge how much weight that estimate's precision actually deserves. Some statisticians recommend reporting a prediction interval alongside the usual confidence interval in this situation specifically, since a prediction interval communicates the likely range of a true effect in a new, similar study, which is often more clinically relevant than the narrower range the confidence interval describes around the pooled mean itself. A clinician deciding whether to apply a pooled finding to an individual patient is implicitly asking the prediction-interval question (how much the true effect might plausibly vary in a comparable setting) rather than the narrower confidence-interval question about the precision of the average effect across the studies that happened to be included, a distinction that should be made explicit whenever a meta-analysis is being translated into practical guidance. Reporting both intervals side by side, rather than only the narrower confidence interval, gives a guideline panel or individual clinician a more complete and transparent account of what the pooled evidence can and cannot promise about any single future case. This is a modest additional step at the reporting stage, but it directly addresses one of the more common misreadings of meta-analytic results: mistaking a narrow confidence interval around the average for a guarantee about how any one new patient or setting will respond.
Key points
Key points
Effect size calculation converts the result of each study into a common, outcome-appropriate metric before pooling; the choice of metric and the application of precision-based weighting shape the pooled result as much as the selection of included studies.
Practical application
Practical application
- The metric should be matched to the outcome type before data extraction begins. The protocol should specify whether binary outcomes will be pooled as odds ratios or risk ratios, and whether continuous outcomes require a standardised mean difference.
- Inconsistently reported results should be converted to the chosen metric systematically. Documented conversion formulas, rather than ad hoc approximations, should be used when a study reports results in a form different from that chosen for pooling.
- The weighting method should be reported alongside the pooled estimate. Authors should state whether a fixed-effect or random-effects (e.g., DerSimonian-Laird) model was used, since the two can produce materially different pooled results and confidence intervals.
Frequently asked questions
Raw numbers do not account for differences in study precision, measurement instruments, or baseline risk across studies, so pooling them directly can produce a misleading result; standardised effect size metrics address this by putting every study on a common, comparable scale.
Both are standardised mean difference metrics; Hedges' g applies a small-sample correction that Cohen's d does not, making it the more commonly recommended choice when included studies have small sample sizes.
Only when the outcome is relatively rare; as outcome frequency increases, the odds ratio diverges further from the risk ratio and can overstate the apparent effect size, so the choice should be made deliberately rather than by habit.
Not necessarily. Weighting is generally based on the inverse of each study's variance, which correlates with but is not identical to sample size; an unusually precise smaller study can outweigh a larger, less precise one.
Not reliably. Two meta-analyses using different effect size metrics for conceptually similar outcomes are not directly comparable without converting one to the other's scale, which should be done explicitly rather than assumed.
References
- Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. Introduction to Meta-Analysis. Chichester: John Wiley & Sons; 2009. read 10 October 2026. View the publisher listing
- Hedges LV, Olkin I. Statistical Methods for Meta-Analysis. Orlando: Academic Press; 1985. read 10 October 2026. View the publisher listing
- Deeks JJ, Higgins JPT, Altman DG (editors). Cochrane Handbook, Chapter 6: Choosing effect measures and computing estimates of effect. read 10 October 2026. View Chapter 6
- Cohen J. Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Hillsdale, NJ: Lawrence Erlbaum Associates; 1988. read 10 October 2026. View the publisher listing
- Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Statistics in Medicine. 2002;21(11):1539-1558. read 10 October 2026. View on Statistics in Medicine
- Deeks JJ, Higgins JPT, Altman DG (editors). Cochrane Handbook, Chapter 10: Analysing data and undertaking meta-analyses. read 10 October 2026. View Chapter 10
- DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials. 1986;7(3):177-188. read 10 October 2026. View on Controlled Clinical Trials
- Viechtbauer W. Conducting meta-analyses in R with the metafor package. Journal of Statistical Software. 2010;36(3):1-48. read 10 October 2026. View on Journal of Statistical Software
- Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. read 10 October 2026. View on The BMJ
- Guyatt GH, Oxman AD, Vist GE, et al. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008;336(7650):924-926. read 10 October 2026. View on The BMJ
Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.
How to cite this article
Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.
Pubrica Academy. (2026, October 10). Effect Size Calculation in Meta-Analysis: Metrics, Conversions and Weighting. Pubrica. https://pubrica.com/academy/concepts-definitions/what-is-effect-size-calculation-in-meta-analysis/
Pubrica Academy. “Effect Size Calculation in Meta-Analysis: Metrics, Conversions and Weighting.” Pubrica, 10 Oct. 2026, pubrica.com/academy/concepts-definitions/what-is-effect-size-calculation-in-meta-analysis/.
Pubrica Academy (2026) Effect Size Calculation in Meta-Analysis: Metrics, Conversions and Weighting. Pubrica. Available at: https://pubrica.com/academy/concepts-definitions/what-is-effect-size-calculation-in-meta-analysis/ (Accessed: 15 October 2026).
Pubrica Academy. “Effect Size Calculation in Meta-Analysis: Metrics, Conversions and Weighting.” Pubrica. October 10, 2026. https://pubrica.com/academy/concepts-definitions/what-is-effect-size-calculation-in-meta-analysis/
Pubrica Academy, ‘Effect Size Calculation in Meta-Analysis: Metrics, Conversions and Weighting’ (Pubrica, 10 October 2026) <https://pubrica.com/academy/concepts-definitions/what-is-effect-size-calculation-in-meta-analysis/> accessed 15 October 2026.
Pubrica Academy. Effect Size Calculation in Meta-Analysis: Metrics, Conversions and Weighting. [Internet]. Pubrica; 2026 Oct 10 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/concepts-definitions/what-is-effect-size-calculation-in-meta-analysis/
Pubrica Academy, “Effect Size Calculation in Meta-Analysis: Metrics, Conversions and Weighting,” Pubrica, Oct. 10, 2026. [Online]. Available: https://pubrica.com/academy/concepts-definitions/what-is-effect-size-calculation-in-meta-analysis/
@misc{/la_effect_size_calculation_meta_analysis_2026,
author = {{Pubrica Academy}},
title = {Effect Size Calculation in Meta-Analysis: Metrics, Conversions and Weighting},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/concepts-definitions/what-is-effect-size-calculation-in-meta-analysis/},
note = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.
Effect Size Metric Decision Sheet
Use this when deciding which effect size metric to pool in a meta-analysis.
[ ] Outcome type identified: binary, continuous, or correlational [ ] Metric selected to match outcome type (odds/risk ratio, mean difference/SMD, correlation) [ ] Checked whether included studies use different instruments, requiring a standardised mean difference [ ] Conversion formulas documented for any study reporting in a different form [ ] Weighting method (fixed-effect vs. random-effects) decided and stated alongside the pooled result
Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.
How Pubrica can help
Pubrica’s biostatistics team selects and documents the effect size metric and weighting method for a meta-analysis before pooling, so that the resulting estimate is interpretable and defensible under peer review.