Ethnic diversity is increasing in most advanced countries, driven mostly by sharp increases in immigration. In the long run immigration and diversity are likely to have important cultural, economic, fiscal, and developmental benefits. In the short run, however, immigration and ethnic diversity tend to reduce social solidarity and social capital. New evidence from the US suggests that in ethnically diverse neighbourhoods residents of all races tend to ‘hunker down’. Trust (even of one's own race) is lower, altruism and community cooperation rarer, friends fewer. In the long run, however, successful immigrant societies have overcome such fragmentation by creating new, cross-cutting forms of social solidarity and more encompassing identities. Illustrations of becoming comfortable with diversity are drawn from the US military, religious institutions, and earlier waves of American immigration.
One of the most important challenges facing modern societies, and at the same time one of our most significant opportunities, is the increase in ethnic and social heterogeneity in virtually all advanced countries. The most certain prediction that we can make about almost any modern society is that it will be more diverse a generation from now than it is today. This is true from Sweden to the United States and from New Zealand to Ireland. In this article, I want to begin to explore the implications of that transition to a more diverse, multicultural society for ‘social capital’– the concept for which I have been honored by the Skytte Prize committee.1
I begin with a word or two about this concept, which has been the subject of an exponentially expanding and controversial literature over the last fifteen years. I prefer a ‘lean and mean’ definition: social networks and the associated norms of reciprocity and trustworthiness.2 The core insight of this approach is extremely simple: like tools (physical capital) and training (human capital), social networks have value. Networks have value, first, to people who are in the networks. For example, economic sociologists have shown repeatedly that labor markets are thoroughly permeated by networks so that most of us are as likely to get our jobs through whom we know as through what we know. Indeed, it has been shown that our lifetime income is powerfully affected by the quality of our networks (Granovetter 1973, 1974; Burt 1992, 1997; Lin 1999, 2001). Similarly, much evidence is accumulating about the health benefits of social ties (House et al. 1988; Berkman 1995; Seeman 1996; Berkman & Glass 2000).
What makes social networks even more interesting, however, is that they also have implications for bystanders. For example, criminologists have taught us the power of neighbourhood networks to deter crime (Sampson et al. 1997; Sampson 2001). My wife and I have the good fortune to live in a neighbourhood of Cambridge, Massachusetts, that has a good deal of social capital: barbecues and cocktail parties and so on. I am able to be in Uppsala, Sweden, confident that my home is being protected by all that social capital, even though – and this is the moment for confession – I actually never go to the barbecues and cocktail parties. In other words, I benefit from those social networks even though I am not actually in them myself. In the language of economics, social networks often have powerful externalities.
Social capital comes in many forms, not all fungible. Not all networks have exactly the same effects: friends may improve health, whereas civic groups strengthen democracy. Moreover, although networks can powerfully affect our ability to get things done, nothing guarantees that what gets done through networks will be socially beneficial. Al Qaeda, for instance, is an excellent example of social capital, enabling its participants to accomplish goals they could not accomplish without that network. Nevertheless, much evidence suggests that where levels of social capital are higher, children grow up healthier, safer and better educated, people live longer, happier lives, and democracy and the economy work better (Putnam 2000, Section IV). So it seems worthwhile to explore the implications of immigration and ethnic diversity for social capital.
In this article, I wish to make three broad points:
• Ethnic diversity will increase substantially in virtually all modern societies over the next several decades, in part because of immigration. Increased immigration and diversity are not only inevitable, but over the long run they are also desirable. Ethnic diversity is, on balance, an important social asset, as the history of my own country demonstrates.
• In the short to medium run, however, immigration and ethnic diversity challenge social solidarity and inhibit social capital. In support of this provocative claim I wish to adduce some new evidence, drawn primarily from the United States. In order to elaborate on the details of this new evidence, this portion of my article is longer and more technical than my discussion of the other two core claims, but all three are equally important.
• In the medium to long run, on the other hand, successful immigrant societies create new forms of social solidarity and dampen the negative effects of diversity by constructing new, more encompassing identities. Thus, the central challenge for modern, diversifying societies is to create a new, broader sense of ‘we’.
The Prospects and Benefits of Immigration and Ethnic Diversity
Figure 1 provides illustrative evidence that immigration has grown remarkably across the advanced nations of the world over the last half century. This chart shows the trends in six different countries, selected more or less at random, with quite different historical trajectories: the United States, Ireland, Sweden, Germany, Britain and France. Although these countries began at somewhat different starting points in the 1960s (France relatively higher, Ireland relatively lower), the general pattern is a clear convergence toward a much higher number of immigrants as a fraction of the total population.
Of course, not all immigrants are ethnically different from the native population: Danish immigrants do not significantly alter the ethnic mix in Sweden, nor do Canadian immigrants in the United States. Conversely, much of the ethnic diversity in the United States, especially black-white diversity, is entirely unrelated to immigration since the ancestors of most African-Americans have been in the United States longer than the ancestors of most white Americans. So diversity and immigration are not identical, and in our subsequent, more detailed analyses we will need to make that distinction more explicit and rigorous. Nonetheless, as a general rule, the mounting wave of immigration depicted in Figure 1 has increased ethnic diversity in the receiving countries. Moreover, because immigrant groups typically have higher fertility rates than native-born groups, ethnic diversity in virtually all of these countries would still increase in the years ahead, even if all new immigration were somehow halted (Smith & Edmonston 1997).
So our societies will inevitably be more ethnically diverse tomorrow than they are today. And that diversity will be a valuable national asset.3 It is not merely that national cuisine is enhanced by immigration, or even that culture of all sorts is enhanced by diversity, though culture and cuisine in my own country provide powerful evidence of those benefits.
• Creativity in general seems to be enhanced by immigration and diversity (Simonton 1999). Throughout history, for example, immigrants have accounted for three to four times as many of America's Nobel Laureates, National Academy of Science members, Academy Award film directors and winners of Kennedy Center awards in the performing arts as native-born Americans (Lerner & Roy 1984; Simonton 1999, Chapter 6; Smith & Edmonston 1997, 384–5). If we were to include second-generation immigrants (i.e. the children of immigrants), the contribution of immigrants would be even greater. Many (though not all) of the scores of studies of collective creativity in work groups (in business, education and so on) find that diversity fosters creativity (Webber & Donahue 2001; O’Reilly et al. 1997; Williams & O’Reilly 1998). Scott Page (2007) has powerfully summarized evidence that diversity (especially intellectual diversity) produces much better, faster problem-solving.
• Immigration is generally associated with more rapid economic growth. The economics profession has debated the short-run economic consequences of immigration for native workers. While there are important distributional effects to be considered, especially the impact of immigration on low-wage native workers in the US, the weight of the evidence suggests that the net effect of immigration is to increase national income. One recent study, for example, suggests that the income of native-born Americans rises more rapidly, ceteris paribus, if they are living in places with more immigrants than if they are living in places with fewer immigrants.4
• In advanced countries with aging populations, immigration is important to help offset the impending fiscal effects of the retirement of the baby-boom generation (Smith & Edmonston 1997, Chapters 6 and 7). In my country, for example, young immigrant workers (documented and undocumented) contribute financially to our Social Security system, but will not draw benefits for several decades, if at all, thus mitigating the otherwise unsustainable imbalance in the medium term between outflow and inflow into our national coffers.5 This effect is even more important in the more rapidly aging nations of Europe and East Asia.
• New research from the World Bank has highlighted yet another benefit from immigration, one of special relevance to the Nordic countries that have long played a disproportionate role on issues of global development. This new research suggests that immigration from the global South to the richer North greatly enhances development in the South, partly because of remittances from immigrants to their families back home and partly because of the transfer of technology and new ideas through immigrant networks. So powerful is this effect that despite ‘brain drain’ costs, increasing annual northward immigration by only three percentage points might produce net benefits greater than meeting all our national targets for development assistance plus cancelling all Third World debt plus abolishing all barriers to Third World trade (World Bank 2005; Pritchett 2006).
In short, immigration and multicultural diversity have powerful advantages for both sending and receiving countries. Yet what about the effects on social capital?
