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

Start one
Contents

Academy · Research Data, Analytics & AI · P Values vs Confidence Intervals

The Relationship Between P Values and Confidence Intervals: Mathematical Correspondence and Consistency Checks

A two-sided P value and a confidence interval calculated from the same model and the same data are mathematically linked rather than independent pieces of evidence. Consequently, a 95% interval that excludes the null value corresponds exactly to a P value below 0.05 for the same comparison.

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

A P value and a confidence interval, when both are calculated from the same statistical model applied to the same data, are directly related mathematically rather than being two separate, independently informative results. A 95% confidence interval that excludes the null hypothesis value (commonly zero for a difference, or one for a ratio) corresponds exactly to a two-sided P value below 0.05 for the same comparison, and an interval that includes the null value corresponds to a P value of 0.05 or higher1. The two statistics address closely related questions using the same underlying standard error, but they present that information in different forms; an understanding of how they are linked prevents the common error of treating a mismatch between them as evidence that an analysis has gone wrong.

Learning objectives

  • The exact mathematical correspondence between a 95% confidence interval and a two-sided P value at 0.05
  • Why a P value and an interval calculated from the same data should never genuinely ‘disagree’
  • The additional information conveyed by the confidence interval that the P value alone does not provide

Common misconceptions

  • “A P value and a confidence interval are two independent checks that can validly disagree.” When calculated consistently from the same model and data, a confidence interval and its corresponding P value are derived mathematically from the same test statistic and standard error; an apparent disagreement between them (for example, a ‘significant’ P value alongside an interval that includes the null) almost always indicates a calculation, rounding or reporting inconsistency rather than two genuinely independent pieces of evidence pointing in different directions.

Both a confidence interval and a two-sided P value are constructed from the same test statistic, the point estimate divided by its standard error, compared against the same underlying sampling distribution1. A 95% confidence interval is defined as the range of null-hypothesis values that would not be rejected by a two-sided test at the 0.05 significance level; by construction, therefore, the interval excludes the null value if and only if the same test produces a P value below 0.051. This correspondence holds for any pair of significance level and confidence level that sum to 100% (for example, a 90% confidence interval corresponds to a two-sided P value threshold of 0.10), not because the two statistics happen to agree, but because one is derived mathematically as the inversion of the other2.

Apparent mismatches as indicators of error

Because the two statistics are mathematically linked rather than independently estimated, a reported result showing a “significant” P value below 0.05 alongside a 95% confidence interval that includes the null value (or the reverse) is not evidence of conflicting signals in the data; it indicates that an inconsistency has been introduced at some point, whether in the calculation, in rounding, in the specific model used for each statistic or in a transcription step1. Greenland and colleagues’ guide to common statistical misinterpretations identifies the treatment of P values and confidence intervals as separate, potentially disagreeing sources of evidence as a conceptual error, emphasizing that any genuine discrepancy between a correctly calculated P value and its correspondingly calculated confidence interval should prompt a check of the methods rather than a substantive interpretation3. A frequent practical cause of apparent mismatches is the use of different statistical methods, for example a normal approximation for the interval and an exact test for the P value, which can produce small numerical inconsistencies near the significance boundary even though both statistics are legitimately calculated.

Information conveyed by the confidence interval but not by the P value

Although the two are mathematically linked with respect to whether the null value falls inside or outside the interval, the confidence interval conveys substantially more information than the single P value, because it displays the entire range of effect sizes compatible with the data, not only whether the null value is among them2. Du Prel and colleagues, writing for clinical researchers, note that two results with an identical P value can have very different confidence intervals, one narrow and centred close to a clinically trivial effect and another wide and spanning a clinically important range; the shared P value alone conceals this information entirely4. This is the principal practical reason why methodological guidance from Cochrane, the American Statistical Association (ASA) and clinical epidemiology consistently recommends reporting both statistics together rather than substituting one for the other.

Back-calculation of a confidence interval from a published P value

The mathematical link between the two statistics is sufficiently close that Altman and Bland published a practical method, widely used by systematic reviewers, for deriving an approximate 95% confidence interval directly from a study’s reported P value and point estimate when the original authors did not report an interval5. This back-calculation is possible only because the two statistics are not independent: the method inverts the same test-statistic relationship that connects a P value to its corresponding interval. Meta-analysts commonly rely on this technique when pooling older trials that reported only a P value, which is itself indirect evidence of how closely the two statistics are linked, since a meaningful interval could not be reconstructed in this way from a P value if the two estimated unrelated quantities.

Implications for meta-analysis reporting

In meta-analysis, the relationship between a pooled P value and a pooled confidence interval holds at the level of the pooled estimate, but it does not automatically extend to the individual P values and intervals of the component studies, which is a frequent source of confusion2. It is entirely possible, and not contradictory, for every included study to report a non-significant P value while the pooled meta-analysis produces a confidence interval that excludes the null value and a correspondingly significant pooled P value, because pooling increases the effective sample size underlying the combined estimate2. Reviewers unfamiliar with this property sometimes incorrectly identify a meta-analysis’s significant pooled result as inconsistent with its “all non-significant” individual studies, when this pattern is precisely what a meta-analysis pooling underpowered individual studies is designed to resolve.

Minor discrepancies arising from exact and approximate methods

In practice, a small, genuine divergence between a reported P value and its corresponding confidence interval can arise without any error when the two statistics are calculated using different underlying methods: a normal approximation for one and an exact method, such as an exact binomial calculation for a small number of events, for the other2. Exact methods are generally considered more reliable for small samples or rare events, for which the normal approximation can perform poorly; when a discrepancy appears near a significance boundary, therefore, verifying whether the P value and the interval were calculated using the same approach (both exact, or both based on the same normal approximation) is an appropriate first step before concluding that the underlying evidence is contradictory3.

