An Information Theoretic Approach to Computer Simulation Sensitivity Analysis

Abstract

In this paper, statistical information theory-based procedures are applied to sensitivity analysis in computer simulation. Information theory, through use of the conditional entropy functional, provides a non- parametric approach to qualitatively assessing the sensitivity of the distributional relationships of the input and output processes of a simulation model. Since the conditional entropy functional quantifies the amount of uncertainty in the distribution of a set of random variables, it can be used as the basis for a methodology to assess the relative strengths of the statistical dependencies among the input/output processes. The application of information theory in this paper focuses on assessing the uncertainty in the simulation output processes attributable to the simulation input processes. This approach to sensitivity analysis is illustrated by an example.