Research Article Volume 13 Issue 2
Department of Statistics, Assam University, Silchar, Assam, India
Correspondence: Rama Shanker, Department of Statistics, Assam University, Silchar, India
Received: May 04, 2024 | Published: May 21, 2024
Citation: Prodhani HR, Shanker R. Weighted Pratibha distribution with properties and application in flood dataset. Biom Biostat Int J. 2024;13(2):52-57. DOI: 10.15406/bbij.2024.13.00414
In this paper, a weighted Pratibha distribution introduced to modelling life time data using the weighted transformation technique in Pratibha distribution. Its statistical properties including survival function, hazard function, reverse hazard function, mean residual life function stochastic ordering, moments related measures such as moments about origin, moments about mean, coefficient of variation, skewness, kurtosis, index of dispersion have been studied. Parameters are estimated by method of maximum likelihood estimation. A simulation study has been carried out to test the consistency of the parameters obtained through the maximum likelihood method. The goodness of fit of the distribution has been presented with a real lifetime dataset and the goodness of fit shows better fit over weighted Sujatha distribution, weighted Komal distribution, weighted Lindley distribution, weighted Garima distribution and weighted Aradhana distribution.
Keywords: Pratibha distribution, statistical properties, maximum likelihood estimation, simulation, application
Fisher1 initially developed the concept of weighted distributions to represent the ascertainment bias. Subsequently, Rao2 extended this idea cohesively while modelling statistical data in which standard distributions were not appropriate for recording these observations with equal probability. To capture the observations in such instances’, weighted models were created using a weighted function. Biased data will arise from the frequency distribution of data recorded such as at least one boy child per family, at least one girl child per family, at least one migration per family, etc.
Assume that the original observation y is based on a distribution with probability density function (pdf) gy(x,η1) , where η1 may be a vector of parameters and that the observation x is recorded based on a probability that is re-weighted by weight function w(x,η2) >0 , where η2 is a new parameter vector.
f(x;η1,η2)=D w(x;η2) gy(x;η1)
where D is a constant used to normalize. Note that these kinds of distributions are referred to as weighted distributions. The simple size-biased distributions or length-biased distributions are the weighted distributions with the weight function w(x)=x . A few broad probability models that produce weighted probability distributions were examined by Patil3 and Rao,4 along with applications and the fact that w(x)=x is a natural outcome in sampling-related situations.
In distribution theory, it is highly helpful to add a shape parameter to an existing distribution using a weighted approach. The existing distribution exhibits increased flexibility and tractability tendencies with the inclusion of a parameter. Weighted distributions are used to model heterogeneity, clustered sampling, and extraneous variance in the dataset.
Weighted versions of one parameter lifetime distributions have been derived by several researchers using the weight function w(x,ω)=xω−1 . For examples, Ghitany et al.5 proposed weighted Lindley distribution (WLD) from Lindley distribution of Lindley,6 Eyob and Shanker7 suggested weighted Garima distribution (WGD) from Garima distributions of Shanker,8 Ganaie et al.9 suggested weighted Aradhana distribution (WAD) from Aradhana distribution of Shanker,10 Shanker and Shukla11 suggested weighted Sujatha distribution (WSD) from Sujatha distribution of Shanker,12 Shanker et al.13 suggested weighted Komal distribution (WKD) from Komal distribution of Shanker,15 Shanker et al.14 suggested weighted Uma distribution (WUD) from Uma distribution of Shanker17 respectively. It has been noted that, depending on a conceptual or applied angle, these weighted distributions did not provide a suitable fit in certain datasets. Therefore, search for better weighted distribution corresponding to recent lifetime distribution is required.
Shanker16 introduced a one parameter Pratibha distribution with statistical properties and applications and observed that Pratibha distribution provides better fit than exponential distribution, Lindley distribution, Sujatha distribution, Shanker distribution by Shanker,18 Akash distribution by Shanker19 and Garima distribution. The pdf and cdf of Pratibha distribution are given by
f(x;η)=η3[η3+η+2](η+x+x2)e−ηx;x>0,η>0F(x;η)=1−[1+ηx(ηx+η+2)η3+η+2] e−η x;x>0,η>0
The primary goal is to study the weighted Pratibha distribution (WPD) and examine its properties. The WPD is being proposed because it is expected to provide a better fit than the weighted counterpart of the Lindley, Sujatha, Komal, and Garima distributions, given that the Pratibha distribution provides the highest degree of fit over these distributions.
