Given increasing immigration-driven diversity in most advanced democracies, social scientists and policy makers have posed questions about the possible negative effects of diversity on social policy and democratic participation. Scholars have examined the welfare state and redistribution (Gilens, Reference Gilens1999; Hero and Tolbert, Reference Hero and Tolbert1996; Lieberman, Reference Lieberman1998; Quadagno, Reference Quadagno1994; Alesina and Glaeser, Reference Alesina and Glaeser2004; Alesina et al., Reference Alesina, Glaeser and Sacerdote2001), but also, more recently, collective-mindedness, by which we mean phenomena such as social trust, civic engagement and political participation (Hooghe et al., Reference Hooghe, Reeskens, Stolle and Trappers2009; Knack and Keefer, Reference Knack and Keefer1997; Putnam, Reference Putnam2000). In this article, we address the latter and ask whether diversity undermines the willingness of citizens to trust one another, to participate in collective endeavours and to be politically engaged.
Our analysis uses a cross-national, cross-sectional time-series dataset that combines individual-level and country-level information. We theorize and investigate variation in immigration's effects on collective-mindedness, and we propose explanations for cross-national differences, building on prior research on comparative civic engagement and social trust (Curtis et al., Reference Curtis, Baer and Grabb2001; Curtis, Grabb, and Baer Reference Curtis, Grabb and Baer1992; Delhey and Newton, Reference Delhey and Newton2005; Hooghe, et al., Reference Hooghe, Reeskens, Stolle and Trappers2009; Paxton, Reference Paxton2002, Reference Paxton2007; Rothstein and Uslaner, Reference Rothstein and Uslaner2005; Ruiter and de Graaf, Reference Ruiter and de Graaf2006; Schofer and Fourcade-Gourinchas, Reference Schofer and Fourcade-Gourinchas2001). The over-time dimension of our analysis is innovative and offers a distinct advantage: we track changes within countries as well as across countries. Our findings show that there is nothing inevitable about declining collective-mindedness in the face of increasing diversity. Indeed, we find that countries with an institutional or policy context promoting economic equality and recognition and accommodation of immigrant minorities experience less dramatic or no declines in collective-mindedness.
Theoretical Background and Motivation
Recently, political scientists and economists have sounded the alarm about the negative repercussions of ethno-racial diversity for social capital and democratic vitality, suggesting that heterogeneity affects “public collective-mindedness”—attitudes about others in society or engagement in social and political actions in the name of a general collective good. Called “social capital” by some, public collective-mindedness has been linked not only to the welfare state and redistribution, but also to health, crime, economic productivity and children's educational success (Castiglione et al., Reference Castiglione, van Deth, Wolleb, Castiglione, van Deth and Wolleb2008; Knack and Keefer, Reference Knack and Keefer1997; Putnam, Reference Putnam2000; Uslaner, Reference Uslaner2002).
Most prominently, Robert Putnam's recent work (Reference Putnam2007) argues that diversity reduces social capital. Such declines are problematic, according to Putnam, because they undermine people's ability to solve collective problems. While Putnam (Reference Putnam2000, Reference Putnam2007) defines social capital narrowly, as social networks and the associated norms of reciprocity and trustworthiness, his empirical investigation is capacious, incorporating trust in others and various measures of civic and political engagement. Putnam reports that those living in diverse areas have less trust in neighbours and local leaders, lower political efficacy, lower levels of voter registration, lower expectations of others' co-operation on collective problems and less likelihood of working on community projects (Reference Putnam2007: 149–51). Critically, contact with “out-group” members reduces trust in both out-group and in-group members. “Diversity seems to trigger not in-group/out-group division, but anomie or social isolation; … people living in ethnically diverse settings appear to ‘hunker down’” (Putnam Reference Putnam2007: 149). While there are several measures of engagement in Putnam's study that increase with diversity, the primary argument rests on indicators that decrease with diversity.
Earlier empirical work on the United States provides additional support for this argument. Alesina and La Ferrara (Reference Alesina and La Ferrara2002) find that individuals in communities with greater racial and income heterogeneity (but not ethnic fractionalization) report less trust in others. A parallel analysis (Alesina and La Ferrara Reference Alesina and La Ferrara2000) of civic participation finds broadly similar results. Costa and Kahn (Reference Costa and Kahn2003) report negative correlations between racial fractionalization or “birthplace” fractionalization and volunteering or membership, but no correlation between fractionalization and trust. In most cases, Costa and Kahn report that income inequality also has an independent negative effect.Footnote 1 These earlier studies support the contention that “In the short to medium run … immigration and ethnic diversity challenge social solidarity and inhibit social capital” (Putnam Reference Putnam2007: 138).
