1Departmentof Statistics, Ahmadu Bello University, Zaria, Kaduna State, Nigeria 2Department of Mathematics, Usmanu Danfodiyo University, Sokoto, Sokoto State, Nigeria
Received: December 16, 2019 Accepted: February 19, 2020
Abstract: In this research work, a continuous probability distribution as an improvement of Topp-Leone distribution was developed. Thus, density function and the cumulative function of the constructed distribution were obtained. We also show the validity of the new distribution. However, some properties of constructed distribution such as, survival and hazard functions, moment, and probability weighted moment and order statistics were derived. Parameters were obtained using method of maximum likelihood estimation. Keywords: Beta-G family, Topp-Leone distribution, Beta-Topp Leone, density function, moments
Introduction distribution would receive wider attraction in reliability and
Probability distribution theory, preference for a particular mechanical engineering. probability distribution in modeling real life phenomena could Jafari et al. (2014) gave the densities and properties of the be based on either the distribution is tractable or the Beta Gompertz distribution. Distributions like the Exponential distribution is flexible (Oguntunde et al., 2016). distribution, Generalised Exponential distribution, The tractability of a probability distribution may be useful in Generalised Gompertz distribution, Gompertz distribution and theory because such distribution would be easy to work with; the Beta Exponential distribution were discovered to be especially when it comes to simulation of random samples, special cases of the Beta Gompertz distribution. A simulation but to practitioners and some other stakeholders, the flexibility study was conducted to investigate the generalization of the of probability distributions could be of interest. In fact, it is proposed distribution. An application to a data set on lifetime preferable to make use of probability distributions that best fit of 50 devices revealed that the Beta Gompertz distribution has the available data set than to transform the existing data set as a better fit than all its sub-models. this may affect the originality of the data set. Because of this, Nekoukhou (2016) studied a discrete