In the present study, we established a new statistical model named as weighted inverse Maxwell distribution (WIMD). Its several statistical properties including moments, moment generating function, characteristics function, order statistics, shanon entropy has been discussed. The expression for reliability, mode, harmonic mean, hazard rate function has been derived. In addition, it also contains some special cases that are well known. Moreover, the behavior of probability density function (p.d.f) has been shown through graphs by choosing different values of parameters. Finally, the performance of the proposed model is explained through two data sets. By which we conclude that the established distribution provides better fit.
2. Weighted Analogue of Inverse Maxwell Distribution with Applications
Aijaz et al. 147
Making the substitution
1
𝑦2 = 𝑧 , so that
1
𝑦3 𝑑𝑦 = −
𝑑𝑧
2
and 𝑦 =
1
√𝑧
, we have
=
2
√𝜋
𝜃
3
2 ∫ 𝑦(
3−𝑘
2
)−1
∞
0
𝑒−
𝜃
𝑧 𝑑𝑧
𝐸(𝑦 𝑘) =
2
√𝜋
𝜃
𝑘
2Γ (
3 − 𝑘
2
) (1.3)
Weighted inverse Maxwell distribution (WIMD)
The weighted distributions have many practical applications to analyze real life data and have been used in various fields
such as biomedicine, ecology, reliability. Whenever existing statistical models does not providing a reliable results then
weighted analogue is used for the improvement of such distributions. Recently various authors have proposed and
analyzed different types of weighted as well as length biased distribution. Das et al introduced the Length biased weighted
Weibull distribution (2011). Abd El-Monsef et al (2015), proposed and studied various structural properties of the weighted
Kumaraswamy distribution. Afaq et al(2016) worked on the Length biased weighted Lomax distribution. Weighted inverse
Rayleigh distribution has been studied by Kawsar fatima et al (2017).Aijaz ahmad dar et al(2018), has introduced
characterization and estimation of weighted Maxwell distribution. Mudasir et al (2018), has introduced weighted version
of generalized inverse Weibull distribution and studied its several statistical properties. Hesham M et al (2017), has
established length biased Erlang distribution and discussed its various properties. In this paper, we create new distribution
which is a Weighted Inverse Maxwell distribution (WIMD) and discussing its several statistical properties.
Definition: suppose 𝑌 is a continuous random variate with p.d.f 𝑓(𝑦), then p.d.f of weighted variate 𝑌𝑤 is defined as.
𝑓𝑤(𝑦) =
𝑤(𝑦)𝑓(𝑦)
𝐸[𝑤(𝑦)]
, 𝑦 > 0 (2.1)
Where 𝑤(𝑦) is a non–negative weight function and 𝐸[𝑤(𝑦)] < ∞. Consider the weight function 𝑤(𝑦) = 𝑦 𝑘
for this
distribution.
The probability distribution function of weighted inverse Maxwell distribution is obtained by using equation (1.1),(1.3) and
(2.1), follows as.
𝑓𝑤(𝑦, 𝜃) =
2𝜃
3−𝑘
2
Γ(
3−k
2
)
𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
, 𝑦 > 0, 𝑘, 𝜃 > 0 (2.2)
The cumulative distribution function (c.d.f) of weighted inverse Maxwell distribution is obtained as.
𝐹𝑤(𝑦, 𝜃) = ∫
2𝜃
3−𝑘
2
Γ (
3−k
2
)
𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
𝑑𝑦
𝑦
0
⇒ 𝐹𝑤(𝑦, 𝜃) =
Γ (
3−k
2
,
θ
y2)
Γ (
3−k
2
)
(2.3)
Where Γ(𝜃, 𝑦) = ∫ 𝑦 𝜃−1
𝑒−𝑦
𝑑𝑦
∞
𝑦
is an upper incomplete gamma function.
Fig (1.1) and (1.2) illustrates the probability density function of Weighted Inverse Maxwell distribution.
