Neutrino Atlas
EN
  1. Terms
  2. Analysis & measurement
  3. Statistics & inference
  4. Confidence level (CL)
  5. Feldman–Cousins method

Feldman–Cousins method

ConceptLevel 4 of 4Also called: unified approach

A frequentist method for building confidence intervals that moves smoothly from upper limits to two-sided intervals.

Proposed in 1998 for small-signal problems in neutrino oscillation searches, it orders outcomes by a likelihood ratio, avoiding empty or non-physical intervals and the 'flip-flopping' between limit types. It is widely used for oscillation contours and rare-event limits.

In plain words

Imagine a search for a tiny signal on top of background. Before seeing the data, many physicists used to decide: if the signal looks significant, quote a two-sided interval; if not, quote an upper limit. Making that choice after looking at the data quietly breaks the promised confidence level. The Feldman–Cousins method gives one recipe that decides automatically and keeps the promise.

How it works

For every possible true signal μ, the method ranks the possible outcomes n by a likelihood ratio and adds them to the acceptance region in that order until the desired probability is reached:

R = P(n | μ) / P(n | μbest)μbest is the physically allowed value that best fits n, for a counting experiment max(0, n − b) with known background b.

The confidence interval is then the set of μ whose acceptance region contains the observed n. Small counts automatically give an upper limit, large counts a two-sided interval, and the result is never empty.

Example

With an expected background of 3 events and 0 observed, the classical recipe for a 90% upper limit gives a negative number. Feldman–Cousins gives a sensible interval from 0 to 1.08 signal events.

History and use

Gary Feldman and Robert Cousins published the 'unified approach' in Physical Review D in 1998, with neutrino oscillation searches as a central example. It became standard for oscillation contours, where parameters such as sin²2θ sit against the physical boundary at zero, and is used by T2K and NOvA. Because it requires many simulated pseudo-experiments for each parameter point, it is computationally expensive. Experiments that observe less background than expected can obtain unusually strong limits, so the authors recommended quoting sensitivity alongside.

Frequently asked questions

Is Feldman–Cousins frequentist or Bayesian?

Frequentist. It is a particular ordering rule within Neyman's construction and guarantees correct coverage without a prior.

Why is it called the unified approach?

Because one procedure covers both upper limits and two-sided intervals, removing the need to choose between them after looking at the data.

Original papers

  1. TheoryG. J. Feldman and R. D. Cousins, A Unified approach to the classical statistical analysis of small signals, Phys. Rev. D 57, 3873 (1998)DOI 10.1103/PhysRevD.57.3873arXiv:physics/9711021

Mentioned in 8 pages

Last updated: · Editorial standard

Cite this page

Text

Neutrino Atlas (2026). “Feldman–Cousins method”. Retrieved 22 November 2026, from https://neutrino-atlas.com/terms/feldman-cousins/

BibTeX

@misc{neutrinoatlas2026termsfeldmancousins,
  title        = {Feldman–Cousins method},
  author       = {{Neutrino Atlas}},
  year         = {2026},
  url          = {https://neutrino-atlas.com/terms/feldman-cousins/},
  note         = {Accessed 2026-11-22},
  language     = {english}
}