Contents
On this page
- Direct rearrangement from the standard error (SE) of the difference to the pooled SD
- The equal-variances assumption underlying the calculation
- Implications of unequal variances
- Application of the conversion in data extraction
- A worked example with realistic values
- Relationship of this method to the other conversions in the series
Academy · Research Data, Analytics & AI · SD from SE of the Difference
Estimation of the Standard Deviation (SD) from the Standard Error of the Difference Between Means
When a paper reports the standard error of the difference between two group means directly, recovery of a pooled standard deviation (SD) is the simplest conversion in this series: a single algebraic rearrangement with no distributional conversion step, provided that the variances of the two groups were assumed equal when the standard error was calculated.
Occasionally a source paper reports the standard error of the difference between two group means (SEdiff) directly, without separately stating the SD of either group, a p-value, a t-value, or a confidence interval. Of the SD-estimation scenarios covered in this cluster, this is the most direct: because the standard error of a difference between two means, under the common assumption of equal variances, is defined in terms of the pooled SD and the two sample sizes, the pooled SD can be recovered through a single algebraic rearrangement, without any intermediate distributional conversion.
Learning objectives
- The one-line rearrangement that recovers a pooled SD directly from a reported SE of the difference
- Why this calculation assumes equal variances between the two groups
- What changes when the reported SE instead comes from an unequal-variances (Welch-type) calculation
Common misconceptions
- “Having the SE of the difference already is basically the same as having the pooled SD.” The two quantities are related mathematically through the sample sizes but are not interchangeable without the rearrangement step: the SE of the difference reflects the combined sampling variability of both groups as scaled by n1 and n2, so the same SE corresponds to a different implied pooled SD depending on the size and balance of the two samples.
Direct rearrangement from the standard error (SE) of the difference to the pooled SD
Under the standard assumption that both groups share a common underlying variance, the standard error of the difference between two independent sample means is defined directly in terms of that shared (pooled) SD and the two sample sizes1:
SEdiff = SDpooled × √(1/n₁ + 1/n₂)
Rearranging this equation to solve for the pooled SD constitutes the entire calculation12:
SDpooled = SEdiff / √(1/n₁ + 1/n₂)
This is the final step shared by every other method in this cluster, each of which first recovers SEdiff from a p-value, t-value, z-score, or confidence interval. Having SEdiff reported directly omits all of those earlier steps and begins at the final one.
The equal-variances assumption underlying the calculation
The formula above is specifically the equal-variances version of the standard error of a difference, which is what underlies a standard pooled-variance independent-samples t-test13. When a paper does not specify how its reported SEdiff was calculated, the equal-variances assumption is a reasonable default, consistent with the presentation of the standard two-sample comparison in most introductory and applied biomedical statistics2.
The assumption is important because it is what allows a single pooled SD, rather than two separate group SDs, to be recovered from one reported value. If the true variances of the two groups differ substantially, the pooled SD remains a meaningful summary of the combined spread for extraction purposes, but it should not be mistaken for the individual SD of either group, which this method cannot recover separately from SEdiff alone1.
Implications of unequal variances
When a paper explicitly states that it used an unequal-variances or Welch-type correction for its SE calculation, often alongside Welch's t-test rather than the standard pooled-variance version, the SEdiff reported no longer decomposes into a single pooled SD using the formula above, because it was built from each group's separate variance divided by its own sample size, rather than one shared pooled variance34. In that case, recovering a single pooled SD from SEdiff alone is not possible without further information, such as each group's individual sample size and some assumption about the ratio of their variances.
Two options are reasonable in this situation: to treat the SEdiff as usable only within its own unequal-variance framework, if the review’s software or method supports direct SE-based pooling for that outcome, or to flag the study as unable to contribute a usable pooled SD to a standard mean-difference meta-analysis and to consider it for narrative synthesis instead5.
Application of the conversion in data extraction
In practice, reviewers should first determine whether the paper states which variance assumption underlies its reported SEdiff, adopting the equal-variances assumption by default only when the paper does not address the point and the comparison is described as a standard independent-samples analysis13. Once that is confirmed, the rearrangement above gives the pooled SD directly, with no further distributional conversion needed.
