Contents
On this page
- The core formula for recovering the standard error from a confidence interval
- Selection of the critical value according to sample size
- Distinguishing a single-group interval from a between-group difference interval
- Application of the conversion in data extraction
- A worked comparison of large-sample and small-sample intervals
- Interpretation of an apparently asymmetric confidence interval
Academy · Research Data, Analytics & AI · SD from Confidence Intervals
Estimation of the Standard Deviation from an Effect Estimate and Its Confidence Interval
A reported confidence interval contains the standard error within its width, so recovery of a standard deviation (SD) from a confidence interval is principally a matter of dividing the width by the correct critical value: 1.96 for a large-sample 95% interval, or the appropriate t-distribution value for a smaller sample with specified degrees of freedom.
A confidence interval is constructed directly from a point estimate and its standard error; the width of the interval therefore encodes the standard error used to construct it. When a paper reports a mean difference with a 95% confidence interval but no separately stated SD or SE, the widely used method of Altman and Bland reverses that construction to recover the standard error and, from it, a pooled SD; the principal practical decision is which critical value corresponds to the way the interval was constructed.
Learning objectives
- The direct formula for recovering a standard error from the width of a reported confidence interval
- Why the correct critical value depends on sample size and which distribution the interval used
- The difference between a CI around a single group's mean and a CI around a between-group difference
Common misconceptions
- “Any 95% confidence interval can be converted to a standard error by dividing its width by 2x1.96.” That constant specifically assumes a large-sample interval built from the normal distribution; a confidence interval built using a t-distribution critical value for a small sample uses a larger multiplier that depends on the specific degrees of freedom, so applying the flat 1.96 constant to a small-sample CI understates the true standard error.
The core formula for recovering the standard error from a confidence interval
The upper and lower bounds of a confidence interval are constructed symmetrically around the point estimate as the estimate plus or minus a critical value multiplied by the standard error, so the full width of the interval (upper minus lower) equals exactly twice the critical value multiplied by the standard error12. Rearranging this relationship gives the standard error directly:
SE = (upper − lower) / (2 × critical value)
For a 95% confidence interval built using the normal approximation, appropriate for large samples, the critical value is approximately 1.96, giving the commonly cited shortcut SE = (upper−lower)/3.921. Once the SE has been recovered, conversion to a pooled SD for a between-group difference follows the relationship used throughout this cluster: SDpooled = SE / √(1/n₁ + 1/n₂)3.
Selection of the critical value according to sample size
The 1.96 constant is a large-sample, normal-distribution critical value, and the Cochrane Handbook states explicitly that a CI constructed for a smaller sample with a t-distribution instead requires the corresponding t critical value for the actual degrees of freedom of the comparison3. For a comparison with, say, 10 degrees of freedom, the appropriate two-sided 95% critical value from the t-distribution is larger than 1.96, meaning that applying the flat normal-distribution constant to a small-sample CI will produce an SE estimate, and therefore an SD estimate, that is too small13.
The guidance of Altman and Bland on moving between confidence intervals and p-values makes the same point: the numerical relationship between the width of an interval, its confidence level, and the distribution used to construct it is exact, and shortcuts that disregard the distribution and degrees of freedom introduce a systematic, rather than random, error into the recovered SE2. When extracting data from a small study, reviewers should look for the reported or implied degrees of freedom and use the corresponding t critical value rather than defaulting to 1.96 for convenience.
Distinguishing a single-group interval from a between-group difference interval
A further distinction must be made before any of this arithmetic begins: a confidence interval may be constructed around the mean of a single group or around the difference between the means of two groups, and the two cases use the standard error differently in the final SD calculation3. For the mean of a single group, the recovered SE relates to that group’s SD through SE = SD/√n, so SD = SE × √n for that group alone. For a between-group difference, the recovered SE is the standard error of the difference, and obtaining a pooled SD requires the sample sizes of both groups through the SDpooled = SE/√(1/n₁+1/n₂) relationship shown above13.
Misidentifying the type of interval reported, by treating a single-group CI as a between-group difference CI or vice versa, applies the wrong subsequent formula and produces an incorrect SD that nonetheless appears credible. The results text and table labels of the paper, specifically whether the stated quantity is a mean or a mean difference, should be checked before either formula is applied.
