Time sequential estimation of the exponential mean under random withdrawals (Q1082762)

In the paper the problem of estimation of the mean exponential lifetime \(\theta\) under random censorship is considered. It is assumed that the censoring variable Y is independent of the survival time X and the investigator observes the datum (Z,\(\delta)\), where \(Z=\min (X,Y)\) and \(\delta =1\) or 0 according as \(Z=X\) or Y. Let \(Z_{(1)},Z_{(2)},...,Z_{(n)}\) denote the order statistic corresponding to the random sample \(Z_ 1,Z_ 2,...,Z_ n\) and \(\delta_{[j]}=1\) if \(Z_{(j)}\) is uncensored and \(=0\) otherwise. Based on the data \(\{(Z_{(i)},\delta_{[i]}):\) \(1\leq i\leq k\}\) the estimator \({\hat \theta}_{n,k}\) of \(\theta\) is constructed. The loss function is given by \(L_{n,k}=a({\hat \theta}_{n,k}-\theta)^ 2+bn+cV_{n,k}\), where \(V_{n,k}=\sum^{k}_{i=1}Z_{(i)}+(n-k)Z_{(k)}\), a,b,c are given constants. Under some assumptions the estimator \({\hat \theta}_{n,k}\) of \(\theta\) is obtained such that this procedure is asymptotically risk efficient as \(c\to 0\). Pertinent asymptotic properties of the sequential estimator and the stopping times of the proposed scheme are also derived. The results generalized those obtained by \textit{P. K. Sen} [Commun. Stat., Theory Methods A 9, 27-38 (1980; Zbl 0431.62053)] when censorship is absent.