On efficient estimation in regression models (Q1314466)

The regression model with smooth regression function and smooth error and covariate distribution is considered. This model includes nonlinear parametric models, nonparametric models, partly linear additive regression models and semiparametric problems. The goal is to study the problem of efficiently estimating functionals of the regression function which may also depend on the distribution of the covariate. For this purpose the concept of least dispersed regular estimation is applied. The resulting efficient estimators are characterized by their efficient influence function. The author provides a general procedure for constructing efficient estimators. The usefulness of the proposed unifying approach for the efficiency considerations in regression models is illustrated by constructing efficient estimators for the special regression models mentioned above.