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

Start one
Contents

Academy · Research Data, Analytics & AI · SD from Five-Number Summary

Estimation of the Standard Deviation from the Five-Number Summary (Minimum, Q1, Median, Q3, Maximum)

The complete five-number summary (minimum, first quartile, median, third quartile and maximum) is the most informative data-extraction scenario addressed by the estimators of Wan and colleagues, and it yields a measurably more accurate standard deviation (SD) than either the range or the interquartile range alone.

Pubrica AcademyWritten by a biostatistics reviewer Reviewed bya meta-analysis methodologist Published10 October 2026 Last checked10 October 2026 Reading time8 minutes Sources10 peer-reviewed

When a source paper reports the full five-number summary, namely the sample minimum, first quartile (Q1), median, third quartile (Q3), and maximum, together with the sample size n, more order-statistic information is available than in any other common SD-estimation scenario. The 2014 methodology paper by Wan, Wang, Liu and Tong treats this as its richest estimation case, combining the range-based signal (from the minimum and maximum) with the interquartile-range-based signal (from Q1 and Q3) in a single weighted estimator that performs better than either signal alone.

Learning objectives

  • Why having all five summary statistics together improves accuracy beyond using the range or IQR alone
  • How Wan and colleagues' estimator weights the range-based and IQR-based components
  • What Luo and colleagues' 2018 refinement adds for the accompanying mean estimate

Common misconceptions

  • “If you already have the range, adding the IQR to the calculation does not meaningfully help.” The validation work of Wan and colleagues showed that their combined five-number-summary estimator reduced estimation error relative to the range alone, because the quartiles carry independent information about the shape of the centre of the distribution that the two extreme values cannot capture.

Advantage of five summary statistics over two or three

Every other SD-estimation method in this cluster works from a partial view of the shape of a dataset: the range describes only the two most extreme points, and the interquartile range describes only the middle 50%. With the minimum, Q1, median, Q3, and maximum available together, it is possible to assess both the distance of the extremes from the bulk of the data and the concentration of the middle half of the data, two distinct pieces of information about the same distribution1.

Wan and colleagues formalised this reasoning through a direct empirical test: across simulated datasets of varying sample size and underlying distribution shape, their five-number-summary estimator consistently produced SD estimates closer to the true value than either a range-only or an IQR-only estimator applied to the same data1. The practical implication for data extraction is clear: whenever a source paper reports the full five-number summary rather than a subset of it, the combined estimator is more accurate than an approach that discards some of the available statistics.

Operation of the weighted combination estimator

The approach of Wan and colleagues does not simply average a range-based SD estimate and an IQR-based SD estimate. It derives a single formula that uses all five order statistics and n directly, based on the expected values of each order statistic under a normal distribution, and the effective weighting of the formula between range-based and IQR-based information varies with sample size1. For smaller samples, in which the extreme values (minimum and maximum) carry proportionally more information relative to their expected position under normality, the range-based component contributes more to the final estimate; for larger samples, the quartile-based component becomes more informative, because quartiles from larger samples are more stable estimates of the corresponding population values12.

This sample-size-adaptive weighting is the reason the formula cannot be reduced to a simple rule of thumb in the manner of the IQR/1.35 or range/d₂ approximations; it requires all five reported values and the sample size, processed through the published formula or an implementation of it, rather than an estimate made by hand1. Several freely available meta-analysis toolkits and online calculators implement the Wan formula directly, which removes the need to work through the underlying order-statistic algebra by hand.

Implications for estimation of the mean

Because this is the richest data scenario in the cluster, it also supports the most accurate available estimate of the mean, not only of the SD. The original 2014 method of Wan and colleagues already improves on the simple (min+max)/2 and median-only mean estimates of the older Hozo approach by incorporating the quartiles as well13. The 2018 follow-up paper by Luo, Wan, Liu and Tong refined this further, deriving an optimal linear combination of the five order statistics specifically for estimating the mean, validated across a wider range of simulated sample sizes and distribution shapes than the original 2014 paper covered4.

