Research Article Volume 11 Issue 3
1Department of Mathematics, Noida International University, Gautam Budh Nagar, India
2Department of Statistics, Assam University, Silchar, India
Correspondence: Kamlesh Kumar Shukla, Department of Mathematics, Noida International University, India
Received: May 31, 2022 | Published: August 9, 2022
Citation: Shukla KK, Shanker R. On statistical properties and applications of Poisson-Pranav distribution. Biom Biostat Int J. 2022;11(3):93-98. DOI: 10.15406/bbij.2022.11.00360
In this paper, Poisson-Pranav distribution, a Poisson mixture of Pranav distribution, has been proposed. Its moments and moments-based measures including coefficients of variation, skewness, kurtosis, index of dispersion have been obtained and their behavior illustrated graphically. The estimation of parameter of the proposed distribution has been discussed using both the method of moment and the method of maximum likelihood. The simulation study has also been presented in order to illustrate the performance of maximum likelihood estimator. The goodness of fit of the proposed distribution has been explained with two count datasets and its fit was found quite satisfactory over Poisson distribution, Poisson-Lindley distribution, Poisson-Akash distribution and Poisson-Ishita distribution.
Keywords: Pranav distribution, moments-based measures, properties, estimation of parameter, simulation, goodness of fit
The Pranav distribution is defined by its probability density function (pdf) and the Cumulative density function (cdf)
f1(x,θ)=θ4θ4+6 (θ+x3)e−θx ;x>0,θ>0f1(x,θ)=θ4θ4+6(θ+x3)e−θx;x>0,θ>0 (1.1)
F1(x,θ)=1−[1+θx(θ2x2+3θx+6)(θ4+6)] e−θx ;x>0,θ>0F1(x,θ)=1−[1+θx(θ2x2+3θx+6)(θ4+6)]e−θx;x>0,θ>0 ( 1.2)
It should be noted that the Pranav distribution, a convex combination of exponential (θ)(θ) and gamma (4,θ)(4,θ) distributions, has been proposed by Shukla1 for modeling lifetime data. Important statistical properties of Pranav distribution including its shapes, moments, skewness, kurtosis, hazard rate function, mean residual life function, stochastic ordering, mean deviations, Bonferroni and Lorenz curves, Renyi entropy measure and stress-strength reliability are available in Shukla.1 The Pranav distribution has been found to provide a better fit for survival time data over exponential distribution, Lindley distribution introduced by Lindley,2 Akash distribution proposed by Shanker3 and Ishita distribution suggested by Shanker and Shukla.4 The pdf of Lindley, Akash, and Ishita distributions has been presented in Table 1.
Lifetime distributions Pdf |
Mixtures of distributions |
Introducer (year) |
|
Lindley |
f(x)=θ2θ+1(1+x)e−θ x;x>0,θ>0f(x)=θ2θ+1(1+x)e−θx;x>0,θ>0 |
exponential (θ)(θ) and gamma (3,θ)(3,θ) distributions |
Lindley2 |
Akash |
f(x)=θ3θ2+2(1+x2)e−θ x;x>0,θ>0f(x)=θ3θ2+2(1+x2)e−θx;x>0,θ>0 |
exponential (θ)(θ) and gamma (3,θ)(3,θ) distributions |
Shanker3 |
Ishita |
f(x;θ)=θ3θ3+2(θ+x2)e−θx ;x>0, θ>0f(x;θ)=θ3θ3+2(θ+x2)e−θx;x>0,θ>0 |
Exponential (θ)(θ) and gamma (3,θ)(3,θ) distributions |
Shanker & Shukla4 |
Table 1 The pdf of Lindley, Akash and Ishita distributions
Ghitany, et al5 have studied in detail on Lindley distribution. Shanker et al.6 have detailed comparative study on modeling of various lifetime data using exponential and Lindley distributions. Further, Shanker et al.7 have detailed comparative study on modeling of real lifetime data using Akash, Lindley and exponential distributions.
During recent decades several one parameter lifetime distributions have been introduced in statistics literature and the Poisson mixture of these distributions, namely Poisson-Lindley distribution (PLD) proposed by Sankaran,8 Poisson-Akash distribution (PAD) introduced by Shanker9 and Poisson-Ishita distribution (PID) suggested by Shukla & Shanker,10 are some among others.
The probability mass function (pmf) of PLD, PAD and PID has been presented in Table 2.