Immigration and Diversity Foster Social Isolation
In the theoretical toolkit of social science we find two diametrically opposed perspectives on the effects of diversity on social connections. The first, usually labelled the ‘contact hypothesis’, argues that diversity fosters interethnic tolerance and social solidarity. As we have more contact with people who are unlike us, we overcome our initial hesitation and ignorance and come to trust them more. Some of the most striking evidence in support of the contact hypothesis came originally from a famous study of the American soldier during the Second World War. White soldiers were asked how they would feel about having black soldiers serving in the same platoon with them. As Table 1 shows, among white soldiers who in fact had no contact with black soldiers, most opposed the idea. On the other hand, white soldiers who had been assigned to units with black soldiers were much more relaxed about the idea of racial integration (Stouffer 1949).
Table 1. Attitudes of White Soldiers in United States Army in the Second World War toward Racial Integration
Extent of contact with black troops
Percentage opposed to mixing black and white platoons in their company
Percentage opposed to a general policy of mixing black and white platoons
Same division, but not same regiment as black troops
Same regiment, but not same company as black troops
Same company as black troops
Evidence of this sort suggested to social psychologists, beginning with Gordon Allport in the 1950s, the optimistic hypothesis that if we have more contact with people of other ethnic and racial backgrounds (or at least more contact in the right circumstances), we will all begin to trust one another more.6 More formally, according to this theory, diversity reduces ethnocentric attitudes and fosters out-group trust and solidarity. If black and white children attend the same schools, for example, race relations will improve. This logic (and the sort of evidence presented in Table 1) was an important part of the legal case that led the United States Supreme Court to require racial desegregation in the famous Brown v. Board of Education case in 1954. For progressives, the contact theory is alluring, but I think it is fair to say that most (though not all) empirical studies have tended instead to support the so-called ‘conflict theory’, which suggests that, for various reasons – but above all, contention over limited resources – diversity fosters out-group distrust and in-group solidarity. On this theory, the more we are brought into physical proximity with people of another race or ethnic background, the more we stick to ‘our own’ and the less we trust the ‘other’ (Blumer 1958; Blalock 1967; Giles & Evans 1986; Quillian 1995, 1996; Brewer & Brown 1998; Taylor 1998; Bobo 1999; Bobo & Tuan 2006).
The evidence that diversity and solidarity are negatively correlated (controlling for many potentially confounding variables) comes from many different settings:
• Among Peruvian micro-credit cooperatives, ethnic heterogeneity is associated with higher default rates; across Kenyan school districts ethno-linguistic diversity is associated with less voluntary fundraising; and in Himalayan Pakistan, clan, religious, and political diversity are linked with failure of collective infrastructure maintenance (Karlan 2002; Miguel & Gugerty 2005; Khwaja 2006).
• Across American census tracts, greater ethnic heterogeneity is associated with lower rates of car-pooling, a social practice that embodies trust and reciprocity (Charles & Kline 2002).
• Within the Union (northern) Army in the American Civil War, the casualty rate was very high and the risks of punishment for desertion were very low, so the only powerful force inhibiting the rational response of desertion was loyalty to one's fellow soldiers, virtually all of whom were other white males. Across companies in the Union Army, the greater the internal heterogeneity (in terms of age, hometown, occupation, etc.), the higher the desertion rate (Costa & Kahn 2003a).
Advocates of the conflict and contact theories clearly disagree about the balance of the empirical evidence, but in their shared focus on ethnocentric attitudes, they share one fundamental assumption – namely that in-group trust and out-group trust are negatively correlated. I believe this assumption is unwarranted and may have obscured some of the most interesting and unexpected consequences of diversity for social capital. In order to explain why, I need to remind you of an important distinction now commonly made in the field of social capital – that is, the distinction between ‘bonding’ social capital (ties to people who are like you in some important way) and ‘bridging’ social capital (ties to people who are unlike you in some important way). So, my bonding social capital consists of my ties to other white, male, elderly professors, and my bridging social capital reflects my ties to people of a different generation or a different race or a different gender.
Too often, without really thinking about it, we assume that bridging social capital and bonding social capital are inversely correlated in a kind of zero-sum relationship: if I have lots of bonding ties, I must have few bridging ties, and vice versa. As an empirical matter, I believe that assumption is often false. In other words, high bonding might well be compatible with high bridging, and low bonding with low bridging.7 In the United States, for example, whites who have more non-white friends also have more white friends.8 This article is not the place for an extended discussion of that empirical issue, but the theoretical point helps to clarify the relationship between diversity and social capital.
Contact theory suggests that diversity erodes the in-group/out-group distinction and enhances out-group solidarity or bridging social capital, thus lowering ethnocentrism. Conflict theory suggests that diversity enhances the in-group/out-group distinction and strengthens in-group solidarity or bonding social capital, thus increasing ethnocentrism. However, virtually none of the hundreds of empirical studies of this broad topic has ever actually measured in-group attitudes. Instead, researchers have typically measured out-group attitudes (positive or negative) and have simply assumed that in-group attitudes must vary inversely. Thus, they have presumed (without evidence) that their measures of out-group attitudes were straightforward measures of ethnocentrism.9 However, once we recognize that in-group and out-group attitudes need not be reciprocally related, but can vary independently, then we need to allow, logically at least, for the possibility that diversity might actually reduce both in-group and out-group solidarity – that is, both bonding and bridging social capital. We might label this possibility ‘constrict theory’ (a term suggested by my colleague, Abby Williamson).
I now present some initial evidence from the United States on the issue of how diversity (and by implication, immigration) affects social capital. The evidence comes from a large nationwide survey, the Social Capital Community Benchmark Survey, carried out in 2000, with a total sample size of roughly 30,000. Embedded within the nationwide sample is a representative national sample of 3,000, as well as smaller samples representative of 41 very different communities across the United States, ranging from large metropolitan areas like Los Angeles, Chicago, Houston and Boston to small towns and rural areas like Yakima, Washington, rural South Dakota and the Kanawha Valley in the mountains of West Virginia. While these 41 sites vary with respect to geographic scope from two inner city neighbourhoods to several largely rural states, for the most part they represent metropolitan areas.10 These sites are shown in Figure 2 and Table 2. These community sites differ in many ways (size, economic profile, region, educational levels, etc.), but for our purposes it is important that they differ greatly in their ethnic diversity. For example, Los Angeles and San Francisco (roughly 30–40 percent white) are among the most ethnically diverse human habitations in history, whereas in our rural South Dakota county (95 percent white) celebrating ‘diversity’ means inviting a few Norwegians to the annual Swedish picnic.
Table 2. Social Capital Community Benchmark Survey Sites
Central Oregon (mid-sized town)
Charlotte region/14 county
Detroit Metro/7 county
East Bay (urban neighbourhood)
East Tennessee (rural region)
Grand Rapids (city)
Indiana (selected counties)
Los Angeles County
New Hampshire (statewide)
North Minneapolis (urban neighbourhood)
Rural SE South Dakota county
San Diego County
San Francisco (city)
St Paul Metro
Total sample size
Another important methodological feature of this survey is that it was conducted simultaneously with the national census of 2000, and virtually every interview in our survey was ‘geo-coded’ (i.e. for the vast majority of our respondents, we know exactly where they live, and thus we know the demographic characteristics of the census tract within which they live).11 Thus, we know not only the race, education, income, marital status and so on of our respondents, but also the race, education, income, marital status and so on of their neighbours. The variability of the thousands of census tracts within which our respondents live is even greater than the variability across the 41 sample communities. Some respondents live in neighbourhoods that are almost completely homogeneous, while others live in neighbourhoods that are extremely diverse in every respect. For our detailed and most sophisticated analyses presented below, we use the individual as the unit of analysis, linking his or her attitudes and behavior to the characteristics of his or her neighbourhood. For expository purposes, however, I begin by using the community as the unit of analysis, showing how the diversity of a community is linked to the average level of social capital in that community.
