Further problems with the ILD test: ‘hypercongruence’ effect and multiple comparisons

RAMÍREZ, M. J.

CLADISTICS (PRINT)

Blackwell Publishing

Año: 2006 vol. 22 p. 289 - 295

0748-3007

The incongruence length difference (ILD) test may produce artificially large significance values with the addition not only of uninformative characters, but also of informative characters not relevant to the groups in conflict.  Previously reported problems with the ILD test involved cases of false positives, reporting high incongruence when none is expected.  Under certain conditions, the test can suffer of the opposite problem (false negatives), reporting non-significant values in cases of high incongruence.  These opposing effects can be combined in a dataset, such that a comparison over all partitions appears as congruent, while some of the pairwise comparisons are reported as significantly incongruent.