Improving the Precision of Bounds for Generalization
Friday, November 9, 2018: 11:30 AM-12:15 PM
Atrium - Exhibit Level (Marriott Wardman Park)
*Names in bold indicate Presenter
Participant: Wendy Chan, University of Pennsylvania
Description of Research: When study samples are not randomly selected from a population, this introduces a bias in treatment impacts. Evidence based policy often rely on estimates of treatment impacts to inform decisions. However, if the estimates rely on assumptions that cannot be validated empirically, this raises the question of whether the estimates are useful. My research considers an alternative perspective to estimation without the dependence on untestable assumptions. Although this framework, based on bounding, can yield wide bounds, I explore the ways that observable administrative data can be leveraged to improve the precision of estimates. This has implications for data accessibility since inference based on fewer assumptions can be improved with more information. An important goal of my research is to illustrate the ways that greater accessibility to data leads to better estimation procedures that do not necessarily rely on assumptions.