Guest Speaker: Credible or Confounded? Applying sensitivity analyses to improve research and its evaluation under imperfect identification

Event Date: 

Friday, May 3, 2019 - 12:30pm to 2:00pm

Chad Hazlett
Assistant Professor
Statistics and Political Science
UCLA

Credible or Confounded? Applying sensitivity analyses to improve research and its evaluation under imperfect identification

Abstract: Social scientists pose important questions about the effects of potential causes, but often cannot eliminate all possible confounders in defense of causal claims. Sensitivity analyses can be useful in these circumstances, providing a route to rigorously investigate causal questions despite imperfect identification. Further, if more widely adopted, these tools have the potential to improve upon standard practice for communicating the robustness of causal claims, while suggesting new ways for readers and reviewers to judge research. We illustrate these uses of sensitivity analysis in an application that examines two potential causes of support for the 2016 Colombian referendum for peace with the FARC. Conventional regression analyses find "statistically and substantively significant" estimated effects for both causes. Yet, sensitivity analyses reveal very weak confounders could overturn one cause (exposure to violence), while extremely powerful confounders are needed to overturn the other (political affiliation with the deal's champion).

Bio: Chad Hazlett is an assistant professor at UCLA in Statistics and Political Science. His work focuses on developing research methods for causal inference in the social sciences, including a focus on civil conflict and mass atrocities.