Statistical significance (σ)
How unlikely an observed excess would be from background alone, expressed in Gaussian standard deviations.
A p-value is converted to an equivalent number of σ. 3σ (p ≈ 1.3 × 10⁻³) is called 'evidence', and 5σ (p ≈ 2.9 × 10⁻⁷, one-sided) is the conventional threshold for 'observation' or discovery. Examples include Daya Bay's θ₁₃ at 5.2σ, OPERA's ντ appearance at 6.1σ and IceCube's NGC 1068 at 4.2σ.
In plain words
When a detector sees more events than background alone predicts, the question is whether that excess is real or just bad luck. Significance answers it on a common scale: how many standard deviations of a bell curve the result lies from the background expectation. Two standard deviations happen by chance in a few percent of experiments; five happen roughly once in three and a half million.
From p-value to σ
- 2σ: p ≈ 2.3 × 10⁻² (one-sided)
- 3σ: p ≈ 1.3 × 10⁻³, conventionally 'evidence'
- 5σ: p ≈ 2.9 × 10⁻⁷, conventionally 'observation'
The σ scale is only a translation of the p-value; it does not assume that the data themselves are Gaussian. A quick estimate for a counting experiment with small signal s on large background b is Z ≈ s/√b.
Examples from neutrino physics
- March 2012: Daya Bay reports a non-zero θ₁₃ at 5.2σ; RENO follows weeks later at 4.9σ.
- 2017: COHERENT observes CEvNS at 6.7σ.
- 2018: OPERA completes its tau-neutrino appearance result with ten candidates at 6.1σ.
- 2020: Borexino detects CNO-cycle solar neutrinos at about 5σ.
- 2022: IceCube reports neutrinos from the galaxy NGC 1068 at 4.2σ, evidence but not yet discovery.
Why 5σ
The high bar protects against two dangers that the σ number alone does not capture: unaccounted systematic effects, and the look-elsewhere effect of having searched in many places. Several 3σ hints in neutrino physics have faded with more data, while the 6σ faster-than-light signal seen by OPERA in 2011 turned out to be an equipment fault. Significance measures surprise under a model, not truth.
Frequently asked questions
Is a 5σ result certainly correct?
No. 5σ means a statistical fluctuation is very unlikely. A hidden systematic error or a flawed background model can still produce a large significance, which is why independent confirmation matters.
Why do some results quote 'local' and 'global' significance?
The local value refers to one place in the search, the global value corrects for all the places that were examined. The global value is always smaller.
Original papers
- TheoryG. Cowan, K. Cranmer, E. Gross and O. Vitells, Asymptotic formulae for likelihood-based tests of new physics, Eur. Phys. J. C 71, 1554 (2011)DOI 10.1140/epjc/s10052-011-1554-0arXiv:1007.1727
Go deeper
- p-valueThe probability, assuming only background, of obtaining a result at least as extreme as the one observed.
- Look-elsewhere effectThe inflation of apparent significance when one searches many places (energies, sky directions, parameters).
Related terms
People
On the timeline
Mentioned in 15 pages
Lessons 1
Milestones 5
People 1
Last updated: · Editorial standard
Cite this page
Text
Neutrino Atlas (2026). “Statistical significance (σ)”. Retrieved 22 November 2026, from https://neutrino-atlas.com/terms/sigma-significance/BibTeX
@misc{neutrinoatlas2026termssigmasignificance,
title = {Statistical significance (σ)},
author = {{Neutrino Atlas}},
year = {2026},
url = {https://neutrino-atlas.com/terms/sigma-significance/},
note = {Accessed 2026-11-22},
language = {english}
}