A note on the construction of asymptotically linear estimators (Q1096991)

The recent results on asymptotically efficient estimation in semiparametric models have induced an interest in constructing asymptotically linear estimators. The main reason for that is that an asymptotically linear estimator whose influence function is the efficient score function is asymptotically efficient. Efficient, asymptotically linear estimators have been proposed by several researchers but they require several estimates of the influence function each based only on a fraction of the sample. However, for moderate sample sizes it is preferable to use estimates of the influence function based on all data than to use several estimates of the influence function, each based only on a fraction of the sample. In the paper it is discussed how to design asymptotically linear estimators that use estimates of the influence function based on the entire sample. Sufficient conditions are given which are slightly stronger than those in the author's previous paper, Ann. Stat. 14, 1139- 1151 (1986; Zbl 0612.62062). These conditions are also necessary. The applicability of the conditions is demonstrated in several examples.