You are on page 1of 6

Statistics Research Letters (SRL) Volume 3, 2014

www.srl-journal.org

Right Censored Bayes Estimator for Lomax


Model
Gyan Prakash
Department of Community Medicine
S. N. Medical College, Agra, U. P., India
ggyanji@yahoo.com
Abstract
Properties of the Bayes estimators are studied for Lomax
model in the present article. Both known and unknown case
of scale parameter are consider here for obtaining the Bayes
estimators of shape parameter. A Right item failure censored
data from the proposed model is considered and the
performances of the procedures are illustrated by simulation
technique.

forecasting the temperatures records. Recently,


Prakash & Singh (2013) presented some Bayes
prediction length of intervals for the Pareto model.

Bayes Estimator; Invariant Squared Error Loss Function; LINEX


Loss Function; Right Item Failure Censoring

The objective of the present paper is to perform a


comparative study between two different asymmetric
loss functions. The Bayes estimators for shape
parameter are obtained here when scale parameter is
considered to be known as well as unknown. A Right
Item-failure censoring criterion is used for the
estimation and the illustration of the procedures, a
simulation study are carried out.

Introduction

The Model and Posterior Densities

The Pareto distribution and their close relatives


provide a very flexible family of fattailed
distributions, which may be used as a model for the
income distribution of higher income group. The
Pareto model plays an important role in socio
economic studies. It is often used as a model to
analyse areas including city population distribution,
stock price fluctuation, oil field locations and military
areas. The Pareto distribution has a decreasing failure
rate, so it has been used for model survival after some
medical procedures (the ability to survive for a longer
time appears enhanced, one survives longer after
certain medical procedures).

The probability density function of the considered


Lomax model is given as

Keywords

Davice & Feldstien (1979) have viewed the Pareto


distribution as a potential model for life testing
problems. This distribution has established its
important role in variety of other problems such as
size of cities and firms (Steindle, 1965), business
mortality (Lomax, 1954), service time in queuing
system (Harries, 1967). Freiling (1966) applied the
Pareto law to study the distributions of nuclear
particles. Harries (1968) used this distribution in
determining times of maintenance service while Dyer
(1981) found that the two-parameter Pareto
distribution transformation is equivalent to the twoparameter exponential distribution. Madi & Raqab
(2004) considered the Pareto distribution as the

f ( x ; , =
) ( x + ) ( + 1) ; x 0, > 0, > 0. (2.1)
Here, is known as the shape parameter and be the
scale parameter. The given model (2.1) is newly
proposed Lomax distribution which is the result of
mixture of Exponential distribution with the
parameter , and the parameter is distributed as
the Gamma distribution with parameters and .
In life testing, the observations usually occurred in
ordered manner such a way that weakest items fail
first and then the second one and so on. Let us
suppose that, the n items are put to test under the
model (2.1) without replacement. In which only
ordered m( n) items are fully measure, while the
remaining (n m) items are censor. These (n m)
censored lifetimes will be order separately. This
process is called as the Right Item failure-censoring
scheme.
Now, let us consider a sequence of independent
random sample of size n such as x (1) ,

x (2) ,...,x (m 1) ,x (m) ,x (m + 1) ,..., x (n) from the model (2.1).


All n items are put to test without replacement and
first m (x (1) ,x (2) ,...,x (m 1) , x (m) ) items are fully

23

www.srl-journal.org

Statistics Research Letters (SRL) Volume 3, 2014

measured while remaining

(n m)

(x (m + 1) ,x (m + 2) ,...,x (n) )

items are censored. Based on the above the likelihood


function is thus obtained as

L ( ,|x ) f (x i ;,)
=i 1

L ( ,|x ) m n exp (

(1 F (x i ;,) )
=i m +1

(2.2)
T1 T0 ) ,

where

=
T1

) and T 0 log ( x (i) + ) .


log ( x (i) +=
n

i=1

i=1

When scale parameter is known, the conjugate prior


density for parameter is taken as a Gamma
distribution with probability density function

e
; > 0, > 0, > 0.
()

=
g()

(2.3)