analog of the beta- several efforts have been made in recent years to ensure that Rayleigh distribution. The distribution contains the the existing standard theoretical distributions are modified and generalized discrete Rayleigh and discrete Rayleigh extended (Merovci, 2013); this could increase their flexibility distributions as special sub-models. The author discussed and enhance the capability to model real life data sets. some distributional and moment properties of the new discrete To extend an existing standard distribution, there are various distribution as well as its order statistics. So, the hazard rate approaches that could be adopted. For instance, the flexibility function of his model can be increasing, bathtub-shaped and of a distribution can be increased by means of generalization upside-down bathtub. Estimation of the parameters was which involves using the available generalized family of illustrated. distributions. Usman et al. (2019) constructed a two parameters continuous When a distribution is generalized, extra shape parameter(s) distribution using Topp-Leone distribution as a baseline. They from the family of distributions used would have been added. derived density function and the cumulative function of the The role of these additional shape parameter(s) is to vary the constructed distribution (Burr X-Topp Leone distribution). tail weight of the resulting compound distribution, thereby Also, they show the validity of the new distribution. inducing it with skewness. The flexibility can also be increased by modifying the existing distribution. For instance, Methodology two or more standard distributions can be combined as in the Following the family proposed by Eugene et al. (2002). case of convolution, quotient, or product of independent Supposed 𝐹(𝑥) and 𝑓(𝑥) are the cumulative density function random variables. Also, some distributions are found to be (cdf) and probability density function (pdf) of any baseline functions of one or more other distributions; the composition distribution, then, the cdf and the pdf of the beta generalized of the student t-distribution is a common example (Sun, family is given by equation 1 and 3. 2011). 