Information Issues in Model Specification

Abstract

This paper explores the tradeoffs and information issues in model specification, i.e., choosing the appropriate level of detail and precision of decision support models. For many real-time and distributed decision problems, decision-makers face a wide spectrum of information gathering choices that vary in the amount and quality of information as well as their associated costs and delays. Modeling these problems raises several operational and strategic questions including: • how much and what type of information to acquire before making a decision. • what protocol to use for exchanging information among multiple decision-makers. • what organization and information structure (e.g., centralized, coordinated, distributed) to use for effective information sharing and decision-making. We highlight the modeling issues and tradeoffs using examples from production planning, distributed processing, and network routing, and draw upon research in diverse fields, including information economics and game theory, knowledge logic in computer science, and distributed computation, to obtain model specification insights. Our discussions emphasize and illustrate two main model specification themes. First, acquiring additional information often has diminishing returns (in terms of the quality of decisions suggested by the model); therefore, an imprecise model based on partial information might be more appropriate than using a detailed and accurate model that identifies ‘optimal’ decisions. Concepts from information economics and team theory provide a framework for analyzing this tradeoff. We also briefly discuss some heuristic methods to identify effective information gathering strategies. Our second theme applies to systems consisting of several decision-makers who make interdependent decisions. In this context, the decision model for each agent must be based on an understanding of what information to exchange, how frequently, and what protocol to use for exchanging information. We illustrate how complete information sharing among distributed decision-makers might even be impossible because of a possible dead-lock in decision-making. This game-theoretic phenomenon has implications for designing the organization structure, information systems, and communication protocols to support multi-agent decision-making.