Now posterior density for parameter is obtain as

( |x ) =

L (|x) g ( )
L (|x) g ( ) d

Bayesian analysis. It has always been recognized that


the most commonly used loss function, squared error
loss function (SELF) is in appropriate in many
situations. The Bayes estimator of a parameter under
SELF is the posterior mean. If SELF is taken as a
measure of inaccuracy then the resulting risk is often
too sensitive to the assumptions about the behaviour
of the tail of the probability distribution. To overcome
this difficulty, a useful asymmetric loss function based
on the squared error loss function is defined for any
estimate corresponding to the parameter as

( )

(3.1)
.
2
The loss function defined in (3.1) is known as the
invariant squared error loss function (ISELF). The
Bayes estimator for parameter corresponding to
ISELF under the posterior ( |x) is obtained as

L ,

( ) E ( )

m+

=
I

where T m = T 1 + n log .

1
(2.5)
g1 ( ,)
= g ( ) ; 0 , > 0 .

Here, we are not going to debate or justify the
questions of the proper choice of the prior distribution.
For situations where life tester has no prior
information about the parameter, they may use the
quasi-density prior. Therefore, we consider here a
diffuse prior.

The joint posterior density for the parameters and


is thus obtained as

L ( ,|x) g1 ( ,)
L ( ,|x) g1 ( ,) d d

m + 1 e T m e T 0
1 ( , |x ) =
;
(m + )

(m + )

(2.6)

d .

=0

Bayes Estimator (Scale Parameter Known)


Case 1: Invariant Squared Error Loss Function (ISELF)
The choice of the loss function may be crucial in

24

1
( |x) d

=
( |x) d
2

m+2
.
Tm

(3.2)

When scale parameter is considered as the random


variable, the joint prior density function for both
parameters and is taken as

where = e T 0 ( T m )

I = E 1

( Tm )
( |x) =
m + 1 e T m ; (2.4)
(m + )

1 ( ,|x ) =

( )
=

The risk corresponding to the Bayes estimator I


under ISELF is

( )

((T ) )
((T ) ) + 1.
2

R I I = (m + 2) 2 E (Tm )
2 (m + 2) E (Tm )

( )

The close form of the risk R I I

(3.3)

does not exist.

However, a simulation study has been carried out in


Section 4 for studying the properties of the Bayes
estimator.
Case 2: LINEX Loss Function (LLF)
When positive and negative errors have different
consequences, the use of squared error loss function
(SELF) in Bayesian estimation may not be appropriate.
In addition, in some estimation problems
overestimation
is
more
serious
than
the
underestimation, or vice-versa. To deal with such
cases, a useful and flexible class of asymmetric loss
function (LINEX loss function (LLF)) is given as

L ( ) = e a a 1 ; = .
The shape parameter of the LLF is denoted by 'a' and
be any estimate of the parameter . The negative

(positive) value of 'a', gives more weight to


overestimation (underestimation) and its magnitude
reflect the degree of asymmetry. It is also seen that, for

Statistics Research Letters (SRL) Volume 3, 2014

a = 1 , the function is quite asymmetric with


overestimation being costly than underestimation. For
small values of |a|, the LLF is almost symmetric and

www.srl-journal.org

0.50,1.00,2.00 and presented in Table 2.


TABLE 1: RISK OF THE BAYES ESTIMATOR
KNOWN

not far from the SELF.

n = 15

The Bayes estimator corresponding to the LLF is


obtained as

( )

(m + ) E (Tm )

{ ln (

1 + aTm 1

Again, close form of the risk R L

To assess and study the properties of the Bayes


estimator for the shape parameter (when scale
parameteris known) a simulation study has been
performed. The random samples are generated as
follows:
1. Generate through prior density g ( ) for the
selected set of prior parameters and as ( , ) =

(0.25, 0.50), (4, 2), and (9,3) . The selections of prior


parametric values meet the criterion that the prior
variance should be unity.
2. Using obtained from (1) with considered values of
= 0.50(0.50)2.50; to generate 10,000 random samples
of size n = 15 from the model (2.1).

5. Similar trend has also been seen when ( )


decreases for other fixed parametric values.
6. The Bayes risk for L under LLF is obtained for a =

0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
2.00 08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03
0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
1.00 08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03
0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
0.50 08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03

Numerical Analysis (Scale Parameter


Known)

values are fixed.

