26–29 Jun 2007
CERN
Europe/Zurich timezone

Event Weighting with near-optimal variance for background-subtracted data

Not scheduled
1m
CERN

CERN

Speaker

James Linnemann (Michigan State University)

Description

It is possible to find event weighting schemes which produce parameter estimates with variance nearly the same as a ML estimate. But there are situations in which a full ML estimate is inconvenient, usually for computational reasons (iteration over large data sets for example). If an variable x associated with the events is a candidate discriminating variable (that is, its distribution for signal and background differ, so that s(x) is not equal to b(x)), a weight function can be defined using s(x) and b(x) which allows estimation of a signal fraction or a number of signal events in a sample with a variance approaching that of a maximum likelihood estimate of the same quantity. In the case in which there is an external estimate of the amount of background in the sample, this is also possible, with improved variance. We derive a formula for this case and discuss it in the context of more general results on event weighting from earlier papers by Barlow and by Tkachov. The specific context came from gamma ray astronomy but the techniques may also be of interest at the LHC.

Author

James Linnemann (Michigan State University)

Co-author

Dr Andrew Smith (University of Maryland)

Presentation materials

There are no materials yet.