Suchergebnisse

AbstractThe insufficient attention that too often is paid to the characteristics required of attribute weights prevents multiattribute decision rules from accurately representing preferences. The weighting summation or linear model requires weights to be proportional to the relative value of unit changes in their attribute value functions. Only then will weights accurately reflect the trade‐offs decision makers are willing to make.A number of methods of choosing weights are critiqued from this perspective. Applications of weighting methods in power plant siting are surveyed, and a siting study of western Maryland is presented. Two weighting methods were used, one deriving weights from trade‐offs made by decision makers and the other asking decision makers to choose weights on a scale of 0 to 10. The locations picked by the two methods differ strikingly.

This paper develops a model for estimating and correcting attribute-weighting biases (such as the splitting bias) that result from the use of value trees when structuring value function weight elicitation. The model is based on the conjecture that attribute weights are influenced by tree structure and a subject's use of the "anchor-and-adjust" heuristic, meaning that the subject starts with an equal allocation of weight among attributes in each tree partition and then adjusts the weights to reflect his or her innate preferences. Adjustments tend to be insufficient, resulting in attribute weights that are closer in value to each other than if the anchor-and-adjust heuristic was not employed. Weights corresponding to environmental and economic attributes of electric system expansion alternatives are elicited from employees of an electric utility and used to illustrate the existence and correction of value tree-induced attribute-weighting biases. Two weight sets are elicited from each subject, one using a nonhierarchical assessment and the other using a hierarchical one. The model results support the hypothesis that a bias exists that is consistent with the anchor-and-adjust heuristic. An analysis of rankings of alternatives and value losses caused by using elicited versus model-estimated debiased weight sets is provided.

Bayesian Monte Carlo (BMC) decision analysis adopts a sampling procedure to estimate likelihoods and distributions of outcomes, and then uses that information to calculate the expected performance of alternative strategies, the value of information, and the value of including uncertainty. These decision analysis outputs are therefore subject to sample error. The standard error of each estimate and its bias, if any, can be estimated by the bootstrap procedure. The bootstrap operates by resampling (with replacement) from the original BMC sample, and redoing the decision analysis. Repeating this procedure yields a distribution of decision analysis outputs. The bootstrap approach to estimating the effect of sample error upon BMC analysis is illustrated with a simple value‐of‐information calculation along with an analysis of a proposed control structure for Lake Erie. The examples show that the outputs of BMC decision analysis can have high levels of sample error and bias.

As probability elicitation becomes widely used, methods other than one-on-one interviews are being used to elicit expert probabilities. This paper considers biases that may arise when probabilities are elicited in an online or workbook setting. We develop a prescriptive model in which the elicited probability is a convex combination of the expert's underlying probability with elements of partition dependence and two anchors arising from responses to previous questions ("carryover" bias). Our model, applied to two data sets, allows us to estimate the amount of the various biases in a set of elicited probabilities from experts. We find that both the format of the questions—whether they appear on the same or separate pages/screens—and the ordering of the questions can affect the amount of bias. Our research addresses biases in the presence of multiple anchors and provides guidance on manipulating the availability of anchors. The results demonstrate the persistence of anchoring even with careful questionnaire design; thus, the proposed model-based methods are useful to suggest corrections for the resulting biases.