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 Exact P-Value

Estimation of the Standard Deviation from an Effect Estimate and Its P-Value

An exact two-sided p-value for a mean difference can be converted into a standard error, and subsequently into a pooled standard deviation (SD), by first converting the p-value into the z-score that would have produced it under a normal approximation.

Pubrica AcademyWritten by a biostatistics reviewer Reviewed bya meta-analysis methodologist Published10 October 2026 Last checked10 October 2026 Reading time7 minutes Sources10 peer-reviewed

When a source paper reports a mean difference between two groups with an exact p-value for that difference, but no SD, standard error, or confidence interval, the p-value contains sufficient information to recover an approximate SD. The method reverses the logic of a standard significance test: rather than calculating a p-value from a standard error, it calculates, from the p-value and the mean difference, the standard error that must have produced it, under the normal-approximation assumptions that the Cochrane Handbook sets out for this situation.

Learning objectives

  • The two-step calculation that converts a p-value and a mean difference into an estimated SD
  • Why this method requires an exact p-value, not a reported threshold such as 'p < 0.05'
  • The specific assumptions this method relies on, and when they do not hold

Common misconceptions

  • “A p-value reported as 'p < 0.05' carries the same information as an exact p-value for this purpose.” The method in the Cochrane Handbook requires the specific numeric p-value because it converts that exact figure into a specific z-score through the inverse normal distribution function; a threshold statement such as 'p < 0.05' only bounds the z-score rather than fixing it, and is therefore insufficiently precise for back-calculating a usable standard error.

The two-step conversion from p-value to standard deviation

The first step converts the exact, two-sided p-value into the z-score that a standard normal distribution would need to produce that exact p-value, using the inverse of the normal cumulative distribution function1:

z = Φ−1(1 − p/2)

The second step uses the z-score together with the reported mean difference (MD) to back-calculate the standard error of the difference and then the pooled SD, provided that the sample sizes of the two groups, n₁ and n₂, are also known1:

SE = MD / z
SDpooled = SE / √(1/n₁ + 1/n₂)

This sequence of conversions, from p-value to z-score, from z-score and mean difference to standard error, and from standard error to pooled SD, is the sequence the Cochrane Handbook sets out for this scenario, and every step depends on the specific numeric p-value reported rather than a rounded or thresholded version of it12.

The requirement for an exact p-value rather than a threshold

A paper that reports "p < 0.05" is telling a reader only that the p-value is somewhere below 0.05, not what its specific value is; the corresponding z-score for p = 0.05 is approximately 1.96, but a paper reporting p = 0.001 and one reporting p = 0.049 would both be truthfully described as "p < 0.05" despite corresponding to very different z-scores, and therefore very different implied standard errors13. Applying this conversion to a thresholded p-value, by substituting the boundary value as though it were the exact reported figure, introduces an error of unknown magnitude and direction, which is why the Cochrane Handbook explicitly limits this method to cases in which the exact p-value is stated1.

The companion paper by Altman and Bland on converting between p-values and confidence intervals makes the same point from the opposite direction: because the relationship between a p-value, an effect estimate, and its precision is exact, any step that loses precision (rounding, thresholding, or reporting only a significance category) breaks the chain needed to reconstruct the underlying standard error accurately4. Where only a thresholded p-value is available, reviewers should regard the p-value as offering no usable route to SD estimation and should look instead for another reported statistic, such as a confidence interval or test statistic, if one exists.

Assumptions underlying the conversion

This method assumes that the original significance test was based on a normal approximation (a z-test), or that the sample was large enough for the result of a t-test to approximate the normal case closely, since it applies the standard normal inverse function rather than a t-distribution inverse, which would also require the degrees of freedom1. For the smaller samples common in subgroup analyses, this assumption introduces some inaccuracy; the t-value-based conversion method addresses this limitation directly by using the t-distribution and the reported degrees of freedom instead of the normal approximation.

The method also assumes the reported p-value tested the mean difference between the same two groups whose sample sizes are being used in the final pooled-SD step, and that the comparison was a simple two-group test rather than an adjusted analysis (such as a regression coefficient's p-value), where the SE being recovered would reflect the adjusted model rather than the raw group-level spread5. Applying this conversion to the p-value of an adjusted analysis and treating the result as an unadjusted SD would misrepresent what the figure describes.

Application of the conversion in data extraction

In practice, reviewers should first confirm that the exact p-value is explicitly stated in the source, rather than inferred or rounded, and that the test referenced is a simple two-group comparison of the relevant mean difference15. Where a paper reports p = 0.000, a rounding artifact rather than a true zero, treating that figure as the exact value produces a nonsensical result (division by an undefined or near-infinite z-score); this case should therefore be flagged and, where possible, the source author contacted for the unrounded figure rather than an estimate substituted3.

As with every imputed SD in this cluster, the resulting figure should be labelled clearly as derived in the review’s extraction table. The PRISMA 2020 expectation that synthesis methods be reported transparently applies directly to such derived statistics, since a reader evaluating the pooled result needs to know which included studies’ SDs were reported directly and which were back-calculated from a p-value under these assumptions67.

