Abstract for: Addressing parameter uncertainty in SD models with fit-to-history and Monte-Carlo sensitivity methods
We present a practical guide, including a step-by-step flowchart, for establishing uncertainty intervals for key model outcomes in the face of uncertain parameters. The process starts with Powell optimization (e.g., using VensimTM) to find a set of uncertain parameters (the “optimum” parameter set or OPS) that minimize the model fitness error relative to available reference behavior data. Optimization also helps refine parameter uncertainty ranges. Next, Markov Chain Monte Carlo (MCMC) or conventional Monte Carlo (MC) randomization creates a sample of parameter sets that fit the reference behavior data nearly as well as the OPS. Under the MC method, the entire parameter space is explored broadly (millions of runs), and the results are sorted to select qualified parameter sets (QPS) based on model-fitness. We analyze the statistical properties of the QPS parameter distributions to ensure their centrality relative to the uncertainty ranges. The full set of QPS outputs are graphed against the reference behavior data. Alternative policies and scenarios are run using the OPS and all QPS, and confidence intervals determined for model outcomes. We demonstrate the method with a substantial model currently in the publication process, and we discuss how such analyses can be communicated for use by policy/decision makers.