Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 96: Data, AI, Computing 7 (uncertainties, likelihoods)

T 96.4: Vortrag

Donnerstag, 7. März 2024, 16:45–17:00, Geb. 30.33: MTI

Combining data with unknown correlations — •Lukas Koch — JGU Mainz

The combination of data points is a regular necessity in particle physics: be it to calculate an "average" of multiple measurements of the same thing, or to do model tests and fits to one or more data sets with multiple data points each. Under ideal circumstances, the uncertainties and correlations between all data points -- i.e. the joint likelihood function -- is known. In that case, it is trivially possible to do "the right thing" and, e.g., use the Mahalanobis distance -- or "chi-squared" -- calculated with the known covariance matrix in order to do statistical tests with the expected properties. In reality, at least some of that information is missing, e.g. when there is no information about the correlation between the results from two separate experiments which share some systematics, or -- especially for older publications -- when there is no publicly available covariance matrices for an experimental result. Applying the M-distance under the assumption of no correlation can lead to undercoverage in this case. In this talk, I will present the use of alternative test statistics that behave conservative in these circumstances, and thus could be a more robust choice when faced with this issue.