One last methodological preliminary: For present purposes, we adopt the basic fourfold categorization of race and ethnicity that was used in the concurrent census: Hispanic, non-Hispanic white, non-Hispanic black and Asian. This classification scheme, like all such schemes, is ‘socially constructed’– that is, it is not God-given, or biological, or timeless and unchanging, or uniquely defensible. Indeed, the social construction of ethnicity will be an important part of my concluding remarks. However, this typology has two advantages for present purposes. First, it is widely used in public and private discourse in contemporary America, and second, it allows us to treat respondents (from our survey) and their neighbours (from the census) in parallel ways.
I begin with probably the least surprising, but in some respects most misleading, finding from our survey. Figure 3 arrays our 41 communities according to their ethnic diversity and the average level of inter-racial trust expressed by our respondents in those communities. We asked every respondent how much he or she trusted whites, blacks, Asian-Americans and Hispanics (or Latinos), and we know the respondent's own ethnicity, so this measure is simply the average trust expressed toward the other three ethnic categories.12 Obviously, Figure 3 shows a strong positive relationship between inter-racial trust and ethnic homogeneity.13 Inter-racial trust is relatively high in homogeneous South Dakota and relatively low in heterogeneous San Francisco or Los Angeles. The more ethnically diverse the people we live around, the less we trust them. This pattern may be distressing normatively, but it seems to be consistent with conflict theory. Had we stopped our inquiry at this point (as previous studies of conflict and contact theory have done), we would have rejected contact theory, at least in its simplest form,14 and accepted conflict theory. However, as we shall see momentarily, the story is actually more complicated.
Figure 4 is directly comparable to Figure 3, except that here our measure of social capital is trust in ‘people who live in your neighbourhood’. Because of de facto residential segregation, most Americans’ neighbours are of the same race as their own. And yet Figure 4 shows virtually the same pattern. The differences across our 41 sites are very substantial in absolute terms. In highly diverse Los Angeles or San Francisco, for example, roughly 30 percent of the inhabitants say that they trust their neighbours ‘a lot’, whereas in the ethnically homogeneous communities of North and South Dakota, 70–80 percent of the inhabitants say the same. In more diverse communities, people trust their neighbours less.
Figure 5 repeats the story, but with one important difference. Now we ask about trust in people of the respondent's own race: how much do whites trust other whites, blacks other blacks, Hispanics other Hispanics, and Asians other Asians? This figure charts an entirely unexpected correlation for it shows that in-group trust, too, is lower in more diverse settings. Whereas Figures 3 and 4 are inconsistent with contact theory, Figure 5 is inconsistent with conflict theory. In other words, in more diverse settings, Americans distrust not merely people who do not look like them, but even people who do.
Finally, Figure 6 completes the story by arraying community diversity and ‘ethnocentric trust’– that is, trust in one's own race minus trust in other races.15 This figure clearly shows that ethnocentric trust is completely uncorrelated with ethnic diversity. Thus, it suggests that neither conflict theory nor contact theory corresponds to social reality in contemporary America. Diversity seems to trigger not in-group/out-group division, but anomie or social isolation. In colloquial language, people living in ethnically diverse settings appear to ‘hunker down’– that is, to pull in like a turtle. Figures 3 to 6, taken together, suggest support for what I earlier tentatively labelled ‘constrict theory’.
So far I have limited my presentation to evidence regarding social trust, and it is true that the most impressive and substantial patterns we have so far discovered involve trust of various sorts, including even trust of shop clerks. However, a wide array of other measures of social capital and civic engagement are also negatively correlated with ethnic diversity.16 In areas of greater diversity, our respondents demonstrate:
• Lower confidence in local government, local leaders and the local news media.17
• Lower political efficacy – that is, confidence in their own influence.18
• Lower frequency of registering to vote, but more interest and knowledge about politics and more participation in protest marches and social reform groups.19
• Less expectation that others will cooperate to solve dilemmas of collective action (e.g., voluntary conservation to ease a water or energy shortage).20
• Less likelihood of working on a community project.21
• Lower likelihood of giving to charity or volunteering.22
• Less happiness and lower perceived quality of life.24
• More time spent watching television and more agreement that ‘television is my most important form of entertainment’.25
To be sure, some dimensions of social capital and civic engagement seem relatively unaffected by ethnic diversity in American communities. For example, organizational activity of various sorts, including religious activity, is essentially uncorrelated with diversity, once we control for confounding variables, and as I have already noted, several measures of political engagement are positively correlated with diversity.26 Nevertheless, a reasonably coherent, consistent image emerges from this analysis.27
Diversity does not produce ‘bad race relations’ or ethnically-defined group hostility, our findings suggest. Rather, inhabitants of diverse communities tend to withdraw from collective life, to distrust their neighbours, regardless of the colour of their skin, to withdraw even from close friends, to expect the worst from their community and its leaders, to volunteer less, give less to charity and work on community projects less often, to register to vote less, to agitate for social reform more, but have less faith that they can actually make a difference, and to huddle unhappily in front of the television. Note that this pattern encompasses attitudes and behavior, bridging and bonding social capital, public and private connections. Diversity, at least in the short run, seems to bring out the turtle in all of us.
This conclusion is provocative, but the graphic evidence presented thus far (bivariate, aggregate analysis) is open to numerous objections. The first and most important is that so far I have used community as the unit of analysis, but that approach obscures a crucial issue – namely, is it who is living in a community that matters (a compositional effect), or who they are living around (a contextual effect)? This question can be resolved only by moving to the individual level of analysis, in which we seek to predict an individual's social connectedness from both his or her personal characteristics (race, age, geographic mobility, etc.) and his or her neighbours’ characteristics (age, race, mobility, etc.).
Second, the diverse communities in our study are clearly distinctive in many other ways apart from their ethnic composition. Diverse communities tend to be larger, more mobile, less egalitarian, more crime-ridden and so on. Moreover, individuals who live in ethnically diverse places are different in many ways from people who live in homogeneous areas. They tend to be poorer, less educated, less likely to own their home, less likely to speak English and so on. In order to exclude the possibility that the seeming ‘effect’ of diversity is spurious, we must control, statistically speaking, for many other factors. Our ability to control simultaneously and reliably for many factors, both individual and aggregate, is enhanced by our much larger sample of respondents than is typical in social surveys. Yakima, Washington, for example, is highly diverse, but relatively small, so our sample there helps distinguish the effects of size and diversity.
These first two methodological objections can be dealt with most efficiently in the context of multivariate analysis. In our ‘standard model’ we have included simultaneously controls at both the individual and the census tract level for:
In addition, we control for region of the country; the respondent's gender, financial satisfaction and work hours; the population density and the Gini index of income inequality in his or her census tract; and two measures of the crime rate in the respondent's county.28 Obviously, it is impossible here to present the full array of statistical evidence for each of the dozens of dependent variables we have examined, but the multivariate analysis we carried out is illustrated in Table 3.
Table 3. Predicting Trust in Neighbours from Individual and Contextual Variables
Notes: Question was ‘How much can you trust people in your neighbourhood?’ N = 23,260. Adj. R2= 0.26.
R owns home (v. rent)
R's education (years)
R's ethnicity: black
Census tract poverty rate
R's satisfaction with current finances
R's ethnicity: Latino
R's household income ($100,000)
County: Non-violent Crimes per Capita
Census tract Herfindahl Index of Ethnic Homogeneity
Census Tract Population Density (100,000 per sq. mi)
Census Tract Percent Living Same Town as Five Years Earlier
R's decades in this community
Census Tract Percent Renters
Census Tract Percent Bachelor's Degree
R is Spanish-speaker
R is female
Census Tract Gini Coefficient for Household Income
Census Tract Average Commute Time (hours)
R's ethnicity: Asian
Census Tract Percent United States Citizens
County: Violent Crimes per Capita
Census Tract Percent Over 65
R is a citizen
R's average monthly work hours
R is resident of South
R is resident of Midwest
R is resident of West
R's commuting time (hours)
Here we seek to predict trust in neighbours (as measured on the full 4-point scale) from our standard array of individual and aggregate-level variables. Not surprisingly, the strongest predictors (controlling for everything else) are individual-level variables: age (younger people are less trusting), ethnicity (blacks and Hispanics are less trusting) and socioeconomic class (the educated, the well-off, and homeowners are more trusting). All of these individual-level patterns are well-established from past research. Next in importance are several contextual variables: poverty (less trust among inhabitants of poorer neighbourhoods), crime (less trust in high-crime areas) and ethnic diversity (less trust among inhabitants of ethnically heterogeneous neighbourhoods).
