Weibull-Bayesian Estimation Based on Maximum Ranked Set Sampling with Unequal Samples

A modification of ranked set sampling (RSS) called maximum ranked set sampling with unequal sample (MRSSU) is considered for the Bayesian estimation of scale parameter α of the Weibull distribution. Under this method, we use Linex loss function, conjugate and Jeffreys prior distributions to derive the Bayesian estimate of α. In order to measure the efficiency of the obtained Bayesian estimates with respect to the Bayesian estimates of simple random sampling (SRS), we compute the bias, mean squared error (MSE) and asymptotic relative efficiency of the obtained Bayesian estimates using simulation. It is shown that the proposed estimates are found to be more efficient than the corresponding one based on SRS.

Keywords

Share and Cite:

Biradar, B. and Shivanna, B. (2016) Weibull-Bayesian Estimation Based on Maximum Ranked Set Sampling with Unequal Samples. Open Journal of Statistics, 6, 1028-1036. doi: 10.4236/ojs.2016.66083.

1. Introduction

In certain practical problems, actual measurements of a variable interest are costly or time-consuming, but the ranking items according to the variable is relatively easy with- out actual measurement. Under such circumstances McIntyre  proposed a sampling scheme called ranked-set sampling (RSS) which can be employed to gain more information than simple random sampling (SRS), while keeping the cost of, or the time constraint on, the sampling about the same. In RSS; one first draws units at random from the population and partition them into m sets of m units. The m units in each set are ranked without making, an actual measurement. The first set of m units are ranked and the smallest is selected for actual quantification. From the second set of m units, the unit ranked and the second smallest is measured, and so on. This method of selection continues until the unit ranked largest is measured from the m-th set. If a large sample is required, then the procedure can be repeated r times to obtain a sample of size . These chosen elements are called ranked set sampling. The mathe- matical support and statistical theory was provided by Takahasi and Wakimoto  . Dell and Clutter  studied theoretical aspects of this technique on the assumption of perfect and imperfect judgment ranking. Shaibu and Muttlak  used median and extreme ranked set sampling method for estimating the parameters of normal, expo- nential and gamma distributions. Al-Omari et al.  Used extreme ranked set sampling method to find the estimates of the population mean. Islam et al.  Obtained the modified maximum likelihood estimator of location and scale parameters depend on selected ranked set sampling for normal distribution. Ibrahim and Syam  used stratified median ranked set sampling method for estimating the population mean.

Some research works have investigated ranked set sampling from a Bayesian point of view. Varian  and Zellner  introduced Bayesian estimation by using asymmetric loss functions. Al-Saleh and Muttlak  obtained the Bayesian estimates of the exponential distribution. Ahmed  obtained the Bayesian estimators of log-normal distribution based on RSS and SRS using Bayes risk. Sadek et al.  , and Sadek and Alharbi  used the asymmetric loss function to obtain the Bayesian estimate of the exponential and Weibull distributions respectively, based on SRS and RSS. Al-Hadhrami and Al-Omari  showed that the Bayesian estimation of the mean of normal distri- bution based on moving extreme ranked set sampling (MERSS) is more efficient than SRS. Hassan  obtained the maximum likelihood estimator and Bayesian estimates of shape and scale parameters of the exponentiated exponential distribution based on SRS and RSS. For more research work on Bayesian one may refer to Mohammadi and Pazira  , Ghafoori et al.  , Said Ali Al-Hadhrami and Amer Ibrahim Al-Omari  , Mohie El-Din et al.  .

In this paper, we derive the Bayesian estimates of the Weibull scale parameter α based on gamma and Jeffreys prior distributions by MRSSU method proposed by Biradar and Santosha  . In Section 2, the preliminaries are discussed. The Bayesian estimates under SEL and LINEX loss functions of the parameter of Weibull distribution using SRS and MRSSU are presented in Section 3. Simulation results and Conclusions are presented in Section 4 and 5 respectively.

2. Preliminaries

Let be a sequence of independent and identically distributed (iid) random variables from a Weibull distribution with probability density function (pdf) (1)

And cumulative distribution function (cdf) (2)

where is the scale parameter and is shape parameter.

In order to derive, and to measure the performance of an estimator we use squared error, loss function (SEL) (see, Berger  ) and Linex loss function.

The Linex loss function for the parameter can be expressed as where ; is an estimate of and, c and d are shaped and scale parameters. The sign and magnitude of the shape parameter c indicate that the direction and degree of symmetry, respectively. When the value of c is zero, the Linex loss function is approximately squared error loss, when c is less than zero, the Linex loss function gives more weight to under-estimation against over-estimation, and it is reversed when c value is greater than zero. The conjugate prior for , is considered, whose pdf is given by (3)

where . If , then becomes the Jeffreys prior.

