Evolution of Long‐Term Social Vulnerability After Tropical Cyclones in the United States Authors: Lingke Jiang, Robbie M. Parks, Yoshira Ornelas Van Horne, Xiao Wu Published: 2026 ABSTRACT AbstractTropical cyclones pose a substantial threat to the health and welfare of communities across the United States. While existing research has focused primarily on the short‐term impacts of tropical cyclones, typically within days or months after exposure, there remains a gap in understanding their long‐term effects on the socioeconomic and demographic composition of communities–a crucial factor in measuring community resilience and recovery. In this study, we conducted a synthetic control analysis using data from all tropical cyclones that occurred in the United States during 2005–2018. We provide evidence that social vulnerability initially increased after tropical cyclones, but exposed regions recovered over time and became less socially vulnerable in the long term. For exposed regions, the social vulnerability index (SVI) increased by 2.4% (95% confidence interval (CI), 0.7%–4.1%) in the year after tropical cyclones, and persisted at elevated levels for at least 5 years. However, SVI decreased over time and, 12 years after exposure, fell even lower than expected in the absence of a tropical cyclone, decreasing to −1.0% (95% CI, −1.9%–0.0%). Our study suggests that this long‐term trend may reflect post‐cyclone gentrification, where vulnerable communities are displaced and replaced by more affluent, less vulnerable populations. Introduction Hurricanes and tropical storms, collectively known as tropical cyclones, are fast‐spinning storm systems with a low‐pressure center that produce strong winds and spiraling thunderstorms (Anthes, 2016) and can cause extensive infrastructure damage. From 1980 to 2024, the costs related to tropical cyclones in the United States (US) are estimated to have exceeded $1.3 trillion (NOAA National Centers for Environmental Information, 2024). The risk of tropical cyclones for nearly 127 million people living in coastal communities in the US is further aggravated by climate change, which has been shown to increase hurricane frequency for the Gulf and Lower East Coast regions (Balaguru et al., 2023). The immediate effects of tropical cyclones on individual health are widely documented and have been associated with physical injuries, mental stress, and increased risk of diseases (Huang et al., 2024; Parks et al., 2022, 2023; Shultz et al., 2005). A 2022 study revealed that death rates increased in the month following tropical cyclones for several causes, including cardiovascular, neuropsychiatric, and respiratory diseases (Parks et al., 2022). A following large‐scale mortality evaluation for all tropical cyclones in the contiguous US during 1930 and 2015 found a robust increase in excess mortality that persists for up to 15 years (Young & Hsiang, 2024). Housing conditions may worsen after tropical cyclones, and access to healthcare services and medication supplies may experience severe disruptions (Shultz et al., 2018). The existing literature has documented economic impacts based on data from community‐level income and employment (Coffman & Noy, 2012; Groen et al., 2020; Yabe et al., 2020). Collectively, tropical cyclones create devastating and wide‐ranging impacts on society, affecting both individuals and communities. The impacts of tropical cyclone exposures vary greatly between communities, revealing underlying inequality (Parks et al., 2022). Historically marginalized communities not only suffer greater impacts after tropical cyclones, but also receive unequal benefits from recovery programs (Burrows et al., 2023; Reid, 2013). In the US, communities with predominantly black or low‐income populations have historically faced greater challenges in securing federal aid and housing insurance following disasters (Greenberg, 2014). Post‐disaster recovery efforts following major hurricanes often deepen urban inequality through a process known as “crisis‐driven urbanization” (Greenberg, 2014). Following Hurricanes Katrina (2005) and Sandy (2012), narrative evidence shows that redevelopment in affected areas tends to be highly uneven, with fortified wealthy neighborhoods, catalyzed gentrification and displacement in low‐income areas, and a pattern where disasters become opportunities for powerful coalitions to pursue redevelopment projects (Aune et al., 2020; Van Holm & Wyczalkowski, 2019). Furthermore, the housing crisis in New Orleans after Katrina saw massive displacement and a severe shortage of rental unit replacement (Comerio, 2014). Paradoxically, this inequitable rebuilding may be fueled by the very aid designed to mitigate disaster impact. While most storms depress local economies, the largest hurricanes often don't show the same negative impact on