3. Weighted Analogue of Inverse Maxwell Distribution with Applications
Int. J. Stat. Math. 148
Special Cases of Weighted Inverse Maxwell Distribution
Case 1: For = 0, then weighted inverse Maxwell distribution (2.2) reduces to inverse Maxwell distribution with p.d.f as .
𝑓(𝑦, 𝜃) =
4
√𝜋
𝜃
3
2
1
𝑦4
𝑒
−
𝜃
𝑦2
Case 2: For 𝑘 = 1, then weighted inverse Maxwell distribution (2.2) reduces to length biased inverse Maxwell distribution
with p.d.f as .
𝑓(𝑦, 𝜃) = 2𝜃
1
𝑦3
𝑒
−
𝜃
𝑦2
Case 3: For = 2,then weighted inverse Maxwell distribution (2.2) reduces to area biased inverse Maxwell distribution with
p.d.f as .
𝑓(𝑦, 𝜃) = 2√
𝜃
𝜋
1
𝑦2
𝑒
−
𝜃
𝑦2
STATISTICAL PROPERTIES
Moments of weighted inverse Maxwell distribution
Let 𝑋 be a random variable from weighted inverse Maxwell distribution with parameter, then its 𝑟 𝑡ℎ
moment is given as.
𝜇 𝑟 = 𝐸(𝑌 𝑟) = ∫ 𝑦 𝑟
∞
0
𝑓𝑤(𝑦, 𝜃, 𝑘)𝑑𝑦
Now using (2.2), we get
= ∫ 𝑦 𝑟
∞
0
2𝜃
3−𝑘
2
Γ (
3−k
2
)
𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
𝑑𝑦
=
2𝜃
3−𝑘
2
Γ (
3−k
2
)
∫ 𝑦 𝑟+𝑘−1
1
𝑦3
𝑒
−
𝜃
𝑦2
𝑑𝑦
∞
0
By making the substitution
𝜃
𝑦2 = 𝑡 sothat
1
𝑦3 𝑑𝑦 = −
1
2𝜃
𝑑𝑡 and 𝑦 = √
𝜃
𝑡
, then the integration yields as.
=
𝜃
𝑟
2
Γ (
3−k
2
)
∫ 𝑡
[1−(𝑟+𝑘)]
2
∞
0
𝑒−𝑡
𝑑𝑡
=
𝜃
𝑟
2
Γ (
3−k
2
)
∫ 𝑡[
3−(𝑟+𝑘)
2
]−1
𝑒−𝑡
𝑑𝑡
∞
0
Therefore
𝜇 𝑟 = 𝐸(𝑌 𝑟) =
𝜃
𝑟
2
Γ (
3−k
2
)
Γ (
3 − (𝑟 + 𝑘)
2
)
Substituting 𝑟 = 1,2 we get
𝑴𝒆𝒂𝒏 = 𝐸(𝑌) =
𝜃
1
2
Γ (
3−k
2
)
Γ (
2 − 𝑘
2
)
𝐸(𝑌2) =
𝜃
Γ (
3−k
2
)
Γ (
1 − 𝑘
2
)
Variance = 𝐸(𝑌2) − [𝐸(𝑌)]2
=
𝜃
Γ(
3−k
2
)
Γ (
1−𝑘
2
) − [
𝜃
1
2
Γ(
3−k
2
)
Γ (
2−𝑘
2
)]
2
Moment generating function of weighted inverse Maxwell distribution
Let y be a random variable from weighted inverse Maxwell distribution, then the moment generating function of ydenoted
by 𝑀 𝑌(𝑡) is given as.
4. Weighted Analogue of Inverse Maxwell Distribution with Applications
Aijaz et al. 149
𝑀 𝑌(𝑡) = 𝐸(𝑒 𝑡𝑦) = ∫ 𝑒 𝑡𝑦
𝑓𝑤(𝑦)𝑑𝑦
∞
0
= ∫ (1 + 𝑡𝑦 +
(𝑡𝑦)2
2!
