ASYMPTOTIC BEST LINEAR UNBIASED ESTIMATION FOR THE LOG-GAMMA DISTRIBUTION

Σχετικά έγγραφα
On Generating Relations of Some Triple. Hypergeometric Functions

Homework for 1/27 Due 2/5

Solutions: Homework 3

Other Test Constructions: Likelihood Ratio & Bayes Tests

1. For each of the following power series, find the interval of convergence and the radius of convergence:

INTEGRATION OF THE NORMAL DISTRIBUTION CURVE

α β


SUPERPOSITION, MEASUREMENT, NORMALIZATION, EXPECTATION VALUES. Reading: QM course packet Ch 5 up to 5.6

LAD Estimation for Time Series Models With Finite and Infinite Variance

Statistical Inference I Locally most powerful tests

Lecture 17: Minimum Variance Unbiased (MVUB) Estimators

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

On Certain Subclass of λ-bazilevič Functions of Type α + iµ

4.6 Autoregressive Moving Average Model ARMA(1,1)

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

On Inclusion Relation of Absolute Summability

ST5224: Advanced Statistical Theory II

Introduction of Numerical Analysis #03 TAGAMI, Daisuke (IMI, Kyushu University)

A study on generalized absolute summability factors for a triangular matrix

Diane Hu LDA for Audio Music April 12, 2010

Solutions to Exercise Sheet 5

2 Composition. Invertible Mappings

SOLUTIONS TO MATH38181 EXTREME VALUES AND FINANCIAL RISK EXAM

: Monte Carlo EM 313, Louis (1982) EM, EM Newton-Raphson, /. EM, 2 Monte Carlo EM Newton-Raphson, Monte Carlo EM, Monte Carlo EM, /. 3, Monte Carlo EM

SOLUTIONS TO MATH38181 EXTREME VALUES AND FINANCIAL RISK EXAM

n r f ( n-r ) () x g () r () x (1.1) = Σ g() x = Σ n f < -n+ r> g () r -n + r dx r dx n + ( -n,m) dx -n n+1 1 -n -1 + ( -n,n+1)

The Heisenberg Uncertainty Principle

Bessel function for complex variable

Gauss Radau formulae for Jacobi and Laguerre weight functions

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

Ordinal Arithmetic: Addition, Multiplication, Exponentiation and Limit

EE512: Error Control Coding

IIT JEE (2013) (Trigonomtery 1) Solutions

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

Ψηφιακή Επεξεργασία Εικόνας

p n r

6. MAXIMUM LIKELIHOOD ESTIMATION

Exercises to Statistics of Material Fatigue No. 5

6.1. Dirac Equation. Hamiltonian. Dirac Eq.

6.3 Forecasting ARMA processes

ANOTHER EXTENSION OF VAN DER CORPUT S INEQUALITY. Gabriel STAN 1

Solve the difference equation

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

C.S. 430 Assignment 6, Sample Solutions

Variance Covariance Matrices for Linear Regression with Errors in both Variables. by J.W. Gillard and T.C. Iles

The Equivalence Theorem in Optimal Design

A New Class of Analytic p-valent Functions with Negative Coefficients and Fractional Calculus Operators

Binet Type Formula For The Sequence of Tetranacci Numbers by Alternate Methods

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

Research on Economics and Management

derivation of the Laplacian from rectangular to spherical coordinates

Concrete Mathematics Exercises from 30 September 2016

Second Order Partial Differential Equations

Section 8.3 Trigonometric Equations

Outline. Detection Theory. Background. Background (Cont.)

