Other Test Constructions: Likelihood Ratio & Bayes Tests

Σχετικά έγγραφα
ST5224: Advanced Statistical Theory II

Theorem 8 Let φ be the most powerful size α test of H

Statistical Inference I Locally most powerful tests

Solution Series 9. i=1 x i and i=1 x i.

Econ 2110: Fall 2008 Suggested Solutions to Problem Set 8 questions or comments to Dan Fetter 1

2 Composition. Invertible Mappings

STAT200C: Hypothesis Testing

6. MAXIMUM LIKELIHOOD ESTIMATION

Μηχανική Μάθηση Hypothesis Testing

Ordinal Arithmetic: Addition, Multiplication, Exponentiation and Limit

Last Lecture. Biostatistics Statistical Inference Lecture 19 Likelihood Ratio Test. Example of Hypothesis Testing.

SOLUTIONS TO MATH38181 EXTREME VALUES AND FINANCIAL RISK EXAM

An Introduction to Signal Detection and Estimation - Second Edition Chapter II: Selected Solutions

557: MATHEMATICAL STATISTICS II RESULTS FROM CLASSICAL HYPOTHESIS TESTING

Exercise 2: The form of the generalized likelihood ratio

Bayesian statistics. DS GA 1002 Probability and Statistics for Data Science.

557: MATHEMATICAL STATISTICS II HYPOTHESIS TESTING

Lecture 34 Bootstrap confidence intervals

Homework 3 Solutions

The Simply Typed Lambda Calculus

Example Sheet 3 Solutions

SOLUTIONS TO MATH38181 EXTREME VALUES AND FINANCIAL RISK EXAM

derivation of the Laplacian from rectangular to spherical coordinates

Lecture 12: Pseudo likelihood approach

Estimation for ARMA Processes with Stable Noise. Matt Calder & Richard A. Davis Colorado State University

C.S. 430 Assignment 6, Sample Solutions

Inverse trigonometric functions & General Solution of Trigonometric Equations

k A = [k, k]( )[a 1, a 2 ] = [ka 1,ka 2 ] 4For the division of two intervals of confidence in R +

A Note on Intuitionistic Fuzzy. Equivalence Relation

4.6 Autoregressive Moving Average Model ARMA(1,1)

Lecture 2: Dirac notation and a review of linear algebra Read Sakurai chapter 1, Baym chatper 3

Lecture 21: Properties and robustness of LSE

Areas and Lengths in Polar Coordinates

6.3 Forecasting ARMA processes

Statistics 104: Quantitative Methods for Economics Formula and Theorem Review

Every set of first-order formulas is equivalent to an independent set

Solutions to Exercise Sheet 5

Approximation of distance between locations on earth given by latitude and longitude

CHAPTER 25 SOLVING EQUATIONS BY ITERATIVE METHODS

Problem Set 3: Solutions

New bounds for spherical two-distance sets and equiangular lines

ENGR 691/692 Section 66 (Fall 06): Machine Learning Assigned: August 30 Homework 1: Bayesian Decision Theory (solutions) Due: September 13

SCITECH Volume 13, Issue 2 RESEARCH ORGANISATION Published online: March 29, 2018

Chapter 6: Systems of Linear Differential. be continuous functions on the interval

Fourier Series. MATH 211, Calculus II. J. Robert Buchanan. Spring Department of Mathematics

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 6/5/2006

Section 8.3 Trigonometric Equations

EE512: Error Control Coding

Jesse Maassen and Mark Lundstrom Purdue University November 25, 2013

Areas and Lengths in Polar Coordinates

5.4 The Poisson Distribution.

Απόκριση σε Μοναδιαία Ωστική Δύναμη (Unit Impulse) Απόκριση σε Δυνάμεις Αυθαίρετα Μεταβαλλόμενες με το Χρόνο. Απόστολος Σ.

