Education and trust

Abstract

Contemporary research on trust has come to assume that education has a universal positive effect on trust. Using the survey item that has dominated the trust literature –‘whether one believes most people can be trusted or one can never be too careful’ – education is often found to be one of the strongest predictors of trust, more important than age, income, wealth, health or any another individual characteristic. Thus there are indeed reasons to believe that education sometimes increases the propensity to trust other people. However, this article argues that there are limits to the positive effect derived from education. Using the fifth wave of the World Values Survey, it is demonstrated that there is no positive effect from education on trust within the family, among friends or among persons living in one’s neighbourhood. In the latter case, a negative effect is found. It is also demonstrated that the positive effect on ‘generalized trust’ is largely a phenomenon found in low-corruption countries. The article demonstrates that in high-corruption countries, education decreases trust in other people – both generalized trust and trust in more proximate relationships.

Keywords

Corruption education generalized trust neighbourhoods radius of trust

Introduction

The nature, consequences, and origins of trust have become dominant themes within social science. Political scientists have studied how trust in fellow citizens facilitates cooperation and stabilizes democracy (Putnam et al., 1993; Offe, 1999; Uslaner, 2002). Economists have studied how trust generates economic development and growth (Knack and Keefer, 1997; Nooteboom, 2007; Bjørnskov, 2012) and sociologists have studied how trust permeates the in- and out-group relationships within communities, diaspora and nations (Hooghe, 2007; Oorschot et al., 2006; Stolle, 1998; Delhey and Newton, 2003; Coffé and Benny, 2007). Though the origins of social trust are subject to controversy, and findings vary from study to study, there is general agreement on the positive effect derived from education. Education is largely believed to increase the propensity to trust other people (see, for example, Huang et al., 2009, for a review of previous research). This gives hope for the future as both the enrolment in and duration of education increases throughout the world. Exact figures are difficult to establish, but according to the UNESCO Institute for Statistics, gross enrolment rates in tertiary education, for example, have increased from 56% to 76% in developed countries in the period from 2000 to 2012 and from 12% to 26% in developing countries.¹ Contemporary societies are thus in the process of an educational revolution.

In order to understand the nature of trust and to explore whether education is a potent tool for building social cohesion, this article reports on studies of how education influences trust across various social relationships and across countries. The overall thesis is that the positive effects education has on trust are more limited than is often implied in the literature. Our contention is that education has a positive impact on trust only in specific types of social relationships and in specific institutional contexts. Previous studies have mainly focused on what has been labelled generalized trust or social trust, which typically has been measured in surveys asking respondents whether ‘most people can be trusted’ or ‘one can never be too careful’. The strong correlations between the proportion answering ‘most people’ can be trusted and all kinds of positive outcomes measures have turned this measure of generalized trust into the dominant indicator of the level of social capital in a given society. However, recent studies have argued that a more explicit measure needs to distinguish between the radius and the intensity of trust (Fukuyama, 1996; Delhey et al., 2011; Frederiksen, 2012). Intensity and radius are found to be independent constructs even at the level of trust in strangers (Van Hoorn, 2014), making the independent study of these two factors important with regard to assessing the impact of education. The most positive outcome of trust is believed to be found in the case where the citizens have both a wide radius of trust, meaning that socially distant groups are included, and a high level of trust, meaning that much faith is put into these distant groups. The most problematic condition is found in the case where citizens have both a narrow radius and low level of trust (Delhey et al., 2011). Following this logic, we distinguish theoretically between mechanisms whereby education could increase respectively the radius of trust and the intensity of trust, and empirically uses data that measure levels of trust across various close and distant relationships.²

In the first section, the relationship between trust and education is submitted to theoretical scrutiny and a number of theses are developed. The second section presents the data and the applied methods. The third section provides basic descriptive measures, and in the following two sections multilevel regressions are applied, respectively analysing educational effects on trust in ‘people you meet for the first time’ and ‘most people’ (the two standard measures for generalized trust), and how education influences trust in persons of another nationality and religion, trust in the family, persons known personally and a person from one’s neighbourhood. In the sixth section, we discuss the strength of our results based on a number of additional models. Finally, the findings are summarized and discussed in the conclusion.

