October 6th, 2004, 7:22 pm
The example I always give is imagine you wanted a sample of good students (the new random variable you wish to simulate) out of a large group (some other random variable you know how to simulate already). One way of doing it is to give them an exam (your test statistic), if they pass, you take them; if they fail, you reject them. I think that's intuitive way of understanding how it works; and the proof of why the rejection algorithm would always give you the desired distribution can be found in almost any textbook on simulation.