Distributions |
Pmf |
Mixtures of distributions |
Introducer (year) |
PLD |
P(x,θ)=θ2(x+θ+2)(θ+1)x+3 ;x=0,1,2,...,θ>0P(x,θ)=θ2(x+θ+2)(θ+1)x+3;x=0,1,2,...,θ>0 |
Poisson mixture of Lindley |
Sankaran8 |
PAD |
P(x, θ)=θ3θ2+2⋅x2+3x+(θ2+2θ+3)(θ+1)x+3 ;x=0,1,2,...,θ>0 |
Poisson mixture of Akash |
Shanker9 |
PID |
P(x,θ)=θ3(θ3+2).x2+3x+(θ3+2θ2+θ+2)(θ+1)x+3 ;x=0,1,2,...,θ>0 |
Poisson mixture of Ishita |
Shukla & Shanker10 |
Table 2 Pmfs of PLD, PAD and PID
Detailed study of PLD, PAD and PID are available in Ghitany & A1 Mutairi,11 Shanker,9 & Shukla & Shanker,10 respectively. Shanker and Hagos12 has detailed study on applications of PLD in various fields of knowledge.
The main reasons and motivation of introducing Poisson-Pranav distribution (PPD) are (i) it has been observed that Pranav distribution gives a better fit than exponential, Lindley, Akash and Ishita distributions and (ii) it is expected that PPD would prove to be a better model for over PLD, PAD and PID.
This paper has been divided into eight sections. The second section deals with the derivation of the pmf of PPD and its behaviour for varying values of parameter. The third section deals with raw moments and central moments of PPD and behaviour of mean and variance, coefficients of variation, skewness, kurtosis and index of dispersion for varying values of parameter. Increasing hazard rate and unimodality property of the PPD has been discussed in section four. The sections five and six deals with estimation of parameter using both the method of moment and maximum likelihood, and simulation study, respectively. Finally, the goodness of fit of the distribution and its comparative study along with conclusions have been presented in sections seven and eight respectively.
Poisson-Pranav distribution
Assuming that the parameter λ of the Poisson distribution follows Pranav distribution, the Poisson mixture of Pranav distribution can be obtained as
P(X=x)=∞∫0e−λλxx!θ4θ4+6(θ+λ3)e−θ λdλ (2.1)
=θ4(θ4+6)x!∞∫0e−(θ+1)λ(θ λx+λx+3)dλ
=θ4(θ4+6)x3+6x2+11x+θ(θ+1)3+6(θ+1)x+4 ;x=0,1,2,...,θ>0 (2.2)
This is named as Poisson-Pranav distribution (PPD)”. The pmf of PPD presented in Figure 1.
Moments
The r th factorial moment about origin of PPD (2.2) can be obtained as
μ(r)′=E[E(X(r)|λ)] , where X(r)=X(X−1)(X−2)...(X−r+1)
Using (2.1), the r th factorial moment about origin of PPD (2.2) can be obtained as
μ(r)′=E[E(X(r)|λ)]=θ4θ4+6∞∫0[∞∑x=0x(r)e−λλxx!](θ+λ3)e−θ λdλ
=θ4(θ4+6)∞∫0λr[∞∑x= re−λλx−r(x−r)!](θ+λ3)e−θ λdλ
Taking x+r in place of x within the bracket, we get
μ(r)′=θ4θ4+6∞∫0λr[∞∑x=0e−λλxx!](θ+λ3)e−θ λdλ
=θ4θ4+6∞∫0λr(θ+λ3)e−θ λdλ
After simplification, the r th factorial moment about origin of PPD can be expressed as
μ(r)′=r![θ4+(r+1)(r+2)(r+3)]θr(θ4+6) ;r=1,2,3,.... (3.1)
Substituting r=1,2,3, and 4 in (3.1), the first four factorial moments about origin can be obtained and using the relationship between factorial moments about origin and moments about origin, the first four moment about origin of PPD are obtained as
μ1′=θ4+24θ(θ4+6)
μ2′=θ5+2θ4+24θ+120θ2(θ4+6)
μ3′=θ6+6θ6+6θ4+24θ2+360θ+720θ3(θ4+6)
μ4′=θ7+14θ6+36θ5+24θ4+24θ3+840θ2+4320θ+5040θ4(θ4+6)
The relationship between moments about mean and the moments about origin of PPD gives the moments about mean as
μ2=σ2=θ9+θ8+30θ5+84θ4+144θ+144θ2(θ4+6)2
μ3=(θ14+3θ13+2θ12+36θ10+270θ9+396θ8+324θ6+1944θ5+648θ4+864θ2+2592θ+1728)θ3(θ4+6)3
μ4=(θ19+10θ18+18θ17+9θ16+42θ15+852θ14+3132θ13+2808θ12+540θ11+11880θ10+34992θ9+20736θ8+2808θ7+59184θ6+132192θ5+93312θ4+5184θ3+98496θ2+186624θ+93312)θ4(θ4+6)4
The coefficient of variation (C.V) , coefficient of Skewness (√β1) , coefficient of Kurtosis (β2) , and index of dispersion (γ) of the PPD can be obtained using following formula:
C.V=σμ′1 ,√β1=μ3μ23/2
β2=μ4μ22 γ=σ2μ′1 .
