Coping with Abundant Missing Entries in Phylogenetic Inference Using Parsimony

Abstract

?When cladistic data sets include taxa with abundant missing entries, parsimony anal? ysis may yield multiple equally optimal trees and necessitate the use of consensus methods,to summarize relationships that are common to the multiple trees. Determination of those relation? ships that are common to the equally parsimonious trees and are thus unambiguously supported by the parsimonious interpretation of the data may not be possible using consensus methods that are widely employed by systematists. Thus, missing data may have an obfuscatory effect upon phylogenetic relationships. This problem can be ameliorated or overcome by adopting a strategy of safe taxonomic reduction. In this approach, taxa that can have no effect upon the relationships inferred for other taxa but that may increase the numbers of equally most-parsimonious trees are identified. Eliminating such taxa through the application of a series of safe deletion rules may reduce the number of equally most-parsimonious trees and thereby facilitate the consensus rep? resentation of unambiguous relationships supported by the data. The methods are illustrated by reanalysis of cladistic data for the Saurischia. [Phylogeny; parsimony; underdetermination; equiv? alence; safe taxonomic reduction; consensus; Saurischia.] Missing entries are a common feature of cladistic data sets. Characters commonly need to be coded as missing for some taxa because the relevant parts are not pre? served or have not been examined. In ad? dition, some characters may be inapplica? ble in some taxa (e.g., tooth characters in birds and turtles, limb characters in snakes and caecilians) and must also be coded as missing for these taxa (Platnick et al., 1991). In both cases, the coding reflects ig? norance about the relationships of a char? acter state (known or unknown) to other character states (Wilkinson, 1992a). Amounts of missing data vary greatly from study to study. Although missing en? tries are not rare in neontological cladistic data, the greatest abundance of missing entries are encountered in paleontological data sets that include poorly preserved fos? sils as terminal taxa. In a survey of 30 pa? leontological and 51 neontological data sets drawn from the literature on tetrapod phylogeny, the average level of missing data, expressed as a percentage of the total number of data points, is 12.6% for the pa? leontological data (rang?, 0-52%) and 2.29% (0-12.25%) for the neontological data. 1 E-mail: mark.wilkinson@bris.ac.uk. When analyzed using parsimony, miss? ing entries contribute nothing positive to the analysis. According to Swofford (1990: 17), only those characters that have nonmissing values will affect the location of any taxon on the tree. In turn, the place? ment of taxa on the tree, as determined by informative characters, can lead to parsi? monious reconstructions of the values of missing entries. The principles of parsi? monious reconstruction of missing entries underpin the approach to the reconstruc? tion of unpreserved fossil morphology re? cently advocated by Bryant and Russell (1992). However, missing entries are not always innocuous to phylogenetic inference using parsimony analysis. It has been widely recognized that including poorly known taxa (with much missing data) in comput? erized parsimony analyses may lead to a dramatic increase in the numbers of equal? ly most-parsimonious trees (MPTs) and an accompanying loss of resolution in consen? sus trees that are used to summarize com? mon elements of the MPTs (Gauthier, 1986; Nixon and Wheeler, 1992; Novacek, 1992a, 1992b; Wilkinson, 1992a; Wilkinson and Benton, 1995). Several authors have there? fore excluded poorly known taxa from parsimony analyses to minimize the num-