Integrated, pattern-oriented behavior analysis and testing software (BATS) for system dynamics models

Analysis of model behavior is mainly conducted in a pattern-based manner in system dynamics (SD) methodology. In pattern-based evaluation of model outputs, similarity of the overall behavior pattern (e.g. growth-and-decline, S-shaped growth, oscillations) and of specific pattern characteristics (e.g. inflection points, stabilization level, periods, amplitudes) are more important than point-by-point similarity measures such as sum of squared errors. Although some output analysis tools that address this special pattern focus are available, these tools and methods lack usability and are fragmented. In this study, a new analysis environment is proposed, and this environment is developed into an integrated standalone software. This software, namely Behavior Analysis and Testing Software (BATS), integrates a behavior pattern classification algorithm and a selection of statistical methods for analysis of steady-state behavior patterns. Apart from enabling comparison of dynamical patterns with these algorithms and methods, BATS includes structured processes that enable the user to conduct automated hypothesis testing, behavior space exploration, and sensitivity analysis. In its current state, BATS can seamlessly communicate with SD modeling software (e.g. Vensim) as well as other common data sources. The thesis also provides illustrative examples of how BATS can assist the modeler and/or analyst in various tasks in several stages of modeling; model credibility by indirect structure testing, output evaluation by behavior pattern tests, behavior sensitivity analysis, model calibration, policy analysis and design. Considering its pattern focus, user-friendly interface, communication capabilities with data sources, and automation capabilities, BATS can be an important contribution to the analysis toolset of SD methodology.