- Access to this full-text is provided by Springer Nature.
- Learn more
Download available
Content available from Scientific Reports
This content is subject to copyright. Terms and conditions apply.
1
Vol.:(0123456789)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports
Social capital’s impact on COVID‑19
outcomes at local levels
Timothy Fraser 1*, Courtney Page‑Tan 2 & Daniel P. Aldrich 3
Over the past thirty years, disaster scholars have highlighted that communities with stronger social
infrastructure—including social ties that enable trust, mutual aid, and collective action—tend to
respond to and recover better from crises. However, comprehensive measurements of social capital
across communities have been rare. This study adapts Kyne and Aldrich’s (Risk Hazards Crisis Public
Policy 11, 61–86, 2020) county‑level social capital index to the census‑tract level, generating social
capital indices from 2011 to 2018 at the census‑tract, zipcode, and county subdivision levels. To
demonstrate their usefulness to disaster planners, public health experts, and local ocials, we paired
these with the CDC’s Social Vulnerability Index to predict the incidence of COVID‑19 in case studies in
Massachusetts, Wisconsin, Illinois, and New York City. We found that social capital predicted 41–49%
of the variation in COVID‑19 outbreaks, and up to 90% with controls in specic cases, highlighting its
power as diagnostic and predictive tools for combating the spread of COVID.
Why do some communities experience greater outbreaks of COVID-19? As of the time of writing on February
1, 2022, COVID-19 has led to at least 74.3 million cases and 884,300 known deaths in the US, and over 5.67
million deaths worldwide1,2. ese outcomes represent major challenges to communities’ resilience, referring
to residents’ “capacity to adapt existing resources and skills to new situations and operating conditions”3. Past
studies highlighted that pandemic outcomes, like many disaster resilience outcomes, depend on more than the
capacity of governments and health care workers to respond4,5 and the social vulnerability of residents by race,
class, age, and gender6–9. However, crisis outcomes also correlate with the strength of community resources such
as social infrastructure accessible to members10–13. Social capital—the social ties that enable trust, reciprocity, and
collective action14—serves as a key resource residents can draw upon before, during, and aer crisis to ensure
that they obtain mutual aid from family, friends, and neighbors and also gain access to key public or semi-public
goods from lawmakers15,16. Past scholars measured social capital using regional level responses from the Global
Social Survey17, custom neighborhood surveys in disaster zones18, and aggregate measures at the county level19.
However, the impact of COVID-19 varies widely by neighborhood. Scholars and local ocials need a clearer
measure of social ties at increasingly hyperlocal levels.
is study introduces new measures for social capital for every census tract, zipcode, and county subdivision
in the United States from 2010 to 2018. ese indices characterize the strength of bonding, bridging, and linking
social capital, as well as overall levels of social capital. Rather than reinventing the wheel, these measures draw
directly from the methodology of Kyne and Aldrich’s19 validated county level Social Capital Indices, extending it
to the census tract level. en, we average these census tract measures up to the zipcode and county subdivision
level to create broad coverage for measuring social capital. Finally, we apply these indices to predict COVID-19
outcomes in dierent regions of the United States, showing frequent associations with COVID-19 test positivity
rates.
is study makes three main contributions to the literature. First, we extend Kyne and Aldrich’s19 county level
measures of social capital to the census tract, zipcode, and county subdivision, enabling comparison of tens of
thousands of communities in terms of bonding, bridging, and linking social ties. is adds an important resource
to scholars and policymakers involved in disaster and pandemic response eorts, easily paired with the CDC’s
Social Vulnerability Index20 and the Baseline Resilience Indicators (BRIC)21.
Second, we nd that these local level measures of social capital are closely correlated with a key diagnostic
measure of COVID-19 spread, the percentage of COVID-19 tests returned positive. Our models accounted for up
to 90% of the variation in COVID-19 spread. is builds on recent research that suggests that social capital helps
residents adopt new behavioral norms like social distancing and masking and reduce COVID-19 spread10–13,
while verifying it with a measure that adjusts for testing capacity.
OPEN
1Political Science Department, Northeastern University, Boston, MA 02115, USA. 2Security and Emergency Services
Department, Embry-Riddle Aeronautical University, Daytona Beach, FL 32114, USA. 3Security and Resilience
Program, Department of Political Science, School of Public Policy & Urban Aairs, Northeastern University, Boston,
MA 02115, USA. *email: timothy.fraser.1@gmail.com
Content courtesy of Springer Nature, terms of use apply. Rights reserved
2
Vol:.(1234567890)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
ird, we nd that social capital has divergent eects depending on the type of social capital present in a
community and the local context of each city or region. First, each kind of social capital index produces strong,
signicant associations with test positivity rates in multiple cases, and bonding social capital is linked to reduced
COVID-19 spread in regions at large, matching past studies10,12,22. However, in specic cities the eect of bonding
social capital frequently changes: communities with strong bonding ties promote COVID-19 spread through
insular social networks, while strong bridging ties help reduce that spread in urban environments. is matches
results from disaster studies, which highlight that while bonding social capital can help some communities, bridg-
ing ties aid most communities by fostering trust, reciprocity, and pro-social behavioral changes across dierent
racial, ethnic, religious, and political lines23–25.
Literature review
Since the COVID-19 pandemic reached the US in early 2020 (or arguably earlier26), local, state, and federal
government agencies have made tracking key COVID-19 outcomes a top priority, including case rates, death
rates, test positivity rates, excess deaths rates, and vaccination rates. States governments across the country have
digitized measurements of these new pandemic resilience outcomes, creating online dashboards for health
ocials to more easily communicate with policymakers, the press, and the public, and to engender greater
trust and transparency in crisis information ow. While these dashboards provoked early pushback from state
government ocials in denial about the scope of the crisis, they have become powerful tools for highlighting
hotspots of vulnerability, as it became clear quickly that communities of color, working class neighborhoods,
front line workers in crowded areas, the aged, and residents with pre-existing conditions faced high health or
nancial risks from COVID. While some dashboards (see for example this Alabama dashboard27), embraced the
CDC’s Social Vulnerability Index, other key indicators of disaster resilience have been conspicuously absent in
pandemic planning dashboards and data monitoring. Below, we review past explanations from disaster scholar-
ship for pandemic resilience, including mobility, health care capacity, quality of health, governance capacity,
political partisanship, and social capital.
First, some communities might see higher rates of COVID-19 spread due to greater population mobility.
While past crises, like SARS and the Avian u28,29, highlighted that travel can spread viruses across borders,
this was complicated in the ongoing pandemic by the fact that many persons infected with COVID-19 remain
asymptomatic30. Among symptomatic patients, 95% of individuals manifest symptoms between 2 and 11days
aer contact with COVID, with a median of 5 days31–33. Recent studies of COVID-19 spread have shown that
lockdowns, compulsory34 or not35, greatly reduced mobility and likely altered the course of the epidemic.
Second, communities with better staed and funded health care systems might respond more adroitly to
the pandemic. e capacity of health care systems, measured by better hospital quality and larger physician
workforces36,37, has been linked to better response quality both before4 and during COVID-1938. Similarly, more
capable governments which can eectively allocate resources, purchase appropriate protective gear, manage
information ow, enforce lockdowns, and ensure high shares of residents with health insurance may see better
health and pandemic outcomes34,39,40. Meta-analyses over the last 2years have highlighted common comorbidities
of COVID-1941–43. For example, Shakaib and colleagues’ meta analysis43 ranked comorbidities by prevalence as
(1) hypertension (~ 28.6% of patient deaths), (2) heart conditions and stroke (14.6% and 8.3%); diabetes (13.2%),
smoking and chronic lung disease (8.1% and 3.2%), chronic kidney and liver diseases (7.2% and 2.7%), and being
immunocompromised (4.8%). Obesity also co-occurs frequently with severe cases; in studies of New York City
hospital patients with severe cases of COVID-19, 35.8% to 60.6% reported obesity44,45; obese COVID-19 patients
tend to see 37% higher rates of death during hospitalization and risk of contracting pneumonia46.
ird, communities may see dierent mobility patterns depending on their levels of political partisanship.
Several notable Republican national and state election ocials consistently downplayed the COVID-19 pan-
demic, despite multiple superspreader events at the White House in 202047. ese sentiments played out among
the general public: By mid-July 2020, just 45% of Republican voters ranked COVID-19 as a major threat to the
country compared to 85% of Democrats48. Multiple studies have highlighted that Republican elected ocials
and voters have been less likely to adopt social distancing and other COVID-19 prevention measures than their
Democratic peers49–51.
