Research Article Volume 13 Issue 3
Department of Statistics, Assam University, Silchar, India
Correspondence: Rama Shanker, Department of Statistics, Assam University, Silchar, India
Received: July 01, 2024 | Published: July 15, 2024
Citation: Shanker R, Prodhani HR. On weighted Amarendra distribution with properties and applications. Biom Biostat Int J. 2024;13(3):76-85. DOI: 10.15406/bbij.2024.13.00418
In this paper some statistical properties including nature of hazard function, mean residual life function, and moments based descriptive statistical constants including coefficient of variation, skweness, kurtosis and index of dispersion of the weighted Amarendra distribution has been discussed. Two methods of estimation, namely, maximum likelihood estimation and maximum product spacing estimation have been discussed. The simulation study has been presented to know the consistency of the estimator of parameters given by the two methods of estimation. Confidence intervals of the parameters are given. The goodness of fit of the distribution has been demonstrated with two real lifetime datasets and the goodness of fit shows that weighted Amarendra distribution provides better fit over weighted Pratibha distribution, weighted Komal distribution, weighted Lindley distribution, weighted Garima distribution, weighted Sujatha distribution, weighted Akash distribution and Gamma distribution.
Keywords: amarendra distribution, hazard function, mean residual life function, moments based measures, estimation of parameters, applications
In distribution theory, it is very much useful and practical to add a shape parameter to an existing distribution using a weighted approach because the existing distribution exhibits increased flexibility and tractability tendencies with the inclusion of a shape parameter. Weighted distributions are used to model heterogeneity, clustered sampling, and extraneous variance in the dataset. Fisher1 was the first person to introduce the concept of weighted distributions and it was Rao2 who popularize the concept with several practical and real life examples with some mathematical treatment of weighted distributions. Generally, weighted versions of one parameter lifetime distributions have been derived by several researchers using the weight function w(x,α)=xα−1w(x,α)=xα−1 or w(x,α)=xαw(x,α)=xα and α=1 or α=0α=1 or α=0 thus the corresponding weighted distribution will reduce to the origina distribution for. For examples, Ghitany et al3 proposed weighted Lindley distribution (WLD) from Lindley distribution of Lindley,4 Shanker and Shukla5 proposed weighted Akash distribution (WAkD) from Akash distribution of Shanker,6 Eyob and Shanker7 suggested weighted Garima distribution (WGD) from Garima distribution of Shanker,8 Ganaie et al.9 suggested weighted Aradhana distribution (WArD) from Aradhana distribution of Shanker.10 It is to be noted that the WArD derived by Ganaie et al.9 was having some serious drawbacks and Shanker et al.11 pointed out those drawbacks of WArD and discussed several interesting properties of WArD and suggested some interesting applications. Further, Shanker and Shukla12 suggested weighted Sujatha distribution (WSD) from Sujatha distribution of Shanker,13 Shanker et al.14 suggested weighted Komal distribution (WKD) from Komal distribution of Shanker,15 Shanker et al.16 suggested weighted Uma distribution (WUD) from Uma distribution of Shanker,17 Prodhani and Shanker18 suggested weighted Pratibha distribution (WPD) from Pratibha distribution of Shanker,19 respectively. While testing the goodness of fit of these weighted distributions, it has been observed that in certain datasets, these weighted distributions do not provide a suitable fit due to either distributional nature of weighted distributions or the stochastic nature of the data. Therefore, there is a need for the further weighted version of the existing distribution. Keeping this in mind, an attempt has been made to have detailed study on weighted Amarendra distribution.
Shanker19 introduced a one parameter Amarendra distribution defined by its probability density function (pdf) and cumulative density function (cdf) as
f(x;θ)=θ4θ3+θ2+2θ+6(1+x+x2+x3)xα−1e−θx;x>0,θ>0f(x;θ)=θ4θ3+θ2+2θ+6(1+x+x2+x3)xα−1e−θx;x>0,θ>0 F(x;θ)=1−[1+θ3x3+θ2+(θ+2)x2+θ(θ3+θ2+2θ+6)xθ3+θ2+2θ+6]e−θx;x>0,θ>0F(x;θ)=1−[1+θ3x3+θ2+(θ+2)x2+θ(θ3+θ2+2θ+6)xθ3+θ2+2θ+6]e−θx;x>0,θ>0
Mohiuddin et al.21 derived weighted Amarendra distribution (WAD) using weighted technique with weight functionw(x)=xαw(x)=xα from Amarendra distribution and discussed it statistical properties such as survival function, hazard function, Mill’s ratio, moments based measure such as mean, variance, harmonic mean, moment generating function, characteristics function , order statistics, entropy measures, Bonferroni and Lorenz curves, likelihood ratio test, maximum likelihood estimation, and represent goodness of fit on two datasets and compared WAD with Amarendra distribution and concluded that WAD provides a better fit over Amarendra distribution.