Theoretical links between education and the radius and intensity trust

A first proposition is that education affects trust through the social conditions associated with different levels of education. The predominant hypothesis is that the socio-economic status associated with different levels of education leads to different types of experience that support either trust or mistrust. Newton (2004) has labelled this type of hypothesis the ‘success and wellbeing theory’ (see also Zmerli and Newton, 2011). Two basic arguments underpin this hypothesis. The first is that trust is based on experience (Misztal, 1996; Hardin, 2002) and, because different socio-economic positions afford people different types of experience, trust varies accordingly (Hardin, 1992; Uslaner, 2002; Stolle et al., 2008). This argument also links trust to differences in early socialisation caused by differences in the educational level and parental socio-economic status. That trust is created in early childhood is one of the dominant theories within the literature on trust (e.g. Becker, 1996; Jones, 1996; Uslaner, 2002). Thus, the higher level of trust among the highly educated reflects that the highly educated respondents tend to have highly educated parents who socialize their children differently than do less educated parents. This could be labelled an inter-generational indirect effect derived from education. The second argument is that trust and potential disappointment are less threatening to those with a secure socio-economic position than to those whose livelihood may be threatened by misplacing trust (Coleman, 1990; Uslaner, 2002, 2008; Cook et al., 2005). Both the arguments underpinning the success and wellbeing theory lead to the thesis that education increases the intensity of trust across all types of social relationship, regardless of social proximity (Thesis 1).

A second proposition is that education has a direct effect not (only) caused by social conditions. One finds various arguments in the literature. It has been argued that education affects the personal capacity or propensity to trust, because higher levels of education lead to higher levels of reflexivity which, in turn, is supportive of trust in unknown fellow citizens (Luhmann, 1979; Giddens, 1990, 1991). A fairly similar approach suggests that cognitive capacity is a driver of trust levels. In the seminal work by Lewis and Weigert (1985), it is argued that trust is underpinned both emotionally and cognitively. However, if trust concerns generalized relationships or issues that have little emotional significance, trust is primarily an issue of cognition. Empirical investigations of the relationship between trust and cognition (Yamagishi, 2001) have suggested that a relationship exists between cognitive capacity and level of trust. People who display more sensitivity to indicators of deceit and the ability to correctly identify trustworthiness also express higher levels of generalized trust. Introducing the concept of ‘cognitive mobilization’, Inglehart (1970) suggested that increasing levels of education lead from a parochial to a cosmopolitan outlook that accepts and trusts increasingly abstract and complex systems of social relationships. Following this lead, Delhey et al. (2011) and Freitag and Traunmüller (2009) argued that higher levels of education and cognitive mobilization increase levels of generalized trust. This leads to the thesis that education has a positive effect on trust only in distant relationships and no effect on trust in more proximate relationships. This entails that education basically expands the radius of trust, making people with more education more trusting with regard to more distant or generalized social relationships (Thesis 2).