𝐺(𝑥) = 𝐼𝐺(𝑥) (𝑎, 𝑏) 1 Akinsete et al. (2008) defined beta-Pareto distribution. The Where: 𝐼𝐺(𝑥) (𝑎, 𝑏) is regarded as the incomplete beta authors discussed various properties of the distribution. Their function ratio distribution was found to be unimodal and has either a unimodal or a decreasing hazard rate. The expressions for the Mathematically, mean, mean deviation, variance, skewness, kurtosis and 1 𝐹(𝑥) 𝑎−1 𝐼𝐺(𝑥) (𝑎, 𝑏) = ∫ 𝑥 (1 − 𝑥)𝑏−1 𝑑𝑥 2 entropies were obtained. The relationship between these 𝐵(𝑎,𝑏) 0 moments and the parameters were provided. The method of (Sangsanit & Bodhisuwan, 2016) maximum likelihood was also proposed to estimate the And 1 parameters of the distribution. The distribution was applied to 𝑔(𝑥) = [𝐹(𝑥)]𝑎−1 [1 − 𝐹(𝑥)]𝑏−1 𝑓(𝑥) 3 𝐵(𝑎,𝑏) two flood data sets. Where: 𝑓(𝑥) and 𝐹(𝑥) denote the pdf and cdf of baseline Hanook et al. (2013) derived densities and various statistical model properties of the Beta Inverse Weibull distribution. The relationships that exist between its parameters, mean, In this paper, Topp Leone Distribution is considered as a skewness, variance and kurtosis were also investigated. The baseline distribution. If a random variable 𝑇 is distributed as shape of the distribution was unimodal and the model parameters were successfully estimated. the Topp-Leone and bounded on [0,1] (Sangsanit & Though, an application to real life data sets was not provided Bodhisuwan, 2016). Let 𝑋 be a continuous random variable but the authors claimed that the Beta Inverse Weibull with pdf 𝑓(𝑥). TL distribution has pdf written by;
FUW Trends in Science & Technology Journal, www.ftstjournal.com
e-ISSN: 24085162; p-ISSN: 20485170; April, 2020: Vol. 5 No. 1 pp. 306 – 309 306 Continuous Probability Distribution
𝑓(𝑥) = 𝛽(2 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽−1 4 Reliability analysis of the BTLD
Where 𝛽 > 0 is a shape parameter. The associated cdf is; The task of survival is the possibility that the system or person 𝐹(𝑥) = (2𝑥 − 𝑥 2 )𝛽 5 does not fail after a certain time. The task of survival is given: 𝑆(𝑋) = 𝑃(𝑋 > 𝑥) = 1 − 𝐺(𝑥) 17 Proposed beta-topp leone distribution (BTLD) Applying the BTLD in (3.7), the survival function for the Using the idea of Eugene et al., (2002), we propose a new BTLD is obtained as: 𝐵(𝑎,𝑏)−𝛽(2𝑥−𝑥 2 )(𝑎,𝑏) continuous probability distribution called the Beta-Topp 𝑆(𝑥) = 18 𝐵(𝑎,𝑏) Leone Distribution (BTLD).Thus, to obtain the pdf and the Hazard function is the probability that a component will fail cdf of the new model, we substitute our baseline distribution or die for an interval of time. The hazard function is defined in equation 3 and 5. So that: 1 as; 𝑔(𝑥)𝐵𝑇𝐿𝐷 = [𝛽(2 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽𝑎−1 ][1 − 𝑓(𝑥) 𝑔(𝑥) 𝐵(𝑎,𝑏) ℎ(𝑥) = = 19 𝑆(𝑥) 1−𝐺(𝑥) (2𝑥 − 𝑥 2 )𝛽 ]𝑏−1 6 𝛽(2−2𝑥)(2𝑥−𝑥 2 )𝛽𝑎−1 [1−(2𝑥−𝑥 2 )𝛽 ]𝑏−1 is the pdf with 3 parameters. ℎ(𝑥) = 20 𝐵(𝑎,𝑏)−𝛽𝐺(𝑥)(𝑎,𝑏) And 1 (2𝑥−𝑥 2 )𝛽 𝑎−1 𝐺(𝑥)𝐵𝑇𝐿𝐷 = ∫ 𝑥 (1 − 𝑥)𝑏−1 𝑑𝑥 = 𝐵(𝑎,𝑏) 0 𝛽(2𝑥−𝑥 2 )(𝑎,𝑏) Linear representation is the cdf 7 In this section, we derive expansions for the cdf and pdf of the 𝐵(𝑎,𝑏) For 0 ≤ 𝑥 ≤ 1, 𝛽 > 0, where 𝑎 > 0 𝑎𝑛𝑑 𝑏 > 0 are the shape BTLD that are useful to study its statistical properties using an parameters. idea by Shittu and Adepoju (2013). Model validity check of the proposed BTLD Applying 4 and 5 into 3, we have: 𝛽(2−2𝑥)(2𝑥−𝑥 2 )𝛽−1 𝑎−1 A probability distribution is a function that describes the 𝑔(𝑥)𝐵𝑇𝐿𝐷 = [(2𝑥 − 𝑥 2 )𝛽 ] [1 − 𝐵(𝑎,𝑏) likelihood of obtaining the possible values that a random (2𝑥 − 𝑥 ] 2 )𝛽 𝑏−1 variable can assume. In other words, the values of the variable 21 were based on the underlying probability distribution (Jim, Consider the series expansion 2019). As such, a function can serve as a probability function 𝑏−1 𝑏 − 1 (−1)𝑖 (2𝑥 [1 − (2𝑥 − 𝑥 2 )𝛽 ] = ∑∞ 𝑖=0 ( ) − 𝑥 2 )𝛽𝑖 22 of a continuous random variable 𝑋 if its values, 𝑓(𝑥) satisfies 𝑖 Where, the binomial coefficient is defined for any real the conditions: number. 𝑓(𝑥) ≥ 0, for −∞ < 𝑥 < ∞ ∞ Thus, 21 becomes ∫−∞ 𝑔(𝑥)𝑑𝑥 = 1 8 𝛽(2−2𝑥)(2𝑥−𝑥 2 )𝛽−1 𝑎−1 = [(2𝑥 − 𝑥 2 )𝛽 ] ∑∞ 𝑖 2 𝛽𝑖 𝑖=0(−1) (2𝑥 − 𝑥 ) 23 It follows from the above that if 𝑋 is a continuous random 𝐵(𝑎,𝑏) 𝑎−1 variable, then the probability that 𝑋 takes on any one We can expand [(2𝑥 − 𝑥 2 )𝛽 ] as follows: particular value is zero, whereas the interval probability that 𝑋 𝑎−1 𝑎 − 1 (−1)𝑚 [1 − [1 − (2𝑥 − 𝑥 2 )𝛽 ]] = ∑∞ 𝑚=0 ( ) [1 − lies between two different values, say, 𝑎 and 𝑏, is given by 𝑚 𝑏 𝑚 (2𝑥 − 𝑥 2 )𝛽 ] 24 𝑃(𝑎 < 𝑋 < 𝑏) = ∫ 𝑔(𝑥)𝑑𝑥 = 1 𝑎 Thus, with the limit of the BTLD (0,1), we have; Also, 1 𝑚 𝑚 𝑃(0 < 𝑋 < 1) = ∫0 𝑔(𝑥)𝐵𝑇𝐿𝐷 𝑑𝑥 = 1 9 [1 − (2𝑥 − 𝑥 2 )𝛽 ] = ∑∞ 𝑛 2 𝛽𝑛 𝑛=0 ( 𝑛 ) (−1) (2𝑥 − 𝑥 ) 25 From 6 and 9, we have: 1 1 Applying 