Evidence beyond the United States is far more mixed. Alesina and La Ferrara (Reference Alesina and La Ferrara2000: 848) note that countries with high levels of trust, high associational activity and strong norms of civic co-operation are also economically equal and “ethnically homogeneous.”Footnote 2 Studying Canada, Soroka and colleagues (Reference Soroka, Johnston, Banting, Kay and Johnston2006) find that ethno-racial diversity somewhat lowers generalized trust in others (though not trust in government), but Kazemipur (Reference Kazemipur2006) finds a positive association between municipal-level diversity and trust across Canadian cities. Costa and Kahn (Reference Costa and Kahn2003:107) note that some countries seem to conform to the American pattern (Finland, Germany, and the United Kingdom), but that others do not, such as Sweden (relatively high heterogeneity, high membership) or Portugal (low heterogeneity, low membership). An analysis of European countries by Hooghe and colleagues (Reference Hooghe, Reeskens, Stolle and Trappers2009) finds that of 26 static and dynamic indicators of diversity, only one weakly correlates with lower levels of trust, a null finding echoed in Gesthuizen and colleagues' study (Reference Gesthuizen, van der Meer and Scheepers2009) that evaluated a range of collective-mindedness indicators. Conversely, in an analysis of 60 industrialized and less developed countries, Delhey and Newton (Reference Delhey and Newton2005) find a consistent, negative effect of ethnic heterogeneity on generalized trust, even after taking out the homogeneous, and trusting, Nordic countries of Europe.Footnote 3 In sum, the comparative research provides no definitive response to the question of whether the diversity-and-social capital linkage is a generalized phenomenon or a case of American exceptionalism. Our research aims to elucidate the competing empirical evidence by theorizing why societies might differ in their reaction to immigration-related diversity and applying our analysis to multiple countries over time.
Mediating Institutions: Theorizing Cross-National Variation
While a universal, negative response to diversity is possible, we believe that national contexts mediate reactions to diversity.Footnote 4 We concur with Hooghe (Reference Hooghe2007: 712) that scholars must specify the contexts in which we find a relationship between heterogeneity and social capital, rather than lament the negative relation. Institutional arrangements matter. As Rothstein and Stolle (Reference Rothstein, Stolle, Castiglione, van Deth and Wolleb2008) theorize and Delhey and Newton demonstrate, “good government is an essential structural basis of trust” (Reference Delhey and Newton2005: 323).Footnote 5 We know that immigrants' civic and political engagement depends on the policy environment and institutional context of the receiving society (Bloemraad, Reference Bloemraad2006; Ireland, Reference Ireland1994; Koopmans et al., Reference Koopmans, Statham, Giugni and Passy2005). In a similar way, societies may filter how the majority understands and responds to immigration. We advance two types of country-level factors that could mediate the response to immigration: those that shape economic security and those that address potential conflicts related to cultural diversity. We expand on each in turn.
Economic security and inequality
Advanced democracies vary in the extent to which labour markets are regulated (for instance, through minimum wage laws and employment protection legislation) and the state cushions the fall in instances of labour market failure (through, for example, welfare state income transfers) (DiPrete, Reference DiPrete2002). Liberal, market-oriented regimes such as the United States tend to have less regulation and less cushion than other advanced democracies, and both contribute to the comparatively high degree of income inequality in the United States (Luxembourg Income Study, 2007). A number of scholars contend that increasing economic insecurity and inequality account for the post-World War II decline in social trust in the United States (Arneil, Reference Arneil2006; Uslaner, Reference Uslaner and Warren1999; Uslaner and Brown, Reference Uslaner and Brown2005; Wuthnow, Reference Wuthnow and Putnam2002). Such inequality could heighten any negative response to immigration.
Prior research suggests two mechanisms through which economic inequality might diminish social capital. One is social-psychological: greater inequality may undermine the sense of shared fate or solidarity needed for social trust and collective action (Delhey and Newton, Reference Delhey and Newton2005; Uslaner and Brown, Reference Uslaner and Brown2005), and it could diminish optimism and a sense of control over one's life, which Uslaner (Reference Uslaner2002) maintains are foundation stones for generalized trust. A second mechanism is grounded in the politics of economic scarcity: people may perceive real stakes in ethnic group differences because of threat to group position or due to general economic conditions, regardless of individual self-interest (Citrin et al., Reference Citrin, Green, Muste and Wong1997; Quillian, Reference Quillian1995). We hypothesize that in countries with less income inequality, the possible negative consequences of increasing immigration are attenuated.Footnote 6 This suggests that the United States (and other more market-oriented regimes) might display a more negative diversity effect than countries where residents enjoy greater economic security and equality.
