Replace 'Nan-induced r=0, p=1' with nan to distinguish from true 0 cases

[kbpi314]

Statistically speaking, a r=0 and p=1 is different than r=nan