It has been observed that there are several statistical properties of WAD which has not been studied by Mohiuddin et al.21 including moments based measures such as coefficient of skweness, kurtosis, index of dispersion; nature of hazard function and the mean residual life function. Further, there are two serious drawbacks of the WAD proposed by Mohiuddin et al.,21 namely (i) The goodness of fit was compared with Amarendra distribution which is not justifiable due to the fact that a comparison of weighted distribution with unweighted distribution is completely illogical, (ii) WAD was compared with Amarendra distribution without K-S and p-value, and concluded that WAD gives better fit over Amarendra distribution, which is unreasonable and such conclusion would never be acceptable to researchers in statistics.
In this paper, a WAD is proposed using weight functionw(x)=xα−1w(x)=xα−1 from Amarendra distribution. Some of its important statistical properties such as hazard function, mean residual life function, moments based measures including coefficient of variation, skewness, kurtosis and index of dispersion have been derived and discussed. Parameters are estimated by the method of maximum likelihood estimation and maximum product spacing estimation. A simulation study is carried out to show the consistency of the estimator the parameters by maximum likelihood estimation and maximum product spacing estimation. Confidence interval of the parameters has been presented with profile plot of the parameters. Two real lifetime datasets have been presented to explain the applications of WAD and the goodness of fit of WAD has been compared with several weighted distributions including WPD, WKD, WLD, WGD, WSD, WAkD and gamma distribution (GD).
The weighted Amarendra distribution (WAD) can be obtained using weighted technique with weight functionw(x)=xα−1w(x)=xα−1 from Amarendra distribution. The pdf and cdf of WAD can be expressed as
f(x;θ,α)=θα+3(1+x+x2+x3)xα−1e−θx[θ3+αθ2+α(α+1)θ+α(α+1)(α+2)]Γ(α);x>0,θ>0,α>0f(x;θ,α)=θα+3(1+x+x2+x3)xα−1e−θx[θ3+αθ2+α(α+1)θ+α(α+1)(α+2)]Γ(α);x>0,θ>0,α>0 F(x;θ,α)=1−θ3Γ(α,θx)+θ2Γ(α+1,θx)+θ Γ(α+2,θx)+Γ(α+3,θx)[θ3+αθ2+α(α+1)θ+α(α+1)(α+2)]Γ(α);x>0,θ>0,α>0F(x;θ,α)=1−θ3Γ(α,θx)+θ2Γ(α+1,θx)+θΓ(α+2,θx)+Γ(α+3,θx)[θ3+αθ2+α(α+1)θ+α(α+1)(α+2)]Γ(α);x>0,θ>0,α>0
where θθ is a scale parameter and αα is shape parameter of the distribution. If we takeα=α+1α=α+1 , we can get the weighted Amarendra distribution proposed by Mohiuddin et al.20 When α=1α=1 ,WAD reduces to Amarendra distribution. The behaviours of the pdf and cdf of WAD are shown in the following Figures 1 & 2 respectively. For increasing values of shape parameter αα , kurtosis is lower and the curve tends to zero at faster rate. This shows that it positively skewed distribution and becomes symmetrical for increasing values of shape parameter αα .
The r th moment about origin of WAD can be obtained as
μr′=E(Xr)=∞∫0xrg(x;θ,α)dx
=[θ3+(α+r)θ2+(α+r)(α+r+1)θ+(α+r)(α+r+1)(α+r+2)]Γ(α+r)θr[θ3+αθ2+α(α+1)θ+α(α+1)(α+2)]Γ(α);r=1,2,3...