3.4515
2.6929
2.1960
2.7500
1.7506
1.7279
2.6269
1.7339
1.4535

3.2934
2.5696
2.0954
2.6240
1.6704
1.6488
2.5066
1.6545
1.3869

3.0767
2.4005
1.9575
2.4513
1.5605
1.5403
2.3417
1.5456
1.2956

2.8165
2.1975
1.7919
2.2440
1.4285
1.4100
2.1436
1.4149
1.1860

L UNDER LLF WHEN IS

n = 15
a m

simulation study has also been carried out in next


Section for studying the properties of the Bayes
estimator L .

tend be closer as m ( ) increases for other parametric

3.5385
2.7608
2.2513
2.8193
1.7947
1.7714
2.6931
1.7776
1.4901

KNOWN

4. The values of the Bayes risk for the I under ISELF

= 2.50 = 2.00 = 1.50 = 1.00 = 0.50

TABLE 2: RISK OF THE BAYES ESTIMATOR

(3.5)
)} + a 1.
( ) does not exist. A

3. The Bayes risk under ISELF have been obtained for


the selected set of censored sample size m = 05,08,12;
and presented in Table 1.

0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03

1
L = ln e a ( |x) d
a

m
+

(3.4)
=
L
.
ln 1 +
T
a
m

Further, the risk corresponding to the Bayes estimator


L under LLF IS
m+

R L L= e a E (Tm ) 1 +

T
m

I UNDER ISELF WHEN IS

= 2.50 = 2.00 = 1.50 = 1.00 = 0.50


2.7648
2.1573
1.7591
2.2030
1.4024
1.3841
2.0929
1.3323
1.3149
2.5433
1.9845
1.6182
2.0266
1.2900
1.2732
1.9253
1.2255
1.2096
2.3395
1.8255
1.4885
1.8642
1.1866
1.1713
1.7710
1.1273
1.1127

2.6968
2.1043
1.7159
2.1488
1.3679
1.3501
2.0415
1.2996
1.2826
2.4808
1.9357
1.5784
1.9768
1.2583
1.2419
1.8780
1.1954
1.1799
2.2820
1.7806
1.4519
1.8184
1.1574
1.1425
1.7275
1.0996
1.0853

2.5733
2.0079
1.6373
2.0504
1.3052
1.2883
1.9480
1.2401
1.2239
2.3672
1.8470
1.5061
1.8863
1.2007
1.1850
1.7920
1.1406
1.1259
2.1775
1.6990
1.3854
1.7351
1.1044
1.0902
1.6484
1.0492
1.0356

2.4040
1.8758
1.5296
1.9155
1.2193
1.2035
1.8198
1.1585
1.1434
2.2114
1.7255
1.4070
1.7622
1.1217
1.1070
1.6741
1.0655
1.0518
2.0342
1.5872
1.2942
1.6209
1.0317
1.0185
1.5399
0.9802
0.9675

2.2007
1.7171
1.4002
1.7535
1.1162
1.1017
1.6659
1.0605
1.0467
2.0244
1.5796
1.2880
1.6132
1.0268
1.0134
1.5325
0.9754
0.9628
1.8621
1.4530
1.1847
1.4838
0.9444
0.9324
1.4097
0.8973
0.8857

7. The behaviours of Bayes estimator L are similar to


those of the estimator . Further, the increasing trend
I

in the magnitude is also seen when 'a' increases but


the increment in magnitude is nominal.
8. The risk magnitude of the Bayes estimator L
(under LLF) is smaller than that of the Bayes estimator
(under ISELF). Hence, the Bayes risk of the
I

estimator I is obtained again under the loss LLF and


presented in the Table 3. It is also noted that the risk

25

www.srl-journal.org

Statistics Research Letters (SRL) Volume 3, 2014

magnitude corresponding to the Bayes estimator L is


smaller than that of the Bayes estimator . Other
I

I1 =

( m + 2)

; =

properties are similar.