Fourth, some communities see worse resilience to crisis due to their levels of social vulnerability6. In both
pandemic and disaster studies, socially vulnerable residents fare worse in terms of resilience to crisis due to
employment challenges, limited mobility, and the fear of discrimination or retribution from neighbors or local
authorities. Groups more socially vulnerable to disaster including women and single parents47, families facing
unemployment or poverty52, the elderly53, the LGBTQ+ community54, the disabled55, and racial, religious, or
ethnic minorities56. During COVID, Black Americans have overwhelmingly felt a higher burden7–9,57, with 1 in
1000 African Americans dead from COVID-1958 at the same time as crushing nancial loss, with over 40% of
Black-owned businesses in the US closed permanently this past year59. However, previous research60 has found
that state and local governance has made a dierence in the pandemic for socially vulnerable communities.
Counties considered highly vulnerable based on the CDC’s Social Vulnerability Index that enacted face mask
requirements, gathering restrictions, and stay-at-home orders experienced reductions in the average number of
COVID-19 deaths, compared to similarly high vulnerability counties that did not enact these non-pharmaceutical
interventions.
Finally, some communities, even facing high social vulnerability or weak health care response, have managed
better pandemic outcomes than others due to their social capital. Social capital—the social ties that bind together
a community and enable trust, reciprocity, and collective action14—serves as a key resource for residents respond-
ing to disaster24, closely correlated with health outcomes16 and recovery outcomes aer crisis18,61. Recent survey
and aggregator level research has demonstrated that residents and communities with stronger social capital were
Content courtesy of Springer Nature, terms of use apply. Rights reserved
3
Vol.:(0123456789)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
more likely to socially distance62, see fewer cases and fatalities10,11, and slower spread of COVID-1912. However,
disaster studies acknowledge the Janus-faced nature of social capital, in which social capital can generate both
positive and negative outcomes depending on the type of social ties fosters63. Indeed, recent studies of social
capital highlight that vertical social ties to government ocials are especially key, relative to others22,64.
Social capital comes in three forms: bonding, bridging, and linking social capital65. Bonding social capital
refers to close ties between members of the same social circles, facilitating trust and mutual aid among friends
and family members23,66. Bridging social capital describes association ties between members of dierent social
groups, built through workplaces67, unions68, volunteering17, sports clubs14, and local associations69. Bridging
ties are the lifeblood of democracy, helping residents build shared stake in their community and enabling close
cooperation during and aer crisis24,69. Finally, linking social capital refers to vertical ties connecting residents
to local, state, and national authorities15. ese linking ties help residents access key public goods from elected
ocials and instill trust in government25,70, which has been linked to better public compliance with public health
protocols during outbreaks71, including SARS72, Ebola73, and COVID-19 crises64. We might expect that linking
and bridging ties aid COVID-19 response, while bonding, insular ties might limit the spread of information and
lead to less resilient response under certain conditions63.
Results
is study examines what kinds of communities experience greater outbreaks of COVID-19, and to what degree
local level measures of social capital and vulnerability can predict those outbreaks. As our main outcome of
interest, this study investigates COVID-19 test positivity rates (the percentage of tests that come back positive)74,
approximating the spread of COVID-19, widely adopted in past studies of community spread75–79. While case
incidence rates are biased because some states test more than others, test positivity rates adjust for the number of
total tests performed in an area. ey are also now widely available in some cities and regions at extremely local
geographies—the census tract, zipcode, and county subdivision levels being the focus of this paper.
is study adapts existing, validated methodologies for measuring social capital at the county level19 to new
measures at the census tract, zipcode, and county subdivision levels. is study generated social capital indices
at the census tract, zipcode, and county subdivision levels, and applied them to case studies within and across
counties in Wisconsin, New York, Illinois, and Massachusetts. Each index was measured at the census tract
Figure1. Geographic distribution of social capital indices at the census-tract level. Violins depict distributions
of social capital indices for each of the 9 US Census-bureau designated geographic divisions in the US.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
4
Vol:.(1234567890)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
level, and their distributions are visualized across US Census Divisions in Fig.1 to demonstrate their degree of
geographic variation.
As shown in Fig.2, these case studies demonstrated considerable variation in overall social capital com-
pared to the national median, with urban areas showing especially troublingly low levels of social capital. Below
we explore how this variation might shape resilience to COVID-19, summarizing ndings from these models
(TablesB1–B3). Model tables in the Supplementary Information appendix report the projected increase in log-
test positivity rates given an increase of one standard deviation in the predictor; all eect sizes are comparable
among predictors.
Variation in COVID‑19 spread explained. ese models explained considerable amounts of variation
in the outcome, as shown by the R2 statistics in TablesB1–B3. For models across counties in TableB1, our fully
specied models explained 47–55% of variation in test positivity rates. However, even simpler versions explained
high levels of variation too: Our models in TableB2 with just social capital measure predicted 41–49% (Models
1A, 2A, and 3A), increasing to 43–54% when paired with social vulnerability indicators (Models 1B, 2B, and
3B) (see “Methods”). is is considerable, especially considering the great dierences between dierent kinds
of counites involved (eg. all of Wisconsin). For models within counties, this includes anywhere from 8 to 89% of
variation in the outcome, and 30–90% including xed or random eects. e best explained cases were Brooklyn
(90%), Queens (81%) and Manhattan (87%)..
Figure2. Variation in overall social capital among case studies. is ugure depicts communities as zipcodes,
census tracts, or county subdivisions. Shading represents social capital, measured as a modied Z-score,
showing distance from the national median standardized by the median absolute dierence (MAD). White
indicates national median level of social capital. Blue shows MAD-calculated standard deviations above median,
while red depicts below median. Maps made in R (version 4.0.3) using the sf package (version 1.0-6)81.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
5
Vol.:(0123456789)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
Figure3. Variation explained by model covariates over time. Points depict variation explained (R2 statistic)
from dozens of fully specied OLS models for each time-step. Indicates substantial variation explained by model
covariates, separate from time.
Figure4. Marginal eects of social capital on COVID-test positivity rates. Bands depict marginal eect on
test positivity rates as specied type of social capital varies by 2 standard deviations around the mean, holding
all other variables at means and modes. Bands reveal varied eect of social capital subtypes depending on
geography.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
6
Vol:.(1234567890)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
Figure3 contextualizes this by displaying daily versions of these models among each case study. Our New
York City models explain the model variation (60–80% in most zipcodes), all other models explained anywhere
between 5 and 55% of variation, each crossing at least 40% at some point in the study period. Even when regressed
as dozens of individual models, in Fig.3, we see that model covariates can explain quite high levels of variation
over time.
Eects of social capital on COVID‑19 spread. is analysis produced three general trends, highlighted
in Fig.4 using marginal eects, calculated holding all covariates at their means while varying each specic type
of social capital. To ensure comparable eect sizes, all predictors in all models were rescaled as Z-scores before
modeling, so that beta coecients show the projected increase in the logged outcome as a predictor increases by
1 standard deviation from the mean (frequently called ‘standardized beta coecients’). Similarly, Fig.4 projects
the test-positivity rate with 95% condence intervals for an otherwise average community as each type of social
capital independently increases from −2 to 0 to 2 standard deviations from the mean level in each sample.
First, aer adjusting for social vulnerability, health care capacity and conditions, governance capacity, and par-
tisanship, we nd that linking social capital, a close proxy for trust in government, is oen negatively associated
with test positivity rates. is occurs frequently across all within county models, and is shown by the consistently
downward trending yellow bands in Fig.4. Signicant negative associations were found in TableB3 for Madison
(beta = −0.13, p < 0.001), Brooklyn (beta = −0.04, p < 0.001), Queens (b eta = −0.28, p < 0.003), and Manhattan
(beta = −0.04, p < 0.001). Similar negative trends were found at large in TableB1 for Wisconsin (beta = −0.13,
p < 0.001) and New York City (beta = −0.006, albeit with low signicance at p = 0.823); the main exception was
Massachusetts, which saw higher test-positivity rates (beta = 0.06, p = 0.058).
Second, we see mixed track records for bonding and bridging social capital. At large, bonding and bridging
ties were linked to lower test-positivity rates: Our across-county models in TableB1 show that communities
with strong bonding social ties saw lower test positivity rates, with signicant associations for Massachusetts
(beta = −0.14, p < 0.007) and Wisconsin (beta = −0.04, p < 0.001). Even moreso, communities with strong bridging
ties saw signicant negative associations for Massachusetts (beta = −0.85, p < 0.001), New York City (beta = −0.11,
p < 0.003), and Wisconsin (beta = −0.32, p < 0.001).
