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

Start one
Contents

Academy · Research Data, Analytics & AI · Pooled SD Calculation

Calculating the Pooled Standard Deviation When Combining Two Subgroups

Combining two subgroups into one overall group, such as a male and a female arm of the same trial, requires a specific formula. The pooled standard deviation is not a simple average of the two subgroup standard deviations (SDs), because it must also account for the difference between the subgroup means.

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

A pooled standard deviation is the single SD that describes a combined group formed by merging two reported subgroups, such as male and female arms of the same study, or two recruitment sites, into one overall sample. This problem differs from the two-group comparison on which most meta-analysis guidance focuses: rather than comparing subgroup A with subgroup B, the task is to combine A and B into a single dataset and report the mean and SD that the combined dataset would have had, using only the summary statistics already published for each subgroup.

Learning objectives

  • The formula for combining the means and SDs of two subgroups into a single pooled mean and SD
  • Why the formula requires a term for the difference between the two subgroup means, in addition to their sample sizes
  • Circumstances in which this calculation is inappropriate and a standard two-group comparison is required

Common misconceptions

  • “Pooling two subgroups' SDs is just a weighted average of the two numbers.” The Cochrane Handbook's combining-groups formula includes a specific term proportional to the squared difference between the two subgroup means; omitting that term and simply weight-averaging the two SDs systematically understates the SD of the merged group whenever the subgroup means differ.

Rationale for combining two subgroups

This calculation is required whenever a published paper reports an outcome separately for two subgroups but the reviewer extracting data for a meta-analysis, a systematic review, or a secondary analysis requires a single value for the whole sample. A trial might report blood pressure change separately for participants under and over 65, or a cohort study might report cholesterol separately by sex, while the review's unit of analysis is the entire enrolled population rather than either subgroup alone1. Contacting the original authors for the combined figure is often impractical within a review timeline, and the participant-level data underlying published subgroup tables are rarely available; the data-extraction guidance of the Cochrane Handbook addresses precisely this situation with a direct algebraic solution4.

The same need arises outside systematic reviews. A hospital auditing length of stay across two wards, or a trialist combining two recruitment sites before running a single overall analysis, faces the same arithmetic problem: two reported means, two reported SDs, two sample sizes, and the need for the single combined mean and SD that the full dataset would have produced had it not been divided. Because the underlying data are unavailable, that combined figure can be obtained only from a formula built on summary statistics, not from a shortcut that approximates it.

Formulas for the pooled mean and pooled SD

Given subgroup 1 with sample size n₁, mean M₁, and SD SD₁, and subgroup 2 with n₂, M₂, and SD₂, the Cochrane Handbook's combining-groups method first finds the combined sample size and the combined mean exactly, since means combine by simple weighting4:

N = n₁ + n₂
M = (n₁M₁ + n₂M₂) / N

The combined SD requires more than a weighted average of SD₁ and SD₂, because it must capture both the internal spread of each subgroup and the additional spread introduced by the two subgroups being centred on different means45:

SD = √{ [(n₁−1)SD₁² + (n₂−1)SD₂² + (n₁n₂/N)(M₁−M₂)²] / (N−1) }

The first two terms inside the brackets are each subgroup's own sum of squared deviations from its own mean, carried over directly from the reported SDs. The third term converts those two separate internal variances into the variance of one dataset measured against a single combined mean, and it is the term most frequently omitted in manual calculations.

Role of the difference between subgroup means in the formula

In the case of two subgroups of equal size with identical SDs but different means, weight-averaging the two SDs would return that same SD for the combined group, implying that the merge added no spread. This is incorrect: when the subgroup means differ, individuals in subgroup 1 who lie exactly at their own subgroup's mean are, by definition, some distance from the combined mean M, and the same applies to subgroup 2. That additional distance constitutes real variance in the merged dataset, and it is precisely what the (n₁n₂/N)(M₁−M₂)² term restores45. The larger the difference between the two subgroup means, the larger this correction, and the more a naive weighted-average SD will understate the true combined SD.

This has a direct practical consequence for reviewers: the error introduced by omitting the mean-difference term is not random noise that cancels out across many extractions but a one-directional bias that always reduces the estimated combined SD, which in a meta-analysis narrows confidence intervals and overstates precision for precisely those comparisons in which the two subgroups differed appreciably8. The magnitude of the term also depends on the balance between n₁ and n₂ relative to N: it is largest when the two subgroups are of approximately equal size, because n₁n₂/N is maximised at n₁=n₂, and it approaches zero as one subgroup comes to dominate the total sample.

Common errors in combining subgroups

The most frequent error is to treat this as a two-group comparison problem and to apply a t-value or z-score conversion formula, when the task is a merge rather than a comparison49. If the review question is whether the two subgroups differ from each other, the subgroup data should remain separate and be used in a standard mean-difference calculation; the combining formula applies only when the review's unit of analysis is the whole, undivided sample and the subgroup division in the source paper is incidental to that.

A second common error is applying the combining formula more than twice in sequence without re-deriving N and M at each step, which compounds small rounding errors from the published summary statistics. When three or more subgroups require merging, it is more accurate to generalise the formula across all subgroups simultaneously, pooled against the single grand mean, than to merge subgroups two at a time and treat each intermediate result as though it were a single reported subgroup free of accumulated rounding. A third, more basic error is extracting an SD from a table that in fact reports the standard error, which removes most of the true spread before the combining formula is applied; values should always be checked against the distinction between SD and SE described by Altman and Bland before any of this arithmetic begins6.