Weighted Pratibha distribution
Let a random variable X~ WPD (η,ω) with the weight function w(x,ω)=xω−1 , the pdf and cdf of WPD can be expressed as
f(x;η,ω)=ηω+2[η3+ηω+ω(ω+1)]Γ(ω)(η+x+x2)xω−1e−ηx;x>0,η>0,ω>0F(x;η,ω)=1−η3Γ(ω,ηx)+ηΓ(ω+1,ηx)+Γ(ω+2,ηx)[η3+ηω+ω(ω+1)]Γ(ω);x>0,η>0,ω>0
where η is a scale parameter and ω is shape parameter of the distribution. When ω=1 , WPD reduced to Pratibha distribution with parameter η . The plots of pdf and cdf of WPD are shown in the following Figures 1 & 2 respectively. From the Figure 1, it is clear that when η=0.1 and for increasing values of ω , the pdf has unimodal and positively skeweed natures. When ω=0.5 and η<0.1 , it has monotonically increasing natures and for ω=0.5 and η≥1 , the pdf have bimodal and positively skewwed natures. The most important feature of WPD is that it is unimodal and bimodal for different values of parameters and in general flood dataset shows unimodal or bimodal shapes depending upon the time period of the flood and WPD would be the best choice for modeling data of flood.
Survival function
The survival function of WPD can be obtained as
S(x;η,ω)=η3Γ(ω,ηx)+ηΓ(ω+1,ηx)+Γ(ω+2,ηx)[η3+ηω+ω(ω+1)]Γ(ω);x>0,η>0,ω>0
Hazard function
The hazard function of WPD can be obtained as
h(x;η,ω)=ηω+2(η+x+x2)xω−1e−ηxη3Γ(ω,ηx)+ηΓ(ω+1,ηx)+Γ(ω+2,ηx);x>0,η>0,ω>0
The plots of hazard function of WPD are graphically shown in the following Figure 3. It shows different shapes including monotonically increasing, decreasing, upside bathtub and downside bathtub and it means that the distribution is applicable for modelling data of these natures.
Reverse hazard function
r(x;η,ω)=ηω+2(η+x+x2)xω−1e−ηx[η3+ηω+ω(ω+1)]Γ(ω)−η3Γ(ω,ηx)+ηΓ(ω+1,ηx)+Γ(ω+2,ηx);x>0,η>0,ω>0
Mean residual life function
Mean residual life function of WPD can be obtained as
m(x;η,ω)=η3Γ(ω+1,ηx)+ηΓ(ω+2,ηx)+Γ(ω+3,ηx)η[η3Γ(ω,ηx)+ηΓ(ω+1,ηx)+Γ(ω+2,ηx)]−x;x>0,η>0,ω>0 .
The plots of mean residual life function are shown in the following Figure 4. It is quite obvious that mean residual life function is monotonically decreasing.
Moments related measures
The rth raw moment (moment about origin) of WPD, after little algebraic simplification, can be obtained as
μr′=∞∫0xrf(x;η,ω) dx=Γ(ω+r)[η3+η(ω+r)+(ω+r)(ω+r+1)]ηr[η3+ηω+ω(ω+1)]Γ(ω);r=1,2,3...
