Iraqi Journal of Statistical Science (16) 2010

p.p. [13-28]

Bayes Estimator of one parameter Gamma distribution

under Quadratic and LINEX Loss Function
Wael Abdul Lateef Jasim*

ABSTRACT
In this paper we derive Bayes' estimator for the Scale
parameter θ in Gamma distribution when α is known and equal
2, i.e. Χ 1 , Χ 2 ,..., Χ n ~ Ga (2, θ) , we take α = 2 to estimate one
parameter of gamma distribution which is θ (Scale parameter),
where gamma distribution is considered as an important model of
the life time models . These estimators are obtained depending on
squared error and LINEX loss function , Then comparisons of
risks for θ under squared and LINEX loss function have been
made . Simulation study is given to illustrate that the proposed
estimators θ̂ LB is preferable to θ̂ SB for the sample sizes n = 10,20,30
from above distribution with parameters (α = 2, θ = 1) and for all
values of " a " (a = ±0.5,±1,±2) .
‫ﻤﻘﺩﺭ ﺒﻴﺯ ﻟﻤﻌﻠﻤﺔ ﻭﺍﺤﺩﺓ ﻤﻥ ﺘﻭﺯﻴﻊ ﻜﺎﻤﺎ ﺘﺤﺕ ﺩﺍﻟﺔ ﺍﻟﺨﺴﺎﺭﺓ ﺍﻟﺘﺭﺒﻴﻌﻴﺔ ﻭﺩﺍﻟﺔ ﺨﺴﺎﺭﺓ‬
‫ﻟﻨﻜﺱ‬
‫ﺍﻟﻤﺴﺘﺨﻠﺹ‬
α ‫ ﻓﻲ ﺘﻭﺯﻴﻊ ﻜﺎﻤﺎ ﻋﻨـﺩﻤﺎ‬θ ‫ﺘﻡ ﻓﻲ ﻫﺫﺍ ﺍﻟﺒﺤﺙ ﺍﺸﺘﻘﺎﻕ ﻤﻘﺩﺭﺍﺕ ﺒﻴﺯ ﻟﻤﻌﻠﻤﺔ ﺍﻟﻘﻴﺎﺱ‬
α = 2 ‫ ﻭﻗﺩ ﺘﻡ ﺍﺨـﺫ‬، Χ 1 , Χ 2 ,..., Χ n ~ Ga (2, θ) ‫( ﻫﺫﺍ ﻴﻌﻨﻲ‬2) ‫ﻤﻌﻠﻭﻤﺔ ﻭﺘﺴﺎﻭﻱ‬
‫ ﺍﺫ ﺍﻥ ﺘﻭﺯﻴـﻊ ﻜﺎﻤـﺎ‬،(‫ )ﻤﻌﻠﻤﺔ ﺍﻟﻘﻴﺎﺱ‬θ ‫ﻟﺘﻘﺩﻴﺭ ﻤﻌﻠﻤﺔ ﻭﺍﺤﺩﺓ ﻤﻥ ﺘﻭﺯﻴﻊ ﻜﺎﻤﺎ ﻭﻫﻲ‬
‫ ﻭﻗﺩ ﺘﻡ ﺍﻟﺤﺼﻭل ﻋﻠﻰ ﻫﺫﻩ ﺍﻟﻤﻘﺩﺭﺍﺕ ﺒﺎﻻﻋﺘﻤـﺎﺩ‬. ‫ﻴﻌﺘﺒﺭ ﻨﻤﻭﺫﺠﺎ ﻤﻬﻤﺎ ﻟﻨﻤﺎﺫﺝ ﺍﻟﺤﻴﺎﺓ‬
‫ ﺜﻡ ﺘﻤﺕ ﺒﻌﺩ ﺫﻟﻙ ﻤﻘﺎﺭﻨﺔ ﺍﻟﺨﻁﻭﺭﺍﺕ‬، ‫ﻋﻠﻰ ﺩﺍﻟﺔ ﺍﻟﺨﻁﺄ ﺍﻟﺘﺭﺒﻴﻌﻴﺔ ﻭﺩﺍﻟﺔ ﺨﺴﺎﺭﺓ ﻟﻨﻜﺱ‬
‫ ﻟﻘﺩ ﻗﻤﻨﺎ ﺒﺩﺭﺍﺴﺔ ﻤﺤﺎﻜﺎﺓ ﻭﺘﻭﻀﻴﺢ‬. ‫ﺘﺤﺕ ﺩﺍﻟﺔ ﺍﻟﺨﺴﺎﺭﺓ ﺍﻟﺘﺭﺒﻴﻌﻴﺔ ﻭﺩﺍﻟﺔ ﺨﺴﺎﺭﺓ ﻟﻨﻜﺱ‬
n = 10,20,30 ‫ وﻻﺣﺠ ﺎم ﻋﻴﻨ ﺎت‬θ̂ SB ‫ ه ﻲ ﻤﻔﻀﻠﺔ ﻋﻠـﻰ‬θ̂ LB ‫ﺍﻥ ﺍﻟﻤﻘﺩﺭﺍﺕ ﺍﻟﻤﻘﺘﺭﺤﺔ‬
. (a = ±0.5,±1,±2) " a " ‫( وﻟﺠﻤﻴﻊ ﻗﻴﻢ‬α = 2, θ = 1) ‫ﻣﻦ اﻟﺘﻮزﻳﻊ اﻋﻼﻩ وﺑﻤﻌﻠﻤﺎت‬

