Covalidation of Dissimilarly Structured Models

Samuel A. Wright


A methodology is presented which allows comparison between models constructed under different modeling paradigms. Consider the following situation: Two models are constructed to study different aspects of the same system. One model simulates a fleet of aircraft moving a given combination of cargo and passengers from an onload point to an offload point. A second model is a linear programming model that optimizes the aircraft and route selection required for the same scenario. We develop a methodology to structure the comparison between large-scale models such as these. Models that compare favorably using this methodology are deemed covalid. Models that perform similarly under the same input conditions are covalid in a narrow sense. Models that are covalid (in this narrow sense) hold the potential to be used in an iterative fashion to improve the input (and thus, the output) of one another We prove that, under certain regularity conditions, this method of output/input crossflow converges, and if the convergence is to a valid representation of the real-world system, the models are covalid in a wide sense. Further, if one of the models has been independently validated, then we may effect a validation by association of the other model through this process.