Skip to main content

The Geography of Social Capital in America

The Geography of Social Capital in America

April 11, 2018

Additional Material:
PDF Version

Download Data (xlsx), Interactive Maps and Rankings

Social capital is almost surely an important factor driving many of our nation’s greatest successes and most serious challenges. Indeed, the withering of associational life is itself one of those challenges. Public policy solutions to such challenges are inherently elusive. But at present, policymakers and researchers lack the high-quality contemporary measures of social capital available at the state and local levels to even try proposing solutions that are attuned to associational life.

This report describes a new social capital index created to rectify this problem. It details the construction of the index, presents maps summarizing the geographic distribution of social capital, and establishes that the index is consistently—and often strongly—related to a range of economic, social, and demographic indicators. The report also presents the geographic distribution of several subcomponents of social capital, including family unity, family interaction, social support, community health, institutional health, collective efficacy, and philanthropic health.

The Social Capital Project is concurrently providing the state and county data underlying each index, as well as the indices and subindices themselves. It is our hope that the availability of the index will inspire researchers to focus more on social capital and its relationship to other features of economic and social life. And we hope it will aid policymakers as they seek to address the country’s needs.

Among the findings:

  • The top fifth of states, in terms of social capital scores, are home to just nine percent of Americans, while 29 percent live in bottom-fifth states.
  • We have social capital scores for 2,992 of 3,142 counties, containing 99.7 percent of the American population. Just eight percent of Americans live in the top fifth of these counties, while 39 percent of the population lives in the bottom fifth of counties. Nearly six in ten (59 percent) of Americans live in the bottom two fifths of counties, compared with 24 percent living in the top two fifths.
  • The 12 states with the highest social capital scores are distributed across two continuous blocs: nine states running from Utah, through Wyoming and Colorado, across the Dakotas and Nebraska, and over to Iowa, Minnesota, and Wisconsin; and the three Northern New England states of Maine, New Hampshire, and Vermont. These states tend to rank highly across all seven subindices as well. Utah has the highest social capital score, followed by Minnesota and Wisconsin.
  • Of the 11 states with the lowest levels of social capital, ten of them fall within a contiguous bloc of states running from Nevada, across the Southwest and South over to Georgia and Florida. New York is the only state in the bottom 11 that is outside this group. Louisiana has the lowest social capital score, followed by Nevada, Arizona, and New Mexico.
  • Of the nine states ranked just above this bottom group, seven border and extend the southern bloc, filling out most of the rest of the South. The 17 southern states in the bottom 20 are home to 45 percent of Americans and 74 percent of Americans in bottom-fifth counties. Six in ten (59 percent) of people in the 17 states live in bottom-fifth counties. Only 17 of 1,338 counties in these states are in the top fifth.
  • Our indices are not dominated by any single subindex, and our state and county indices appear to be approximating social capital in the same general way.
  • Among the component variables underlying the state index, the strongest associations with the index itself across states were for the volunteer rate (0.86), heavy television watching by children (-0.81), the share of adults who made charitable contributions (0.80), the share with emotional and social support (0.80), heavy usage of electronics among children (-0.77), the share of adults that are married (0.75), the share of children living with a single parent (-0.72), and the share of births that were to unwed mothers (-0.71).
  • At the county level, the highest correlates of social capital were violent crime (-0.73), the share of children with a single parent (-0.71), the share of adults currently married (0.69), voting rates (0.59), and nonprofits plus congregations (0.57).
  • Despite the outsized role that religious communities have played in social capital investment, indicators of religious adherence and commitment were generally weakly (or even negatively) correlated with our social capital scores, both at the state and county levels. This may suggest that social capital organized around religion may be displaced by secular sources of social capital, that the availability of resources provided by secular social capital weakens religious commitment, or that people in distressed places turn to religious communities for the support that is missing in other parts of their lives. This question is a subject for future Social Capital Project research.
  • Our social capital indices correlate strongly with earlier social capital indices across states and counties, and with other indices such as the Family Prosperity Institute’s Family Prosperity Index, Opportunity Nation’s Opportunity Index, and the Economic Innovation Group’s Distressed Communities Index.
  • We show the correlations of our indices and subindices with 59 state-level and 50 county-level benchmarks reflecting a range of economic, social, demographic, educational, health, and other outcomes.
  • Our index is a clear improvement on the Penn State index, based on this benchmarking, but remarkably, Robert Putnam’s state index from Bowling Alone, published nearly two decades ago, has slightly higher benchmark correlations than ours. Because our index captures the health of family life, and because it is based on up-to-date and freely available data (including at the county level), we still prefer it to the Putnam measure. The fact that the correlation between the two state-level indices is 0.81 reassuringly suggests that very different approaches to social capital measurement capture the same essential construct.