Application of the conversion in data extraction
In practice, reviewers should first identify what the CI surrounds (a single mean or a difference), then identify the corresponding sample size or sizes and the critical value appropriate to the confidence level reported (99% and 90% intervals use different critical values from a 95% interval, and the formula should be adjusted accordingly)13. Once these elements are confirmed, the SE and SD follow directly from the formulas above.
As with the other derived statistics in this cluster, this estimate should be labelled clearly in the review’s extraction table as derived rather than directly reported, and any study that relies on a small-sample, t-distribution-based CI, where the correction to the critical value matters most, is a reasonable candidate for a sensitivity analysis, consistent with the general expectation of transparency that PRISMA 2020 sets for any statistic entering a pooled synthesis through derivation rather than direct extraction45.
A worked comparison of large-sample and small-sample intervals
Consider a large trial that reports a mean difference of 5.0 units with a 95% CI of 3.2 to 6.8, with a total sample large enough for the normal-approximation critical value of 1.96 to be appropriate. The recovered SE is (6.8−3.2)/(2×1.96) = 3.6/3.92 ≈ 0.9181. If each group had n=100, the pooled SD is SDpooled = 0.918 / √(1/100+1/100) = 0.918/√0.02 = 0.918/0.1414 ≈ 6.49.
Consider now a much smaller pilot study with n₁=n₂=8 (14 degrees of freedom) that reports the same mean difference of 5.0 and an apparently identical 95% CI of 3.2 to 6.8. The correct critical value in this case is the two-sided 95% t-distribution value for 14 degrees of freedom, approximately 2.145, not 1.963. Using the correct value gives SE = 3.6/(2×2.145) = 3.6/4.29 ≈ 0.839, whereas applying 1.96 in error would give SE ≈ 0.918, overstating the implied standard error and understating the precision that this small study reported. This direction of error is the opposite of what might be intuitively expected, and the example illustrates that the critical value matched to sample size changes the answer by more than a rounding amount13.
Interpretation of an apparently asymmetric confidence interval
The formulas above assume a symmetric confidence interval, where the point estimate sits exactly midway between the lower and upper bounds, which is the standard case for a mean or mean difference on its natural scale1. Some effect measures, most commonly odds ratios, risk ratios, and hazard ratios, are symmetric on a logarithmic scale rather than on their reported natural scale; their confidence intervals therefore appear asymmetric around the point estimate in a results table, even though the underlying calculation was entirely standard3.
For a mean difference, an interval that appears genuinely asymmetric warrants investigation before this conversion is applied, since it may indicate that the interval was constructed by a method other than the standard symmetric approach, for instance a bootstrap confidence interval, which does not relate to the standard error through the same fixed critical-value formula2. Checking the method the paper states for constructing its interval, rather than assuming that the standard formula applies, prevents a mismatched conversion from entering a pooled analysis.
A simple arithmetic check, applicable regardless of the method used to construct the interval, is to confirm that the point estimate lies approximately midway between the reported lower and upper bounds. For a standard symmetric interval, it should lie almost exactly at the midpoint, and any meaningful departure is evidence that the interval was not constructed in the standard way, which should prompt a review of the paper’s stated methodology before this conversion is applied12. This check is quick and identifies a meaningful proportion of mismatched or misread confidence intervals before they are carried through to a derived SD. Together with confirmation of the interval’s stated confidence level and construction method, these checks address most of the practical sources of error in this conversion.
Key points
Key points
A confidence interval's width already encodes its standard error; dividing that width by twice the correct critical value recovers the SE, but the critical value must match the interval's actual confidence level and sample size (1.96 for large-sample 95% CIs, a larger t-distribution value for smaller samples).
Practical application
Practical application
- The quantity that the confidence interval surrounds, a single group mean or a between-group difference, should be identified. The two cases lead to different subsequent SD formulas.
- The confidence level and the appropriate critical value should be identified, rather than a default of 1.96 assumed. A small-sample interval constructed with a t-distribution requires the matching critical value for its degrees of freedom.
