IB 200A Principals of Phylogenetic Systematics Spring 2010 Bayesian Phylogenetics Nick Matzke I. Background: Philosophy of Statistics What is the point of statistics? And what are you doing when you reach a statistical conclusion? These questions are basically never asked or answered in most introductory statistics classes, from middle school through many graduate courses. The questions only became apparent to me when I began to realize that the field of statistics is not like basic mathematics, even though at first it seems like just an application of the math you learned in high school. In basic math, answers are either right or wrong, and that’s it. In statistics, the “right ” method (and thus answer) can often be a matter of opinion. In statistics, there are • judgment calls, • background philosophies, • uncooperative data (e.g., data that don’t fit ideal criteria, such as independence, or following a standard distribution – especially e.g. biological and spatial data), • uncooperative calculations (e.g., non-integrable functions, calculations that take too long, problems that involve evaluating more possibilities than there are atoms in the universe) • important, practical decisions that depend upon the conclusion reached, despite all of the above Question: What are some practical decisions that rely upon statistical conclusions? In science? In biology? In phylogenetics? E.g.: Ou et al. (1992). Molecular epidemiology of HIV transmission in a dental practice. Science 256, 1165–1171. doi: 10.1126/science.256.5060.11652 de Oliveira et al. (2006). HIV-1 and HCV sequences from Libyan outbreak. Nature, 444, 836-837. doi:10.1038/444836a Received: 4 November 2006; Accepted 24 November 2006; Published online 6 December 2006. http://www.nature.com/nature/journal/v444/n7121/full/444836a.html3 An excellent, broad, and sophisticated-yet-introductory discussion of statistics and its application to evolutionary science is Elliot Sober’s (2008) Evidence and Evolution: The logic behind the science. The chapters: 1. Evidence 2. Intelligent design 3. Natural selection 4. Common ancestry Sober’s main point is that it is extremely important to be extremely clear on what question, exactly, you are asking. The grand debates between different statistical “schools of thought ” – Bayesian, Likelihoodism, Frequentism – and about what specific methods are appropriate are often much more resolvable if you think carefully about what question you have, and what information (data) you have or can get. Sober (2008), p. 3: The statistician Richard Royall begins his excellent book on the concept of evidence (Royall 1997:4) by distinguishing three questions: (1) What does the present evidence say? (2) What should you believe? (3) What should you do? […] answering question (2) requires more than an answer to (1), and answering question (3) requires more than an answer to (2). The best feature of Sober is the extremely clear introduction to three major statistical “schools of thought,” and discussion of the strengths and weaknesses of each in numerous specific real-world situations (including inferring common ancestry versus separate ancestry, and inferring the action of natural selection). II. Bayesianism, Likelihoodism, Frequentism Except for basic probability, essentially all the statistics that any of you learned in high school and college was frequentist (without saying so). So for most people, frequentist statistics – ideas like chi-squared tests, t-tests, regression, ANOVA, and testing of null hypotheses – simply is “statistics. “ Strangely, frequentist statistics is actually the youngest school of thought, and its dominance is a recent phenomenon, dating only to the early/mid-20th century. Frequentism definitely benefited from being the favored approach during the explosion of professional science over the last 100 years, and frequentism was particularly strong in biology, especially genetics and population genetics. The famous evolutionary biologist Sir Ronald A. Fisher was also probably the most important founder/promoter of frequentism. E.g. Wikipedia quotes Richard Dawkins calling him “the greatest of Darwin's successors,” and someone else calling him “a genius who almost single-handedly created the foundations for modern statistical science.” (Note equation of frequentism with “modern statistical science! ”) Another Wikipedia gem: L.J. Savage, "I occasionally meet geneticists who ask me whether it is true that the great geneticist R.A. Fisher was also an important statistician" (Annals of Statistics, 1976).4 (In addition to helping to found population genetics, frequentist statistics, inventing Fisher's Fundamental Theorem of Natural Selection, and being knighted, Fisher was also an avid lifelong eugenicist, and a lifelong devout Anglican; the concept linking all of this together is “Progress,” but that is a different lecture…) Bayes Bayesianism is actually much older, dating back at least to the 1700s and discussions of games of chance and probabilities. The name comes from the Reverend Thomas Bayes (1702-1761), who proposed a special case of what came to be called “Bayes’ theorem” in his posthumous Essay Towards Solving a Problem in the Doctrine of Chances (1764). Bayes’ theorem is easiest to understand by starting with basic probability and conditional probability. Basic Probability Let’s first remember some basic probability. • P(E) = P(event) = “Probability than an event occurs in a trial” Often writers talk about the P(data) or P(observations) instead of P(event). • Probabilities of exclusive events must sum to 1, so P(E) + P(not E) = 1 Discussion Questions: • What is P(heads)? • What is P(rolling a 1) = P(event = 1) = P(1)? Conditional Probability In reality, to answer the questions above, we need some model or hypothesis before we can calculate the probability. This is: • P(event given some model/hypothesis) = P(event | hypothesis) = P(E | H) • “model” and “hypothesis” get used interchangably E.g., the probability of getting a 1 on a 6-sided fair die is • P(event = 1 | “6-sided fair die”) = 1/6, or • P(E|H) = 1/6, where E=”rolling a 1” and H= ”die is six-sided and fair” What is the probability of rolling a 1 if the die is randomly picked from 2 dice, where 1 die is 6-sided and fair, and
View Full Document
Unlocking...