It is sadly true in the United States that poverty, crime and diversity are themselves intercorrelated, but Table 3 shows that even comparing two equally poor (or equally rich), equally crime-ridden (or equally safe) neighbourhoods, greater ethnic diversity is associated with less trust in neighbours. We should take the precise numeric estimates here with more than a grain of salt, but in round numbers Table 3 implies that in terms of the effect on neighbourly trust, the difference between living in an area as homogeneous as Bismarck, North Dakota, and one as diverse as Los Angeles is roughly as great as the difference between an area with a poverty rate of 7 percent and one with a poverty rate of 23 percent, or between an area with 36 percent college graduates and one with none. Even holding constant affluence and poverty, diversity per se has a major effect. Every one of the correlates of ethnic homogeneity listed above (civic collaboration, altruism, personal friendship, confidence in local institutions, happiness, television-watching and so on) passes this same stringent multivariate, multilevel test.
Methodologically speaking, the analysis of contextual effects is one of the thorniest thickets in contemporary social science. I do not have time or space here to elaborate on all the serious threats to the validity of these claims that my colleagues and I have considered, or to adduce all the evidence that has led us (at least so far) to reject those threats. Nevertheless, it may be useful simply to list several prominent issues and briefly indicate our verdict.
People mostly choose where to live, and that simple fact opens up a hornets’ nest of methodological problems with correlational analysis since people with a certain characteristic may choose to live in distinctive areas. For example, the fact that people with children live nearer to schools does not mean that proximity to a school caused them to become parents. In our case, however, selection bias is prima facie implausible as an explanation for our results. For selection bias to produce a negative correlation between diversity and sociability, paranoid, television-watching introverts would have to choose disproportionately to live in mixed neighbourhoods. Phrased differently, a self-selection interpretation of our results would require, for example, that when non-whites move into a previously all-white neighbourhood, the first whites to flee (or the most reluctant to move in) would be the most trusting, and the last to flee would be the least trusting; or alternatively, that ethnic minorities and immigrants would selectively choose to move into neighbourhoods in which the majority residents are most irascible and misanthropic. Common sense suggests that the opposite is more likely; if anything, selection bias probably artificially mutes the underlying causal pattern. In short, taking self-selection into account, our findings may underestimate the real effect of diversity on social withdrawal.29
Different Strokes for Different Folks?
We considered the possibility that the effects of diversity on social capital might vary from group to group. Perhaps people in poor neighbourhoods are more sensitive to diversity than people in upscale neighbourhoods (or the reverse). Perhaps women are more likely to hunker in the presence of diversity than men (or the reverse). Perhaps conservatives are more allergic to diversity than liberals (or the reverse). Perhaps the basic relationship is different for different racial and ethnic groups. Perhaps younger people are less upset by diversity than older generations. Our base model directly controls for most of these variables, but the more subtle question here involves interaction effects: Does the relationship between diversity and sociability vary between men and women, upscale and downscale neighbourhoods, liberals and conservatives, whites and non-whites, young people and older generations?
The short answer is basically ‘no’. The same pattern appears within each of these demographic groups. To be sure, the strength of the core patterns varies somewhat from group to group, partly perhaps as a function of sample size and reduced variance. Thus, for example, the impact of diversity on trust and sociability seems to be somewhat greater in lower-status neighbourhoods, but for measures of altruism the negative impact of diversity seems somewhat greater in upper-status areas. Diversity seems to affect men and women equally, though with minor variation across different indicators of sociability. The impact of diversity on sociability seems somewhat greater among conservatives, but it is significant among liberals, too. The impact of diversity is definitely greater among whites, but is visible as well among non-whites.
Broadly speaking, contemporary ethnic diversity in American communities reflects (in roughly equal measure) two quite different historical processes: the African slave trade of the seventeenth and eighteenth centuries and the growing immigration of Latinos and Asians into the United States in the twentieth and twenty-first centuries. Although all four racial-ethnic categories are represented in all parts of the country, African-Americans are disproportionately represented in the Southeast and the urban areas of the North, whereas Latinos and Asian-Americans are concentrated in the Southwest and West. Thus, in gross terms, variance in our basic measure of ethnic diversity can be partitioned into two distinct factors: the percentage of blacks in a given area and the percentage of immigrants in a given area. It is important to ask whether these two different types of diversity, with their very different historical matrices, have different effects on social capital.
Thus, we replicated our multivariate, multilevel analyses, but included both ‘percent black’ and ‘percent immigrant’ in place of our core measure of ethnic diversity. For the primary indicators of social capital discussed earlier (i.e. social trust, community attachment and sociability) each of these two separate measures of diversity has a significant and independent negative effect, though percent immigrant seems to have a somewhat more consistent and powerful effect. In the interests of parsimony, therefore, I have presented our findings in this article simply in terms of ethnic diversity, rather than distinguishing among different types of diversity. In subsequent work it will be desirable to seek to decompose the underlying patterns much further: what is the effect (for example) of Latino neighbours on blacks’ trust of Asians? And what is the effect of (say) Mexican neighbours on Cubans’ trust of whites? On the other hand, these further decompositions will be complicated by increasingly severe problems of sample size and multicollinearity. At this stage in our work, we have discovered no patterns at this level of disaggregation that would call into question our core finding that ethnic diversity itself seems to encourage hunkering.
We initially suspected that the effects of diversity might be greater for older generations raised in a less multicultural era, whereas younger cohorts would be less discombobulated by diversity. Among twenty-something respondents in 2000, diversity appears to lower trust somewhat less than it does among older respondents. However, every successively older cohort from age 30 to age 90 showed essentially equal effects, so Americans raised in the 1970s seem fully as unnerved by diversity as those raised in the 1920s. Moreover, people in their 20s are exceptionally mobile (as they go to college and take jobs), so their current residence is probably a noisier proxy for their actual social context. Consequently, even their slightly lower contextual sensitivity might well be merely a passing life cycle effect, not a harbinger of enduring change. We have unearthed no convincing evidence of generational differences in reactions to diversity.
In claiming that ethnically diverse neighbourhoods produce hunkering, we use the census tract as a proxy for ‘neighbourhood’. However, the real neighbourhoods in which people experience their daily lives likely vary from census tracts. Obviously, no nationwide survey could gather contextual data on personally defined ‘neighbourhoods’ for all respondents, so it is difficult to address this issue empirically. However, insofar as the error introduced by this mismatch between objectively and subjectively defined contexts is more or less random, the net effect is that our results underestimate the real effects of diversity. Moreover, we have replicated all our key findings using county as the contextual variable, and the results are virtually identical, though slightly less sharp, like a photograph that is slightly out of focus.30 That is precisely what we would expect if the error introduced by mismatch were random because the mismatch is undoubtedly greater when context is defined at the more gross level of county than at the finer-grained level of census tract.31 Moreover, the fact that we find the same contextual effect using two such different measures of context suggests that the pattern is impressively robust. The presumption is that if we could magically define the boundaries of each respondent's neighbourhood personally and attach relevant neighbourhood characteristics, the negative effects of diversity might look even more pronounced.
To confirm the robustness of the relationship between social capital and ethnic diversity, we exploited an entirely different dataset: a measurement of social capital for every county in America compiled by Anil Rupasingha, Stephen J. Goetz and David Freshwater (RGF) at Pennsylvania State University (Rupasingha et al. 2006). The RGF measure, based on the density of civic and non-profit organizations, voting turnout and cooperation with the census, includes no measures of individual attitudes and behaviour, but it is strongly correlated with an independent survey-based measure of social trust.32 The advantage of the RGF dataset is that it covers all 3,111 counties in the continental United States. Controlling for education levels, poverty, urbanization, commuting time, total population (logged), residential mobility and region, the RGF social capital measure is strongly negatively correlated with both immigration and ethnic diversity.33 This entirely independent confirmation strengthens our confidence that our core finding is not dependent upon a restrictive definition of ‘context’.
