You are on page 1of 13

SYNOPSIS

On

PROPERTIES AND APPLICATIONS OF ORDERED


RANDOM VARIABLES

Submitted by

MOHD AMIR

for

Ph. D. in Statistics
(2020)

Under the supervision of


Dr. Haseeb Athar

Department of Statistics and Operations Research


Aligarh Muslim University
Aligarh
Introduction

A set of random variables are called ordered random variables if they are
arranged in non-decreasing or decreasing order of magnitude. Ordered random
variables play a very important role in statistical theory and application.
Ordinary order statistics, record values, progressive type-II censored order
statistics and sequential order statistics are various models of ordered random
variables and have found remarkable uses in a variety of fields such as
reliability theory, survival analysis, financial economics, structural engineering
and so on. These models are used in description of various methodology and
estimation theory in Statistics due to their un-sophistication and robustness,
even at the cost of some loss of efficiency.

Sometimes, the observed data are not in its complete form but are available in
the form of some particular like an ordinary order (ascending or descending
order), records or censored, etc. In such situations, an appropriate model of the
ordered random variable is used to estimate the population parameters. These
models would belong to ordinary order statistics, record values, censored order
statistics, sequential order statistics and so on.

Suppose we have a set of n random variables say, X 1 , X 2 , X 3 ,. . . , X n , and if we


arrange them in ascending order of magnitude and written as,

X ( 1) ≤ X (2) ≤ X (3) ≤ .. . ≤ X (n )

The general notation of the k th order statistic is X ( k ) . Note X ( k ) is different from X k.


X k is the k th random variable from our set, whereas X ( k ) is the k thorder statistic

from our set. X ( k ) takes the value of X ( k ) if X k is the k th random variable when the
realizations are arranged in ascending order.

The 1st order statistic X ( 1) is the set of the minimum values from the realization of
the set of n random variables. The nth order statistic X ( n) is the set of the
maximum values (nth minimum values) from the realization of the set of n
random variables.

Record Values

Extreme values frequently occur in lifetime data analysis. The choice of taking
record values can be justified in many real-life situations, where the data are
available in the form of extremes i.e. only those data are recorded which are
higher/lower than the previous highest/lowest value. Record values are widely
used in extreme weather conditions, sports, economics, insurance, reliability
and in many other real-life perspectives. It may also be helpful as a model for
successively largest insurance claims in non-life insurance, for highest water
levels or highest temperatures. Record values are also used in the reliability
theory.
The concept of record value was first introduced by Chandler (1952). Further
contributions continued with the work of many authors and researchers, among
them Balakrishnan et al. (1993), Arnold et al. (1998), Balakrishnan and Chan
(1998), Rao and Shanbhag (1998), Ahsanullah (2004), Raqab (2002), Ahmadi et
al. (2009) and Wergen (2014) are notable.

Definition
Let { X i } ,i ∈ N , be sequence of iid continuous random variables with distribution
function F ( x) and pdf f ( x ) . Let Y j=max ⁡(X 1 , X 2 ,.. . , X j ) for j ≥ 1. We say X j is an
upper record if Y j>Y j−1 . An analogous definition deals with lower record value.
One can transform the upper record by replacing the original sequence of X j by

−X j , j≥ 1 or if P ( X j> 0 ) =1 for all j then by


1
{
Xj }
, j≥ 1 , the lower record value of

this sequence will correspond to the upper record values of original sequence.
The indices at which the upper record value occur are given by the record time
{ U ( n ) } , n>0 , i .e . , X U (n) is the nth upper record, where
U ( n )=min[ j<U ( n−1 ) , n>1] and U ( 1 )=1.

The distribution of U ( n) does not depend on F . Further, we will denote L(n) as


the indices where the lower record values occur. By assumption U ( 1 )=L ( 1 )=1.
The distribution of L(n) also does not depend on F .