+ ⋯ ) 𝑓𝑤(𝑦)𝑑𝑦
∞
0
= ∫ ∑
𝑡 𝑟
𝑟!
∞
𝑟=0
∞
0
𝑦 𝑟
𝑓𝑤(𝑦)𝑑𝑦
= ∑
𝑡 𝑟
𝑟!
∞
𝑟=0
∫ 𝑦 𝑟
𝑓𝑤(𝑦)𝑑𝑦
∞
0
= ∑
𝑡 𝑟
𝑟!
∞
𝑟=0
𝜃
𝑟
2
Γ (
3−k
2
)
Γ (
3 − (𝑟 + 𝑘)
2
)
Characteristics function of weighted inverse Maxwell distribution
Let y be a random variable from weighted inverse Maxwell distribution, then the characteristics function of y denoted by
𝜙 𝑌(𝑡) is given as.
𝜙 𝑌(𝑡) = 𝐸(𝑒 𝑖𝑡𝑦) = ∫ 𝑒 𝑖𝑡𝑦
𝑓𝑤(𝑦)
∞
0
𝑑𝑦
= ∫ (1 + 𝑖𝑡𝑦 +
(𝑖𝑡𝑦)2
2!
+ ⋯ ) 𝑓𝑤(𝑦)𝑑𝑦
∞
0
= ∫ ∑
(𝑖𝑡) 𝑟
𝑟!
∞
𝑟=0
∞
0
𝑦 𝑟
𝑓𝑤(𝑦)𝑑𝑦
= ∑
(𝑖𝑡) 𝑟
𝑟!
∞
𝑟=0
∫ 𝑦 𝑟
𝑓𝑤(𝑦)𝑑𝑦
∞
0
= ∑
(𝑖𝑡) 𝑟
𝑟!
∞
0
𝜃
𝑟
2
Γ (
3−k
2
)
Γ (
3 − (𝑟 + 𝑘)
2
)
Harmonic mean of weighted inverse Maxwell distribution
The harmonic mean (H) is given as:
1
𝐻
= 𝐸 (
1
𝑌
) = ∫
1
𝑦
𝑓𝑤(𝑦)
∞
0
𝑑𝑦
1
𝐻
= ∫
1
𝑦
∞
0
2𝜃
3−𝑘
2
Γ (
3−k
2
)
𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
𝑑𝑦
=
2𝜃
3−𝑘
2
Γ (
3−k
2
)
∫ 𝑦 𝑘−5
𝑒
−
𝜃
𝑦2
𝑑𝑦
∞
0
Making the substitution
𝜃
𝑦2 = 𝑧, after solving the integral we get.
1
𝐻
=
1
𝜃
1
2
Γ(
3−𝑘
2
)
Γ (
4 − 𝑘
2
)
Mode of weighted inverse Maxwell distribution
Taking the log of weighted inverse Maxwell distribution, we get
ln 𝑓𝑤(𝑦, 𝜃) = ln 2𝜃
3−𝑘
2 − ln Γ (
3 − 𝑘
2
) + (𝑘 − 4) ln 𝑦 −
𝜃
𝑦
(4.1)
Differentiate (4.1),w.r.t y we get
𝜕 ln 𝑓𝑤(𝑦, 𝜃)
𝜕𝑦
=
𝑘 − 4
𝑦
+
2𝜃
𝑦3
(3.1)
Equating (3.1) to zero and solve for y, we obtain
𝑦0 = √(
2𝜃
4 − 𝑘
)
5. Weighted Analogue of Inverse Maxwell Distribution with Applications
Int. J. Stat. Math. 150
Reliability measures
Suppose Y be a continuous random variable with c.d.f 𝐹𝑤(𝑦) , 𝑦 ≥ 0 .then its reliability function which is also called survival
function is defined as
𝑆 𝑤(𝑦) = 𝑝𝑟(𝑌 > 𝑦) = ∫ 𝑓𝑤(𝑦)
∞
0
𝑑𝑦 = 1 − 𝐹𝑤(𝑦)
The survival function of weighted inverse Maxwell distribution is given as
𝑆 𝑤(𝑦) = 1 −
Γ (
3−k
2
,
θ
y2)
Γ (
3−k
2
)
=
𝛾 (
3−k
2
,
θ
y2)
Γ (
3−k
2
)
(4.1)
The hazard rate function of the random variable y is given as
𝐻 𝑤(𝑦) =
𝑓𝑤(𝑦)
𝑆 𝑤(𝑦)
( 4.2 )
Substituting (2.2) and(4.1),into (4.2), we get.
