Practical Statistics for Particle Physics Analyses: Search for New Physics (4/4)

<!--HTML-->This will be a 4-day series of 2-hour sessions as part of CERN's Academic Training Course. Each session will consist of a 1-hour lecture followed by one hour of practical computing, which will have exercises based on that day's lecture. While it is possible to follow just the lectures or just the computing exercises, we highly recommend that, because of the way this course is designed, participants come to both parts. In order to follow the hands-on exercises sessions, students need to bring their own laptops. The exercises will be run on a dedicated CERN Web notebook service, SWAN (swan.cern.ch), which is open to everybody holding a CERN computing account. The requirement to use the SWAN service is to have a CERN account and to have also access to Cernbox, the shared storage service at CERN. New users of cernbox are invited to activate beforehand cernbox by simply connecting to https://cernbox.cern.ch. A basic prior knowledge of ROOT and C++ is also recommended for participation in the practical session. Day 4: Search for New Physics A large fraction of Particle Physics Publications are devoted to searches for New Physics. Such analyses can result in discovery (e.g. Higgs boson), exclusion (e.g. SUSY for a large range of parameters), or be ambiguous. This last talk deals with the statistical issues relevant to such analyses. Topics involved include: * p-values: Complaints about p-values; p-values and L-ratios; Critical values. * Blind analyses * Upper Limits, including CLs * The Look Elsewhere Effect * Why 5 sigma for discovery? The example of the Higgs discovery, and the measurement of its mass and spin-parity are discussed.