Académique Documents
Professionnel Documents
Culture Documents
(6)
One is interested in an item for which there is a significant
individual variation in ability to with stand shocks. There may
be no practical way to inspect an individual item to determine
its threshold y. In the case, the threshold must be a random
variable. The shock survival probability are given by
0
i k
P X y g x H x dx
1
k
g By convolution (7)
Therefore S t P T t is the survival function
which gives the probability that the cumulative antigenic
diversity will fail only after time t.
0 k
P {There are exactly k contacts in (0, t] *P (The
total cumulative antigenic diversity (0, t]}
It may be happen that successive shock become
increasingly effective in causing damage, even though they are
independent. This means that V
k
(t), the distribution function
of the k
th
damage is decreasing in k = 1,2, for each t. A
renewal process is a counting process such that the time until
the first event occurs has some distribution F, the time
between the first and second event has, independently of the
time of the first event, the same distribution F, and so on.
When an event occurs, we say that a renewal has taken place.
It is also known from renewal process that
P (exactly k policy decision in (0, t]) = F
k
(t) F
k+1
with
F
0
(t)
= 1
0
k i
k
P T t V t P X y
Data that measure the length of time until the occurrence
of an event are called lifetime, failure times or survival data.
L (T) =1-S (t)
Taking Laplace transform of L (T), we get
1
0
1
1
k
k k
k
F t F t g
1
1
1 1
k
k
k
g F t g (8)
Laplace transforms l
*(s)
is given by
1
1
1
1
g f s
l s
g f s
(9)
Using the results discussed in section 6, let the random
variable U denoting inter arrival time of Mittag-Leffler
distribution with parameter . Now
1
f s ,
substituting in the equation (9) we get,
1
1
1
g
l s
S g
0
1
/
1
1
s
d
l s
ds
g
1
1
1
E T
g
(10)
2
2
0 2 2
2
2
/
1
1
s
d
E T l s
ds
g
2
2
V T E T E T
2
2
2
2 1
1
1
1
1
V T
g
g
(11)
The inter arrival time of the threshold follows Mittag-
Leffler distribution. Using the result given in equation (5),
based on the assumption, the Laplace transforms of the
Mittag-Leffler is given by
1
1
.
1
. ~ . , 1 ~
1
g Mit g Mit .
2
1
E T (12)
Similarly
2
2
2
2
1
V T (13)
Bonfring International Journal of Industrial Engineering and Management Science, Vol. 4, No. 3, August 2014 157
ISSN 2277-5056 | 2014 Bonfring
F. Numerical Illustration
Table 1: Variation in and for Changes in when
=2 Fixed
= 0.5 = 1.0 = 1.5
E (T) V (T) E (T) V (T) E (T) V (T)
0.1 14.142 200.0 13.23 177.77 12.612 159.06
0.2 7.071 50.00 6.67 44.44 6.306 39.765
0.3 4.714 22.22 4.44 19.75 4.204 17.673
0.4 3.535 12.5 3.33 11.11 3.153 9.941
0.5 2.828 8.00 2.67 7.11 2.522 6.362
0.6 2.357 5.55 2.22 4.938 2.102 4.418
0.7 2.020 4.081 1.90 3.628 1.801 3.246
0.8 1.767 3.125 1.67 2.778 1.577 2.485
0.9 1.571 2.469 1.48 2.19 1.401 1.963
1.0 1.414 2.00 1.33 1.778 1.261 1.590
Figure 1: Variation in for Changes in when =2 Fixed
Figure 2: Variation in for Changes in when =2 Fixed
As interarrival time increases keeping = 0.5, = 1.0
and = 1.5 fixed the expected time to recruitment and its
variance decreases. The model discussed in this paper,
indicate how the real life situations which may arise in the
management of manpower planning.
a. Numerical Computation of Two Parameter Generalized
Logistic Distribution
An extensive simulation study is carried out to compare
the performance of different methods maximum likelihood
Estimation, Method of Moment Estimators, Estimators based
on percentiles, Least Squares and Weighted Least Squares
Estimations, L-Moment Estimators and unbiased estimator
together with the results discussed in section 4. The average
relative estimates and average mean square errors of =0.5,
=1.0, =1.5, =2.0, =2.5, when is known and the results
are given in Table 2 and Figure 3.
