I am an Assistant Professor of Economics at UCLA. I am also a research affiliate of the Human Capital and Economic Opportunity Global Working Group (HCEO). My research interests are policy evaluation, causality and applied econometrics. A constant theme in my research is the identification, estimation and inference of causal effects. More recently, I have studied how to apply revealed preference analysis to enhance causal inference on social experiments. I have studied a range of social experiments including the Perry Pre-school Intervention, High/Scope Comparison Study, Abecedarian Project, Nurse-Family Partnership, Primeira Infancia Melhor (in Brazil), Oportunidades (Mexico), Moving to Opportunity (MTO).
Noncompliance as a Rational Choice: A Framework that Exploits Compromises in Social Experiments to Identify Causal Effects.
Noncompliance is a pervasive problem in social experiments which hinges the identification of causal effects. This paper offers a framework in which noncompliance is not portrayed as a drawback, but a key ingredient of identification analysis. The method uses revealed preference analysis to exploit the incentives generated by the design of social experiments in order to nonparametrically identify causal parameters. The framework is used to evaluate the Moving to Opportunity, the largest housing experiment in the US. Moving to Opportunity was designed to investigate the casual effect of relocating disadvantaged families from high-poverty neighborhoods to low-poverty communities. Substantial noncompliance prevents the evaluation of neighborhood effects, that is the causal effect of residing in different neighborhoods types. Nevertheless, noncompliance still allows for the evaluation of voucher effects, that is the causal effect of being offered a voucher. Previous literature shows that voucher effects on labor market outcomes are not statistically significant. This paper exploits the incentives of the MTO intervention to identify neighborhood effects. Although voucher effects are not statistically significant, neighborhood effects are. The result reconciles MTO with a growing literature attesting the impact of neighborhood quality on economic well-being. The framework can be broadly applied to exploit economic incentives in multiple choice models with heterogeneous agents and categorical instrumental variables. I show that all causal parameters can be estimated by standard 2SLS applied to suitable data transformations.
Instrumental Variables and Causal Mechanisms: Unpacking the Effect of Trade on Workers and Voters
(with Christian Dipel, Robert Gold, and Stephan Heblich)
Instrumental variables (IV) are commonly used to identify treatment effects, but standard IV estimation cannot unpack the complex treatment effects that arise when a treatment and its outcome together cause a second outcome of interest. For example, IV estimations have been used to show that import exposure to low-wage countries has adversely affected high-wage countries’ labor markets. They have also been used to show that such import exposure has polarized voters. However, they cannot answer the question to what extent the latter is a consequence of the former. We propose a new identification strategy that allows us to do so, appealing to one additional identifying assumption and requiring no additional instruments. Applying this framework, we estimate that across specifications labor market adjustments explain virtually all of the effect of import exposure on voting. This enables us to provide rigorous evidence that the correct policy response to voter polarization has to be focused on labor markets.
Inference with Imperfect Randomization: The Case of the Perry Preschool Program
(with James Heckman and Azeem Shaik)
This paper considers the problem of making inferences about the effects of a program on multiple outcomes when the assignment of treatment status is imperfectly randomized. By imperfect randomization we mean that treatment status is reassigned after an initial randomization on the basis of characteristics that may be observed or unobserved by the analyst. We develop a partial identification approach to this problem that makes use of information limiting the extent to which randomization is imperfect to show that it is still possible to make nontrivial inferences about the effects of the program in such settings. We consider a family of null hypotheses in which each null hypothesis specifies that the program has no effect on one of several outcomes of interest. Under weak assumptions, we construct a procedure for testing this family of null hypotheses in a way that controls the familywise error rate - the probability of even one false rejection - in finite samples. We develop our methodology in the context of a reanalysis of the HighScope Perry Preschool program. We find statistically significant effects of the program on a number of different outcomes of interest, including outcomes related to criminal activity for males and females, even after accounting for the imperfectness of the randomization and the multiplicity of null hypotheses.
Randomized Biased-Controlled Trials
The literature on the design of social experiments largely benefits from standard theory of randomized controlled trials (RCTs). In it, noncompliance is usually interpreted as a departure of a perfect randomization that generates selection bias and often prevents the identification of causal parameters (Duo et al., 2008; Deaton, 2016). Unfortunately, noncompliance is rather a rule than an exception in social experiments (Heckman et al, 2014). I develop a general framework that combines the incentives induced by the design of social experiments with revealed preference analysis to nonparametrically identify causal parameters. The method differs from typical RCT design as noncompliance is not a drawback, but a key ingredient of identification analysis. I show that the choice criteria such as the Ordered Monotonicity of Imbens and Angrist (1994) and Unordered Monotonicity of Heckman and Pinto (2016) can be traced to particular properties of experimental design. The method broadly applies to unordered choice models with categorical instrumental variables and multiple treatments.
Mediation Analysis in IV Settings With a Single Instrument
(Christian Dippel, Robert Gold, and Stephan Heblich)
The method of Two-stage Leas Squares (2SLS) employs instrumental variables (IV) to evaluate the causal effects of an endogenous treatment on an outcome of interest. The method is widely adopted by empirical economists who typically investigate a data set consisting of the instrument variable, the treatment variable and multiple outcomes. Although a single instrument enables the identification of treatment effects on multiple outcomes, it cannot identify the causal effect of one outcome on another. In particular, the standard IV model cannot unpack the causal effects that arise when the treatment variable and an outcome together cause a second outcome of interest. We present a simple identification strategy that enables the researcher to use instrumental variables to identify the causal relation among outcomes. Our method does not require additional instruments. It exploits an identifying assumption regarding unobserved error terms that maintains the endogeneity of the treatment variable while allowing for endogeneity among outcomes. This paper offers a novel application of the well-known method of 2SLS for a class of IV model with a single instrument and multiple outcomes.
The Effects of Two Influential Early Childhood Interventions on Health and Healthy Behaviour
(with Gabriella Conti and James Heckman), The Economic Journal, 2016, Vol. 126, pp. 28-65.
Causal Analysis After Haavelmo
(with James Heckman), Econometric Theory, 2015, Vol. 31, pp 115-151.
Econometric Mediation Analyses: Identifying the Sources of Treatment Effects from Experimentally Estimated Production Technologies with Unmeasured and Mismeasured Inputs
(with James Heckman), Econometric Reviews, 2015, Vol. 34, pp 6-31.