ILLUSTRATIVE
PRIOR ELICITATIONElicit a prior with roulette chips
Fitted prior: Beta(2.7, 14.2)mean 16% · SD 9 points 0%10%20%30%40%50% Response rate
1 · The expert places 20 chips where they think the response rate may fall
2 · bayprior fits a Beta prior to the chips, ready to pool or check against data
EXPERT POOLINGPool several experts into one prior
Expert A · mean 10% Expert B · mean 30% Overlap 0.17 · they disagree Linear pool · mean 20% Logarithmic pool · mean 17% 0%10%20%30%40%50%60% Response rate
Two experts hold different views, and bayprior measures how far apart they are
Linear pooling keeps both views: a wide prior that admits either expert may be right
Logarithmic pooling blends them: a narrower prior between the two
ROBUST PRIORSA robust prior reacts when data conflict
Prior Standard: estimate 40%Robust: estimate 79% 0%20%40%60%80%100% Response rate New data
Weight on the informative part
80%<1%
Prior part: 80% informative, 20% vague. Data so far fit the informative part
8 of 10 respond: conflict, so the weight moves to the vague part and the estimate follows the data