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

The Neutrix Product of the Distributions r. x λ

CRASH COURSE IN PRECALCULUS

Certain Sequences Involving Product of k-bessel Function

HOMEWORK 4 = G. In order to plot the stress versus the stretch we define a normalized stretch:

Queensland University of Technology Transport Data Analysis and Modeling Methodologies

On a four-dimensional hyperbolic manifold with finite volume

FORMULAS FOR STATISTICS 1

Aquinas College. Edexcel Mathematical formulae and statistics tables DO NOT WRITE ON THIS BOOKLET

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

K. Hausdorff K K O X = SDA. symbolic data analysis SDA SDA. Vol. 16 No. 3 Mar JOURNAL OF MANAGEMENT SCIENCES IN CHINA

HW 3 Solutions 1. a) I use the auto.arima R function to search over models using AIC and decide on an ARMA(3,1)

( y) Partial Differential Equations

CHAPTER 103 EVEN AND ODD FUNCTIONS AND HALF-RANGE FOURIER SERIES

Second-order asymptotic comparison of the MLE and MCLE of a natural parameter for a truncated exponential family of distributions

Μηχανική Μάθηση Hypothesis Testing

Homework 3 Solutions

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

FREE VIBRATION OF A SINGLE-DEGREE-OF-FREEDOM SYSTEM Revision B

Uniform Estimates for Distributions of the Sum of i.i.d. Random Variables with Fat Tail in the Threshold Case

Στα επόμενα θεωρούμε ότι όλα συμβαίνουν σε ένα χώρο πιθανότητας ( Ω,,P) Modes of convergence: Οι τρόποι σύγκλισης μιας ακολουθίας τ.μ.

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

Introduction to the ML Estimation of ARMA processes

5.4 The Poisson Distribution.

Finite Field Problems: Solutions

Jesse Maassen and Mark Lundstrom Purdue University November 25, 2013


CHAPTER 25 SOLVING EQUATIONS BY ITERATIVE METHODS

HOMEWORK#1. t E(x) = 1 λ = (b) Find the median lifetime of a randomly selected light bulb. Answer:

Statistics 104: Quantitative Methods for Economics Formula and Theorem Review

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

ANSWERSHEET (TOPIC = DIFFERENTIAL CALCULUS) COLLECTION #2. h 0 h h 0 h h 0 ( ) g k = g 0 + g 1 + g g 2009 =?

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

An Inventory of Continuous Distributions

Εργαστήριο Ανάπτυξης Εφαρμογών Βάσεων Δεδομένων. Εξάμηνο 7 ο

Appendix to On the stability of a compressible axisymmetric rotating flow in a pipe. By Z. Rusak & J. H. Lee

DESIGN OF MACHINERY SOLUTION MANUAL h in h 4 0.

«ΑΓΡΟΤΟΥΡΙΣΜΟΣ ΚΑΙ ΤΟΠΙΚΗ ΑΝΑΠΤΥΞΗ: Ο ΡΟΛΟΣ ΤΩΝ ΝΕΩΝ ΤΕΧΝΟΛΟΓΙΩΝ ΣΤΗΝ ΠΡΟΩΘΗΣΗ ΤΩΝ ΓΥΝΑΙΚΕΙΩΝ ΣΥΝΕΤΑΙΡΙΣΜΩΝ»

forms This gives Remark 1. How to remember the above formulas: Substituting these into the equation we obtain with

MATH 38061/MATH48061/MATH68061: MULTIVARIATE STATISTICS Solutions to Problems on Matrix Algebra

Lecture 34 Bootstrap confidence intervals

Potential Dividers. 46 minutes. 46 marks. Page 1 of 11

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

Homework 4.1 Solutions Math 5110/6830

Transcript:

