SEMI-PARAMETRIC ESTIMATION OF Px,y({(x,y)/x >y}) FOR THE POWER FUNCTION DISTRIBUTION

IJESM Volume 3, Issue 2 ISSN: 2320-0294
_________________________________________________________
A Quarterly Double-Blind Peer Reviewed Refereed Open Access International e-Journal - Included in the International Serial Directories
Indexed & Listed at: Ulrich's Periodicals Directory ©, U.S.A., Open J-Gage, India as well as in Cabell’s Directories of Publishing Opportunities, U.S.A.
International Journal of Engineering, Science and Mathematics
http://www.ijmra.us
94
June
2014
SEMI-PARAMETRIC ESTIMATION OF Px,y({(x,y)/x >y})
FOR THE POWER FUNCTION DISTRIBUTION
Dhanya M*
E S Jeevanand**
Abstract:
The stress-strength model describes the life of a component which has a random strength X and
is subjected to random stress Y, in the context of reliability. The component will function
satisfactorily whenever X>Y and it fails at the instant the stress applied to it exceeds the strength.
R=P(Y<X) is a measure of component reliability .In this paper, we obtain semi parametric
estimators of the reliability under stress- strength model for the Power function distribution
under complete and censored samples. We illustrate the performance of the estimators using a
simulation study.
Key words: Stress strength model, Power function distribution, Censored sample, Semi
parametric estimates.
*
Department of Management Studies, Federal Institute of science and Technology, Angamaly ,
Kochi-683577, India
**
Department of Mathematics, Union Christian College, Aluva-2 , Kerala, India

_________________________________________________________
95
June
2014
1. Introduction
The stress-strength model describes the life of a component which has a random strength X and
is subjected to random stress Y, in the context of reliability. The component will function
satisfactorily whenever X>Y and it fails at the instant the stress applied to it exceeds the strength.
R=P(Y<X) is a measure of component reliability .Another interpretation of R is that it measures
the effect of the treatment when X is the response for a control group and Y refers to the
treatment group. It has many applications in engineering concepts such as structures
,deterioration of rocket motors, static fatigue of ceramic components ,the ageing of concrete
pressure vessels. For more real life application in engineering ,reliability, quality control,
medicine and psychology we refer Jeevanand (1997,1998) ,Kotz et .al. (2003), Nadarajah (2005),
Nadarajah and Kotz (2006).
Most of the studies relating to estimation of the reliability measures are restricted to the
parametric framework, either in the classical or Bayesian methods. It is often felt that statistical
models based on conventional parametric distributions are not flexible enough to provide a
reliable description of survival data. This has led to a wide spread use of non-parametric
estimators, notably the Kaplan- Meier estimator for the survival function and the Nelson – Aalen
estimator for the cumulated hazard. Semi parametric estimators are suggested as a compromise
to the conventional parametric and non-parametric estimators.
Power laws appear widely in physics, biology, earth and planetary sciences, economics
and finance, computer science, demography and the social sciences. The city populations size,
the sizes of earthquakes, moon craters, solar flares, computer files and wars, the frequency of
use of words in any human language, the frequency of occurrence of personal names in most
cultures, the numbers of papers scientists write the number of citations received for the papers,
the number of hits on web pages, the sales of books, music recordings sales of branded
commodity, the numbers of species in biological taxa, people’s annual incomes and a host of
other variables all follow power-law distributions. Power law implies that the small occurrence
are extremely common, where as large instance are extremely rare. The power function
distribution is given by pdf

_________________________________________________________
96
June
2014
In this paper, we present semi parametric estimators for reliability of Power function
distribution using complete and censored sampling scheme. In the next section least square
estimation based on survival function are obtained. In section 3 , we consider the estimation
under complete sampling scheme. In section 4, estimation under type I censoring scheme are
discussed. In section 5,we illustrate the performance of the estimators using a simulation study.
2. Least square estimator based on Survival function
Let X and Y are independent power function distribution with pdf
(2.1)
and
(2.2)
P(X>Y) = (2.3)
Afify (2003) derived the least square estimator using the regression of survival function on the
observations, for the Pareto distribution. We used this approach to derive another estimator for
the reliability function R. The survival function at the sample points 1 2, ,..., nx x x ,`from a power
function distribution is
(2.4)
Taking logarithms on both sides of (2.4) we get
ix 1ln ln ix 
 
