Statistical Individuals and Simulated Individuals: Analysing Agent-Based Demographic Models with Gaussian Process Emulators

Event-history demography is concerned with statistical individuals, whose life courses can be inferred from empirical information. In contrast, agent-based models study simulated individuals, for whom certain behavioural rules are assumed. We wish to bring these two approaches closer together by proposing a method to analyse the rule-based outcomes statistically. We present a Semi-Artificial Model of Population (SAMP), which augments the Wedding Ring agent-based model of partnership formation by statistical data on natural population change in the United Kingdom. We utilise a Gaussian process emulator - a statistical model of the SAMP - to analyse the impact of selected parameters on two key model outputs: population size and share of agents with partners. Emulators permit a statistical analysis of model properties and help select plausible parameter values, despite the non-linearities and feedback loops present in agent-based models. A sensitivity analysis is also attempted, aiming to assess the relative importance of different parameters. The resulting multi-state model of population dynamics has an enhanced predictive capacity, but with some trade-offs between the outputs considered. The proposed methods allow for generating coherent, multi-level agent-based scenarios aligned with selected aspects of the demographic reality.