A Human Impacts-Driven Framework for Quantifying Disaster-Dependent Social Vulnerability: A Case Study of Hurricane Maria in Puerto Rico

Abstract

We propose a supervised learning approach to statistically quantify the magnitude of extreme events on vulnerable communities using publicly available panel data directly reflective of the different dimensions and manifestations of social suffering. The manifestations along these dimensions include suicides, substance abuse, excess mortality, unemployment, and others. Our modified treatment-effect model allows counterfactual baseline conditions to be posited from which a quantitative multi-faceted measure of social suffering can be determined. This capability is greatly beneficial to policymakers who have to allocate scarce resources to strengthen a geographic region consisting of several administrative units. Our work represents a distinct deviation from the established approach of assessing social vulnerabilities of communities subject to extraordinary events that rely on composite indices (such as SoVI) based on published census data. Recent academic research points out that such approaches are ad hoc, do not consider the dynamics of actual events, and lack formal validation of the variables and models. Further, there is an ongoing challenge for indices to be validated with dependent variables that proxy realizations of vulnerability. We describe the supervised treatment-effect approach in detail and illustrate its applicability using panel data encompassing the 2017 Hurricane Maria event across various municipalities in Puerto Rico. Our statistical modeling methodology stands apart since it explicitly and more realistically captures the social hardships of the actual event as manifested by different published panel data indicators. Our methodology is flexible enough to accommodate individual preferences of different stakeholders in how they assign importance to different manifestations of social hardship.