Using attributes to predict objectives in preference models

Prescriptive decision analysis suggests identifying the fundamental objectives – what the decision maker really cares about – and then constructing a value hierarchy by decomposing these objectives until quantifiable attributes can be identified. In many decision contexts the decision maker is presented with a list of attributes without an opportunity to consider her fundamental objectives. In this paper we explore an approach where a decision maker is given prespecified attributes and then identifies her objectives. She assesses multiattribute models to predict performance levels on each objective and a preference model over these objectives. We use simulation to explore what happens when a decision maker applies this two-step approach to model the relationships between a given set of attributes and her objectives instead of attempting to directly estimate the attribute weights in a choice problem. These simulation results suggest that the explicit consideration of objectives results in less error in expressions of preference than the direct weighting of attributes unless the number of attributes and objectives in the decision context is small.