2
1 1
1
ln ( ) ln ln
n
i i
i
R x x  

 
1
( ) ( ) 1 ( ) , 1,2,3,...i i i
x
R x P X x i

    

_________________________________________________________
97
June
2014
Differentiating partially with respect to the unknown parameters and β, then equating the
results equal to zero we get normal equations as
2
1 1
1 1 1
ln ( )ln( ) (ln ) ln ln 0
n n n
i i i i
i i i
R x x x x  
  
    
2 2
1 1 1
1 1
ln ( ) ln ln 0
n n
i i
i i
n
R x x
  

   
    (2.8)
These normal equations can be solved numerically to obtain the values of 1LS

. In a similar
manner we will obtain the value of 2LS

. Thus
(2.9)
3. Estimation of R using complete sampling
In this section we obtained the estimate based on sample from the
power function density (2.1). An estimate of the survival function ( )iS x is 1-
where is the th
i order statistic and ,the empirical distribution function. In
order to avoid log(0), D’Agostino and Stephans (1986) suggested that can be
approximated by, where 0≤c≤1. In this paper we take the three popular
values for c considered in Wu (2001), Faucher and Tyson (1988) ,viz c=0,0.3 and 0.5.Then (2.7)
and (2.8) becomes
2
21 1
1
1 1
1
ln( ) ln ln 0
1 2
n n
i
i i
n c i n
x
n c
 
 
   
  
  
  
(3.2)
These normal equations can be solved numerically to obtain Proceeding similarly
with and 0≤c≤1 one can obtain the estimate of
. So the estimate of R is obtained by

_________________________________________________________
98
June
2014
1
1 2
cs
cs cs
CR

 


 


(3.3)
4.Estimation of R using type I censored sample
Finaly here we propose a semi parametric estimator for the reliability function of the power
function distribution (2.1) when the strength is censored at a pre determined time T. Suppose n
components with power function life times are put on test and we observe the number of
components failed at each time point x, x+k, x+2k,….up to the time T. Define, nj as the number
of components still functioning at the time xj = x+jk, j=0,1,2,…,xj ≤ T and d-j as the number of
components whose failure occurs in the time interval (xj-1, xj). Then the Kaplan –Meir estimator
of the survival function S(t) (see Jan et al.(2005)) for a given t is
S*(t) = (4.1)
For t<t1,S*(t)=1,where mj as the number of components still functioning at the time y, yj= y+jk,
and dj as the number of components whose failure occurs in the time interval (yj-1, yj).
( )
2
1 1
1 : 1 1
ln(1 ) ln (ln ) ln ln( ) 0
j
n n n
j
i i i
i j x x i ij
n
x x x
d
  
   
 
     
 
   
( )
2 2
1 1 1
1 : 1
( ln(1 ) ln ln 0
j
n n
j
i
i j x x ij
n n
x
d
  

    
     
These normal equations can be solved numerically to obtain . Similarly we will
obtain
This leads to the estimate of R as
1
1 2
TCS
TCS
TCS TCS
R

 


 


(4.4)

_________________________________________________________
99
June
2014
5. Simulation results
In this section, we present the results of a simulation study in order to compare the performance
of these estimators. We perform a simulation study of 2000 samples of sizes n = 25; 50; 100 and
200 generated from (2.1) and (2.2). We have presented the simulation results concerning the bias
and mean square errors of all these estimators. In all the simulation results presented here, the
bias of an estimator can be determined as the (average value of the estimate report in the table -
True value). The variance of an estimator was determined as the sample variance obtained from
all the simulations carried out. Finally, the mean square error of estimator is (variance of the
estimator + (Bias)2
). The bias and mean squared errors of the estimators under complete
sampling and censored sampling are presented in Table 1.
Table 1 : Means and MSEs (in parentheses) of the estimates of R under complete and censored
sampling
(α1, α2,β) (2,2.5,15) (3,3.2,15) (2.6,1.8,15) (1.9,2.6,15)
True R 0.55555 0.51612 0.40909 0.57778
n = 25 LSR

cR

n = 50 LSR

cR

0.39032
(0.0223)
0.41263
(0.0056)
0.40263
(0.0031)
0.40621
(0.0123)
0.40111
(0.0037)
0.43207
(0.0324)
0.46160
(0.0046)
0.4560
(0.0029)
0.44608
(0.0214)
0.44363
(0.0024)
0.56284
(0.0239)
0.57057
(0.0049)
0.56356
(0.0031)
0.55602
(0.0117)
0.55233
(0.0027)
0.37476
(0.0221)
0.40989
(0.0045)
0.39389
(0.0027)
0.40603
(0.0113)
0.40522
(0.0033)