Putting r= 1,2,3,4 we obtain the first four moments about origin as follows μ1′=α[θ3+(α+1)θ2+(α+1)(α+2)θ+(α+1)(α+2)(α+3)]θ[θ3+αθ2+α(α+1)θ+α(α+1)(α+2)] μ2′=α(α+1)[θ3+(α+2)θ2+(α+2)(α+3)θ+(α+2)(α+3)(α+4)]θ2[θ3+αθ2+α(α+1)θ+α(α+1)(α+2)] μ3′=α(α+1)(α+2)[θ3+(α+3)θ2+(α+3)(α+4)θ+(α+3)(α+4)(α+5)]θ3[θ3+αθ2+α(α+1)θ+α(α+1)(α+2)] μ4′=α(α+1)(α+2)(α+3)[θ3+(α+4)θ2+(α+4)(α+5)θ+(α+4)(α+5)(α+6)]θ4[θ3+αθ2+α(α+1)θ+α(α+1)(α+2)]
Now using the relationship between raw moments and central moments we obtain the central moments of WAD as μ2=α(α6+2α5θ+3α4θ2+4α3θ3+3α2θ4+2αθ5+θ6+9α5+14α4θ+18α3θ2+24α2θ3+9αθ4+2θ5+31α4+34α3θ+33α2θ2+44αθ3+6θ4+51α3+34α2θ+18αθ2+24θ3+40α2+12αθ+12α)θ2(α3+α2θ+αθ2+θ3+3α2+αθ+2α)2 μ3=2α(α9+3α8θ+6α7θ2+10α6θ3+12α5θ4+12α4θ5+10α3θ6+6α2θ7+3αθ8+θ9+12α8+30α7θ+51α6θ2+74α5θ3+81α4θ4+78α3θ5+61α2θ6+18αθ7+3θ8+60α7+120α6θ+165α5θ2+198α4θ3+180α3θ4+150α2θ5+111αθ6+12θ7+162α6+246α5θ+255α4θ2+238α3θ3+153α2θ4+84αθ5+60θ6+255α5+273α4θ+189α3θ2+128α2θ3+42αθ4+234α4+156α3θ+54α2θ2+24αθ3+116α3+36α2θ+24α2)θ3(α3+α2θ+αθ2+θ3+3α2+αθ+2α)3 μ4=3α(α13+4α12θ+10α11θ2+20α10θ3+31α9θ4+40α8θ5+44α7θ6+40α6θ7+31α5θ8+20α4θ9+10α3θ10+4α2θ11+αθ12+θ13+20α12+72α11θ+166α10θ2+316α9θ3+450α8θ4+528α7θ5+524α6θ6+408α5θ7+264α4θ8+136α3θ9+46α2θ10+12αθ11+2θ12+173α11+548α10θ+1136α9θ2+2016α8θ3+2550α7θ4+2640α6θ5+2332α5θ6+1560α4θ7+885α3θ8+396α2θ9+76αθ10+8θ11+856α10+2328α9θ+4220α8θ2+6856α7θ3+7396α6θ4+6464α5θ5+4908α4θ6+2648α3θ7+1268α2θ8+520αθ9+40θ10+2691α9+6108α8θ+9362α7θ2+13700α6θ3+11955α5θ4+8264α4θ5+5136α3θ6+1904α2θ7+616αθ8+240θ9+5628α8+1029α7θ+12758α6θ2+16572α5θ3+10834α4θ4+5296α3θ5+2560α2θ6+3448αθ7+7943α7+11180α6θ+10468α5θ2+1912α4θ3+5120α3θ4+1344α2θ5+480αθ6+7480α6+7560α5θ+4744α4θ2+4672α3θ3+976α2θ4+4504α5+2896α4θ+912α3θ2+768α2θ3+1568α4+480α3θ+240α3)θ4(α3+α2θ+αθ2+θ3+3α2+αθ+2α)4
Thus, the coefficient of variation (C.V), coefficient of skewness(√β1), coefficient of kurtosis(β2), and index of dispersion(γ) of WAD are obtained as CV=√μ2μ1′=√α(α6+2α5θ+3α4θ2+4α3θ3+3α2θ4+2αθ5+θ6+9α5+14α4θ+18α3θ2+24α2θ3+9αθ4+2θ5+31α4+34α3θ+33α2θ2+44αθ3+6θ4+51α3+34α2θ+18αθ2+24θ3+40α2+12αθ+12α)α[θ3+(α+1)θ2+(α+1)(α+2)θ+(α+1)(α+2)(α+3)] √β1=μ3μ232 =2α(α9+3α8θ+6α7θ2+10α6θ3+12α5θ4+12α4θ5+10α3θ6+6α2θ7+3αθ8+θ9+12α8+30α7θ+51α6θ2+74α5θ3+81α4θ4+78α3θ5+61α2θ6+18αθ7+3θ8+60α7+120α6θ+165α5θ2+198α4θ3+180α3θ4+150α2θ5+111αθ6+12θ7+162α6+246α5θ+255α4θ2+238α3θ3+153α2θ4+84αθ5+60θ6+255α5+273α4θ+189α3θ2+128α2θ3+42αθ4+234α4+156α3θ+54α2θ2+24α3+116α3+36α2θ+24α2)[α(α6+2α5θ+3α4θ2+4α3θ3+3α2θ4+2αθ5+θ6+9α5+14α4θ+18α3θ2+24α2θ3+9αθ4+2θ5+31α4+34α3θ+33α2θ2+44αθ3+6θ4+51α3+34α2θ+18αθ2+24θ3+40α2+12αθ+12α)]32 =3α(α13+4α12θ+10α11θ2+20α10θ3+31α9θ4+40α8θ5+44α7θ6+40α6θ7+31α5θ8+20α4θ9+10α3θ10+4α2θ11+αθ12+θ13+20α12+72α11θ+166α10θ2+316α9θ3+450α8θ4+528α7θ5+524α6θ6+408α5θ7+264α4θ8+136α3θ9+46α2θ10+12αθ11+2θ12+173α11+548α10θ+1136α9θ2+2016α8θ3+2550α7θ4+2640α6θ5+2332α5θ6+1560α4θ7+885α3θ8+396α2θ9+76αθ10+8θ11+856α10+2328α9θ+4220α8θ2+6856α7θ3+7396α6θ4+6464α5θ5+4908α4θ6+2648α3θ7+ 1268α2θ8+520αθ9+40θ10+2691α9+6108α8θ+9362α7θ2+13700α6θ3+11955α5θ4+8264α4θ5+5136α3θ6+1904α2θ7+616αθ8+240θ9+5628α8+1029α7θ+12758α6θ2+16572α5θ3+10834α4θ4+5296α3θ5+2560α2θ6+3448αθ7+7943α7+11180α6θ+10468α5θ2+1912α4θ3+5120α3θ4+1344α2θ5+480αθ6+7480α6+7560α5θ+4744α4θ2+4672α3θ3+976α2θ4+4504α5+2896α4θ+912α3θ2+768α2θ3+1568α4+480α3θ+240α3)[α(α6+2α5θ+3α4θ2+4α3θ3+3α2θ4+2αθ5+θ6+9α5+14α4θ+18α3θ2+24α2θ3+9αθ4+2θ5+31α4+34α3θ+33α2θ2+44αθ3+6θ4+51α3+34α2θ+18αθ2+24θ3+40α2+12αθ+12α)]2 γ=μ2μ1′ =α(α6+2α5θ+3α4θ2+4α3θ3+3α2θ4+2αθ5+θ6+9α5+14α4θ+18α3θ2+24α2θ3+9αθ4+2θ5+31α4+34α3θ+33α2θ2+44αθ3+6θ4+51α3+34α2θ+18αθ2+24θ3+40α2+12αθ+12α)θα(α3+α2θ+αθ2+θ3+3α2+αθ+2α)[θ3+(α+1)θ2+(α+1)(α+2)θ+(α+1)(α+2)(α+3)]
The behaviors of coefficien of variation, coefficient of skewness, coefficient of kurtosis, index of dispersion for differnet values of the parameters of WAD are presented in the Figure 3. For fixed value of α and increasing values of θ, coefficient of variation and coefficient of kurtosis are increasing, and for fixed value of θand increasing values of αcoefficient of variation decreases and coefficient of kurtosis is first decreases after that increases like U-shape. For all values of the parameter θ and α , coefficient of skewness and index of dispersion are always decreasing.
Reliability function
The survival function of the reliability function of WAD can be obtained as
R(x;θ,α)=1−F(x;θ,α)
=θ3Γ(α,θx)+θ2Γ(α+1,θx)+θΓ(α+2,θx)+Γ(α+3,θx)[θ3+αθ2+α(α+1)θ+α(α+1)(α+2)]Γ(α);x>0,θ>0,α>0
The graphical representation of reliability function is presented in Figure 4.
Hazard function
The hazard function of WAD can be obtained as
h(x;θ,α)=f(x;θ,α)R(x;θ,α)
=θα+3(1+x+x2+x3)xα−1e−θxθ3Γ(α,θx)+θ2Γ(α+1,θx)+θΓ(α+2,θx)+Γ(α+3,θx);x>0,θ>0,α>0
The graphical representation of hazard function is presented in Figure 5.
From the Figure 5, it is clear that for any value of θ and α<1 , it has V-shaped hazard function and for any values of θ and α≥1 it has increasing hazard function.
