Sample Selection in the Estimation of Air Bag and Seat Belt Effectiveness
针对安全带和气囊效果估计中因仅收集致命事故数据导致的样本选择偏差，本文提出将样本限制在有人死于其他车辆的事故中，从而得到更一致的估计，发现安全带比以往更有效，而气囊效果则低于普遍认知。
Measurement of seat belt and air bag effectiveness is complicated by the fact that systematic data are collected only for crashes in which a fatality occurs. These data suffer from sample selection since seat belt and air bag usage influences survival rates which in turn determine whether a crash is included in the sample. Past researchers either ignored sample selection or adopted indirect estimation methods subject to other important biases. We propose a simple, but novel, solution to the selection problem: limiting the sample to crashes in which someone in a different vehicle dies. Under relatively weak conditions, consistent estimates can be obtained from this restricted sample. Empirically, we find seat belts to be more effective in saving lives than most previous estimates. Air bags, however, appear to be less effective than generally thought. If our coefficients can be generalized to all crashes, the cost per life saved with seat belts is approximately $30, 000, compared to $1.6 million for air bags.