Matrix Representation of Data in Experiment

Similar documents
Linear regression. Daniel Hsu (COMS 4771) (y i x T i β)2 2πσ. 2 2σ 2. 1 n. (x T i β y i ) 2. 1 ˆβ arg min. β R n d

ECONOMETRIC THEORY. MODULE XIII Lecture - 34 Asymptotic Theory and Stochastic Regressors

1 General linear Model Continued..

Lecture 22: Review for Exam 2. 1 Basic Model Assumptions (without Gaussian Noise)

Properties and Hypothesis Testing

STATISTICAL PROPERTIES OF LEAST SQUARES ESTIMATORS. Comments:

Statistical and Mathematical Methods DS-GA 1002 December 8, Sample Final Problems Solutions

Asymptotic Results for the Linear Regression Model

Probability 2 - Notes 10. Lemma. If X is a random variable and g(x) 0 for all x in the support of f X, then P(g(X) 1) E[g(X)].

Efficient GMM LECTURE 12 GMM II

The variance of a sum of independent variables is the sum of their variances, since covariances are zero. Therefore. V (xi )= n n 2 σ2 = σ2.

( θ. sup θ Θ f X (x θ) = L. sup Pr (Λ (X) < c) = α. x : Λ (x) = sup θ H 0. sup θ Θ f X (x θ) = ) < c. NH : θ 1 = θ 2 against AH : θ 1 θ 2

Chapter 1 Simple Linear Regression (part 6: matrix version)

Linear Regression Demystified

[412] A TEST FOR HOMOGENEITY OF THE MARGINAL DISTRIBUTIONS IN A TWO-WAY CLASSIFICATION

MA Advanced Econometrics: Properties of Least Squares Estimators

TAMS24: Notations and Formulas

Lecture 7: Properties of Random Samples

Section 14. Simple linear regression.

Lecture 11 and 12: Basic estimation theory

UNIVERSITY OF NORTH CAROLINA Department of Statistics Chapel Hill, N. C. FOR MULTIVARIATE POPULATIONS. J. N. Srivastava.

Slide Set 13 Linear Model with Endogenous Regressors and the GMM estimator

Statistical Inference Based on Extremum Estimators

Statistics 203 Introduction to Regression and Analysis of Variance Assignment #1 Solutions January 20, 2005

Lecture 33: Bootstrap

11 Correlation and Regression

First, note that the LS residuals are orthogonal to the regressors. X Xb X y = 0 ( normal equations ; (k 1) ) So,

f(x i ; ) L(x; p) = i=1 To estimate the value of that maximizes L or equivalently ln L we will set =0, for i =1, 2,...,m p x i (1 p) 1 x i i=1

Estimation of the Mean and the ACVF

University of California, Los Angeles Department of Statistics. Practice problems - simple regression 2 - solutions

Simple Linear Regression

Economics 241B Relation to Method of Moments and Maximum Likelihood OLSE as a Maximum Likelihood Estimator

Lecture 3. Properties of Summary Statistics: Sampling Distribution

Lecture 6 Chi Square Distribution (χ 2 ) and Least Squares Fitting

11 THE GMM ESTIMATION

Topic 9: Sampling Distributions of Estimators

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS

Econ 325/327 Notes on Sample Mean, Sample Proportion, Central Limit Theorem, Chi-square Distribution, Student s t distribution 1.

arxiv: v1 [math.pr] 13 Oct 2011

Expectation and Variance of a random variable

CEE 522 Autumn Uncertainty Concepts for Geotechnical Engineering

Lecture 6 Chi Square Distribution (χ 2 ) and Least Squares Fitting

Lecture 2: Monte Carlo Simulation

A Relationship Between the One-Way MANOVA Test Statistic and the Hotelling Lawley Trace Test Statistic

Some Basic Probability Concepts. 2.1 Experiments, Outcomes and Random Variables

Geometry of LS. LECTURE 3 GEOMETRY OF LS, PROPERTIES OF σ 2, PARTITIONED REGRESSION, GOODNESS OF FIT

Exponential Families and Bayesian Inference

Chapter 2 The Monte Carlo Method

ECON 3150/4150, Spring term Lecture 3

Algebra of Least Squares

Simple Linear Regression Matrix Form

Statistical Inference (Chapter 10) Statistical inference = learn about a population based on the information provided by a sample.

Chapter Vectors

The Method of Least Squares. To understand least squares fitting of data.

Problem Set 4 Due Oct, 12

Inverse Matrix. A meaning that matrix B is an inverse of matrix A.