This is important in meta-analysis because a pooled mean-difference analysis requires both a mean and an SD for every included study, and errors in either propagate into the pooled estimate. Using the most accurate available estimator for both, rather than a method that estimates the SD well but leaves the mean to a cruder approximation, reduces the total error that a single imputed study contributes to the final result of a review56.

Application of the estimator in practice

In practice, reviewers should first confirm that all five values and n are available and consistently reported, since a common extraction error, arising when a paper uses ambiguous labelling, is to confuse the reported median with the mean, which invalidates the entire subsequent estimate1. Once these values are confirmed, applying the Wan or Luo formula, by hand, in a spreadsheet implementation, or through a published calculator, yields both the mean and the SD required for pooling.

As with every imputed statistic in this cluster, the normality assumption underlying the derivation still applies, and the result should be checked against the reported median: an estimated mean that differs substantially from the reported median indicates either a data-entry error or skew severe enough that the normal-distribution-based estimator should be interpreted cautiously17. Even with the richest available input data, this estimator remains an approximation based on a distributional assumption, and the PRISMA 2020 expectation that synthesis methods be reported transparently applies to it exactly as it does to the simpler range-only and IQR-only cases8.

A worked example of the improvement over simpler methods

Consider a study of n=30 that reports min=10, Q1=22, median=30, Q3=38, max=55. A range-only estimate using the control-chart divisor for n=30 (d₂ ≈ 4.09) gives SD ≈ (55−10)/4.09 ≈ 11.0. An IQR-only estimate using the fixed constant gives SD ≈ (38−22)/1.35 ≈ 11.91. These two single-signal estimates differ by close to one unit, which itself suggests that the underlying data may depart somewhat from normality, possibly reflecting right skew, as indicated by the gap between the maximum (55) and the median (30) being much larger than the gap between the median and the minimum.

The combined five-number-summary estimator of Wan and colleagues does not simply average these two figures; it weights the range-based and IQR-based information according to the sample size and the expected order-statistic relationships for n=30, and for a sample of this size it typically yields a value closer to the IQR-based estimate, since quartiles from 30 observations are more stable than the two extreme order statistics12. The practical lesson concerns less the exact resulting value than the diagnostic value of computing both single-signal estimates first: when they disagree noticeably, as in this example, the disagreement provides early warning that the study outcome may be skewed enough to warrant the same sensitivity-analysis caution applied to the simpler single-signal methods elsewhere in this cluster.

A further practical check specific to the five-number summary is to confirm that the five values are internally consistent before any formula is applied: the minimum should not exceed Q1, Q1 should not exceed the median, the median should not exceed Q3, and Q3 should not exceed the maximum. A violation of this ordering, for instance a reported Q1 larger than the reported median, almost always indicates a transcription error in the source table or in the extraction, and identifying it before a derived SD is computed prevents the error from being carried into a pooled estimate13.

This ordering check is also a fast way to catch a surprisingly common source paper error: a results table where a column header has shifted by one position during typesetting, so that what is labelled "Q3" is actually the maximum, and what is labelled "max" is actually missing a value entirely. A five-number summary that otherwise looks plausible but fails this basic ordering test should be treated as a flag to re-check the original source table, or supplementary materials, before using any of its five values in a calculation at all1. This ordering check is quick to perform and is among the least costly safeguards against an extraction error that would otherwise go undetected.

Key points

Key points

The full five-number summary supports the most accurate available SD (and mean) estimator in this cluster, because Wan and Luo's formulas combine range-based and quartile-based information with sample-size-adaptive weighting rather than relying on either signal alone.

