An ideal model of a decision is to have all the relevant factors considered with respect to all the alternatives. However, for many decisions with large (even infinite) number of alternatives, the ideal decision model is not possible. To support a decision maker common models are either a detailed model or an aggregate model. A detailed model can consider all factors explicitly for only a subset of the alternatives. An aggregate model can handle all the alternatives but not all of the factors. The results of those selected alternatives in a detailed model are more credible than results produced in an aggregate model because all the factors are explicitly considered. However, there is always a risk of choosing an alternative that would give a lower value than one of those not included in the subset being analyzed. In an aggregate model, some factors cannot be considered explicitly; therefore, the results are not as credible as those of detailed model are. This dissertation examines an approach for incorporating the results from a detailed model into those of an aggregate model to be called an improved-aggregate model. As an approach to build an improved-aggregate model, this dissertation proposes to use Bayesian updating for specified parameters of an aggregate model. However, the major issue in adopting Bayesian inference is that the observed data are not suitable for deriving the likelihood functions of the target parameters; a novel approach is proposed in this dissertation to solve this problem.; This proposed approach has broad applicability in a variety of problem areas including resource allocation, conceptual system design, R&D investment, portfolio management, and screening alternatives. This research is the first known approach to introduce the results of a detailed analysis paradigm into the results of a second through Bayesian updating.
展开▼
