Robustness testing of model based multiple criteria decisions: fundamentals and applications

Robustness or insensitivity is a desirable property of decisions; however, most texts on robustness and/or sensitivity analysis do not define it precisely. A broad literature in this field concentrates on robust design or decisions (including robust optimization). This paper focuses on robustness testing, that is, checking whether a design has actually resulted in robust properties of the system if some of basic assumptions are changed. We propose a general framework of such robustness testing and show that robustness is a property of the relation between three (classes of) models: a basic model of the decision situation, a second model of possible perturbations of the first model, and a model of implementation of the decision, optionally taking into account some measurements of the impact of perturbations. Typical approaches to robustness or sensitivity analysis assume tacitly that the first two models can be combined and analyze parameters deviations in one combined model. However, the role of the first two models can be asymmetric if some optimization of the decision is performed on the first model. We extend this framework, intended originally for single criteria (scalar) optimization to multiple criteria (vector) optimization. The proposed approach is illustrated by diverse examples