Introduction: Why Build a Social Capital Index?

Discussions about American society, to the extent they involve facts, revolve around problems reflected in economic, demographic, and political measures. What gets defined as a problem, which causes attract interest, what consequences are deemed worrisome, and how effective are attempted remedies—all of these depend on having reasonable measurements of the things under study.

The result is that our understanding of the world is framed by measureable problems, causes, and consequences, and is less attuned to those that are more difficult to measure. For example, the development of gross domestic product (GDP) is one of the great successes in the history of measurement. 1 But today’s debates are often hindered by the imbalance between well-measured economic variables such as GDP and less well-measured social, cultural, and psychological ones.

Social capital—the aspects of our relationships that produce benefits for us—falls into this second group. Economic factors and outcomes are important, but if we neglect the health of our associational life, we will misdiagnose the causes of many problems and tend to focus on economic priorities over social ones. Measuring “social capital,” however, is no simple matter. Different people—different researchers—use the phrase to mean different things. And many aspects of what gets lumped under “social capital” that are quantifiable are infrequently included in household surveys or administrative data.

Yet, the various attributes and resources to which “social capital” refers are likely to be important. It is incumbent on researchers to develop high-quality measurements of social capital, as well as the more specific things to which it refers. Absent these measures, policymakers will never have a complete picture of how the nation is faring.

This report introduces a new index of social capital and describes its construction. It presents state- and county-level estimates of social capital and its subcomponents. Finally, it assesses the extent to which these measures correlate with a range of social, economic, demographic, and other benchmarks. We are providing the data behind our indices and subindices; it is our hope that they will be used by other researchers and policymakers to gain a more complete and accurate picture of the nation’s challenges.


What is “Social Capital”?

As discussed in our flagship report, “What We Do Together,” the basic idea of social capital as something important that is related to social relationships, social networks, and civil society has a long history. 2 The reference to capital suggests that key to the concept is the conjecture that aspects of our associational life are productive for us. 3 Some scholars have described social capital as inhering in our social networks, as an attribute of collectives. 4 Communities may be said to have more or less productive social capital, or social capital that is differentially productive for the particular ends valued by community members. Others have put the focus on the individual, so that a person’s social capital may be characterized as more or less productive for them. 5 These different emphases may be reconciled by positing social capital as a feature of individual relationships, so that an individual’s social capital is typified by the aggregation of her relationships, and community-wide social capital is the aggregation of all the relationships across members.

But what actually constitutes social capital is not consistently defined across researchers. For example, consider “trust.” Is trust an element of social capital—a characteristic intrinsic to relationships that is productive—or is it the consequence of a community having productive social capital (something that social capital produces)? Depending on the researcher, social capital may or may not include the content of relationships, the structure of relationships, or the number of relationships.

It is also likely that different elements of social capital—networks or shared values, for instance—have different causes and effects. And different forms of associational life—families, communities—may be more or less important as incubators of social capital. Different aspects of social capital may even be in tension with each other; social-capital-building within families can come at the expense of social capital investment in neighbors, for instance.

We take a pragmatic approach to these issues. In our understanding of social capital, close and nurturing relationships with other people almost self-evidently provide benefits. Therefore social capital is likely to be “greater” or more productive in families, communities, and organizations with an abundance of close, supportive relationships. Social capital is also likely to be reflected in cooperative activities. These activities may be informal (e.g. conversing or working together with neighbors), or formal (e.g. membership in groups or service on a committee). Some cooperative activities may be formalized in institutions (e.g. governments, schools, news media, corporations), including nonprofit organizations specifically meant to deliver benefits or to represent interests. Social capital is also reflected in trust in other people, confidence in institutions, mutual generosity, high collective efficacy, and low social disorganization.

In our view, places where these features of social life come together have “high,” or “more,” or more productive social capital—features of social life that provide benefits to community and family members. Places with a dearth of these features have “low,” or “less,” or less productive social capital. We try to minimize the extent to which “social capital” reflects value judgements; what is productive social capital for some—criminal networks, for instance—may appear to others to be deeply problematic. Many of the indicators that go into our index are about the extent to which people do things together, without regard to what they are doing.

Nevertheless, there is no getting around the fact that any specific way of measuring social capital will involve normative considerations as to what to include or exclude. And other ambiguities are unavoidable. Our index takes a high violent crime rate as reflecting low social capital—a diminished ability to maintain social order—but it could also reflect tight and effective social networks taking the form of gangs.