1 Inferential Methods for Correlation and Regression Analysis

Goodness-of-Fit Tests and Categorical Data Analysis (Devore Chapter Fourteen)

, then cv V. Differential Equations Elements of Lineaer Algebra Name: Consider the differential equation. and y2 cos( kx)

First Year Quantitative Comp Exam Spring, Part I - 203A. f X (x) = 0 otherwise

(3) If you replace row i of A by its sum with a multiple of another row, then the determinant is unchanged! Expand across the i th row:

Last time: Moments of the Poisson distribution from its generating function. Example: Using telescope to measure intensity of an object

CEU Department of Economics Econometrics 1, Problem Set 1 - Solutions

Solution to Chapter 2 Analytical Exercises

Linear Regression Models

Topic 9: Sampling Distributions of Estimators

Because it tests for differences between multiple pairs of means in one test, it is called an omnibus test.

Topic 9: Sampling Distributions of Estimators

Problem Set 2 Solutions

Machine Learning Brett Bernstein

TMA4245 Statistics. Corrected 30 May and 4 June Norwegian University of Science and Technology Department of Mathematical Sciences.

4. Hypothesis testing (Hotelling s T 2 -statistic)

FACULTY OF MATHEMATICAL STUDIES MATHEMATICS FOR PART I ENGINEERING. Lectures

Open book and notes. 120 minutes. Cover page and six pages of exam. No calculators.

5 : Exponential Family and Generalized Linear Models

APPLIED MULTIVARIATE ANALYSIS

17. Joint distributions of extreme order statistics Lehmann 5.1; Ferguson 15

CS284A: Representations and Algorithms in Molecular Biology

ECE 8527: Introduction to Machine Learning and Pattern Recognition Midterm # 1. Vaishali Amin Fall, 2015

Stochastic Simulation

Machine Learning Theory (CS 6783)

(all terms are scalars).the minimization is clearer in sum notation:

Module 1 Fundamentals in statistics

Probability and statistics: basic terms

Math 152. Rumbos Fall Solutions to Review Problems for Exam #2. Number of Heads Frequency

Circle the single best answer for each multiple choice question. Your choice should be made clearly.

MATH10212 Linear Algebra B Proof Problems

Session 5. (1) Principal component analysis and Karhunen-Loève transformation

ECE 901 Lecture 14: Maximum Likelihood Estimation and Complexity Regularization

Lecture 12: September 27

Statistical Properties of OLS estimators

ECE 901 Lecture 13: Maximum Likelihood Estimation

Maximum Likelihood Estimation

Lecture 18: Sampling distributions

Bayesian Methods: Introduction to Multi-parameter Models

Direction: This test is worth 150 points. You are required to complete this test within 55 minutes.

A statistical method to determine sample size to estimate characteristic value of soil parameters

A RANK STATISTIC FOR NON-PARAMETRIC K-SAMPLE AND CHANGE POINT PROBLEMS

Lecture 20: Multivariate convergence and the Central Limit Theorem

LECTURE 8: ASYMPTOTICS I

Transcription:

Matrix Represetatio of Data i Experimet Cosider a very simple model for resposes y ij : y ij i ij, i 1,; j 1,,..., (ote that for simplicity we are assumig the two () groups are of equal sample size ) Y X x3 1, 1 if 1 The first colum of the matrix X is all 1 s i.e. X 1 1 0 1 1 0 1 0 1 1 0 1 Sice the sum of the last two colums of X is the first colum of X, we have that rakx We seek a vector is miimized. 1 of estimates SSE S e y ij y ij i1 j1 ow where SSE y ij i i1 j1 y ij i We must therefore solve SSE 0, SSE 1 0, SSE 0 i.e. 3 ormal equatios (.E. s). 7

These become y ij 1...1 i1 j1 y 1 y 1j 1... j1 y y j...3 j1 otice the patter these ormal equatios follow; we will see similar patters throughout this course. We ca summarize these 3 ormal equatios usig oe matrix equatio: i.e. XY XX 1 OTE: XX 0 is sigular so we caot ivert it. (We ca use a pseudo-iverse but it is ot 0 uique). We are cocered with 1 1 3 gives us 1 y 1 y ow we ca write ad i1 y ij YY j1 YI x Y y YJ xy where J x is a square matrix of order with all etries beig 1 s. ow we ca write YY YA 1 YYA YYA 3 Y where A 1 J x A 1 J x J x J x J x 8

