Errors on Errors in Particle Physics

by Prof. Glen Cowan (Royal Holloway, University of London)

Friday 13 Feb 2026, 13:00 → 14:00 Europe/London

Bohr conference Room

A type of statistical model is described in which one assigns error bars to error bars, e.g., in a fit or a combination of measurements.  This is particularly relevant for systematic uncertainties whose values are assigned using an ad hoc procedure, such as variation of a renormalisation scale or the difference between two different analysis methods ("two-point systematics").  The new model changes the results of an analysis in interesting and useful ways:  confidence intervals for fitted parameters become sensitive to the internal compatibility of the data, and the influence of outliers is reduced compared to what is found with standard least-squares.  The model assumes that variances related to systematic uncertainties are not fixed constants but rather gamma-distributed estimates.  This is equivalent to replacing the usual Gaussian for measurement errors by a Student's t distribution, which has longer tails.  Applications related to some recent and ongoing measurements in Particle Physics are explored.