Goodness-of-Fit: Why does it work so well?

Fred James (CERN) - INFN Genova, 14 April 2003

Of all the statistical tools in the physicist's toolbox, the oldest and most often used is Pearson's Chi-square Test. This is just one member of a class known in statistics as goodness-of-fit (gof) tests. Gof is the part of statistics that is the most successful in practice but has the weakest theoretical foundation. In the Bayesian framework, it is even "illegal". In the classical framework, there is a major problem because the most important property of the test -- its power -- is undefined.  The people who use it (correctly and successfully) are often unable to explain what the results mean. We will try to clarify the meaning.


Last update 18 May 2003 - Maria Grazia Pia