Causal Recreation Demand Estimation with Cellphone Mobility Data
研究提出一种结合经验贝叶斯和随机系数logit模型的方法，解决休闲需求模型中零市场份额问题，并利用手机数据评估2021年亨廷顿海滩漏油事件导致的临时关闭对游客福利的因果影响。
<h3>Abstract</h3> Accurately estimating the welfare impacts of environmental changes in recreation demand modeling requires robust causal inference methods. However, a persistent challenge remains as the zero market share issue restricts the causal inference application to broader regions and longer periods. To address this, we integrate empirical Bayes posterior mean estimation into a two-step random coefficient logit model, ensuring that sites with low or zero visitation are properly incorporated without distorting demand estimates. We apply this framework to the 2021 Huntington Beach oil spill, using high-frequency cellphone data to track changes in beach visits. By combining the Synthetic Difference-in-Differences (SDID) approach with empirical Bayes-adjusted market shares, we examine the causal effects of temporary beach closures on visitor welfare. Our findings reveal that Huntington Beach experienced the largest and most prolonged welfare loss, with an estimated aggregate loss of $1.0 million and weekly losses of $83 thousand persisting beyond the initial closure. In contrast, Newport Beach and Laguna Beach exhibited faster recoveries. This study advances recreation demand modeling by refining demand estimation for low-visit sites and strengthening causal inference techniques for environmental disruptions, ultimately providing a more reliable framework for assessing the economic costs of beach closures and other environmental policy interventions.