Practical application

Practical application

  1. All five values and the sample size should be confirmed to be reported consistently and unambiguously. Particular attention should be given to papers that label a mean as a median or vice versa, which will invalidate the entire subsequent estimate.
  2. The Wan (2014) or Luo (2018) formula should be used rather than a manual average of range-based and IQR-based estimates. The correct formula weights the two components by sample size in a manner that a simple average cannot replicate.
  3. The estimated mean should be checked against the reported median. A large discrepancy indicates either a data-entry error or skew severe enough to warrant cautious interpretation of the estimate.
  4. An available calculator or spreadsheet implementation should be used rather than manual derivation of the order-statistic algebra. Several published meta-analysis toolkits implement these formulas directly, reducing the risk of calculation error.

Frequently asked questions

When all five values are available, yes; the simulations of Wan and colleagues showed the combined estimator outperforming either single-signal method on the same data, because it uses strictly more information about the shape of the distribution.

The 2014 paper by Wan and colleagues also covers several reduced scenarios, such as median, range and n, or median, IQR and n, each with its own formula; the full five-number case is the most information-rich of the set.

Not strictly; the formulas can be implemented in a spreadsheet, though several free online calculators and meta-analysis packages already implement them directly, which reduces the risk of manual calculation error.

The extraction should first be re-checked for labelling errors; if the values are confirmed as correct, the discrepancy probably reflects skew, and the study may be better handled narratively or through a methods-without-meta-analysis approach than by forcing it into the pooled estimate.

References

  1. 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
  2. Luo D, Wan X, Liu J, Tong T. Optimally estimating the sample mean from the sample size, median, mid-range, mid-quartile range, and/or range. Statistical Methods in Medical Research. 2018;27(6):1785-1805. read 10 October 2026. View on Statistical Methods in Medical Research
  3. 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
  4. Higgins JPT, Li T, Deeks JJ (editors). Chapter 6: Choosing effect measures and computing estimates of effect. In: Cochrane Handbook for Systematic Reviews of Interventions, version 6.4. Cochrane, 2023. read 10 October 2026. View the Cochrane Handbook, Chapter 6
  5. Deeks JJ, Higgins JPT, Altman DG (editors). Chapter 10: Analysing data and undertaking meta-analyses. In: Cochrane Handbook for Systematic Reviews of Interventions, version 6.4. Cochrane, 2023. read 10 October 2026. View the Cochrane Handbook, Chapter 10
  6. Walter SD, Yao X. Effect sizes can be calculated for studies reporting ranges for outcome variables in systematic reviews. Journal of Clinical Epidemiology. 2007;60(8):849-852. read 10 October 2026. View on Journal of Clinical Epidemiology
  7. Altman DG, Bland JM. Standard deviations and standard errors. BMJ. 2005;331(7521):903. read 10 October 2026. View on The BMJ
  8. 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
  9. Higgins JPT, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327(7414):557-560. read 10 October 2026. View on The BMJ
  10. Altman DG, Bland JM. How to obtain the confidence interval from a P value. BMJ. 2011;343:d2090. 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). Estimation of the Standard Deviation from the Five-Number Summary (Minimum, Q1, Median, Q3, Maximum). Pubrica. https://pubrica.com/academy/estimation-of-standard-deviation-with-minimum/

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.

Five-Number-Summary SD Worksheet

This method applies when a source paper reports min, Q1, median, Q3, max, and n together.

[ ] Confirmed all five values and n are present and unambiguously labelled
[ ] Checked the paper is not mislabelling a mean as a median or vice versa
[ ] Ran the five values through the Wan (2014) or Luo (2018) estimator for mean and SD
[ ] Sense-checked the estimated mean against the reported median
[ ] Labelled both the mean and SD as imputed if used in a pooled analysis

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 data extraction team applies the validated Wan and Luo five-number-summary estimators, rather than a manually averaged approximation, whenever a source reports the full five-number summary, maximising the accuracy of every imputed SD that enters a client’s meta-analysis.

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.

Use of every statistic reported in the source paper during extraction

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.