income, likely because they trigger massive federal aid packages that, while stabilizing the economy, can also finance this uneven redevelopment (Callahan et al., 2024). Understanding the long‐term effects of tropical cyclone exposures on the dynamics of social vulnerability in affected communities is crucial to disaster management, public health, and social welfare. However, our knowledge of how socioeconomic and demographic compositions evolve in these affected communities over time remains limited (Ghosh et al., 2022), for two main reasons. First, previous studies have focused mainly on the relatively short‐term impacts of tropical cyclones (Bell et al., 2023; Ghosh et al., 2022, 2022; Lawrence et al., 2019; Parks et al., 2021, 2022, 2023; Weinberger et al., 2024). Second, a lack of harmonized longitudinal data and suitable analytical methods has limited most studies to assessing single tropical cyclone events rather than the cumulative effects of multiple disasters (Coffman & Noy, 2012; Groen et al., 2020; Karbownik & Wray, 2019). Here, we empirically quantify the impacts of all tropical cyclone exposures that occurred during 2005–2018 in the contiguous US on the following multi‐year changes of social vulnerability in affected regions, as measured by the Centers for Disease Control and Prevention (CDC) social vulnerability index (SVI), a comprehensive metric derived from numerous socioeconomic and demographic factors to characterize community social vulnerability at county‐level (B. E. Flanagan et al., 2011). Collecting multi‐year tropical cyclones, meteorological, socioeconomic, and demographic data and applying an extended synthetic control approach, we assessed how social vulnerability in counties affected by tropical cyclones would have changed if they had not been exposed to tropical cyclones and then compared these counterfactual changes with their actual evolution. To validate the significance of the effects, we conducted placebo tests to rule out the possibility that the observed effect could have occurred by chance alone (Abadie et al., 2010). We made all data and statistical modeling procedures publicly available. Materials and Methods Data Social vulnerability refers to the potential susceptibility of communities to negative effects caused by external stresses on human health, which helps to identify communities that need support before, during, or after disasters. The SVI provided by the Centers for Disease Control and Prevention/Agency for Toxic Substances and Disease Registry (CDC/ATSDR) is a comprehensive measure of community social vulnerability based on numerous US census variables from both the decennial US Census and the 5‐year American Community Survey (ACS), as detailed in Table 1 (Centers for Disease Control and Prevention Agency for Toxic Substances Disease Registry, 2025). These variables were categorized into four themes: Socioeconomic Status, Household Characteristics, Racial and Ethnic Minority Status, Housing Type and Transportation. The overall SVI is an additive index constructed by summing the percentile ranking of these four themes, which were then ranked against all other counties to generate the final percentile ranking. SVI is a percentile ranking that ranges from 0 to 1, with values closer to 0 indicating lower vulnerability and those closer to 1 indicating higher vulnerability. The methods to construct the SVI have been detailed in the literature (B. E. Flanagan et al., 2011). Table 1Compositions of Current Overall Social Vulnerability IndexSocioeconomic statusBelow 150% povertyUnemployedHousing cost burdenNo high school diplomaNo health insuranceHousehold characteristicsAged 65 years and olderAged 17 years and youngerCivilian with a disabilitySingle‐parent householdsEnglish language proficiencyRacial and ethnic minority statusHispanic or Latino (of any race)Black or African American, not Hispanic or LatinoAsian, not Hispanic or LatinoAmerican Indian or Alaska Native, not Hispanic or LatinoNative Hawaiian or Pacific Islander, not Hispanic or LatinoTwo or more races, not Hispanic or LatinoOther races, not Hispanic or LatinoHousing type and transportationMulti‐unit structuresMobile homesCrowdingNo vehicleGroup quarters Compositions of Current Overall Social Vulnerability Index Data on wind exposure in the contiguous US were obtained from packages hurricaneexposure (version 0.1.1) and hurricaneexposuredata (version 0.1.0) in R, with full coverage of space and time during our study period (Anderson, Schumacher, et al., 2020; Anderson, Yan, et al., 2020). Briefly, the data were generated by an exhaustive assessment of tropical cyclones recorded in the Atlantic Hurricane Database (HURDAT2) based on wind field modeling and validation against observations from weather stations (Landsea & Franklin, 2013). In line with literature (Parks