Monkeying with the goodness-of-fit test

Abstract

The familiar Σ(OBS - EXP)2/EXP goodness-of-fit measure is commonly used to test whether an observed sequence came from the realization of n independent identically distributed (iid) discrete random variables. It can be quite effective for testing for identical distribution, but is not suited for assessing independence, as it pays no attention to the order in which output values are received. This note reviews a way to adjust or tamper, that is, monkey-with the classical test to make it test for independence as well as identical distribution - in short, to test for both the i's in iid, using monkey tests similar to those in the Diehard Battery of Tests of Randomness (Marsaglia 1995).