Signaling, Random Assignment, and Causal Effect Estimation
22 Pages Posted: 10 Mar 2020
Date Written: February 18, 2020
Causal estimates from randomization are commonly viewed as ideal. However, their primacy for many real-world decisions is questionable. This is because random assignment, in eliminating self-selection, also eliminates signaling. However, outside experiments, agents make discretionary decisions, generating signaling content which alters beliefs, payoffs, and causal effects. Therefore, if the objective is informing optimal discretionary decisions, rather than predicting outcomes under forced actions or noisy mistakes, random assignment can be misleading. In applications from finance, labor, and macroeconomics, we show signaling can amplify, attenuate, or reverse the sign of causal effects derived from random assignment, with corresponding implications for optimal actions.
Keywords: signal, random assignment, causal effect, selection
JEL Classification: D82, E6, G3, J24
Suggested Citation: Suggested Citation