et al., 2023), a county is considered exposed to a tropical cyclone in a given year if the maximum sustained wind in that county reached or exceeded the gale force (≥34 knots, or 63 km/hr) on the Beaufort scale on any given day of that year when the tropical cyclone was at the point of closest approach to that county. Given that the spatial scale of tropical cyclones substantially exceeds that of individual counties (tropical cyclone gale‐force wind fields typically extend 150–300 km from the storm center whereas US county size averages around 2,900 km2) (Wang & Toumi, 2022), variations in wind speed are negligible across the average county size. Tropical cyclone exposure data are available for full space and time coverage during 1995–2018, but SVI data are available only for 2000, 2010, and then biennially from 2014 to 2020. This highlights the necessity to incorporate various demographic covariates for our synthetic control study, as a long trajectory of pre‐exposure outcomes is not available in contrast to the settings of the traditional synthetic control method (Botosaru & Ferman, 2019; Kaul et al., 2015). Furthermore, there are two missing entries for SVI during the study period, for Clearfield County (FIPS code 42033), Pennsylvania, in 2014, and for Rio Arriba County (FIPS code 35039), New Mexico, in 2018. We excluded these two counties from our analysis because during our study period, Clearfield County experienced only one yearly exposure to tropical cyclones, and Rio Arriba County encountered zero exposure. We obtained data on temperature and precipitation across the US for 1995–2018 from the Gridmet data set, with full time and space coverage during the pre‐exposed period of this study (Abatzoglou, 2013). GridMET data were developed by spatially interpolating meteorological observations from a variety of observatory networks using the Parameter‐elevation Regressions on Independent Slopes Model (PRISM) (Abatzoglou, 2013; Daly et al., 2008). We calculated spatially averaged monthly values of each variable across grid cells (4‐km resolution) and converted the gridded data to the county level. The summer months were defined from June to August. We compiled all demographic and socioeconomic raw data at the county level from several public sources. Data that reflect basic demographic characteristics were obtained from CDC Wonder (United States Department of Health and Human Services (US DHHS), Centers for Disease Control and Prevention (CDC), National Center for Health Statistics (NCHS), 2021), including Percentage Black Population, Percentage Hispanic Population, Male/Female Ratio, Population Size, with full time and space coverage over the pre‐exposed period. Population density was calculated from the Population Size. Per Capita Income was obtained from the Bureau of Economic Analysis (BEA) (U.S. Bureau of Economic Analysis, 2022), and was adjusted for inflation using the year 2000 as baseline and July values derived from the Consumer Price Index (CPI) inflation calculator, according to the US Bureau of Labor Statistics (BLS) (U.S. Bureau of Labor Statistics, 2022). The Percentage Below Poverty Threshold was obtained as the Small Area Income and Poverty Estimates (SAIPE) from the US Census Bureau (U.S. Census Bureau, 2021); and the percentage of over age 25 years with a high‐school diploma was obtained from the US Department of Agriculture (USDA) as educational attainment (U.S. Department of Agriculture, 2025). For all outcomes, exposures, and covariates, associated county identification codes were made consistent, mutually exclusive, and collectively exhaustive over time. Creating Synthetic Controls First, we constructed the synthetic control region from a weighted set of counties in the donor pool (i.e., the full set of unexposed counties) that most closely resembled the exposed region in pre‐exposure covariates over time. Briefly, the synthetic control method is a quasi‐experimental design that aims to construct a weighted combination of unexposed units as controls to which the exposed unit is compared. This comparison is used to estimate what would have happened to the exposed unit if it had not been exposed. Here, we formed the synthetic control region that had historical trajectories of tropical cyclone exposure and SVI, as well as meteorological, demographic, and socioeconomic profiles similar to those of the exposed counties. We divided the time series of each unit into three periods (shown in Figure 1). Specifically, (a) the pre‐exposure period serves the source of calculating weights for each unexposed unit to create synthetic control units that mirrors the exposed unit as closely as possible; (b) the exposure period defines the exposure status (i.e., the exposed unit is the unit experiencing tropical cyclones in this period); (c) the evaluation period