The main idea behind the statistics investigated in this report is the following. In the multinomial case, one would compare the numbers of success runs of various lengths with the expected numbers of success runs of corresponding lengths, given a fixed total number of success runs. On the other hand, given a fixed number of digits, let us compare the observed numbers of success runs with the numbers expected conditional on the observed total number of success runs. This might result in a sort of "condi¬tional multinomial" distribution. It turns out that this idea works.
展开▼