*
Assistance Lecturerlechnical Institute/ Mosul
Received:18/2 /2009 ____________________Accepted: 6 /9 / 2009
[14] ________________Bayes Estimator of one parameter...

1- Introduction
In Reliability studies the models which are used in life
testing include the Exponential, Gamma, Lognormal,
…distributions. If the failure is mainly due to aging or wearing
out process, then it's reasonable in many applications to choose
one of the above mentioned distributions (see Chhikara & Folks
(1977), Sinha & Kale (1980), Von Alven (ed.)(1964), Sherif &
Smith (1980) .
The Gamma distribution is the most widely used in life
experiment , it has probability density function (p.d.f) with two
parameters α and θ is :
x α −1 x
f ( x ; α , θ) = α
exp(− ) x ≥ 0 , α, θ ≥ 0
θ Γα θ
In this paper we will use the gamma distribution with one
parameter θ where α is known and equal 2, so it has probability
density function (p.d.f) as follows :
x x
f ( x; θ) = exp(− ) x ≥ 0 , α, θ ≥ 0 (1)
θ 2
θ
And the distribution function of X is :
t t
x x t t t
F( t ) = ∫ f ( x; θ)dx = ∫ exp(− )dx = 1 − exp(− ) − exp(− ) t≥0
0 0 θ
2
θ θ θ θ
The Reliability function, the probability of no failure before time
t is :
t t
R ( t ) = 1 − F( t ) = [1 − ] ∗ exp(− )
θ θ
Then the Hazard function which is the failure rate of a gamma
distribution is :
t t
exp(− )
f (t) θ t
= θ
2
H( t ) = = (2)
R (t) t t θ(θ − t )
[1 − ] ∗ exp(− )
θ θ
In Bayesian estimation, we consider two types of loss functions.
The first is squared error loss function (quadratic loss) which
classified as a symmetric function and associates equal
importance to the losses for overestimation and underestimation
of equal magnitude. The second is the LINEX (linear-
 Iraqi Journal of Statistical Science (16) 2010 _____________ [15]

exponential where the name LINEX is justified by the fact that is

this loss function rises approximately linearly on one side of zero
and approximately exponentially on the other side) loss function
which is asymmetric, was introduce by Varin (1975). These loss
functions were widely used by several authors; among them are
Rojo (1987), Basu and Ebrahimi (1991), Pandy(1997), Soliman
(2000) and Nassar and Eissa (2004).
The quadratic loss for Bayes estimate of a parameter θ , The
posterior mean assuming that exists, denoted by θs . The LINEX
loss function may be expressed as :

L( ∆ ) ∝ exp( c∆ ) − c∆ − 1 c≠0 (3)

Where ∆ = θˆ − θ . The sign and magnitude of the shape parameter
c reflects the direction and degree of asymmetry, respectively.
(If c > 0 ), the overestimation is more serious than
underestimation, and vice-versa). For c closed to zero, the
LINEX loss is approximately squared error loss and therefore
almost symmetric.
The posterior expectation of the LINEX loss function equation
(3) is :
Ε[ L(θˆ − θ)] ∝ exp(cθˆ )Ε[exp( −cθ)] − c(θˆ − Ε(θ)) − 1 (4)
By result of Zellner(1986), the Bayes estimator of θ , denoted by
θ̂ L under the LINEX loss is the value θ̂ which minimize (4), is
given by :
1
θˆ L = − ln{Ε[exp(−cθ)]} (5)
c
When the expectation Ε[exp(−cθ)] exists and finite [see Calabria
and Pulcini(1996)].