However, for within-county models in TableB3, the evidence for bonding and bridging ties diverges.
Bridging ties were linked to lower COVID-19 spread in Madison (beta = −0.05, p = 0.100) and Manhattan
(beta = −0.08, p < 0.001), with negative trends in Chicago with limited signicance (beta = −0.13, p = 0.696) and
Queens (beta = 0.03, p = 0.756). Similarly, bonding ties were linked to lower COVID-19 spread in Manhattan
(beta = −0.18, p < 0.001), with less signicant negative trends in Chicago (beta = −0.103, p = 0.684) and Madison
(beta = −0.06, p = 0.153).
However, several cases diverged in TableB3. Greater bonding ties were linked to greater disease spread in
Brooklyn (beta = 0.04, p < 0.001), the Bronx (beta = 0.08, p = 0.012), and Queens (beta = 0.66, p < 0.001). And
greater bridging ties were linked to greater spread in Brooklyn (beta = 0.03, p < 0.001). Indeed, these communities
were hard hit by COVID-19 early; Brooklyn is home to Spring Creek Towers, the subsidized housing develop-
ment which suered an outbreak leading their zipcode to face the highest death rate from COVID-19 in New
York in the rst 3 months of the outbreak, while the Bronx overall faced the highest rates for COVID-19 cases,
hospitalizations, and deaths in the rst 3 months80.
Figure5. COVID-19 test positivity rates in New York City. Zipcodes of 5 boroughs based on whether above or
below the median level of social capital in the NYC area. Blue lines depicting counties, a color scale depicting
distance from the median COVID-19 test positivity rate (shaded white), averaged over time. Maps made in R
(version 4.0.3) using the sf package (version 1.0-6)81.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
7
Vol.:(0123456789)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
Yet this is not necessarily a contradictory nding. Past studies of disaster indicate that bonding social capital
is ckle24; it can help family, friends, and members of the same social circle team up and share resources, but it
can also promote insular social circles and inhibit the spread of quality information. In other words, we might
expect that in historically marginalized communities where trust in local and national authorities is limited, that
bonding social capital might not help these communities as much as bridging or linking social capital.
Spatial dierences in eects of social capital on covid‑19 spread. Figure5 highlights these trends
in three panels, which highlight using black borders the zipcodes with the highest bonding (le panel), bridging
(center panel), and linking social capital (right panel), compared to the COVID-19 test positivity rate, colored
from blue (far below median) to white (median) to red (far above median). We see that bridging ties aided
some parts of the city: many Bronx and Brooklyn zipcodes, with summarily low bridging social ties, saw high
COVID-19 outbreak levels, while Manhattan zipcodes with strong bridging ties saw low COVID-19 outbreaks.
In contrast, Staten Island experienced high test positivity rates on average, despite having strong bonding, bridg-
ing, and linking social capital, possibly leading to New York City’s overall skewed, positive eects of bridging and
linking social capital on COVID-19 spread.
ese ndings about bonding and bridging social capital at rst glance defy past ndings in disaster studies,
where aer adjusting for bonding, bridging, and linking social capital, bridging ties, fostered through associa-
tions, typically encourage greater resilience to crisis than bonding, in-group ties. However, the nature of COVID-
19 is especially challenging to bridging social capital. While linking social ties, as usual, help promote trust in
public health and disaster response, bridging ties are usually indicated by connectedness to multiple dierent
parts of society. In this case, that makes bridging residents in a community both capable of great good, reaching
out to neighbors and supporting each other, but also capable of spreading the virus to new groups.
However, as highlighted by bridging social capital’s negative eects on COVID-19 spread within the Manhat-
tan, Madison, New York City, and Wisconsin at large, many communities with strong bridging social ties are
developing important innovations to respond to COVID, ranging from virtual religious services and clubs to
masked volunteer outreach and volunteering by teens out of school. For example, the Jewish Center located in
Manhattan’s Upper West Side, which typically can accommodate 500 of its faithful, now can only accommodate
60 members who must pre-register and have their temperature taken on site before entering. Similarly, the Dar
Al-Dawah mosque in Astoria, Queens now limits their in-person attendance to 64 people, requiring temperature
checks and hand sanitization at the door for entry. ose who worship in-person are required to wear a mask
and bring their own prayer rugs and place them in designated spots that allow for six feet of separation. Dar
Al-Dawah has also oered additional services during religious holidays to accommodate higher demand for
in-person, community worship82.
In other words, while strong bridging ties may not mean an automatic transition to resilience to the pandemic,
they represent an important reservoir that city and health care ocials can draw on to mobilize communities,
provide aid, and address the "twin pandemics" of COVID’s outright eects and its disproportionate eects on
Black Americans.
Discussion
is study measured bonding, bridging, and linking social capital for each census tract, zipcode, and county
subdivision between 2010 and 2018 in the United States. en, we applied these indices to predicting COVID-
19 test positivity rates in Massachusetts, New York, Illinois, and Wisconsin, zooming in further to several urban
areas, including Chicago, IL, Madison, WI, and each of the ve boroughs of New York City. Broadly, we found
that linking social capital is frequently tied to less COVID-19 spread (with Massachusetts as a contrarian case),
while bonding and bridging social demonstrate mixed eects depending on the underlying social and racial
histories of these communities.
In particular, we nd that the eects of social capital are not uniform but divergent depending on whether
those social ties link in-groups (bonding), bridge dierent groups (bridging), or engender trust in ocials (link-
ing), and the local context in which they are built.
Contributions to the literature. Strong evidence was found supporting our hypothesis that social capital
indices are strong, signicant predictors of COVID-19 spread rates. Across counties, communities with stronger
bonding social capital see lower rates of COVID-19 spread. Recent research on matched samples of US coun-
ties showed similar results, where bonding social capital seemed to help families and friends shelter and rein-
forced norms to wear masks and physical distance22. is largely conrms the trends found in past studies10,12,
including those which use robust metrics like excess death rates22. It is worth noting that overall, when analyzed
descriptively in Fig.6’s scatterplots, each type of social capital was negatively correlated with COVID-19 spread.
e main exception in Fig.6 is New York City, where linking social capital was related to worse spread. ere
are several reasons why this might be the case. First, Staten Island is home to 5% of New York City’s population,
but residents accounted for a quarter of COVID-19 fatalities in late 2020. e politically conservative borough
has struggled with low resident participation in public health measures, including diculty enforcing restric-
tions on indoor dining, residents refusing to wear masks83. And in Brooklyn, public health messaging has had
diculty penetrating some communities despite strong social ties. For example, tensions between Mayor Bill
DeBlasio and the Orthodox Jewish community over COVID-19 have ared since March 2020, when the mayor
sent police to break up a local rabbi’s funeral and disperse crowds to prevent COVID-19 transmission. Such
confrontations have been attributed to a decline in trust in public health measures and government response,
and helped circulate rumors discouraging COVID-19 testing84.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
8
Vol:.(1234567890)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
However, aer applying statistical controls, we found that each type of social capital produced varying eects
depending on geography. When we zoomed into urban settings, bonding social ties took on mixed eects. In
three out of six urban areas studied, communities with stronger bonding ties on average saw higher rates of
COVID-19 spread, matching past ndings in disaster studies, where bonding social ties can promote insular
response24.
Further, bridging social capital showed varying eects, leading to lower spread in Massachusetts, Chicago,
Brooklyn, the Bronx, and Wisconsin (although these last two eects had lower statistical signicance, where
p > 0.10). We theorize that bridging ties help ameliorate the deleterious eects of bonding social capital when
present, but that since many bridging ties are formed in civil society groups and associations, they were occasion-
ally the cause of early superspreader events. is study suggests that local planners should carefully inventory
their community resources when responding to pandemics, in order to proactively channel bridging and linking
social capital early into better public response.
Limitations. Finally, this study came with several limitations. First, we relied on aggregate indicators
sourced annually from the American Community Survey to make inferences about the state of social capital in
communities. Future studies should ground-truth these measures by comparing them with survey responses in
local communities. Second, this study also relied on a handful of indicators at the zipcode or county level, which
were then averaged down onto census tracts, due to lack of availability. is primarily aected the bridging and
linking indices, which fortunately still retain much variability within counties.