24 and 25 in 21, we have: ∫0 [𝛽(2 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽𝑎−1 ][1 − (2𝑥 − ∞ 𝐵(𝑎,𝑏) 2 𝛽 𝑏−1 𝛽(2 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽−1 𝑏−1 𝑎−1 𝑚 𝑥 ) ] 𝑑𝑥 = 1 10 = ∑ ( )( )( ) 𝛽 1 𝐵(𝑎, 𝑏) 𝑖 𝑚 𝑛 ∫ (2 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽𝑎−1 [1 − (2𝑥 − 𝑥 2 )𝛽 ]𝑏−1 𝑑𝑥 = 1 𝑖,𝑚,𝑛=0 𝐵(𝑎,𝑏) 0 (−1)𝑖+𝑚+𝑛 (2𝑥 − 𝑥 2 )𝛽𝑖 (2𝑥 − 𝑥 2 )𝛽𝑛 26 11 ∞ 𝑑𝑦 𝑑𝑦 𝛽(2 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽−1 (𝑘 + 1) 𝑏−1 𝑎−1 𝑚 Let 𝑦 = 2𝑥 − 𝑥 2 , = 2 − 2𝑥 => 𝑑𝑥 = = ∑ ( )( )( ) 𝑑𝑥 (2−2𝑥) 𝐵(𝑎, 𝑏)(𝑘 + 1) 𝑖 𝑚 𝑛 Using the above terms in 11, we obtain: 𝑖,𝑚,𝑛=0 𝛽 1 (−1)𝑖+𝑚+𝑛 (2𝑥 − 𝑥 2 )𝛽(𝑖+𝑛) 27 ∫0 𝑦 𝛽𝑎−1 (1 − 𝑦 𝛽 )𝑏−1 𝑑𝑦 = 1 𝐵(𝑎,𝑏) 12 𝑑𝑧 Let 𝑧 = 𝑦 𝛽 , 𝑑𝑧 = (𝛽𝑦 𝛽−1 )𝑑𝑦, 𝑑𝑦 = Let 𝛽𝑦 𝛽−1 𝑏−1 𝑎−1 𝑚 ∑∞ 𝑖,𝑚,𝑛=0( 𝑖 )( 𝑚 )( 𝑛 )(−1) 𝑖+𝑚+𝑛 From the above terms, the equation follows: = ∅𝑖,𝑚,𝑛 (𝑎, 𝑏) 28 𝛽𝑎−1 𝐵(𝑎,𝑏)(𝑘+1) 1 1 𝑑𝑧 ∫ 𝑧 𝛽 (1 − 𝑧)𝑏−1 =1 13 𝐵(𝑎,𝑏) 0 𝑦 𝛽−1 1 1 𝑎− 1 −1+ 1 And ∫ 𝑧 𝛽 𝑧 𝛽 (1 − 𝑧)𝑏−1 𝑑𝑧 =1 14 𝛽(2 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽−1 (𝑘 + 1)(2𝑥 − 𝑥 2 )𝛽(𝑖+𝑛) = 𝜋𝑘 (𝑥) 29 𝐵(𝑎,𝑏) 0 1 1 Thus, ∫ 𝑧 𝑎−1 (1 − 𝑧)𝑏−1 𝑑𝑧 = 1 𝐵(𝑎,𝑏) 0 15 𝑔(𝑥) = ∅𝑖,𝑚,𝑛 (𝑎, 𝑏)𝜋𝑘 (𝑥) 30 Since 𝐵(𝑎, 𝑏) is also know as the normalizing constant Moment because it makes the integral equal to 1 Let 𝑋 denote a continuous random variable, the Sth moment of Thus, 1 𝑋 is given by: ∫0 𝑍 𝑎−1 (1 − 𝑍)𝑏−1 𝑑𝑧 = 𝐵(𝑎, 𝑏) 16 ∞ 𝐸(𝑥 𝑠 ) = ∫−∞ 𝑥 𝑠 𝑔(𝑥)𝑑𝑥 31 Then ∞ 1 = ∫−∞ ∅𝑖,𝑚,𝑛 (𝑎, 𝑏)𝜋𝑘 (𝑥)𝑑𝑥 32 × 𝐵(𝑎, 𝑏) = 1, hence the proof. 1 𝐵(𝑎,𝑏) 1 = (𝑘 + 1)∅𝑖,𝑚,𝑛 (𝑎, 𝑏) ∫0 𝑥 𝑠 𝛽(2𝑥 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽−1 (2𝑥 − However, since ∫0 𝑔(𝑥)𝐵𝑇𝐿𝐷 𝑑𝑥 = 1, it means the proposed 𝑥 2 )𝛽(𝑖+𝑛) 𝑑𝑥 33 BTLD is valid. Let 𝑈 = (𝑘 + 1)∅𝑖,𝑚,𝑛 (𝑎, 𝑏) FUW Trends in Science & Technology Journal, www.ftstjournal.com e-ISSN: 24085162; p-ISSN: 20485170; April, 2020: Vol. 5 No. 1 pp. 306 – 309 307 Continuous Probability Distribution
Therefore, we have Where
1 𝑈 ∫0 𝑥 𝑠 𝛽(2 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽[[𝑖+𝑛]+1]−1 𝑑𝑥 34 (2𝑥 − 𝑥 2 )𝛽(𝑗+1)−1 = Let consider ∑∞ 𝑘 𝛽(𝑗 + 1) − 1 (2𝑥)[𝛽(𝑗+1)−1]−𝑘 2𝑘 𝑘=0(−1) ( ) 𝑥 55 𝛽[[𝑖+𝑛]+1]−1 𝑘 [1 − [1 − (2𝑥 − 𝑥 2 )]] = Then, put 55 in 54 ∑∞ 𝛽[[𝑖 + 𝑛] + 1] − 1 𝑞 (−1)𝑞+𝑣 (2𝑥 ∞ 𝑞,𝑣=0 ( ) ( ) − 𝑥 2 )𝑣 35 𝛽(𝑖 + 𝑛) 𝑖 𝑞 𝑣 2[𝛽(𝑗+1)−1]−𝑘 𝛽 ∑ (−1)𝑗+𝑗+𝑘 ( )( ) Substitute 35 in to 34, it gives: 𝑖 𝑗 𝑖,𝑗,𝑘=0 𝛽[[𝑖 + 𝑛] + 1] − 1 𝑞 (−1)𝑞+𝑣 1 𝑠 𝛽(𝑗 + 1) − 1 1 𝑠+[𝛽(𝑗+1)−1]+𝑘 (2 = 𝑈 ∑∞𝑞,𝑣=0 ( )( ) ∫0 𝑥 𝛽(2 − ( ) ∫0 𝑥 − 2𝑥)𝑑𝑥 56 𝑞 𝑣 𝑘 2 𝑣 2𝑥)(2𝑥 − 𝑥 ) 𝑑𝑥 36 𝛽[[𝑖 + 𝑛] + 1] − 1 𝑞 (−1)𝑞+𝑣 1 𝑠+𝑣 (1 Order statistics = 2𝛽𝑈 ∑∞𝑞,𝑣=0 ( )( ) ∫0 𝑥 − 𝑞 𝑣 Suppose 𝑋1 , 𝑋2 , … , 𝑋𝑛 is a random sample from any BTLD. 