Imagine a highway where everyone drives 5 mph over the speed limit. You match the flow — it feels safe, normal, barely a violation. Now one driver blasts past at 30 over. Most people don't match that speed exactly, but the average creeps up to 10 over, then 15. A few match the outlier. A few go further. Within minutes the whole road has shifted. That's the core mechanism this paper demonstrates in a controlled setting: people don't just copy rule-breaking, they use observed violations as a new floor and sometimes build above it. The committed claim: in a controlled incentivized experiment, exposure to flagrant rule-breaking causes individuals not merely to imitate the violation but to escalate beyond it, and agent-based simulations show this escalation dynamic can collapse collective compliance from a small minority of escalators. This isn't the first paper on social norms contagion, but the escalation-beyond-observation finding and the simulation of collective collapse are the new contributions. The experiment recruited 750 US participants, split roughly evenly between Republicans and Democrats, for an online task. Participants saw a 'shooting star' cross 100 squares and had to pick the brightest — but rewards increased for picking squares further right of the correct answer. After an initial choice, they saw another participant's choice (including party affiliation) and could update. Six rounds total. The key finding: 23% of initial choices already targeted the maximum-reward square. Seeing others follow rules slightly reduced cheating. Seeing small violations slightly increased it. But seeing flagrant violations — someone picking a square far to the right — caused large increases in rule-breaking, with some participants exceeding the violation they observed. The effect was stronger when the observed violator shared the participant's political affiliation. The researchers then fed the observed behavioral parameters into agent-based computer simulations to model how these dynamics would play out in larger populations over time. The result: a minority of escalating individuals — those who consistently exceed observed violations — can cause an entire collective to spiral from general compliance into widespread rule-breaking. About a third of participants consistently followed the rules regardless of what they saw, which acts as a stabilizing force, but the simulations suggest it may not be enough when escalators are present. The integrity picture is mixed. The experiment uses a real incentive structure (money for cheating), which is stronger than hypothetical scenarios, but the task is abstract — no real-world penalties, no reputation at stake, no authority figure watching. Prof Stephen Reicher of St Andrews, not involved in the study, flagged exactly this: the abstraction means real-world relevance remains unproven. He also noted the results actually cut against simple contagion models — influence depended on who the other person was (politically aligned vs. not), which is more nuanced than 'monkey see, monkey do.' The simulation component is a same-team model calibrated from their own experimental data, not an independent validation. The political-alignment moderation is interesting but underexplored. The paper finds stronger effects when the observed violator shares your political identity, consistent with social identity theory, but the experiment wasn't designed to deeply probe this — it's one binary variable (same vs. different party) in a US sample. The mechanism could be trust, shared norms, or simple ingroup favorability. The paper doesn't disentangle these, and a non-US sample might show different identity boundaries entirely. The practical upshot for governance and institutional design: if you want to prevent norm collapse, the critical variable isn't the average level of rule-breaking — it's the visibility of the worst offenders. One publicly visible flagrant violator shifts the reference point for everyone watching, and some watchers will escalate further. The policy implication, which the paper gestures toward but doesn't formalize, is that enforcement should prioritize the most visible and extreme violations, not the average case.
