We derive testable implications of instrument validity in just identified treatment effect models with endogeneity and consider several tests. The identifying assumptions of the local average treatment effect allow us to both point identify and bound the mean potential outcomes (i) of the always takers under treatment and (ii) of the never takers under non-treatment. The point identified means must lie within their respective bounds, which provides us with four testable inequality moment constraints. Furthermore, we use dominance/equality restrictions on potential outcomes across subpopulations to increase the power to detect violations of instrument validity. Finally, we adapt our testing framework to the identification of distributional features (as local quantile treatment effects). A brief simulation study and an application to labor market data are also provided.

Previous research found that less accommodating caseworkers are more successful in placing unemployed workers into employment. This paper explores the causal mechanisms behind this result using semiparametric mediation analysis. Analyzing rich linked job seeker-caseworker data for Switzerland, we find that the positive employment effects of less accommodating caseworkers are not driven by a particularly effective mix of labor market programs but, rather, by other dimensions of the counseling process, possibly including threats of sanctions and pressure to accept jobs.

Standard sample selection models with non-randomly censored outcomes assume (i) an exclusion restriction (i.e., a variable affecting selection, but not the outcome) and (ii) additive separability of the errors in the selection process. This paper proposes tests for the joint satisfaction of these assumptions by applying the approach of Huber and Mellace (2011) (for testing instrument validity under treatment endogeneity) to the sample selection framework. We show that the exclusion restriction and additive separability imply two testable inequality constraints that come from both point identifying and bounding the outcome distribution of the subpopulation that is always selected/observed. We apply the tests to two variables for which the exclusion restriction is frequently invoked in female wage regressions: non-wife/husband's income and the number of (young) children. Considering eight empirical applications, our results suggest that the identifying assumptions are likely violated for the former variable, but cannot be refuted for the latter on statistical grounds.

Summary In the presence of an endogenous binary treatment and a valid binary instrument, causal effects are point identified only for the subpopulation of compliers, given that the treatment is monotone in the instrument. With the exception of the entire population, causal inference for further subpopulations has been widely ignored in econometrics. We invoke treatment monotonicity and/or dominance assumptions to derive sharp bounds on the average treatment effects on the treated, as well as on other groups. Furthermore, we use our methods to assess the educational impact of a school voucher program in Colombia and discuss testable implications of our assumptions. Copyright © 2015 John Wiley & Sons, Ltd.

Using a comprehensive simulation study based on empirical data, this article investigates the finite sample properties of different classes of parametric and semiparametric estimators of (natural) direct and indirect causal effects used in mediation analysis under sequential conditional independence assumptions. The estimators are based on regression, inverse probability weighting, and combinations thereof. Our simulation design uses a large population of Swiss jobseekers and considers variations of several features of the datagenerating process (DGP) and the implementation of the estimators that are of practical relevance. We find that no estimator performs uniformly best (in terms of root mean squared error) in all simulations. Overall, so-called "g-computation" dominates. However, differences between estimators are often (but not always) minor in the various setups and the relative performance of the methods often (but not always) varies with the features of the DGP.