2- LINEX loss function and its properties

Thompson and Basu (1996) identified a family of loss
functions L(∆) , where ∆ is either the estimation error (θˆ − θ) or
the relative estimation error (θˆ − θ) / θ , such that
• L(0) = 0
• L(∆) > (<)L(−∆) > 0 for all ∆ > 0
[16] ________________Bayes Estimator of one parameter...

• L(⋅) is twice differentiable with L′(0) = 0 and L′′(∆) > 0 for all
∆≠0 .
• 0 < L′(∆) > (<) − L′(−∆) > 0 for all ∆ > 0 .
Such loss function is useful whenever the actual losses are
nonnegative , increases with estimation error, overestimation is
more (less) serious than under estimation of the same magnitude
and losses increase at a faster (slower) rate with overestimation
error .
Considering the loss function
L∗ ( ∆ ) ∝ b exp( a∆ ) + c∆ + d
and with the restriction L∗ (0) = 0 , (L∗ )′(0) = 0 , we get d = −b and
c = −ab ,see Thompson and Basu (1996). The resulting loss
function is:
L∗ ( ∆ ) ∝ b[exp( a∆ ) − a∆ − 1] (6)
Which is considered as a function of θ and θ̂ , is called the
LINEX loss function, a and b are constants with b > 0 so that the
loss function is nonnegative . The shape of the LINEX loss
function (6) is determined by the constant a , and the value of b
will be taken equal one (i.e. b = 1 ) .
In figure 1, values of exp( a∆ ) − a∆ − 1 are plotted against ∆ for
selected values of a . It is seen that for a > 0 , the curve rises
almost exponentially when ∆ > 0 and almost linearly when ∆ < 0
. on the other hand, if a < 0 the function rises almost
exponentially when ∆ < 0 and almost linearly when ∆ > 0 . So the
sign of a reflects the direction of asymmetry, a > 0(a < 0) if
overestimation is more (less) serious than underestimation; and
its magnitude reflects the degree of asymmetry. The important
view is that for small values of a the function is almost
symmetric and not far from a squared error loss (SEL).
The expanding exp( a∆ ) ≈ 1 + a∆ + a 2 ∆2 / 2, L(∆) ≈ a 2 ∆2 / 2 , a SEL
function.
 Iraqi Journal of Statistical Science (16) 2010 _____________ [17]

3- Bayes' Estimator for θ

In this section we derived the estimation of unknown
parameter θ of the Gamma distribution based on random sample
of size n .The likelihood function is given by :
n n

n ∏Χ i − ∑ Χi
L( Χ θ) = ∏ f ( Χ i θ) = i =1
exp( i =1
) (7)
i =1 θ 2n θ
The natural logarithm of the likelihood function (7) :
l = ln L( Χ θ) = ∑ ln Χ i − 2n ln θ −
∑Χ i

⇒ θˆ ML =
∑Χ i
(8)
2n
Here we consider the non-informative prior f (θ) which is derived
from (7) as follows :
∂ 2 l 2 n 2∑ Χ i − ∂ 2l 2n
= − ⇒ F.I = −Ε[ ]= 2
∂θ 2 θ 2 θ3 ∂θ 2
θ
[18] ________________Bayes Estimator of one parameter...

Then the prior distribution of θ is :

f (θ) = F.I = 2n / θ 2 ∝ 1 / θ (9)
By combining the prior distribution f (θ) in (9) with the
likelihood function L( Χ θ) , using Bayes theorem we get the
posterior distribution :
π(θ Χ) ∝ L( Χ θ).f (θ)

⇒ π(θ Χ) ∝ θ −( 2 n +1) exp(−

∑Χ i
)
θ
Which is the kernel distribution of Inverse gamma distribution,
then
(∑ Χ i ) 2 n ∑Χ i
π( θ Χ ) = θ −( 2 n +1) exp(− ) θ>0 (10)
Γ 2n θ
And from (10), the expectation of the posterior distribution
above, we get :
Ε(θ Χ) = θˆ Bayes =
∑Χ i
(11)
2n − 1
we observe that, if n is large and approach to ∞ , then Bayes'
estimator will be same as the estimator of Maximum Likelihood,
i.e.
if n → ∞ , ⇒ θˆ Bayes = θˆ ML .