Cultural threat and multiculturalism policies
In his analysis of US immigration policy, Zolberg (Reference Zolberg, Hirschman, Dewind and Kasinitz1999, Reference Zolberg2006) argues that the politics of immigration breeds strange bedfellows: free-market proponents on the economic right join with socio-cultural progressives to support migration, while those on the economic left join with social conservatives worried about immigration. Migration raises the specter of economic threat but also generates cultural fears of fragmentation (Huntington, Reference Huntington2004; Zolberg and Woon, Reference Zolberg and Woon1999).
Scholars debate whether policies—particularly around “multiculturalism”—attenuate or exacerbate perceptions of cultural threat. Multicultural proponents contend that under democratic government by the majority, minorities face disadvantages of recognition and accommodation, requiring culturally specific minority group rights (Kymlicka, Reference Kymlicka1995, Reference Kymlicka2001; Taylor, Reference Taylor and Gutmann1994). The primary purpose of such accommodations is to prevent majority domination, but scholars hint that these policies may mute negative reactions to immigration-generated diversity, thereby facilitating generalized trust and inter-group engagement in civic and political life. Starting in the 1970s, “old” (Canada and Australia, for example) and “new” (Sweden and the Netherlands, for example) immigration countries began to embrace multiculturalism. These governments provided funds for ethnic and immigrant organizations and non-majority language learning, accommodations for religious and cultural minorities, and public recognition.
Since the late 1990s, however, commentators observe a “backlash” against multiculturalism (Brubaker, Reference Brubaker2001; Entzinger, Reference Entzinger, Joppke and Morawska2003; Joppke, Reference Joppke2001). Critics blame multicultural policies for exacerbating social divisions, fuelling divisiveness, retarding immigrants' integration and, in some cases, undermining a country's liberal democratic values (Barry, Reference Barry2001; Gitlin, Reference Gitlin1995; Hollinger, Reference Hollinger2000; Huntington, Reference Huntington2004). From this viewpoint, multicultural policies exacerbate any negative relationship between diversity and trust, though increasing divisiveness could fuel rather than temper civic and political engagement.
Empirically, there is some evidence that multiculturalism promotes collective-mindedness among immigrants by providing them with instrumental support and symbolic legitimacy, which could increase overall “stocks” of social capital as immigrant populations grow (Bloemraad, Reference Bloemraad2006; Vermeulen and Berger, Reference Vermeulen, Berger, Ramakrishnan and Bloemraad2008). It is less clear how multiculturalism affects social trust for majority populations. Hooghe and colleagues (Reference Hooghe, Reeskens, Stolle, Banting, Courchene and Seidle2007) find that multiculturalism policies have no direct effect, positive or negative, on cross-national differences in trust and ethnocentrism, but they do not test whether such policies mediate the diversity-trust link. We thus identify two competing hypotheses: one suggests that multiculturalism promotes trust and engagement in the context of diversity, while the other suggests that it ignites a backlash, lowering aggregate trust and possibly willingness to engage in collective endeavours. We also consider a third hypothesis: that the interaction between diversity and multiculturalism does not affect trust and engagement in the same way. Multiculturalism policies might increase general distrust by highlighting differences, but this might motivate people to join a group, rather than hunker down.
Conceptualizing social capital
The possibility of differential effects for trust, on the one hand, and collective engagement, on the other, raises an important issue about conceptualizations of social capital. As Castiglione notes (Reference Castiglione, Castiglione, van Deth and Wolleb2008: 558), definitions of social capital range from “thin” structural accounts focused on networks (see Lin Reference Lin2001) to morally “thick” notions of social trust and cohesion (see Uslaner Reference Uslaner2002). Theorists thus debate the centrality of trust for social capital, and empirical studies differ on whether trust and engagement co-vary (for example, Putnam Reference Putnam2000) or whether trust and engagement are independent from each other (Uslaner, Reference Uslaner, Castiglione, van Deth and Wolleb2008; Uslaner and Brown, Reference Uslaner and Brown2005). Our own view is that in some contexts distrust may be a rational, useful sentiment that fuels civic and political engagement, but for this analysis we purposely remain agnostic, running models on three possible indicators of social capital: generalized trust, civic engagement and non-electoral political participation.Footnote 7
Data and Methods
To investigate how institutional contexts mediate the relationship between diversity and collective-mindedness, we conduct a cross-sectional time-series analysis of social trust, organizational memberships and political action. We pay particular attention to changes over time and how these changes vary across countries with different characteristics. Individual-level data come from the World Values Surveys (WVS) (European Values Study Foundation and World Values Survey Association, 2006), conducted in a range of countries since the early 1980s. There are up to four waves of data for each country, from the early 1980s, around 1990, the mid-1990s, and around 2000.Footnote 8 We limit our analysis to advanced democracies for which we have data for multiple waves, so our analyses include 17 to 19 countries, depending on the outcome. The countries include Anglo settler societies (Australia, Canada and the United States); pre-2004-expansion EU-15 countries (Austria, Belgium, Denmark, Finland, France, Germany, Ireland, Italy, the Netherlands, Portugal, Spain, Sweden and the United Kingdom); two non-EU Western European countries (Norway and Switzerland); and Japan.Footnote 9 Because of item availability, in the analysis of political engagement we drop Switzerland and, in the analysis of organizational memberships, Australia and Switzerland. At the country-year level, the sample size ranges from 47 to 60, and at the individual level from 66,573 to 77,756, depending on the outcome. We merge individual-level WVS data with macro-level variables compiled from several sources, including the United Nations (2005), the World Bank (2007), the Luxembourg Income Study (2007) and secondary sources, as outlined below.
