A new method to elicit probability distribution beliefs

  • Thomas' website
  • A new method to elicit probability distribution beliefs
  • Abstract:
    I introduce a method to elicit belief distributions. With this method an elicitee's payoff not only depends on the realised state and the elicited distribution, but also on a random two-level partitioning of the state-space. As the number of relevant states or 'bins' increases, the set of best-response distributions converges asymptotically to the elicitee's true belief. I compare the introduced method with both the well known "Quadratic scoring rule" (Brier 1950 and e.g. Selten 1998), an the "Karni method" (Karni, 2009) and argue that the introduced method gives stronger incentives to elicitees in situations where there are many 'bins' (think for example of eliciting an expert's belief distribution over possible oil prices in cents on a future date).