𝐻 𝑤(𝑦) =
2𝜃
3−𝑘
2 𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
𝛾 (
3−k
2
,
θ
y2)
Shannon’s Entropy of weighted inverse Maxwell distribution
The concept of information entropy was introduced by Shanon in 1948. The entropy can be interpreted as the average
rate at which information is produced by a random source of data and is given by
𝐻 𝑤(𝑥, 𝜃) = −𝐸[log 𝑓𝑤(𝑥, 𝜃)]
= −𝐸 [log (
2𝜃
3−𝑘
2
Γ (
3−k
2
)
𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
)]
= − log (
2𝜃
3−𝑘
2
Γ (
3−k
2
)
) − (𝑘 − 4)𝐸(log 𝑦) + 𝐸 (
𝜃
𝑦2
) (5.1)
Now
𝐸(log 𝑦) = ∫ log 𝑦 𝑓𝑤(𝑦)
∞
0
𝑑𝑦
= ∫ log 𝑦
∞
0
2𝜃
3−𝑘
2
Γ (
3−k
2
)
𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
𝑑𝑦
=
2𝜃
3−𝑘
2
Γ (
3−k
2
)
∫ log 𝑦 𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
𝑑𝑦
∞
0
Making the substitution
𝜃
𝑦2 = 𝑡 so that
1
𝑦3 𝑑𝑦 = −
1
2𝜃
𝑑𝑡, and 𝑦 = (
𝜃
𝑡
)
1
2
=
1
Γ (
3−k
2
)
∫ log (
𝜃
𝑡
)
∞
0
1
2
𝑡
1−𝑘
2 𝑒−𝑡
𝑑𝑡
After solving the integral, we get
=
1
2
[log 𝜃 − 𝜓 (
3 − 𝑘
2
)] (5.2)
Also 𝐸 (
𝜃
𝑦2) = 𝜃 ∫
1
𝑦2
∞
0
𝑓𝑤(𝑦 , 𝜃)𝑑𝑦
=
2𝜃
3−𝑘
2 𝜃
Γ (
3−k
2
)
∫
1
𝑦2
𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
𝑑𝑦
∞
0
6. Weighted Analogue of Inverse Maxwell Distribution with Applications
Aijaz et al. 151
Making substitution
𝜃
𝑦2 = 𝑡 , we get
=
1
Γ (
3−k
2
)
∫ 𝑡
3−𝑘
2
∞
0
𝑒−𝑡
𝑑𝑡
Therefore 𝐸 (
𝜃
𝑦2) =
3−𝑘
2
(5.3)
Substituting the values (5.2), (5.3) in (5.1), we get
𝐻 𝑤(𝑥, 𝜃) = − log
2𝜃
3−𝑘
2 𝜃
Γ (
3−k
2
)
−
𝑘 − 2
2
[log 𝜃 − 𝜓 (
3 − 𝑘
2
)] + (
3 − 𝑘
2
)
Where 𝜓(. ) denotes the digamma function.
Order statistics of weighted inverse Maxwell distribution
Let us suppose 𝑌1, 𝑌2, 𝑌3 … , 𝑌𝑛 be random samples of size n from weighted inverse Maxwell distribution with p.d.f 𝑓(𝑦) and
c.d.f 𝐹(𝑦). Then the probability density function of 𝑘 𝑡ℎ
order statistics is given as.