Data limitations require us to focus on nation–states as the unit of analysis. This contrasts with much work that focuses on smaller, sub-national units (regions, cities, neighbourhoods, and so forth). While not always articulated, the sub-national focus derives from a theoretical perspective centred on inter-personal interactions. We contend that collective-mindedness also stems from national level institutional arrangements and policies, making this a critical level of analysis.Footnote 10 Ideally, we would want cross-national data at a sub-national level to study local and national context effects, but we know of no available data source that includes the range of countries we do and that measures immigration (or any other form of ethno-racial diversity) at a sub-national level over time. It remains an empirical question whether people are responsive to changing immigrant presence in their neighbourhoods, in the country as a whole, or both, with such reactions plausibly dependent on institutional contexts. Our results are not strictly comparable to studies that focus on sub-national units or use other measures of diversity, but our study provides an important starting point for highlighting the ways in which national contexts can mediate social attitudes and actions.
Dependent variables
We selected dependent variables that tap the main components of collective-mindedness, are available over multiple waves of the WVS, and overlap with prior research on social capital. We use a standard measure of generalized trust, which asks, “Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people?” While we recognize the shortcomings of this formulation (for example, that “most people” remains undifferentiated and that we might trust the same person differently in different contexts or for different tasks), our analytic strategy requires consistent measurement over two decades. This is the best available variable, and it is used in the majority of previous studies we have discussed. Every WVS wave includes the generalized trust question, resulting in 19 countries and 60 country-years. We code “most people can be trusted” as one and “you need to be very careful” as zero. The mean level of trust across all surveys is .403, ranging from .174 in Portugal to .639 in Norway. Table 1 shows the mean value of each dependent variable by country.
Table 1 Levels of Social Trust, Organizational Membership, and Political Action, by Country
Source: World Values Survey, averages across multiple waves
Organizational membership captures the concept of civic engagement and is one way to tap social networks. We use a series of items about membership in specific types of organizations. Unfortunately, these membership items are not available in the third wave of the WVS, so we have only 17 countries and 47 country-years available for this analysis.Footnote 11 We create a dichotomous variable indicating whether a respondent belongs to any of six types of organization (social welfare service organizations for elderly, handicapped, or deprived people; religious or church organizations; education, arts, music, or cultural organizations; Third World development or human rights organizations; conservation, environment, or animal rights organizations; and youth work organizations). The mean membership level across all surveys is .360, ranging from .169 in Portugal to .675 in the Netherlands.
While political participation is not strictly social capital, many scholars use it in their analyses because it is seen as flowing from trust and civic engagement or because it taps trust or engagement in a political system. We include it in our analysis for these reasons, and because commentators have suggested that culturally distinct migrant groups (often Muslims in Europe, sometimes Mexicans in the United States) undermine democratic vitality (Bawer, Reference Bawer2006; Huntington, Reference Huntington2004). Our measure comes from a standard battery of items about specific non-electoral political actions: signing a petition, joining a boycott and attending a lawful demonstration.Footnote 12 The original questions have three response categories: “have done,” “might do” and “would never do.” We collapse “might do” and “would never do,” and our summary variable indicates whether someone has ever done any of these actions. We exclude Switzerland because the item is available only for one wave, resulting in 18 countries and 58 country-years. The mean value for the political action measure, across all surveys, is .589, ranging from .339 in Spain to .768 in Australia.
We analyze each dependent variable separately because we do not think that they necessarily tap the same underlying phenomenon, nor do we believe that diversity necessarily affects each in the same way. Indeed, empirically, the three outcomes are only weakly correlated.Footnote 13 At the individual level, the bivariate correlations are .13, .15 and .18, for trust and political action, trust and organizational membership, and organizational membership and political action, respectively; at the country-year level, they are .02, .32 and .29. These figures, especially the weak relationship between social trust and political action, lend credence to our argument that high levels of participation do not require generalized trust in other people.
