Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Data, Analytics & AI · SMD Estimation

Standardized Mean Difference (SMD) Estimation in Meta-Analysis: Estimating Means and Standard Deviations from Medians and Ranges

When a trial reports only a median, range and sample size rather than a mean and standard deviation, two peer-reviewed formulas, one published in 2005 and a more accurate one published in 2014, allow meta-analysts to estimate the values that a standardized mean difference requires. The trial can therefore be retained in the pooled estimate rather than excluded.

Pubrica AcademyWritten by a biostatistics methodologist Reviewed byan evidence synthesis statistician Published10 October 2026 Last checked10 October 2026 Reading time8 minutes Sources10 peer-reviewed

A standardized mean difference (SMD) expresses the difference between two group means in units of their pooled standard deviation, which lets a meta-analysis combine outcomes measured on different scales, such as two different depression questionnaires, into a single comparable effect size1. Correct estimation of an SMD depends entirely on the availability of a usable mean and standard deviation for each group. In practice, however, a meaningful proportion of eligible trials report medians, ranges or interquartile ranges instead, particularly for skewed outcomes, and these statistics must be converted before they can enter the pooled calculation. Although narrow, this technical problem is consequential: excluding every trial that reports a median rather than a mean is not a neutral choice, because the trials most likely to report medians, namely those with skewed or outlier-prone outcome distributions, may differ systematically from those that report means. Naive exclusion may therefore bias the pooled estimate rather than merely reduce its sample size.

Learning objectives

  • The quantity that the standardized mean difference standardizes, and why this makes it necessary for pooling across different scales
  • The two most widely used formulas for estimating a mean and standard deviation from a median, range, or interquartile range
  • The conditions under which each of the two formulas performs better, based on direct simulation evidence

Common misconceptions

  • “A trial that reports only a median and range cannot contribute to a meta-analysis’s pooled standardized mean difference.” The 2014 paper by Wan and colleagues in BMC Medical Research Methodology provides and validates formulas for estimating the sample mean and standard deviation from the sample size, median, and range and/or interquartile range. It extends the earlier 2005 method of Hozo and colleagues, specifically so that trials reporting only these summary statistics can be converted and included in a pooled SMD rather than excluded from the synthesis.

The quantity standardized by the SMD and its significance

The standardized mean difference divides the raw difference between two group means by a pooled, or sometimes a single-group, standard deviation, producing a unit-free effect size that can be pooled across studies even when they measured the same underlying construct with different instruments1. Cohen's original formulation, widely known as Cohen's d, and Hedges's small-sample-corrected version, Hedges's g, both follow this same logic but differ in a correction factor that matters most when individual study sample sizes are small2. Borenstein and colleagues’ widely used meta-analysis textbook adds a related caution relevant to interpretation: because the SMD is unit-free, authors should be cautious in comparing its magnitude across very different clinical contexts, since an SMD of 0.5 does not necessarily carry the same clinical meaning for a pain outcome as for a cognitive-test outcome, even though the two values are mathematically identical3. This standardization is what allows, for example, a meta-analysis of depression interventions to pool a trial that used the Hamilton Depression Rating Scale with one that used the Beck Depression Inventory, which a raw mean difference could not validly achieve because the two scales have different ranges and units3. Morris’s widely cited work on effect sizes for pretest-posttest-control group designs extends this standardization logic to a common but more complex case, in which a meta-analysis must combine change-from-baseline outcomes across studies with different baseline variances, and shows that the choice of standard deviation (the pretest, the posttest or a pooled change-score variance) can itself materially affect the resulting SMD if not handled consistently across the included studies8.

Estimation when a trial does not report a mean or standard deviation

Calculating any version of the SMD requires a mean and a standard deviation for each group, but trials reporting skewed outcomes (hospital length of stay and pain scores are common examples) often report a median and range or interquartile range instead, on the reasonable statistical grounds that the mean and SD are less representative for a skewed distribution4. Trial authors usually choose to report a median and range rather than a mean and standard deviation for a sound reason: for a genuinely skewed distribution, the median is a more representative measure of central tendency than the mean, which can be influenced substantially by a small number of extreme values. The reporting choice that creates the meta-analyst’s problem is therefore often good practice at the level of the individual trial, which is why a validated conversion method, rather than a blanket policy of excluding such trials, is the appropriate response. Before Hozo, Djulbegovic and Hozo published their widely used estimation method in 2005, meta-analysts in this situation generally had two unsatisfactory options: to exclude the trial from the pooled analysis entirely, discarding real evidence, or to request raw data from the original authors, which is frequently impractical given publication timelines and author turnover5. The scale of this problem is not trivial: skewed outcomes, such as length of hospital stay, time-to-event measures reported as medians and various symptom severity scores, are common precisely in the clinical areas, including critical care, oncology and surgery, where meta-analysis is most needed to resolve genuinely uncertain treatment questions. A validated conversion method is therefore important well beyond a narrow statistical detail.