𝑣 𝑥)(2 − 𝑥) 𝑑𝑥 37 Let 𝑋𝑟 : 𝑛 denote the order statistics. The pdf of 𝑋𝑟 : 𝑛 can be 𝛽[[𝑖 + 𝑛] + 1] − 1 𝑞 (−1)𝑞+𝑣 expressed as: Let 𝑄 = 2𝛽𝑈 ∑∞ 𝑞,𝑣=0 ( )( ) 38 𝑞 𝑣 𝑛! 𝑎𝑛𝑑 𝑦 = 1 − 𝑥, 𝑥 = 1 − 𝑦, 𝑑𝑥 = −𝑑𝑦 𝑓𝑟 : 𝑛(𝑋, 𝑎, 𝑏, 𝛽) = 𝑓(𝑋)𝐵𝑇𝐿𝐷 [𝐹(𝑋)𝐵𝑇𝐿𝐷 ]𝑖−1 [1 − (𝑖−1)!(𝑛−𝑖)! Substituting above terms in to equation 43 gives 𝐹(𝑋)𝐵𝑇𝐿𝐷 ]𝑛−1 57 0 = 𝑄 ∫1 (1 − 𝑦)𝑠+𝑣 𝑦(1 + 𝑦)𝑣 − 𝑑𝑦 39 Where, Substitute 6 and 7 in 57 𝑏−1 𝑤 𝑣 𝑛!𝛽(2−2𝑥)(2𝑥−𝑥 2 )𝛽𝑎−1 [1−(2𝑥−𝑥 2 )𝛽 ] 𝛽(2𝑥−𝑥 2 )(𝑎,𝑏) 𝑖−1 (1 + 𝑦)𝑣 = ∑∞ 𝑤 (−1) (𝑤 ) 𝑦 𝑤 40 = [ ] [1 − (𝑖−1)!(𝑛−𝑖)!𝐵(𝑎,𝑏) 𝐵(𝑎,𝑏) Using 40 gives: 𝑛−𝑖 𝑤 𝑣 1 𝛽(2𝑥−𝑥 2 )(𝑎,𝑏) = 𝑄 ∑∞𝑤 (−1) (𝑤 ) ∫0 (1 − 𝑦) 𝑠+𝑣 𝑤+1 𝑦 𝑑𝑦 41 [ ]] 58 𝐵(𝑎,𝑏) Thus, Consider 𝑤 𝑣 1 (𝑠+𝑣+1)−1 (𝑤+2)−1 = 𝑄 ∑∞𝑤 (−1) (𝑤 ) ∫0 (1 − 𝑦) 𝑦 𝑑𝑦 42 𝛽(2𝑥−𝑥 2 )(𝑎,𝑏) 𝑛−𝑖 𝑛 − 𝑖 𝛽(2𝑥−𝑥2 )(𝑎,𝑏) 𝑖 Γ[(𝑠+𝑣+1)−1]Γ(𝑤+2)−1 [1 − [ ]] = ∑∞ 𝑖 𝑖 (−1) ( )[ ] 59 =𝑄 43 𝐵(𝑎,𝑏) 𝑖 𝐵(𝑎,𝑏) Γ[[(𝑠+𝑣+1)−1]+[(𝑤+2)−1]] and [1 − (2𝑥 − 𝑥 2 )𝛽 ] 𝑏−1 = ∑∞ 𝑗 𝑏−1 2 𝛽𝑗 60 Mean 𝑗=0(−1) ( 𝑗 ) (2𝑥 − 𝑥 ) The mean of the BTLD can be obtained from Sth moment of the distribution when s=1 as follows: Then 58 becomes Γ[(𝑣+2)−1]Γ(𝑤+2)−1 𝜇1 = 𝐸(𝑋) = 𝑄 44 𝑛! ∑∞ 𝑖+𝑗 (𝑛−1)(𝑏−1) Γ[[(𝑣+2)−1]+[(𝑤+2)−1]] 𝑖,𝑗=0(−1) 𝑖 𝑗 = (𝑖−1)!(𝑛−𝑖)!𝐵(𝑎,𝑏) 𝛽(2 − 2𝑥)(2𝑥 − Also the second moment of BTLD is obtained from the sth 2 )(𝑎,𝑏) 2𝑖−1 moment of the distribution when s=2: 𝑥 2 )𝛽(𝑎+𝑗)−1 [ 𝛽(2𝑥−𝑥 ] 61 Γ[(𝑣+2)−1]Γ(𝑤+2)−1 𝐵(𝑎,𝑏) 𝜇2 = 𝐸(𝑋 2 ) = 𝑄 45 Γ[[(𝑣+2)−1]+[(𝑤+2)−1]] Variance If 𝑉𝑎𝑟(𝑋) = 𝐸(𝑋 2 ) − [𝐸(𝑋)]2 46 (2𝑥 − 𝑥 2 )𝛽(𝑎+𝑗)−1 2 ∞ Γ[(𝑣+2)−1]Γ(𝑤+2)−1 Γ[(𝑣+2)−1]Γ(𝑤+2)−1 =𝑄 − [𝑄 ] 47 𝛽(𝑎 + 𝑗) − 1 Γ[[(𝑣+2)−1]+[(𝑤+2)−1]] Γ[[(𝑣+2)−1]+[(𝑤+2)−1]] = ∑(−1)𝑘 ( ) (2𝑥)[𝛽(𝑎+𝑗)−1]−𝑘 𝑥 2𝑘 𝑘 Probability weighted moment (PWM) 𝑘=0 The PWM of X following the Beta-G distribution is formally defined by: Then ∞ 2[𝛽(𝑎+𝑗)−1]−𝑘 𝛽𝑛! ∑∞ 𝑖+𝑗+𝑘 (𝑛−1)(𝑏−1)(𝛽(𝑎+𝑗)−1) 𝑔(𝑥)𝐵𝑇𝐿𝐷 = ∫−∞ 𝑥 𝑠 [𝐺(𝑥)]𝑟 𝑔(𝑥)𝑑𝑥 48 𝑖,𝑗,𝑘=0(−1) 𝑖 𝑗 𝑘 = (𝑖−1)!(𝑛−𝑖)!