L-moment estimation for parametric survival models given censored data

An estimator for the rth L-moment, λ_r, given right censored data is proposed, which takes the form of λ̂_r=Σ_j=1ⁿT_(j)u_j(r), where the T_(j)'s are the ordered censored or failure times. We present the L-moments for several common survival distributions. Under certain regularity conditions, it is shown that λ̂_r converges in probability to λ_r, and has an asymptotic normal distribution. We further develop exact bootstrap estimators of mean and variance of λ̂_r. The procedure is illustrated by an application to head-and-neck cancer survival data. Monte Carlo simulation studies show that our L-moment estimator may be used to characterize distributions and provide an alternative approach to parameter estimation.