4- Bayes' Estimator of θ under squared error loss

function
Under squared error function L(θˆ , θ) = (θˆ − θ) 2 , and by using
(10),we have
L(θˆ , θ) = (θˆ − θ) 2
Then we minimize
∫ (θˆ − θ) π(θ Χ )dθ
2

The Bayes' estimator of θ denoted by θ̂ SB (SB=Squared Bayes) is

the posterior mean
θˆ SB = Ε π (θ) =
∑Χ i
(12)
2n − 1
 Iraqi Journal of Statistical Science (16) 2010 _____________ [19]

5- Bayes' Estimator of θ under LINEX loss function

θˆ
Suppose that ∆ = − 1 , where θ̂ is an estimate of θ .
θ
consider the loss function :
L ( ∆ ) = exp( a∆ ) − a∆ − 1 a≠0
Under this LINEX loss function ,the posterior mean of the loss
function L(∆) with respect to π(θ Χ) in (10) is :
θˆ θˆ
∞
Ε[L(∆)] = ∫ {exp[a ( − 1)] − a ( − 1) − 1}π(θ Χ)dθ (13)
0
θ θ
By using the integration by part we get :

θˆ θˆ
∞ ∞ ∞
Ε[L(∆)] = ∫ {exp[a ( − 1)]}π(θ Χ)dθ − ∫ a ( − 1)π(θ Χ)dθ − ∫ π(θ Χ)dθ
0
θ 0
θ 0

θˆ θˆ
= exp(−a )Ε[exp{a ( )}] − aE[( ) − 1] − 1 (14)
θ θ
The value of θ̂ that minimizes the posterior expectation of the
loss function L(∆) denoted by θ̂ LB (LB=LINEX Bayes) is
obtained by solving the equation :
∂Ε[L(∆)] a θˆ 1
= Ε[e −a exp(a ( )] − aE( ) = 0
∂θˆ θ θ θ
1 θˆ 1
⇒ Ε[ exp{a ( )}] = e a E ( ) (15)
θ θ θ
Provided that all expectations exist and finite, then we will use
(10) and (15) to find the expectations and get the optimal
estimate for θ ,
θˆ θˆ
∞
1 1
Ε[ exp{a ( )}] = ∫ exp{a ( )}π(θ Χ)dθ
θ θ 0
θ θ
(∑ Χ i ) 2 n ∞
θˆ − ∑ Χi
∫0
−( 2 n + 2 )
= θ exp{a ( )} exp( )dθ
Γ 2n θ θ
(∑ Χ i ) 2 n ∞
1
= ∫θ
−( 2 n + 2 )
exp{− (∑ Χ i − aθˆ )dθ
Γ 2n 0
θ
1 θˆ (∑ Χ i ) 2 n Γ 2n + 1 2n ( ∑ Χ i ) 2 n
⇒ Ε[ exp{a ( )}] = ∗ =
θ θ Γ 2n (∑ Χ i − aθˆ ) 2 n +1 (∑ Χ i − aθˆ ) 2 n +1
[20] ________________Bayes Estimator of one parameter...

And also
1
∞
1 (∑ Χ i ) 2 n ∞ − ( 2 n + 2 ) − ∑ Χi
E( ) = ∫ π(θ Χ)dθ =
Γ 2n ∫0
θ exp( )dθ
θ 0
θ θ
(∑ Χ i ) 2 n Γ 2n + 1
= ∗
Γ 2n (∑ Χ i ) 2 n +1
2n
=
∑ Χi
Then from (15), we have :
2n ( ∑ Χ i ) 2 n 2n a (∑ Χ i ) 2 n +1
= e ⇒ = ea
(∑ Χ i − aθˆ ) 2 n +1 ∑ Χ i (∑ Χ i − aθ)
ˆ 2 n +1

aθˆ 2 n +1 aθˆ
a
−
∴ (1 − ) = e −a ⇒ (1 − )=e 2 n +1

∑ Χi ∑ Χi
⇒ θˆ LB =
∑Χ
(1 − e
i
)
−
a
2 n +1
(16)
a
For more information (see Canfield (1970),Varin (1975) and
Zellner (1986)) .