Homework for 1/27 Due 2/5

Fractional Colorings and Zykov Products of graphs

Exercises to Statistics of Material Fatigue No. 5

Uniform Convergence of Fourier Series Michael Taylor

6.1. Dirac Equation. Hamiltonian. Dirac Eq.

ORDINAL ARITHMETIC JULIAN J. SCHLÖDER

Numerical Analysis FMN011

Concrete Mathematics Exercises from 30 September 2016

Reminders: linear functions

Chapter 6: Systems of Linear Differential. be continuous functions on the interval

w o = R 1 p. (1) R = p =. = 1

ECE598: Information-theoretic methods in high-dimensional statistics Spring 2016

Homework 8 Model Solution Section

3.4 SUM AND DIFFERENCE FORMULAS. NOTE: cos(α+β) cos α + cos β cos(α-β) cos α -cos β

Matrices and Determinants

12. Radon-Nikodym Theorem

Practice Exam 2. Conceptual Questions. 1. State a Basic identity and then verify it. (a) Identity: Solution: One identity is csc(θ) = 1

Partial Differential Equations in Biology The boundary element method. March 26, 2013

Trigonometric Formula Sheet

The ε-pseudospectrum of a Matrix

Congruence Classes of Invertible Matrices of Order 3 over F 2

Math 446 Homework 3 Solutions. (1). (i): Reverse triangle inequality for metrics: Let (X, d) be a metric space and let x, y, z X.

ω ω ω ω ω ω+2 ω ω+2 + ω ω ω ω+2 + ω ω+1 ω ω+2 2 ω ω ω ω ω ω ω ω+1 ω ω2 ω ω2 + ω ω ω2 + ω ω ω ω2 + ω ω+1 ω ω2 + ω ω+1 + ω ω ω ω2 + ω

Introduction to the ML Estimation of ARMA processes

Nowhere-zero flows Let be a digraph, Abelian group. A Γ-circulation in is a mapping : such that, where, and : tail in X, head in

Depth versus Rigidity in the Design of International Trade Agreements. Leslie Johns

Math221: HW# 1 solutions

TMA4115 Matematikk 3

DESIGN OF MACHINERY SOLUTION MANUAL h in h 4 0.

Notes on the Open Economy

Math 6 SL Probability Distributions Practice Test Mark Scheme

More Notes on Testing. Large Sample Properties of the Likelihood Ratio Statistic. Let X i be iid with density f(x, θ). We are interested in testing

If we restrict the domain of y = sin x to [ π, π ], the restrict function. y = sin x, π 2 x π 2

Lecture 15 - Root System Axiomatics

MATH423 String Theory Solutions 4. = 0 τ = f(s). (1) dτ ds = dxµ dτ f (s) (2) dτ 2 [f (s)] 2 + dxµ. dτ f (s) (3)

Second Order RLC Filters

CHAPTER 48 APPLICATIONS OF MATRICES AND DETERMINANTS

Figure A.2: MPC and MPCP Age Profiles (estimating ρ, ρ = 2, φ = 0.03)..

If we restrict the domain of y = sin x to [ π 2, π 2

Phys460.nb Solution for the t-dependent Schrodinger s equation How did we find the solution? (not required)

PARTIAL NOTES for 6.1 Trigonometric Identities

Parametrized Surfaces

ESTIMATION OF SYSTEM RELIABILITY IN A TWO COMPONENT STRESS-STRENGTH MODELS DAVID D. HANAGAL

ΕΙΣΑΓΩΓΗ ΣΤΗ ΣΤΑΤΙΣΤΙΚΗ ΑΝΑΛΥΣΗ

The challenges of non-stable predicates

Lecture 7: Overdispersion in Poisson regression

The Pohozaev identity for the fractional Laplacian

Dynamic types, Lambda calculus machines Section and Practice Problems Apr 21 22, 2016

Transcript:

Other Test Constructions: Likelihood Ratio & Bayes Tests Side-Note: So far we have seen a few approaches for creating tests such as Neyman-Pearson Lemma ( most powerful tests of H 0 : θ = θ 0 vs H 1 : θ = θ 1 ) Two methods for uniformly most powerful (UMP) tests Method I: Based on Neyman-Pearson Lemma (may work for R p -valued parameters θ and tests of H 0 : θ Θ 0 R p vs H 1 : θ Θ 0 ) Method II: Monotone Likelihood Ratio (may work for real-valued θ R and tests H 0 : θ θ 0 vs H 1 : θ > θ 0 or H 0 : θ θ 0 vs H 1 : θ < θ 0 ) The tests above are often rooted in comparing likelihoods to make a testing decision (e.g., Neyman- Pearson Lemma). We next consider a very general testing procedure based on comparing the ratio of two likelihoods. A.) Likelihood Ratio Tests Definition: Let f(x θ), θ Θ R p, be the joint pdf/pmf of X = (X 1,..., X n ) (the parameter θ can be vector-valued) and let Θ 0 be a nonempty proper subset of Θ. Then, the likelihood ratio statistic (LRS) for testing H 0 : θ Θ 0 R p vs H 1 : θ Θ 0 is defined as λ(x) = max f(x θ) θ Θ 0 max f(x θ). θ Θ Note that if ˆθ MLE of θ over entire Θ & θ maximum of f(x θ) over θ Θ 0 we may write λ(x) = f(x θ) f(x ˆθ) Definition: A size α likelihood ratio test (LRT) for testing H 0 : θ Θ 0 R p vs H 1 : θ Θ 0 is defined as 1 if λ(x) < k ϕ(x) = γ if λ(x) = k 0 if λ(x) > k where γ [0, 1] and 0 k 1 are constants determined by max θ Θ0 E θ ϕ(x ) = α. 1

Example: Let X 1,..., X n be iid Gamma(α = 3, θ), θ > 0. Find a size α LRT for H 0 : θ = θ 0 vs H 1 : θ θ 0. 2

Example: Let X 1,..., X n be iid Exponential(θ, ν), θ > 0, ν R with common pdf { 1 θ f(x θ, ν) = e (x ν)/θ if x ν 0 otherwise Find a size α LRT for H 0 : ν = ν 0 vs H 1 : ν ν 0 (where ν 0 R is fixed). 3

B.) Large Sample Properties of LRT Tests (for calibration) The following result describes the asymptotic distribution of the likelihood ratio statistic (under appropriate regularity conditions) & may be used to calibrate a LRT in a simple fashion when the sample size is sufficiently large. Theorem: Let X 1,... be iid random vectors with common pdf/pmf f(x θ), θ Θ R (the parameter θ can be vector-valued). Let λ n (X 1,...,X n) denote the likelihood ratio statistic based on X 1,...,X n for testing H 0 : θ Θ 0 R p vs H 1 : θ Θ 0, where Θ 0 has the form { } Θ 0 = θ = (θ 1,..., θ p ) Θ : θ 1 = θ1, 0..., θ r = θ }{{} r 0 for some θ1, 0..., θr 0 and r p hypothesized values for first r p parameters Then, under the Cramér-Rao type regularity conditions, it holds that: if H 0 is true, 2 log λ n (X 1,...,X d n) χ 2 r as n. Remark: The above limiting distribution suggests the following testing procedure based on the (1 α)-quantile of a χ 2 r distribution, denoted as χ 2 1 α(r) for which P ( χ 2 r χ 2 1 α(r) ) = 1 α, P ( χ 2 r > χ 2 1 α(r) ) = α. Namely, for n large (e.g., say n 30 observations), ϕ(x 1,...,X 1 if 2 log λ n (X 1 n) =,...,X n) > χ 2 1 α(r) 0 otherwise is an approximate size α LRT for testing H 0 : θ 1 = θ 0 1,..., θ r = θ 0 r vs H 1 : θ i θ 0 i for some 1 i r. 4

