Required simulated population ratios for valid assessment of shared autonomous vehicles’ impact using agent-based models

The potential impact of shared autonomous vehicles (SAV) on daily lives, urban systems, and transportation has drawn increasing attention, as their widespread adoption is anticipated. Agent-based simulation analysis has emerged as a key tool for exploring SAV operation methods and analyzing their impact. Due to high computational costs, simulations of Mobility-on-Demand services like SAVs, typically use only a fraction of the total population. However, the size of simulated population ratio can affect the analysis results. In this study we evaluated simulated population ratio bias by performing simulations with different simulated population ratios and comparing the results. The analysis focused on a small local city in Japan, where total travel demand was generated through travel demand model based population synthesis. Mode preferences for SAV and other travel options were derived from a mode choice model estimated using data from a previously published stated preference survey. Results revealed that it is necessary to handle travel demand on a scale close to reality when performing a simulation using SAV with dynamic ridesharing (DRS). A 60% simulated population ratio was required to reduce bias to less than 10%. On the other hand, when dynamic ride-sharing is not considered, the simulation bias remained under 10% even with a simulated population ratio as low as 10%. Additionally, we found that for a given population ratio, the magnitude of bias varied across evaluation metrics, with the empty vehicle kilometers traveled (VKT) ratio being particularly sensitive.