Putting r=1,2,3,4 , the first four raw moments are obtained as
μ1′=ω[η3+η(ω+1)+(ω+1)(ω+2)]η[η3+ηω+ω(ω+1)]
μ2′=ω(ω+1)[η3+η(ω+2)+(ω+2)(ω+3)]η2[η3+ηω+ω(ω+1)]
μ3′=ω(ω+1)(ω+2)[η3+η(ω+3)+(ω+3)(ω+4)]η3[η3+ηω+ω(ω+1)]
μ4′=ω(ω+1)(ω+2)(ω+3)[η3+η(ω+4)+(ω+4)(ω+5)]η4[η3+ηω+ω(ω+1)]
The central moments of WPD, after simple algebraic simplification, can be obtained as
μ2=ω[η6+2η4ω+2η3ω2+2η4+8η3ω+η2ω2+2ηω3+ω4+6η3+η2ω+6ηω2+4ω3+4ηω+5ω2+2ω]η2(η3+ηω+ω2+ω)2
μ3=2ω[η9+3η7ω+3η6ω2+3η7+15η6ω+3η5ω2+6η4ω3+3η3ω4+12η6+3η5ω+21η4ω2+13η3ω3+3η2ω4+3ηω5+ω6+15η4ω+16η3ω2+9η2ω3+12ηω4+5ω5+6η3ω+6η2ω2+15ηω3+9ω4+6ηω2+7ω3+2ω2]η3(η3+ηω+ω2+ω)3
μ4=3ω[2η12+η12ω+4η10ω2+4η9ω3+12η10ω+24η9ω2+6η8ω3+12η7ω4+6η6ω5+8η10+60η9ω+22η8ω2+76η7ω3+56η6ω4+12η5ω5+12η4ω6+4η3ω7+40η9+16η8ω+160η7ω2+158η6ω3+76η5ω4+101η4ω5+44η3ω6+6η2ω7+4ηω8+ω9+96η7ω+164η6ω2+148η5ω3+272η4ω4+160η3ω5+46η2ω6+36ηω7+10ω8+56η6ω384η5ω2+287η4ω3+252η3ω4+118η2ω5+116ηω6+38ω7+104η4ω2+172η3ω338ω4+8ω3]η4(η3+ηω+ω2+ω)4
Thus, the coefficient of variation (C.V), coefficient of skewness (√β1) , coefficient of kurtosis (β2) , and index of dispersion (γ) of WPD are obtained as
CV=√ω[η6+2η4ω+2η3ω2+2η4+8η3ω+η2ω2+2ηω3+ω4+6η3+η2ω+6ηω2+4ω3+4ηω+5ω2+2ω]ω[η3+η(ω+1)+(ω+1)(ω+2)]
√β1=2ω[η9+3η7ω+3η6ω2+3η7+15η6ω+3η5ω2+6η4ω3+3η3ω4+12η6+3η5ω+21η4ω2+13η3ω3+3η2ω4+3ηω5+ω6+15η4ω+16η3ω2+9η2ω3+12ηω4+5ω5+6η3ω+6η2ω2+15ηω3+9ω4+6ηω2+7ω3+2ω2]ω[η6+2η4ω+2η3ω2+2η4+8η3ω+η2ω2+2ηω3+ω4+6η3+η2ω+6ηω2+4ω3+4ηω+5ω2+2ω]
β2=3ω[2η12+η12ω+4η10ω2+4η9ω3+12η10ω+24η9ω2+6η8ω3+12η7ω4+6η6ω5+8η10+60η9ω+22η8ω2+76η7ω3+56η6ω4+12η5ω5+12η4ω6+4η3ω7+40η9+16η8ω+160η7ω2+158η6ω3+76η5ω4+101η4ω5+44η3ω6+6η2ω7+4ηω8+ω9+96η7ω+164η6ω2+148η5ω3+272η4ω4+160η3ω5+46η2ω6+36ηω7+10ω8+56η6ω384η5ω2+287η4ω3+252η3ω4+118η2ω5+116ηω6+38ω7+104η4ω2+172η3ω338ω4+8ω3]ω2[η6+2η4ω+2η3ω2+2η4+8η3ω+η2ω2+2ηω3+ω4+6η3+η2ω+6ηω2+4ω3+4ηω+5ω2+2ω]2
γ=ω[η6+2η4ω+2η3ω2+2η4+8η3ω+η2ω2+2ηω3+ω4+6η3+η2ω+6ηω2+4ω3+4ηω+5ω2+2ω]ωη(η3+ηω+ω2+ω)[η3+η(ω+1)+(ω+1)(ω+2)]
When η≤1.3,ω≤1.3 , variance is greater than the mean. The plots of coefficient of variation, skewness, kurtosis and index of dispersion are shown in the following Figure 5.
Figure 5 Coefficient of variation, coefficient of skewness, coefficient of kurtosis, index of dispersion for differnet values of the parameters of WPD.