forms the outcome comparisons between the exposed unit and the synthetic control units to estimate what would have happened to the exposed unit if it had not been exposed. To ensure sufficient comparability, we balanced over 10 years of pre‐exposure covariate trajectories between the exposed regions and their synthetic controls. Table 2 shows the comparison of the pre‐exposure characteristics of the actual exposed region with those of the synthetic control regions in 2005 and 2018 (the first and last years of exposure). Overview of the quasi‐experimental design. Exposed and unexposed units are defined by tropical cyclone exposure within the year period. We create synthetic controls as a weighted set of unexposed counties that maintain similar trajectories on pre‐exposure social vulnerability index (SVI), tropical cyclone exposure status, and other meteorological and socioeconomic covariates. These synthetic controls are then used as counterfactuals in the evaluation period to estimate the effects of tropical cyclones on SVI. Table 2Covariate Means Prior to Exposures in 2005 and 2018Covariate2005 exposure cohort2018 exposure cohortExposed (n = 217)Unexposed (n = 2,998)Synthetic controlExposed (n = 303)Unexposed (n = 2,912)Synthetic controlPast TC exposure, mean0.260.040.260.210.030.19Baseline SVI, percentile0.770.480.770.750.470.74Summer temperature, °C27.022.826.826.323.026.0Summer precipitation, mm/day4.72.94.64.33.04.3High school graduate, %71.377.171.375.681.075.8Black, %27.87.727.433.26.832.9Hispanic, %5.66.05.64.37.74.7Below poverty, %18.613.918.519.214.919.3Per capita income, USD21,72423,63621,69323,64326,15423,680Male‐to‐female ratio1:1.031:1.021:1.031:1.021:1.011:1.02Age <15 years, %21.520.821.519.719.819.8Age ≥65 years, %13.815.013.914.616.014.7Population density, per km2 149251149221259249Population size115,03287,795117,09275,48697,44679,064 Note. All variables except for SVI were averaged from 1995 to the year prior to the exposure year. Per‐capita income was adjusted for inflation using the year 2000 as baseline and July values. Covariate Means Prior to Exposures in 2005 and 2018 Note. All variables except for SVI were averaged from 1995 to the year prior to the exposure year. Per‐capita income was adjusted for inflation using the year 2000 as baseline and July values. The validity of the synthetic control approach depends on the synthetic control region being representative of the exposed region (Abadie, 2021). The key assumption underlying the causal validity of this synthetic control design is that had the exposed regions not been exposed to tropical cyclones for that particular year, their future SVI would have matched that of the synthetic control region, and as a result, any future divergence in SVI for the exposed versus synthetic control regions was attributed to tropical cyclone exposure of that year. The validity of this assumption depends on two factors: first, the availability of adequate unexposed units to build a synthetic control region capable of consistently tracking the historical trajectories of the exposed region before exposure (Abadie, 2021); and second, the absence of unobserved confounders that could cause future divergence in SVI between exposed and synthetic control regions. For the first point, we assessed the quality of the synthetic control design via covariate balance to check the degree to which the distributions of pre‐exposure covariate trajectories are similar across exposure and synthetic control regions (Ho et al., 2007). Figure 2 shows the covariate balance checks throughout the pre‐exposure period for all variables and all exposure years, using mean absolute standardized mean difference (ASMD) as a metric (Parast et al., 2020). Except for three exposure years out of the 14‐year exposure period, the mean ASMD was well below 0.1, rendering our synthetic control approach effective in achieving covariate balance across a wide variety of covariates. Further sensitivity analyses also found that the main results remained almost unchanged even if we excluded these three less balanced years (detailed in Supporting Information S1). This is because only units that are similar in both observed and unobserved confounding factors are expected to exhibit comparable outcome trajectories over extended periods (Abadie, 2021). Adjusting for historical SVI trajectories allows us to account for such types of unmeasured confounders, which represents a major advantage of the synthetic control method. Mean absolute standardized mean difference for all covariates for exposure years. Estimating Causal Effect First, we formally define the causal effect of interest. Let N denote the total number of study units (i.e., counties) and T denote the total number of time points (i.e., years) in the time series. Define the year of exposure as T0, where 1