Working Paper: NBER ID: w26109
Authors: Joshua Schwartzstein; Adi Sunderam
Abstract: We present a framework for analyzing “model persuasion.” Persuaders influence receivers’ beliefs by proposing models (likelihood functions) that specify how to organize past data (e.g., on investment performance) to make predictions (e.g., about future returns). Receivers are assumed to find models more compelling when they better explain the data, fixing receivers’ prior beliefs over states of the world. Model persuaders face a key tradeoff: models that better fit the data given receivers’ prior beliefs induce less movement in receivers’ beliefs. This tradeoff means that a receiver exposed to the true model can be most misled by persuasion when that model fits poorly—for instance when there is a lot of data that exhibits randomness. In such cases, a wrong model often wins because it provides a better fit. Similarly, competition between persuaders tends to neutralize the data because it pushes towards models that provide overly good fits and therefore do not move receivers’ beliefs much. The fit-movement tradeoff depends on receiver characteristics, so with multiple receivers a persuader is more effective when he can send tailored, private messages. We illustrate with examples from finance, business, politics, and law.
Keywords: Model Persuasion; Belief Formation; Bayesian Inference
JEL Codes: G11; G41; L15; M3
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| competition among persuaders (D72) | models that provide overly good fits (C52) |
| models that provide overly good fits (C52) | do not significantly alter receivers' beliefs (D80) |
| models which better fit the data (C52) | less movement in receivers' beliefs (D83) |
| wrong model can be more compelling than the true model (C52) | belief formation (D83) |