One-sided matching under strategic reporting Insights from Amsterdam school choice

One-sided matching systems assign people to indivisible resources such as school seats on the basis of submitted preferences. Designers of such systems face a well-known tension between efficiency and strategyproofness: mechanisms that protect truthful reporting can produce poor matches, while efficient alternatives may invite strategic misreporting of preferences. School choice systems have largely resolved this tension in favour of strategyproofness, but Amsterdam's experience with Random Serial Dictatorship (RSD), where some students were placed at schools far down their lists, shows that this choice can be costly.
This thesis asks how one-sided matching systems should be evaluated and designed when attempts to improve allocations incentivize strategic misreporting. This question is addressed from formal, institutional, behavioural, and computational perspectives, using Amsterdam secondary school choice as a motivating case. The thesis first studies rank-minimizing matching as an efficient alternative to RSD, finds a misreporting strategy that requires only first-choice popularity information, and shows that misreporting can create unfair outcomes. It then analyzes Amsterdam's placement guarantee and shows that an institutional safeguard layered on a strategyproof mechanism can also create incentives to misreport. Next, the risk of misreporting is empirically evaluated through a survey of 140 parents in Amsterdam, finding that transparency about risks of misreporting can deter rather than enable manipulation, and that parents in competitive tracks are more risk-averse. Learned matching methods using Deep Q-Networks and Generative Flow Networks are then shown to reach efficiency–robustness tradeoffs unavailable to classical mechanisms. Finally, the thesis proposes a framework for responsible civic AI research, arguing that civic engagement is intertwined with impactful, interdisciplinary research.