Sakhyā : The Idia Joural of Statistics 994, Volume 56, Series B, Pt. 3, pp. 34-322 ASYMPTOTIC BEST LINEAR UNBIASED ESTIMATION FOR THE LOG-GAMMA DISTRIBUTION By N. BALAKRISHNAN McMaster Uiversity, Hamilto P.S. CHAN The Chiese Uiversity of Hog Kog SUMMARY. I this paper, we discuss Beett s determiatio of optimal asymptotic weights for the BLUEs of the locatio scale parameters based o geeral Type-II cesored samples the apply this method to the log-gamma distributio.. Itroductio Let Y be a log-gamma rom variable with probability desity fuctio g(y) = (y µ)/σ σγ(κ) e e +κ(y µ)/σ, <y<, <µ<,σ >0,κ>0,...(.) where µ is the locatio parameter, σ is the scale parameter, κ is the shape parameter, or, equivaletly, ( g R (y) = κκ /2 κ bγ(κ) exp y u b ) κe (y u)/b κ, <y<,...(.2) where u = µ + σ log k, b = σ κ....(.3) Lawless (980, 982) has illustrated the usefuless of the log-gamma model i (.) as a life-test model discussed the maximum likelihood estimatio of the parameters; see also Pretice (974). Youg Bakir (987) have discussed the log-gamma regressio model. Oe may also refer to Lawless (980) DiCiccio (987) for some valuable work o the iferece for a related geeralized gamma distributio. Paper received. October 992; revised August 993. AMS (990) subject classificatio. 62F0. Keywords phrases. Log-gamma distributio, asymptotic best liear ubiased estimators (ABLUEs), type-ii cesored samples, life-test data.

best liear ubiased estimatio 35 Recetly, Balakrisha Cha (994) have studied order statistics from the log-gamma distributio determied their meas, variaces covariaces, used these quatities to determie the best liear ubiased estimators (BLUEs) of µ σ i (.) based o complete as well as Type-II cesored samples. They have also discussed the liear estimatio of µ σ based o k optimally selected order statistics. I this paper, we discuss asymptotic approximatios to the BLUEs of µ σ based o Type-II cesored samples. 2. Asymptotic best liear ubiased estimatio I this sectio, we shall cosider a Type II cesored sample Y r+: Y r+2: Y :...(2.) from a populatio with c.d.f F ((y µ)/σ) p.d.f f((y µ)/σ)/σ briefly describe Beett s determiatio of optimal asymptotic weights for the BLUEs of µ σ. Let us deote X i: = (Y i: µ)/σ, α i: = E(X i: ), β i,j: =Cov (X i:,x j: ). The the best liear ubiased estimators of µ σ for the cesored sample i (2.) are (see Lloyd, 952; David, 98, p. 29; Balakrisha Cohe, 99, p. 80-8) give by { α µ Ωα Ω α Ωα } Ω = (α Ωα)( Ω) (α Ω) 2 Y...(2.2) where σ = { Ωα Ω Ωα } Ω (α Ωα)( Ω) (α Ω) 2 Y...(2.3) Y = (Y r+:,y r+2:,...,y : ), α = (α r+:,α r+2:,...,α : ), = (,...,) r s, Ω =(ω i,j )=(β i,j: ), r + i, j s. The variaces covariace of µ σ i (2.2) (2.3) are (see Lloyd, 952; David, 98, p. 30; Balakrisha Cohe, 99, p. 8) give by { var(µ )=σ 2 α } Ωα (α Ωα)( Ω) α Ω) 2,...(2.4) { var(σ )=σ 2 } Ω (α Ωα)( Ω) α Ω) 2,...(2.5)

