Identifying Causal Effects in Information Provision Experiments

Download:

• Paper

• Code available upon request

Abstract:

Information provision experiments are a popular way to study causal effects of beliefs on behavior. Researchers estimate these effects using TSLS. I show that existing TSLS specifications do not estimate the average partial effect; they have weights proportional to belief updating in the first-stage. If people whose decisions depend on their beliefs gather information before the experiment, the information treatment may shift beliefs more for people with weak belief effects. This attenuates TSLS estimates. I propose researchers use a local-least-squares (LLS) estimator that I show consistently estimates the average partial effect (APE) under Bayesian updating, and apply it to Settele (2022).

Figure 2. LLS Estimates of the Average Partial Effect are 70% Larger than TSLS

Citation

Dylan Balla-Elliott (2023). “Identifying Causal Effects in Information Provision Experiments.” https://doi.org/10.48550/arXiv.2309.11387 .