As with the other derived statistics in this cluster, the pooled SD should be labelled clearly as derived from a reported SE rather than from a directly reported SD, and the variance assumption used should be recorded alongside it in the extraction table, so that subsequent reviewers, or editors checking the review’s methods, can identify the assumption on which each included study’s derived SD rests, consistent with the transparency PRISMA 2020 expects of any methodological choice that affects the synthesis67.
A worked example with realistic values
Consider a paper that reports a mean difference of 2.4 units between two groups of n₁=25 and n₂=28, with a directly stated standard error of the difference of 0.95, described in the methods as derived from a standard independent-samples comparison. Applying the rearrangement directly gives SDpooled = 0.95 / √(1/25+1/28) = 0.95 / √(0.04+0.0357) = 0.95/√0.0757 = 0.95/0.2752 ≈ 3.451. This single step, without intermediate distributional conversion, constitutes the entire calculation once the equal-variances assumption has been confirmed as reasonable for this comparison.
Consider now a second paper that reports an identical SEdiff of 0.95 for the same two sample sizes but states explicitly that Welch’s correction for unequal variances was used. In this case the formula above cannot be applied as shown, because the reported SEdiff reflects each group's own separate variance divided by its own sample size rather than one shared pooled variance scaled by the combined sample sizes34. Applying the same rearrangement regardless would produce a value that resembles a plausible pooled SD but corresponds neither to the variance of either group nor to any meaningful combined quantity; this is the silent, difficult-to-detect error that the distinction is intended to prevent.
Relationship of this method to the other conversions in the series
Every other SD-estimation method in this cluster, whether starting from a p-value, a t-value, a z-score, or a confidence interval, ultimately recovers this same quantity, the standard error of the difference, before applying the final rearrangement shown here16. A p-value requires an inverse-normal step first; a confidence interval requires division by a critical value first; a t-value or z-score requires only its own one-line rearrangement first. When a paper reports SEdiff directly, none of those earlier steps is required, which is why this scenario is the simplest in the cluster and, in effect, the common end point of every other method in it.
It follows that, when two different statistics are reported for the same comparison (for example, both a p-value and a directly stated SE of the difference), the directly reported SE should be preferred, since it avoids an unnecessary conversion step and the approximation or distributional assumption that step would introduce25. As a general extraction principle across this cluster, reviewers should use the most directly reported statistic available, resorting to derivation methods such as those covered elsewhere only when no more direct statistic is reported.
This order of preference should be stated explicitly in a review’s data-extraction protocol, so that every reviewer on a team applies the same rule consistently: a directly reported SD first, a directly reported SE of the difference second, and an estimate derived from a p-value, t-value, z-score, or confidence interval only when neither of the first two is available. Documenting this hierarchy, and recording the level from which each included study’s figure was obtained, provides the methods section with a concrete, verifiable account of how much of the pooled estimate rests on directly reported rather than derived statistics, which is the transparency readers require to judge the confidence that the final pooled result warrants67. Recording the level from which each figure was obtained is a minor addition to an extraction table, but it meaningfully strengthens the overall methodological transparency of a review and requires little additional effort once the extraction has been completed carefully.
Key points
Key points
A directly reported standard error of the difference converts into a pooled SD with a single rearrangement, SD_pooled = SE_diff / sqrt(1/n1+1/n2), but only when the SE was calculated under the equal-variances assumption; an unequal-variances (Welch-type) SE does not decompose into one pooled SD this way.
Practical application
Practical application
- The variance assumption stated for the reported SE of the difference should be checked. The equal-variances assumption should be adopted by default only when the comparison is described as standard and the paper does not address the point.
- The formula SD_pooled = SE_diff / sqrt(1/n1+1/n2) should be applied directly. No distributional conversion step is needed beyond this single rearrangement.
- Where an unequal-variances (Welch-type) SE is confirmed, this formula should not be applied. The study should instead be identified for narrative synthesis or for a method that natively supports unequal-variance SEs.
