Bonn 2010 – wissenschaftliches Programm
T 76.4: Vortrag
Mittwoch, 17. März 2010, 14:50–15:05, JUR N
How good is your fit? — •Frederik Beaujean1, Allen Caldwell1, Daniel Kollár2, and Kevin Kröninger3 — 1Max-Planck-Institut für Physik, München — 2CERN — 3II. Physikalisches Institut, Universität Göttingen
The main goals of a typical data analysis are to extract the possible values of parameters within the context of a model including an estimate of the parameter uncertainties, and to draw conclusions on the validity of the model as a representation of the data.
The Bayesian Analysis Toolkit, BAT, is a C++ library developed to evaluate the posterior probability distribution for models and their parameters using Markov Chain Monte Carlo. This allows for straightforward parameter estimation, limit setting and uncertainty propagation.
In this talk we provide an introduction to the "goodness-of-fit" problem and show how to attack it using p values, both in the frequentist and Bayesian approach. We discuss common pitfalls in the use of p values and demonstrate BAT's capabilities to compute them. Various p value definitions are implemented; in addition users can easily define their own p value tailored to their specific problem.
The discussion will be illustrated by a real life physics example, the lifetime measurement of unstable particles.