Sensitivity analysis techniques for models of human behavior.

Abstract

Human and social modeling has emerged as an important research area at Sandia National Laboratories due to its potential to improve national defense-related decisionmaking in the presence of uncertainty. To learn about which sensitivity analysis techniques are most suitable for models of human behavior, different promising methods were applied to an example model, tested, and compared. The example model simulates cognitive, behavioral, and social processes and interactions, and involves substantial nonlinearity, uncertainty, and variability. Results showed that some sensitivity analysis methods create similar results, and can thus be considered redundant. However, other methods, such as global methods that consider interactions between inputs, can generate insight not gained from traditional methods.

System dynamics analysis on the effectiveness of vaccination and social mobilization policies for COVID-19 in the United States

“…System dynamics is very useful when creating highly abstract simulations of complex systems, especially social engineering [50,51] through computer simulation technology. SD can be used to get insight about human behavior and contribute to understanding of social systems for behavioral forecasting [52]. This capability is especially useful to model the situations around COVID-19 and generate helpful insight.…”

Application of a variance‐based sensitivity analysis method to the Biomass Scenario Learning Model

“…The literature reports on numerical sensitivity analysis of model parameters, which is considered in this study, in addition to analysis of the sensitivity of models to graphical functions (Eker et al ., ) and the sensitivity of model behavior patterns to changes in model parameters (Hekimoğlu and Barlas, ; Walgrave, ). Literature on numerical sensitivity analysis of model parameters includes application and review of several statistical methods (Bier, ), the use of regression and design of experiments (Kleijnen, ), the use of statistical screening methods (Ford and Flynn, ) for factor prioritization (Taylor et al ., ), and together with design of experiments (Jalili and Ford, ). This analysis contributes to the literature by presenting a case study on the application of variance decomposition‐based sensitivity of a system dynamic model.…”

Sensitivity analysis of graphical functions

“…The studies on multivariate sensitivity analysis focused on the use of statistical tools such as regression analysis and design of experiments (DoE) (Kleijnen, ) or Taguchi and Latin hypercube methods (Clemson et al ., ) for the sampling of input values to investigate the input–output relationships. Bier () provides a list of several statistical sensitivity analysis methods and presents an insightful comparison of these. In addition to numerical sensitivity (changes in the numerical values of output results in response to changes in inputs), several studies focused on behavioral sensitivity (Ozbas et al ., ; Huang et al ., ; Hekimoglu and Barlas, ), which is demonstrated by a change in the behavior pattern of the output in response to changes in inputs.…”