Lecture 15 - Hypothesis Testing

Slides:



Advertisements
Similar presentations
An Introduction to Phylogenetic Methods
Advertisements

Likelihood ratio tests
1 General Phylogenetics Points that will be covered in this presentation Tree TerminologyTree Terminology General Points About Phylogenetic TreesGeneral.
Maximum Likelihood. Likelihood The likelihood is the probability of the data given the model.
Statistical Significance What is Statistical Significance? What is Statistical Significance? How Do We Know Whether a Result is Statistically Significant?
“Inferring Phylogenies” Joseph Felsenstein Excellent reference
Business Statistics - QBM117
Distance Methods. Distance Estimates attempt to estimate the mean number of changes per site since 2 species (sequences) split from each other Simply.
The z-Test What is the Purpose of a z-Test? What are the Assumptions for a z- Test? How Does a z-Test Work?
Statistical Significance What is Statistical Significance? How Do We Know Whether a Result is Statistically Significant? How Do We Know Whether a Result.
Evaluating Hypotheses Chapter 9 Homework: 1-9. Descriptive vs. Inferential Statistics n Descriptive l quantitative descriptions of characteristics ~
Course overview Tuesday lecture –Those not presenting turn in short review of a paper using the method being discussed Thursday computer lab –Turn in short.
C82MCP Diploma Statistics School of Psychology University of Nottingham 1 Overview of Lecture Independent and Dependent Variables Between and Within Designs.
Probabilistic methods for phylogenetic trees (Part 2)
© 2013 Pearson Education, Inc. Active Learning Lecture Slides For use with Classroom Response Systems Introductory Statistics: Exploring the World through.
Lecture 13 – Performance of Methods Folks often use the term “reliability” without a very clear definition of what it is. Methods of assessing performance.
Today Concepts underlying inferential statistics
Probability Population:
Statistics 11 Hypothesis Testing Discover the relationships that exist between events/things Accomplished by: Asking questions Getting answers In accord.
Tuesday, September 10, 2013 Introduction to hypothesis testing.
HYPOTHESIS TESTING: A FORM OF STATISTICAL INFERENCE Mrs. Watkins AP Statistics Chapters 23,20,21.
Phylogeny Estimation: Traditional and Bayesian Approaches Molecular Evolution, 2003
Lecture 15 - Hypothesis Testing A. Competing a priori hypotheses - Paired-Sites Tests Null Hypothesis : There is no difference in support for one tree.
The Probability of a Type II Error and the Power of the Test
Individual values of X Frequency How many individuals   Distribution of a population.
6.1 - One Sample One Sample  Mean μ, Variance σ 2, Proportion π Two Samples Two Samples  Means, Variances, Proportions μ 1 vs. μ 2.
Tree Confidence Have we got the true tree? Use known phylogenies Unfortunately, very rare Hillis et al. (1992) created experimental phylogenies using phage.
A brief introduction to phylogenetics
1 Psych 5500/6500 The t Test for a Single Group Mean (Part 1): Two-tail Tests & Confidence Intervals Fall, 2008.
Hypothesis Testing State the hypotheses. Formulate an analysis plan. Analyze sample data. Interpret the results.
Correlation Assume you have two measurements, x and y, on a set of objects, and would like to know if x and y are related. If they are directly related,
Not in FPP Bayesian Statistics. The Frequentist paradigm Defines probability as a long-run frequency independent, identical trials Looks at parameters.
Testing alternative hypotheses. Outline Topology tests: –Templeton test Parametric bootstrapping (briefly) Comparing data sets.
Simple examples of the Bayesian approach For proportions and means.
Test of a Population Median. The Population Median (  ) The population median ( , P 50 ) is defined for population T as the value for which the following.
Lecture 19 – Species Tree Estimation
Step 1: Specify a null hypothesis
Hypothesis Testing for Proportions
Size of a hypothesis test
Chapter 4. Inference about Process Quality
Hypothesis Testing II: The Two-sample Case
CHAPTER 6 Statistical Inference & Hypothesis Testing
PCB 3043L - General Ecology Data Analysis.
Chapters 20, 21 Hypothesis Testing-- Determining if a Result is Different from Expected.
Hypothesis Testing and Confidence Intervals (Part 1): Using the Standard Normal Lecture 8 Justin Kern October 10 and 12, 2017.
Central Limit Theorem, z-tests, & t-tests
AP STATISTICS REVIEW INFERENCE
Two-sided p-values (1.4) and Theory-based approaches (1.5)
Hypothesis Tests for a Population Mean in Practice
STA 216 Generalized Linear Models
Elementary Statistics
Problems: Q&A chapter 6, problems Chapter 6:
Discrete Event Simulation - 4
Summary and Recommendations
Inference Concepts Hypothesis Testing.
Statistics for the Social Sciences
Reasoning in Psychology Using Statistics
Lecture 10/24/ Tests of Significance
Reasoning in Psychology Using Statistics
Computing and Statistical Data Analysis / Stat 7
CHAPTER 12 Inference for Proportions
CHAPTER 12 Inference for Proportions
Lecture 8 – Searching Tree Space
What are their purposes? What kinds?
Inference Concepts Hypothesis Testing.
Reasoning in Psychology Using Statistics
STA 291 Summer 2008 Lecture 18 Dustin Lueker.
Testing Hypotheses I Lesson 9.
Summary and Recommendations
Evaluation David Kauchak CS 158 – Fall 2019.
Presentation transcript:

Lecture 15 - Hypothesis Testing A. Competing a priori hypotheses Null Hypothesis : There is no difference in support for one tree over the other. Test statistic: In Parsimony (or minimum evolution); d = LengthA – LengthB In Likelihood; d = lnLA – lnLB

Early methods looked at the difference in optimality between the two trees on a site-by-site basis. Thus, these are called paired-sites method. Normal (mean = 0) KH-Test: Parsimony – normal distribution Tree # 1 2 Length 354 357 [So the test statistic is 3 steps]   Kishino-Hasegawa test: Tree Length diff s.d.(diff) t P* --------------------------------------------------------------------- 1 354 (best) 2 357 3 9.65019 0.3109 0.7560 * Probability of getting a more extreme T-value under the null hypothesis of no difference between the two trees (two-tailed test). Example So we would accept the Ho that there is no difference in support for either tree.

Same data, but using ML as criterion. KH-Test: Likelihood – normal distribution Tree 1 2 -ln L 2558.60898 2529.81009 [So the test statistic is 28.8 lnL units]   Kishino-Hasegawa test: KH test using normal approximation, two-tailed test KH-test Tree -ln L Diff -ln L P --------------------------------------------- 1 2558.60898 28.79889 0.000* 2 2529.81009 (best) * P < 0.05 Here, the same data and test lead to rejection of Ho under a different optimality criterion KH-Test: Likelihood – RELL bootstrapping Tree 1 2 -ln L 2558.60898 2529.81009   Kishino-Hasegawa test: KH test using RELL bootstrap, two-tailed test Number of bootstrap replicates = 1000 KH-test Tree -ln L Diff -ln L P --------------------------------------------- 1 2558.60898 28.79889 0.396 2 2529.81009 (best)

trees that were examining. Same Data, Same Two Trees, Different Inferences Note that we’ve been using these tests as intended; we have two a priori trees that were examining.

Most applications of these tests have involved comparing a suboptimal tree to the best tree. This results in use of a mismatched null distribution and the tests should not be used in this way. SH Test - corrects for the a priori requirement. Uses a collection of trees (rather than just ML v. Hypothesis). If there are only two trees in the collection, the ML tree and the tree we’re interested in testing, the SH test is identical to the KH test. Uses RELL bootstrap to generate an average lnL for the collection. Uses this average to center the null distribution. The test is very conservative, though, and very sensitive to the collection of trees used in centering the null distribution. Shimodaira has applied multiscale bootstrapping to correct for this bias in the AU (almost unbiased) test

Parametric Bootstrap Tests Generate the null distribution via simulation under the hypothesis being tested. Assuming some hypothesis is true, what is the probability of an observed test statistic? Ancient Fragmentation Hypothesis PNW Biogeography Northern Dispersal Hypothesis N. Rockies N. Cascades S. Cascades Southern Dispersal Hypothesis N. Rockies S. Cascades N. Cascades

Parametric Bootstrap Tests This tree has a likelihood score of lnL = -1593.01499

Parametric Bootstrap Tests Can we reject an ancient fragmentation? Force the tree to fit the predictions of that hypothesis, we get a much worse tree: lnL = -1612.50229 d = lnL(uconstrained) – lnL(constrained) = - 1593.01499 – ( - 1612.50229) = 19.487 lnL units. If the true tee actually were the constrained tree (i.e., if the hypothesis being examined were actually true), what is the probability that we would see a tree that is 19.487 lnL units better, simply due to stochasticity? So we simulate many replicate data sets under the hypothesis. We first run an unconstrained search to find the ML tree for the replicate. We then run a search constrained to find the best tree consistent with the hypothesis for each replicate.

Parametric Bootstrap Tests Can we reject an ancient fragmentation? We can reject a southern dispersal hypothesis as well and are left with the northern dispersal hypothesis that the tree suggests.

Bayesian Hypothesis Testing If we run a typical MCMC, we have sample of topologies that represent the posterior distribution of trees. The proportion of the trees in the distribution that are consistent with the topological predictions of each hypothesis provide the posterior probability of that hypothesis. Remember the difference in Bayesian and frequentist perspectives, and the different interpretations of p-values this causes.

An Approximate LRT of Hypotheses Anismonv and Gascuel (2006. Syst. Biol., 55:539) produced a fast approximate approach. Most phylogenetic hypotheses rely the presence/ absence of a particular internal branch. Test statistic would be 2(lnl1 - lnl0). A more conservative test statistic: 2(lnl1 - lnl2). Power and accuracy analyses & robustness to model violations.