𝑓𝑋(𝑘)
(𝑦, 𝜃) =
𝑛!
(𝑘 − 1)! (𝑛 − 𝑘)!
[𝐹(𝑦)] 𝑘−1[1 − 𝐹(𝑦)] 𝑛−𝑘
𝑓(𝑥) . 𝑘 = 1,2,3, … . , 𝑛 (6.1)
Now using the equation (2.2) and (2.3) in (6.1). the probability of 𝑘 𝑡ℎ
order statistics of weighted inverse Maxwell
distribution is given as
𝑓𝑤(𝑘)(𝑦, 𝜃) =
2 𝑛! 𝜃
3−𝑘
2 𝑦 𝑘−4
(𝑘 − 1)! (𝑛 − 𝑘)!
𝑒
−
𝜃
𝑦2
[Γ (
3−𝑘
2
)]
𝑛 [𝛾 (
3 − k
2
,
θ
y2
)]
𝑘−1
[Γ (
3 − 𝑘
2
,
𝜃
𝑦2
)]
𝑛−𝑘
Then, the p.d.f of first order 𝑌1 weighted inverse Maxwell distribution is given as
𝑓1 𝑤
(𝑦, 𝜃) =
2𝑛𝜃
3−𝑘
2 𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
[Γ (
3−𝑘
2
)]
𝑛 [Γ (
3 − 𝑘
2
,
𝜃
𝑦2
)]
𝑛−1
And the p.d.f of nth order 𝑌𝑛 weighted inverse Maxwell distribution is given as
𝑓𝑛 𝑤
(𝑦, 𝜃) =
2𝑛𝜃
3−𝑘
2 𝑦 𝑘−4
𝑒
−
𝜃
𝑦2
[Γ (
3−𝑘
2
)]
𝑛 [𝛾 (
3 − k
2
,
θ
y2
)]
𝑛−1
ESTIMATION OF PARAMETER
Method of Moments
In order to obtain the sample moments of weighted inverse Maxwell distribution, we equate population moments with
sample moments
𝜇1 =
1
𝑛
∑ 𝑦𝑖
𝑛
𝑖=1
𝑌̅ =
𝜃
1
2
Γ (
3−k
2
)
Γ (
2 − 𝑘
2
)
𝜃̂ = [𝑌̅
Γ (
3−k
2
)
Γ (
2−𝑘
2
)
]
1
2
Method of maximum likelihood Estimation:
The estimation of parameters of weighted inverse Maxwell distribution is doing by using the method of maximum likelihood
estimation. Suppose 𝑌1, 𝑌2, 𝑌3 … 𝑌𝑛 be random samples of size n from weighted inverse Maxwell distribution. Then the
likelihood function of weighted inverse Maxwell distribution is given as.
𝑙 = ∏ 𝑓(𝑦𝑖, 𝜃)
𝑛
𝑖=1
= ∏
2 𝜃
3−𝑘
2
Γ (
3−k
2
)
𝑦𝑖
𝑘−4
𝑒
−𝜃 ∑
1
𝑦 𝑖
∞
𝑖=0
𝑛
𝑖=1
(7.1)
7. Weighted Analogue of Inverse Maxwell Distribution with Applications
Int. J. Stat. Math. 152
The log likelihood function of (7.1), is given as
ln 𝑙 = ln (
2 𝜃
3−𝑘
2
Γ (
3−k
2
)
)
𝑛
+ ∑ ln 𝑦𝑖
𝑘−4
− ∑
𝜃
𝑦𝑖
2
𝑛
𝑖=1
𝑛
𝑖=1
= 𝑛 ln 2 + 𝑛 (𝜃
3−𝑘
2 ) ln 𝜃 − 𝑛 ln Γ (
3 − k
2
) + (𝑘 − 4) ∑ ln 𝑦𝑖 − 𝜃 ∑
1
𝑦𝑖
2
𝑛
𝑖=1
𝑛
𝑖=1
Differentiate w.r.t 𝜃 , we get
𝜕 ln 𝑙
𝜕𝜃
=
𝑛(3 − 𝑘)
2𝜃
− ∑
1
𝑦𝑖
2
𝑛
𝑖=1
(7.2)
Now equating (7.2), to zero, we get
⇒
𝑛(3 − 𝑘)
2𝜃
− ∑
1
𝑦𝑖
2
𝑛
𝑖=1
= 0
𝜃̂ =
𝑛(3 − 𝑘)
2
∑ 𝑦𝑖
2
𝑛
𝑖
DATA ANALYSIS
Data 1: In this section we provide an application which explains the performance of the newly developed distribution.