Reverse hazard function
The reverse hazard function of WAD can be obtained as
r(x;θ,α)=f(x;θ,α)F(x;θ,α) =θα+3(1+x+x2+x3)xα−1e−θx[{θ3+αθ2+α(α+1)θ+α(α+1)(α+2)}Γ(α)−{θ3Γ(α,θx)+θ2Γ(α+1,θx)+θΓ(α+2,θx)+Γ(α+3,θx)}];x>0,θ>0,α>0
Mean residual life function
The mean residual life function of WAD can be obtained as
m(x;θ,α)=1S(x;θ,α)∞∫xtf(t;θ,α)dt−x
=θ3Γ(α+1,θx)+θ2Γ(α+2,θx)+θΓ(α+3,θx)+Γ(α+4,θx)θ3Γ(α,θx)+θ2Γ(α+1,θx)+θΓ(α+2,θx)+Γ(α+3,θx)−x;x>0,θ>0,α>0
The graphical representation of mean residual life function is presented in Figure 6.
Maximum Likelihood Estimation
Let (x1,x2,...,xn) be a random sample from WAD. The log-likelihood function of WAD can be expressed as
logL=n[(α+3)logη−log(θ3+αθ2+α(α+1)θ+α(α+1)(α+2))]−nlog(Γ(α))+n∑i=1log(1+xi+xi2+xi3)+(α−1)n∑i=1log(xi)−θˉx
This gives
∂logL∂θ=n(α+3)θ−n{3θ2+2αθ+α(α+1)}θ3+αθ2+α(α+1)θ+α(α+1)(α+2)−nˉx=0
∂logL∂α=nlogθ−n{θ2+(2α+1)θ+(3α2+6α+2)}θ3+αθ2+α(α+1)θ+α(α+1)(α+2)+nψ(α)+n∑i=1logxi=0
The log-likelihood equations presented here are not readily solvable because it is not in closed form, necessitating the use of maximization techniques using R software. Iterative solutions are employed to optimize the likelihood function until sufficiently close parameter values are achieved. These equations can be solved using Fisher’s scoring method. For Fisher's scoring method, the following approach is undertaken
∂2logL∂θ2=−n(α+3)θ2−n[{θ3+αθ2+α(α+1)θ+α(α+1)(α+2)}(6θ+2α)−{3θ2+2αθ+α(α+1)}2][θ3+αθ2+α(α+1)θ+α(α+1)(α+2)]2
∂2logL∂θ∂α=−nθ−n[{θ3+αθ2+α(α+1)θ+α(α+1)(α+2)}(2θ+2α+1)−n{3θ2+2αθ+α(α+1)}{θ2+(2α+1)+(3α2+6α+2)}][θ3+αθ2+α(α+1)θ+α(α+1)(α+2)]2=∂2logL∂α∂θ
∂2logL∂α2=−n[{θ3+αθ2+α(α+1)θ+α(α+1)(α+2)}(2θ+6α+6)−{θ2+(2α+1)+(3α2+6α+2)}2][θ3+αθ2+α(α+1)θ+α(α+1)(α+2)]2+nψ′(α)
For finding the MLEs (ˆθ,ˆα)
of parameters (θ,α)
of WAD, following equations can be solved
(∂2logL∂θ2∂2logL∂θ ∂α∂2logL∂θ ∂α∂2logL∂α2)ˆθ=θ0ˆα=α0(ˆθ−θ0ˆα−α0)=(∂logL∂θ∂logL∂α)
whereθ0 and are the initial values of θ and α;. These equations are solved iteratively till close estimates of parameters are obtained.
Maximum product spacing estimation
The maximum product spacing estimates (MPSE) (ˆθ,ˆα)
of parameters (θ,α)
can be obtained numerically by maximizing the following function with respect to θ and α.
MPSE=1n+1n+1∑i=1log[F(xi,θ,α)−F(xi−1,θ,α)]
To assess the consistency of maximum likelihood estimators (MLE) and maximum product spacing estimators (MPSE) for WAD, a simulation study has been conducted. The investigation involved examining mean estimates, biases (B), mean square errors (MSEs), and variances of the MLE and MPSE for WAD, utilizing the specified formulas.
Mean=1nn∑i=1ˆHi, B=1nn∑i=1(ˆHi−H), MSE=1nn∑i=1(ˆHi−H)2,Variance=MSE−B2
where H=(θ,α) and ˆHi=(ˆθi,ˆαi) .
