An Internal Validation Assessment of Scale Across Social Vulnerability Index Model Structures

Proactive and equitable planning for natural hazards is vital, as these events can cause mass destruction and severely impact livelihoods. To aid hazard preparedness decision-making, organizations can utilize tools like a social vulnerability index (SVI), developed to identify vulnerable populations to ensure that those with inherent social inequities are considered in planning. However, SVI construction involves various approaches that introduce epistemic uncertainty, potentially affecting resulting decisions. While progress has been made in understanding how construction processes affect index results, the spatial elements of SVI models are underexplored, with conflicting views on the influence of scale selection. This study addresses this gap by evaluating how changes in the selection of scalar properties (areal units and geographic boundaries) and indicator selection impact SVI ranks for two indices, the Center for Disease Control SVI (CDC SVI) and the University of South Carolina Hazards Vulnerability and Resilience Institute SVI (HVRI SoVI). We examine these changes across three model structures: hierarchical with z-score standardization, hierarchical with percentile ranking normalization, and inductive with z-score standardization, employing an uncertainty and sensitivity analysis. When altering scalar and indicator properties, we found the inductive model less robust than hierarchical models. We also observed indicator selection as the primary driver of variability in SVI ranks across all model structures. However, we found significant yet mixed effects of scale selection and interaction effects on variability in SVI ranks. Our findings emphasize the critical role of scale selection in shaping index outcomes and underscore the need for critical evaluation in SVI creation to advance equitable hazard management.

针对自然灾害开展主动且公平的规划至关重要，此类灾害事件往往会造成大规模破坏，并严重影响民众生计。为辅助灾害备灾决策，相关机构可借助社会脆弱性指数（Social Vulnerability Index，SVI）这类工具——该指数旨在识别脆弱群体，确保规划过程中纳入存在固有社会不平等的人群。然而，SVI的构建存在多种不同方法，这些方法会引入认知不确定性，进而可能对最终决策产生影响。尽管学界在探究构建流程对指数结果的影响方面已取得进展，但SVI模型的空间维度仍未得到充分探索，且针对尺度选择的影响存在相互矛盾的观点。本研究针对这一研究空白展开探究，通过评估尺度属性（空间单元与地理边界）选取及指标选取的变化，对两类SVI排名产生的影响：这两类指数分别为美国疾病控制与预防中心SVI（CDC SVI）以及南卡罗来纳大学灾害脆弱性与韧性研究所SVI（HVRI SoVI）。本研究采用不确定性与敏感性分析方法，针对三类模型结构开展上述变化的相关检验：基于z-score标准化的层级模型、基于百分位排名归一化的层级模型，以及基于z-score标准化的归纳式模型。在改变尺度与指标属性的实验中，我们发现归纳式模型的鲁棒性弱于层级模型。同时我们还发现，在所有模型结构中，指标选取都是导致SVI排名出现差异的主要因素。不过，我们发现尺度选择及其交互效应会对SVI排名的差异产生显著且复杂的影响。本研究结果凸显了尺度选择在塑造指数结果中的关键作用，并强调在SVI构建过程中开展严谨评估的必要性，以推动公平的灾害管理工作。