Cited by 1,946Statistical Methods in Clinical TrialsStatistical Methods in EpidemiologyMeta-analysis and systematic reviews

Statistics notes: Absence of evidence is not evidence of absence

Open access
Altman, Douglas G, Bland, J Martin·1995·BMJ

Abstract

Randomized controlled clinical trials are conducted to determine whether differences of clinical importance exist between selected treatment regimens. When statistical analysis of the study data finds a P value greater than 5%, it is convention to deem the assessed difference nonsignificant. Just because convention dictates that such study findings be termed nonsignificant, or negative, however, it does not necessarily follow that the study found nothing of clinical importance. Subject samples used in controlled trials tend to be too small. The studies therefore lack the necessary power to detect real, and clinically worthwhile, differences in treatment. Freiman et al. found that only 30% of a sample of 71 trials published in the New England Journal of Medicine in 1978-79 with a P value greater than 10% were large enough to have a 90% chance of detecting even a 50% difference in the effectiveness of the treatments being compared, and they found no improvement in a similar sample of trials published in 1988. It is therefore wrong and unwise to interpret so many negative trials as providing evidence of the ineffectiveness of new treatments. One must instead seriously question whether the absence of evidence is a valid justification for inaction. Efforts must be made to look for quantification of an association rather than just a P value, especially when the risks under investigation are small. The authors cite a recent trial comparing octreotide and sclerotherapy in patients with variceal bleeding, as well as the overview of clinical trials evaluating fibrinolytic treatment for preventing reinfarction after acute myocardial infarction as examples.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1136/bmj.311.7003.485

Cites

5 of 5 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Null findings represent the absence of evidence rather than evidence of an absence (i.e., evidence of no effect).

    “null findings represent the absence of evidence rather than evidence of an absence (i.e., evidence of no effect; Altman & Bland, 1995).”

    From Does preregistration improve the credibility of research findings? (2020) · cites this paper as Altman & Bland, 1995

  • With classic null-hypothesis significance tests, a nonsignificant result does not provide support for the null hypothesis but only adds evidence that H0 cannot be rejected.

    “It should be pointed out, however, that with classic null-hypothesis significance tests (NHSTs), finding a nonsignificant result does not provide support for the null hypothesis (H0), but only adds evidence that H0 cannot be rejected (Altman & Bland, 1995).”

    From Impact of Police Academy Training on Recruits’ Integrity (2016) · cites this paper as Altman & Bland, 1995

  • Interpreting negative findings is challenging because absence of evidence is not evidence of absence.

    “In fact, interpreting negative findings is particularly challenging, because absence of evidence is not evidence of absence 93 .”

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Authors
Altman, D. G., & Bland, J. M.
Affiliations
St George's, University of London (GB)
Venue
BMJ
Publisher
BMJ
Published
1995-08-19
Volume
311
Issue
7003
Pages
485
ISSN
  • 0007-1447 (ISSN-L)
  • 0959-8138
  • 1468-5833
OpenAlex
W1525161423
Cited by
1,946
References
5
Citations by year
2026: 195 · 2025: 82 · 2024: 88 · 2023: 108 · 2022: 105 · 2021: 107 · 2020: 106 · 2019: 104 · 2018: 78 · 2017: 74 · 2016: 70 · 2015: 60 · 2014: 76 · 2013: 73 · 2012: 68
FWCI
15.94
Type
journal article
Language
en
Open access
Locations
  • BMJ (publishedVersion)
  • PubMed Central (publishedVersion)
  • PubMed (publishedVersion)
  • HighWire Press Open Archive (submittedVersion)
Topics
  • Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods in Clinical Trials
  • Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods in Epidemiology
  • Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Meta-analysis and systematic reviews
Keywords
statistical power, non-significant results, negative clinical trials, treatment ineffectiveness, clinical significance, p-value interpretation, sample size, octreotide, sclerotherapy, variceal bleeding, thrombolysis, acute myocardial infarction, reinfarction prevention
MeSH terms
Data Interpretation, Statistical, Statistics as Topic, Randomized Controlled Trials as Topic, Sample Size, Controlled Clinical Trials as Topic

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review