Learning from Viral Information

Motivated by social media, we study an equilibrium model of agents interacting with and learning from each other's signals.

Rational agents arrive sequentially, observe a signal (corresponding to a news story) and a sample of predecessors' signals (corresponding to a news feed), and decide which of these signals to endorse.

The observed sample is jointly determined by predecessors' endorsement behavior and a sampling rule (capturing a platform algorithm).

We focus on how often the sampling rule selects more viral (i.e., widely endorsed) signals.

Showing agents viral signals can increase information aggregation, but it can also generate steady states where most endorsed signals are wrong.

These misleading steady states self-perpetuate, as agents who observe wrong signals develop wrong beliefs, and thus rationally continue to endorse them.

We highlight several consequences of our results for social-media platforms.