THE IMPACT OF MORPHOLOGICAL VARIANTS ON A CLADISTIC HYPOTHESIS WITH AN EXAMPLE FROM A MYOLOGICAL DATA SET

The larger the sample size per operational taxonomic unit (OTU), the higher the probability of identifying the correct character state for that OTU. Large numbers of characters are essential to assure the accuracy and stability of a phylogenetic reconstruction. Both large sample sizes and large numbers of characters are desirable, but given limited time and resources, cladistic analysis often involves a trade-off between the number of specimens one can study per OTU versus the number of characters one has time to discover. For a given data set, what balance of sample size and character number will yield an acceptably small error (the probability that a recognized clade is false) and yet achieve sufficient power (the probability of recognition of a valid clade)? Myological data compiled for a study of arvicoline rodents were manipulated to determine the effect of character number reduction, the effect of deliberate inclusion of incorrect character states (mistakes), and the interaction of these two effects in an attempt to provide one answer to this question. The results indicate that with sample sizes of four specimens per OTU (some of which were at supraspecific levels), the frequency of mistakes is sufficient to cause unacceptable increases in error and losses of power in resulting cladograms. However, the error and power of the resulting cladogram are more sensitive to reductions in character number than to the inclusion of an occasional incorrect character state. Beyond a minimal increase in sample size (to five specimens per OTU in this example), it is more advantageous to search for new characters than it is to increase the number of specimens surveyed. Because adequate sample size is a function of the taxonomic level of the study and the data structure of the characters, these specific results should not be broadly applied, but the methods used to determine the appropriate sample size for the arvicoline data are applicable to other studies. We must devote more effort to increasing the reliability of our data sets to assure highly reliable cladograms.