This research, consisting of two projects, will develop innovative econometric methods to study the effects of public policy interventions. The first project studies simple matching methods for treatment effects estimation and missing data imputation. New analytical techniques for the study of the distribution of simple matching estimators of average treatment effects have opened the way to solve two outstanding problems: (a) The properties of the most popular matching procedures to estimate the effects of and other treatment effects---simple matching estimators on the estimated propensity score---are not well understood. This project will fill this gap in the literature by extending recent development to the case when matching is done on estimated covariates (as the propensity score). (b) The CPS and other large surveys use a matching procedure known as 'hot deck' to impute missing values. Large biases in simple matching estimators of treatment effects indicate that hot deck imputation procedures will produce similar biases. This research will study the properties of an improved hot deck procedure that is free of these biases.
Studies of the effects of policy interventions often take place at an aggregate level and use economies not exposed to the intervention of interest as a control group. This approach is subject to two problems: (i) there is some ambiguity about how to choose control groups with researchers using subjective measures of affinity to select between possible controls. (ii) If aggregate data are used for the analysis, the usual cross-section standard errors, which measure only uncertainty about aggregate variables, do not apply. Uncertainty in these models is about the ability of the control group to reproduce the behavior of the exposed economies in absence of the intervention. This project will study the properties of data-driven procedures to construct adequate control groups. This procedure limits the discretion in the choice of the control group, forcing the empirical researcher to demonstrate the affinities between the exposed and non-exposed economies using observed quantifiable characteristics. In addition, this project will study how to measure uncertainty about the quality of the comparison control groups using time series information.
The properties of propensity score matching estimators and estimators based on data imputed by 'hot deck 'are important problems empirical research in economics. Moreover, the study of these properties will entail the development and application of new analytical methods in econometrics. The methods developed in this research will also be used to write software for the STATA econometric package for wider use. The project will also help in the efficient development and implementation of public policy.
