You are on page 1of 12

This article was downloaded by: [Monash University Library]

On: 18 September 2013, At: 06:59


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer
House, 37-41 Mortimer Street, London W1T 3JH, UK

Communications in Statistics - Theory and Methods


Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/lsta20

Some inference results on pr(x < y) in the bivariate


exponential model
a b c
A. M. Awad , M. M. Azzam & M. A. Hamdan
a
Yarmouk University, Irbid, Jordan
b
University of Jordan, Amman, Jordan
c
University of Jordan, Amman, Jordan
Published online: 27 Jun 2007.

To cite this article: A. M. Awad , M. M. Azzam & M. A. Hamdan (1981) Some inference results on pr(x < y) in the bivariate
exponential model, Communications in Statistics - Theory and Methods, 10:24, 2515-2525

To link to this article: http://dx.doi.org/10.1080/03610928108828206

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of
the Content. Any opinions and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied
upon and should be independently verified with primary sources of information. Taylor and Francis shall
not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other
liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
COMMUN. STATIST.-THEOR. METH., A10(24), 2515-2525 (1981)

SOYE INFERENCE RESULTS ON Pr(X < Y)


IN THE BIVARIATE EXPONENTIAL MODEL

A. M. Awad M. M. Azzam
Yarmouk University University of Jordan
Irbid, Jordan Amman, Jordan
M. A. Hamdan
University of Jordan
Amman, Jordan
Downloaded by [Monash University Library] at 06:59 18 September 2013

Key Words and Phrases: e s t i m a t o r ; t e s t s f o r e q u a l i t y o f means;


b i v a r i a t e exponential: t r i n o m i a l , r e l i a b i l i t y .

ABSTRACT

This paper provides three different estimators for Pr(X < Y)


when X and Y have a bivariate exponential distribution. The asymp-
totic variances of the three estimators are also derived. A test
for the equality of the means of X and Y and confidence limits for
the difference of the two means are presented. Our results are
directly applicable in a reliability context with underlying bi-
variate exponential distribution.

1. INTRODUCTION

Recently a number of papers dealt with the problem of esti-


mating P1 = Pr(X < Y) in the normal case, that is, when X and Y
are independent normal variates. Church and Harris [ Z ] derived

Copyright @ 1981 by Marcel Dekker, Inc.


2516 AWAD, AZZAM, AND HAMDAN

the maximum likelihood estimator (NLE) of PI, while Downton [3]


derived the uniformly minimum variance unbiased estimator (UMVE).
Both papers are based on the assumption that the distribution of
Y is known and a random X-sample is observed. Woodward and
Kelly [ 9 ] derived an alternative expression for the UMVU esti-
mator of Downton.
The problem of estimating P1 in the exponential case, that
is, X and Y are independent exponential variates has been con-
sidered in some papers. Tong [ 8 ] derived two expressions for
the W U E of PI. Johnson [ 4 ] made a correction to one of the
expressions given by Tong. Kelly, and Schucany [ 5 ] derived the
MLE and the UMVUE for PI.
Downloaded by [Monash University Library] at 06:59 18 September 2013

In this paper we derive three estimators of P1 when X and


Y have a bivariate exponential distribution. The bivariate
exponential distribution was introduced by Marshall and Olkin
[6] as a bivariate model for the lifetimes of the components in
a two-component system. It was assumed that the system is sub-
jected to shocks governed by three independent Poisson processes
with parameters hl, A 2 and X3. It was further assumed that
these three types of shocks were fatal to the first component,
the second component and to both components, respectively. This
model was called by Marshall and Olkin "The fatal shock model".
They also described a "non-fatal shock model" and a "Residual
life independent of age model", which lead to the same form of
the bivariate exponential distribution. If X and Y are the
lifetimes of the two components, then,

-- exp [ - Xlx - X 2y - X3 max(x,y)].