Comparison of two formulas for estimating the mean and standard deviation

Hozo, Djulbegovic, and Hozo's 2005 BMC Medical Research Methodology paper provided formulas for estimating the mean and variance of a sample from its median, range, and sample size, under the assumption that the underlying distribution is reasonably close to normal or at least symmetric5. Wan, Wang, Liu, and Tong's 2014 paper in the same journal extended this work by also incorporating the interquartile range, when available, alongside the median and range, and by validating the resulting formulas through simulation studies across a range of sample sizes and underlying distributions6. Their simulations found that incorporating the interquartile range, where a trial reports one, generally produced more accurate estimates of both the mean and the standard deviation than relying on the median and range alone, particularly as sample size grew, making their 2014 method the generally preferred approach when a trial reports both quantities6. Both papers remain among the most cited methodological references in evidence synthesis because they addressed a problem for which no satisfactory alternative then existed: an author faced with a trial reporting only a median could either exclude it, losing statistical power and possibly introducing selection bias into the set of included trials, or convert it using a method whose accuracy had been tested through simulation rather than assumed.

Application of the conversion formulas in evidence synthesis

Neither formula is a substitute for obtaining the original mean and standard deviation when this is genuinely possible; both are explicitly presented by their authors as a best estimate for use when direct extraction is not an option, not as an equally valid alternative to the trial’s own summary statistics. Applying either formula to a specific trial requires verification of the assumption on which each depends: that the underlying outcome distribution is not so severely skewed that a mean and SD would be a misleading summary of it46. The Cochrane Handbook's chapter on analysing data recommends that authors report explicitly which trials in a meta-analysis had their summary statistics estimated through these conversion methods rather than extracted directly, so that readers and any subsequent sensitivity analysis can see how much of the pooled estimate rests on converted, rather than directly reported, data7. A sensitivity analysis excluding the converted trials, conducted alongside the main pooled estimate, is the most direct means of determining whether the conversion materially changed the overall conclusion, and it is inexpensive relative to the risk that an undisclosed, conversion-driven shift in the main result goes unnoticed by readers who cannot know that a conversion occurred. The same Handbook chapter also recommends checking the plausibility of a converted trial’s reported range against the other included studies before relying on the conversion, since an implausibly wide or narrow range, sometimes indicating a typographical error in the original paper or a unit mismatch, will distort the estimated mean and standard deviation however correctly the conversion formula is applied9. Reporting the result of the sensitivity analysis alongside the main estimate, rather than conducting it privately and omitting it if it happens to agree, is what allows a reader to verify that the conversion did not drive the review’s conclusion, instead of having to take this on trust. This disclosure is not burdensome once the conversion has been performed; it mainly requires authors to maintain, during data extraction, a list of the included trials that contributed converted rather than directly reported summary statistics, so that the information is available for reporting rather than having to be reconstructed retrospectively from extraction notes after the main analysis is complete.

Key points

Key points

A standardized mean difference requires a usable mean and standard deviation (SD) for each group. When a trial reports only a median, range or interquartile range, the 2014 formulas of Wan and colleagues, which build on and generally outperform the 2005 method of Hozo and colleagues, allow that trial to contribute to the pooled estimate rather than being discarded.

Practical application

Practical application

  1. The 2014 method of Wan et al. should be used when both the median and range and the interquartile range are reported. Their simulations indicated that this combination generally produced more accurate mean and SD estimates than the range alone, particularly at larger sample sizes.
  2. The 2005 method of Hozo et al. should be used only when the interquartile range is unavailable. Their original formulas remain valid when only the median, range, and sample size are reported, but offer less precision than methods incorporating the interquartile range.
  3. Every converted trial should be identified explicitly in the methods and results. Authors should report which studies’ summary statistics were estimated rather than extracted directly, and conduct a sensitivity analysis excluding them to assess the effect of the conversion on the pooled estimate.