( )

2.9932
2.3354
1.9044
2.3850
1.5182
1.4984
2.2658
1.4424
1.4235
2.7533
2.1484
1.7519
2.1940
1.3965
1.3784
2.0843
1.3268
1.3095
2.5328
1.9763
1.6115
2.0181
1.2846
1.2680
1.9173
1.2204
1.2046

2.8561
2.2284
1.8172
2.2758
1.4487
1.4298
2.1620
1.3763
1.3583
2.6272
2.0500
1.6717
2.0935
1.3325
1.3153
1.9888
1.2660
1.2495
2.4168
1.8858
1.5377
1.9257
1.2258
1.2099
1.8295
1.1645
1.1494

2.6682
2.0818
1.6976
2.1260
1.3534
1.3357
2.0197
1.2857
1.2689
2.4543
1.9151
1.5617
1.9557
1.2448
1.2288
1.8579
1.1827
1.1673
2.2578
1.7617
1.4365
1.7990
1.1451
1.1303
1.7091
1.0879
1.0738

2.4425
1.9057
1.5540
1.9462
1.2389
1.2227
1.8489
1.1770
1.1616
2.2467
1.7531
1.4296
1.7903
1.1395
1.1249
1.7008
1.0827
1.0686
2.0668
1.6127
1.3150
1.6468
1.0482
1.0347
1.5645
0.9959
0.9830

( )

2 ( |,x) d

26

m+1

T
T
e m e 0 d d
(m + )

m+1
T
T
e m e 0 d d
(m + )

) +1

(5.2)

(5.3)

Similarly, the Bayes estimator for the shape parameter


, corresponding to the LLF is given as

1
m+1
L1 = ln e a
e Tm e T 0 d d

(m + )
a

1
m T 0
L1 =
ln ( Tm + a )
e d
a

and the risk with respect to LLF is

(5.4)

1
1

m T 0
a

R L L
e
E (Tm ) ( Tm + a )
e
d
=

m T 0
e d + a 1.
+ E (Tm ) ln ( Tm + a )
(5.5)

m+1
2 ( |,x) = e Tm e T 0 d .
(m + )
The Bayes estimator of the shape parameter
corresponding to the posterior density 2 ( |,x) when

2 ( |,x) d

Case 2: LINEX Loss Function (LLF)

2 ( |,x) = 1 ( ,|x) d

a (m + 2) E (Tm ) ( ) + a 1.

and is given in (2.6). The Marginal posterior density


for the parameters is obtained as

R L L
e a E (Tm ) ea (m + 2)
=

The joint posterior density 1 ( ,|x) for parameters

scale parameter is considered random variable is

(5.1)

.
d

and

Case 1: Invariant Squared Error Loss Function (ISELF)

( Tm ) m + 2

2 (m + 2) E (Tm ) 1 ( )

Bayes Estimator (Unknown Scale Parameter)

I1 =

T0

2
R I II = (m + 2) 2 E (Tm ) 2 ( )

= 2.50 = 2.00 = 1.50 = 1.00 = 0.50


3.0686
2.3943
1.9524
2.4451
1.5565
1.5362
2.3229
1.4787
1.4594
2.8227
2.2025
1.7960
2.2493
1.4317
1.4131
2.1368
1.3602
1.3425
2.5966
2.0261
1.6521
2.0690
1.3170
1.3000
1.9656
1.2512
1.2350

I1

n = 15
0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
2.00 08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03
0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
1.00 08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03
0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
0.50 08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03

( Tm ) m +1 d

The risk expression corresponding to the Bayes


estimator under ISELF and LLF is

I UNDER LLF WHEN IS

KNOWN

T0

TABLE 3: RISK OF THE BAYES ESTIMATOR

a m

( )

The close forms of the risk under the ISELF and LLF
for both Bayes estimator again do not exist. To study
the properties of the Bayes estimators, a simulation
study has been performed in next section.
Remark
The magnitude of risk for the Bayes estimator I
(known scale parameter case) under the risk criteria
ISELF is higher than that under LLF. Hence, in the
case when scale parameter is considered to be random
variable, the risk for Bayes estimator is obtained
I1

here for both risk criterion.


Numerical Analysis (Unknown Scale
Parameter)
When both parameters are considered as the random
variable, a simulation study also has been carried out
for studying the properties of Bayes estimators as:
1. The values of have been generated for the similar
set of values of the prior parameters and consider

Statistics Research Letters (SRL) Volume 3, 2014

www.srl-journal.org

in Section 4.
2. Generate a random value of from the uniform
U ( 0, 1/ ) for the given values of prior parameter

( = 0.10, 0.50, 1.00, 2.00, 5.00).