A third proposition is that the positive effect of education on generalized trust is largely a matter of a qualitative change in the type of relationships towards which the highly educated are oriented. Usually, the link between the network of social relationships and trust has been theorized along the lines of Tönnies’ Gemeinschaft–Gesellschaft dichotomy: in more rural areas, community persists, supporting higher trust levels, whereas in more urban areas, community is eroded and trust is consequently lower (Portes, 1998; Putnam, 2000). This would not explain a positive effect derived from education, since we would expect people with higher education generally to live in more urbanized areas, which would suggest a low level of trust. However, according to Georg Simmel, the shift from Gemeinschaft to Gesellschaft is associated with a qualitative change in social relationships: social relationships change from in-depth qualitative relationships to superficial quantitative relationships (Simmel, 1910). The pre-objectified relationships of Gemeinschaft are characterized by a high degree of familiarity demarcated by the unfamiliar: the stranger. In the objectified relationships of Gesellschaft, unfamiliarity is the greater part of every social relationship: most people you meet are strangers. Simmel’s key point is that this also changes what we expect of social relationships and what type of social relationships we will trust (Simmel, 1971). This point is repeated in Giddens’ (1990) and Bauman’s (2013) work on globalization, modernity and the disengagement of the affluent and educated from the ties of communities and geographical situatedness. The upper and lower strata of society cease to share a place or community as a hub of social networks. Increasingly, the upper strata are oriented towards and connected to national and international networks, whereas the lower strata remain identified with, and are connected through, place and community. Thus from this perspective we would expect both a direct and indirect influence of education on trust. Directly, the lower and higher educated would presumably have more trust within the social relationships towards which they are respectively more strongly oriented. An extension of this thesis is that familiarity with community and disengaged relationships, respectively, will increase trust within these relationships and thus education indirectly underpins an increase in the level of trust. In terms of modelling, the prediction is that higher education could have a detrimental effect on close relationships (Thesis 3). This would mean that the general notion of the trust radius as a cut-off point on the linear relationship between trust and social distance may be flawed. Rather, the radius of trust may have an inner as well as an outer threshold (Frederiksen, 2012) and be less linear and more topographical than usually assumed. This would correspond with findings on trust radius and level as independent constructs (Van Hoorn, 2014).

A fourth proposition is that a positive effect from education on trust only applies in specific countries or in specific institutional contexts. Parsons and Bales (1956), Luhmann (1979), Barber (1983), Giddens (1990), Mistzal (1996) and Uslaner (2002) agree that trust in other people is affected and stabilized by social institutions and, moreover, by knowledge of and experience with these institutions. Social institutions at all levels are argued to secure the pre-contractual foundations of cooperation, enforce norm-compliance and reduce the level of uncertainty associated with interaction in general. In theory, numerous interactions between level of education and level of trust are conceivable. However, we will limit ourselves to one of the standard theses within institutional theory: namely, the level of corruption matters (see, for example, Uslaner, 2008; Morris and Klesner, 2010). We expect an interaction effect with education and the level of corruption. Based on Thesis 2, the argument is that as the higher educated become more reflexive and cognitively mobilized, they will be more sensitive to and critical towards failing social institutions. Another simple argument is that education gives insights into whether fellow citizens are actually trustworthy or not. A similar interaction effect has been found for trust in political institutions. Across European countries, Hakhverdian and Mayne (2012) found education to have a positive effect in low corruption countries but a negative effect in high corruption countries. In terms of social trust, this would suggest that the radius of trust may expand in low corruption countries, but may actually retract in high corruption countries. To our knowledge, this detrimental effect of corruption has not previously been studied. In terms of intensity, we expect corruption, and the interaction with education, to be detrimental to trust in all kinds of relationship (Thesis 4).

Data, measurement and methods

Theses about the education effect are tested by means of the fifth wave of the World Values Survey (WVS) collected between 2005 and 2009. The unique feature of this data set is that the respondents were asked not only about trust in ‘most people’, but also about trust in ‘your family’, ‘people you know personally’, ‘your neighbourhood’, ‘people of another religion’, ‘people of another nationality’ and ‘people you meet for the first time’. The questions were introduced by the statement ‘I’d like to ask you how much you trust people from various groups’. These items are conveniently ranked from the closest (‘your family’) to the most distant category of others (‘people you meet for the first time’). The respondents have not been asked directly about the radius of trust, but these data allow us at least to study the educational effects across different human relationships (Delhey et al., 2011, worked on the same data). In contrast to Delhey et al. (2011), the present article does not aggregate the items into a fixed index in respect of in-group (the first three measures) and out-group (the last three measures) trust. The reason is that the effect from education is heterogeneous across both the first three and the last three items (see below), which supports Van Hoorn’s (2014) correction of Delhey et al.’s (2011) work on radius of trust. We use trust in ‘people you meet for the first time’ as our primary indicator for ‘generalized trust’ but, despite its problems, we have also conducted our analyses using the standard measure of trust in ‘most people’. The latter analyses are not shown but they are referred to in the text and available in the online appendix.