36. balakrisha p.s. cha { cov(µ,σ )= σ 2 α } Ω (α Ωα)( Ω) (α Ω) 2....(2.6) As, i, j such that i/ p i, j/ p j, 0 <p i < p j <, we have the expected value of X i: the covariace of X i: X j: up to order / as (see David, 98, p. 80; Arold Balakrisha, 989, p. 75-76) α i: = F (p i )=G i...(2.7) β i,j: = p iq j,i j, +2f(G i )f(g j )...(2.8) where q i = p i, G i = F (p i ) for r + i s. The Ω may be algebraically worked out usig (2.8), is give by { ω i,i = f 2 (G i ) p i+ p i + ω i,i = ω i,i = f(gi)f(gi ) p i p i, ω i,j = 0 otherwise, p i p i },...(2.9) where p r = 0 p + =. Usig (2.9) lettig r/ λ ()/ λ 2 as, it ca be easily show that Ω = α Ω = α Ωα = i=r+ {f(g i+ ) f(g i )} 2 p i+ p i + f 2 (G r+ ) p r+ + f 2 (G ) p = λ 2 λ Ψ 2 (v)du + f 2 (G(λ )) λ + f 2 (G(λ 2)) λ 2 = I (say) {f(g i+ ) f(g i )}{G i+ f(g i+ ) G i f(g i )} p i+ p i i=r+ + Gr+f 2 (G r+) p r+ + Gf 2 (G ) p = λ 2 λ Ψ(v)( + vψ(v))du + G(λ )f 2 (G(λ )) λ + G(λ 2)f 2 (G(λ 2 )) λ 2 = I 2 {G i+ f(g i+ ) G i f(g i )} 2 i=r+ + G2 r+ f 2 (G r+) p r+ p i+ p i + G2 f 2 (G ) p = λ 2 λ ( + vψ(v)) 2 du + G2 (λ )f 2 (G(λ )) λ + G2 (λ 2 )f 2 (G(λ 2 )) λ 2 = I 22 ;...(2.0)...(2.)...(2.2)

best liear ubiased estimatio 37 i the above formulae, v = G(u) =F (u), G (u) =df (u)/du =/f(g(u)), Ψ(v) =f (v)/f(v). The fact that d{+vf(v)}/du =+vψ(v) is also used i the derivatio of these formulae. From Eqs. (2.2) (2.3), we ca the write the BLUEs of µ σ as where, for i = r +,..., s, γ i = (I I 22 I2 2 ) I 22 µ = σ = j=r+ i=r+ i=r+ γ i Y i:...(2.3) δ i Y i:...(2.4) ω i,j I 2 j=r+ α j: ω i,j...(2.5) δ i = (I I 22 I2 2 ) I j=r+ α j: ω i,j I 2 j=r+ ω i,j....(2.6) Usig (2.9), the quatities i γ i δ i ca be further simplified as j=r+ ω i,j f(g(ξ i )) d2 f(g(ξ i)) dξ dξi 2 i = φ (ξ i )dξ i (say) φ (ξ i )/...(2.7) with i =[ξ i ]+, [m] idicatig the itegral part of m, where j=r+ j=r+ Ω r+,j φ (ξ r+ )dξ r+ + f 2 (G r+) p r+ f (G r+ ) φ (ξ r+ ) + φ,r+ ω,j φ (ξ )dξ + f 2 (G ) p + f (G ) φ(ξ) + φ,, φ,r+ = f 2 (G r+ ) p r+ f (G r+ ), φ, = f 2 (G ) p + f (G )....(2.8)...(2.9)

38. balakrisha p.s. cha Moreover, j=r+ α j: ω i,j f(g(ξ i )) d2 {G(ξ i )f(g(ξ i ))} dξ dξi 2 i = φ 2 (ξ i )dξ i (say) φ 2 (ξ i )/,...(2.20) where j=r+ j=r+ α j: ω r+,j φ 2 (ξ r+ )dξ r+ + G r+f 2 (G r+ ) p r+ {f(g r+ )+G r+ f (G r+ )} + φ 2,r+ φ 2(ξ r+ ) α j: ω,j φ 2 (ξ )dξ + Gf 2 (G ) p +{f(g )+G f (G )} + φ 2,, φ 2(ξ ) φ 2,r+ = G r+f 2 (G r+ ) p r+ {f(g r+ )+G r+ f (G r+ )},...(2.2)...(2.22) φ 2, = G f 2 (G ) + {f(g )+G f (G )}. p I the above derivatios, we may ote that the fuctios φ φ 2 are of the form (with v = G(u) =F (u) Ψ(v) =f (v)/f(v), as before) φ (u) = f(v) d2 f(v) du 2 = Ψ (v)...(2.23) φ 2 (u) = f(v) d2 {vf(v)} du 2 = {Ψ(v)+vΨ (v)}....(2.24) Hece, from Eqs. (2.5) (2.6), we may determie asymptotically the coefficiets γ i δ i (except for the two extreme coefficiets, viz., for i = r + s) by settig u = p i = i/( + ) i the followig cotiuous weight fuctios γ(u) = φ (u)i 22 φ 2 (u)i 2 (I I 22 I 2 2 )...(2.25) δ(u) = φ 2(u)I φ (u)i 2 (I I 22 I 2 2 )....(2.26)