Non-linearity and Inequality?
We suspected that the effects of ethnic diversity might be non-linear, perhaps reflecting ‘tipping point’ effects, so that an increase of non-white immigrants (for example) from 0 to 5 percent might not have the same impact as an increase from 10 to 15 percent or from 47.5 to 52.5 percent. In fact, we found no empirical evidence for such non-linear effects of diversity in our analyses.
In exploring the effects of diversity, we have obviously concentrated on ethnic diversity. However, an equally important and directly analogous set of questions might be – indeed, should be – posed about the effects of economic diversity. What is the relationship between neighbourhood economic inequality and social capital? This query is especially important because (as I have explained elsewhere) the correlation between economic equality and social capital is virtually ubiquitous, both across space and across time, both in the United States and around the world (Putnam 2000, 358–60; Costa & Kahn 2003b).
Our standard statistical model includes measures of economic inequality, particularly the Gini index of income inequality, and its effects are often quite parallel to, and independent of, the effects of ethnic diversity. Generally speaking, people who live in neighbourhoods of greater economic inequality also tend to withdraw from social and civic life. On the other hand, the relationships involving economic diversity seem to be somewhat more complex than those involving ethnic diversity, which are the focus of our concern in this article. First, the correlations between social capital and economic inequality are less consistent than those between social capital and ethnic diversity. Second, the correlations we find between economic inequality and social capital appear to be non-linear, with some pronounced tipping points, unlike the patterns involving ethnic diversity. Third, unlike ethnic diversity, the effects of income inequality seem to be interactive at the tract and county levels.
Most fundamentally, however, economic inequality does not appear to be a significant confounding variable in our analyses of ethnic diversity. First, as I have already noted, our standard model directly controls for both income inequality and poverty. Second, we have been able to discover no significant interactive effects between economic inequality and ethnic diversity – that is, our core finding that diversity produces hunkering is equally true both in communities with great economic disparities and in those that are relatively egalitarian. Economic inequality is very important, but it does not appear to cause, amplify or obscure the apparent effects of ethnic diversity on social capital.
Pot-holes and Playgrounds
We considered the possibility that public amenities might be rarer in more diverse neighbourhoods, perhaps for political reasons, and that this absence of amenities (not diversity itself) might undermine social capital. Ethnically homogeneous neighbourhoods might have a more congenial ratio of playgrounds to pot-holes. We have found no perfect nationwide measures of local amenities within each census tract, but we were able to construct several ZIP-code-level measures of schools, libraries, civic associations, small shops, sports clubs, religious institutions, day care facilities and other sites of social interaction.34 If anything, such community resources turn out to be positively correlated with ethnic diversity, so they cannot account for our core finding, and in fact, when added to our standard model, do not.
Our data set has a complicated, nested structure, including one large national sample and 41 smaller community samples. One of the methodological challenges is that the nested structure introduces the possibility of biased standard errors since observations within each community are not independent.35 Moreover, conventional multiple regression assumes that the effect of the key explanatory variable (diversity, in our case) does not vary from community to community, whereas, in fact, variation in reactions to diversity from one community to another community is an open and interesting question. To address this issue, we pursued four strategies:
• We replicated the analysis on the (N = 3,003) national sample alone. The core results are fully confirmed, although the significance levels are obviously attenuated by the smaller sample size.
• We ran the standard model separately within each community sample. These samples are much smaller (and therefore more vulnerable to random error). Moreover, variation in ethnic diversity is much lower within any community than nationwide, and multicollinearity among contextual variables is a more serious problem within any given community. (Bismarck has no rich, transient, Asian-American neighbourhoods, for example, and Los Angeles has no poor, low-crime, all-white neighbourhoods.) Consequently, standard errors are much higher in this setting. Nevertheless, within 26 of the 41 community samples, diversity was associated with low trust, controlling for all standard covariates, although the link achieved conventional statistical significance in only a few cases.
• We estimated a random-intercept, random-coefficient Hierarchical Linear Model (HLM) for the pooled, 41-site sample. This approach produces an estimate of the diversity effect that is essentially a weighted average of the coefficients within each community. This estimate is highly significant, though slightly lower than the full-sample OLS coefficient.
• Based on the estimated HLM model, we calculated the empirical Bayes or shrinkage estimate of the effect of diversity on trust within each community, and that effect is negative in 39 of the 41 communities. To be sure, the negative effects of diversity seem to be more pronounced in some communities than in others, and those differences across communities should be quite instructive, so we intend to explore them more fully in subsequent work. Nevertheless, the core finding that diversity encourages hunkering seems highly robust.
In short, we have tried to test every conceivable artifactual explanation for our core finding, and yet the pattern persists. Many Americans today are uncomfortable with diversity.
One powerful limitation on this analysis, however, deserves more substantial discussion, for it sheds an entirely new light on our central concern about the effects of immigration and ethnic diversity on modern societies. All our empirical analysis to this point has involved ‘comparative statics’– that is, we have compared people living in places with different ethnic mixes at one point in time– namely different American communities in the year 2000.36 Although our evidence does suggest that it is the level of diversity that matters, not the rate of change, we have not yet considered any ‘dynamic’ evidence about the effects of immigration and diversity over long periods of time within a single place (whether a single community or the nation as a whole). Exploring the dynamics, as opposed to the comparative statics, of diversity and social capital requires entirely different methods, and my research group has only begun to explore those avenues. For example, several of my colleagues have undertaken case studies of the effects of immigration and diversity over time on the social and political life of various local communities in the United States.37 Moreover, we have only begun to explore highly relevant evidence from such distant domains as experimental social psychology and the history of previous waves of immigration. Thus, my comments in the final third of this article are necessarily preliminary. However, these ideas are, I believe, crucial to any final interpretation of our ‘comparative statics’ evidence.
Becoming Comfortable with Diversity
Social psychologists and sociologists have taught us that people find it easier to trust one another and cooperate when the social distance between them is less.38‘When social distance is small, there is a feeling of common identity, closeness, and shared experiences. But when social distance is great, people perceive and treat the other as belonging to a different category’ (Alba & Nee 2003, 32). Social distance depends in turn on social identity: our sense of who we are. Identity itself is socially constructed and can be socially de-constructed and re-constructed. Indeed, this sort of social change happens all the time in any dynamic and evolving society. For example, religious evangelism, social mobilization and political campaigning all involve the intentional transformation of identities.
Changed identity can also lead to changed behaviour. For example, the more university graduates identify with their alma mater, the greater their alumni donations (Mael & Ashforth 1992, as cited in Kramer 2006). Although the linkage between identity and social capital is only beginning to be explored, it is an important frontier for research. The relationship between the two is almost certainly powerful and reciprocal: Whom you hang out with probably affects who you think you are, and who you think you are probably affects whom you hang out with.
Diversity itself can only be conceived in terms of socially constructed identities. We saw that earlier when we were forced to define ‘diversity’ in our research in terms of the currently canonical four ethno-racial categories in the United States Census. However, how people are assigned by others to racial and ethnic categories has varied greatly over time and space. Thus, adapting over time, dynamically, to immigration and diversity requires the reconstruction of social identities, not merely of the immigrants themselves (though assimilation is important), but also of the newly more diverse society as a whole (including the native born).
Please allow me several personal anecdotes to illustrate that identities are socially constructed and malleable. Several of my grandchildren were raised in Costa Rica, the children of an American mother (my daughter) and a Costa Rican father. A few years ago they moved to Pittsburgh and at the end of the first week of school, my granddaughter Miriam came home and asked my daughter: ‘People keep calling me “Hispanic.” What do they mean? I tell them “No, I'm Costa Rican.”’ My daughter, a social historian by profession, but also a mom, knew she had to answer the question seriously, and she replied: ‘“Hispanic” is how North Americans refer to people whose parents came from Latin America.’‘Oh,’ asked Miriam, ‘is Daddy Hispanic?’‘Yes,’ replied my daughter. After a pause, Miriam asked: ‘Are you Hispanic?’ and my daughter replied ‘No.’ After a much longer pause came Miriam's inevitable question: ‘Am I Hispanic?’‘That's a difficult question, isn't it?’ replied my daughter. Miriam was learning about the complicated way in which Americans today divide up the world, and in the process she was reconstructing her own social identity.