The acceptance-rejection method of simulation study has been employed to generate data. This method is commonly used in simulation studies to produce random samples from a target distribution. The method for generating random samples from the WAD involves the following steps:
The biases and MSEs of the MLE and MPSE of the parameters decreases for increasing sample size as evident in Table 1. This supports the first-order asymptotic theory of MLE. From the Table 1, it observed that in case of the parameter θ , MLE provides the better estimate as compared to MPSE and in case of the parameter α, MPSE provides the better estimate as compared to MLE.
Parameter |
Sample size |
MLE |
MPSE |
||||
Mean |
Biased |
MSE |
Mean |
Biased |
MSE |
||
θ=1.5 |
20 40 60 80 100 |
1.48704 1.48878 1.49006 1.49189 1.49376 |
-0.01295 -0.01121 -0.00993 -0.00810 -0.00623
|
0.00034 0.00027 0.00026 0.00021 0.00017 |
1.48465 1.48806 1.48906 1.49116 1.49304 |
-0.01534 -0.01193 -0.01093 -0.00883 -0.00695 |
0.00037 0.00028 0.00026 0.00021 0.00018 |
α=1 |
20 40 60 80 100 |
0.98180 0.98481 0.98619 0.98940 0.99162 |
-0.01819 -0.01518 -0.01380 -0.01059 -0.00837 |
0.00052 0.00040 0.00036 0.00031 0.00026 |
0.98250 0.98499 0.98639 0.98967 0.99167 |
-0.01749 -0.01500 -0.01360 -0.01032 -0.00832 |
0.00038 0.00032 0.00027 0.00022 0.00020 |
θ=1.7 |
20 40 60 80 100 |
1.69411 1.69488 1.69756 1.69853 1.69967 |
-0.00588 -0.00511 -0.00243 -0.00146 -0.00032 |
0.00038 0.00029 0.00026 0.00023 0.00020 |
1.68324 1.68611 1.68800 1.69117 1.69298 |
-0.01675 -0.01388 -0.01199 -0.00882 -0.00701 |
0.00090 0.00059 0.00049 0.00038 0.00031 |
α=2.3 |
20 40 60 80 100 |
2.29306 2.29351 2.29465 2.29503 2.29610 |
-0.00693 -0.00649 -0.00534 -0.00496 -0.00389 |
0.00134 0.00100 0.00086 0.00076 0.00066 |
2.29615 2.29672 2.29773 2.29857 2.29894 |
-0.00384 -0.00327 -0.00226 -0.00142 -0.00105 |
0.00306 0.00203 0.00147 0.00129 0.00108 |
Table 1 Descriptive constants of the parameters of WAD
Variance-Covariance matrix for the parameters θ=1.5, α=1 and θ=1.7, α=2.3 are given by
θ αθα(0.000130.000060.000060.00020) and θ αθα(0.000200.000030.000030.00066)
To test the goodness of fit of WAD , we have considered following two real lifetime datasets.
Dataset 1: The following symmetric data, discussed by Murthy et al,21 relating to the failure times of windshields. The values are as follows.
0.04, 0.3, 0.31, 0.557, 0.943, 1.07, 1.124, 1.248, 1.281, 1.281, 1.303, 1.432, 1.48, 1.51, 1.51,1.568, 1.615, 1.619, 1.652, 1.652, 1.757, 1.795, 1.866, 1.876, 1.899, 1.911, 1.912, 1.9141,0.981, 2.010, 2.038, 2.085, 2.089, 2.097, 2.135, 2.154, 2.190, 2.194, 2.223, 2.224, 2.23, 2.3,2.324, 2.349, 2.385, 2.481, 2.610, 2.625, 2.632, 2.646, 2.661, 2.688, 2.823, 2.89, 2.9, 2.934,2.962, 2.964, 3, 3.1, 3.114, 3.117, 3.166, 3.344, 3.376, 3.385, 3.443, 3.467, 3.478, 3.578,3.595, 3.699, 3.779, 3.924, 4.035, 4.121, 4.167, 4.240, 4.255, 4.278, 4.305, 4.376, 4.449,4.485, 4.570, 4.602, 4.663, 4.694.
The total time to test (TTT) plots and the histogram of the original dataset 1 and the corresponding simulated dataset are shown in the Figure 7.
Dataset-2: The following bi-modal, a set of complete data discussed by Murthy et al22 reports the lifetimes of 20 electronic components. The observations are:
0.03, 0.12, 0.22, 0.35, 0.73, 0.79, 1.25, 1.41, 1.52, 1.79, 1.80, 1.94, 2.38, 2.40, 2.87, 2.99,3.14, 3.17, 4.72, 5.09.