- The standard error should be computed as SE = (upper-lower) / (2 x critical value). This is the direct rearrangement of the construction of the interval.
- The SD should be derived using the formula appropriate to the interval type (single group or difference). SD = SE x sqrt(n) for a single group; SD_pooled = SE / sqrt(1/n1+1/n2) for a difference.
Frequently asked questions
It is an approximation that becomes progressively less accurate as sample size decreases, because a small-sample interval is more likely to have been constructed with a t-distribution critical value larger than 1.96; where the degrees of freedom are known or can be inferred, the corresponding t critical value is more accurate.
The method still applies, but the critical value changes to match that confidence level (for example, approximately 1.645 for a large-sample 90% interval, or approximately 2.576 for a large-sample 99% interval) and must be substituted into the same formula.
The paper’s labelling and the accompanying point estimate should be checked: a CI reported alongside the mean value of a single group is a single-group interval, whereas one reported alongside a stated difference between two groups is a between-group difference interval, and each requires a different formula to obtain the SD.
If a paper reports the SE directly, this conversion is unnecessary; the method applies specifically when only the bounds of the CI, and not the SE itself, are stated, and it reconstructs the SE from which those bounds were derived.
References
- 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
- Altman DG, Bland JM. Standard deviations and standard errors. BMJ. 2005;331(7521):903. read 10 October 2026. View on The BMJ
- 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
- 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
- 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
- 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
- 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
- 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
- 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 from an Effect Estimate and Its Confidence Interval. Pubrica. https://pubrica.com/academy/estimation-of-standard-deviation-only-with-confidence-intervals/
Pubrica Academy. “Estimation of the Standard Deviation from an Effect Estimate and Its Confidence Interval.” Pubrica, 10 Oct. 2026, pubrica.com/academy/estimation-of-standard-deviation-only-with-confidence-intervals/.
Pubrica Academy (2026) Estimation of the Standard Deviation from an Effect Estimate and Its Confidence Interval. Pubrica. Available at: https://pubrica.com/academy/estimation-of-standard-deviation-only-with-confidence-intervals/ (Accessed: 15 October 2026).
Pubrica Academy. “Estimation of the Standard Deviation from an Effect Estimate and Its Confidence Interval.” Pubrica. October 10, 2026. https://pubrica.com/academy/estimation-of-standard-deviation-only-with-confidence-intervals/
Pubrica Academy, ‘Estimation of the Standard Deviation from an Effect Estimate and Its Confidence Interval’ (Pubrica, 10 October 2026) <https://pubrica.com/academy/estimation-of-standard-deviation-only-with-confidence-intervals/> accessed 15 October 2026.
Pubrica Academy. Estimation of the Standard Deviation from an Effect Estimate and Its Confidence Interval. [Internet]. Pubrica; 2026 Oct 10 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/estimation-of-standard-deviation-only-with-confidence-intervals/
Pubrica Academy, “Estimation of the Standard Deviation from an Effect Estimate and Its Confidence Interval,” Pubrica, Oct. 10, 2026. [Online]. Available: https://pubrica.com/academy/estimation-of-standard-deviation-only-with-confidence-intervals/
@misc{/la_standard_deviation_estimation_from_confidence_intervals_2026,
author = {{Pubrica Academy}},
title = {Estimation of the Standard Deviation from an Effect Estimate and Its Confidence Interval},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/estimation-of-standard-deviation-only-with-confidence-intervals/},
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.
CI-to-SD Conversion Worksheet
This method applies when a source reports a confidence interval for a mean or mean difference but no SD or SE.
[ ] Identified whether the CI surrounds a single group's mean or a between-group difference [ ] Confirmed the stated confidence level (90%, 95%, 99%) and matched the correct critical value [ ] Checked whether the CI was built using a normal (large-sample) or t-distribution (small-sample) critical value [ ] Computed SE = (upper-lower) / (2 x critical value) [ ] Converted to SD using the matching formula for a single group or a difference [ ] Labelled the resulting SD as derived in the extraction table
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 matches each reported confidence interval to its correct critical value and interval type before converting it to a usable SD, avoiding the systematic understatement that the 1.96 approximation produces in smaller studies.