Within the literature on causal statistical inference, an important goal is to examine the causal mechanisms or channels through which the treatment or intervention affects the outcome of interest. Net (or direct) causal effects measure the effect of the treatment on the outcome while blocking the effect of the treatment on the variable that represents the mechanism. Hence, net effects are useful in learning about the ways in which the treatment causally affects the outcome and, as a result, can be used for policy purposes in the design, development, and evaluation of interventions. Despite their evident importance, the latest theoretical developments on the definition, identification, and estimation of this type of effects have taken place in fields outside economics. The research objectives of this proposal are to: (i) introduce developments on net and mechanism effects in other fields to economics, employing familiar econometric language; (ii) contribute to the literature on these causal effects by providing new conditions for their partial and point identification (nonparametric as well as parametric) under different treatment assignments, allowing for heterogeneous effects; and, (iii) provide relevant applications and simulations of the methods that can guide future applied research.
More specifically, this project first employs the potential outcomes framework (Neyman, 1923; Rubin, 1974) and related concept of principal stratification (Frangakis and Rubin, 2002) to introduce into economics the concepts of net average treatment effect (NATE) and mechanism average treatment effect (MATE). These two effects decompose the total average treatment effect (ATE). Related concepts have been previously introduced in other fields by Robins and Greenland (1992) and Pearl (2001). Second, this project provides new results for the nonparametric partial identification of net treatment effects under minimal assumptions. Subsequently, the project presents different sets of assumptions that allow parametric as well as nonparametric point identification of NATE. The nonparametric results follow from an application of the insights in the seminal work of Imbens and Angrist (1994) and Angrist, Imbens and Rubin (1996) on local average treatment effects. The analysis is done for the cases in which the treatment is randomly assigned, when selection is based on observables, and when selection is based on unobservables. This last case has not been considered in the recent literature on net effects in other fields, and it is important in economics. Finally, applications of the methodologies are provided based on substantive and well known empirical problems employing data sets available to the investigators. They will be complemented by Monte Carlo simulations. These simulations offer further insights on the application of the methods and the robustness of alternative identification assumptions.
This project makes a clear and significant contribution to the econometrics literature on causal inference. It also provides linkages among econometrics, statistics, epidemiology, artificial intelligence, and other areas that emphasize estimation of causal effects. The proposed research is also likely to impact many areas of economics and other sciences by adding new tools to the kit of applied researchers. This is particularly the case in those fields in which the estimation of treatment effects is important for analysis and policymaking, such as labor, public, agricultural and health economics, among others. To this effect, the project stresses the inclusion of empirical applications and simulations that will provide guidance for applied researchers as they implement the methods introduced in this proposal.
Broader Impacts: The results and methods developed in this project will be used as a teaching tool in graduate courses in economics and agricultural economics at the University of Florida and the University of Miami. The research assistants selected for the project ¯who will help with coding the statistical programs for the empirical applications and simulations¯will be well-trained in this regard, and every effort will be made to ensure that they come from underrepresented groups (which is anticipated to be no problem given the existing diversity at the principal investigators? institutions). The project also broadens the participation of underrepresented groups because both PIs, in addition to being junior faculty, are members of such a group. Finally, the project will promote synergies between the University of Miami and the University of Florida, and particularly between economics and agricultural economics, as the PIs are members of each department.