The total time to test (TTT) plots and the histogram of the original dataset 2 and the corresponding simulated dataset are shown in the Figure 8.
In order to compare lifetime distributions, values of−2logL , Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Consistent Akaike Information Criterion (CAIC), Hannan-Quinn Information Criterion (HQIC), Kolmogorov-Smirnov Statistics (K-S) and the corresponding probability value (p-value) for the above data set has been computed. The formulae for computing AIC, BIC, CAIC, HQIC and K-S are as follows:
AIC=−2logL+2p ,BIC=−2logL+plog(n) ,CAIC=−2logL+2pnn−p−1
HQIC=−2logL+2plog[log(n)] ,K-S=Supx|Fm(x)−Fo(x)|
where,p= number of parameters, n= sample size, Fm(x)= empirical cdf of considered distribution and Fo(x)= cdf of considered distribution.
The ML estimates and the MPS estimates of the parameters along with their standard errors (in parenthesis) of the considered distributions for datasets 1 and 2 are given in Tables 2 & 3 respectively. The goodness of fit measures for the datasets 1 and 2 for the considered distributions are presented in Tables 4 & 5 respectively. It is clear from tables 4 and 5 that WAD has the least, AIC, BIC, CAIC, HQIC and K-S values as compared to the WPD, WKD, WLD, WGD, WSD, WAkD and GD, therefore WAD provides the best fit as compared to these considered distributions for the two datasets. The fitted plot of the considered distributions, Q-Q plot, P-P plot and ECDF plot of the dataset-1 and 2 are shown in the Figure 9, which also support the hypothesis that WAD provides best fit among the considered distributions. The confidence Interval of the parameters of WAD for the dataset-1 & 2 are given in Table 6. The profile plots of WAD for the datasets 1 and 2 are given in Figure 10.
Distributions |
MLE |
MPSE |
||
ˆθ |
ˆα |
ˆθ |
ˆα |
|
WAD |
1.7924 (0.2066) |
2.3564 (0.4593) |
1.6634 (0.1912) |
2.0582 (0.4179) |
WPD |
1.6158 (0.2149) |
2.8242 (0.5245) |
1.4798 (0.2004) |
2.4801 (0.4863) |
WKD |
1.4528 (0.2084) |
3.2582 (0.5113) |
1.3282 (0.1945) |
2.9428 (0.4763) |
WLD |
1.4583 (0.2099) |
3.0683 (0.4945) |
1.3318 (0.1956) |
2.7609 (0.4589) |
WGD |
1.4721 (0.2104) |
3.2138 (0.4846) |
1.3490 (0.1964) |
2.9218 (0.4491) |
WSD |
1.6075 (0.2089) |
2.6928 (0.4807) |
1.4115 (0.2150) |
3.7482 (0.5559) |
WAkD |
1.6750 (0.2084) |
2.7341 (0.4712) |
1.9195 (0.2251) |
3.3442 (0.5249) |
GD |
1.3556 (0.2101) |
3.4823 (0.5018) |
1.2333 (0.1961) |
3.1818 (0.46781) |
Table 2 ML estimates and MPS estimates of parameters with their standard errors (in parenthesis) of the parameters of the considered distribution of the dataset-1
Distributions |
MLE |
|
MPSE |
|
ˆθ | ˆα | ˆθ | ˆα | |
WAD |
1.3253 (0.2375) 1.3253 (0.2375) |
0.7877 (0.2795) 0.7877 (0.2795) |
1.1598 (0.2025) 1.1598 (0.2025) |
0.5902 (0.2177) 0.5902 (0.2177) |
WPD |
1.0191 (0.2206) |
0.8391 (0.3227) |
0.8574 (0.1833) |
0.5973 (0.2486) |
WKD |
0.7510 (0.2096) |
1.0218 (0.3258) |
0.5978 (0.1719) |
0.7814 (0.2601) |
WLD |
0.7762 (0.2164) |
0.9515 (0.3079) |
0.6197 (0.1779) |
0.7277 (0.2441) |
WGD |
0.75425 (0.2250) |
1.0728 (0.3071) |
0.5923 (0.1883) |
0.8547 (0.2479) |
WSD |
1.0260 (0.2259) |
0.8422 (0.2908) |
1.1353 (0.2071) |
2.2794 (0.2527) |
WAkD |
1.1117 (0.2352) |
0.9511 (0.3011) |
1.9573 (0.2530) |
2.2681 (0.2449) |
