We study a 2004 program designed to motivate Chinese bureaucrats to reduce accidental deaths. Each province received a set of "death ceilings" that, if exceeded, would impede government officials' promotions. For each category of accidental deaths, we observe a sharp discontinuity in reported deaths at the ceiling, suggestive of manipulation. Provinces with safety incentives for municipal officials experienced larger declines in accidental deaths, suggesting complementarities between incentives at different levels of government. While realized accidental deaths predict the following year's ceiling, we observe no evidence that provinces manipulate deaths upward to avoid ratchet effects in the setting of death ceilings.