Collider physics measurements follow a systematic methodology: define a signal process, identify backgrounds, design event selection criteria (cuts or multivariate classifiers) to maximize signal significance, estimate backgrounds from data-driven methods or simulation, and extract the signal through fits to discriminating distributions. Statistical methods (hypothesis testing, confidence intervals, profile likelihood) quantify the significance of observations and the precision of measurements.
Collider physics analysis is the methodology for extracting physics results from the millions of collision events recorded by particle detectors. The process begins with a trigger -- a real-time selection that reduces the event rate from ~1 billion collisions per second to a few thousand events per second that are recorded to disk. Trigger selections must be efficient for the physics of interest while rejecting the overwhelming rate of soft QCD events.
Event reconstruction converts raw detector signals into physics objects: electrons, muons, photons, jets, and missing transverse energy (from neutrinos or other invisible particles). Each object type has specific identification criteria (isolation, shower shape, track quality) and calibrations. The performance of object reconstruction -- efficiency, fake rate, energy/momentum resolution -- is measured in data using standard candle processes (Z -> ll, J/psi -> mu mu, W -> e nu) and parameterized for use in the analysis.
The core of any analysis is the signal extraction strategy. Analysts define selection criteria (cuts on kinematic variables, or more commonly, multivariate classifiers trained on simulated signal and background) to enhance the signal-to-background ratio. The remaining background is estimated using data-driven methods in control regions or from validated simulations. The signal yield is then extracted by fitting a discriminating distribution (invariant mass, BDT output, neural network score) in the signal region, typically using a binned or unbinned maximum likelihood fit. Systematic uncertainties -- from jet energy scale, luminosity, PDF choices, theoretical cross sections, and many other sources -- are included as nuisance parameters in the fit.
Statistical interpretation follows the CLs method or Bayesian framework. For discovery, the test statistic is the profile likelihood ratio comparing signal+background to background-only hypotheses, and the significance is quoted in units of sigma. For upper limits (when no signal is observed), the CLs method provides 95% confidence level upper bounds on the signal cross section. For parameter measurements, profile likelihood scans or Bayesian posteriors give confidence intervals. The statistical tools (RooFit, RooStats, pyhf) are shared across experiments and embody decades of experience in handling the complex likelihood models of modern particle physics.