Generalized order statistics

The concept of generalized order statistics (gos) was first introduced by Kamps
(1995). Generalized order statistics includes various important models related to
ordered random variables like order statistics, record values, sequential order
statistics, order statistics with non-integral sample size and progressive type II
censored order statistics etc.
n−1

Let m= ( m1 , m2 , . . . ,mn−1 ) ∈ Rn−1 , M r =∏ m j ,1 ≤ r ≤ n−1,


n ∈ N , n ≥ 2, k ≥ 1 , ~ be the
j=r

parameters such that γ=k +n−r + M r >0 , for all r ∈ {1 , 2 ,. . . ,n−1 } .

Then X ( 1 , n , ~
m , k ) , X ( 2, n , ~
m , k ) ,. . . , X ( n , n , ~
m , k ) are called generalized order

statistics (gos) if their joint pdf is given by (Kamps, 1995)

(∏ )
n−1
f x (1 , n ,~m , k ) , X (2 ,n , ~m ,k ) ,.. ., X (n , n , ~m, k ) ( x1 , x 2 , . .. , x n )=k γj ¿
j=1

on the cone F−1 ( 0 )< x1 ≤ x 2 ≤. . .≤ x n ≤ F−1 ( 1 ) of Rn

Variants of generalized order statistics:

mr , 1≤ r ≤ n−1 k γ r , 1≤ r ≤ n−1 models


0 1 n-r+1 Order statistics

0 α −n+ 1 α −r +1 Order statistics with


non-integral sample
size

( n−r +1 ) α r −( n−r ) α r +1−1 αn ( n−r +1 ) α r Sequential order


statistics

−1 1 1 Record values,
occurrence times of
no-homogeneous
Poisson process,
minimal repair,
relevation
transformation
−1 k∈N k k −th record values

β r− βr +1−1 βn βr Pfeifer’s record


values

β r k r −β r +1 k r+1−1 β n kn βr kr k n−record values

α r k r −α r+1 k r +1−1 α n kn α r kr Ordering via


truncation
0 , if r ≠ r 1 v−n1−n+1 v−r +1 ,if r ≤ r 1 Progressive type II
n1 , if r=r 1 v−n1−r+ 1, if r > r 1 censoring

Estimation Theory

A branch of statistics is called “estimation theory” that deals with estimating the
values of some unknown characteristic called parameter based on a measured
empirical testimony (population) which is random in nature. By knowing the
values of such parameters, one can make statistical inference about the
premeditated population. The term ‘Estimation’ appeared in variety of fields but
in the theory of statistics, it become popular during the period 1920–30. Fisher
(1922) and other bring this infrastructure which depict the problem of
estimation as the combination of some statistical procedures which are as
follow:

 Specification of the statistical model, which is called the ‘distribution of


population’.
 Estimation of the parameter (say θ).
 Problem of obtaining sampling distribution of the estimators of θ , for
comparing their performance and eventually to reach at one estimator of θ
, using these sampling distributions, which is the most desirable sense.

Further, in estimation theory, there are two broad categories of interpretation


regarding the estimation of the population parameters which are:

 Classical (also called conventional or frequentist or non-Bayesian)


estimation.
 Bayesian estimation.

Classical or non-Bayesian estimation

In this approach, estimation of the parameters is directly obtained from the


sample data by emphasizing the frequency of data. It is also called ‘frequentist
approach’. This approach has been associated with the frequentist interpretation
of probabilities. Mathematically, a point estimator can be defined as:

Let g(θ)be a measurable function of θ on parameter space Θ ⊆R . A point


estimator of g(θ) is a function T =T ( X )=T ( X 1 , X 2 , . .. , X n ) , defined for all possible
values of X, taking the values in g ( Θ )={ g ( θ )|θ ∈Θ } . In another words, if X =x is
the observed values and T ( X ) is an estimator of g ( θ ) , then T ( x) is called a
estimate of g ( θ ) .
Properties of estimators

An estimator is said to be a good estimator if its estimated value is near to the


true value parameter. There may be a number of estimators available for the
given parameter. Therefore, it is important to find a good estimator. There are
some criteria which should be satisfied by a good estimator:

 Unbiasedness
 Consistency
 Sufficiency
 Efficiency

Bayesian Estimation

Bayesian analysis combines two sources of information about the unknown


parameters of interest. The first one is the sample data (likelihood function) and
the second is prior distribution, which represents additional information that is
available to analyst. Suppose we have a random sample of size mi.e.,
(x 1 , x 2 ,. . . , x m ). Which we regard as independent identically distributed random

variables with distribution function F (X∨ λ) and pdf f ( x∨λ), where λ is a


labelling parameter, real valued or a vector valued.

Since we are uncertain about the true value of the parameter λ , we will
consider λ as a random variable with distribution function F (λ) or pdf f ( λ ) . The
parameter space is denoted by Ω . The rules of probability are used directly to
make inferences about the parameters. Probability statements about parameters
must be interpreted as “degree of belief”. The prior distribution must be
subjective. Let us suppose that m items are put to test and it is assumed that their
recorded life items having sample size mfrom a population with pdf f ( x| λ )to be
specific, we will assume λ being a real valued. It is also agreed λ as random
variable having pdf f ( λ ) .
Moments and recurrence relations of ordered random variables

Order statistics and their moments have received attention from the beginning
of last century since Galton (1902) and Pearson (1902) studied the distribution
of the difference of the successive order statistics. The exact moment of order
statistics assumes significant importance in the statistics literature and have
been tabulated numerically for several distributions. For significant
contributions on this topic, one may be referred to David and Nagarajah (2003),
Sarhan and Greenberg (1962), Arnold and Balakrishnan (1989) and Arnold et
al. (1992).

Linear functions of the order statistics of which the range is a simple example,
are extremely useful in the estimation of parameters. Knowledge of the mean,
variance, and covariances of the order statistic involved allows us to find the
expected value and variances of the linear function, and hence permits us to
obtain estimators and their efficiencies. The means are also of interest in
selection problems and in so-called scoring procedures, where due to
uncertainty about the underlying distribution, ordered observations
X (i) ,i=1,2, … , n are replaced by their scores E( Z (i ) ), Z(i) being ordered variates

from some standardised distribution such as the unit normal.

There are mainly three reasons due to which recurrence relations and identities
have great importance, as follows

 Reduces the amount of direct computation and hence reduces the time
and labour.
 Provide some simple checks to test the accuracy of computation of
moments of order statistic.
 They express the higher order moments in terms of lower order moments
and hence make the evaluation of higher order moments easy.
Objective of Research

 To develop the moments, recurrence relation and properties of


generalized order statistics and record values by using some new
continuous distribution.

 To illustrate the applications on record values and find the moments by


using a family of distribution.

 To apply the models of ordered random variables in estimation problems.

 For practical purpose and uses of study we can include numerical values
of different parameters of continuous distribution.

 These numerical values can be found out by using any computer language
like R, Mathematica, Python, etc.

Review of literature

 Developments of order statistics through the early 1960’s were described


in the volume edited by Sarhan and Greenberg (1962a). A comprehensive
survey of order statistics can be found in David (1981). Interesting related
studies are described in Harter (1970), Galambos (1978), Barnett and
Lewis (1984), Castillo (1988), and Balakrishnan and Cohen (1991).
 The significant contributions on moments of order statistics are made by
David and Nagarajah (2003), Sarhan and Greenberg (1962), Arnold and
Balakrishnan (1989) and Arnold et al. (1992). Shah (1966) derived the
product moments of order statistics from logistic distribution. Later on,
Shah (1970) discussed the recurrence relation for the moments of all
order statistics from logistic distribution. Khan et al. (1983a) found
general results for the kth moment of order statistics without considering
any particular distribution. Kamps (1995a) investigated the recurrence
relations for moments of generalized order statistics based on non-
identically distributed random variables, which contains order statistics
and record values as special case.
 Chandler (1952) shown several properties of record values and noted
their Markovian structure. Two books on records by Ahsanullah (1995)
and Arnold et al. (1998) are worth mentioning. Several other researchers
have discussed the concept of record values in the literature. For example,
Ahsanullah and Raqab (1999), Awad and Raqab (2000), Lee (2001),
Bieniek and Szynal (2002), Athar et al. (2003), AL-Hussaini and Ahmad
(2003a), Ahsanullah (2004), Shawky and Abu-Zinadah (2008),
Ahsanullah et al. (2010), Faizan and Khan (2011), Lee and Kim (2011),
Minimol and Thomas (2013), Bashir and Akhtar (2014), Paul and
Thomas (2015), Khan et al. (2015), Kumar (2016), Chacko and
Muraleedharan (2018).