A reviewer's checklist for detecting inconsistencies

A peer reviewer or statistical editor checking a manuscript for this type of inconsistency can perform a brief, mechanical check: take the reported point estimate and its confidence interval, back-calculate the implied P value using the method described by Altman and Bland, and compare the back-calculated value with the P value reported separately by the authors4. A close match confirms internal consistency; a substantial mismatch warrants a direct query to the authors before publication, since it frequently arises from a transcription error, from the undisclosed use of different statistical tests for the two figures or, less commonly, from a genuine error in one of the two calculations that would otherwise remain unnoticed by readers who do not perform this cross-check.

A worked numerical illustration demonstrates this check. Suppose that a manuscript reports an odds ratio of 1.8 with a 95% confidence interval of 1.05 to 3.08 and separately states a P value of 0.03 for the same comparison; back-calculating the implied P value from the point estimate and interval using Altman and Bland’s method should produce a value close to 0.03, which confirms internal consistency5. If the manuscript instead reported a P value of 0.11 alongside the same interval excluding the null value of 1.0, the mismatch would be immediately evident from this single calculation and would warrant a direct query to the authors before publication, since an interval that excludes the null value at the 95% level cannot legitimately be paired with a two-sided P value above 0.05 from the same underlying model1. The same cross-check should be applied not only to a manuscript’s primary outcome but to every secondary and subgroup result reported in a results table, since a single transcription error, for example copying the wrong row from a statistical output file into a typeset table, is considerably more likely to occur in a less closely scrutinized secondary outcome than in the principal result that authors, editors and reviewers have already checked several times4.

Key points

Key points

A P value and a confidence interval calculated from the same model and data are mathematically linked rather than independently informative, so an apparent disagreement between them indicates a calculation or reporting inconsistency to be checked rather than a genuine conflict in the evidence. This link also means that a confidence interval can be back-calculated from a P value and used as a reviewer's cross-check, and that a meta-analysis's pooled result can legitimately be significant even when no individual included study was.

Practical application

Practical application

  1. Treat an apparent mismatch between a P value and its interval as a prompt to check methods, not as a finding. If a significant P value and an interval that includes the null appear together, the model, rounding and calculation method used for each should be rechecked before the discrepancy is reported as real.
  2. Use consistent methods to calculate both statistics. Mixing a normal-approximation interval with an exact-test P value, or similar combinations, should be avoided, since this is a common and avoidable source of inconsistency near the significance boundary.
  3. Report the interval even when the P value alone would suffice for a significance claim. The width and position of the interval convey information on clinical relevance that the P value cannot, so its omission discards substantive information even when both point to the same conclusion about significance.

Frequently asked questions

Not when both are calculated correctly and consistently from the same model and data at corresponding levels (e.g., P < 0.05 and a 95% interval); if this appears to occur, it indicates a calculation or methodological inconsistency in the derivation of the two statistics, not two genuinely different conclusions from the same evidence.

Because the link guarantees agreement only on the single question of whether the null value falls inside the interval; the width and exact bounds of the interval convey additional information about the range of plausible effect sizes that the P value, as a single number, cannot communicate.

References

  1. du Prel JB, Hommel G, Rohrig B, Blettner M. Confidence interval or p-value? Part 4 of a series on evaluation of scientific publications. Deutsches Arzteblatt International. 2009;106(19):335-339. read 10 October 2026. View on Deutsches Arzteblatt International
  2. Greenland S, Senn SJ, Rothman KJ, Carlin JB, Poole C, Goodman SN, Altman DG. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. European Journal of Epidemiology. 2016;31(4):337-350. read 10 October 2026. View on European Journal of Epidemiology
  3. Gardner MJ, Altman DG. Confidence intervals rather than P values: estimation rather than hypothesis testing. BMJ. 1986;292(6522):746-750. read 10 October 2026. View on The BMJ
  4. 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
  5. Deeks JJ, Higgins JPT, Altman DG (editors). Chapter 10: Analysing data and undertaking meta-analyses. In: Cochrane Handbook for Systematic Reviews of Interventions. Cochrane, 2024. read 10 October 2026. View on Cochrane Training
  6. Wasserstein RL, Lazar NA. The ASA Statement on p-Values: Context, Process, and Purpose. The American Statistician. 2016;70(2):129-133. read 10 October 2026. View on Taylor & Francis Online
  7. DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials. 1986;7(3):177-188. read 10 October 2026. View on ScienceDirect
  8. Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Statistics in Medicine. 2002;21(11):1539-1558. read 10 October 2026. View on PubMed
  9. Riley RD, Higgins JPT, Deeks JJ. Interpretation of random effects meta-analyses. BMJ. 2011;342:d549. read 10 October 2026. View on The BMJ
  10. Schulz KF, Grimes DA. Sample size calculations in randomised trials: mandatory and mystical. The Lancet. 2005;365(9467):1348-1353. read 10 October 2026. View on The Lancet

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). The Relationship Between P Values and Confidence Intervals: Mathematical Correspondence and Consistency Checks. Pubrica. https://pubrica.com/academy/relationship-between-confidence-intervals-and-p-values/

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 / Confidence Interval Consistency Check

This checklist is intended for use when a P value and a confidence interval reported together appear to disagree.

[ ] Confirmed both statistics were calculated from the same statistical model
[ ] Checked whether the same approximation method (normal, t, exact) was used for both
[ ] Rechecked rounding at each calculation step
[ ] Recalculated the P value directly from the interval's bounds as a cross-check
[ ] If a genuine inconsistency remains, treated it as a methods issue to resolve before reporting

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 cross-checks reported P values against their corresponding confidence intervals for internal consistency before a manuscript goes to peer review.

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.

Pubrica can verify whether reported P values and confidence intervals are internally consistent.

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.