Skip to main content

About my research

Job market paper

This paper focuses on the estimation of the average treatment effect on the treated (ATT) in evaluation studies under unconfoundedness. As an alternative to traditional matching and reweighting methods, I propose a constrained Dirichlet process mixture of normals (DPMN) model to consistently estimate the covariates distribution in the treatment group and match the control units to the treated so that the distributions of covariates are stochastically equivalent. I use this approach to build a matching estimator and a reweighting estimator with desirable properties. First, the DPMN matching estimator meets the balancing property by construction. Second, since DPMN yields consistent estimates of the propensity score, the reweighting estimator is semi-parametrically efficient. Traditional matching and propensity-score based methods are two-step approaches, which may result in incorrect standard errors. In this paper, the whole algorithm is integrated into a single efficient Markov Chain Monte Carlo scheme: the resulting marginal standard errors can account for errors arising from the first step estimation. I illustrate this new method with Monte Carlo experiments and an empirical application of the LaLonde(1986) data. The DPMN reweighting estimator is found to have a performance comparable to conventional reweighting estimators. I also find that the DPMN matching estimator is less biased and more efficient than traditional matching estimators, as a result of improved balance.

Research papers

In this paper we investigate how the Research Excellence Framework (REF), last held in 2014 to assess the research quality in British higher education institutions over the period 2008-2013, perceive economics journals in their assessment system. Exploiting on-line published data on submitted research outputs of different REF quality standards, which is only available at the institutional level, we propose a novel algorithm within an ordered probit framework that allows us to distinguish the censored REF standards for each individual submission and to estimate how economics journals were perceived by the Economics and Econometrics sub-panel and the Business and Management Studies sub-panel. In particular, we develop an efficient Markov Chain Monte Carlo (MCMC) sampling scheme for the inference and also suggest a robust and weakly informative prior distribution to overcome the potential separation problem. This is the first paper to employ a standard regression model to directly predict the perception of journal quality for the REF 2014 exercise. The estimated results can be viewed as a directory for determining to what extent each economics journal meets the criteria set by the REF 2014. Our proposed method can be generalised to other generalised linear models where the outcomes are censored at an aggregate level.

Empowering women and enhancing children's early development are two important goals that are often pursued via independent policy initiatives in developing countries. In this paper we study a unique approach that pursues both goals at the same time: empowering mothers through tools that also advance their children's development. A program operated by AVSI, an Italian NGO, in a poor neighborhood of Quito, Ecuador, targets parents of children from birth to age 5. It provides family advisor-guided parent training sessions once every two weeks for groups of six to eight mothers and their children. We find that the program empowered women in various dimensions, including higher labor force participation and employment, higher likelihood of a full-time job in the formal-sector and higher wages. Treated mothers are also more likely to continue their education, make independent decisions regarding their own finances, have greater role in intra-household decisions, especially on issues involving children's education and discipline and increase parental inputs into their children's development. We find that treated children improve their cognitive and non-cognitive skills, for example, they are less likely to repeat a grade or temporarily drop-out from schooling, are less absent from and have improved behaviors in school, have better attitudes towards learning, and achieve higher scores on cognitive tests. Applying a recently suggested factor model of children's relative non-cognitive skills reaffirms our finding of significant gains in children non-cognitive skills. All results hold when we estimate aggregate treatment impacts, use summary indices instead of individual outcomes in order to account for multiple inference, when we use entropy balancing to adjust for differences in pre-treatment covariates, and when we use other robustness checks.

Work in progress

  • "Teachers in Brazil", with Fernanda Brollo, Roland Rathelot, and Victor Lavy
  • "Bayesian Approach to the Non-linear Panel Data Model", with Mingli Chen