Explanatory variables
Our primary focus is macro-comparative, but our models include individual-level control variables for gender, marital status, age, education, income, employment status and religious affiliation. Gender and marital status are dichotomous variables indicating being female and currently married. Age is a simple continuous variable. The education variable measures the age at which an individual left full-time education, bottom-censored at 12 and top-censored at 21.Footnote 14 We include a rough control for a respondent's household income, corresponding approximately to country-specific income deciles. There are more missing responses for income (around 17 per cent) than for other variables, so we include a variable indicating missing income rather than dropping these cases; the income coefficient thus pertains only to those with non-missing income information. Employment status has three categories: employed, unemployed and out of the labour force. Employed is the reference category. Finally, the religion variable taps whether a respondent is religiously affiliated, and if so, with which broad religion. No religious affiliation is the reference, and other categories include Catholic, Protestant, and “other.” Means of all explanatory variables, by country, are reported in an appendix.
Research on diversity and public collective-mindedness is broadly concerned with human diversity of race, ethnicity, culture, religion, language or national origin. Such diversity can be a longstanding feature of a particular country—Belgium's two language groups or the United States' African American population—or “new” diversity from contemporary immigration. Unfortunately, much previous research does not differentiate between longstanding diversity and immigration. We concentrate our analysis on immigrants since immigration policy is perceived, by scholars and policy makers, as one of the few levers available to control diversity. Our key independent variable is the percentage of the population that is foreign born, specific to country and year. This information comes from the United Nations (2005), and is available for each country every five years. We interpolate values for years that lie within these five-year periods.Footnote 15 We use the foreign-born measure since it is the simplest measure of immigration, allowing for the most straightforward interpretation of results. Because immigration statistics were not standardized across countries until very recently, this is also the most reliable measure. We recognize, however, that the native-born population might not view all immigrants alike: a Swede in Norway, for example, is likely seen less as an outsider than a member of the country's significant Pakistani-origin population. For this reason, we conduct a sensitivity analysis using an alternative measure that excludes immigrants from wealthy, industrialized countries. The two measures of the immigrant population, though not identical in the cross section, track each other closely over time, such that the results are largely similar regardless of which measure we use.Footnote 16
Table 2 reports the size of foreign-born populations from 1980 to 2000 by country. It reveals dramatic variation. For example, Japan starts and ends the period with the smallest foreign-born population (1.3 percent in 2000). In contrast, almost 22 percent of Switzerland's population was foreign born in 2000. Rates of change also vary widely, from a slight decrease in Belgium and France, to an increase of more than 5 percentage points in Austria, the Netherlands, and the United States.
Table 2 Percent Foreign Born 1980–2000, by Country
Source: United Nations (2005)
Our other key independent variables derive from our hypotheses that economic inequality and multiculturalism mediate the relationship between immigration-driven diversity and collective-mindedness. We use two indicators of economic conditions, GDP per capita and a Gini coefficient. Figures for GDP per capita, expressed in purchasing power parities (in thousands of constant 2000 U.S. dollars), come from the World Bank (2007). The Gini figures come from the Luxembourg Income Study (2007).Footnote 17 Gini figures are not available for every calendar year, and we use available data from the early 1980s to the present to interpolate figures for other years. Gini coefficients measure inequality in income distribution, with values closer to zero representing greater equality and values closer to one greater inequality. Among these countries, the Gini coefficient in 2000 ranged from .225 (Denmark) to .385 (Portugal). We use Gini coefficients to capture a range of underlying dynamics, from redistribution policies to labour market regulations.Footnote 18
To evaluate the cultural threat and multiculturalism hypothesis, we rely on a typology developed by Banting and colleagues (Reference Banting, Johnston, Kymlicka, Soroka, Banting and Kymlicka2006), which categorizes countries as strong, moderate or weak on multiculturalism policies. Banting and colleagues enumerate eight types of policy: formal affirmation of multiculturalism; multicultural school curricula; insertion of ethnic representation/sensitivity in public media or licensing; exemption codes for ethno-religious minorities (of dress, Sunday store closing, and so forth); dual citizenship; state funding for minority cultural activities; funding of bilingual or mother tongue language instruction; and affirmative action for disadvantaged groups. Canada and Australia rank as the only two “strong” multicultural states; the United States, Belgium, the Netherlands, Sweden and the United Kingdom rank as “moderate”; and France, Austria, Denmark, Germany, Finland, Ireland, Italy, Japan, Norway, Portugal, Spain and Switzerland are “weak.” We group together “strong” and “moderate” countries to indicate those which have taken some steps to accommodate and recognize minorities.Footnote 19Table 3 presents the association between the two key macro-level variables.