Our conceptualization of social capital keeps associational life central. Two implications follow from this focus. First, our index affords greater importance than is often given to family relationships. The family is ultimately the most intimate form of social life, and the bedrock for other social capital investment. Second, while our index includes various measures of “civic engagement,” it excludes those indicators of civic engagement that do not involve associational life. For example, we ignore measures of civic or political knowledge, as well as those that emphasize following current events or news. In this way, we try to draw intuitive boundaries around the concept of social capital.


Past Efforts to Develop a Social Capital Index

Ours is not the first effort to construct an index of social capital. Robert Putnam’s foundational 2000 book Bowling Alone featured a state-level index. 6 It included 14 indicators in five categories: community organizational life, engagement in public affairs, community volunteerism, informal sociability, and social trust. Putnam’s index was a simple average of the 14 scores (after standardizing them to put them on a common scale). These measures covered the second half of the 1970s, the 1980s, and the first half of the 1990s, but generally not the same years.

The surveys that Putnam consulted for these data were not always designed to be representative of every state, however. That is to say, some surveys are designed so that the participants are broadly representative only of the American population. Those surveys will include many people from many states, but for any given state, it is not necessarily the case that the participants represent the state’s population well. Further, the measures are out of date, since Bowling Alone was published in 2000, and updating the index would require purchasing data that is not otherwise publicly available.

In a 2000 paper, economists Alberto Alesina and Eliana La Ferrara included a state-level map displaying social capital index levels, divided into four unequally-sized categories. 7 They used measures of group participation, trust, and presidential election voting rates, all from the General Social Survey (GSS). Unfortunately, the GSS is not designed to be representative of each U.S. state; it is representative of the nation as a whole.

In a 2006 paper, Daniel Kim and several coauthors updated Putnam’s work and created two state-level social capital indices from 10 of Putnam’s indicators. 8 This smaller group still represented all five of Putnam’s original categories. One index included community volunteerism, informal sociability, and social trust, and the other included engagement in public affairs. Both indices included community organizational life. As was the case in Bowling Alone, some of the data comes from surveys that were not designed to represent every state. In a subsequent paper, Kim and Chul-Joo Lee created another state-level index, using the Annenberg National Health Communication Survey (covering 2005-2008). 9 The index indicated the average number of formal and informal groups, out of 15 different types, in which adults participated. However, this survey, like the GSS, was not designed to be representative of each state.

Also in 2006, the National Conference on Citizenship, in association with the Center for Information and Research on Civic Learning and Engagement and the Saguaro Seminar, introduced a “Civic Health Index.™” 10 It was comprised of 40 indicators, grouped into nine categories. Most of these categories are clearly related to social capital: “connecting to civic and religious groups,” “trusting other people,” “giving and volunteering,” “connecting to others through family and friends,” “participating in politics,” and “trusting and feeling connected to major institutions.” Others, however, while reasonable in an index of civic health, reflect social capital much less directly, including “staying informed,” “understanding civics and politics,” and “expressing political views.”

The Civic Health Index™ generally weights all of the indicators within a category equally and then weights the category scores equally to compute the index. Index values were estimated at the national level from 1975 to 2004. The index declined by over seven percent from 1975 to 1995, then made up over half of that decline by 2003. No state or county estimates are available.

The Legatum Prosperity Index™ has been assessing nations around the world since 2007, and beginning in 2008, social capital has been represented via a social capital subindex. 11 This subindex has changed over time, but among the indicators included have been donations, volunteering, membership in groups, trust, helping strangers, marital status, importance of religion and friends, having reliable friends, voter turnout, voicing opinions, and being treated with respect. The most recent index compares 149 countries, but no data are available at the state or county level. The social capital subindex relies on Gallup data, which must be purchased.

The most influential social capital index in recent years has been one originally produced by Anil Rupasingha, Stephan Goetz, and David Freshwater and subsequently updated by Penn State University’s Northeast Regional Center for Rural Development. 12 This index is available at the county level—the first available at this level of disaggregation. The most recent version incorporates data on presidential voting rates in the 2012 election, mail-in response rates for the 2010 decennial census, domestically-oriented non-profit organizations per capita, and group membership organizations and recreational establishments (“associations”) per capita. We discuss some shortcomings of this index in detail in the next section.


The Need for a New Social Capital Index

This brief review highlights the shortcomings of these previous sources of social capital estimates. Several indices rely on data that is out of date. A number of them either include a limited range of social capital indicators or include indicators that are less obviously about social capital. In particular, the health of family life is under-examined by the earlier measures. And some depend on data sources that are not freely available.

Several of the indices do not provide estimates below the national level. Of the indices available at the state level, all rely at least in part on surveys that cannot be assumed to represent state populations well.

