Replication data for: Natural Hazards and Economic Losses: Why Correcting Sample Selection Matters

Economic losses from natural disasters vary by countries, and it has been hypothesized that institutional, political, and other national conditions and policies all play a role in determining the severity of loss. Many empirical studies for understanding the determinants of disaster losses, however, suffer from endogeneity and selection bias, which can potentially make their results method-dependent. To demonstrate, we investigate the relationship between disaster propensity, wealth, and economic loss from a panel data collected by [Neumayer et al., 2014]. We first demonstrate that the original data is subject to endogeneity and selection bias, reconstruct the dataset, and apply Heckman correction. The bias-corrected estimated impact of disaster propensity changes direction from the original result by [Neumayer et al., 2014] — countries that experience more frequent disasters tend to suffer from greater economic damage, holding everything else equal. We suggest that disaster propensity could be an indicator of vulnerability, or a sign of insufficient prevention and mitigation measures. Although we cannot provide any definitive explanation for the phenomenon, our result shows that correcting selection bias matters when dealing with natural disasters data. For future work, a more sophisticated construction of the latent propensity variable and the application of quantile regression for endogenous selection models could broaden our understanding.自然灾害引发的经济损失存在显著国别差异，既有研究假说指出，制度环境、政治格局及其他国家层面的国情与政策，均会对灾害损失的严重程度产生影响。然而，诸多旨在探明灾害损失决定因素的实证研究，均面临内生性与选择偏误问题，这类问题可能导致研究结果高度依赖所采用的研究方法。为验证这一论断，本文基于Neumayer等人2014年采集的面板数据(Panel Data)，探究灾害发生倾向(disaster propensity)、财富水平与经济损失三者间的关联。本文首先证实原始数据存在内生性与选择偏误，随后对数据集进行重构，并采用赫克曼修正(Heckman Correction)方法开展偏误修正。经偏误修正后的灾害发生倾向估计影响方向，与Neumayer等人2014年的原始研究结果截然相反：在其他条件不变的情况下，灾害发生频率更高的国家往往会承受更为严重的经济损失。本文认为，灾害发生倾向可作为国家灾害脆弱性的表征，或是防灾减灾措施不足的信号。尽管本文无法对该现象给出确定性解释，但研究结果表明，在处理自然灾害相关数据时，修正选择偏误具有重要意义。未来可通过更精细化地构建潜在倾向变量，以及将分位数回归应用于内生性选择模型，进一步拓展相关研究的认知边界。