Table 3 Associations between Macro-Level Independent Variables
Note: Gini means are calculated at the country-year level.
Modelling strategy
We use multi-level logistic regression models, where individuals are nested within country-year-specific surveys.Footnote 20 The model takes the general form
![$$
\displaylines{
ln[P_{ij} /(1 - P_{ij} )] = \beta _0 + \beta _1 d_i + \beta _2 X_{ij} + \beta _3 X_i + \cr
\;{\kern 1pt} \beta _4 * foreign\;born_i + \cr
\;{\kern 1pt} \beta _5 X_i * foreign\;born_i + \epsilon _{ij} + \zeta _i \cr}
$$](https://static.cambridge.org/binary/version/id/urn:cambridge.org:id:binary:20160201115338218-0582:S0008423910000077jra_eqnU001.gif?pub-status=live)
where i and j index surveys and individuals, respectively and P is the probability of one of the collective-mindedness measures as a function of country dummy variables in d i, individual- and macro-level variables in X ij and X i, the foreign-born variable, and interactions between other macro-level variables in X i and the foreign born variable. The terms ε and ζ are the individual- and macro-level error terms, respectively.
We include country fixed effects because our primary interest is in explaining within-country variation over time in levels of trust, membership and engagement, and not in explaining stable differences between countries. Any unmeasured features of a country that are constant over time are absorbed into the country fixed effects. This means that the main effects of time-constant country-level variables cannot be estimated, although we can and do examine interactions of these variables with the foreign-born variable, since the latter does vary over time.
Results
We begin with a baseline model shown in Table 4. This model includes only individual-level controls and country fixed effects. Results are not particularly surprising and are consistent with previous studies of similar outcomes (Ruiter and de Graaf, Reference Ruiter and de Graaf2006; Schofer and Fourcade-Gourinchas, Reference Schofer and Fourcade-Gourinchas2001). Women and men are indistinguishable in terms of levels of social trust; women have somewhat higher levels of organizational membership than men, but men are more likely than women to engage in political actions. Aside from some statistically insignificant results, being married, older, more educated and in a higher income bracket make a person more trusting of others, more likely to belong to an organization, and more likely to engage in political actions, whereas being unemployed or out of the labour force has a negative effect.Footnote 21 Religious affiliation has very different effects on the three outcomes. Consistent with prior research, religious affiliation boosts the probability of organizational membership, though this is more true for Protestant and “other” than for Catholic affiliation. The opposite is true for political action: the non-affiliated have the highest rates of participation, followed by Protestants, “others” and Catholics. Finally, Protestants have higher levels of trust than the other three groups, which are indistinguishable from one another.
Table 4 Log Odds of Social Trust, Organizational Membership, and Political Action (Baseline Model)
Note: Continuous variables are centred at their means.
* p < .05;
** p < .01;
# p < .10 (two-tailed test).
The models in Table 5 begin to answer our key questions of interest: whether changes in a country's immigrant population size affect levels of trust and engagement, and whether this relationship varies across institutional contexts. Model 5.1 is identical to the baseline model but adds the immigration variable along with a control for GDP.Footnote 22 We see that the immigration variable has an insignificant effect on trust and political action, but a significant, positive effect on membership. So the initial cross-national evidence here suggests no reason to believe that immigration erodes social capital and, in fact, may actually boost some forms of participation. Note, however, that model 5.1 shows only average effects across all countries; the possibility remains that the immigration effect could vary systematically across different kinds of societies. That is, we might see a negative effect of immigration in one subset of countries, but a null or even positive effect elsewhere, which produce no net effect of immigration when we look at all countries together. We turn to this question of varying immigration effects for the remainder of our discussion. To answer it, we examine interactions between the foreign-born variable and institutional and policy context variables. We do this for income inequality and multiculturalism policies separately (models 5.2 and 5.3), and then we turn to more complex models with multiple macro-level interactions (model 5.4).
Table 5 Effects of Foreign Born Population Size on Log Odds of Social Trust, Organizational Membership, and Political Action
Notes: All models include individual controls and country fixed effects, not shown here. Omitted categories are male, unmarried, available income information, employed, no religious affiliation, USA, and low multiculturalism. Continuous variables are centred at their means.