A second story: I grew up in a small town in the Midwest in the 1950s. Of the 150 students in my senior class, I knew the religion of virtually every one. Even now, when I have long forgotten their names, I can generally remember who was a Catholic, who was a Methodist and so on. Nor was that some personal quirk of mine, because in fact most of my classmates knew everyone else's religion. My own children, who went to high school in the 1980s, knew the religion of hardly any of their classmates. Why the difference? To solve the mystery, you need to know that over those thirty years religious endogamy (the practice of marrying only within one's faith) has largely faded in America, at least among mainline Protestants and Catholics and Jews. In the 1950s, for the most important aspect of any adolescent's life – mating – it was essential to keep track of one's peers’ religious affiliations. By the 1980s, religion was hardly more important than left- or right-handedness to romance. Very few of us keep track of the handedness of other people because it seldom matters to our social interactions. People know whether they themselves are left- or right-handed, but it is not an important badge of social identity. Similarly, though most Americans know their own religious affiliation, for younger Americans that affiliation is less salient socially.
In that sense, Americans have more or less deconstructed religion as a salient line of social division over the last half century, even though religion itself remains personally important. In fact, our own survey evidence suggests that for most Americans their religious identity is actually more important to them than their ethnic identity, but the salience of religious differences as lines of social identity has sharply diminished. As our religious identities have become more permeable, we have gained much religiously bridging social capital, while not forsaking our own religious loyalties. To be sure, deconstructing divisive racial and ethnic identities will not be so quick and simple, but an extraordinary achievement of human civilization is our ability to redraw social lines in ways that transcend ancestry. It is my hypothesis that a society will more easily reap the benefits of immigration, and overcome the challenges, if immigration policy focuses on the reconstruction of ethnic identities, reducing their social salience without eliminating their personal importance. In particular, it seems important to encourage permeable, syncretic, ‘hyphenated’ identities; identities that enable previously separate ethnic groups to see themselves, in part, as members of a shared group with a shared identity.39
To illustrate that this is not a purely platitudinous prescription, let me mention briefly some historical success stories from my own country. First, the United States Army today has become a relatively colour-blind institution. Systematic surveys have shown that the average American soldier has many closer inter-racial friendships than the average American civilian of the same age and social class (United States Department of Defense 1997; Moskos & Butler 1996). Yet barely thirty years ago the Army was not a race-relations success story. During the Vietnam War, one heard frequently of inter-racial ‘fragging’– that is, deadly attacks with fragmentation hand grenades among soldiers of different races. We need to learn more details about this case, but even this brief sketch suggests that something that the Army has actually done during the last thirty years has had the effect of reconstructing social identities and increasing social solidarity even in the presence of ethnic diversity. Strict enforcement of anti-discrimination and anti-defamation policies is a key part of the story, but I suspect that a new emphasis on shared identities that cross racial lines may also have been important.
A second example is equally striking. Historically, Americans worshipped in such complete racial segregation that it was proverbial among sociologists of religion that ‘11:00 am Sunday is the most segregated hour in the week’. In recent years, however, many churches, especially evangelical megachurches, have become substantially more integrated in racial terms. During ongoing research on the changing role of religion, my colleagues and I have attended numerous services over the last several years in churches across America. In many large evangelical congregations, the participants constituted the largest thoroughly integrated gatherings we have ever witnessed. It remains true that most church-goers in America (53 percent) report that all or almost all of the people in their congregation are of the same race. However, younger people and those who attend evangelical megachurches (and Catholic parishes) report significantly more racial integration.40 It seems likely that this undoing of past segregation is due, at least in part, to the construction of religiously based identities that cut across (while not effacing) conventional racial identities.
A last example is historically more complicated, but ultimately more relevant to our contemporary interests. A century ago America also experienced a large, sustained wave of immigration that massively increased our ethnic diversity in traditional terms, with the arrival of millions of immigrants of different ‘races’– a term that then referred to the Italian and Polish Catholics, Russian Jews and others who were swarming into a previously White Anglo Saxon Protestant (WASP)-dominated society. Though I have not found any comparable survey evidence for that period, my strong suspicion is that that period also witnessed a good deal of hunkering, even within the immigrant communities. Yet fifty years later, the grandchildren of the WASPs and of the immigrants were comfortable in one another's presence.
The best quantitative evidence concerns ethnic endogamy. At the turn of the last century in-marriage was ‘castelike for new ethnics from east and southern Europe’, whereas by 1990 only ‘one-fifth [of white Americans] have spouses with identical [ethnic] backgrounds’.41 Conversely, the cultures of the immigrant groups permeated the broader American cultural framework, with the Americanization of St Patrick's Day, pizza and ‘Jewish’ humour. In some ways ‘they’ became like ‘us’, and in some ways our new ‘us’ incorporated ‘them’. This was no simple, inevitable, friction-less ‘straight-line’ assimilation, but over several generations the initial ethnic differences became muted and less salient so that assimilation became the master trend for these immigrant groups during the twentieth century.42
Recounting exactly how that happened would require another article longer than this one. Such an essay would not tout the American experience in the twentieth century as an undiluted triumph, but America has been, as the historian David Hollinger (2000, 208) argues, ‘a formidable engine of ethno-racial change’.43‘American identity’, observes Charles Hirschman, ‘is rooted not in nationhood but rather in the welcoming of strangers’, as embodied in the Statue of Liberty (Hirschman 2005: 595).
That longer article would also have to address the complicated racial dynamics raised by so-called ‘whiteness studies’, or in the words of one leading scholar: ‘how America's immigrants became white’.44 This accommodation of the immigrants is sometimes said to have coincided with increased prejudice and discrimination against African-Americans, but was that link causally necessary or merely coincidental? Such an article would need to address the question of how the pace of assimilation was affected, if at all, by the long pause in American immigration between 1924 and 1964. It would explore the intriguing and unexpected history of American flag worship and the Pledge of Allegiance, a civic practice that was sought by (among other groups) American socialists as a way to symbolize that embracing American ideals (‘one nation indivisible with liberty and justice for all’) made you a perfectly good American even if you were not a WASP (Ellis 2005). Such an article would explore the effects of ‘Americanization’ in public schools, as well as the transition in American nationalism during the 1930s and 1940s from ‘ethnic nationalism’ to ‘civic nationalism’ (Mirel 2002). It would reckon with the effects of the Second World War on American popular culture, including the ubiquitous movie foxhole that always seemed (and not by accident) to contain a Jew from Brooklyn, an Italian from Chicago and a Swede from North Dakota. It would explore the role played by political parties and religious institutions, especially the Catholic Church. It would grapple with the divergent meanings of assimilation, and the fact that Americans today are far more comfortable than Europeans with hyphens (Alba & Nee 2003; Alba 2005). It would weigh potential differences between the twentieth- and twenty-first-century waves of immigration, such as the possibly more visible distinctiveness of contemporary migrants, the structural economic differences, the increase of transnational ties, and the ideological and policy differences (such as affirmative action) between the two eras.
And most fundamentally and most controversially, that longer historical analysis would need to re-open one of the questions that I earlier set aside: To what extent are the two different forms of diversity in America today (i.e. that involving recent immigrants and that involving African-Americans) really analogous? I have argued that the effects of these two forms of diversity on social capital seem largely similar in contemporary America. The historical origins of the two forms are, however, obviously different, and that might well mean that the most effective public responses to the underlying issues must also be different.
Some tough research questions have been raised by my analysis that I have not yet answered. We need to learn more about the many possible mechanisms – from physiological to political – that link diversity and hunkering. We need to chart the conditions under which this linkage is strong, weak, or even non-existent. Is it equally true in all countries, for example? Even within the US, does diversity in the workplace or in church or in school have the same effects as the neighbourhood diversity I have examined in this article? We need to explore whether and when bonding and bridging social capital might be negatively related, because in such circumstances diversity could well produce a more classic in-group/out-group divide that I have not found in the contemporary US. We need to examine more closely the interaction between economic and ethnic diversity, asking whether diversity may have a more deleterious effect when ethnic divisions coincide more fully with economic ones. And most fundamentally, we need much more systematic research to confirm the hypotheses in this third section of my essay linking institutions, identity, and social capital.
