Improving Microsimulation in Policy Analysis: Using Social Behavioral Factors & Multiple Predictands to Enhance Behavioral-Dynamic Microsimulation in Public Policy

Abstract

In this dissertation, I offer methodological modifications to some of the current microsimulation model (MSM) practices in policy analysis including: 1) enhancing the approaches to estimating the behavior of a reacting population by adding social behavioral factors to the standard economic behavioral factors commonly used, and 2) as a design feature of the dynamic (multi-time period) aspects of the model, applying a multiple predictand process employed in tropical cyclone models to improve forecast-to-forecast continuity. Including social behavioral factors to complement the economic behavioral factors of the behavioral logic model could account for human behavior that is not otherwise captured at the foundation of common public policy microsimulations. Likewise, a multiple predictand dynamic process looks at the aging of a microsimulation model holistically. It gives preference to key variables when generating the initial forecast equation then acknowledges that, as the forecast intervals advance to new cycles, variables not previously considered significant may become significant and therefore should be included for the benefit of the late cycles. The objective is to design a microsimulation model that satisfies four conditions: The model should 1) estimate the core economic behaviors of the reacting population, 2) estimate the interacting social behaviors of the reacting population with a baseline for comparison, 3) generate dynamic cycle estimates for the years following the initial cycle, and 4) allow analysis on the degree to which the MSM predicts nongroup health insurance take-up following the implementation of the Affordable Care Act (ACA). The resulting structure to satisfy these four conditions we will call the Dynamic Choice Hybrid Economic Social Simulation (Dynamic-CHESS) model. The findings in this analysis suggest that adding social behavioral factors improves the accuracy of the model, with the factors of asymmetric information with respect to language and health insurance literacy potentially diverting the largest share of the offered population. The version of the Dynamic-CHESS model with social behavioral factors has a higher classification rate than the version without them, and it ultimately produces a nongroup health insurance take-up estimate that is consistently more in line with the actuals over the 3-year research period than its counterpart.