How Do Groups Speak

&

How are They Understood?

Discussion

ASSA Meetings

January 2026

Theory Motivation

• With a single expert disclosing evidence we can generically construct fully revealing outcomes via skepticism
• But with disclosure by a group (with heterogeneous interests) we have to make a decision over who to be skeptical of
• This is the opposite result from Cheap Talk, where multiple senders can theoretically lead to full revelation

• From Onuchic and Ramos, the equilibrium outcomes will depend on:
  • preferences of the group members
  • how they reach decisions

Behavioral Motivation

• We know that lab participants do not make full use of skepticism (Jin, Luca and Martin  2021)
  • This is partially mechanical as it's a boundary test, but the result is clear
• What happens with groups where attributing blame is more complicated?
  • Are we more skeptical of participants with more decision power?
  • How does the degree of our skepticism respond to the voting rule?

Really nice Results!

• In ordinal terms we find:
  1. Most skepticism in the individual treatments, though similar level to the Leader treatment
  2. More skepticism in Unilateral than Consensus
  3. Asymmetric updating in the Leader treatment

• Data suggests that cognitive load is related to subjects ability to be skeptical
  • Wonder if there is a way to make this less ad hoc?

Critiques

• A lot of the incentivized rounds do not contain informative data on skepticism, lower opportunity to learn
• Incentives are relatively weak
  • This is almost every paper testing a theory with an expectation over an initial uniform prior
  • How robust are conclusions when we have approximate null predictions for skepticism in several treatments
• Survey questions: not clear what they are incentivizing?
  • elicits modal belief with a uniform eqbm. posterior
• Not clear that the Individual control is right comparison
• With a frequentist design, I wonder if this would converge more clearly to equilibrium?
• Would be good to make clear the contribution/identification over Behnk, Hao and Reuben (2022)

Lots of Interesting Questions Remain

• In larger groups how much decision power is necessary for blame?
• Is skepticism in proportion to disclosure power or does it respond asymmetrically?
• If we have unilateral/consensus + within-group transfers what happens?
• If we switch to a frequentist version of this would it all converge to the equilibrium outcomes?

Thank you!  🙏