6- The Decision Theory, Risk function & Risk Efficiency

The decision theoretic approach begins with a careful
definition of all the elements of a decision problem. It is
imagined that there is a decision-maker who is to choose an
action from a set Α . He is to do this based upon observation of a
random variable, or data Χ . This Χ (typically a vector
Χ 1 ,......., Χ n ) has a probability distribution which depends on an
unknown parameter θ . Here θ denotes a state of nature. The set
of all possible values of θ is the parameter space Θ .
The decision is to be made by a statistical decision function (or
rule) d ; this is a function which specifies d( x ) as the action to be
taken when the observed data is Χ = x . On taking action θˆ = d( Χ)
the decision-maker incurs a loss of L(θ, θˆ ) . A good decision
function is one that has a small value of the risk function :
 Iraqi Journal of Statistical Science (16) 2010 _____________ [21]

⎡ L(θˆ , θ)f (X; θ)dX

⎢R∫Χ
⎢
R (θ, θˆ ) = Ε[L(θ, θˆ )] = ⎢or (17)
⎢
⎢ ∑ L(θ, θ)p(X; θ)
ˆ
⎣ x∈R Χ
Clearly if R (θ, θˆ 1 ) ≤ R (θ, θˆ 2 ) for all θ and R (θ, θˆ 1 ) < R (θ, θˆ 2 ) for
some θ then the risk efficiency for θ̂1 is better than the risk
efficiency for θ̂ 2 . We say that θ̂ 2 is inadmissible. For more
information see Weber (2007) .

7- The Risk Efficiency of θ̂ LB with respect to θ̂ SB under

squared error loss function
The risk functions of the estimators under squared error
loss are denoted by R (θˆ LB , θ) and R (θˆ SB , θ) ,are given by :
∞
R S (θˆ LB , θ) = ∫ (θˆ LB − θ) 2 f ( x 1 , x 2 ,....., x n θ)dx 1dx 2 ......dx n (18)
0

Let S = ∑ Χ i , because x i , i = 1,2..., n are identically distributed and

independent from gamma distribution with parameters (2, θ) then
S = ∑ Χ i ~ Gam(2n , θ) , so that :
∞
S 2 n −1 S
R S (θˆ LB , θ) = ∫ (θˆ 2LB − 2θˆ LB θ + θ 2 ) ∗ 2 n exp(− )dS
0 θ Γ 2n θ
∞ ∞ ∞
S 2 n −1 S ˆ θ∗ S
2 n −1
S S 2 n −1 S
= ∫ θˆ 2LB ∗ exp( − ) dS − 2 ∫ θ LB exp( − ) dS + ∫ θ 2
∗ exp(− )dS
0 θ Γ 2n
2n
θ 0 θ Γ 2n
2n
θ 0 θ Γ 2n
2n
θ
Thus
a a
− −
θ [2n (2n + 1)(1 − e
2 2 n +1 2
) ] θ [4n (1 − e
2 2 n +1
)]
R S (θˆ LB , θ) = − + θ2
a2 a
a a
− −
2n (2n + 1)(1 − e ) [4n (1 − e
2 n +1 2
)] 2 n +1
⇒ R S (θˆ LB , θ) = θ 2 [ 2
− + 1] (19)
a a
By the same way we can find R S (θˆ SB , θ) under squared error loss :
∞
R S (θˆ SB , θ) = ∫ (θˆ SB − θ) 2 f ( x 1 , x 2 ,....., x n θ)dx 1dx 2 ......dx n (20)
0

Let S = ∑ Χ i then :
[22] ________________Bayes Estimator of one parameter...

∞
S 2 n −1 S
R S (θˆ SB , θ) = ∫ (θˆ SB
2
− 2θˆ SB θ + θ 2 ) ∗ 2 n exp( − )dS
0 θ Γ 2n θ
∞ ∞ ∞
ˆ S2n−1 S ˆ S2n−1 S S S
= ∫ θSB ∗ 2n
2
exp(− )dS − 2∫ θSBθ ∗ 2n exp(− )dS + ∫ θ2 ∗ 2n exp(− )dS
0 θ Γ2n θ 0 θ Γ2 θ 0 θ θ
Thus
2n (2n + 1)θ 2 4nθ 2
R S (θˆ SB , θ) = − + θ2
(2n − 1) 2
(2n − 1)
2n (2n + 1) 4n
⇒ R S (θˆ SB , θ) = θ 2 [ − + 1] (21)
(2n − 1) 2
(2n − 1)
The risk efficiency of θ̂ LB with respect to θ̂ SB under squared
error loss function is denoted by :
a a
− −
2n (2n + 1)(1 − e 2 n +1 ) 2 [4n (1 − e 2 n +1 )]
ˆ [ − + 1]
R (θ , θ) a2 a (22)
RE S (θˆ LB , θˆ SB ) = S LB =
R S (θˆ SB , θ) 2n (2n + 1) 4n
[ − + 1]
(2n − 1) 2 (2n − 1)