Figure6. Social capital (mostly) reduces COVID-19 spread. Bivariate Scatterplots of social capital index scores
compared to test positivity rates over, with lines of best t depicting weak-to-strong negative associations for all
except linking social capital in New York City.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
9
Vol.:(0123456789)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
ird, as discussed in the “Methods” section, we relied on county averages to ll in some indicators of
bridging social capital at the census tract level; future studies should identify comparable correlates or principal
components to approximate such measures. Fourth, this study focused on data from 2010 to 2018, due to the
limited availability of bridging and linking indicators, but future studies should extend this backwards to 2000.
Finally, this observational study showed associations between social capital and COVID-19 spread, but fur-
ther studies are needed to verify whether these associations could be causal. Other factors may also inuence
COVID-19 spread depending on geographic context, including pollution85,86 and disaster damage87; natural
and quasi-experiments may be helpful tools for further disentangling social capital’s eects from confounders22.
Despite these limitations, these local level indices produce a considerably granular picture of the state of social
capital in the US, far more so than previously available, and are a powerful tool for predicting the spread and
abatement of COVID-19.
Policy implications. In summary, this study analyzed the state of social capital in the United States between
2010 and 2018, generating estimates for bonding, bridging, and linking social capital in 72,877 census tracts over
9years. rough geographic aggregation, we have produced estimates at the census tract, zipcode, and county
subdivision level, which can serve as a resource for scholars and policymakers involved in disaster and pandemic
response and recovery eorts. Further, this study identied through 3 regional studies and 6 urban case studies
the considerable statistical power of these indices in predicting the spread of COVID-19.
We found considerable evidence that bonding social capital reduces COVID-19 spread at large, but especially
in our urban case studies, this trend reversed. Instead, several cities displayed divergent trends, known as the
Janus-faced nature of social capital63, where the insular, homophilous social networks fostered by bonding social
capital were associated with greater COVID-19 spread (eg. Brooklyn, the Bronx, & Queens), while the bridging
networks, trust, and reciprocity fostered by bridging social capital were associated with reduced COVID-19
spread in others cities (Madison, Manhattan, New York City at large, etc.),. Finally, this study found continued
evidence that linking social capital is oen negatively associated with COVID-19, matching past ndings that
trust in government and public health ocials is key to pandemic prevention and response22,64,71–73. However,
the positive association between linking ties and COVID-19 spread in Massachusetts hint that these trends are
not set in stone and may uctuate depending on local crisis conditions.
ese ndings highlight that while horizontal social ties usually aid in community resilience, local policy-
makers should pay special attention to encouraging trust and reciprocity among residents. is is especially vital
considering that rising political polarization has been accompanied by pandemic denial from some state ocials
and rejection of social distancing and mask mandates among residents49–51.
Future studies should apply these indices to examine how social ties aected the movements and patterns
of residents, economic activity, and physical and mental health during the pandemic—all important adjacent
indicators of risk and resilience during the COVID-19 crisis. Further, scholars and policymakers should use these
indices to identify communities as similar as possible to their own in terms of social capital in order to make
projections about their communities’ recovery trajectories. Finally, scholars should apply these indices to diag-
nosing and explaining a wide variety of other community resilience, highlighting the close relationships between
social capital and policy-relevant social outcomes, including health16, political polarization49–51,88,89, adaptation
to climate change90, and resilience to future disasters25,55,91. We hope that these indices galvanize social capital
scholars to examine the role of social capital in their communities’ recovery and to encourage greater attention
to the role of residents and community planning in recovery and response to crisis.
Methods
is study examines what kinds of communities experience greater outbreaks of COVID-19, measuring to what
degree local-level measures of social capital and vulnerability predict those outbreaks. Below, we (1) outline our
methodology for adapting these measures to the local level, and (2) describe our modeling procedures for case
studies demonstrating the uses of these indices in predicting COVID-19 spread.
Measuring social capital at the local level. To measure social capital, we took the average of three
subindices (bonding, bridging, and linking) in each census tract, built out of 20 indicators total, listed in Table1.
ese indicators were outlined and validated for the original county level social capital index19, but we describe
them below as well.
First, to represent bonding social capital, we use 9 indicators describing how similar residents in a community
are in terms of race, age, class, gender, language, and communication capacity, because homophilous communi-
ties tend to have strong bonding social ties98,99. To represent race similarity, we used a fractionalization approach
to measure how fractionalized a community is into dierent racial categories, where 0 represents homogeneity
and 1 represents heterogeneity92. We repeated this approach for ethnicity similarity, using the share of residents
which identied as Latino or Hispanic, or not92, as well as similarity between genders by income95. To represent
educational equality, we calculated the negative absolute dierence between the share of residents with a college
education compared to those which did not graduate high school93. Each of the aforementioned negative meas-
ures were then reverse-coded, so that a low value denotes low heterogeneity while high values denote homoge-
neity. Finally, we measured four more positive measures of homogeneity. To represent employment equality, we
calculated the absolute dierence between the share of the employed and unemployed labor force, because this
indicates that most of the labor force is similarly employed96. To represent language competency, we used the
percentage of residents who speak English prociently; to represent communication capacity, we used the share
of households with a telephone; while to represent age, we used that share of residents below 65years of age21,94.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
10
Vol:.(1234567890)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
Second, to represent bridging social capital, we use 6 indicators of membership in associations that can
bridge dierent parts of society. We measured the number of religious organizations93, civic organizations, social
advocacy organizations96, charitable organizations, and unions68, normalized per 10,000 residents. Each of these
was obtained from the Zipcode Business Patterns census, and then averaged to the census tract level. Finally, we
supplement this with one county tally not available at the zipcode level, the percentage of residents participating
in a fraternal order.
ird, to represent linking social capital, we use 5 indicators of connection and representation. To represent
political linkages, we gathered the share of voting age eligible citizens94, while to represent government linkages,
we measured the share of local, state, and federal government employees per capita97. We supplemented these
census tract level measures with one county level measure of political activities, namely the share of residents
who attended a political rally, speech, or organized protest that year95.
is culminated in four indices, including bonding, bridging, and linking social capital, as well as overall
social capital, which represents the average of its three subindices. To combine indicators, we use the mean,
because this best approximates the midpoint of a small number of indicators for a given census tract. is is
better compared to the median, which could potentially completely ignore the inuence of the lowest indicator
in favor of more commonly occurring values (for an example, see Fig.A1). Each index was measured at the
census tract level, and their distributions are visualized across US Census Divisions in Fig.1 to demonstrate
their degree of geographic variation.
is data comes with several caveats. e indicators for these indices, presented in Table1, are based on
various tallies pre-aggregated by the census to the lowest level at which they would provide these data (eg. tract,
zipcode, or county subdivision). It was not possible to test the geographic distributions of these tallied units (eg.
people or organizations) within a census tract, zipcode, or subdivision; that would require individual records
for each organization, but the census does not provide these, so as to preserve the privacy of respondents. ese
indices cannot explore spatial variation within a community; as such data becomes available, we encourage future
studies to test such questions. Fortunately, they are well suited to expose spatial variation in social capital among
communities, as shown below in Fig.1, and later, in Fig.2.
We present the distributions of our indicators using density plots in Appendix Fig.A2; indicators largely
retained a bell curve shape, except for a handful, such as language competency and communication capacity.
However, we view this as important evidence of variation in aspects of bonding social capital; it is important
to up- or downweight- communities’ bonding indices based on these observed trends, since such factors are
Table 1. Indicators for census-tract level social capital indices. a Filled in missing census tract values with
average value from census tracts in that county. Missing data tally reects aer imputing county median for
specially marked zipcode bridging social capital measures. b Used county level measure, because comparable
measures were unavailable at local levels.
