Accounting for missing data in M-estimation: a general matching approach |
| |
Authors: | Anton Flossmann |
| |
Institution: | (1) Department of Economics, University College London, Gower Street, London, WC1E 6BT, UK;(2) University of St.Gallen, SIAW, Bodanstrasse 8, 9000 St.Gallen, Switzerland |
| |
Abstract: | 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 (IPW). An alternative
semiparametric M-estimator which involves local polynomial matching techniques is proposed and its asymptotic distribution
is derived. Like IPW, the proposed estimation approach has a double robustness property for the estimation of unconditional
means. Monte Carlo evidence suggests slightly better finite sample properties of the semiparametric M-estimator relatively
to IPW. A version of the proposed estimator is applied to estimate the impact of noncognitive skills on wages in Germany for
two different educational treatment regimes. |
| |
Keywords: | |
本文献已被 SpringerLink 等数据库收录! |
|