8- The Risk Efficiency of θ̂ LB with respect to θ̂ SB under

LINEX loss function
The risk functions of the estimators θ̂ LB and θ̂ SB under
LINEX loss are denoted by R L (θˆ LB , θ) and R L (θˆ SB , θ) , where the
subscript L denotes risk relative to LINEX loss and are given as
follows :
θˆ LB
∞
a [( ) −1] θˆ LB (23)
R L (θˆ LB , θ) = ∫ {e θ
− a[( ) − 1] − 1}f ( x 1 , x 2 ,....., x n θ)dx 1dx 2 ......dx n
0
θ
Let S = ∑ Χ i , and as a mentioned above S = ∑ Χ i ~ Gam(2n, θ) ,
so that :
θˆ LB
∞
a [( ) −1] θˆ LB S 2 n −1 S
R L (θˆ LB , θ) = ∫ {e θ
− a[( ) − 1] − 1} ∗ 2 n exp(− )dS
0
θ θ Γ 2n θ
θˆ LB
∞
a [( ) −1] S 2 n −1
S θˆ ∞
S 2 n −1 S
∞
S 2 n −1 S
∞
S 2 n −1 S
= ∫e θ
∗ exp( − )dS − ∫ a ( LB ) ∗ 2 n exp( − )dS + ∫ a ∗ 2 n exp( − )dS − ∫ 2 n exp( − )dS
0 θ 2 n Γ 2n θ 0
θ θ Γ 2 n θ 0 θ Γ 2 n θ 0 θ Γ 2n θ

Thus
a a
− −
R L (θˆ LB , θ) = e 2 n +1 − 2n (1 − e 2 n +1 ) + a − 1
(24)
By the same way we can find R L (θˆ SB , θ) under squared error loss
 Iraqi Journal of Statistical Science (16) 2010 _____________ [23]

θˆ SB
∞
a [( ) −1] θˆ SB (25)
R L (θˆ sB , θ) = ∫ {e θ
− a[( ) − 1] − 1}f ( x 1 , x 2 ,....., x n θ)dx 1dx 2 ......dx n
0
θ
Let S = ∑ Χ i then :
θˆ SB
∞
a [( ) −1] θˆ SB S 2 n −1 S
R L (θˆ SB , θ) = ∫ {e θ
− a[( ) − 1] − 1} ∗ 2 n exp( − )dS
0
θ θ Γ 2n θ
θˆ
∞
a [( SB ) −1] S 2 n −1
S θˆ SB ∞
S 2 n −1 S S 2 n −1
∞
S S 2 n −1
∞
S
= ∫e θ
∗ exp( − ) dS − ∫ a ( ) ∗ exp( − ) dS + ∫ a ∗ exp( − ) dS − ∫ exp( − )dS
0 θ Γ 2n
2n
θ 0
θ θ Γ 2n
2n
θ 0 θ Γ 2n
2n
θ 0 θ Γ 2n
2n
θ
Thus
a −2n a
R L (θˆ SB , θ) = e −a (1 − ) − −1 (26)
2n − 1 2n − 1
The risk efficiency of θ̂ LB with respect to θ̂ SB under
LINEX loss function is denoted by :
a a
− −
ˆ ˆ R L (θˆ LB , θ) e 2 n +1 − 2n (1 − e 2 n +1 ) + a − 1
RE L (θ LB , θ SB ) = = (27)
R L (θˆ SB , θ) e −a (1 − a ) − 2 n − a − 1
2n − 1 2n − 1

9- Numerical Example
We generated Ν = 500 samples of sizes n = 10,20,30 from
equation (1) with θ = 1 , we used Minitab to generate these
samples and we take randomly the samples of size n = 10,20,30 ,
respectively and then the risk functions are computed for the
estimators θ̂ LB and θ̂ SB under the LINEX loss function and
squared error loss , and also computed the risk efficiency to
compare between the LINEX loss and squared error loss function
to check which estimator is inadmissible under theses functions .
The results are explained in tables from 1 to 6 .
[24] ________________Bayes Estimator of one parameter...