Figure 5 illustrates that for fixed values of ω and increasing values of η, the coefficient of variation, coefficient of skewness and coefficient of kurtosis are monotonically increaseing, wheres as for fixed values of η and increasing values of ω, coefficient of variation, coefficient of skewness and coefficient of kurtosis are monotonically decreaseing. On the other hand for fixed values of ω and increasing values of η and ficxed values η and incrasing values of ω index of dispersion is decreasing.
Maximum likelihood estimation
Let (x1,x2,...,xn) be a random sample from WPD. The log-likelihood function of WPD can be expressed as
logL=n[(ω+2)logη−log(η3+ηω+ω(ω+1))]−nlog(Γ(ω))+n∑i=1log(η+xi+xi2)+(ω−1)n∑i=1log(xi)− ηn∑i=1xi
This gives
∂logL∂η=n(ω+2)η−n(3η2+ω)η3+ηω+ω(ω+1)+n∑i=11η+xi+xi2−n∑i=1xi=0
∂logL∂ω=nlog(η)−n(η+2ω+1)η3+ηω+ω(ω+1)−nψ(ω)+n∑i=1logxi=0
where ψ(ω)=∂∂ωlog(Γ(ω)) is a digamma function.
The log-likelihood equations presented here are not readily solvable 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∂η2=−n(ω+2)η2−6nη[η3+ηω+ω(ω+1)]−n(3η2+ω)2[η3+ηω+ω(ω+1)]2−n∑i=11(η+xi+xi2)2
∂2logL∂ω2=−2n[η3+ηω+ω(ω+1)]−n(η+2ω+1)2[η3+ηω+ω(ω+1)]2−nψ′(ω)
∂2logL∂η∂ω=−nη−6nη[η3+ηω+ω(ω+1)]−n(η+2ω+1)(3η2+ω)[η3+ηω+ω(ω+1)]2=∂2logL∂ω∂η
For finding the MLEs (ˆη,ˆω) of parameters (η,ω) of WPD, following equations can be solved
(∂2logL∂η2∂2logL∂η∂ω∂2logL∂ω∂η∂2logL∂ω2)ˆη=η0ˆω=ω0(ˆη−η0ˆω−ω0)=(∂logL∂η∂logL∂ω)
where η0 and ω0 are the initial values of η and ω . These equations are solved iteratively till close estimates of parameters are obtained.
A simulation study
To assess the effectiveness of maximum likelihood estimators for WPD, a simulation study has been conducted. The investigation involved examining mean estimates, biases (B), mean square errors (MSEs), and variances of the maximum likelihood estimates (MLEs) for WPD, 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=(η,ω) and ˆHi=(ˆηi,ˆωi) .
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 WPD involves the following steps:
The biases, MSEs, and variances of the MLEs of the parameters decrease for increasing sample size as evident in Tables 1 & 2. This supports the first-order asymptotic theory of MLEs.
Parameter |
Sample size |
Mean |
Bias |
MSE |
Variance |
|
25 |
0.09587 |
-0.00412 |
0.00001 |
0.00000 |
50 |
0.09600 |
-0.00399 |
0.00001 |
0.00000 |
|
100 |
0.09629 |
-0.00370 |
0.00001 |
0.00000 |
|
200 |
0.09745 |
-0.00254 |
0.00001 |
0.00000 |
|
300 |
0.097994 |
-0.00200 |
0.00000 |
0.00000 |
|
|
25 |
1.48705 |
-0.01294 |
0.00049 |
0.00033 |
50 |
1.49011 |
-0.00988 |
0.00038 |
0.00028 |
|
100 |
1.49332 |
-0.00667 |
0.00028 |
0.00023 |
|
200 |
1.49509 |
-0.00490 |
0.00021 |
0.00019 |
|
300 |
1.49674 |
-0.00325 |
0.00016 |
0.00015 |
Table 1 Descriptive constants of WPD for η=0.1,ω=1.5
Parameter |
Sample size |
Mean |
Bias |
MSE |
Variance |
|
25 |
0.23411 |
0.03411 |
0.00117 |
0.00001 |
50 |
0.23359 |
0.03359 |
0.00113 |
0.00001 |
|
100 |
0.23324 |
0.03324 |
0.00111 |
0.00001 |
|
200 |
0.23299 |
0.03299 |
0.00109 |
0.00001 |
|
300 |
0.23205 |
0.03205 |
0.00103 |
0.00001 |
|
|
25 |
0.30356 |
0.00356 |
0.00026 |
0.00025 |
50 |
0.30297 |
0.00297 |
0.00019 |
0.00018 |
|
100 |
0.30267 |
0.00267 |
0.00014 |
0.00014 |
|
200 |
0.30211 |
0.00211 |
0.00012 |
0.00011 |
|
300 |
0.30130 |
0.00130 |
0.00010 |
0.00010 |
Table 2 Descriptive constants of WPD for η=0.1,ω=1.5
Variance-Covariance matrix for the prameters η=0.1,ω=1.5 and η=0.2,ω=0.3 respectively as
η ωηω(0.0000030.0000050.0000050.000193) and η ωηω(0.0000090.0000010.0000010.000100)
To test the goodness of fit of WPD , we have considered a real lifetime dataset from flood discharge. The following right-skewed dataset discussed by Montfort,20 presents the maximum annual flood discharges of the North Saskatchewan in units of 1000 cubic feet per second of the north Saskatchewan river at Edmonton over a period of 47 years.
