Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 40: Experimental methods II
T 40.5: Talk
Tuesday, March 16, 2021, 17:00–17:15, To
Studies of parameter uncertainty correction methods in unbinned maximum likelihood fits to events weighted by the SPlot technique — Peter Buchholz, Mazuza Ghneimat, •Tim-Philip Hücking, Iskander Ibragimov, and Wolfgang Walkowiak — Universität Siegen, Siegen, Deutschland
The sPlot technique is used in HEP to e.g. separate signal and background events in control variable distributions on a statistical basis. This separation is achieved by applying sWeights as event weights. The sWeights are calculated based on a discriminating variable fit. In case of fitting the sWeighted distributions using an unbinned maximum likelihood approach, the covariance matrix, defined as the inverse Hessian matrix of the likelihood function, needs to be corrected.
Uncertainty correction methods proposed by Eadie et al. and Langenbruch are studied, utilizing RooFit. The studies are based on pseudo experiments employing a simple statistical model. The uncertainty correction method proposed by Eadie et al. is found to be sufficient only in few cases. Langenbruch proposed two uncertainty correction methods: The first one accounts for the presence of the sWeights in the fit and the second one additionally for the uncertainties of the sWeights. This second method is found to work best.
For certain distributions of the control variable, fits to sWeighted distributions may still show some problematic behavior, as observed for the case of two Gaussian distributions.