Statistical tests of the IBM PC pseudorandom number generator

The widespread use of IBM personal computers in biomedical research and medical practice is evidenced by a variety of journal articles. Many of the application programs are written in the BASIC language, they model a stochastic process, and they depend on the Microsoft random number generator. Unfortunately, the Microsoft generator has exhibited a number of flaws including, for some seeds, a lack of uniformity of generated sequences of numbers, and serial correlation within such sequences. This paper provides evidence of these problems for several seeds, and provides suggestions for acceptable seeds which ensure that the generated sequences of numbers pass two fundamental statistical tests for randomness.