I am a Senior Data Scientist in the Pricing and Applied Economics Team at Starbucks . I work at the intersection of econometrics and machine learning. I received my PhD in Economics from the University of Florida. My research spans the fields of econometrics and microeconomics.
My job market paper [here] focuses on the missing outcome problem and deriving informative bounds using covariates without imposing shape restrictions with applications to consumer confidence. Another part of my research focuses on network formations and extending the partial identification on bounding the centrality measures when networks suffer from the missing link problem.
Abstract: The missing outcome problem is a pervasive problem in economics that arises in many situations and it hinders the researcher's apacity to recover the population moments. The literature has focused on the identifying power of shape restrictions which can be invoked in empirical studies. In this study, we propose a novel approach of partial identification that does not rely on shape restrictions but exploits the variation in the sample of University of Michigan Index of Consumer of Sentiment that resulted from the substitution of landlines with cellphones in telephone surveys. We provide conditions under which the bounds are improved and extend our approach to treatment effects literature.
Abstract: Networks with missing links can arise due to the boundary specification, non-response in surveys, and fixed choice survey design. We model the link formation of a friendship network with missing links arising from non-response in surveys and fixed choice problem and construct sharp upper bounds on the centrality measures. We provide identification conditions to overcome degree heterogeneity. Finally, we apply our proposed approach to study a friendship network in North Carolina.
Abstract: Undercoverage occurs when population members do not appear in the sample frame. For instance, the substitution of landlines with cellphones that took place in the last two decades increased the undercovered population in telephone surveys because survey practices tended to exclude cellphones. This undercoverage problem is related to the identification problem with incomplete information studied by Manski (2009). This paper shows how to construct an identification region a la Manski to assess the extent of the undercoverage problem for a population mean and for the coverage error considered in the survey research literature; and it shows that the widths of both regions are the same. The identification region considers all the possible values that the mean of the undercovered population can take and provides a neat summary of the potential extent of the coverage error. We illustrate the approach using two indices of consumer confidence during the period 2003-2018.
Abstract: Empirical models find a positive effect of political knowledge on turnout, however this may not reflect the true causal effect. The identification of the average causal effect of political knowledge on turnout is hindered by a key identification issue: the endogenous acquisition of political knowledge. Using survey data from the 2016 American National Election Studies (ANES), we employ a nonparametric bounding method to overcome this identification challenge. This method relies on weak and credible assumptions to partially identify the average treatment effect (ATE) of political knowledge on turnout. Specifically, we provide informative bounds by exploring in a sequential fashion the identifying power of different assumptions. We find that the joint combination of the monotone treatment response, monotone treatment selection, and monotone instrumental variable, allows us to improve, in the sense that the estimated upper bound is lower than the point estimates reported in the literature.
Abstract: Linear-in-means models, first proposed by Manski (1993), are used in modelling social interactions. Manski shows that identification of the endogenous, contextual, and correlated effects is not possible when we condition on individual member characteristics to identify the groups. Different from Manski we use a nonlinear linear-in-means model to identify the main effects under several very mild conditions and estimate our model using a GMM approach.