Only the Penn State index provides county-level estimates. But after researching the index, we were dissatisfied. The Penn State indicators tap a limited range of the concepts invoked by “social capital.” The index includes nothing on family health, volunteerism, charitable giving, informal community engagement, social support, or collective efficacy. Presidential voting and census mail-in rates are tenuous indicators of social capital, as they relate people primarily to federal, rather than local institutions.

With only four indicators, problems in any one of them can seriously affect the resulting index. Several of the indicators suffer from interpretive or data issues. For instance, places with many nonprofit organizations may have high civic engagement, but that might also simply reflect that they have a lot of problems to address. In addition, to the extent that nonprofit organizations are professionally run, they may actually crowd out informal volunteerism and a sense of obligation to fellow community members. Further, the data used by Penn State are from IRS registrations, and a large number of religious nonprofits are not required to register. (Nor are the smallest nonprofits.) 13 Some faith communities, such as the Church of Jesus Christ of Latter-day Saints (also known as the "LDS" or "Mormon" Church), deliberately oppose registration. 14 One study of Indiana found that registered nonprofits in the IRS data included only 60 percent of nonprofits they were able to identify from other sources. 15

The establishment data only counts places with paid employees and an Employer Identification Number (EIN). The distinction between an “establishment” and an organization relying on voluntary service is potentially a profound problem. For example, in the 2015 establishment data, Utah has just 658 religious establishments. But data on religious congregations (described below) indicates that in 2010 there were over 5,500 congregations in the state. 16 The reason for the discrepancy may be that the organization of the Mormon Church relies on volunteers rather than on employees. Or it may be that because religious organizations often are not required to register with the IRS, many do not have EINs. The distinction between an “establishment” and an organization relying on voluntary service also likely affects non-religious nonprofit organizations, such as parent-teacher organizations and civic membership groups.

Finally, the establishments measure constructed by the Penn State researchers includes a variety of athletic and recreational establishments, including golf courses, fitness centers, and bowling alleys. While those kind of inherently social establishments reflect social capital, they are very different than the membership organizations otherwise counted in their establishment measure (including labor unions, political organizations, civic organizations, and the like). Further, what the researchers have included and excluded seems arbitrary. Left out are movie theaters, theater and dance companies, racetracks, zoos, theme parks, arcades, casinos, skiing facilities, museums, libraries, bars, and dance clubs.

When we compared the Penn State index to a variety of benchmark indicators, it was only moderately or weakly related to them. The correlation of the most recent version of the index with county poverty rates is -0.34, for instance (where -1.00 would indicate, roughly, that variation in social capital completely explains variation in poverty rates). As we will see below, our index is more strongly correlated than the Penn State index is with most of our benchmarks, often much more strongly.

We confirmed we could replicate the Penn State index independently, which revealed that census response rates were actually negatively correlated with the Penn State social capital score. 17 That was another red flag, since the hypothesized relationship—the reason for its inclusion in the index—was that higher response rates indicate greater social capital (i.e., stronger norms regarding the responsibilities of citizenship, or greater confidence and trust in the federal government).

We also estimated a corresponding state-level index using Penn State’s approach, and this time all four indicators were positively correlated with the index. 18 The state-level correlations with our benchmarks were stronger than the county-level ones, but still lower than what we expected. In particular, when we looked at how the state we know best, Utah, was ranked along the Penn State index and its components, we saw large discrepancies with other research. For instance, Utah is ranked first on the Family Prosperity Initiative’s Family Prosperity Index. 19 According to the U.S. Religion Census, administered by the Association of Statisticians of American Religious Bodies, Utah has the highest rate of religious adherence in the country, and it is ranked 7th in terms of congregations per capita. 20 Similarly, research using the Current Population Survey (CPS) indicates that Utah has the highest rate of religious volunteerism, but it also ranks the state 15th in terms of secular volunteerism. 21 Putnam’s index from Bowling Alone ranks it at 14th in terms of social capital, and it is in Alesina and La Ferrara’s top group of nine states.

Yet, the Penn State index ranks Utah 20th in terms of census response rates, 45th in terms of presidential voting rates, second worst in terms of nonprofits, and worst in terms of associations. Given these rankings, Utah ranks worst overall on the Penn State index. 22 The establishment data that is the basis for one of the four inputs into the Penn State index ranks Utah last in the nation in terms of religious organizations per capita.

With such large state-level discrepancies, it is hard to imagine that the county-level Penn State index is reliable for all but the most disparate comparisons. Our conclusion was that a better social capital index was needed than those currently available.

