Creating and detecting specious randomness

Abstract

AbstractWe present a new test of non-randomness that tests both the lower and the upper critical limit of a $\chi ^{2}$ χ 2 -statistic. While checking the upper critical value has been employed by other tests, we argue that also the lower critical value should be examined for non-randomness. To this end, we prepare a binary sequence where all possible bit strings of a certain length occurs the same number of times and demonstrate that such sequences pass a well-known suite of tests for non-randomness. We show that such sequences can be compressed, and therefore are somewhat predictable and thus not fully random. The presented test can detect such non-randomness, and its novelty rests on analysing fixed-length bit string frequencies that lie closer to the a priori probabilities than could be expected by chance alone.