The data set has been taken from Gross and Clark (1975), which signifies the relief times of 20 patients getting an
analgesic. We use previous data to associate the fit of the newly developed model with inverse Maxwell distribution.
The data are follows.
1.1,1.4,1.3,1.7,1.9,1.8,1.6,2.2,1.7,2.7,4.1,1.8,1.5,1.2,1.4,3.0,1.7,2.3,1.6,2.0.
In order to compare the two distribution models, we consider the criteria like AIC (Akaike information criterion, AICC
(corrected Akaike information criterion) and BIC (Bayesian information criterion. The better distribution corresponds to
lesser AIC, AICC and BIC values.
𝐴𝐼𝐶 = −2𝑙𝑛𝐿 + 2𝑘, 𝐴𝐼𝐶𝐶 = 𝐴𝐼𝐶 +
2𝑘(𝑘+1)
(𝑛−𝑘−1)
,𝐵𝐼𝐶 = −2𝑙𝑛𝐿 + 𝑘𝑙𝑛
Table 1: ML estimates and Criteria for Comparison
Distribution Estimates Standard Error -2logL AIC AICC BIC
Weighted inverse Maxwell distribution 0.09297414
1.65287637
0.1146933
0.2608437
43.90867 45.90867 46.1308922 45.2097
Inverse Maxwell distribution 0.2070503 0.1690519 69.44787 71.44787 71.6700922 70.7489
Data 2: The data set is on the breaking stress of carbon fibres of 50 mm length (GPa). The data has been previously used
by Cordeiro and Lemonte (2011) and Al-Aqtash et al.(2014). The data is as follows:
0.39, 0.85, 1.08, 1.25, 1.47, 1.57, 1.61, 1.61, 1.69, 1.80, 1.84, 1.87, 1.89, 2.03, 2.03, 2.05, 2.12, 2.35, 2.41, 2.43, 2.48,
2.50, 2.53, 2.55, 2.55, 2.56, 2.59, 2.67, 2.73, 2.74, 2.79, 2.81, 2.82, 2.85, 2.87, 2.88, 2.93, 2.95, 2.96, 2.97, 3.09, 3.11,
3.11, 3.15, 3.15, 3.19, 3.22, 3.22, 3.27, 3.28, 3.31, 3.31, 3.33, 3.39, 3.39, 3.56, 3.60, 3.65, 3.68, 3.70, 3.75, 4.20, 4.38,
4.42, 4.70, 4.90.
Table 2:ML estimates and Criteria for Comparison
Distribution Estimates Standard Error -2logL AIC AICC BIC
Weighted Inverse Distribution 0.02658982
1.99391471
0.03755243
0.11173997
238.6447 240.6447 240.866922 240.464244
Inverse Maxwell Distribution 0.07914596 0.06461209 403.0806 405.0806 405.302822 404.900144
From Table 1 and 2, it has been observed that the Weighted inverse Maxwell model have the lesser AIC, AICC, -
2logL and BIC values as compared to inverse Maxwell distribution. Hence, we can conclude that Weighted inverse
Maxwell distribution leads to a better fit as compared to inverse Maxwell model.