Constructing a New State Social Capital Index

For the better part of the past year, the Social Capital Project has been gathering county- and state-level data on a range of social, economic, demographic, health, religious, and other indicators. Broadly speaking, we looked for indicators related to family structure and stability, family interaction and investment, civil society, trust and confidence in institutions, community cohesion, institutions, volunteerism, and social organization. There are not many surveys that provide such variables using samples designed to represent every state or county. Nor are there many censuses or administrative data sources that capture the entire population of interest in all states or counties. The limited availability of data was a fundamental constraint that removed much of the hard work that otherwise might have gone into choosing among many dozen possible indicators.

Starting from around 20 county-level measures and an additional 50 state-level indicators, we eventually settled on seven at the county level and 25 at the state level. These indicators are from data collected by various sources between 2006 and 2016, primarily from 2013 forward. The details of how we selected these variables can be found in the Appendix. Table 1 provides an overview of the indicators and their derivation. Appendix Table A1 displays the full range of variables we considered and gives their sources.

Our state social capital index includes seven dimensions, represented by five subindices and two stand-alone indicators. These dimensions were chosen partly out of data constraints, but we also considered the ways in which past researchers had theorized about social capital and associational life. 23 We then combined these seven component variables to create an index score for each state.


Subindices

We transformed the original values of each indicator to “standard scores,” by subtracting the mean value and dividing by the standard deviation of the indicator’s distribution (a measure of variation). The mean of each standard score is zero, and the standard deviation is one. Standardizing the scores puts them on a comparable scale, allowing us to combine multiple indicators despite their initially having different distributions (including different minimum and maximum values). We reversed the polarity of certain measures, such as the share of births to single mothers, so that a larger positive standard score always corresponded with “more” social capital.

Each subindex is comprised of a weighted sum of standard scores. Simply adding multiple standard scores gives them equal weight in contributing to the subindex, but we wanted indicators to be weighted more or less depending on how well they reflected the concept embodied in the subindex. The weights are selected through a statistical technique called “principal components analysis” (PCA). Specifically, the weights are estimated so that the resulting subindex accounts for the maximum possible “variance,” or variability, across the original scores. Some information in the original set of indicators is lost by using this “first principal component score” as the subindex, but the loss is minimized versus any other set of weights. It is analogous to finding the best angle from which to photograph a three-dimensional object so that the two-dimensional rendering retains the most information.

In the domain of family health, we created a “family unity” subindex and a “family interaction” subindex. The family unity subindex combines state-level data from the American Community Survey (2012-2016) on the share of births that are to unwed mothers (weight of 0.57), the percentage of children living in families headed by a single parent (0.60), and the percentage of women ages 35-44 who are married (and not separated) (0.57). (The weights could, in theory, range between -1.0 and 1.0, and they reflect the extent to which an indicator is correlated with the subindex itself. Ideally, the weights should be sizable and should all have the same sign.) 24 The subindex accounts for 91 percent of the variability across the original three variables of which it is comprised.

The family interaction subindex combines data from the 2016 National Survey of Children’s Health on the share of children ages 0-5 read to every day by a family member (weight of 0.47), the share of children who watch television or videos or play video games at least four hours a day (0.65), and the percentage of children who use computers, cell phones, and other electronic devices for purposes other than schoolwork at least four hours a day (0.60). Our family interaction subindex accounts for 68 percent of the variability across the original three variables.

We created a social support subindex, comprised of several indicators from multiple sources. It includes the share of adults who sometimes, rarely, or never “get the social and emotional support [they] need,” taken from 2006 and 2010 Behavioral Risk Factor Surveillance System data (weight of 0.50). It also includes, from the 2013 Civic Engagement Supplement to the CPS, the share of adults who do favors for neighbors at least once a month (0.49) and the share who trust most or all of their neighbors (0.54). The last indicator is the average number of “close” friends that adults report having in the 2008 Civic Engagement Supplement to the CPS (0.47). The resulting index accounts for 70 percent of the original variability across the four variables that comprise it.

Our community health subindex incorporates information on the share of adults who reported volunteering for an organization in the past year (weight of 0.33), the share who attended a public meeting to discuss community affairs (0.38), and the share who worked with neighbors to improve the community (0.39), all from the 2015 Volunteer Supplement to the CPS. It also includes the share of adults who served on a committee or as an officer of a group in the past year, from the 2013 Civic Engagement Supplement to the CPS (0.38). From the 2008 Civic Engagement Supplement to the CPS, we include the share who attended a public meeting where political issues were discussed (0.39) and the share who participated in a march, protest, rally, or demonstration (0.29). Our community health subindex accounts for 65 percent of the variability across eight indicators.

