The modified moment method for multiply censored samples (Q1897883)

Assume that \(T_ 1, \dots, T_ n\) are positive i.i.d. random variables with the distribution function \(F\). We write \(F\in {\mathcal K}_ 1\) if \(F(t; u,b)= F_ 0 ((\varphi (t)- u)/b)\), and \(F\in {\mathcal K}_ 2\) if \(F(t; \theta)= F_ 0 (t/\theta)\), where \(F_ 0\) is a parameter-free distribution function, \(\varphi\) is a continuous increasing on \((0, +\infty)\) function such that \(\varphi (0)=- \infty\). Denote \(S(t)= 1- F(t)\) as the survival function. Assume that the censoring times \(U_ 1, \dots, U_ n\) are random variables or constants. The random variables \(T_ i\) and \(U_ j\) may be dependent. Denote \(X_ i= T_ i\wedge U_ i= \min (T_ i, U_ i)\), and let \(\delta_ i= \chi (T_ i\leq U_ i)\) denote the indicator of the set \(\{T_ i\leq U_ i\}\). Then the set of the \(n\) random variables \((X_ 1, \delta_ 1), \dots, (X_ n, \delta_ n)\) is the so-called multiply censored sample (m.c.s.). In the partial case of a type I censored sample \(U_ i= t\), for a type II censored sample \(U_ i= T_ n\), where \(T_{rn}\) is the \(r\)-th order statistic from \(T_ 1, \dots, T_ n\). There is a large number of papers considering estimation of location and scale parameters \(u\) and \(b\) in the case of type II censored samples. The most frequently used are the maximum likelihood and best linear unbiased or best linear invariant estimators (m.l.e., b.l.u.e., b.l.i.e.). In the case of multiply censored samples, m.l.e. are used as a rule. Graphical and least squares methods can also be used. In this paper modified moment estimators (m.m.e.) of parameters \(u\), \(b\), \(\theta\) are given; asymptotical properties of m.m.e. are investigated; estimators of the m.m.e. variances are given; and graphical goodness of fit tests for the classes \({\mathcal K}_ 1\) and \({\mathcal K}_ 2\) are proposed. Calculation of the m.m.e. is considerably simpler than calculation of the above mentioned estimators.