TABLE 4: RISK OF THE BAYES ESTIMATOR

I UNDER ISELF WHEN IS

UNKNOWN

0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03

n = 15
= 5.00 = 2.00 = 1.00 = 0.50 = 0.10
3.7322
2.9120
2.3746
2.9737
1.8930
1.8684
2.8405
1.8749
1.5717

3.6405
2.8404
2.3162
2.9006
1.8465
1.8225
2.7707
1.8288
1.5331

3.4738
2.7103
2.2101
2.7677
1.7619
1.7390
2.6438
1.7450
1.4629

3.2452
2.5320
2.0647
2.5856
1.6460
1.6246
2.4698
1.6302
1.3666

2.9707
2.3178
1.8901
2.3669
1.5068
1.4872
2.2609
1.4923
1.2510

TABLE 5: RISK OF THE BAYES ESTIMATOR L UNDER LLF WHEN IS


KNOWN

n = 15
a

0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
2.00 08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03
0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
1.00 08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03
0.25, 0.50
05 04, 02
09, 03
0.25, 0.50
0.50 08 04, 02
09, 03
0.25, 0.50
12 04, 02
09, 03

= 5.00 = 2.00 = 1.00 = 0.50 = 0.10


3.0434
2.3747
1.9364
2.4250
1.5437
1.5236
2.3038
1.4665
1.4474
2.7996
2.1845
1.7813
2.2308
1.4200
1.4015
2.1193
1.3490
1.3315
2.5752
2.0094
1.6385
2.0520
1.3062
1.2893
1.9495
1.2409
1.2248

2.9686
2.3163
1.8888
2.3654
1.5058
1.4861
2.2472
1.4305
1.4118
2.7308
2.1308
1.7375
2.1760
1.3851
1.3671
2.0672
1.3158
1.2988
2.5119
1.9600
1.5982
2.0016
1.2741
1.2576
1.9016
1.2104
1.1947

2.8326
2.2102
1.8023
2.2571
1.4368
1.4180
2.1443
1.3650
1.3471
2.6057
2.0332
1.6579
2.0763
1.3217
1.3045
1.9725
1.2555
1.2393
2.3968
1.8702
1.5250
1.9099
1.2157
1.2000
1.8145
1.1550
1.1400

2.6462
2.0648
1.6837
2.1086
1.3423
1.3247
2.0032
1.2752
1.2585
2.4342
1.8994
1.5488
1.9397
1.2347
1.2187
1.8427
1.1729
1.1578
2.2391
1.7471
1.4247
1.7842
1.1357
1.1210
1.6951
1.0790
1.0650

2.4224
1.8902
1.5413
1.9303
1.2288
1.2127
1.8338
1.1673
1.1521
2.2283
1.7387
1.4178
1.7756
1.1303
1.1156
1.6868
1.0737
1.0599
2.0497
1.5993
1.3042
1.6333
1.0396
1.0262
1.5517
0.9877
0.9749

3. Using the above generated values of and


obtained in steps (1) & (2), we generate the 10,000
random samples of size n = 15 form the model (2.1).
4. For the similar set of selected values of m and 'a'
(Section 4), the risk corresponding to the Bayes
estimator is obtained and presented in the Tables 46
respectively.

5. It has been observed from Table 4-6 that all the


properties of the Bayes estimator are similar to those
of the corresponding Bayes estimators obtained when
scale parameter is known.
6. The magnitude of the risk is wider than that in the
case of known scale parameter. However, the
difference between the risk magnitude of
corresponding to Bayes estimator in both known and
unknown case of scale parameter is least.
7. Again it is noted that the magnitude of risk under
the LLF is robust corresponding to the risk criterion
ISELF.
TABLE 6: RISK OF THE BAYES ESTIMATOR

I UNDER LLF WHEN IS

KNOWN

n = 15
a m ,

02, 0.50
05 04, 0.25
08, 0.125
02, 0.50
2.00 08 04, 0.25
08, 0.125
02, 0.50
12 04, 0.25
08, 0.125
02, 0.50
05 04, 0.25
08, 0.125
02, 0.50
1.00 08 04, 0.25
08, 0.125
02, 0.50
12 04, 0.25
08, 0.125
02, 0.50
05 04, 0.25
08, 0.125
02, 0.50
0.50 08 04, 0.25
08, 0.125
02, 0.50
12 04, 0.25
08, 0.125

