| dc.description.abstract | A wide class of biogeographic or phylogeographic studies predicts the simultaneous divergence of co-distributed taxa. Typically, a geological event, or a climate-related change in geography, is hypothesized to have structured a broad range of biota, many components of which may only be distantly related to each other. Direct assessment of these predictions is precluded in many studies by the lack or paucity of appropriate fossils for calibration when estimating divergence times in a phylogenetic context. However, even without direct divergence time estimation of all the relevant splits, there might be sufficient information in the data to estimate the probability that these groups diverged simultaneously if the datasets are treated in a parallel, coordinated, and integrated fashion, rather than independently. This study investigates the statistical framework and methods used to address this issue. Most current statistical phylogeographic methods rely on the coalescent as an underlying model. While the coalescent is robust to a range of violations of some of its assumptions, such as the Wright- Fisher demographic model, and, morever, has been elaborated or extended to allow the relaxing of some of its other assumptions, little has been done to assess and quanitfy how violations of these assumptions affect phylogeographic analysis in general, and phylogeographic model selection in particular. One of the major problems in evaluating the performance of phylogeographic methods with respect to their responses or behavior when the assumptions of the coalescent are violated is the lack of a rich or flexible non-coalesccent based spatially-explicit simulation engine. The first chapter of my dissertation is thus focussed on developing and producing such a simulator: a forward-time, agent-based, spatially-explicit simulation program that generates genealogies for multiple loci evolving in populations of multiple sexual diploid species on a spatio-temporally environmentally-heterogenous landscape. The second chapter of the dissertation assesses the performance of an Approximate Bayesian Computation approach to simultaneous divergence time testing model selection. It profiles the performance this approach under a variety of conditions, ranging from ones in which its model assumptions are completely met, to ones in which they are selectively violated in varying degrees. While there currently are no full- or exact-likelihood methods that address this question, under the special controlled circumstances of the study it was possible to adapt an existing program to provide some indication of how a full-likehood method may work in contrast. The third chapter of this work presents a program that simultaneously estimates the divergence time between sister populations of multiple species in parallel. This program uses a Bayesian statistical framework to analyze data from multiple genetic loci, integrating over uncertainty in gene trees, divergence times, and demographic parameters. If limited to two species, the program allows for reverse-jump MCMC to sample from models of different dimensionality with respect to the divergence time, so as to explicitly estimate the posterior probability of simultaneous divergence vs. non-simultaneous divergence. | |