In order to measure intensity of trust, the respondents could answer on a four-point scale: ‘trust completely’, ‘trust somewhat’, ‘do not trust very much’ and ‘do not trust at all’. In analysing the cross-group differences, we use a simple scale from 0 to 3. A mean of 3 implies that all within the group answered ‘trust completely’, and a mean of 0 implies that all answered ‘do not trust at all’. Those answering ‘don’t know’ were excluded.

Education is difficult to measure across nations, so we simply used the nine-group ranking provided by the WVS. The success and wellbeing of the respondents is measured by the subjective placement of the household into income deciles (because the WVS does not include an actual measurement of household income) and subjective state of health (1–5, from ‘very poor’ to ‘very good’), an indirect measure of the limitations and access to (and ability to succeed in) numerous types of social arena (Frederiksen, 2011). We used the standard World Bank indicator to measure the level of corruption in the given nation state. This so-called Control of Corruption Index is originally scaled from –2.5 to +2.5; that is, the least corrupt countries score highest. In order to simplify interpretation, we have multiplied by –1 so the most corrupt countries score highest. In the section on the sensitivity test, we also used Vanhanen’s index of democratization as an alternative measure for the quality of the national institutions. Finally, all models include Gini-coefficients (the standard UNDP Gini-coefficients), because previous research has shown income distributions to be the strongest single determinant of cross-national differences in trust levels (Rothstein and Uslaner, 2005; Bjørnskov, 2007).³

We used all 51 countries available in the fifth WVS which asked these trust questions (n = 73,322). The countries that did not ask the questions were Japan, New Zealand, Iran, Iraq, Guatemala and Hong Kong. We did not have the income decile of the respondents from Argentina and Jordan and therefore these two countries are not included in the statistical models; and we did not have the Gini-coefficient from Andorra. Thus, the effective number of countries is 48. As the survey data have a nested structure (sampling within countries), we applied multilevel regression models. In order to facilitate the interpretations and the comparisons between the models, we treated the dependent trust variables as continuous variables. All independent variables in the models were first rescaled to have a minimum of zero and a maximum of one, and were thereafter centred at their grand mean. A reported regression coefficient for an independent variable thus shows the effect on the dependent variable when the independent variable changes from its minimum value in the sample to its maximum value. We applied both random intercept and random slope models. Whilst the WVS is one of the few data sets that provide enough countries to allow the use of multilevel regressions, 48 cases nevertheless sets clear limits as to the number of level-two variables and interactions effects that can be modelled. In the models presented, two level-two variables are included (corruption and level of economic inequality) and, in the full model, one interaction variable is included. However, a number of sensitivity tests using other level-two variables have also been conducted. Finally, it should be noted that in terms of establishing causality we are naturally limited by the cross-sectional nature of our data. Panel data are better suited to link changes in education to changes in trust. Thus, what are provided in this article are indications of potential causality. The strength is that we are able to provide such indicators across a larger number of countries.

Descriptive measures

This section provides some basic descriptive statistics. The intensity of trust varies considerably when respondents are asked about different groups. Almost everyone trusts their family completely (mean 2.8), while most people are sceptical towards people they meet for the first time (mean 1.0). Between these two extremes fall respectively ‘people you know personally’ (2.0), ‘people in the neighbourhood’ (1.9), ‘people of another religion’ (1.4) and ‘people of another nationality’ (1.2). This ranking is fairly stable across the 51 countries. It is also fairly consistent across educational groups; both the most and least educated have the most trust in their family and the least trust in persons they meet for the first time (see Figure 1). These analyses are strictly explorative and, without taking notice of the multilevel structure of the data, we do not therefore report any results from tests of statistical significance.

Figure 1.

Intensity of trust in different ranges, mean on index from 0–3 by educational level: n = 73,322.

