|
<li>Then, an elicitation method is chosen. Simple methods are the most effective. One effective approach is the quartile method. The expert first decides on a lower and upper limit of possible values for the quantity to be estimated. Because the quartiles are elicited before the median, this minimizes the effects of the “anchoring and adjustment heuristic” <span class="citation">(O’Hagan et al. <a href="#ref-ohagan2006uncertain">2006</a>)</span>, whereby experts tend to anchor their subsequent estimates of quartiles based on their first judgement of the median. Following this, a median value is decided on, and lower and upper quartiles are elicited. The SHELF package has functions to display these quartiles graphically, allowing the expert to adjust them at this stage if necessary. It is important for the expert to confirm that, in his/her judgement, the four partitioned regions that result have equal probability.</li> |
In the section on prior elicitation it states that limits are chosen by the the expert, then:
Because the quartiles are elicited before the median, this minimizes the effects of the “anchoring and adjustment heuristic”
It goes on to say, then medians are estimated then quartiles, so I think the statement above should be "bounds are elicited before the median".