Index Concept Indicator Eect on index Level Years Missing data (%)aLiterature
Bonding
Race similarity Race fractionalization (0 = homogeneity, 1 = het-
erogeneity) − Tract 2010–18 0.9% 92
Ethnicity similarity Ethnicity fractionalization (0 = homogeneity,
1 = heterogeneity) − Tract 2010–18 0.9% 92
Education equality Negative absolute dierence between % of resi-
dents with college education vs. did not graduate
high school − Tract 2010–18 1.4% 93,94
Race/income inequality Gini coecient (0 = equality, 1 = inequality) − Tract 2010–18 1.2% 21
Employment equality Absolute dierence between % employed and
unemployed labor force + Tract 2011–18 1.0% 95
Gender income similarityaGender income fractionalization (0 = homogene-
ity, 1 = heterogeneity) − Tract 2010–18 3.4% 68
Language competency % Procient English Speakers + Tract 2010–18 1% 94
Communication capacity % Households with telephone + Tract 2010–18 1.5% 21
Non-elder population % Below 65years of age + Tract 2010–18 1% 94
Bridging
Religious Organizations Religious organizations per 10,000 persons + Zipcode 2012–16 0.4% 93
Civic OrganizationsaCivic organizations per 10,000 persons
Social Advocacy organizations per 10,000
persons + Zipcode 2012–16 0.7% (9.3%)
2.5% (22.0%) 96
Social embeddedness—charitable tiesaCharitable organizations per 10,000 persons + Zipcode 2012–16 2.3% (20.7%) 68
Social embeddedness—Fraternal tiesbMember of fraternal order (% of total) + County 2010 0% 68
Social embeddedness—Union ties Unions per 10,000 persons + Zipcode 2012–16 3% 68
Linking
Political Linkage % of voting age citizens eligible for voting + Tract 2010–18 1% 94
Local government linkage % local government employees (per capita) + Tract 2010–18 1% 97
State government linkage % state government employees (per capita) + Tract 2010–18 1% 97
Federal government linkage % federal government employees (per capita) + Tract 2010–18 1% 97
Political linkage-political activitiesb% Attended political rally, speech, or organized
protest + County 2010 0% 95
Content courtesy of Springer Nature, terms of use apply. Rights reserved
11
Vol.:(0123456789)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
closely involved in building in-group ties. e exception was for rates, including all bridging social capital indica-
tors (organizations normalized by population) and 3 linking indicators (government employees normalized by
population). Because rates are always right-skewed, we log-transformed these (before rescaling from 0 to 1) to
be more comparable to other variables. Finally, to deal with outliers, while retaining the general shape of these
distributions, we capped all indicators at their 2.5th and 97.5th percentiles.
Missing data. While 15 out of 20 indicators were missing 1% of data points or less, 3 indicators were missing
considerable amounts of data. ese variables included social advocacy groups (missing 22%), charitable organi-
zations (20.7%), and civic organizations (9.3%). While complete data is obviously preferable, these associational
measures are vital to the measurement of bridging social capital. Despite these missing data points, most of these
census tracts were surrounded by other census tracts where this data was widely available. As a result, we use a
two-stage imputation process, rst imputing missing data for these ve variables alone with each census tract’s
county median value. is reduces missing data considerably, and ensures that our indices are geographically
consistent. Aer county median imputation, social advocacy groups lack only 2.5% of data points, charitable
organizations lack just 2.3%, and civic organizations lack just 0.7%. ese are much better levels of missing data.
To ll in remaining data points, we leverage the statistical power of our dataset of 72,877 census tract obser-
vations from 2010 to 2018, using multiple imputation using the Amelia II soware (version 1.76) in R (version
4.0.3)100. Using time-series imputation and melding results across 5 imputed datasets improves imputation for
missing data over time, because it lls in data points from, for example, 2013, by taking into consideration sen-
sible values given that census tract’s values in 2012 and 2014. While complete data is preferable, this two-handed
strategy helps keep missing data points geographically similar to those in their county and temporally similar
to observations of the same census tract over time. Our indices reect local level variation considerably better
than would a local level index that excludes these indicators.
Aggregation. Finally, we aggregated these census tract measures to the zipcode and county subdivision
levels as well. Since these geographies do not overlap perfectly, for each zipcode or county subdivision, we identi-
ed all census tracts that overlap with these jurisdictions and took the median index score for each index, which
best captures the most likely index score for that tract given its surroundings. We repeated this process with the
CDC’s 2018 Social Vulnerability Index20,101 and its four subcomponents, to enable comparisons between social
capital and social vulnerability indices at the census tract, zipcode, and county subdivision levels.
Variables. Finally, to demonstrate the value of these granular social capital indices, we modeled the eects
of social capital on COVID-19 spread in several case studies, to be discussed further below. is study uses the
following variables to represent key factors in COVID-19 spread.
As our outcome variable, we used the test positivity rate—the percentage of tests that return positive. is is a
useful outcome to measure, because it highlights how widespread infection is in a community, while controlling
for the amount of testing in a locale74. In contrast, straight case rates and death rates may miss unidentied cases.
(Other ideal measures, such as excess deaths rates, are not currently available below the county level). Test posi-
tivity rates are a useful proxy for the spread of COVID-19, because these rates closely correlate with COVID-19
case rates, rates of residents with COVID-19 antibodies77, hospital admissions rates75, and death rates76. Rates
of positive tests have been used to measure spread in studies in US counties102, Louisiana county subdivisions87,
and New York City zipcodes78,79, among others.
Since this outcome is right skewed, we log transformed it and applied linear models, as discussed below.
Because our outcome is a percentage, these models already adjust for the size of the population.
For each case study, we tested the eects of bonding, bridging, and linking social capital on test positivity
rates, controlling for the CDC’s for 4 social vulnerability sub-indices. ese include (1) socioeconomic status, (2)
minority status and language, (3) household composition and disability, and (4) housing type and transport20,94.
Additionally, we controlled for population mobility, measured over time at the county level, using the aver-
age daily change in workplace mobility, estimated with Google android user movement103. ese measures have
been frequently used to study COVID-19 and mobility (eg.11,51,62). We averaged movement between two and
eleven days prior to each observation, since 95% of individuals manifest symptoms between 2 and 11days aer
contact with COVID-1931–33.
Finally, we also controlled for several other constant county level traits. ese include health care capacity,
which we measured by averaging two rescaled indicators, including the number of primary care physicians per
100,000 residents in 201736 and the number of preventable hospital stays per 100,000 residents in 2017, which
we reverse scaled104. We also measured overall quality of health by averaging seven indicators. ese include the
precentage of residents identied as current smokers in 2017, drinking excessively in 2017, who reported being
physically inactive in 2016, who had diabetes in 2016, who were obese in 2016, who reported experiencing poor
physical health over 14days in a month in 2017, and the age-adjusted premature mortality rate (deaths under age
75) between 2016 and 2018. ese measures were gathered from the County Health Rankings105. We also control
for governance capacity, measured by the number of municipal employees per capita from the American Com-
munity Survey in 2018, and for partisanship, measured by the share of residents who voted for the Democratic
presidential candidate in 2016 using data from the MIT Elections Lab106.
Case studies. We applied this array of variables to two types of case studies, including (1) cases across coun-
ties and (2) cases within counties. ese cases were located in the Northeast, Mid-Atlantic, and Midwest, to
highlight indices in dierent regions and geographic levels, where COVID-19 outcome data had been reported
at these granular levels. (While it is rare for entire states to report these outcomes, many cities do).
Content courtesy of Springer Nature, terms of use apply. Rights reserved
12
Vol:.(1234567890)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
First, we examine variation across counties at our three levels of interest, drawing from (1) weekly test positiv-
ity rates over a 7-day period in 1393 Wisconsin census tracts from April 6, 2020 to November 16, 2020, (2) daily
test positivity rates over a 7day period in 178 New York City zipcode tabulation areas within the ve boroughs
from May 18 to October 28, 2020, and (3) weekly test positive rates over a 14day period in 350 Massachusetts
county subdivisions from September 30 to November 19, 2020.
Massachusetts, Wisconsin, and New York City used the rate of molecular tests (PRC tests)107–109, while Chi-
cago’s test positivity rates combine both molecular and antigen test results110. Despite including antigen testing,
Chicago retained the same negative associations for bridging and linking social capital found in Wisconsin,
Madison, Manhattan, Queens, and New York City, although Chicago’s trends demonstrated lower statistical
signicance (p-values = 0.30–0.70) (see “Results”). Further, it is worth noting that the study period (2020) largely
predates mass-availability of antigen tests in local convenience stores, so during this period, we expect that
molecular tests reect well the state of testing across dierent cities.
For these cases, we model the eect of social capital indices, controlling for social vulnerability indices, at
the census tract, zipcode, or county subdivision, and we then control for county-level traits including mobility,
health care capacity, overall health quality, governance capacity, and partisanship.