The behavior of the mean and the variance of PPD have been shown in Figure 2. Clearly PPD is always over-dispersed (variance greater than the mean).
The behavior of C.V, √β1, β2 and γ of the PPD has been shown graphically for different values of parameter θ in Figure 3.
Figure 3 Behavior of coefficient of variation, coefficient of skewness, coefficient of kurtosis and Index of dispersion of PPD for different values of the parameter θ
Increasing hazard rate and unimodality
The PPD (2.2) has an increasing hazard rate (IHR) and thus unimodal. Clearly
P(x+1;θ)P(x;θ)=1θ+1[1+2x2+13x+17x3+6x2+11x+θ(θ+1)3+6]
is a decreasing function in x . Thus P(x;θ) is log-concave which means that the PPD has an increasing hazard rate (IHR) and unimodal. A detailed discussion about interrelationship between log-concavity, unimodality and IHR for discrete distributions are available in Grandell.13
Method of moment estimate (MOME)
Equating the population mean to the sample mean based on random sample (x1,x2,...,xn) the MOME ˜θ of the parameter ˜θ of PPD is the solution of the following fifth degree polynomial equation
ˉx θ5−θ4+6ˉx θ−24=0 ,
where ˉx is the sample mean.
Maximum Likelihood Estimate (MLE)
Let (x1,x2,...,xn) be a random sample of size n from the PPD and let fx be the corresponding observed frequency The likelihood function L and the log-likelihood function of the PPD is given by
L=(θ4θ4+6)n1(θ+1)k∑x=1(x+4)fxk∏x=1[x3+6x2+11x+θ(θ+1)3+6]fx .
logL=nlog(θ4θ4+6)−k∑x=1(x+4)fxlog(θ+1)+k∑x=1fxlog[x3+6x2+11x+θ(θ+1)3+6] .
The first derivative of the log likelihood function is given by
dlogLdθ=12nθ(θ4+6)−n(ˉx+4)θ+1+k∑x=1(4θ3+9θ2+6θ+1)fx[x3+6x2+11x+θ(θ+1)3+6] .
The maximum likelihood estimate (MLE), ˆθ of the parameter θ of PPD is the solution of the following log likelihood equation
dlogLdθ=12nθ(θ4+6)−n(ˉx+4)θ+1+k∑x=1(4θ3+9θ2+6θ+1)fx[x3+6x2+11x+θ(θ+1)3+6]=0
This non-linear equation can be expressed in closed form and hence can be solved iteratively using Newton- Raphson method available in R-software. The MOME can be taken as the initial value of the parameter for Newton-Raphson method.
For a simulation study, we generate N=10,000 pseudo-random sample of sizes n=50, 100, 150, and 200 of a variable X having PPD). Then using Monte Carlo simulation we estimate the average bias and the mean squared error (MSE) of the MLEs of the parameter for θ =1.5, 2, 2.5 and 3.0. The formulas for finding bias and MSE of the parameter θ are
B(⌢θ)=1NN∑j=1(⌢θj−θ), MSE(⌢θ)=1NN∑j=1(⌢θj−θ)2 ).
Using following algorithm, we generate a pseudo-random sample from PPD
Algorithm
Generate,u∼U(0,1)x→0px⇒θ4(θ(θ+1)3+6)(θ+1)4(θ4+6)while(px<u)dox→x+1px1=px*px−1px⇒px+px1whilereturn(x)
The ML estimate, biases and the mean squares error (MSE) of the parameter based on simulated data are presented in Table 3.
