EMPIRICAL EXAMPLE

To demonstrate the potential utility of StrataPhy and to contrast its use with manual searches using PAUP and MacClade, we reanalyzed the Paleogene viverravid data set of Polly (1997) using StrataPhy. Our intention in reanalyzing Polly's data set is not comment on viverravid phylogeny, but to show how the automated procedure provided in a StrataPhy analysis can provide a more extensive search of the potential phylogenetic trees, in a reasonable amount of time.

The data matrix consists of 23 taxa (after redundant taxa were removed from the original matrix with 32 taxa, as in the original study), with 39 morphologic characters and a single stratigraphic character with 14 stratigraphic intervals. In the original analysis, Polly generated a set of most parsimonious cladograms using PAUP, and then manually manipulated them and designated ancestors using MacClade. Using this procedure, he identified two equally parsimonious phylogenetic trees with total length of 91 steps.

We performed a stratocladistic analysis with StrataPhy using 10 random taxon addition sequence replicates and a single heuristic ancestral search per replicate. The hypothetical outgroup included in the matrix was designated as the sole outgroup for the analysis. All morphologic characters were treated as unordered and given equal weight. All other settings were StrataPhy defaults. Whereas analysis time of different data sets will vary tremendously depending on various properties, this analysis was completed in 2 minutes and 22 seconds on a 1.67GHz Apple PowerBook G4 computer. StrataPhy discovered 24 most parsimonious phylogenies from a single "island" (sensu D.R. Maddison 1991) of length 89 – two steps shorter than those reported by Polly.

It should be noted that most of these 24 phylogenies differ in only the assignment of taxa as ancestors, as they correspond to only two cladistic topologies. The topologies of the most parsimonious trees from the StrataPhy analysis were not identical to that from Polly's analysis. In fact, the cladistic topologies of the most parsimonious phylogenies from the StrataPhy stratocladistic analysis are congruent with none of the most parsimonious cladograms produced in the PAUP cladistic search (Polly 1997). If the two most parsimonious trees are converted to their corresponding cladograms where observed taxa are not fixed as ancestors, the total morphological steps for the two trees are one and two steps longer, respectively, than the most parsimonious cladograms from the PAUP search. This result highlights the potential risk of missing the most parsimonious phylogeny inherent in the commonly employed technique of using only the most parsimonious cladograms from a PAUP search as a starting point for manual stratocladistic analysis in MacClade. The simultaneous branch-swapping and ancestral assignment search in StrataPhy allows users to rapidly survey a much broader set of possible optimal trees.