_________________________________________________________
100
June
2014
n = 100 LSR

cR

n=200 LSR

cR

0.41661
(0.0017)
0.41227
(0.0011)
0.42828
(0.0005)
0.43827
(0.00012)
0.42521
(0.0009)
0.44333
(0.0000)
0.44194
(0.0010)
0.46663
(0.0014)
0.45725
(0.0017)
0.46777
(0.0002)
0.44604
(0.0040)
0.4609
(0.0001)
0.48011
(0.0000)
0.48911
(0.00002)
0.57663
(0.0017)
0.56228
(0.0034)
0.57426
(0.0008)
0.57816
(0.0020)
0.56912
(0.0005)
0.59512
(0.0000)
0.59191
(0.0009)
0.38670
(0.0013)
0.41326
(0.0017)
0.41345
(0.0003)
0.41592
(0.00012)
0.40231
(0.0004)
0.42012
(0.0000)
0.42371
(0.0009)
The performance of the estimators are evaluated. It is revealed that the estimators does not seem
very sensitive with variation of the parameters α1, α2 and β. It seems that the MSE and bias of
the estimators become smaller as the sample size increases.

_________________________________________________________
101
June
2014
References
[1] Ali.M.Masoom and Woo J (2005) On the ratio X/(X+Y) for the power function
distribution, Pakistan Journal of Statistics, 21(2), 131-138.
[2] Enis, P. and Geisser, S. (1971) Estimation of probability that Y > X, Journal of American
Statistical Association, 66, 162-168.
[3] Faucher, B. and Tyson, W. R. (1988) On the determination of Weibull parameters.
Journal of Material Science Letters, 7,1199-1203.
[4] Jan, B., Shah, S. W. A., Shah, S. and Qadir, M. F. (2005) Weighted Kaplan Meier
estimation of survival function in heavy censoring, Pakistan Journalof Statistics, 21 (1), 55-
83.
[5] Jeevanand, E.S. and Nair, N.U. (1994) Estimating P[X > Y ] from exponential samples
containing spurious observations, Communications in Statistics
- Theory and Methods, Vol. 23(9), 2629-2642.
[6] Jeevanand, E.S. (1997) Bayes Estimation of P(X2 < X1) for a bivariate Pareto
distribution, The Statistician, A Journal of the Royal Statistical Society, Vol. 46(1) , 93-99.
[7] Jeevanand, E.S. (1998) Bayes estimate of the reliability under stress-strength model for
the Marshal- Olkin bivariate exponential distribution, IAPQR Transactions,1998, 23(2), 133-
136.
[8] Kaplan, E.L. and Meier, P. (1958). Non-parametric estimation from incomplete
observations. J. Amer. Statist. Assoc. 53:457-481.
[9] Kelley, G.D., Kelly, J.A. and Schucany, W.R. (1976) Efficient Estimation of P(Y < X) in
the exponential case, Technometrics, 18, 359-360.
[10]Kim, C., Bae, W., Cho, H., and Park, B. U. (2005) Non-parametric hazard function
estimation using the Kaplan - Meier estimator, Non-parametricStatistics 17 (8), 937 - 948.
[11] Kotz, S., Lumelskii, Y and Pensky, M (2003) The Stress-Strength Model and its
Generalizations, World Scientific, London.

SEMI-PARAMETRIC ESTIMATION OF Px,y({(x,y)/x >y}) FOR THE POWER FUNCTION DISTRIBUTION

More Related Content

What's hot (13)

Similar to SEMI-PARAMETRIC ESTIMATION OF Px,y({(x,y)/x >y}) FOR THE POWER FUNCTION DISTRIBUTION (20)

More from IJESM JOURNAL (20)

Recently uploaded (20)

SEMI-PARAMETRIC ESTIMATION OF Px,y({(x,y)/x >y}) FOR THE POWER FUNCTION DISTRIBUTION