* p < .05;
** p < .01;
# p < .10 (two-tailed test).
We hypothesize that in countries with greater economic insecurity, residents might feel more threatened by immigration, and therefore withdraw more from collective life. As the results of model 5.2 show, we find some support for this contention. This model includes main effects of GDP and percentage of foreign born, as before, and adds effects for Gini and an interaction between Gini and percentage of foreign born. Given the interaction, the main effects of Gini and percentage of foreign born pertain to a situation in which the other variable is set to its mean. For our purposes, the most interesting result here is the negative and significant interaction between Gini and percentage of foreign born across all three outcomes.Footnote 23 This means that the effect of immigration on trust and engagement is more negative (or at least less positive) in contexts of greater income inequality. For social trust, the foreign-born effect becomes negative when a country's Gini coefficient is just below the mean, at about .284. Since the United States has income inequality far above this figure, the negative correlation between diversity and social trust found in prior studies of the United States might stem in part from high inequality. For organizational memberships, the foreign-born effect becomes negative when a country's Gini is about .081 above the mean, at about .367. This is higher than the actual Gini for all countries but Portugal, so here, higher inequality implies less positive rather than more negative effects of immigration. For political action, the foreign-born effect becomes negative when a country's Gini is .072 below the mean, or .214, so countries with the range of inequality we observe will almost always have a negative effect of immigration on political action, with the effect attenuated in more equal countries.
We also hypothesized that multiculturalism policies will mediate the cultural threat of immigration, but this could have different consequences for engagement versus trust. We find that, in fact, such policies do appear to have different effects on different outcomes. Model 5.3 shows that the foreign-born effect on trust is slightly exacerbated in more multicultural states, but these same more multicultural countries experience higher organizational membership with greater immigration. There is no significant effect of multiculturalism on the relationship between immigration and political engagement.Footnote 24 For organizational memberships, a resident of a country without multiculturalism policies will experience no effect of increasing immigration, whereas an otherwise similar person's probability of membership in a more multiculturalist country increases by .053 in response to a one percentage point increase in the size of the foreign-born population.Footnote 25 The different pattern of effects for trust and membership implies that multicultural policies might have participatory benefits but not necessarily generate a strong sense of social cohesion.
Model 5.4 includes both macro-level interactions, and our major conclusions remain the same. The main effect of the foreign-born variable here refers to a non-multiculturalist country with average income inequality. For such a country, as the percentage of foreign born increases, levels of trust, membership and participation do decrease, though not significantly so for trust. But the conclusion that immigration harms collective-mindedness must be qualified. For each outcome, institutional arrangements mediate the foreign-born effect. Any straightforward story about the effect of immigration is impossible.
For all three outcomes, the interaction between income inequality and percentage of foreign born is negative and significant, as we saw in the model with only this single interaction. This suggests that immigration has a more negative or less positive effect in contexts of high income inequality. Multiculturalism policies, on the other hand, have different mediating effects for different outcomes. In the social trust model, there is still a small but statistically insignificant negative interaction between multiculturalism policies and percentage of foreign born, but the effect of immigration on organizational and political participation is actually significantly less negative or more positive in multiculturalist countries. Thus, multiculturalism policies seem to increase civic and political participation in the context of an expanding immigrant population, but have no significant effect on social cohesion (as measured by generalized social trust).
Figure 1 illustrates the magnitude of key effects in Table 5, using coefficients from model 5.4. The predicted probabilities in this figure refer to an individual with baseline characteristics (male, unmarried, employed, and not religiously affiliated, of average age, education and income) in a country with average GDP and with unmeasured characteristics like those of the United States.Footnote 26 For each outcome, we illustrate the effect of a one percentage point increase in the size of the immigrant population (from the mean of 7.8 per cent to 8.8 per cent) in each of four country types: those with and without multiculturalism policies and with “minimum” and “maximum” income inequality (within the range of Gini coefficients in the data, from .20 to .38). The first thing to note is that immigration has no negative effect on any of the outcomes in a hypothetical country with both low income inequality and multiculturalism policies. In fact, for all three outcomes, immigration's effect is positive, and this is particularly so for the probability of organizational membership. In contrast, states with high income inequality and weak multiculturalism policies experience a negative effect of immigration on all three outcomes. The other two country types are less easily summarized since the relative magnitude of the inequality and multiculturalism effects varies across outcomes, but the clear conclusion is that there is no universal effect of immigration on trust, membership and participation.
Figure 1 Mediating Effects of Income Inequality and Multiculturalism on the Relationship between Immigration and Collective-Mindedness
Notes: Coefficients are taken from Model 5.4. Predicted probabilities are for an individual with baseline characteristics in a country with average GDP and unmeasured characteristics like those of the United States.
Overall, our analysis suggests that any relationship between immigration and collective-mindedness is mediated by institutional structures and state policies. If immigration is threatening—economically or culturally—it can lead to “hunkering down,” as documented in the American case, but this is not universally so. Social arrangements fundamentally shape these tendencies.
Implications and Caveats: Is the United States Exceptional?
