Outgroup Analysis and Parsimony

Methods that use outgroups in the reconstruction of phylogeny are described and evaluated by the criterion of parsimony. By considering the character states and relationships of outgroups, one can estimate the states ancestral for a study group or ingroup, even when several character states are found among the outgroups. Algorithms and rules are presented that find the most parsimonious estimates of ancestral states for binary and multistate characters when outgroup relationships are well resolved. Other rules indicate the extent to which uncertainty about outgroup relationships leads to uncertainty about the ancestral states. The algorithms and rules are based on “simple parsimony” in that convergences and reversals are counted equally. After parsimony is measured locally among the outgroups to estimate ancestral states, parsimony is measured locally within the ingroup, given the ancestral states, to find the ingroup cladogram. This two-step procedure is shown to find the ingroup cladograms that are most parsimonious globally; that is, most parsimonious when parsimony is measured simultaneously over the ingroup and outgroups. However, the two-step procedure is guaranteed to achieve global parsimony only when: (a) outgroup relationships are sufficiently resolved beforehand; (b) outgroup analysis is taken to indicate the state not in the most recent common ancestor of the ingroup, but in a more distant ancestor; and (c) ancestral states are considered while the ingroup is being resolved, not merely added afterward to root an unrooted network. The criterion of global parsimony is then applied to evaluate procedures used when outgroup relationships are poorly resolved. The procedure that chooses as ancestral the state occurring most commonly among the outgroups can sometimes yield cladograms that are not globally parsimonious. By the criterion of global parsimony, the best procedure is one that simultaneously resolves the outgroups and ingroup with the data at hand. Finally, simple parsimony can choose among competing hypotheses, but it often fails to indicate how much confidence can be placed in that choice.