GD |
0.6007 (0.2103) |
1.1627 (0.3280) |
0.4481 (0.1705) |
0.9240 (0.2672) |
Table 3 ML estimates and MPS estimates of parameters with their standard errors (in parenthesis) of the parameters of the considered distribution of the dataset-2
Distributions |
−2logL |
AIC |
BIC |
CAIC |
HQIC |
K-S |
P-value |
WAD |
278.73 |
282.73 |
295.75 |
282.87 |
284.72 |
0.07 |
0.74 |
WPD |
282.87 |
286.87 |
299.89 |
287.01 |
288.86 |
0.13 |
0.08 |
WKD |
286.46 |
290.46 |
303.48 |
290.6 |
292.45 |
0.11 |
0.26 |
WLD |
285.49 |
289.49 |
302.51 |
289.63 |
291.48 |
0.12 |
0.17 |
WGD |
286.02 |
290.02 |
303.04 |
290.16 |
292.01 |
0.14 |
0.07 |
WSD |
282.27 |
286.27 |
299.29 |
286.41 |
288.26 |
0.14 |
0.07 |
WAkD |
280.78 |
286.78 |
297.8 |
284.92 |
286.77 |
0.14 |
0.09 |
GD |
287.88 |
291.88 |
304.9 |
292.02 |
293.87 |
0.11 |
0.21 |
Table 4 Goodness of fit of the dataset-1
Distributions |
−2logL |
AIC |
BIC |
CAIC |
HQIC |
K-S |
P-value |
WAD |
63.18 |
67.18 |
80.20 |
67.88 |
67.56 |
0.09 |
0.99 |
WPD |
64.03 |
68.03 |
81.05 |
68.73 |
68.41 |
0.17 |
0.6 |
WKD |
65.33 |
69.33 |
82.35 |
70.04 |
69.72 |
0.14 |
0.74 |
WLD |
65.07 |
69.07 |
82.09 |
69.77 |
69.45 |
0.16 |
0.66 |
WGD |
65.48 |
69.48 |
82.5 |
70.18 |
69.86 |
0.18 |
0.43 |
WSD |
64.02 |
68.02 |
81.04 |
68.72 |
68.4 |
0.17 |
0.57 |
WAkD |
63.62 |
67.62 |
80.64 |
68.32 |
68.01 |
0.16 |
0.68 |
GD |
66.14 |
70.14 |
83.16 |
70.84 |
70.53 |
0.14 |
0.83 |
Table 5 Goodness of fit of the dataset-2
Datasets |
Parameters |
90% CI |
95% CI |
99% CI |
|
|
(Lower, Upper) |
(Lower, Upper) |
(Lower, Upper) |
1 |
ˆθ |
1.4806, 2.1618 |
1.4270, 2.2395 |
1.3279, 2.3978 |
ˆα |
1.6769, 3.1895 |
1.5637, 3.3670 |
1.3580, 3.7307 |
|
2 |
ˆθ |
0.9835, 1.771 |
0.9284, 1.8706 |
0.8287, 2.0775 |
|
ˆα |
0.4189, 1.3496 |
0.3676, 1.4816 |
0.2827, 1.7634 |
Table 6 Confidence interval of the parameters of WAD for the dataset-1 and 2
In this paper, a weighted Amarendra distribution (WAD) has been suggested which contains Amarendra distribution. Its statistical properties including moments based measures such as moments about origin, moments about mean, coefficient of variation, skewness, kurtosis, index of dispersion, reliability function, hazard function, reverse hazard function, mean residual life function have been discussed with graphical representation. Parameters are estimated by the method of maximum likelihood estimation and maximum product spacing estimation. A simulation study is carried out to show the consistency of the estimaor of the parameters by maximum likelihood estimation and maximum product spacing estimation. Confidence interval of the parameters has been presented with profile plot of the parameters. Finally, in application portion, goodness of fit demonstrated on two real lifetime datasets from engineering field and fitted plot of the considered distributions, P-P plot, Q-Q plot, and ECDF plot of dataset-1 and 2 are presented. Its shows that WAD provides a better fit as compared with WPD, WKD, WLD, WGD, WSD, WAkD and GD.
Authors are grateful to the editor in chief and the anonymous reviewer for some minor comments which improved both the quality and the presentation.
The authors declare that they have no conflicts of interest.
None.
©2024 Shanker, 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