19.885, 20.940, 21.820, 23.700, 24.888, 25.460, 25.760, 26.720, 27.500, 28.100, 28.600,
30.200, 30.380, 31.500, 32.600, 32.680, 34.400, 35.347, 35.700, 38.100, 39.020, 39.200,
40.000, 40.400, 40.400, 42.250, 44.020, 44.730, 44.900, 46.300, 50.330, 51.442, 57.220,
58.700, 58.800, 61.200, 61.740, 65.440, 65.597, 66.000, 74.100, 75.800, 84.100, 106.600,
109.700, 121.970, 121.970, 185.560.
The summary of the dataset and its total time in test (TTT) plots are shown in the folowing Table 3 and the Figure 6. The goodness of fit of the WPD along with other weighted and unweighted distributions are shown in the Table 4.
Minimum |
1st Quartile |
Median |
Mean |
3rd Quartile |
Maximum |
19.89 |
30.34 |
40.40 |
51.50 |
61.34 |
185.56 |
Table 3 Goodness of fit of the datasetDescriptive constants of WPD for
Distributions |
MLE |
|
|
AIC |
K-S |
P-value |
|
θ^ |
α^ |
-2log L |
|
|
|
WPD |
0.0717 (0.0148) |
1.7187 (0.7103) |
443.17 |
447.17 |
0.12 |
0.47 |
WSD |
0.0776 (0.0186) |
2.0226 (0.9041) |
443.34 |
447.34 |
0.15 |
0.24 |
WKD |
0.0717 (0.0148) |
0.2055 (0.0521) |
443.18 |
447.18 |
0.27 |
0.00 |
WLD |
0.0717 (0.0148) |
2.7180 (0.7104) |
443.17 |
447.17 |
0.16 |
0.16 |
WGD |
0.0757 (0.0145) |
3.1524 (0.6626) |
443.78 |
447.78 |
0.14 |
0.28 |
WAD |
0.0724 (0.0147) |
0.7822 (0.7061) |
443.31 |
447.31 |
0.26 |
0.00 |
Table 4 Goodness of fit of the dataset
The Table-4 shows that WPD have the least, AIC and K-S values as compared to the WSD, WKD, WLD, WGD, and WAD. So, we conclude that WPD provides a better fit as compared to WSD, WKD, WLD, WGD, and WAD. From the fitted plot and the P-P plot of the considered distribution presented in the Figure 7 & 8 for the dataset also exhibit that WPD provides a better fit as compared to the considered distributions.
In this study a weighted version of Pratibha distribution known as weighted Pratibha distribution (WPD) has been proposed and discussed. Its significant statistical properties such as moments and its related measures, survival function, hazard function, reverse hazard function, and mean residual life function are studied. Parameters of the proposed distribution are estimated using the maximum likelihood estimation. A simulation study is carried out to know the performance of the estimated parameters values. Finally, an example of lifetime dataset relating to flood is carried out for the application and goodness of fit of the proposed distribution and it has been shown that it provides better fit over weighted distributions such as WSD, WKD, WLD, WGD, and WAD. Therefore, the WPD can be considered an important weighted distribution for modelling real dataset relating to flood.
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 there are no conflicts of interest.
None.
©2024 Prodhani, 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