Primary Submission Category: Causal Inference and SUTVA/Consistencies Violations
Understanding Shift-share Designs from the Perspective of Interference
Authors: Ye Wang,
Presenting Author: Ye Wang*
Shift-share designs have been widely adopted by researchers in political economy to evaluate the impacts of random shocks on regions with varying degrees of exposure to these shocks. Examples include how import competition from China affects local labor markets or election results across the United States, and how immigrants from different origins accelerate the advance of innovations in different research fields. Nevertheless, existing approaches are built upon structural restrictions on treatment effects, including additivity and homogeneity of the shocks’ impacts. These restrictions are often violated in practice, leading to biases in both estimation and inference. In this paper, I argue that a shift-share design can be understood as a bipartitie experiment under interference. From this perspective, I introduce novel estimands to capture the expected average treatment effect generated by any shock. These estimands can be nonparametrically identified and consistently estimated without aforementioned restrictions. The proposed method also allows practitioners to examine how the magnitude of the impacts varies within the sample and construct confidence intervals with the correct coverage. I test the method’s performance by replicating Autor, Dorn and Hanson (2013) and detect heterogeneity of treatment effects ignored by previous studies.