Frequently asked questions

Excluding them discards real, eligible evidence and can introduce its own bias if the excluded trials differ systematically from the included ones; validated conversion formulas allow that evidence to inform the pooled estimate, provided that the assumptions of the conversion are reasonable for the outcome in question.

Both methods perform best when the underlying distribution is reasonably symmetric; severe skewness can make the mean and SD a misleading summary regardless of which formula estimates them, which is a limitation of the underlying statistic, not merely of the conversion method.

Hedges's g includes a small-sample correction that reduces upward bias in the effect size estimate, which matters most when individual study sample sizes are small; for larger samples the two converge closely, so the practical difference shrinks as sample size grows.

Yes. The Cochrane Handbook recommends disclosing this explicitly so readers can judge how much of the pooled result depends on estimated, rather than directly reported, summary statistics.

No. Hozo et al.'s and Wan et al.'s methods are specifically for estimating a mean and standard deviation of a continuous outcome from order statistics; binary outcomes require entirely different effect measures, such as risk ratios or odds ratios, and different estimation approaches.

References

  1. Cohen J. Statistical Power Analysis for the Behavioral Sciences. 2nd ed. Lawrence Erlbaum Associates; 1988. read 10 October 2026. View on Routledge
  2. Hedges LV, Olkin I. Statistical Methods for Meta-Analysis. Academic Press; 1985. read 10 October 2026. View on ScienceDirect
  3. Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. Introduction to Meta-Analysis. Wiley; 2009. read 10 October 2026. View on Wiley Online Library
  4. Higgins JPT, Thomas J, Chandler J, et al, eds. Cochrane Handbook for Systematic Reviews of Interventions, version 6.4, Chapter 6: Choosing effect measures and computing estimates of effect. Cochrane; 2023. read 10 October 2026. View on Cochrane Training
  5. Hozo SP, Djulbegovic B, Hozo I. Estimating the mean and variance from the median, range, and the size of a sample. BMC Medical Research Methodology. 2005;5:13. read 10 October 2026. View on BMC Medical Research Methodology
  6. Wan X, Wang W, Liu J, Tong T. Estimating the sample mean and standard deviation from the sample size, median, range and/or interquartile range. BMC Medical Research Methodology. 2014;14:135. read 10 October 2026. View on BMC Medical Research Methodology
  7. Higgins JPT, Thomas J, Chandler J, et al, eds. Cochrane Handbook for Systematic Reviews of Interventions, version 6.4, Chapter 10: Analysing data and undertaking meta-analyses. Cochrane; 2023. read 10 October 2026. View on Cochrane Training
  8. Morris SB. Estimating effect sizes from pretest-posttest-control group designs. Organizational Research Methods. 2008;11(2):364-386. read 10 October 2026. View on Organizational Research Methods
  9. Deeks JJ, Higgins JPT, Altman DG, eds. Cochrane Handbook for Systematic Reviews of Interventions, version 6.4, Chapter 6: Choosing effect measures and computing estimates of effect. Cochrane; 2023. read 10 October 2026. View on Cochrane Training
  10. 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

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). Standardized Mean Difference (SMD) Estimation in Meta-Analysis: Estimating Means and Standard Deviations from Medians and Ranges. Pubrica. https://pubrica.com/academy/concepts-definitions/standardized-mean-difference-smd-estimation/

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.

SMD Conversion Decision Worksheet

For use when an eligible trial reports a median rather than a mean and SD.

[ ] Confirmed the trial reports sample size plus median and range, and/or interquartile range
[ ] Checked whether the outcome's likely distribution is reasonably symmetric (not severely skewed)
[ ] Applied Wan et al. (2014) if interquartile range is available; Hozo et al. (2005) if only range is available
[ ] Flagged this trial as using estimated (not directly reported) mean/SD in the data extraction table
[ ] Planned a sensitivity analysis excluding converted trials to check impact on the pooled SMD

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 applies and documents validated methods for estimating means and standard deviations from trials that report only medians or ranges, and conducts sensitivity analyses so that the pooled standardized mean difference can be justified during peer review.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Pubrica can review whether the trials in a pooled SMD use reported or correctly estimated summary statistics.

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.