Our findings extend and qualify previous studies. Focusing specifically on immigration, our analysis supports the view that immigration decreases trust, civic engagement and political participation in some advanced democracies. However, our analysis also raises important qualifications. In societies of relative income equality, residents are far less likely to withdraw from collective life in the face of immigration. Multiculturalism policies might reduce general social trust in the face of immigration, but they also appear to increase engagement. In countries that have both low income inequality and relatively strong multiculturalism policies, all negative effects of immigration on collective-mindedness disappear; we actually see higher levels of organizational and political participation with immigration. While increases in participation do not necessarily mean that everyone is “getting along” —native-born residents might organize to oppose immigration—trust and engagement both increase with immigration in more economically equal societies.
More generally, our findings speak to a central theoretical claim: the need to take institutions seriously. Positing a general, negative relationship between diversity and collective-mindedness requires a universalist account of human behaviours and attitudes. Such an account, predicated on innate psychological traits, differential interests or fear of social change must be tempered by sociologists' accumulated knowledge about how institutions and social context channel humans' beliefs and actions. The United States might not be unique in experiencing decreased collective-mindedness in the face of diversity, but the American pattern is not the only type of response.
Our research comes with a number of caveats, challenges which we hope will spur further thinking and research on the topic. First, we treat multiculturalism policies as time-invariant. The reality of current political debates suggests otherwise. Since the late 1990s, we have witnessed a backlash against multiculturalism in some countries. Future work can build on the index created by Banting and colleagues (Reference Banting, Johnston, Kymlicka, Soroka, Banting and Kymlicka2006) to track changes in policies over time. We could then use policy changes as “natural experiments” for further analysis.
A second concern is the unit of analysis. Diversity is not necessarily experienced at the level of the nation–state, and this is a problem for an analysis that is unable to focus on sub-national units. Country-level analyses such as ours will hopefully lay the groundwork for targeted, cross-national case studies of how these processes operate in local communities.
We focus on immigration, specifically, because we are skeptical of studies that conflate various kinds of diversity. Our primary analysis relies on an undifferentiated measure of the immigrant population, but immigrants who are more racially, linguistically or religiously different from the mainstream population may well contribute more to a feeling of “diversity.” Sensitivity analysis suggests that our findings are robust to an alternative definition that excludes immigrants who might be considered most similar to mainstream populations. Of course, it is difficult with available data sources to make differentiated measures truly comparable, but we find it reassuring that our main conclusions appear to hold regardless of which of these two definitions of the migrant population we use.
Our data also do not permit us to identify individual immigrants. If higher aggregate levels of trust or engagement have positive consequences on a society regardless of who trusts or is engaged, this is not particularly problematic. But if institutions have different effects on immigrant engagement than native-born engagement, we would want to know who is driving the patterns. Unfortunately, current large-scale comparative datasets do not oversample the foreign-born population, and the WVS does not uniformly identify foreign-born individuals. Given the relatively large changes in trust, membership and engagement that we observe as the immigrant population grows, we do not think that the immigrant population itself could be entirely responsible for the patterns we find. Changes in native-born attitudes and behaviour must be part of the story. Furthermore, if we take the perspective that “social capital” is a public good, our findings are important regardless of the precise mechanism.
In short, much work remains to be done, but the general framework we have developed, considering the ways in which institutional arrangements and policies shape diversity's effects, is one that can and should be incorporated into future research. We conclude by reiterating our main findings: we find no universal link between immigration-generated diversity and collective-mindedness. The direction and strength of the relationship depend on institutional arrangements and policies. This insight is critical if we wish to assess future prospects for social capital in advanced democracies, as these societies grow ever more diverse.
Acknowledgments
We benefited from the feedback of many people, but we especially would like to thank Keith Banting, Christopher Bail, Rafaela Dancygier, Marion Fourcade, Cybelle Fox, Laura Morales, Robert Putnam, Tim Reeskens, Dylan Riley, Dietlind Stolle, Van Tran, Rima Wilkes, Matthew Wright and two anonymous reviewers of this journal. Earlier versions of this article were presented in the departments of sociology at the University of Oxford, Columbia University, and Stanford University, at the 2008 Canadian Sociological Association meetings in Vancouver, the 2008 Summer Workshop on Immigration and Social Change in Higher Disley, UK, the 2008 American Sociological Association meetings in Boston, to the Social Interactions, Identity and Well-being Program of the Canadian Institute for Advanced Research, and at the 2009 Diversity and Democratic Politics Workshop at Queen's University. Financial support for this project came from a Berkeley Futures grant from the University of California, Berkeley.
Table A1 Descriptive Statistics, by Country
Notes: For descriptive statistics of dependent variables, see Table 1. Figures are averaged across all waves.