Example: Let X 1,... be iid N 2(µ, A) random vectors, where µ = (µ 1, µ 2 ) R 2 and A is a known 2 2 positive definite matrix. Find a size α LRT for testing H 0 : 2µ 1 + 3µ 2 = 0 vs H 1 : 2µ 1 + 3µ 2 0. 5

C.) Bayes Tests Let X 1,..., X n have joint pdf/pmf f(x θ), θ Θ R p, and we want to test H 0 : θ Θ 0 R p vs H 1 : θ Θ 0. Let π(θ) be a prior pdf P (θ Θ 0 x) = Θ 0 f θ x(θ)dθ posterior probability that θ Θ 0 P (θ Θ 0 x) = Θ\Θ 0 f θ x(θ)dθ posterior probability that θ Θ 0 Note that P (θ Θ 0 x) + P (θ Θ 0 x) = 1 Then, a Bayes test for testing H 0 : θ Θ 0 vs H 1 : θ Θ 0 is given by 1 if P (θ Θ 0 x) P (θ Θ 0 x) 1 if P (θ Θ 0 x) 1/2 ϕ(x) = = 0 otherwise 0 otherwise Discussion: The Bayes test follows from minimizing the Bayes Risk BR ϕ1 simple test ϕ 1 (x) (i.e., tests where ϕ 1 (x) {0, 1} for any x) of a Consider a loss function L(θ, a) = I {θ Θ0 }I {a=1} + I {θ Θ0 }I {a=0} where a may assume two values: a = 1 means reject H 0 and a = 0 means don t reject H 0. So, the loss L(θ, a) = 0 for a correct decision and L(θ, a) = 1 for an incorrect decision: L(θ, a) = { 0 if θ Θ 0 & a = 0 or if θ Θ 0 & a = 1 1 otherwise The risk function of a simple test ϕ 1 (x) is R ϕ1 (θ) = E θ L(θ, ϕ 1 )) = I {θ Θ0 } P θ (ϕ 1 ) = 1) (X } (X {{} prob. of Type I error +I {θ Θ0 } P θ (ϕ 1 (X ) = 0) }{{} prob. of Type II error The Bayes risk of ϕ 1 (x) w.r.t. π(θ) is: BR ϕ1 = E (θ) R ϕ1 (θ) = Θ R ϕ 1 (θ)π(θ)dθ & the Bayes test ϕ(x) minimizes BR ϕ1 over all simple tests ϕ 1 (x) 6

Alternatively, we can find the Bayes test by minimizing the posterior risk of a simple test ϕ 1 (x) for each fixed x, where the posterior risk is ( ) E θ xl(ϕ 1 (x), θ) = I {θ Θ0 }I {ϕ1 (x)=1} + I {θ Θ0 }I {ϕ1 (x)=0} f θ x(θ)dθ Θ = I {ϕ1 (x)=1} f θ x(θ)dθ + I {ϕ1 (x)=0} f θ x(θ)dθ Θ 0 Θ\Θ 0 = I {ϕ1 (x)=1}p (θ Θ 0 x) + I {ϕ1 (x)=0}p (θ Θ 0 x) For each fixed x, we choose the values ϕ 1 (x) = 1 or 0 of the test to minimize the posterior risk; that is, for each fixed x, we should pick ϕ 1 (x) = 1 if P (θ Θ 0 x) P (θ Θ 0 x) and pick ϕ 1 (x) = 0 if P (θ Θ 0 x) < P (θ Θ 0 x). Note this is the same decision rule as the Bayes test ϕ(x) above. Example: Let X 1,..., X n be iid N(θ, 1), θ R. Find the Bayes test for H 0 : θ θ 0 vs H 1 : θ > θ 0 under the N(µ, τ 2 ) prior for θ, where µ, τ 2, θ 0 are fixed. 7