Further, we estimate, from 2015 County Business Pattern data on establishments, membership organizations per capita (weight of 0.30). Finally, we include a measure of non-profit organizations per capita (weight of 0.36). This was created by summing registered non-religious not-for-profit organizations per capita and religious congregations per capita. The former is from the December 2015 Internal Revenue Service Business Master File (accessed through the Urban Institute’s National Center for Charitable Statistics). Most faith-based organizations, excepting colleges and health care organizations, are not required to register with the IRS. Only half of religious congregations do so, and the share varies by congregation. 25 We therefore added congregations per capita from the 2010 Religious Congregations and Membership Study, conducted by the Association of Statisticians of American Religious Bodies and accessed through the Association of Religion Data Archives.

Finally, we created an institutional health subindex. This subindex combines the rate at which citizen adults of voting age cast ballots in the 2012 and 2016 presidential elections (averaged over the two years, weight of 0.38), the rate at which residents returned the 2010 decennial census questionnaire through the mail (0.44), and the share of adults with “great” or “some” confidence in corporations (0.49), the media (0.38), and public schools (0.53) to do what is right. The voting data is from the U.S. Election Assistance Commission annual reports, the census response rates are from the Census Bureau, and the confidence measures are from the 2013 Civic Engagement Supplement to the CPS. The institutional health subindex accounts for 48 percent of the variability across the original five indicators. The lower proportion that it explains relative to the other subindices may reflect the weaknesses in the presidential voting and census response indicators discussed above.


Stand-Alone Indicators

We included in our state-level social capital index two stand-alone indicators to represent two other dimensions of associational life. The violent crime rate was included to reflect the level of “collective efficacy” (or conversely, of social disorganization). The idea is that communities high in social capital are better positioned to informally police their own community and enforce pro-social norms, and their residents are less likely to do harm to one another. 26 Violent crimes are better reported than crimes generally, which is why we do not use a broader measure. The source for this measure was the Federal Bureau of Investigation’s Uniform Crime Reporting Program.

The second stand-alone indicator reflects philanthropic health—the share of adults who gave more than $25 in the past year to “charitable or religious organizations.” By setting the threshold for giving low, this measure ensures that cross-state differences are not driven by income concentration at the top, where charitable giving is also somewhat concentrated. 27 This measure comes from the 2015 Volunteer Supplement to the CPS.


Combining the Dimensions of Social Capital

The final step was to create the index itself. We standardized the five subindex scores and the two stand-alone indicators to put them all on a common scale. We then ran principal components analysis on these seven variables to create weights for each of them: family unity (0.38), family interaction (0.41), social support (0.45), community health (0.33), institutional health (0.36), collective efficacy (0.28), and philanthropic health (0.40).

Each state’s social capital index score was computed by taking the weighted sum of the seven standard scores and then standardizing this weighted sum. Index scores range from -2.2 to 2.1; a score of, say, 1.5 means that a state lies one-and-a-half standard deviations above the mean index score across states. Roughly, its social capital levels are higher than the average state’s social capital by an amount 1.5 times the typical gap between a state and the average.

There is an unavoidable element of arbitrariness in creating a one-dimensional index to reflect a concept as complex and diffuse as “social capital.” The usefulness of our index depends on its ability to represent a potentially important factor affecting a range of social, economic, and health outcomes. The index represents a “noisy” measure of a fuzzy concept. But it reflects those aspects of its constituent indicators that all measure the same “thing,” and lets them contribute to the measure insofar as they reflect that thing. Our social capital index accounts for 56 percent of the variability across the two stand-alone indicators and the five subindices (each of which accounts for 48 to 91 percent of the variability across its constituent indicators).


Constructing a New County-Level Social Capital Index

While the state index includes seven dimensions of social capital, our county index includes only four because fewer variables are available at the county level. These include three subindices—two of which contain their own subindex—and one stand-alone indicator. The construction of the county-level index is more complicated than for the state-level index. Table 2 shows the variables that go into the index. (See Appendix Table A1 for the full list of county-level variables we considered.)


Subindices

As when we created the state index, we transformed all original values to standard scores, though this time the mean and standard deviation applied to the distribution of values across counties rather than across states. As with the state-level subindices, our county-level subindices are weighted sums of standard scores, with the weights determined through PCA. These weights need not be the same as those produced from state-level data, where it is variation across states that is being analyzed.

We created the same family unity subindex as we did at the state level. The subindex combines county-level data from the American Community Survey (2007-201128 and 2012-2016) on the share of births that are to unwed mothers (weight of 0.52), the percentage of children living in families headed by a single parent (weight=0.62), and the percentage of women ages 35-44 who are married (and not separated) (weight=0.59). Reassuringly, these weights are very similar to those produced in the state-level analyses, suggesting that the state- and county-level subindices are measuring the same underlying construct. It does explain less of the variability in the original three variables than the state-level subindex does—73 percent instead of 91 percent.