Our primary goal in these models is to evaluate and compare the eects of bonding, bridging, and linking
social capital indices across dierent geographic contexts. In contrast, some dierences between control variables’
eects, eg. mobility or governance capacity, are to be expected, due to dierent levels of governance capacity
and COVID precautions (although shared eects among control variables are always a good sign). To test the
consistency of social capital eects, in TableB2, we break down our original TableB1 models across counties,
showing to what degree the eects of social capital indices remain consistent when using just our independent
variables with no controls (Models 1A, 2A and 3A), basic controls (Models 1B, 2B, and 3B), and full controls
(Models 1C, 2C, and 3C, the same as Models 1–3 in TableB1). In the results, we report associations which persist
across each level of controls, as a validity check to ensure our results are not simply due to model specication.
en, for each model, we compared xed eects by date, random eects by jurisdiction, and random eects
by jurisdiction nested within counties. e advantage here is that by nesting within multiple counties, we can
control for county wide traits. We used Hausman tests to choose between xed, random, and nested random
eects. Given a statistically signicant Hausman test (in this case, p < 0.001), we reported xed eects, and oth-
erwise reported nested random eects, which t better than random eects alone. Models tables are presented
in the Supplementary Information.
Second, we examine variation within counties at the zipcode level and census tract level (there are usually
too few county subdivisions to do this otherwise). We draw from 77 census tracts in Madison, WI from April
6, 2020 to November 16, 2020, from 60 zipcodes in Chicago, IL from March 7 to November 14, 2020, and from
44 zipcodes in Manhattan, 25 in the Bronx, 37 in Brooklyn, and 60 in Queens between May 18 and October 28,
2020. (Staten Island could not be modeled individually due to considerable collinearity between multiple social
capital and vulnerability indicators). Each model includes social capital and social vulnerability indicators, with
xed eects by date or random eects by jurisdiction. Since county variables are time-invariant or correlate
with temporal xed eects, they cannot be included in these models, which are intended to use as few variables
as possible, to demonstrate the considerable predictive power of these main indices together. Signicant Haus-
man tests (in this case, p < 0.001) led us to use xed eects for models of Madison, Brooklyn, and Manhattan,
while results of limited statistical signicance (p ~ 0.90) led us to use random eects for Chicago, the Bronx, and
Queens. ese are visualized in Fig.4, using marginal eects111 calculated in the ggeects package (version 1.1.1)
in R (version 4.0.3)114, as described in the “Results”.
Finally, we repeated all nine models (within counties and across counties) as simple OLS models for each
timestep to demonstrate the high predictive power of social capital and controls in each timestep. We use the
R2 statistic, the percentage of variance explained in the outcome, to represent this, and plot this over time for
each sample in Fig.3.
Goodness of t and validity. Figure3 required generating hundreds of models in a loop, but each of these
small OLS regression models fullled necessary assumptions112: First, no model demonstrated problematic col-
linearity in these models; all variance ination factor scores were below 10, the threshold for problematic colline-
arity. Second, there was no problematic heteroskedasticity; by modeling the data separately for each time period,
we removed any temporal relationships; further, heteroskedasticity aects standard errors, but not model-level
statistics like the R2 statistic, remedying any such concerns.
Further, in our reported models in TablesB1–B3, several models had highly collinear variables, usually
between social capital and social vulnerability variables. We performed a series of transformations on these
variables, which reduced collinearity for each variable, measured by the variance ination factor, to below 10,
the indicator for problematic levels, and near 2.5, the gold standard. In a handful of cases, problematic covari-
ates like governance capacity were dropped when transformations would not suce. ese steps are described
beneath each Supplementary Information table. Results of these models are presented above in the main text,
while model tables are listed in TablesB1–B3.
Ethics declarations. is study involved no human subjects and relied on only aggregate, publicly available
data and therefore did not require ethics review.
Data availability
All code necessary for replicating this study will be made available for replication on the Harvard Dataverse
(https:// doi. org/ 10. 7910/ DVN/ OSVCRC).
Content courtesy of Springer Nature, terms of use apply. Rights reserved
13
Vol.:(0123456789)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
Received: 4 September 2021; Accepted: 29 March 2022
References
1. Johns Hopkins. COIVD-19 data in motion (Coronavirus Resource Center, 2021).
2. Allen, J., Almukhtar, S. & Aufrichtig, A. Coronavirus in the US: Latest Map and Case Count (New York Times, 2021).
3. Comfort, L. Shared Risk: Complex Systems in Seismic Response (Emerald Publishing, 1999).
4. Schoenbaum, S. C., Schoen, C., Nicholson, J. L. & Cantor, J. C. Mortality amenable to health care in the United States: e roles
of demographics and health systems performance. J. Public Health Policy 32, 407–429 (2011).
5. Farag, M. et al. Health expenditures, health outcomes and the role of good governance. Int. J. Health Care Finance Econ. 13,
33–52 (2012).
6. Cutter, S. L., Boru, B. J. & Shirley, L. W. Social vulnerability to environmental hazards. Soc. Sci. Q. 84, 242–261 (2003).
7. Yancy, C. W. COVID-19 and African Americans. JAMA 323, 1891–1892 (2020).
8. Wadhera, R. K. et al. Variation in COVID-19 hospitalizations and deaths across New York City boroughs. JAMA 323, 2192–2195
(2020).
9. Chin-Hong, P., Alexander, K. M., Haynes, N., Albert, M. A., e Association of, Black Cardiologists. Pulling at the heart: COVID-
19, race/ethnicity and ongoing disparities. Nat. Rev. Cardiol. 17, 533–535 (2020).
10. Borgonovi, F., Andrieu, E. & Subramanian, S. V. e evolution of the association between community level social capital and
COVID-19 deaths and hospitalizations in the United States. Soc. Sci. Med. 278, 113948 (2021).
11. Borgonovi, F. & Andrieu, E. Bowling together by bowling alone: Social capital and Covid-19. Soc. Sci. Med. 265, 113501 (2020).
12. Varshney, L. R. & Socher, R. COVID-19 growth rate decreases with social capital. medRxiv, 2020.04.23.20077321 (2020).
13. Jean-Baptiste, C., Herring, R. P., Beeson, W. L., Dos Santos, H. & Banta, J. E. Stressful life events and social capital during the
early phase of COVID-19 in the US. Soc. Sci. Humanit. Open 2, 100057 (2020).
14. Putnam, R. D. Bowling Alone: e Collapse and Revival of American Community (Simon and Schuster, 2000).
15. Szreter, S. & Woolcock, M. Health by association? Social capital, social theory, and the political economy of public health. Int.
J. Epidemiol. 33, 650–667 (2004).
16. Kawachi, I., Subramanian, S. V. & Kim, D. Social Capital and Health (Springer, 2008).
17. Lee, J. & Fraser, T. How do natural hazards aect participation in voluntary association? e social impacts of disasters in
Japanese society. Int. J. Disaster Risk Reduct. 34, 108–115 (2019).
18. Iwasaki, K., Sawada, Y. & Aldrich, D. P. Social capital as a shield against anxiety among displaced residents from Fukushima.
Nat. Hazards 89, 405–421 (2017).
19. Kyne, D. & Aldrich, D. P. Capturing bonding, bridging, and linking social capital through publicly available data. Risk Hazards
Crisis Public Policy 11, 61–86 (2020).
20. Flanagan, B. E., Hallisey, E. J., Adams, E. & Lavery, A. Measuring community vulnerability to natural and anthropogenic hazards:
e Centers for Disease Control and Prevention’s Social Vulnerability Index. J. Environ. Health 80, 34–36 (2018).
21. Cutter, S. L., Burton, C. G. & Emrich, C. T. Disaster resilience indicators for benchmarking baseline conditions. J. Homel. Secur.
Emerg. Manag. 7, 23 (2010).
22. Fraser, T., Aldrich, D. P. & Page-Tan, C. Bowling alone or distancing together? e role of social capital in excess death rates
from COVID19. Soc. Sci. Med. 284, 114241 (2021).
23. Hawkins, R. L. & Maurer, K. Bonding, bridging and linking: How social capital operated in New Orleans following Hurricane
Katrina. Br. J. Soc. Work. 40, 1777–1793 (2010).
24. Aldrich, D. P. Building Resilience: Social Capital in Post-Disaster Recovery (University of Chicago Press, 2012).
25. Aldrich, D. P. Black Wave: How Networks and Governance Shaped Japan’s 3/11 Disasters (University of Chicago Press, 2019).
26. Basavaraju, S. V. etal. Serologic testing of U.S. blood donations to identify SARS-CoV-2-reactive antibodies. Clin. Infect. Dis.
(2020).
27. Samford University. Coronvirus COVID-19 in Alabama. https:// exper ience. arcgis. com/ exper ience/ e03f8 7e48a 234fe ebbad 27d0e
e78 24 (2021).
28. Bowen, J. T. & Laroe, C. Airline networks and the international diusion of severe acute respiratory syndrome (SARS). Geog r.