It follows that
and

where
Downloaded by [Monash University Library] at 06:59 18 September 2013

Let P = (PI, P2, P3). In this paper we derive three esti-


mators of P when the distribution of X and Y is given by (1.1).
Section 2 gives the estimator P which is based on the maximum
likelihood estimators of the X's given in [I]. Section 3 gives
the estimator P based on the moment type estimators of the A's
given in [I]. In Section 4 , we introduce a Mann-Whitney type
estimator P
*.
The asymptotic variances of the three estimators are
formally derived. Based on P", we give a test for Ho : X 1 = A2
(Section 5), and confidence limits for X - X (Section 6).
1 2
Some concluding remarks pertaining to comparison of the three
estimators are given in Section 7

2. 5IAXIMUM LIKELIHOOD ESTIMATES

Bemis, Bain and Higgins [ I ] have obtained the maximum likeli-


hood estimates for X1, A2 and X3 as the solutions of the equations
2518 AWAD, AZZAM, AND HANDAN

and

where
Downloaded by [Monash University Library] at 06:59 18 September 2013

and

0 otherwise

Proschan and Sullo [7] gave an iterative procedure for cal-


culating
A n
i1' i2 and XA 3 - A
Since the ?fLE of g(X1, X2, X ) is
A 3
g(X19 X2, A3) where 1 X2, X 3 are the MLE's for X1, X2 and X3,
the \fLE1s for P1, P and P are il/i, i2/i and i3/i,
2 3
respectively.
A A

Formally, the asymptotic variances of P P


1' 2 and P3 can be
obtained from the Fisher information matrix.

3. MOMENT TYPE ESTIMATES

and P may not be easy to find,


Since the MLE's for P1, P
2 3- - -
one may use the moment type estimates A1, X and X given by
2 3
Bemis, Bain and Higgins [l] to get the estimates
To o b t a i n t h e asymptotic v a r i a n c e s of t h e s e e s t i m a t o r s ,
* -
let L = (il, A ~ ,, p = ,,/A, y = A + A , y2 = h 2 + l3 .

d P - 1
-- - .
Downloaded by [Monash University Library] at 06:59 18 September 2013

Then v = A2
-,

Suppose t h e m a t r i c e s A and B a r e given by

and
2520 AWAD, AZZAM, AND HAMDAN

Then following Bemis, Bain and Higgins [ I ] , the


asymptotic variance of P is

-
Var ( P ) = -1 v B A B' v'.
n

4. MANN-WHITNEY TYPE ESTIMATES

It may' be remarked that the variances of the moment


type estimates need a lot of computation. Hence we suggest
using a non-parametric approach which is very easy to apply.
It is clear that n n2 and n3 have a trinomial (n;P P ,P )
1' 2 3
Downloaded by [Monash University Library] at 06:59 18 September 2013

distribution. Hence, the MLE's of PI, P2 and P are


3

P
* *
= n /n, P2 = n /n and P3 = n3/n
Ji
.
1 1 2

These estimates are unbiased and have variances

5. TEST FOR EQUALITY OF MEANS

Bain, Bemis and Higgins [I] introduced a test for zero


correlation in the bivariate exponential distribution based on
the moment type estimator P3 n3/n (which is identical to P3) .
=
*
In some practical situations, it may be necessary to test for
the equality of the two mean failure components in a two com-
ponent system governed by the bivariate exponential model. In
what follows, we shall employ the multinomial MLE's P* and P*
1 2
to derive a simple test for the hypothesis Ho:A1 = A
2'
Clearly, H is equivalent to P1 = P2, and hence the problem
reduces to testing equality of two proportions in a tri-
nomial distribution. The likelihood ratio criterion reduces to
Pr(X < Y).

This is a Mc-Nemnr type statistic, and it can be shown


that -2 In R is asymptotically equivalent to [c.f. Kendall and
Stuart (1973) ]

2
Z is asymptotically distributed as a chi-squared random
variable with one degree of freedom.
Downloaded by [Monash University Library] at 06:59 18 September 2013

6. CONFIDENCE LIMITS FOR (Al - X2)

We may construct asymptotic confidence limits for


1 1
(Al - h2) based on the difference ( - , and this involves
X Y
the asymptotic variance of this difference. A simpler formu-
*
lation may be derived by using P1 - P;. In fact

where

and
D
-
4 means converge in distribution.

Hence approximate (1 - a) 100% confidence limits for


P1 - P2 = (A1 - X2)/A are

where Z is the upper a12 percentage point of N(0,l).


a/ 2
To obtain confidence limits for (A1 - A2) from (6.2) an
2522 AWAD, AZZAM, AND HAMDAN

estimate for X is needed. An estimate for X can be obtained


through estimating XI, A and A using the equations
2 3

These however are only two algebraically independent equations.


