A simulation approach to investigate factors influencing the cost of omitted objectives in multiattribute models

Abstract

Empirical evidence suggests that decision-makers are ill-equipped to identify all relevant objectives in a decision problem. We examine the effect of an incomplete set of objectives using a Monte Carlo simulation to compare a baseline model to a reduced model incorporating only a subset of objectives. We assess the performance of reduced models varying in the number of objectives, the number of alternatives, the correlations among objectives, and attribute weights. Results suggest that missing objectives will most impact multiattribute models with negative correlations between objectives; similarly, models with equally weighted objectives suffer more than models with unequal weights. Decision problems with more objectives tend to be less impacted by missing objectives, given the same proportion of missing objectives. In contrast, decision problems with more alternatives are more impacted for some performance measures but less on others. However, the variation in model performance due to the number of objectives and alternatives is relatively minor compared to the variation due to the nature of the correlation between objectives.