J. 172, 130–144 (2006).
29. Smallman-Raynor, M. & Cli, A. D. e geographical spread of avian inuenza A (H5N1): Panzootic transmission, pandemic
potential, and implications. Ann. Assoc. Am. Geogr. 98, 553–582 (2008).
30. Lavezzo, E. et al. Suppression of COVID-19 outbreak in the municipality of Vo, Italy. Nature 584, 425–429 (2020).
31. Lauer, S. A. et al. e incubation period of coronavirus disease 2019 (COVID-19) from publicly reported conrmed cases:
Estimation and application. Ann. Intern. Med. 172, 577–582 (2020).
32. Guan, W. J., Ni, Z. Y. & Hu, Y. Clinical characteristics of coronavirus disease 2019 in China. N. Engl. J. Med. 382, 1708 (2020).
33. Li, Q., Guan, X., Wu, P. & Wang, X. Early transmission dynamics in Wuhan, China, of novel coronavirus-infected pneumonia.
N. Engl. J. Med. 382, 1199–1207 (2020).
34. May, T. Lockdown-type measures look eective against covid-19. BMJ, m2809 (2020).
35. Yabe, T. et al. Non-compulsory measures suciently reduced human mobility in Tokyo during the COVID-19 epidemic. Sci.
Rep. 10, 18053 (2020).
36. Baicker, K. & Chandra, A. Medicare spending, the physician workforce, and beneciaries’ quality of care. Health A. 23, W4-197
(2004).
37. Goodman, D. C. & Grumbach, K. Does having more physicians lead to better health system performance?. JAMA 299, 335–337
(2008).
38. Miller, I. F., Becker, A. D., Grenfell, B. T. & Metcalf, C. J. Disease and healthcare burden of COVID-19 in the United States. Nat.
Med. 26, 1212–1217 (2020).
39. Bonaccorsi, G. et al. Economic and social consequences of human mobility restrictions under COVID-19. Proc. Natl. Acad. Sci.
USA 117, 15530–15535 (2020).
40. Bollyky, T. J. et al. e relationships between democratic experience, adult health, and cause-specic mortality in 170 countries
between 1980 and 2016: An observational analysis. Lancet 393, 1628–1640 (2019).
41. Ng, W. H. et al. Comorbidities in SARS-CoV-2 patients: A systematic review and meta-analysis. MBio 12, e03647-20 (2021).
42. Cheng, S. et al. Comorbidities’ potential impacts on severe and non-severe patients with COVID-19: A systematic review and
meta-analysis. Medicine 100, e24971 (2021).
43. Shakaib, B. et al. A comprehensive review on clinical and mechanistic pathophysiological aspects of COVID-19 Malady: How
far have we come?. Virol. J. 18, 1–120 (2021).
44. Richardson, S., Hirsch, J. S. & Narasimhan, M. Presenting characteristics, comorbidities, and outcomes among 5700 patients
hospitalized with COVID-19 in the New York City area. JAMA 323, 2052–2059 (2020).
45. Goyal, P. et al. Clinical characteristics of covid-19 in New York City. N. Engl. J. Med. 382, 2372–2374 (2020).
46. Docherty, A. B. et al. Features of 20 133 UK patients in hospital with covid-19 using the ISARIC WHO Clinical Characterisation
Protocol: Prospective observational cohort study. BMJ 369, m1985 (2020).
Content courtesy of Springer Nature, terms of use apply. Rights reserved
14
Vol:.(1234567890)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
47. Enarson, E. rough women’s eyes: A gendered research agenda for disaster social science. Disasters 22, 157–173 (1998).
48. Tyson, A. Republicans Remain Far Less Likely than Democrats to view COVID-19 as a Major reat to Public Health (Pew Research
Center, 2020).
49. Grossman, G., Kim, S., Rexer, J. & irumurthy, H. Political partisanship inuences behavioral responses to governors’ recom-
mendations for COVID-19 prevention in the United States. Proc. Natl. Acad. Sci. USA 117(39), 24144–24153 (2020).
50. Painter, M. & Qiu, T. Political beliefs aect compliance with COVID-19 social distancing orders. J. Econ. Behav. Organ. 185,
688–701 (2021).
51. Clinton, J., Cohen, J., Lapinski, J. S. & Trussler, M. Partisan pandemic: How partisanship and public health concerns aect
individuals’ social mobility during COVID-19. Sci. Adv. 7(2), 1–7 (2021).
52. Deng, H. et al. High-resolution human mobility data reveal race and wealth disparities in disaster evacuation patterns. Humanit.
Soc. Sci. Commun. 8, 144 (2021).
53. Salvati, P. et al. Gender, age and circumstances analysis of ood and landslide fatalities in Italy. Sci. Total Environ. 610–611,
867–879 (2018).
54. Dominey-Howes, D., Gorman-Murray, A. & McKinnon, S. Emergency management response and recovery plans in relation to
sexual and gender minorities in NEW South Wales, Australia. Int. J. Disaster Risk Reduct. 16, 1–11 (2016).
55. Uscher-Pines, L. et al. Disaster preparedness of households with special needs in southeastern Pennsylvania. Am. J. Prev. Med.
37, 227–230 (2009).
56. Fussell, E., Sastry, N. & VanLandingham, M. Race, socioeconomic status, and return migration to New Orleans aer Hurricane
Katrina. Popul. Environ. 31, 20–42 (2009).
57. Karaye, I. M. & Horney, J. e impact of social vulnerability on COVID-19 in the U.S.: An analysis of spatially varying relation-
ships. Am. J. Prev. Med. 59, 317–325 (2020).
58. Peck, P. e virus is showing black people what they knew all along. Atlantic. https:// www. theat lan tic. com/ health/ archi ve/ 2020/
12/ pande mic- black- death- toll- racism/ 617460/ (2020).
59. da Costa, P. N. e covid-19 crisis has wiped out nearly half of black small businesses. Forbes (2020).
60. Page-Tan, C. & Corbin, T. Protective policies for all? An analysis of Covid-19 deaths and protective policies among low-,
medium-, and high-vulnerability groups. Disasters 45, S119–S145 (2021).
61. Page-Tan, C. Bonding, bridging, and linking social capital and social media use: How hyperlocal social media platforms serve
as a conduit to access and activate bridging and linking ties in a time of crisis. Nat. Hazards 105, 2219–2240. https:// doi. org/ 10.
1007/ s11069- 020- 04397-8 (2021).
62. Barrios, J. M., Benmelech, E., Hochberg, Y. V., Sapienza, P. & Zingales, L. Civic capital and social distancing during the covid-19
pandemic. J. Public Econ. 193, 1–45 (2021).
63. Aldrich, D. P., Page-Tan, C. & Fraser, T. In (ed Trump, B. D.) (EPFL International Risk Governance Center, 2018).
64. Fraser, T. & Aldrich, D. P. e dual eect of social ties on COVID-19 spread in Japan. Sci. Rep. 11, 1–12 (2021).
65. Aldrich, D. P. & Meyer, M. A. Social capital and community resilience. Am. Behav. Sci. 59, 254–269 (2015).
66. Cox, R. S. & Perry, K. E. Like a sh out of water: Reconsidering disaster recovery and the role of place and social capital in com-
munity disaster resilience. Am. J. Community Psychol. 48, 395–411 (2011).
67. Granovetter, M. S. e strength of weak ties. Am. J. Sociol. 78, 1360–1380 (1973).
68. Norris, F. H., Stevens, S. P., Pfeerbaum, B., Wyche, K. F. & Pfeerbaum, R. L. Community resilience as a metaphor, theory, set
of capacities, and strategy for disaster readiness. Am. J. Community Psychol. 41, 127–150 (2008).
69. Smiley, K. T., Howell, J. & Elliott, J. R. Disasters, local organizations, and poverty in the USA, 1998 to 2015. Popul. Environ. 40,
115–135 (2018).
70. Tsai, L. L. Solidary groups, informal accountability, and local public goods provision in rural China. Am. Polit. Sci. Rev. 101,
355–372 (2007).
71. Funk, S., Gilad, E., Watkins, C. & Jansen, V. A. A. e spread of awareness and its impact on epidemic outbreaks. Proc. Natl.
Acad. Sci. 106, 6872–6877 (2009).
72. Tai, Z. & Sun, T. Media dependencies in a changing media environment: e case of the 2003 SARS epidemic in China. New
Media Soc. 9, 987–1009 (2007).
73. Vinck, P., Pham, P. N., Bindu, K. K., Bedford, J. & Nilles, E. J. Institutional trust and misinformation in the response to the
2018–19 Ebola outbreak in North Kivu, DR Congo: A population-based survey. Lancet. Infect. Dis 19, 529–536 (2019).
