Most phylogenetic analyses involve limited data (usually much less than 10,000 characters) evolving under a mosaic of selective constraints. In such cases, phylogeny estimation can be particularly susceptible to random and systematic error. Even when data sets can be justifiably combined to increase total character number, vagaries of mutation dynamics among subsets of characters can mislead phylogeny estimation. One possible solution is to create an heterogeneous mutation model for phylogenetic analysis, one that can accommodate observed mutational differences between sets of characters. Here I describe such a model for phylogeny estimation in Myoporaceae using two chloroplast sequence data sets - the rpl16 intron and the trnT-trnL intergenic spacer - and morphological cladistic characters. The method uses data partitioning, likelihood ratio tests, bayesian inference, and a conversion of likelihood parameter values into weighted parsimony step matrices. The heterogeneous model encompasses five mutational categories for characters and generates parsimony trees that can be assessed by conventional bootstrap and Bremer support methods. Comparative results are given for outcomes of independent character partitions, combined equal-weight parsimony analysis, and the combined heterogeneous weighted parsimony analysis. The feasibility of such models for phylogeny estimation and their potential benefits and limitations are discussed.

Key words: rpl16 intron, trnT-trnL spacer, bayesian inference, heterogeneous models, maximum likelihood, weighted parsimony