3. Bayesian Estimates

In this section, we derive the Bayes estimates of the Weibull parameter based on simple random sampling and maximum ranked set sampling with unequal samples by assuming that the shape parameter is known. In each case, we use both conjugate and non-informative prior for the scale parameter. Also, we use the symmetric loss function (squared error loss) and asymmetric loss function (Linear-exponential, Linex) to derive the corresponding Bayesian estimates. And we denote and as posterior densities of, given SRS() and RSS() respectively.

3.1. Bayesian Estimation Based on SRS

Let be a sequence of iid random variables, has the Weibull distribution with parameters () and be the conjugate prior. In this case, the posterior density based on SRS is given by

(4)

Hence, the Bayesian estimation of depend on squared error loss (SEL) is

because the Bayes estimate with respect to SEL is the posterior mean then

(5)

While the Bayesian estimate of based on Linex loss function is

where,

Then,

(6)

3.2. Bayesian Estimation Based on MRSSU

Assume that the variable of interest X has density function and distribution function is known. Let, be m sets of random samples from X, and they are independent. Denote, ,

. Let is taken from the first set, is taken from the second set and is taken from the last set, then be a one cycle MRSSU from X and all’s are independent. In this study we assume that there is no error in ranking. The density of has the same density as the ith order statistic (maximum) of an SRS of size i from, i.e., has the density

Let MRSSU be drawn from Weibull distribution, then the density function of is

Then the joint density of MRSSU in this case due to independence of’s is given by

where and.

Then the posterior density of α is

(7)

The Bayes estimate of based on the squared error loss function is

(8)

Next, in order to derive the Bayesian estimation of based on LINEX loss function, first we need to compute the posterior expectation of from Equation (7) as

(9)

Now the Bayesian estimation of on LINEX is

(10)

where is as derived in Equation (9).

3.3. Bayesian Estimation Based on Non-Informative Prior

The non-informative prior distribution of the parameter is obtained from Equation

(3) and it is given by. Then, we obtain the Bayesian estimates of

in this case as follows:

1) Simple Random Sample:

(11)

and

(12)

2) Maximum ranked set sampling with unequal samples:

(13)

and

(14)

4. Simulation Results

To illustrate the performance of the derived Bayesian estimates of scale parameter of the Weibull distribution with informative and non-informative prior based on SRS and MRSSU, we carry out the Monte Carlo simulations using R-Software version 3.1.1. We compute bias, mean squared error and relative efficiency of the estimators by assuming the shape parameter is known. The numerical results obtained for fixed values of, [and 1] and sample size m [3, 4 and 5] for 1000 runs. The bias of the Bayesian estimates based on SRS and MRSSU are presented in Table 1 and Table 2, and MSE of the Bayesian estimates based on SRS and MRSSU is presented in Table 3 and Table 4.

Table 1. Bias of the Bayesian estimates based on SRS and MRSSU. For (when, ,).

Table 2. Bias of the Bayesian estimates based on SRS and MRSSU. For (when, ,).

The relative efficiency of the Bayesian estimates based on maximum ranked set sampling with unequal samples with respect to simple random sampling can be defined as follows

And are presented in Table 5.

Table 3. MSE of the Bayesian estimates based on SRS and MRSSU. For (when, ,).

Table 4. MSE of the Bayesian estimates based on SRS and MRSSU. For (when, ,).

Table 5. Relative efficiency when and.

5. Conclusions

We present Bayesian estimation based on SRS and MRSSU. The Weibull distribution is used as an application example to illustrate our results. We compute bias, MSE and relative efficiency of the derived Bayesian estimates and then make a comparison between SRS and MRSSU. Our observations of the results are stated in the following points:

1) From Table 1 and Table 2, first, we found that the Bayesian estimates of are all biased. Next, we found that the Bayesian estimates based on Jeffreys prior are less biased than gamma prior. Also, we observed that the Bayesian estimates based on MRSSU are considerably less biased than SRS.

2) From Table 3 and Table 4, it is observed that the mean squared error of all estimates decreases when sample size m increases. Next, we observed that the Bayesian estimates based on MRSSU have a much smaller mean squared error than the corresponding Bayesian estimates based on SRS in all cases considered.

3) From Table 5, we observe that the relative efficiency of the Bayesian estimator based on MRSSU w.r.t. SRS Bayesian estimators are greater than 1 and increases with m. Also, decreases in Linex function as m increases for.

Therefore, we conclude that the Bayesian estimates based on maximum ranked set sampling with unequal samples are more efficient than the corresponding Bayesian estimates of simple random sampling.

Finally, we conclude that the results of the simulation experiment showed that the Bayesian estimates based on maximum ranked set sampling with unequal samples are more efficient, when compared with the Bayesian estimates of simple random sampling.

Acknowledgements

The authors would like to thank the referees for their helpful comments that have led to an improved paper.

Conflicts of Interest

The authors declare no conflicts of interest. 