[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

Let us consider a right censored Type II sample, RCD(n, r) from the Weibull distribution F(x; c, m), where only the smallest r data T_1<T_2<…<T_r have been observed, while the (n-r) ones T_j(> T_r,j = r+1, r+2,…, n) among the sample of size n have not yet been known; and c and m are unknown scale and shape parameters, respectively, of the distribution. The purpose of the present paper is to investigate certain estimators of the parameters which have smaller values of | BIAS | and/or MSE of the estimators than the ones of the previously known conventional methods such as the least square(LSE), the maximum likelihood(MLE) and BLUE for 1/m. To achieve the purpose, we propose to use two kinds of quantities: (i) a combined measure, K, of evaluating simultaneously errors along abscissa and along ordinate, and (ii) appropriately predicted value X^^〜_j for unknown X_j =1nT_j (j = r+1, r+2, …, n). Now, let us presume that {X_1,X_2,…,X_r, X^^〜_<r+1>,…,X^^〜_n} is a set of complete sample with size n, and apply the conventional methods shown above and certain methods of bias reduction. Then, performing Monte Carlo simulation of estimating m and c, and of evaluating BIAS and MSE for 20000 sets of RCD(n, r), we can conclude that the estimators proposed in the present paper especially for m are good enough, since their | BIAS | and MSE are both considerably reduced to lower levels than the ones of the conventional methods in most cases of n and r of our simulation for r = 4,5,…,n; n = 5,7,10,15,20.

View full abstract