Worked example with realistic values

Consider a trial that reports a secondary outcome separately for two age subgroups: 40 participants under 65 with a mean score of 12.0 and SD of 3.0, and 60 participants aged 65 and over with a mean score of 15.0 and SD of 3.5. A reviewer requiring the combined mean and SD for the whole 100-participant sample first computes N = 40 + 60 = 100 and M = (40×12.0 + 60×15.0)/100 = (480+900)/100 = 13.84. A naive weighted average of the two SDs would give (40×3.0 + 60×3.5)/100 = 3.3, but that figure ignores the fact that the two subgroups are centred three points apart on either side of the combined mean.

Applying the full formula, the first two bracket terms are (40−1)×3.0² = 351 and (60−1)×3.5² = 722.75, summing to 1073.75. The mean-difference term is (40×60/100)×(12.0−15.0)² = 24×9 = 216. Adding these gives 1289.75, and dividing by (N−1) = 99 gives 13.03, the square root of which is approximately 3.6145. The correctly combined SD of 3.61 is noticeably larger than the naive weighted-average figure of 3.3, an understatement of roughly 9% that would have propagated into any downstream analysis using the merged dataset; this illustrates why the mean-difference term should not be omitted even when it appears to be a minor refinement.

Repeating the same calculation with the two subgroup sizes reversed (60 participants under 65 and 40 aged 65 and over, with the same means and SDs) shows how the size of the mean-difference term depends on the balance between the two groups. The term (n₁n₂/N)(M₁−M₂)² is maximised when the two subgroups are equal in size and decreases as one group comes to dominate the total, so a review combining a large subgroup with a very small one will obtain a smaller correction from this term than a review combining two subgroups of approximately equal size with the same mean difference45. This principle applies to any extraction involving subgroups of unequal size: the risk of substantially understating the combined SD by omitting the mean-difference term is greatest when the subgroups being merged are closest in size.

Key points

Key points

Combining two subgroups into one requires a formula with a specific term for the squared difference between their means. Omitting that term and averaging the two SDs understates the true spread of the merged group whenever the subgroup means differ.

Practical application

Practical application

  1. The task should be confirmed as a merge rather than a comparison. The combining formula should be used only when the unit of analysis of the review is the whole sample; if the question is whether the subgroups differ, they should be kept separate and analysed with a mean-difference method.
  2. The sample size (n), mean and SD of each subgroup should be extracted exactly as reported. The definitions in the source table should be checked carefully to confirm that the value is an SD and not a standard error (SE) before either number is used in the formula.
  3. The combined sample size (N) and mean (M) should be computed first, and both subgroup SDs and the mean difference then entered into the full SD formula. A simple weighted average of the two SDs should not be substituted; the mean-difference term is required.
  4. For three or more subgroups, the formula should be generalised rather than applied to successive pairs. Sequential pairwise merging compounds rounding error across intermediate steps.

Frequently asked questions

No. A t-test's pooled SD assumes that two groups share one underlying variance and uses it to test whether their means differ. The pooled SD described here characterises the combined dataset formed by merging two reported subgroups into one; it describes a merge rather than an assumption used in a significance test.

The formula applies directly with the actual n1 and n2 of the subgroups; the mean-difference correction term becomes smaller as the imbalance in size increases, because it is scaled by n1n2/N, which falls as one subgroup comes to dominate the total.

Yes, provided that each subgroup's n, mean, and SD are the best available estimates for that subgroup; the combining formula does not depend on whether the inputs were reported directly or first estimated by another method, such as from a reported range or interquartile range.

No. Unlike a standard two-sample t-test's pooled-variance assumption, this combining formula uses each subgroup's own separately reported SD and does not require the two subgroup variances to be equal.

References

  1. 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
  2. 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
  3. 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
  4. 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
  5. Altman DG, Bland JM. Standard deviations and standard errors. BMJ. 2005;331(7521):903. read 10 October 2026. View on The BMJ
  6. 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
  7. 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
  8. 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
  9. 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
  10. 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

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). Calculating the Pooled Standard Deviation When Combining Two Subgroups. Pubrica. https://pubrica.com/academy/calculation-of-pooled-standard-deviation/

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.

Subgroup-Combining Worksheet

This checklist applies when two reported subgroups are merged into one combined mean and SD for extraction.

[ ] Confirmed the task is a merge, not a between-subgroup comparison
[ ] Extracted n1, M1, SD1 and n2, M2, SD2 exactly as reported, double-checked against SE
[ ] Computed N = n1+n2 and M = (n1M1+n2M2)/N
[ ] Computed the mean-gap term (n1n2/N)(M1-M2)^2
[ ] Computed combined SD using the full Cochrane Handbook formula, not a weighted average of SD1 and SD2
[ ] If combining 3+ subgroups, generalised the formula rather than merging pairs sequentially

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 and biostatistics review team recomputes pooled means and SDs from subgroup tables using the full Cochrane Handbook formula, to prevent a meta-analysis or audit from inheriting a combined SD understated by a shortcut weighted average.

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.

Verification that a pooled SD accounts for the difference between subgroup means

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.