Partition-free congruence analysis: Implications for sensitivity analysis

Abstract

A criterion is proposed to compare systematic hypotheses based on multiple sources of information under a diverse set of interpretive assumptions (i.e., sensitivity analysis of b1Wheeler, 1995). This metric, the Meta-Retention Index (MRI), is the retention index (RI) of Farris calculated over the set of conventional homologous qualitative characters (ordered, unordered, Sankoff, etc.) and molecular fragment characters sensub2 b3Wheeler (1996, 1999). The superiority of this measure to other similar measures (e.g., incongruence length difference test) comes from its independence from partition information. The only values that participate in its calculation are the minimum, maximum and observed cost (= cladogram cost) of each character. The partition (morphology, gene locus) from which the variant may have come is irrelevant. In the special cases where there is only a single data partition, this measure is equivalent to the conventional RI; and in the case where there are single fragment characters per partition (contiguous molecular loci as data sets) the measure is identical to the complement of the Rescaled Incongruence Length Difference (RILD) of b4Wheeler and Hayashi (1998). The MRI can serve as an optimality criterion for deciding among systematic hypotheses based on the same data, but different sets of analysis assumptions (e.g., character weights, indel costs). The MRI may lose discriminatory power in situations where a minority of highly congruent characters is given high weight. This situation can be detected and seems unlikely to occur frequently in real data sets. © The Willi Hennig Society 2006.