[ ] Because only values outside of the range of the confidence interval are counted, the method implicitly assumes that values in the distributions can be ordered and represent contiguous ranges. For instance if d1 = [2,2,3,4], d2=[1,1,2,2,4,4,5,5], it's possible that d1 was drawn from d2. This implicit assumption is wrong if the values 1,2,3,4,5 are actually ordinal encodings of unordered categories, i.e. this test does not apply if {1:'apple', 2:'banana', 3:'grape', 4:'orange', 5:'strawberry'}. TODO: compare to chi-squared test.
{1:'apple', 2:'banana', 3:'grape', 4:'orange', 5:'strawberry'}
. TODO: compare to chi-squared test.