74. Johns Hopkins Bloomberg School of Public Health. COVID-19 Testing: Understanding the “Percent Positive”. https:// publi cheal
th. jhu. edu/ 2020/ covid- 19- testi ng- under stand ing- the- perce nt- posit ive#: ~: text= e% 20per cent% 20pos itive% 20is% 20a,with%
20lev els% 20of% 20dis ease% 20tra nsmis sion (2020).
75. Fenga, L. & Gaspari, M. Predictive Capacity of COVID-19 Test Positivity Rate. Sensors 21, 2435 (2021).
76. Al Dallal, A., Al Dallal, U. & Al Dallal, J. Positivity rate: an indicator for the spread of COVID-19. Curr. Med. Res. Opin. 37,
2067–2076 (2021).
77. Chiu, W. A. & Ndeo-Mbah, M. L. Using test positivity and reported case rates to estimate state-level COVID-19 prevalence
and seroprevalence in the United States. PLoS Comput. Biol. 17, e1009374 (2021).
78. Cordes, J. & Castro, M. C. Spatial analysis of COVID-19 clusters and contextual factors in New York City. Spatial Spatio-temporal
Epidemiol. 34, 100355 (2020).
79. Muñoz-Price, L. S. et al. Racial disparities in incidence and outcomes among patients with COVID-19. JAMA Netw. Open 3,
e2021892 (2020).
80. Schwirtz, M. & Cook, L.R. ese N.Y.C. Neighborhoods have the highest rates of virus deaths. New York Times. https:// www.
nytim es. com/ 2020/ 05/ 18/ nyreg ion/ coron avirus- deaths- nyc. html (2020).
81. Pebesma, E. Simple features for R: Standardized support for spatial vector data. R J. 10, 439–446 (2018).
82. Estrin, J. Staying Apart, But Praying Together: As the Coronavirus Surges Again, Houses of Worship in New York Struggle to Serve
their Communities Safely (New York Times, 2020).
83. Farinaci, A. What’s Behind Staten Island’s High COVID-19 Rates? (Spectrum News NYI, 2020).
84. Bellafante, G. When Covid Flared Again in Orthodox Jewish New York (New York Times, 2020).
85. Rovetta, A. & Castaldo, L. Relationships between demographic, geographic, and environmental statistics and the spread of novel
coronavirus disease (COVID-19) in Italy. Curēus 12, e11397 (2020).
86. Kasioumi, M. & Stengos, T. e eect of pollution on the spread of COVID-19 in Europe. Econ. Disasters Clim. Change 6,
129–140 (2021).
87. Page-Tan, C. & Fraser, T. COVID-19 to go? e role of disasters and evacuation in the COVID-19 pandemic. Glob. Environ.
Change 73, 102471 (2022).
88. Panagopoulos, C., Fraser, T., Aldrich, D. P., Hummel, D. & Kim, D. Bridging the divide: Does social capital moderate the impact
of polarization on health?. Polit. Res. Q. https:// doi. org/ 10. 1177/ 10659 12921 10345 61 (2021).
89. Fraser, T., Aldrich, D. P., Panagopoulos, C., Hummel, D. & Kim, D. e harmful eects of partisan polarization on health. PNAS
Nexus 1, pgac011 (2022).
90. Fraser, T., Cunningham, L. & Nasongo, A. build back better? Eects of crisis on climate change adaptation through solar power
in Japan and the United States. Glob. Environ. Polit. 21, 1–22 (2021).
Content courtesy of Springer Nature, terms of use apply. Rights reserved
15
Vol.:(0123456789)
Scientic Reports | (2022) 12:6566 | https://doi.org/10.1038/s41598-022-10275-z
www.nature.com/scientificreports/
91. Page-Tan, C. e role of social media in disaster recovery following Hurricane Harvey. J. Homel. Secur. Emerg. Manage. 18,
93–123 (2021).
92. Alesina, A., Baqir, R. & Easterly, W. Public goods and ethnic divisions. Q. J. Econ. 114, 1243–1284 (1999).
93. Chamlee-Wright, E. & Storr, V. H. Club goods and post-disaster community return. Ration. Soc. 21, 429–458 (2009).
94. Morrow, B. H. Community Resilience: A Social Justice Perspective. http:// www. resil ientus. org/ libra ry/ FINAL_ MORROW_
9‐25‐08_ 12234 82348. pdf (2008).
95. Tierney, K. J., Lindell, M. K. & Perry, R. W. Facing the Unexpected: Disaster Preparedness and Response in the United States (Joseph
Henry Press, 2001).
96. Cutter, S. L., Ash, K. D. & Emrich, C. T. Urban-rural dierences in disaster resilience. Ann. Am. Assoc. Geogr. 106, 1236–1252
(2016).
97. Murphy, B. L. Locating social capital in resilient community-level emergency management. Nat. Hazards 41, 297–315 (2007).
98. Mouw, T. Estimating the causal eect of social capital: A review of recent research. Ann. Rev. Sociol. 32, 79–102 (2006).
99. Jules, P. Social capital and the collective management of resources. Science 302, 1912–1914 (2003).
100. Honaker, J., King, G. & Blackwell, M. Amelia II: A program for missing data. J. Stat. Sow. 45, 1–47 (2011).
101. Geospatial Research, Analysis, and Services Program (GRASP). CDC Social Vulnerability Index 2018 US Database. https://
www. atsdr. cdc. gov/ place andhe alth/ svi/ data_ docum entat ion_ downl oad. html (2018).
102. Khazanchi, R. et al. County-level association of social vulnerability with COVID-19 cases and deaths in the USA. J. Gen. Intern.
Med. 35, 2784–2787 (2020).
103. Google. COVID-19 Community Mobility Reports—Google. https:// www. google. com/ covid 19/ mobil ity/ (2021)
104. Brumley, R. et al. Increased satisfaction with care and lower costs: Results of a randomized trial of in-home palliative care. J.
Am. Geriatr. Soc. 55, 993–1000 (2007).
105. University of Wisconsin Population Health Institute. County Health Rankings Database. https:// www. count yheal thran kings.
org (2019).
106. MIT Election Data & Science Lab. County Presidential Election Returns 2000–2016. https:// elect ionlab. mit. edu (2021)
107. Massachusetts Department of Public Health. COVID-19 Dashboard, September 30, 2020. https:// www. mass. go v/ doc/ covid- 19-
dashb oard- septe mber- 30- 2020/ downl oad (2020).
108. Wisconsin Department of Health Services. COVID-19 Public Use Data Denitions. https:// www. dhs. wisco nsin. gov/ publi catio
ns/ p0267 7a. pdf (2021).
109. NYC Department of Health and Mental Hygiene. NYC Coronavirus Disease 2019 (COVID-19) Data. Github. https:// github.
com/ nyche alth/ coron avirus- data (2021).
110. Illinois National Electronic Disease Surveillance System. COVID-19 Cases, Tests, and Deaths by ZIP Code [Cook County].
Healthdata.gov. https:// healt hdata. gov/ datas et/ COVID- 19- Cases- Tests- and- Deaths- by- ZIP- Code/ kyyi- 5aa8/ data (2020).
111. Soltz, D. & Williams, R. SAGE Research Methods Foundations (eds Atkinson, P. et al.) (Sage Publications, Inc, 2020).
112. Casson, R. J. & Farmer, L. Understanding and checking the assumptions of linear regression: a primer for medical researchers.
Clin. Exp. Ophthalmol. 42, 590–596 (2014).
113. Rosenberg, B. e second White House coronavirus outbreak: Mark Meadows, the Secret Service, and more. Vox (2020).
114. Lüdecke, D. ggeects: Tidy data frames of marginal eects from regression models. J. Open Source Sow. 3, 772 (2018).
Author contributions
T.F., C.P., and D.A. wrote and edited the main manuscript text; T.F. designed methods and analysis; T.F. and C.P.
designed code and collected data; D.A. conceptualized framework and advised research. All authors reviewed
the manuscript.
Competing interests
e authors declare no competing interests.
Additional information
Supplementary Information e online version contains supplementary material available at https:// doi. org/
10. 1038/ s41598- 022- 10275-z.
Correspondence and requests for materials should be addressed to T.F.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional aliations.
Open Access is article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons licence, and indicate if changes were made. e images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.
© e Author(s) 2022
Content courtesy of Springer Nature, terms of use apply. Rights reserved
1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com
Available via license: CC BY 4.0
Content may be subject to copyright.
