Uncertainty and Inference in Agent-Based Models

Agent-Based Models (ABMs) can be used to quantify future risks by projecting observable behavior into the future. This can be achieved by simulating a hypothetical longitudinal study based on cross-sectional data and estimating quantities on dynamic risks (e.g., relative hazard). Such an approach, however, requires assessment of the variation of the estimates, which would naturally have a higher variance than would be achieved in a real longitudinal study. We present a methodology that considers rigorous statistical measurements such as standard errors and uncertainty associated with the fact that the analyzed longitudinal data are a projection of the cross-sectional survey. We illustrate the use of our approach in simulated and real studies.