Goodness-of-fit Pitfalls and Power

Physicists frequently use various tests (especially chi-square) to test how well
a model describes data or how well two datasets agree with each other.
In this talk, a particular class of problems will be used to illustrate some of the
practical issues in performing such tests, and to compare the performance
of several tests. Special attention will be given to the low-statistics case, in
which specification of the null hypothesis can be problematic, leading to
erroneous results from "toy Monte Carlo" calculations.