- The variance assumption should be recorded alongside the derived SD in the extraction table. This practice keeps the methodological basis of every derived figure visible to subsequent reviewers.
Frequently asked questions
Yes, in the sense that it requires only one algebraic rearrangement; every other method in this cluster (from a p-value, t-value, z-score, or confidence interval) must first recover the SE of the difference through an additional step before reaching the same final formula.
It is a reasonable default when the comparison is described as a standard independent-samples analysis without further qualification, since that is the most common underlying method in applied biomedical research, but it remains an assumption that should be recorded as such.
Such a discrepancy suggests that the equal-variances assumption may not hold well in practice, even if the paper used it to calculate its reported SE. It should be flagged as a limitation of the derived pooled SD rather than treated as a reason to abandon the calculation, since the paper’s own analysis made the same assumption.
No. It recovers only a single pooled SD describing the combined spread implied by the reported SE of the difference and the two sample sizes; it cannot separate that combined figure back into two individual group SDs without further information.
References
- 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
- Altman DG, Bland JM. Standard deviations and standard errors. BMJ. 2005;331(7521):903. read 10 October 2026. View on The BMJ
- 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
- 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
- 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
- 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
- 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
- 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
- Mulrow CD. Rationale for systematic reviews. BMJ. 1994;309(6954):597-599. read 10 October 2026. View on The BMJ
- 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). Estimation of the Standard Deviation (SD) from the Standard Error of the Difference Between Means. Pubrica. https://pubrica.com/academy/estimation-of-standard-deviation-from-standard-error/
Pubrica Academy. “Estimation of the Standard Deviation (SD) from the Standard Error of the Difference Between Means.” Pubrica, 10 Oct. 2026, pubrica.com/academy/estimation-of-standard-deviation-from-standard-error/.
Pubrica Academy (2026) Estimation of the Standard Deviation (SD) from the Standard Error of the Difference Between Means. Pubrica. Available at: https://pubrica.com/academy/estimation-of-standard-deviation-from-standard-error/ (Accessed: 15 October 2026).
Pubrica Academy. “Estimation of the Standard Deviation (SD) from the Standard Error of the Difference Between Means.” Pubrica. October 10, 2026. https://pubrica.com/academy/estimation-of-standard-deviation-from-standard-error/
Pubrica Academy, ‘Estimation of the Standard Deviation (SD) from the Standard Error of the Difference Between Means’ (Pubrica, 10 October 2026) <https://pubrica.com/academy/estimation-of-standard-deviation-from-standard-error/> accessed 15 October 2026.
Pubrica Academy. Estimation of the Standard Deviation (SD) from the Standard Error of the Difference Between Means. [Internet]. Pubrica; 2026 Oct 10 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/estimation-of-standard-deviation-from-standard-error/
Pubrica Academy, “Estimation of the Standard Deviation (SD) from the Standard Error of the Difference Between Means,” Pubrica, Oct. 10, 2026. [Online]. Available: https://pubrica.com/academy/estimation-of-standard-deviation-from-standard-error/
@misc{/la_standard_deviation_estimation_from_standard_error_difference_2026,
author = {{Pubrica Academy}},
title = {Estimation of the Standard Deviation (SD) from the Standard Error of the Difference Between Means},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/estimation-of-standard-deviation-from-standard-error/},
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.
SE-of-Difference-to-SD Conversion Worksheet
This method applies when a source reports the standard error of the difference between two group means directly.
[ ] Checked whether the paper states its variance assumption for the reported SE of the difference [ ] Confirmed n1 and n2 for the two compared groups [ ] Computed SD_pooled = SE_diff / sqrt(1/n1+1/n2) [ ] Flagged any confirmed unequal-variances (Welch-type) SE as unsuitable for this formula [ ] Labelled the resulting SD as derived, noting the variance assumption used
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 review team confirms the variance assumption underlying every reported standard error of the difference before converting it to a pooled SD, preventing equal- and unequal-variance cases from being combined in a client’s meta-analysis.