Metagraphs in hierarchical modeling

When using a decision support system (DSS) containing a large model base, a user has to decide which specific models are relevant for any particular task. When the model base is large and diverse, this task can be quite difficult. Furthermore, the number and variety of models available can be confusing. One way to simplify the user's interaction with the DSS, and facilitate more effective use of the system, is through the use of ''views'' of the model base. A view is an abstraction of a model base that limits attention to relationships between certain variables, and its purpose is to guide the development of DSS software that executes lower-level models in response to higher-level requests for information or analysis. Single views may be of use to individual decision makers who wish to abstract information from a model base, and multiple views may be of use to collections of decision makers who are using a model base to integrate their analyses to make a collective decision. In this paper, we show how representation of models using a graph theoretic structure called a metagraph can facilitate the construction and maintenance of model base views. In particular, we show how useful views can be constructed using a projection operation on a metagraph, and also present conditions under which views can be combined. The concepts developed in the paper are illustrated using an example from life cycle costing.