When to use approximation for ROC p-value

[Peter]

Documentation says:

If the smallest group size is larger than minimum_size (default = 10), then P is calculated using a normal approximation.

whereas the code says:

bool ROC::use_exact_method(void) const
{
  return (n_pos() < minimum_size_) && (n_neg() < minimum_size_);
}

which one should we follow? Naively, I'm thinking the docs make more sense than the implementation, but let's check the literature.

[Peter]

• SPSS uses approximation if n>41.
• Hollander and Wolfe says approximation is valid if n⁺ and n^- are large.
• ​StatsDirect uses approximation if n>80 or (n⁺>30 and n^->30).

[Peter]

(In [2612]) change rule for when to use exact method in ROC::p_value. fixes #677