Influence of Implementation on the Properties of Pseudorandom Number Generators with a Carry Bit

We present results of extensive statistical and bit level tests on three implementations of a pseudorandom number generator algorithm using the lagged Fibonacci method with an occasional addition of an extra bit. First implementation is the RCARRY generator of James, which uses subtraction. The second is a modified version of it, where a suggested error present in the original implementation has been corrected. The third is our modification of RCARRY such that it utilizes addition of the carry bit. Our results show that there are no significant differences between the performance of these three generators.