Table 1 : the estimators θ̂ LB , θ̂ SB , the risk efficiencies

RE L (θˆ LB , θˆ SB ) , RE S (θˆ LB , θˆ SB ) under the prior f (θ) for the value of
a=2

n θ̂SB θ̂ LB R L (θˆ LB , θ) R L (θˆ SB , θ)

10 1.0511 0.9071 0.0911 0.1465
20 1.0361 0.9619 0.0443 0.0600
30 1.02 0.9760 0.0324 0.0375
n R S (θˆ LB , θ) R S (θˆ SB , θ) RE L (θˆ LB , θˆ SB ) RE S (θˆ LB , θˆ SB )
10 0.0495 0.0581 1.6081 1.1737
20 0.025 0.027 1.3544 1.08
30 0.0166 0.0175 1.1574 1.0542
Table 2 : the estimators θ̂ LB , θ̂ SB , the risk efficiencies
RE L (θˆ LB , θˆ SB ) , RE S (θˆ LB , θˆ SB ) under the prior f (θ) for the value of
a =1

n θ̂SB θ̂ LB R L (θˆ LB , θ) R L (θˆ SB , θ)

10 1.2042 0.5320 0.0234 0.0318
20 1.1819 0.5553 0.0121 0.0139
30 1.1766 0.5644 0.0082 0.0088
n R S (θˆ LB , θ) R S (θˆ SB , θ) RE L (θˆ LB , θˆ SB ) RE S (θˆ LB , θˆ SB )
10 0.0481 0.0581 1.3589 1.2079
20 0.0245 0.027 1.1487 1.1020
30 0.0165 0.0175 1.0731 1.0606
 Iraqi Journal of Statistical Science (16) 2010 _____________ [25]

Table 3 : the estimators θ̂ LB , θ̂ SB , the risk efficiencies

RE L (θˆ LB , θˆ SB ) , RE S (θˆ LB , θˆ SB ) under the prior f (θ) for the value of
a = 0 .5

n θ̂SB θ̂ LB R L (θˆ LB , θ) R L (θˆ SB , θ)

10 1.1072 0.2474 0.0059 0.0075
20 1.0859 0.2566 0.00305 0.00344
30 1.0738 0.2585 0.0020 0.0022
n R S (θˆ LB , θ) R S (θˆ SB , θ) RE L (θˆ LB , θˆ SB ) RE S (θˆ LB , θˆ SB )
10 0.0478 0.0581 1.2711 1.2154
20 0.0244 0.027 1.1278 1.1065
30 0.0164 0.0175 1.1000 1.0670

Table 4 : the estimators θ̂ LB , θ̂ SB , the risk efficiencies

RE L (θˆ LB , θˆ SB ) , RE S (θˆ LB , θˆ SB ) under the prior f (θ) for the value of
a = −2

n θ̂SB θ̂ LB R L (θˆ LB , θ) R L (θˆ SB , θ)

10 1.0409 0.9881 0.0984 0.1035
20 1.0376 1.0115 0.0495 0.0502
30 1.0292 1.0119 0.0332 0.0336
n R S (θˆ LB , θ) R S (θˆ SB , θ) RE L (θˆ LB , θˆ SB ) RE S (θˆ LB , θˆ SB )
10 0.0499 0.0581 1.0518 1.1643
20 0.0240 0.027 1.0141 1.1250
30 0.0166 0.0175 1.0120 1.0542
[26] ________________Bayes Estimator of one parameter...

Table 5 : the estimators θ̂ LB , θ̂ SB , the risk efficiencies

RE L (θˆ LB , θˆ SB ) , RE S (θˆ LB , θˆ SB ) under the prior f (θ) for the value of
a = −1

n θ̂SB θ̂ LB R L (θˆ LB , θ) R L (θˆ SB , θ)

10 1.2081 1.1194 0.0242 0.0269
20 1.1796 1.1359 0.0123 0.0130
30 1.1121 1.0845 0.0083 0.0085
n R S (θˆ LB , θ) R S (θˆ SB , θ) RE L (θˆ LB , θˆ SB ) RE S (θˆ LB , θˆ SB )
10 0.0482 0.0581 1.1115 1.2053
20 0.0245 0.027 1.0569 1.1020
30 0.0164 0.0175 1.0240 1.0670