RESEARCH PLAN

My research strategy will consist of the following steps

 The first step is to finish the required course work, which consists of papers
on research methodology and on topics related to the research field.
 The next step is to dig wide information on prior research work done in my
field. Great literature would be helpful for producing new ideas and
building some more new work in my research area.
 After getting information from previous work, I will try to solve some
research problems by novel approach.
References
Ahsanullah, M. (1995): Record Statistics. Nova Science Publishers, New York.

Al-Hussaini, E. K. and Ahmad, A. E. B. A. (2003). On Bayesian interval


predication of future records.

Arnold, B.C., Balakrishnan, N. and Nagaraj H. N. (1992): A first course in


order Statistics. John Wiley.

Arnold, B.C., Balakrishnan, N. and Nagaraj H. N. (1998): Records. John Wiley.

Asgharzadeh, A. Fallah, A., Raqab, M. Z., and Valiollahi, R. (2018a). Statistical


inference based on Lindey record data. Statistical papers.

Awad, A. M. and Raqab, M. Z. (2000). Prediction intervals for the future record
values from exponential distribution: comparative study. Journal of Statistical
Computation and Simulation.

Balakrishnan N, Cohen AC (1991) Order Statistics and Inference: Estimation


Methods. Academic Press, San Diego.

Bolstad, W. M. and Curran, J. M. (2017): Introduction to Bayesian Statistics. 3 rd


Edition, John Wiley & Son, Inc., USA.

Chandler, K.N. (1952): The distribution and frequency of record values. J. R.


Statist. Soc., B 14, 220-228.
David, H. A. and Nagaraj, H. N. (2003): Order Statistics, Third Edition, John
Wiley.

Doostparast, M. (2009). A note on estimation based on record data. Metrika,


69(1): 69-80.
Gelman et al. (2014): Bayesian Data Analysis. CRC Press.
Hoff, P.D. (2009): A First Course in Bayesian Statistical Methods. Springer.

Kamps, U (1995): A Concept of Generalized Order Statistics. B.G. Teubner


Stuttgart.

Shahbaz, M.Q., Ahsanullah, M., Shahbaz, S.H. and Al-Zahrani, B.M. (2016).
Ordered Random Variables: Theory and Applications. Atlantis Press, Paris,
France.

Shakil, M. and Ahsanullah, M. (2011). Record values of the ratio of Raylegh


random variables. Pakistan Journal of Statistics.

Shorrok, R. W. (1972). On record values and record times. Journal of Applied


Probability, 9(2):316-326.
Singh, S., and Kumar, D. (2019). Statistical inference based on generalized
Lindey record values. Journal of applied Statistics.

Soliman, A. A., and Al-Abound, F. M. (2008). Bayesian Inference using record


values from Rayleigh model with application. European journal of Operational
Research.

Soliman, A., Amin, E. A. and Abd-El Aziz, A. A. (2010). Estimation and


prediction from inverse Rayleigh distribution based on lower record values.
Applied Mathematical Science.

Wang, B. X. and Ye, Z. S. (2015). Inference on the Weibull distribution based


on record values. Computational Statistics & Data Analysis.

Some recent literature reviews.

You might also like