We also created a community health subindex, though due to data availability issues, the county subindex incorporates less information than the corresponding state-level one. We were concerned that the available county-level indicators of community health did not fully capture the underlying concept. In particular, we lacked the CPS indicators of informal civil society and activities requiring a time commitment that were available at the state level—working together with neighbors, attending public meetings, serving on committees or as officers, volunteering, attending political meetings, and participating in demonstrations. We worried about this omission, in particular, because professionalized services offered through membership organizations and other nonprofit groups might be expected to crowd out informal and time-intensive volunteer activities, potentially leaving the stock of social capital thinner than it might have been. Inherently, formal organizations that serve members’ or clients’ interests allow people to “farm out” social capital activities. To include only a measure of the health of formal organizations would penalize places where community involvement is more informal.

To resolve this concern, we first went back to the state data and created a new subindex of “informal civil society” for each state. The subindex score was the first principal component score combining the six CPS variables above. 29 We then assigned this subindex score to every county within a state. In other words, the only variation in the subindex score is between states, and all counties within a state get the same score.

Next, back in the county data, we created five different candidate subindices, using different combinations of the informal civil society subindex score, membership organizations per capita, non-religious non-profit organizations per capita, congregations per capita, and the combination of non-religious non-profits and congregations. We computed, for each candidate subindex, the population-weighted average subindex score across a state’s counties. Then we correlated each of these state averages with the state-level community health subindex. We selected the subindex out of the five candidates that produced the strongest correlation.

The final county-level community health index combines non-religious non-profits per capita (weight of 0.70), congregations per capita (0.48), and the informal civil society subindex (0.53). The population-weighted average of this subindex across a state’s counties correlated at 0.97 with the state-level community health subindex. For context, the correlation of the state-level subindex with the version of the county community health subindex we favored prior to adding in the informal civil society subindex was 0.75. The county-level community health subindex accounts for 55 percent of the variability in the three original variables that go into it.

Finally, we included an institutional health subindex. As with the community health subindex, we were concerned about the incomplete data we had at the county level. In this case, we lacked information about confidence in institutions. We took the same approach as for community health. In the state data, we created a confidence subindex that included the three institutional confidence variables. 30 We assigned every county in a state the state’s subindex score. Then we created three versions of a county-level institutional health index, using different combinations of presidential voting rates, census response rates, and the confidence subindex.

As before, we created population-weighted state averages across a state’s counties and compared them to the state-level institutional health index. The version that correlated most strongly included presidential voting rates (weight of 0.63), census response rates (0.41), and the confidence subindex (0.66), accounting for 44 percent of the variability in those three measures. 31 The correlation of the population-weighted state average across counties with the state-level institutional health subindex was 0.98.

We did not attempt to create subindices at the county level for family interaction or social support, lacking data. 32


Stand-Alone Indicator

The county-level social capital index includes one stand-alone indicator. As for the state-level index, the violent crime rate was included to reflect the level of collective efficacy in a county. It comes from the Federal Bureau of Investigation’s Uniform Crime Reporting Program.

The charitable giving measure from the CPS is not available at the county level, so it is not included as a stand-alone indicator.


Combining the Dimensions of Social Capital

Computing the county-level index was also a bit more involved than for the state index. We standardized the three subindex scores and the collective efficacy stand-alone indicator to put them all on a common scale. We then ran PCA on these four variables. The weights were 0.53 for family unity, 0.47 for community health, 0.49 for institutional health, and 0.51 for collective efficacy. We took the weighted sum of the four standard scores to get the first iteration of the index, which accounted for 51 percent of the variability in the original four constituent measures.

However, information on violent crime rates was missing for 178 counties (out of 3,142). We were able to compute scores for 103 of these counties by creating an alternative index that left out violent crime. (The weights were 0.55 for family unity, 0.56 for community health, and 0.62 for institutional health. The subindex accounted for 56 percent of the variability in the three original variables.) Where a county lacked a score using the original index, we gave it the score on the alternative index. These two indices were correlated with each other at 0.94, so where states ranked on one was largely where they ranked on the other.

The final county-level index scores range from -4.3 to 2.9, indicating greater dispersion than exists across states.

To assess how the county-level and state-level indices might differ from one another, we created another state-level index using only the three subindices and the stand-alone violent crime indicator that are in the county index. This index correlated with the fuller state-level index at 0.96. 33 We also computed for each state the population-weighted average across counties of the county-level social capital index. The correlation between it and the state-level social capital index was 0.95, and the correlation between it and the state-level index based on the county-level methods was 0.98. Thus, the thinner county-level index likely ranks counties very similarly to the way in which the fuller state-level index would rank them.