𝐵(𝑎,𝑏) (2 − Applying 29 in to 48 ∞ 2 2𝑖−1 = ∫−∞ 𝑥 𝑠 [𝐺(𝑥)]𝛽(𝑖+𝑛) 𝑔(𝑥)𝑑𝑥 49 2𝑥)𝑥 [𝛽(𝑎+𝑗)−1]+𝑘 𝛽(2𝑥−𝑥 )(𝑎,𝑏) [ ] 62 1 𝛽(𝑖+𝑛) 𝐵(𝑎,𝑏) = ∫0 𝑥 𝑠 𝛽(2𝑥 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽−1 [(2𝑥 − 𝑥 2 )𝛽 ] 𝑑𝑥 50 Consider 2[𝛽(𝑎+𝑗)−1]−𝑘 𝛽𝑛! ∑∞ 𝑖+𝑗+𝑘 (𝑛−1)(𝑏−1)(𝛽(𝑎+𝑗)−1) 𝑖,𝑗,𝑘=0(−1) 𝑖 𝑗 𝛽(𝑖+𝑛) 𝑘 𝛽(𝑖 + 𝑛) Let =𝑄 [1 − [1 − (2𝑥 − 𝑥 2 )𝛽 ]] = ∑∞ 𝑖=0(−1) 𝑖 ( ) [1 − (𝑖−1)!(𝑛−𝑖)!𝐵(𝑎,𝑏) 𝑖 Then 𝑖 (2𝑥 − 𝑥 2 )𝛽 ] 51 𝛽(2𝑥−𝑥 2 )(𝑎,𝑏) 2𝑖−1 Also, 𝑄(2 − 2𝑥)𝑥 [𝛽(𝑎+𝑗)−1]+𝑘 [ ] 63 𝐵(𝑎,𝑏) 𝑖 𝑗 𝑖 [1 − (2𝑥 − 𝑥 2 )𝛽 ] = ∑∞ 2 𝛽𝑗 𝑗=0(−1) (𝑗 ) (2𝑥 − 𝑥 ) Parameter estimation 52 The model parameters of the BTLD can be estimated by Using 51 and 52 into 50 leads to: maximum likelihood. Let 𝑋 = (𝑥1 , 𝑥2 , … , 𝑥𝑛 )′ be a random ∞ 1 sample of size n from BTL with parameter vector 𝜃 = =∫ 𝑥𝑠 𝛽(2 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽−1 ∑ (−1)𝑗+𝑗 (𝑎, 𝑏, 𝛽)′ . 0 𝑖,𝑗=0 Then, the likelihood function can be expressed as: 𝛽(𝑖 + 𝑛) 𝑖 (2𝑥 𝐿(𝜃) = ∏𝑛𝑖=1 𝑓(𝑥)𝐵𝑇𝐿𝐷 64 ( )( ) − 𝑥 2 )𝛽𝑗 𝑑𝑥 53 𝑖 𝑗 ∞ 1 𝛽(𝑖 + 𝑛) 𝑖 Using equation 9 in 75, that is the pdf of BTLD, we have: = ∑ (−1)𝑗+𝑗 ( ) ( ) ∫ 𝑥 𝑠 𝛽(2 − 2𝑥) 𝑖 𝑗 0 = [𝛽(𝑎, 𝑏)]−𝑛 [𝛽𝑛 ∑𝑛𝑖=1(2 − 2𝑥)(2𝑥 − 𝑥 2 )𝛽𝑎−1 ] ∑𝑛𝑖=1[1 − 𝑖,𝑗=0 𝑏−1 (2𝑥 − 𝑥 2 )𝛽(𝑗+1)−1 54 (2𝑥 − 𝑥 2 )𝛽 ] 65
FUW Trends in Science & Technology Journal, www.ftstjournal.com
e-ISSN: 24085162; p-ISSN: 20485170; April, 2020: Vol. 5 No. 1 pp. 306 – 309 308 Continuous Probability Distribution
The log-likelihood function is
𝑙(𝜃) = 𝑛𝑙𝑜𝑔𝛽 + ∑𝑛𝑖=1 𝑙𝑜𝑔(2 − 2𝑥) + (𝛽𝑎 − 1) ∑𝑛𝑖=1 log(2𝑥 − 𝑥 2 ) + References (𝑏 − 1) ∑𝑛𝑖=1 𝑙𝑜𝑔 [1 − (2𝑥 − 𝑥 2 )𝛽 ] − 𝑛𝑙𝑜𝑔[𝛽(𝑎, 𝑏)] 66 Akinsete A, Famoye F & Lee C 2008. The beta-Pareto By taking the partial differentiation of 𝑙(𝜃) with respect to distribution: Statistics. A J. Theoretical and Appl. 𝑎, 𝑏, 𝛽 respectively, the components of the unit score vector Stat., 42(6). ′ 𝜇𝜃 = (𝜇𝑎 , 𝜇𝑏 𝑎𝑛𝑑 𝜇𝛽 ) can be obtained as follows: Eugene N, Lee C & Famoye F 2002. Beta-normal distribution 𝑑𝑙 −𝑛 and its application. Theory Commun. in Stat. and 𝜇𝑎 = = ∑𝑛𝑖=0 log(2𝑥 − 𝑥 2 )𝛽 𝑑𝛽(𝑎, 𝑏) 67 Methods, 31(4) 497-501. 𝑑𝑎 𝛽(𝑎,𝑏) 𝑛 𝑛𝑑𝛽(𝑎, 𝑏) Hanook S, Shahbaz M, Moshin M & Kibra BM 2013. A note = 𝛽 ∑ 𝑙𝑜𝑔 (2𝑥 − 𝑥 2 ) − on beta inverse-Weibull distribution. Communi. in Stat.- 𝛽(𝑎, 𝑏) 𝑖=1 Theory and Methods, 42(2): 320-335. = 𝛽 ∑𝑛𝑖=1 𝑙𝑜𝑔 (2𝑥 − 𝑥 2 ) − 𝑛𝜑(𝑎) 68 Jafari A, Tahmasebi S & Alizadeh M 2014. The beta- 𝑑𝑙 𝑛 𝜇𝑏 = = ∑𝑛𝑖=1 𝑙𝑜𝑔[1 − (2𝑥 − 𝑥 2 )𝛽 ] − 𝑑𝛽(𝑎, 𝑏) 69 Gompertz distribution. Revista Colombiana de 𝑑𝑏 𝛽(𝑎,𝑏) 𝑛 Estadistica, 37(1): 139-156. 