This lecture series by Louis Lyons (Imperial College) and Lorenzo Moneta (CERN) will cover the statitiscal methods used in particle physics, and is targeted at graduate students beginning work in particle physics fields including high energy physics, neutrino physics and low background particle physics. The series will consist of 6 sessions, each with a 75 minute lecture and a 60 minute computing practical. This is an updated version of the Academic Training Course given by Lyons and Moneta at CERN in November 2016.
The outline of the lecture series is:
Session 1: Review of introductory topics
What is Statistics? Comparison of Statistics and Probability.
Conditional probability.
Why are uncertainties important?
Combining uncertainties and results.
Improved estimates by incorporating theory.
Probability distributions: Binomial, Poisson and Gaussian
Session 2: Likelihoods
How likelihoods work.
Examples including the mass peak and exponential decay.
Parameter uncertainty estimates and coverage
Multiple parameters
Extended maximum likelihood
Unbinned likelihood
Session 3: χ2 and Goodness of Fit
Introduction
Pearson and Neyman χ2
χ2 with correlated uncertainties for 1-D and 2-D parameter(s)
Goodness of fit and number of degrees of freedom
Other GoF tests including Kolmogorov-Smirnov and Anderson-Darling
Session 4: Bayes and Frequentism
What is 'Probability'?
Bayesian sensitivity to the prior
Bayes and Frequentism for exponential decay
Resulting statements from Bayes and Frequentist approaches
Treatment of systematic uncertainties
Session 5: Search for New Physics
Discovery, exclusion or neither?
p-values
Blind analyses
Upper Limits: CLs, p0 versus p1 plots, Different methods
Discovery criteria
Examples from Higgs search: Local p-values, Higgs mass, Spin-parity
Session 6: Miscellaneous
Learning to love the covariance matrix
Neural networks as example of multivariate analysis.
Questions and discussion