A Unified Approach to the Classical Statistical Analysis of Small
Signals, with Applications to Neutrino Oscillation Experiments
A Unified Approach to the Classical Statistical Analysis of Small
Signals, with Applications to Neutrino Oscillation Experiments
Prof. Robert Cousins, UCLA
In a recent paper by Gary Feldman and myself, we present a classical
confidence belt construction based on a likelihood-ratio ordering
principle. This solves several problems including: undercoverage caused
by using the data to decide whether to quote a one-sided or two-sided
limit; null or non-physical intervals for a non-negative Gaussian variable
such as a mass-squared; and null or non-physical intervals for a Poisson
mean when there is a downward fluctuation in the expected background. The
talk will begin with a reminder of Bayesian and classical intervals and
their differences, proceed with a discussion of the proposed method in the
prototypical cases, and end with application to the complex problem of
limit curves in neutrino oscillation experiments.