Table 6 : the estimators θ̂ LB , θ̂ SB , the risk efficiencies

RE L (θˆ LB , θˆ SB ) , RE S (θˆ LB , θˆ SB ) under the prior f (θ) for the value of
a = − 0 .5

n θ̂SB θ̂ LB R L (θˆ LB , θ) R L (θˆ SB , θ)

10 1.1679 1.0693 0.0060 0.0070
20 1.1347 1.0859 0.0030 0.0032
30 1.1241 1.0917 0.00205 0.00216
n R S (θˆ LB , θ) R S (θˆ SB , θ) RE L (θˆ LB , θˆ SB ) RE S (θˆ LB , θˆ SB )
10 0.0477 0.0581 1.1666 1.2180
20 0.0244 0.027 1.0666 1.1065
30 0.0165 0.0175 1.0536 1.0606

10- Conclusion
1- From tables (1-6), we observe that the risk efficiency
RE L (θˆ LB , θˆ SB ) is greater than 1, which means that the proposed
estimators θ̂ LB is preferable to θ̂ SB for the sample sizes n = 10,20,30
from gamma distribution with parameters (α = 2, θ = 1) and for all
values of " a " (a = ±0.5,±1,±2) .
 Iraqi Journal of Statistical Science (16) 2010 _____________ [27]

2- A symmetric loss function is more appropriate than Squared

error loss function .
3- We note that the risk efficiency RE L (θˆ LB , θˆ SB ) is greater than
the risk efficiency RE S (θˆ LB , θˆ SB ) for all positive values of a , and
the risk efficiency RE S (θˆ LB , θˆ SB ) is greater than the risk efficiency
RE L (θˆ LB , θˆ SB ) for all negative values of a and for all sample sizes
n = 10,20,30 .

11- References
1- Basu, A. P., and Ebrahimi, N. (1991), Bayesian approach to
life testing and reliability estimation using asymmetric loss
function. J. Statist. Plann. Infer., 29, 21–31.
2- Calabria, R., and Pulcini, G. (1996), Point estimation under
asymmetric loss functions for left truncated exponential
samples. Commun. Statist. Theory. Meth., 25, 585–600.
3- Canfield, R. V. (1970), A Bayesian approach to reliability
estimation using a loss function. IEEE Trans. Reliab., R-19,
13–16.
4- Chhikara, R., and Folks, J. (1977), The Inverse Gaussian
Distribution As A Life Model. Technometrics 19, 461-468 .
5- Dey, S., (2007), Inverted Exponential Distribution as a Life
distribution Model from a Bayesian viewpoint. Data Science
Journal. Volume 6, 107-113 .
6- Nassar, M. M., and Eissa, F. H. (2004), Bayesian estimation
for the exponentiated Weibull model, Commun. Statist.-
Theory Meth., 33(10), 2343-2362 .
7- Pandey, B. N. (1997), Testimator of the scale parameter of the
exponential distribution using LINEX loss function.
Commun. Statist. Theor. Meth., 26(9), 2191–2202.
8- Parsian, A., and Kirmani, S. N. U. A. (2002), Estimation under
LINEX loss function. Unpublished manuscript.
9- Sherif, Y. S., and Smith, M. L., (1980), First- Passage Time
Distribution of Brownian motion as a reliability model.
IEEE Transaction of Reliability, R-29, No.5,425-426 .
10- Sinha, S. and Kale, B. (1980) Life Testing and Reliability
Estimation. John Wiley & Sons .
[28] ________________Bayes Estimator of one parameter...

11- Soliman, A. A. (2000), Comparison of LINEX and quadratic

Bayes estimators for the Rayleigh distribution. Commun.
Statist. Theor. Meth., 29(1), 95–107.
12- Varian, H. R. (1975), A Bayesian approach to real estate
assessment. In Studies in Bayesian Econometrics and
Statistics in Honor of Leonard J. Savage, eds. Stephan E.
Fienberg and Arnold Zellner, Amsterdam: North-Holland,
pp. 195–208.
13- Von Alven, W. (ed) (1964) Reliability Engineering by
ARINC. Prentice Hall, Inc., New Jersey .
14- Weber, R., (2007),http://www.statslab.cam.ac.uk/~rrw1/stats/
, Statistical Laboratory, University of Cambridge .
15- Zellner, A. (1986), Bayesian estimation and prediction using
asymmetric loss functions. J. Am. Statist. Assoc., 81, 446–
451.