This paper addresses M-estimation of conditional mean functions when observations are missing at random. The usual approach of correcting for missing data, when the missing data mechanism is ignorable, is inverse probability weighting. An alternative semiparametric M-estimator which involves...