Working Paper: NBER ID: w13981
Authors: Bryan S. Graham; Cristine Campos De Xavier Pinto; Daniel Egel
Abstract: We propose a new inverse probability weighting (IPW) estimator for moment condition models with missing data. Our estimator is easy to implement and compares favorably with existing IPW estimators, including augmented inverse probability weighting (AIPW) estimators, in terms of efficiency, robustness, and higher order bias. We illustrate our method with a study of the relationship between early Black-White differences in cognitive achievement and subsequent differences in adult earnings. In our dataset the early childhood achievement measure, the main regressor of interest, is missing for many units.
Keywords: Missing Data; Cognitive Achievement; Wage Gap; Inverse Probability Weighting; Econometrics
JEL Codes: C14; C21; C23
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| early cognitive skills (G53) | wage gaps (J31) |
| IPT estimate of wage gap (J79) | more precise than unweighted and IPW estimates (C13) |
| IPT method (C67) | corrects for biases due to non-random missingness (C83) |
| analysis of complete cases (C29) | inconsistent and imprecise estimates (C13) |
| IPT method (C67) | more accurate representation of causal relationship between cognitive skills and wage outcomes (J31) |
| early cognitive skills (G53) | adult earnings (J31) |