npc() seems to add 1/B or 2/(B+1) to the p-values spuriously.
pvalues_from_distr[:, j] = 1 - rankdata(distr[:, j], method="min")/(plus1+B) + (1 + plus1)/(plus1+B)
When it is called from sim_npc with plus1=False, the P-value this would give to the largest value of the test statistic is
1-B/B + 1/B = 1/B, which is correct.
If it is called with plus1=True, the P-value for the largest test statistic is
1-B/(B+1) + 2/(B+1) = 3/(B+1), which is not correct.
For the smallest test statistic, when plus1=False,
1 - 1/B + 1/B = 1, which is correct.
When plus1=True,
1 - 1/(B+1) + 2/(B+1) > 1, which is wrong.
This is probably why Liptak was throwing an error.
Need to correct this for the plus1=True case; then the error test for Liptak on lines 179-181 can be deleted, I think.
npc() seems to add 1/B or 2/(B+1) to the p-values spuriously.
pvalues_from_distr[:, j] = 1 - rankdata(distr[:, j], method="min")/(plus1+B) + (1 + plus1)/(plus1+B)
When it is called from sim_npc with plus1=False, the P-value this would give to the largest value of the test statistic is
1-B/B + 1/B = 1/B, which is correct.
If it is called with plus1=True, the P-value for the largest test statistic is
1-B/(B+1) + 2/(B+1) = 3/(B+1), which is not correct.
For the smallest test statistic, when plus1=False,
1 - 1/B + 1/B = 1, which is correct.
When plus1=True,
1 - 1/(B+1) + 2/(B+1) > 1, which is wrong.
This is probably why Liptak was throwing an error.
Need to correct this for the plus1=True case; then the error test for Liptak on lines 179-181 can be deleted, I think.