Stationary Alternatives Research Articles

Statistics, stationarity and random number generation

This paper remarks upon some issues involved in evaluating the "randomness" of numerical sequences. The question of how much to test is addressed, particularly with respect to pseudorandom generators. Historical failures of seemingly random sequences are noted. The dependence of evaluation programme upon proposed use of the sequence is stressed. The meaning and importance of stationarity are considered, and results from statistical distribution theory useful in checking for it, and in further evaluation of a sequence, are described. An example illustrates differences in power of three tests directed against a particular class of stationary alternatives to Normal white noise.