In sum, our state index captures a fuller set of social capital indicators than any previous effort. We could not find a reliable measure of generalized trust at the state or county levels, but we believe we have covered most of the essential domains discussed by past social capital theorists. We considered including measures of segregation by race and income in our indices. The idea is that places where different types of people largely live apart are likely to be missing out on some benefits of social capital. However, research suggests that more diversity actually tends to reduce levels of social capital. 34 Intuitively, it is more difficult to engage with people when they are not “like” us. In the end, we decided not to incorporate segregation into our indices. We view segregation as having an indeterminate effect on levels of the many dimensions of social capital. It seems more likely that segregation affects the distribution of social capital within a state or county. 35


Findings

Table 3a lists the state social capital index ranks and the rankings on the individual subindices of the index. Table 3b lists the county social capital index and subindices as national percentiles. Figure 1 displays the state social capital scores in a map, and Figure 2 displays the county-level data. We have social capital scores for 2,992 of 3,142 counties, containing 99.7 percent of the American population. Before examining the places with the highest and lowest social capital scores, we provide some initial details about the distribution of social capital in America.


Click a column to reorder by index or subindex state ranking (1 is best, 51st is worst). For county rankings within states, click a county in Figure 2. Use the search box to jump to a specific state.

Source: Social Capital Project. Download Data (xlsx)


Table 3b. County-Level Index and Subindices as National Percentiles

Click a column to reorder by index or subindex county percentile (100th percentile is best, 0 is worst). For county rankings within states, click a county in Figure 2. Use the search box to jump to a specific county or state.

Source: Social Capital Project. Download Data (xlsx)


The maps in Figures 1 and 2 display states and counties broken out into five (roughly) equally-sized groups—ten states per group and 598 counties. 36 These groups do not contain the same number of people, however. The states with the lowest social capital include 29 percent of the nation’s population, while the top grouping is home to just nine percent of Americans. Over half the population (56 percent) is in the lowest two groups of states, while 21 percent is in the top two groups. At the county level, 39 percent of the population in non-missing counties lives in the bottom fifth, while just eight percent lives in the top fifth. Nearly six in ten (59 percent) of Americans live in the bottom two fifths of counties, compared with 24 percent living in the top two fifths.

Across states, the social capital scores are strongly correlated with each subindex. The correlations are 0.89 for social support subindex scores, 0.82 for family interaction, 0.80 for philanthropic health, 0.76 for family unity, 0.72 for institutional health, 0.65 for community health, and 0.55 for collective efficacy. The 21 correlations between the seven subindices are all positive, except that community health and collective efficacy, surprisingly, are correlated at -0.11. Otherwise, the correlations range from 0.17 (family unity and community health) to 0.74 (family interaction and community health).

At the county level, social capital scores are also strongly correlated with all four subindex scores. The correlations are 0.76 for the family unity subindex, 0.73 for collective efficacy, 0.71 for institutional health, and 0.65 for community health. The fact that these correlations are all fairly strong means that our state and county indices do not simply reflect a single dimension driving the results. The correlations between the four subindices range from 0.24 (family unity and community health) to 0.47 (family unity and collective efficacy).

Diving deeper into the components of the indices at the state level, the indicators with the strongest correlation to social capital were the volunteer rate (0.86), heavy television watching by children (-0.81), the share of adults who made charitable contributions (0.80), the share with emotional and social support (0.80), heavy usage of electronics among children (-0.77), the share married (0.75), the share of children living with a single parent (-0.72), and the share of births that were to unwed mothers (-0.71). While not included in the index, the share who trust most of their neighbors was correlated at 0.86 with it. At the county level, the highest correlates of social capital were violent crime (-0.73), the share of children with a single parent (-0.71), the share of adults currently married (0.69), voting rates (0.59), and nonprofits plus congregations (0.57). The importance of the absence of many of the key state-level variables at the county level is evident.

A few state-level indicators had low correlations with the index, including membership organizations per capita (0.07), confidence in the media to do what is right (0.20), having participated in a march or demonstration (0.21), and non-religious non-profits and congregations per capita (0.29). Three of these indicators go into the community health subindex, which may explain why it is less strongly correlated with social capital scores than most of the other subindices. Relatedly, the Penn State social capital index relies on variants of the membership organization and non-profits indicators. Our replication of the Penn State index correlates only at 0.37 with our index, as we will see below. At the county level, census response rates—one of the four Penn State components—was correlated with our index at only 0.26, but the correlation between the 2014 Penn State index and our county index was 0.56. We view this as evidence that the relatively thin county-level indices do not measure social capital as strongly as our richer state-level index does.


Figure 1. Social Capital Index and Subindex Scores by State

Coming Soon: Social Capital Index data at the metropolitan/micropolitan, commuting zone, and congressional district levels. Click here to open in a new window