best liear ubiased estimatio 39 For the two extreme cases we may determie asymptotically the coefficiets by the formulas γ r+ = γ(ξ r+ )+ φ,r+i 22 φ 2,r+I 2 I I 22 I2 2...(2.27) γ = γ(ξ )+ φ,i 22 φ 2,I 2 I I 22 I 2 2...(2.28) δ r+ = δ(ξ r+ )+ φ 2,r+I φ,r+i 2 I I 22 I2 2...(2.29) δ = δ(ξ )+ φ 2,I φ,i 2 I I 22 I2 2....(2.30) By usig (2.4) - (2.6) (2.0) - (2.2), we ca easily obtai the asymptotic variaces covariace of the estimators µ σ i (2.3) (2.4) to be var(µ ) I 22 σ 2 = I I 22 I2 2 var(σ ) I σ 2 = I I 22 I2 2...(2.3)...(2.32) cov(µ,σ ) I 2 σ 2 = I I 22 I2 2....(2.33) 3. Derivatio for the log-gamma distributio I this sectio, we derive exact explicit expressios for the various fuctios itroduced i the last sectio; these formulae make the computatio of the asymptotic BLUEs for this case a lot simpler. For this purpose, we establish the followig three lemmas which will be repeatedly used i the algebraic maipulatios to follow i this sectio. Lemma. For κ > 0, let F κ (λ) = λ e κy ey Γ(κ) dy. The F κ+ (λ) =F κ (λ) eκλ eλ Γ(κ +). Lemma 2. For κ > 0, let G κ (λ) = λ ye κy ey dy. Γ(κ)

320. balakrisha p.s. cha The G κ+ (λ) =G κ (λ)+ κ F κ(λ) λeκλ eλ Γ(κ +), where F κ (λ) is as defied i Lemma. Lemma 3. For κ > 0, let H κ (λ) = λ y 2 e κy ey dy. Γ(κ) The H κ+ (λ) =H κ (λ)+ 2 κ G κ(λ) λ2 e κλ eλ Γ(κ +), where G κ (λ) is as defied i Lemma 2. Now, for the stard log-gamma distributio with desity fuctio f(x), with kow shape parameter κ, wehave Hece, the fuctio Ψ(v) becomes f (x) =f(x)(κ e x )....(3.) Ψ(v) = f (v) f(v) = κ ev...(3.2) Let us deote ξ λ ξ λ2 by Fκ (λ ) Fκ (λ 2 ), respectively, where Fκ ( ) is iverse c.d.f of the log-gamma distributio with shape parameter κ. Sice v = Fκ (u) dv =(/f(v))du, applyig Lemma, the fiite itegral i Eq. (2.3) ca be writte as λ Ψ 2 (v)du = ξ λ2 ξ λ (κ 2 2κe v + e 2v )f(v)dv = κ 2 [λ 2 λ ] 2κ 2 [F κ+ (ξ λ2 ) F κ+ (ξ λ )] +κ(κ + )[F κ+2 (ξ λ2 ) F κ+2 (ξ λ )]...(3.3) = κ(λ 2 λ )+ κ Γ(κ) [eκξ λ 2 e κξ λ ] Γ(κ) [e(κ+)ξ λ 2 e (κ+)ξ λ ]. Similarly, usig Lemmas 2, the fiite itegral i (2.4) ca be writte as λ Ψ(v)[ + vψ(v)]du = κ(λ 2 λ ) κ[f κ+ (ξ λ2 ) F κ+ (ξ λ )] +κ 2 [G κ (ξ λ2 ) G κ (ξ λ )] 2κ 2 [G κ+ (ξ λ2 ) G κ+ (ξ λ )] + κ(κ + )[G κ+2 (ξ λ2 ) G κ+2 (ξ λ )] = (λ 2 λ )+κ[g κ (ξ λ2 ) G κ (ξ λ )] + κ(κ ) Γ(κ) [ξ λ 2 e κξ λ 2 ξ λ e κξ λ ] Γ(κ) [ξ λ 2 e (κ+)ξ λ 2 ξ λ e (κ+)ξ λ ]....(3.4)