Table: 2 Average Relative Estimates and its Mean Square
Error when =1
Method
N =0.5 =1.0 =1.5 =2.0 =2.5
MLE
MME
PCE
LSE
WLSE
UBE
10
20
30
50
100
4.132
4.104
3.723
4.071
4.052
4.005
3.501
3.423
3.002
3.074
3.328
3.245
3.652
3.523
3.214
3.435
3.516
3.107
3.504
3.078
2.934
3.062
3.071
2.000
3.358
3.088
2.814
3.073
3.064
2.876
Figure 3
The two parameter Generalized logistic distribution of the
GL (, ) can be easily obtained through the transformation
1
1
ln 1 , x U
where U is a uniform
deviate on (0, 1). The software Mathematica can be used to
generate the generalized logistic random variables and for
computing the roots as well as for computing the minimization
or maximization of the related functions. For generating
Uniform deviate an (0,1), the procedure suggested by
Rubinstein [16]) has been used. Since
is the scale parameter
and all the estimators are scale invariant, we take =1 in all
cases considered. We consider various choices of the shape
parameter =0.5, =1.0, =1.5, =2.0, =2.5 and for sample
sizes n=10, 20, 30, 50 and 100. We compute the average
relative biases and average relative mean square errors over
1000 runs. This number of runs will give the accuracy in the
order
5
1000 0.01 . First we consider the estimation
of when is known. If is known, the maximum likelihood
estimations, percentiles estimations and unbiased best
estimators (UBEs) of can be obtained directly. The least
0
2
4
6
8
10
12
14
16
1 2 3 4 5 6 7 8 9 10
E
x
p
e
c
t
e
d
t
i
m
e
(
T
)
Inter arrival time c
=0.5
=1.0
=1.5
0
50
100
150
200
250
1 2 3 4 5 6 7 8 9 10
V
a
r
i
a
n
c
e
(
T
)
Inter arrival time c
=0.5
=1.0
=1.5
0
1
2
3
4
5
0.5 1 1.5 2 2.5
Series1
Series2
Series3
Series4
Series5
Bonfring International Journal of Industrial Engineering and Management Science, Vol. 4, No. 3, August 2014 158
ISSN 2277-5056 | 2014 Bonfring
squares and weighted least square estimators can be obtained
minimizing with respect to only.
For simulation, the algorithm and results suggested by
Karian et al., [10], Gupta et al., [8], Mudholkar et al., [13], can
be used. It is observed from Table 1 the most of the estimators
usually over estimate , except percentiles estimation, which
under estimates in all cases considered. One can also
observe that for each estimation method, the average relative
biases and average relative mean square errors decrease, as the
sample size increases. As far as biases are concerned, the
unbiased estimators are more or less unbiased as expected
and have minimum mean square errors. The percentiles
estimations provide the best results for small sample size
(n=10) while the unbiased estimators work well otherwise.
The weighted least square estimators work better than the
least square estimators in all cases, considered. Clearly, as
sample size increases, the biases and the mean square errors
decrease for all methods. As a consequence of that, all
estimators are asymptotically unbiased and consistent of
when is known. It is observed that the average biases and
mean square errors of these estimators depend on . For all
methods, as increase, the average relative mean square
errors of
decrease. It is also observed that all methods
overestimate , where as percentiles estimation underestimates
in all cases considered.
Comparing the biases, it is observed that percentiles
estimations perform well for small sample size n=10 and for
all values of considered. For moderate or large sample sizes
20 n , percentiles estimations works well. In the sense
of mean square errors, it is observed that percentiles
estimations performed well for sample sizes and for all
choices of s considered. For moderate and large sample
sizes, maximum likelihood estimators work better than other
estimators except the percentiles estimations. The mean
square errors of the weighted least square estimators are
usually smaller than that of the least square estimators where
as the mean square errors of the method of moment
estimators are usually the maximum. Considering all the
points and taking in to the account that percentiles
estimations and method of moment estimators are easiest to
compute and found that percentiles estimation as a good
estimate of .
Comparing the performance of all the estimators, it is
observed that as far as biases are concerned, least square
estimators work the best in almost all the cases considered for
estimating both and . The performances of the weighted
least square estimators are quite close. For small sample size,
the percentiles estimation performs well when compared with
lest square estimator and weighted least square estimator for
estimating . In the sense of mean square estimators it is
clear that percentiles estimations work the best in almost all
the cases considered for estimating both parameters. The
performances of the maximum likelihood estimators are also
quite close to that of the percentiles estimations for estimating