So, this article is but a prolegomenon to a larger project on how to manage the challenge that immigration and diversity pose to social capital and solidarity. Nevertheless, my hunch is that at the end we shall see that the challenge is best met not by making ‘them’ like ‘us’, but rather by creating a new, more capacious sense of ‘we’, a reconstruction of diversity that does not bleach out ethnic specificities, but creates overarching identities that ensure that those specificities do not trigger the allergic, ‘hunker down’ reaction.45 In this I share the view expressed by Trevor Phillips, chair of the British Commission on Equality and Human Rights, who has been quoted as saying: ‘We need to respect people's ethnicity but also give them, at some point in the week, an opportunity to meet and want to be with people with whom they have something in common that is not defined by their ethnicity’ (Easton 2006).
My argument here is that in the short run there is a tradeoff between diversity and community, but that over time wise policies (public and private) can ameliorate that tradeoff. Even while pressing forward with research to confirm and clarify these arguments, we must also begin to ask about their implications for public policy. This is surely not the place for a comprehensive proposal for immigration reform, but a few comments may illustrate the policy directions suggested by my analysis.
Immigration policy is not just about numbers and borders. It is also about fostering a sense of shared citizenship. Whatever decisions we reach on numbers and borders, America is in the midst of renewing our historical identity as a nation of immigrants, and we must remind ourselves how to be a successful immigrant nation.
• Tolerance for difference is but a first step. To strengthen shared identities, we need more opportunities for meaningful interaction across ethnic lines where Americans (new and old) work, learn, recreate, and live. Community centers, athletic fields, and schools were among the most efficacious instruments for incorporating new immigrants a century ago, and we need to reinvest in such places and activities once again, enabling us all to become comfortable with diversity.
• Most immigrants want to acculturate – to learn English, for example. Expanding public support for English-language training, especially in settings that encourage ties among immigrants and natives of diverse ethnic backgrounds, should be a high priority.
• Since the long-run benefits of immigration and diversity are often felt at the national level (scientific creativity, fiscal dividends, and so forth), whereas the short-run costs (fragile communities, educational and health costs, for example) are often concentrated at the local level, there is a strong case for national aid to affected localities.
• Our field studies suggest that locally based programs to reach out to new immigrant communities are a powerful tool for mutual learning. Religious institutions – and in our era, as a century ago, especially the Catholic church – have a major role to play in incorporating new immigrants and then forging shared identities across ethnic boundaries. Ethnically defined social groups (such the Sons of Norway or the Knights of Columbus or Jewish immigrant aid societies) were important initial steps toward immigrant civic engagement a century ago. Bonding social capital can thus be a prelude to bridging social capital, rather than precluding it. To force civic and religious groups who work with immigrants to serve as enforcement tools for immigration laws, as some have suggested, would be exceptionally counterproductive to the goal of creating an integrated nation of immigrants.
But we need to work toward bridging, as well as bonding. Senator Barack Obama, whose life story embodies ties between immigrant and native-born America, has called for
. . . an America where race is understood in the same way that the ethnic diversity of the white population is understood. People take pride in being Irish-American and Italian-American. They have a particular culture that infuses the (whole) culture and makes it richer and more interesting. But it's not something that determines people's life chances and there is no sense of superiority or inferiority. . . . [I]f we can expand that attitude to embrace African-Americans and Latino-Americans and Asian-Americans, then . . . all our kids can feel comfortable with the worlds they are coming out of, knowing they are part of something larger. (Obama 2007)
Scientific examination of immigration, diversity and social cohesion easily could be inflamed as the results of research become part of the contemporary political debate, but that debate needs to be informed by our best efforts to ascertain the facts. It would be unfortunate if a politically correct progressivism were to deny the reality of the challenge to social solidarity posed by diversity. It would be equally unfortunate if an ahistorical and ethnocentric conservatism were to deny that addressing that challenge is both feasible and desirable. Max Weber instructed would-be political leaders nearly a century ago that ‘Politics is a slow boring of hard boards.’ The task of becoming comfortable with diversity will not be easy or quick, but it will be speeded by our collective efforts and in the end well worth the effort. One great achievement of human civilization is our ability to redraw more inclusive lines of social identity. The motto on the Great Seal of the United States (and on our dollar bill) and the title of this essay –e pluribus unum– reflects precisely that objective – namely to create a novel ‘one’ out of a diverse ‘many’.
The roster of colleagues who have made important contributions to this project is too long to list here, but crucial roles have been played by Tom Sander, Chris Achen, Mahzarin Banaji, Amy Bates, Mark Beissinger, Josh Bolian, Xavier de Souza Briggs, Tami Buhr, Karena Cronin, Ben Deufel, John DiIulio, Sarah Dryden-Peterson, Lew Feldstein, Shaylyn Romney Garrett, Marty Gilens, Dan Hopkins, Darby Jack, Louise Kennedy, Andrew Leigh, Chaeyoon Lim, Gabriel Loiacono, Sean McGraw, Byron Miller, Matt Pehl, Lara Putnam, Pedro Ramos-Pinto, Steve Resch, Nate Schwartz, Thomas Soehl, Anant Thaker, Van Tran, Jessica Wellburn, Abby Williamson and Catherine Wreyford. I am also grateful for the generous support of the Carnegie Corporation, the Ford Foundation, the William and Flora Hewlett Foundation, the Lilly Endowment, the Rockefeller Brothers Fund, the Rockefeller Foundation and more than three dozen community foundations that funded the 2000 Social Capital Community Benchmark Survey.
This article summarizes initial results from a longer-term project. My colleagues and I will elaborate on our evidence and argument in subsequent publications. My intention here is to provide enough evidence that others can evaluate our argument and provide instructive commentary and criticism.
See Putnam (2000, 18–24) for a discussion of the concept of ‘social capital’.
For a thorough, nuanced analysis of diversity as a social and political value, see Schuck (2003).
Of course, one can artificially create a zero-sum relationship between bridging and bonding by asking what proportion of, say, friendships are bridging or bonding, or about relative trust of in-groups and out-groups, but the result is a mathematical trick, not an empirical finding.
This generalization is based on our extensive analysis of the 2000 Social Capital Community Benchmark Survey described later in this article.
An important exception to this critique is Brewer (1999), who emphasizes that in-group and out-group attitudes can be independent of one another.
The 41 community sites were not chosen strictly randomly, but reflected our ability to raise local funds to cover local costs in as wide an array of communities as we could manage. Nevertheless, extensive analysis has failed to unearth any significant differences between the nationally representative sample of 3,000 and the aggregate of the 41 local sites (N~27,000), either in frequency distributions or in relations among variables. Thus, for practical purposes, we treat the entire sample of 30,000 as a single nationwide sample, while confirming key generalizations on the nationally representative sample of 3,000. Several sites, as well as the national sample, over-sampled African-Americans and Latinos. We used both English- and Spanish-speaking interviewers, so we have an unusually broad sample of Latino respondents. All analyses reported here are based on data weighted to reflect population cross-distributions on race, age, education and gender. The AAPOR RR3 response rate was 27.4 percent across all communities, fairly typical for random-digit dialing telephone interviews nowadays. On the effects (often surprisingly small) of response rates on response bias, see Groves (2006). Having explored the representativeness of these data, we believe that the only significant defect is that the responses are modestly (5–10 percent) biased in a ‘pro-community’ direction, probably due to a contextual effect that enhanced normal social desirability bias – that is, since the interview spent 30 minutes on questions of trust and community involvement, some respondents focused on the attractions and duties of community more than they otherwise might have done.
A United States census tract has an average of roughly 4,000 inhabitants, so it can be thought of as a large neighbourhood. Census tract boundaries are generally drawn to reflect local opinion about real neighbourhoods, though inevitably there is slippage in this effort.