best liear ubiased estimatio 32 Fially, usig Lemmas, 2 3, the fiite itegral i (2.5) ca be writte as λ [ + vψ(v)] 2 du = (λ 2 λ )+2κ[G κ (ξ λ2 ) G κ (ξ λ )] 2κ[G κ+ (ξ λ2 ) G κ+ (ξ λ )] + κ 2 [H κ (ξ λ2 ) H κ (ξ λ )] 2κ 2 [H κ+ (ξ λ2 ) H κ+ (ξ λ )] +κ(κ + )[H κ+2 (ξ λ2 ) H κ+2 (ξ λ )] = (λ 2 λ )+2[G κ (ξ λ2 ) G κ (ξ λ )] +κ[h κ (ξ λ2 ) H κ (ξ λ )] + κ Γ(κ) [ξ2 λ 2 e (κ)ξ λ 2 ] ξλ 2 e (κ)ξ λ ] Γ(κ) [ξ2 λ 2 e (κ+)ξ λ 2 ξλ 2 e (κ+)ξ λ ]....(3.5) Hece, we ca evaluate I,I 2 I 22 if we kow the values of G κ (ξ λ ) H κ (ξ λ ). Note that [G κ (ξ λ2 ) G κ (ξ λ ]= λ ye κy e y dy Γ(κ) ye κy e y [H κ (ξ λ2 ) H κ (ξ λ ]= dy λ Γ(κ) ca be easily computed by employig stard umerical methods sice both itegrs are cotiuous i ay ope iterval. Ackowledgemets. The first author would like to thak the Natural Scieces Egieerig Research Coucil of Caada for fudig this research. The authors would also like to thak Prof. Bimal K. Siha a referee for suggestig some chages which led to a improvemet i the presetatio of this paper. Refereces Arold, B. C. Balakrisha, N. (989). Relatios, Bouds Approximatios For Order Statistics. Lecture Notes i Statistics 53, Spriger-Verlag, New York. Balakrisha, N. Cohe, A. C. (99). Order Statistics Iferece: Estimatio Methods, Academic Press, Sa Diego. Balakrisha, N. Cha P. S. (994). Log-gamma order statistics liear estimatio of parameters. Computatioal Statistics & Data Aalysis (Revised). David, H. A. (98). Order Statistics, Secod editio, Joh Wiley & Sos, New York. DiCiccio, T. J. (987). Approximate iferece for the geeralized gamma distributio. Techometrics 29, 32-39. Lawless, J. F. (980). Iferece i the geeralized gamma log-gamma distributio. Techometrics 22, 67-82. (982). Statistical Models & Methods For Lifetime Data, Joh Wiley & Sos, New York.

322. balakrisha p.s. cha LLoyd, E. H. (952). Least-squares estimators of locatio scale parameters usig order statistics. Biometrika 39, 88-95. Pretice, R. L. (974). A log gamma model its maximum likelihood estimatio. Biometrika, 6, 539-544. Youg, D. H. Bakir, S. T. (987). Bias correctio for a geeralized log-gamma regressio model. Techometrics 29, 83-9. Departmet of Mathematics Statistics McMaster Uiversity Hamilto, Otario Caada l8s 4K Departmet of Statistics The Chiese Uiversity of Hog Kog Shati Hog Kog