A worked example with realistic values

Consider a trial with 30 participants per arm (n₁=n₂=30) that reports a mean difference of 4.2 units between groups with an exact two-sided p-value of 0.018, but no SD, SE, or CI. The first step converts p=0.018 into its implied z-score: z = Φ−1(1−0.018/2) = Φ−1(0.991) ≈ 2.371. The second step combines the z-score with the mean difference to obtain the standard error: SE = 4.2/2.37 ≈ 1.77. Finally, the pooled SD is obtained from the SE and the two sample sizes: SDpooled = 1.77 / √(1/30+1/30) = 1.77 / √(0.0667) = 1.77/0.258 ≈ 6.861.

Consider now a second trial that reports the same mean difference of 4.2 and the same sample sizes but a smaller exact p-value of 0.002, implying a larger z-score of approximately 3.09. The same two steps give SE = 4.2/3.09 ≈ 1.36 and SDpooled ≈ 5.27, a noticeably smaller implied SD than in the first example. This result is consistent with expectation: for the same observed mean difference and sample size, a smaller p-value (stronger evidence against no difference) corresponds to a smaller underlying SE and therefore to less variable underlying data14. This relationship provides a useful plausibility check: if a calculation produces a larger implied SD alongside a smaller p-value for an otherwise comparable study, the arithmetic and the extracted inputs should be re-checked before the result is accepted.

The limits of this plausibility check should also be noted: it confirms that the arithmetic is internally consistent, not that the originally reported values are accurate. If the p-value or mean difference in the source paper was mistyped during publication, the conversion will reproduce that error in the derived SD, and the conversion step alone provides no means of detecting it. This is a further reason to record the exact values extracted, together with their source (table number, page, or supplementary file), in the review’s extraction log, so that any later query about an unusual derived figure can be traced to the original paper rather than to recollection56.

Finally, where a study reports both an exact p-value and a confidence interval or t-value for the same comparison, the two derived SD estimates should agree closely. A discrepancy is informative in itself and should prompt a check of whether the two statistics describe the same comparison, the same two groups, and the same underlying test before either derived SD is admitted to a pooled analysis14.

Key points

Key points

An exact, numeric p-value for a two-group mean difference can be converted into a z-score, then a standard error, then a pooled SD, but only when the exact figure (not a threshold like 'p<0.05') is reported and the comparison was a simple two-group test.

Practical application

Practical application

  1. The source should be confirmed to report an exact numeric p-value rather than a threshold statement. A value such as ‘p<0.05’ cannot be converted reliably; only a specific figure such as p=0.023 supports this method.
  2. The exact p-value should be converted to a z-score using the inverse normal cumulative distribution function (CDF). z = Φ^-1(1 - p/2) for a two-sided test.
  3. The z-score should be combined with the reported mean difference to obtain the standard error, and then the pooled SD. SE = MD/z; SD_pooled = SE / sqrt(1/n1 + 1/n2).
  4. A reported value of p=0.000 should be identified as a rounding artifact rather than a usable exact value. The underlying unrounded figure should be sought from the source rather than treating the value of zero literally.

Frequently asked questions

No. This method requires the specific numeric p-value to calculate the exact z-score it implies; a significance category or threshold statement does not fix that value precisely enough to back-calculate a usable standard error.

This conversion assumes a normal-approximation test of a mean difference; a p-value from a non-parametric test such as the Mann-Whitney U test reflects a different underlying statistic and should not be converted using this method without adjustment.

The t-value method uses the t-distribution directly with the reported degrees of freedom, which is more accurate for smaller samples; this p-value method relies on the normal approximation and is most appropriate when the original test was itself based on that approximation or when sample sizes were reasonably large.

No; the formula shown here applies to a two-sided p-value. A one-sided p-value requires a different inverse-normal calculation (z = Φ^-1(1-p)), and the type reported should be confirmed before either version is applied.

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. 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
  3. 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
  4. Altman DG, Bland JM. Standard deviations and standard errors. BMJ. 2005;331(7521):903. read 10 October 2026. View on The BMJ
  5. 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
  6. 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
  7. 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
  8. 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
  9. Mulrow CD. Rationale for systematic reviews. BMJ. 1994;309(6954):597-599. read 10 October 2026. View on The BMJ
  10. 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

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 P-Value. Pubrica. https://pubrica.com/academy/estimation-of-standard-deviation-only-with-p-value/

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.

P-Value-to-SD Conversion Worksheet

This method applies when a source reports an exact p-value and mean difference but no SD, SE, or CI.

[ ] Confirmed the source reports an exact numeric p-value (not a threshold)
[ ] Confirmed the test compares the same two groups whose n1 and n2 you will use
[ ] Flagged p=0.000 as a rounding artifact if present; sought the unrounded value
[ ] Converted p to z using the two-sided inverse normal formula
[ ] Computed SE = MD/z, then SD_pooled = SE / sqrt(1/n1+1/n2)
[ ] 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 back-calculates SDs from exact p-values using the correct Cochrane Handbook formula sequence, and identifies thresholded or rounded p-values that cannot support a reliable conversion before they enter 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.

Precision of the reported p-value for reliable estimation of the standard deviation

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.