A 3 I x Jx 0 0 I x Jx ow ote that A i A i, i 1,,3, i.e. they are all idempotet Also A i A j 0,i j 1,, 3 their associated q.f. s are idepedetly distributed. So is tested usig H o : 1 YA Y YA 3 Y/ ~ F 1.d.f. uder H o these quadratic forms (q.f s.) i a ormal radom variable (r.v.),whe divided by, are distributed as ChiSquared ( ) with degrees of freedom (d.f.) equal to the rak of the matrix associated with the q.f. if idepedet, their ratios follow a F distributio. 9

Liear Models ad Quadratic Forms Mai Results: YY YA 1 YYA YYA 3 Y... 1. a. YA i Y, with proper divisors, are distributed as with d.f. rak(a i b. the quadratic forms YA i Y are idepedet c. the correspodig ratios have F distributios Liear Models: express Y as a liear combiatio of j s i.e. Here: Y is a vector of observatios, y i 1 x i1 x i... p x ip i for i 1,..., i.e. Y X X xp is a matrix of kow costats (X is called the desig matrix), is a vector of parameters ad is a vector of radom errors. I experimetal desig, X cosists of oly 1s ad 0 s. 1. ote:. There is o restrictio that the model be a liear fuctio of the x i 3. There is o restrictio that the x i be idepedet. ote: Study of the experimetal situatio must motivate the model. A liear model ivolves a model equatio with associated assumptios that state the ature of the radom compoet ad the restrictios the parameters must satisfy. 10

Distributio Assumptios: Assumptios re i : y i x ij j i j 1. E i 0iE 0. E i j 0i j ucorrelated 3. E i ihomogeeous variace 4. E I Hece EY X ote that at this poit we make o assumptio of a ormal distributio. If Z is a vector of r.v. s: the mea vector is EZ ad the variace-covariace matrix is so If A is a matrixaz exists ad covz V covz i, z j v ij EAZ AEZ covaz AV A 5. Ofte we assume ormality. We assume the errorshave a multivariate ormal distributo with E 0, ad cov I i.e. 0,I or equivaletly, Y X,I The assumptio of ormality is importat for testig ad estimatio but ot for least squares estimatio or for subdivisio of the sum of squares (SS). The method of maximum likelihood, uder the assumptio of ormality, gives the same estimators of estimable fuctios as the method of least squares. 11

Method of Least Squares Estimate j by choosig estimator j which miimizes the sum of squares of the residuals ( S e SSE Se y i y i where y i x ijj, j 1,..., p YY YY ee j Solve the set of p equatios obtaied by settig Se j I geeral, the j th equatio is x ij y i ie. the ormal equatios are writte i matrix otatio as i1 0. (These equatios are called the ormal equatios.) p s1 i1 XY XX x is x ij s XX is pxp symmetric ad rxx rx umber of liearly idepedet colums of X. A. The Full Rak Case (this is the usual case i multiple regressio but ot i experimetal desig) If the p colums of X are liearly idepedet, rx rxx p thereforexx 1 exists ad the ormal equatios have a uique set of solutios give by XX 1 XY the ad E E XX 1 XY XX 1 XEY XX 1 XX cov cov XX 1 XY XX 1 X I XX 1 X XX 1 XI X XX 1 XX 1 XI X XX 1 XX 1 ote: We should choose experimetal poits with this i mid - i.e. choose X so as to miimize the average variace of the s s 1

The sum of squares of error is give by S e YY YY YYYYYYYY where YX But YY YX YX XX 1 XY YXXX 1 X Y YY ad YY X X XX XX 1 XY XX XX 1 XY YX XX 1 XX XX 1 XY YX XX 1 XY YY or YY Thus S e YYYY Y IXXX 1 X Y or YY YYS e where YY represets the sum of squares due to regressio Thus we have that the total (ucorrected) sum of squares is give by YY Y XXX 1 X YY IXXX 1 X Y S r S e 13

ow ES r E Y XXX 1 X Y XX p ES e E Y IXXX 1 X Y p Sice XX is positive defiite,xx is oegative ad equals zero oly if 0. Uder the assumptio of ormality, S r S e ocetral pd.f cetral pd.f ad the two q.f. s are idepedetly distributed Therefore to test H 0 : 0 we have test statistic F S e S r /p / p ocetral F p,pd.f This result uses the followig regardig quadratic forms i ormal radom variables: If Y, I the YAY md.f iff A is idempotet of rak m YAY ad YBY are idepedetly distributed iff AB 0 ote that we have S e Y IXXX 1 X Y YBY S r Y XXX 1 X Y YAY 14