Robust elicitation of imprecise expert knowledge for water infrastructure management
A project lead by SWIP (Sustainable Water Infrastructure Planning) at Eawag is working on the development of probabilistic pipe decay models to support mid- and longterm infrastructure management and rehabilitation planning. Lack of data is one of the diculties this group is facing, but methods exist to collect information of experienced experts with the help of elicitation. To develop an appropriate elicitation procedure was the goal of this thesis, especially comparing the imprecise method of elicitation, which is in development, with the traditional precise method.
First, a brieng was developed for the experts to be informed about the intention and the procedure of the elicitation. The interview itself was conducted with 15 experts in water infrastructure management, eight in the drinking water sector and seven in the wastewater sector. The interviews were designed to elicit pipe lifetime distributions of dierent pipe types. Important psychological and statistical aspects as the basic axioms of probability theory, possible pitfalls, calibrating the expert and give feedback to recognize inconsistencies in the experts answers, are some among the important considerations.
To evaluate if experts have a distinct preference regarding the precise and imprecise method of elicitation, subjective questions were posed to the experts after the elicitations. To additionally see if the choice of the method had an inuence on the time needed for elicitation or the adaptations done for each probability, these were measured and compared. Finally, to recognize if both methods expressed the same information with regard to the lifetime of the pipes, the elicited precise and imprecise estimates for each pipe type were compared. The intention of this work was to evaluate if the imprecise method is equal or even better than the precise method. If this case was conrmed, then the imprecise method could be recommended for future elicitations.
The answers given for the questions posed after the elicitation, state a preference of the imprecise method (8 experts out of 14). Most experts also found it easier (9 out of 14). Interestingly, a majority thought in a precise way (8 out of 14), but this did not seem to infuence the preference of the method as half of those experts prefered the imprecise method. The results showed that the methods had no inuence on time nor on adaptation. The comparison of the methods with regard to the elicited pipe types showed, that the inuence factors of the dierent areas in Switzerland or the experience of the experts were dominating the elicited estimations.
To sum up, the imprecise method can be recommended for future elicitations, as it is considered equal or better than the precise method. More than half of the experts preferred this method. This is also supported by the easiness of doing the elicitation.