= 2.50 = 2.00 = 1.50 = 1.00 = 0.50


3.4058
2.6574
2.1669
2.7138
1.7275
1.7050
2.5781
1.6412
1.6198
3.1328
2.4445
1.9933
2.4964
1.5890
1.5684
2.3716
1.5097
1.4900
2.8819
2.2487
1.8336
2.2963
1.4617
1.4428
2.1816
1.3887
1.3707

3.3221
2.5921
2.1136
2.6471
1.6850
1.6631
2.5147
1.6009
1.5800
3.0558
2.3844
1.9443
2.4350
1.5499
1.5298
2.3133
1.4726
1.4534
2.8111
2.1934
1.7885
2.2399
1.4258
1.4073
2.1280
1.3546
1.3370

3.1699
2.4734
2.0168
2.5259
1.6078
1.5869
2.3995
1.5276
1.5076
2.9158
2.2752
1.8552
2.3235
1.4789
1.4597
2.2073
1.4052
1.3868
2.6823
2.0929
1.7066
2.1373
1.3605
1.3428
2.0305
1.2926
1.2758

2.9613
2.3106
1.8841
2.3597
1.5020
1.4825
2.2416
1.4271
1.4084
2.7239
2.1255
1.7331
2.1706
1.3816
1.3636
2.0621
1.3127
1.2955
2.5058
1.9552
1.5943
1.9967
1.2710
1.2544
1.8969
1.2075
1.1918

2.7108
2.1152
1.7247
2.1601
1.3750
1.3571
2.0520
1.3064
1.2893
2.4935
1.9457
1.5865
1.9870
1.2647
1.2483
1.8877
1.2017
1.1859
2.2939
1.7898
1.4595
1.8278
1.1635
1.1483
1.7365
1.1054
1.0910

Conclusions
For the study of Bayes estimator under the Right-Item
failure censoring criterion, Lomax model is considered
here as the underlying model. The proposed Lomax
distribution is the result of the mixture of distribution.
Under considered censoring scheme, the Bayes
estimator is obtained for the shape parameter under
two different loss functions for both known and
unknown case of scale parameter. A simulation has
been carried out for study the performances of the
procedures.
27

www.srl-journal.org

In the present article, it is observed that the Bayes risk


corresponding to the entire estimator is least under the
asymmetric loss function LLF; and that the difference
in risk magnitude between the corresponding
estimators under both the case is robust. The
magnitude of the estimate is closer as censored sample
size increases.

Statistics Research Letters (SRL) Volume 3, 2014

Harries, C. M. (1967).Queues with stochastic service


rate.Navel Research Logistics Quarterly, 14, 219-230.
Harris, C. M. (1968).The Pareto distribution as a Queue
discipline. Operations Research, 16, 307-313.
Lomax, K. S. (1954). Business failures. Another example of
the analysis of failure data. Journal of the American
Statistical Association, 49, 847852.

ACKNOWLEDGEMENT

The author would like to thank the learned referee for


their suggestions which had improved significantly
the earlier draft of the manuscript.
REFERENCES

Davis, H. T & Feldstein, M. L. (1979).The generalized Pareto


law as a model for progressively censored survival
data.Biometrika, 66, 299-306.
Dyer, D. (1981). Structural probability bounds for the strong
Pareto law. Canadian Journal of Statistics, 9, 71-77.
Freiling, E. C. (1966). A comparison of the fallout mass-size
distributions calculated by lognormal and power-law
models.U.S. Naval Radiological Defence Laboratory, San
Francisco.

28

Madi, M. T. & Raqab, M. Z. (2004).Bayesian prediction of


temperature

records

using

the

Pareto

model.

Environmetrics, 15 (7), 701710.


Prakash, G. and Singh, D. C. (2013).Bayes prediction
intervals for the Pareto model. Journal of Probability and
Statistical Science, 11(1), 109-122.
Steindle, J (1965). Random processes and the growth of firms,
a study of the Pareto law. New York, Heffner.
Gyan Prakash: Presently working as an Assistant Professor
in Statistics, in the department of Community Medicine, S. N.
Government Medical College, Agra, U. P., India. His
interests are in methods based on classical inference and
testing (probabilistic concepts), data assimilation and
decision-based processes. Prakash holds M.Sc. degree and
Ph. D. in Statistics. His current interest is on Bayesian
analysis and data simulation.