Respondents were asked: ‘(How about) White people (would you say you can you trust them a lot, some, only a little, or not at all)?’‘How about African-Americans or blacks?’‘How about Hispanics or Latinos?’‘How about Asian people?’ Question order was randomized.
Our measure of ethnic homogeneity is a Herfindahl index calculated across the four basic ethnic categories. This standard measure is best interpreted as the likelihood that any two individuals randomly selected from a given community will be from the same category. We replicated our results using another plausible measure of ethnic diversity, García-Montalvo and Reynal-Querol's (2005) index of polarization, and the results are virtually identical.
I say ‘in its simplest form,’ because the more complicated version of contact theory maintains that contact enhances trust only under highly specific conditions: common goals, inter-group cooperation and so forth. As those conditions are narrowed, however, contact theory itself approaches tautology.
. Figure 6 is based on subtracting ‘other’ from ‘own’ racial trust, each as measured on a 4-point scale.
Each of the following generalizations is based not merely on the sort of bivariate aggregate analysis presented in Figures 3 to 6, but on exhaustive multivariate, individual-level analyses as described below.
The questions were: ‘How much of the time do you think you can trust the local government to do what is right – just about always, most of the time, only some of the time, or hardly ever?’‘Do you agree strongly, agree somewhat, disagree somewhat, or disagree strongly: The people running my community don't really care much what happens to me?’‘Would you say you can you trust the local news media a lot, some, only a little, or not at all?’
The question was: ‘Overall, how much impact do you think people like you can have in making your community a better place to live – no impact at all, a small impact, a moderate impact, or a big impact?’
The questions were: ‘Are you currently registered to vote?’‘How interested are you in politics and national affairs – very interested, somewhat interested, only slightly interested, or not at all interested?’‘Could you tell me the names of the two US Senators from your state?’‘Have you in the last twelve months participated in any demonstrations, protests, boycotts, or marches?’‘Did any of the groups that you are involved with take any local action for social or political reform in the past 12 months?’
The question was: ‘Now I'd like to ask you a few questions about the local community where you live. If public officials asked everyone to conserve water or electricity because of some emergency, how likely is it that people in your community would cooperate – very likely, likely, unlikely, or very unlikely?’
The question was: ‘Have you in the last twelve months . . . worked on a community project?’
The measure here is a factor score index combining questions about contributions to charity and frequency of volunteering.
The questions were: ‘Right now, how many people do you have in your life with whom you can share confidences or discuss a difficult decision – nobody, one, two, or three or more?’‘About how many close friends do you have these days? These are people you feel at ease with, can talk to about private matters, or call on for help. Would you say that you have no close friends, one or two, three to five, six to ten, or more than that?’
The questions were: ‘Overall, how would you rate your community as a place to live – excellent, good, only fair, or poor?’‘All things considered, would you say you are very happy, happy, not very happy, or not happy at all?’ In recent years, psychologists and economists have produced a sophisticated literature on the determinants of happiness. It is generally agreed that social connectivity (family, friends and community involvement) is a powerful determinant, so this simple question turns out to be a good, indirect measure of social capital. For overviews of this burgeoning field, see Diener (2000); Layard (2005); Kahneman et al. (1999); Myers and Diener (1995); Blanchflower and Oswald (2004); Powdthavee (2006); Helliwell (2006); Helliwell and Putnam (2004).
The questions were: ‘How many hours per day do you spend watching television on an average weekday, that is, Monday through Friday?’‘Do you agree/disagree with the statement that “Television is my most important form of entertainment”’
For a recent empirical assessment of the links between ethnic diversity and political engagement (or ‘citizenship behavior’), including a review of previous political science treatments of that issue, see Anderson & Paskeviciute (2006). For the argument that heterogeneity enhances some forms of political engagement, while dampening others, see Campbell (2006).
I set aside for fuller analysis in a subsequent publication one important dimension of social capital – namely, inter-racial friendships. At first glance, ethnic diversity is positively correlated with inter-racial friendships. However, to some extent, that correlation represents a quasi-tautological ‘pool’ effect: It is obviously much easier, even randomly, for whites who live in Harlem to have black friends than it is for whites who live in virtually all-white Duluth, Minnesota. When we control for that structural constraint (e.g. by calculating the odds that any respondent would encounter people of other races simply randomly in their local community), then inter-racial friendships (apart from that structural constraint on opportunities for contact) appear to be actually more common in less diverse settings. The statistical methodology for this analysis is, however, far from straightforward, so I forebear from pursuing the issue here. For a useful introduction to this general issue, see McPherson et al. (2001).
The county is the lowest geographical level at which crime rates in America are consistently reported.
We have also undertaken some empirical tests of selection bias. For example, since higher SES people are less constrained in housing markets, any selection bias should be concentrated among them. However, our core findings are equally found among upper and lower SES respondents. We also used our respondents’ reported likelihood of staying in their community over the next five years to explore whether low-trust individuals are poised to flee homogeneous communities, as the selection-bias story implies. In general, low-trust people say they are less likely to stay put, but this is equally true in homogeneous and heterogeneous neighbourhoods.
One partial exception to this generalization involves religious involvement. Using our standard multivariate, multilevel model, tract-level homogeneity tends to predict slightly higher religious involvement, whereas county-level homogeneity tends to predict lower religious involvement. This anomalous pattern appears to reflect regional differences in religiosity.
Whereas the average census tract has a population of roughly 4,000, the average county has a population of roughly 80,000.
Necessarily our survey measure of trust (based on the aggregated DDB survey agree-disagree item ‘most people are honest’) under-represents less populous counties nationwide, so this confirmatory analysis is limited to the 444 most populous counties. Given the noise in both measures, the correlation between the RGF measure and the DDB measure is quite strong (r = 0.37).
Details available upon request from the author.
I am grateful to Mario Luis Small for pointing us to these data.
On analyzing clustered data and hierarchical linear modeling, see Singer (1998); Raudenbush and Bryk (2002); Gelman and Hill (2007). In addition to the strategies outlined in the text, we have adjusted standard errors for intragroup correlation within each community sample and with the exception of one composite measure of volunteering and giving to charity, census tract ethnic heterogeneity remains a highly significant predictor across a dozen diverse indicators of trust and sociability.
We have explored whether ‘old’ diversity is different from ‘new’ diversity by examining differences between neighbourhoods that were already diverse in the 1980 and 1990 censuses and neighbourhoods that had become diverse only in 2000. Although our examination of such ‘lagged’ effects is incomplete, we have so far discovered no evidence that over a span of these two decades ‘older’ diversity has become any less likely to trigger the ‘hunkering’ reaction than more recent diversity.
Benjamin Deufel has explored the impact of Latino immigration into five mid-sized towns in Minnesota, Iowa and North Carolina – so-called ‘new destination’ immigration. Abby Williamson is exploring the impact of both sudden and gradual immigration on the social and political life of a half dozen towns from Lewiston, Maine, to Yakima, Washington. Their work, though still in progress, makes clear that the effects of immigration and diversity over time vary widely from place to place and that those effects depend in part on policies, public and private, within the receiving communities.
Social psychology also suggests that neighbourhood heterogeneity may lead to lower predictability of social behavior and thus to ‘information overload’. (This literature builds on Milgram (1970). I am grateful to Daniel Gilbert for pointing me to this literature.) Information overload in turn leads to systemic shutdown, the physiological counterpart of ‘hunkering down’. In short, unfamiliar difference in the social environment may lead to withdrawal. We intend to explore this avenue in subsequent research.
My approach here is akin to the ‘post-ethnic’ perspective offered by Hollinger (2000).
I shall provide detailed evidence on racial integration in American churches in a forthcoming book on the changing role of religion in American society. The quantitative evidence I report here is drawn from a national survey on religion and civic life conducted in 2006 among 3,000 Americans.
For a historically sophisticated account of these issues, see Hollinger (2000). My argument that the effects of diversity on social capital may be moderated by policies that transform and reinforce national identity is consistent with Miguel (2004), who finds that ethnic diversity dampens the provision of public goods in Kenya, but not in Tanzania, which pursued more serious nation-building policies.