The Elicitation of Subjective Probabilities: A Laboratory Study in an Accounting Context

A Bayesian method for calibration and aggregation of expert judgement

“…To parameterise our model correctly, we need first to define the generic distributions g e and the corresponding unbiased parameters γ * Xeh ∀X ∈ X and ∀e ∈ E. These choices, at an expert level, define the form of the random draw in the calibration model given in equation ( 4) and the first line in the aggregation step given in equation (7). Suppose, as proposed in the calibration section, that the experts have provided 3 quantiles (0.05,0.50 and 0.95) for each X ∈ X and Y ∈ Y; the parameterisation we choose should preserve all of the information that the experts have provided.…”

“…This process should not be done in isolation from the experts and a feedback process is often employed to playback g e and give an opportunity for refinement. Other methods exist for obtaining g e more directly from experts and many authors have considered the best elicitation methods for expert judgement models ([3], [7], [29], [28], [23]). The decision maker's prior will often be elicited and parameterised in a similar way.…”

UT Biomedical Informatics Lab (BMIL) probability wheel

“…For example, it is more meaningful for an expert or anyone to say that an event has a “75% chance of happening” rather than it will “most likely happen [3]. ” Such quantified opinions are helpful to understand a person’s perception of risk, their general beliefs, and expert estimates involving outcomes given different situations [9] [10] [11]. …”