: A statistical tool to unfold data distributions

Abstract

The paper advocates the use of a statistical tool dedicated to the exploration of data samples populated by several sources of events. This new technique, called s Plot, is able to unfold the contributions of the different sources to the distribution of a data sample in a given variable. The s Plot tool applies in the context of a Likelihood fit which is performed on the data sample to determine the yields of the various sources.

Observation ofB0→D¯0K+K−and Evidence forAaij1, Beteta2, Adametz3 et al. 2012Phys. Rev. Lett.292120View full textAdd to dashboardCite

“…Requirements are also placed on the corresponding variables for candidate composite particles ( " D 0 , B 0 ðsÞ ) together with restrictions on the consistency of the decay fit ( 2 vertex ), the flight distance significance ( 2 flight ), and the angle between the momentum vector and the line joining the PV to the B 0 ðsÞ vertex ( cos dir ) [24]. Further discrimination between signal and background categories is achieved by calculating weights for the remaining " D 0 þ À candidates [25]. The weights are used by the NEUROBAYES neural network package [26] to maximize the separation between categories.…”

“…Here the index i runs over all candidates in the fit range, W i is the signal weight for candidate i [25] from the fit shown in Fig. 2, and tot i is the efficiency for candidate i, which depends only on its Dalitz plot position.…”

Improved measurements of the branching fractions forB0→π+π−andB0→KAubert1, Barate2, Bóna3 et al. 2007Phys. Rev. D41130View full textAdd to dashboardCite

“…1 we compare the PDF shapes (solid curves) to the data using the event-weighting technique described in Ref. [17]. For each plot, we perform a fit excluding the variable being plotted and use the fitted yields and covariance matrix to determine the relative probability that an event is signal or background.…”

Measurements of the properties ofΛc(2595),Λc(2625),Aaltonen1, González2, Amerio3 et al. 2011Phys. Rev. D570320View full textAdd to dashboardCite

“…In the second step, we perform a dedicated selection of the four states under study. All neural networks are constructed with the NeuroBayes package [35,36] and trained, only using data, by means of the s Plot technique [37,38]. This technique assigns a weight to each candidate proportional to the probability that the candidate is signal.…”