Thus, we have only the first two equations, say. A third
equation may be (arbitrarily) chosen by the method of moments.
We suggest an equation based on combined information from both
marginals, namely,
Downloaded by [Monash University Library] at 06:59 18 September 2013

-
-L = Xl+X
3
and -& = A2 + X
3
X Y

which lead to the third equation

The first two equations of (6.3) together with (6.5) lead


t0

This gives an estimate of X (based partially on the M L E ) ,


namely,
Substituting (6.6) into (6.2), we obtain

as an approximate (1 - a) 100% confidence interval for


Downloaded by [Monash University Library] at 06:59 18 September 2013

7. CONCLUDING REMARKS

The estimators given in Section 2, {Pi}, are based on the


maximum likelihood estimators {iij, a d hence all information
in the bivariate sample { (Xi, Yi) , i = 1, . . . , n} is utilized,
that is each individual sample element is used in the calcu-
lation of {ii}. The estimators in Section 3, {Pi}, called
moment type estimators, are based on estimating the exponen-
tial parameters through equating C Xi' C Yi and n3 (the number
of bivariate sample elements X = Y.) to their expected values.
i 1
Obviously, the sums C X. and C Y. involve information on the
marginal distributions of X and Y respectively. Thus, only
n3 involves bivariate information used in the estimators { P . ] .
* are indeed
The estimators introduced in Section 4, {Pi),
1

maximum likelihood estimators of the multinomial parameters


{Pi}, which are based on n n2 and ng, the numbers of sample
elenents with X < Yi, Xi > Yi and Xi = Yi, respectively.
i
Thus, the multinomial maximum likelihood estimators, P I, are
entirely based on information derived from the bivariate sample.
Although, we have formally derived above the asymptotic vari-
-
antes of the estimators (P.1 and {P.}, these variances can not
1
2524 AWAD, AZZAM, AND W A N

be analytically compared because their formulation is rather


involved. Such comparison is not necessary, however, since
*
the estimators {P.} are to be recommended as the ML estimators
1
for {P.), which are consistent, unbiased, jointly sufficient
1
and efficient, and their variances are simply {P.(l - Pi)/ni,}.

ACKNOWLEDGMENT

The authors are indebted to the referee for constructive


criticism which helped to improve the first draft of this paper.
Downloaded by [Monash University Library] at 06:59 18 September 2013

BIBLIOGRAPHY
- --

Bemis, B. M., Bain, L. J., and Higgins, J. J. (1972).


Estimation and hypothesis testing for the parameters of
a bivariate exponential distribution. J. --
- bf Amer. Statist.
Assoc., 67,927-929.

Church. J. D. and Harris. B. (1970). The estimation of reli-


ability from stress-strength relationships. Technometrics,
12, 49-54.
-

Downton, F. (1973). The estimation of Pr(Y < X) in the normal


case. Technometrics, 15, 551-558.

Johnson, N. L. (1975). Letter to the editor. Technometrics,


17, 393.
-
Kelly, G. D., Kelly, J. A. and Schucany, W. R. (1976). Effi-
cient estimation of P(Y < X) in the exponential case.
Technometrics, 18,359-360.

Marshall, A. W. and Olkin, I. (1976). A multivariate exponen-


.tial distribution. .J.of Amer. Statist. Assoc., 62, 30-44.

Proschan, F. and Sullo, P. (1976). Estimating the parameters


of a multivariate exponential distribution. J. of Amer.
Statist. Assoc., 2, 465-472.

Tong, H. (1974). A note on the estimation of P(Y < X) in the


exponential case. Technometrics, 16,625.
Woodward, W. A. and Kelly, G. D. ( 1 9 7 7 ) . Minimum variance
unbiased estimation of P(Y < X) in the normal case.
Technometrics, 2, 95-98.

R e c e i v e d O c t o b e r , 1 9 7 9 ; R e v i s e d May, 1 9 8 0 a n d October, 1980;


R e t y p e d May, 1 9 8 1 .

R e f e r e e d b y Gary D. Kelley, Texas Tech University,


L u b b o c k , TX.
Downloaded by [Monash University Library] at 06:59 18 September 2013

You might also like