Re: [Maxima-discuss] "distrib" package

[Soegtrop, M. <mic...@in...>]

Dear Henry,

I thought about it a bit over lunch. In case we are talking about sets of discrete samples, I think the Mann Whitney Wilcoxon test is pretty close to what you want. Assuming you have a sample S drawn from distribution A and a sample T drawn from distribution B, it gives you the probability that the true median of A is higher than the true median of B, without making any assumptions on the distribution. The only difference to what you want is that it compares the median, not the mean. If I remember right, the MMW test starts to give useful results for 2x5 or more samples. Below this, the effectiveness will be low (that is the answer will usually be close to 50%).

Maxima seems to have a function for it (test_rank_sum).

Best regards,

Michael