<p>Prior studies that sought to highlight the benefit of seismic risk mitigation strategies (e.g., retrofits, building code change) during real earthquakes have largely relied on anecdotal evidence. This paper presents a data-driven methodology for quantifying the regional-scale benefit of seismic interventions that leverages the methods, tools, and language of causal inference. From a causal perspective, the seismic intervention is defined as the treatment, and the metric used to quantify performance at scale is the outcome. In addition to the intervention, the additional features that are expected to influence the performance are defined as covariates. Using causal inference, we are able to isolate the effect (or benefit) of the intervention (treatment) on the overall performance (outcome) while controlling for influential covariates or confounders. The framework is implemented to quantify the realized and unrealized inventory-scale benefit of cripple retrofit during the 2014 South Napa earthquake. The building retrofit state and repair costs due to earthquake damage are defined as the treatment and outcome, respectively. The covariates considered are the building and site parameters (e.g., number of stories, age, site shaking intensity during the earthquake) that, in addition to the retrofit state, are expected to influence the level of damage and resulting repair costs. The double machine learning algorithm is adopted as the estimation strategy because of its ability to deal with high-dimensional covariates and confounders. The results of the causal analysis show that the retrofit reduced repair costs in pre-1980 residential buildings with cripple walls by an average of $13,697. The total realized benefit considering all retrofitted buildings is approximately $3.4 million. In other words, this would have been the added inventory level repair cost if these buildings were not retrofitted. Moreover, this reduction in repair costs would have increased by a factor of approximately four (i.e., to <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11069_2025_7531_Article_IEq1.gif" Format="GIF" Height="10" Rendition="HTML" Resolution="72" Type="Linedraw" Width="17" /> </InlineMediaObject> <EquationSource Format="TEX">$\approx$</EquationSource> <EquationSource Format="MATHML"><math> <mo>≈</mo> </math></EquationSource> </InlineEquation> $14.5 million) had the entire inventory been retrofitted prior to the earthquake.</p>

Quantifying the realized and unrealized benefits of seismic interventions using causal inference

期刊论文
出版模式：

开放获取
发表日期：

2025年7月28日

查看全文

Henry Burton,
Sebastian Galicia Madero,
Chenhao Wu,
Rashid Shams,
Chukwuebuka Nweke

摘要

Prior studies that sought to highlight the benefit of seismic risk mitigation strategies (e.g., retrofits, building code change) during real earthquakes have largely relied on anecdotal evidence. This paper presents a data-driven methodology for quantifying the regional-scale benefit of seismic interventions that leverages the methods, tools, and language of causal inference. From a causal perspective, the seismic intervention is defined as the treatment, and the metric used to quantify performance at scale is the outcome. In addition to the intervention, the additional features that are expected to influence the performance are defined as covariates. Using causal inference, we are able to isolate the effect (or benefit) of the intervention (treatment) on the overall performance (outcome) while controlling for influential covariates or confounders. The framework is implemented to quantify the realized and unrealized inventory-scale benefit of cripple retrofit during the 2014 South Napa earthquake. The building retrofit state and repair costs due to earthquake damage are defined as the treatment and outcome, respectively. The covariates considered are the building and site parameters (e.g., number of stories, age, site shaking intensity during the earthquake) that, in addition to the retrofit state, are expected to influence the level of damage and resulting repair costs. The double machine learning algorithm is adopted as the estimation strategy because of its ability to deal with high-dimensional covariates and confounders. The results of the causal analysis show that the retrofit reduced repair costs in pre-1980 residential buildings with cripple walls by an average of $13,697. The total realized benefit considering all retrofitted buildings is approximately $3.4 million. In other words, this would have been the added inventory level repair cost if these buildings were not retrofitted. Moreover, this reduction in repair costs would have increased by a factor of approximately four (i.e., to $\approx$ $\approx$ $14.5 million) had the entire inventory been retrofitted prior to the earthquake.