Sample size(n) |
|
Bias |
MSE |
|
50 |
1.5 |
-0.00146 |
0.0001 |
|
2.0 |
0.00545 |
0.00148 |
||
2.5 |
0.00237 |
0.00282 |
||
3.0 |
0.000516 |
0.000013 |
||
100 |
1.5 |
-0.000343 |
0.000011 |
|
2.0 |
-0.000257 |
0.000006 |
||
2.5 |
0.001433 |
0.000205 |
||
3.0 |
-0.00105 |
0.000111 |
||
150 |
1.5 |
-0.000433 |
0.00028 |
|
2.0 |
-0.00002 |
0.0000006 |
||
2.5 |
0.000123 |
0.0000022 |
||
3.0 |
-0.00308 |
0.001431 |
||
200 |
1.5 |
0.00518 |
0.00537 |
|
2.0 |
0.00118 |
0.00028 |
||
2.5 |
0.00039 |
0.00003 |
||
3.0 |
0.00088 |
0.000156 |
Table 3 Estimated Bias and MSE of MLEs (⌢θ)
This table shows that bias and mean square error tends to zero for increasing sample size and increasing values of parameter. Further, MLE of θ has a negative bias in some cases.
In this section two examples of observed count datasets have been considered for goodness of fit of over-dispersed distributions namely PPD, PLD, PAD and PID. The dataset in Table 4 has been taken from Kemp & kemp14 and dataset in Table 5 has been taken from Loeschke & Kohler15 and Janardan & Schaeffer.16 The fitted plots of the distributions for dataset in Tables 4 & 5 have been presented in Figures 4 & 5, respectively.
No. of errors per group |
Observed frequency |
Expected frequency |
|
|||
|
PD |
PLD |
PAD |
PID |
PPD |
|
0 |
35 |
27.4 |
33 |
33.5 |
33.7 |
34.3 |
1 |
11 |
21.5 |
15.3 |
14.7 |
14.5 |
13.8 |
2 |
8 |
8.4 |
6.8 |
6.6 |
6.5 |
6.3 |
3 |
4 |
2.2 |
2.9 |
2.9 |
2.9 |
3.0 |
4 |
2 |
0.5 |
2.0 |
2.3 |
2.4 |
2.6 |
Total |
60 |
60 |
60 |
60 |
60 |
60 |
ML estimate |
ˆθ=0.7833 |
ˆθ=1.7434 |
ˆθ=2.07797 |
ˆθ=2.1171 |
||
χ2 |
7.98 |
2.2 |
1.4 |
1.33 |
1.07 |
|
d.f. |
1 |
1 |
2 |
2 |
2 |
|
p-value |
|
0.0047 |
0.138 |
0.4966 |
0.514 |
0.5856 |
Table 4 Distribution of mistakes in copying groups of random digits
No. of Chromatid aberrations |
Observed frequency |
|
Expected frequency |
||||
PD |
PLD |
PAD |
PID |
PPD |
|||
0 |
268 |
231.3 |
257 |
260.4 |
260.8 |
264.1 |
|
1 |
87 |
126.7 |
93.4 |
89.7 |
89.3 |
85.9 |
|
2 |
26 |
34.7 |
32.8 |
32.1 |
31.8 |
30.7 |
|
3 |
9 |
6.3 |
11.2 |
11.5 |
11.5 |
11.7 |
|
4 |
4 |
0.8 |
3.8 |
4.1 |
4.2 |
4.6 |
|
5 |
2 |
0.1 |
1.2 |
1.4 |
1.5 |
1.8 |
|
6 |
1 |
0.1 |
0.4 |
0.5 |
0.6 |
0.7 |
|
7+ |
3 |
0.1 |
0.2 |
0.3 |
0.3 |
0.5 |
|
Total |
400 |
400.0 |
400.0 |
400.0 |
400.0 |
400.0 |
|
ML estimate |
ˆθ=0.5475 |
ˆθ=2.380442 |
ˆθ=2.659408 |
ˆθ=2.3362 |
ˆθ=2.5388 |
||
χ2 |
38.21 |
6.21 |
4.17 |
3.61 |
2.17 |
||
d.f. |
2 |
3 |
3 |
3 |
3 |
||
p-value |
0.000 |
0.1018 |
0.2437 |
0.3067 |
0.5375 |
Table 5 Distribution of number of chromatid aberrations (0.2 g chinon 1, 24 hours)
A Poisson mixture of Pranav distribution named Poisson-Pranav distribution (PPD) has been proposed. Its factorial moments, raw moments and central moments have been derived. The statistical constants including coefficients of variation, skewness, kurtosis and Index of have been studied. Method of moment and maximum likelihood has been explained. Goodness of fit of PPD over Poisson distribution (PD), PLD, PAD and PID has been discussed with two examples of observed real datasets. PPD gives much closure fit over the considered distributions.
None.
The authors declared no conflicts of interest.
©2022 Shukla, et al. This is an open access article distributed under the terms of the, which permits unrestricted use, distribution, and build upon your work non-commercially.
2 7