𝑛𝑑𝛽(𝑎, 𝑏) Jim F 2019. Making Statistics Intuitive. = ∑ 𝑙𝑜𝑔[1 − (2𝑥 − 𝑥 2 )𝛽 ] − 𝛽(𝑎, 𝑏) https://statisticsbyjim.com. Retrieved on 19/08/2019. 𝑖=1 Merovci F 2013. Transmuted exponentiated exponential = ∑𝑛𝑖=1 𝑙𝑜𝑔[1 − (2𝑥 − 𝑥 2 )𝛽 ] − 𝑛𝜑(𝑏) 70 𝑑𝑙 𝑛 distribution. Math. Sci. and Applic. E-Notes, 1(2): 112- 𝜇𝛽 = = + ∑𝑛𝑖=1 𝑙𝑜𝑔(2𝑥 − 𝑥 2 ) 𝑎 + (1 − 122. 𝑑𝛽 𝛽 ∑𝑛 2 𝛽 2 𝑖=1(2𝑥−𝑥 ) log(2𝑥−𝑥 ) Nekoukhou V 2016. the beta-rayleigh distribution on the 𝑏) (−1) 71 [1−(2𝑥−𝑥 2 )𝛽 ] lattice of integers. J. Stat. Res. Iran, 12(2): 205-224. 𝛽 𝑛 ∑𝑛 2 2 𝑖=1(2𝑥−𝑥 ) 𝑙𝑜𝑔(2𝑥−𝑥 ) Oguntunde PE, Adejumo AO, Okagbue HI & Rastogi MK = + 𝑎 ∑𝑛𝑖=1 𝑙𝑜𝑔(2𝑥 − 𝑥 2 ) + (1 − 𝑏) [1−(2𝑥−𝑥 2 )𝛽 ] 72 𝛽 2016. Statistical properties and applications of a new lindley exponential distribution. Gazi Univ. J. Sci., 29(4): However, log-likelihood function of this distribution cannot 831-838. be solved analytically because of its complex form but it can Sangsanit Y & Bodhisuwan W 2016. The Topp-Leone be maximized by employing global optimization methods generator of distributions: Properties and inferences. available with softwares like R software, SAS, Mathematical Songklanakarin J. Sci. and Techn., 38(5): 537-548. and so on. Shittu OI & Adepoju KA 2013. On the beta-nakagami distribution. Progress in Applied Mathematics, 5(1): 49- Conclusion 58. In this study, we constructed a continuous probability Sun J 2011. Statistical Properties of a Convoluted Beta- distribution with three shape parameters. The density function Weibull Distribution Theses, Dissertations and and the cumulative function of the constructed distribution Capstones 277. http://mds.marshall.edu/etd/277. were derived. Also, the validity of the new distribution was Usman U, Musa Y, Muhammad AB & Zakari Y 2019. On the illustrated. Some statistical properties of the proposed Burr X-Topp Leone distribution. Annals. Comp. Sci. distribution which include reliability analysis, moments, Series, 17(2): 206-208. probability weighted moment and order statistics have been presented. The method of maximum likelihood was used to estimate the parameters.
Conflict of Interest Authors declare there is no conflict of interest related to this study.
FUW Trends in Science & Technology Journal, www.ftstjournal.com
e-ISSN: 24085162; p-ISSN: 20485170; April, 2020: Vol. 5 No. 1 pp. 306 – 309 309