Click here for my full Research Statement hosted on my Github.

Working Papers

Cluster-Robust Inference when Treatment Effects Vary Across Clusters: A Design-Based Approach, Danko, Robin. [Job Market Paper]

"Clustered standard errors are common tools used by empirical researchers in the social sciences to obtain valid inference. Clusters can be defined by geography or by social strata such as gender. Typically, cluster-robust errors adjust for correlations induced by sampling the outcome from a data-generating process with correlated cluster-level components. The data-generating process is typically assumed to either contain a small, finite, and fixed number of clusters or an infinite number of clusters. In this paper, I show that modeling clusters as finite features of a population the number of conventional cluster-robust standard errors can be severely inflated when the number of clusters is large enough for valid asymptotic approximation. I propose new standard errors that correct this bias."

2SLS Variance Adjusting for Design-Based Uncertainty: Inference and Estimation, Danko, Robin. 

"In the typical approach to inference in the social sciences, researchers assume that a negligibly small sample is drawn from a large population. While natural in many applications, it is less natural in other instances such as when doing inference on statewide data when data on all 50 states is available; in this case the sample does not differ from the population. In this article, I apply a design-based approach to analyze the population variance of the 2SLS estimator. Design-based uncertainty explicitly considers the unknown counterfactual outcomes under alternative treatment schemes. I derive standard errors of the 2SLS estimator that consider both design-based and sampling-based uncertainty. I show that these new standard errors are generally smaller than the usual infinite population sampling-based standard errors and provide conditions under which they coincide."