In a bivariate analysis, it is difficult to see any positive relationship between education and trust. The least educated trust their family as much as the most educated. The most educated do trust people they ‘meet for the first time’ somewhat more than do the least educated, but the difference (measured in this bivariate manner) is not as great as one should expect from the previous literature on education and generalized trust (see Figure 1). The bivariate description, however, does indicate some interesting variations across educational levels. In particular, those with higher education (university) are more inclined to trust people of another nationality than are those with less education. A similar but smaller difference can be found for the item about ‘people of another religion’. Finally, the least educated have more trust in people from their neighbourhood, despite the likelihood of socio-economic conditions being worse in these neighbourhoods compared to those of the higher educated. These initial patterns support the thesis that education has positive effects only on some relationships (Thesis 2) and that education can have a detrimental effect on trust in closer relationships caused by disengagement (Thesis 3), at least in terms of trust in persons from one’s neighbourhood.

In a bivariate analysis, we also found support for the thesis that the educational effect on trust is different across countries (Thesis 4). For the initial descriptive purpose, the 51 countries are divided into three groups: low-corruption countries (scoring –1.5 or below on the corruption index), medium-corruption countries (scoring –1.49 to 0 on the index) and high-corruption countries (scoring above 0 on the index). Using this grouping, Figure 2 indicates the level of trust in ‘people you meet for the first time’ across countries and educational levels. The first visual presentation reveals that the trust levels for the various educational groups vary in high-, medium- and low-corruption countries. In low-corruption countries (the UK, the Netherlands, the USA, Canada, Australia, Norway, Sweden, Finland, Switzerland, Chile, and Germany), the mean trust in ‘people you meet for the first time’ is 1.05 among those with the lowest level of education. In addition, as found by many other studies, the level of generalized trust is higher for those with highest education: in low-corruption countries, the mean trust in ‘people you meet for the first time’ is 1.55. However, the pattern is very different in high-corruption countries (Mexico, Argentina, Brazil, India, Bulgaria, Romania, China, Turkey, Ukraine, Russia, Peru, Ghana, Moldova, Georgia, Thailand, Indonesia, Vietnam, Colombia, Serbia, Egypt, Morocco, Trinidad and Tobago, Ethiopia, Mali, Rwanda and Zambia). In these countries, the mean trust in ‘people you meet for the first time’ is 1.05 among those with the lowest education and 0.94 among those with the highest education. Bivariate analysis thus shows that the education effect is absent. The same is true for medium-corruption countries (France, Italy, Spain, South Africa, South Korea, Poland, Slovenia, Taiwan, Uruguay, Jordan, Cyprus, Andorra, Malaysia and Burkina Faso). Similar descriptive figures for the other trust indicators are available in the online appendix.

Figure 2.

Association between education and trust in ‘people you meet for the first time’, respectively for low-, medium- and high-corruption countries: fifth wave of WVS. n = 51.

The following three sections formally test the relationship between education and various trust items by means of multilevel linear regression. We look first at generalized trust measured by trust in ‘people you meet for the first time’.

Education and generalized trust

The ‘empty model’, with no independent variables and trust in people ‘one meets for the first time’ as the dependent variable, demonstrates that a sizeable part of the variance –12% – is due to country variation (interclass effect)(see Table 1). As expected, the difference in levels (the intercept) across countries is significant and therefore a multilevel model is needed. The first model (Model I) indicates the bivariate effect derived from education (controlled for country-level differences: the interclass effect). There is a statistically significant but modest effect (0.093). A change from the lowest level in education to the highest level (on the nine-point scale) is estimated to increase the trust in ‘people you meet for the first time’ by 0.093 on the 0–3 scale. Thus, in general (across the included 48 countries), education has a very modest effect on generalized trust. This is also reflected in the estimate that education explains close to zero of the variance at the individual level and only two per cent of the variance across countries. Thus, only two per cent of the difference in trust levels across countries can be explained by some countries having more highly educated people than others. In Model II, we isolate the direct effect derived from education (reflexivity and cognitive ability) by controlling for the health and wealth associated with higher education (the indirect effects). In this model, the effect derived from education almost disappears (0.027; see Table 1). However, this is not a correct specification of the effect derived from education. As demonstrated by previous research, in some contexts education does matter. In technical terms, we need to introduce country-level variables, a varying slope for education and cross-level interaction. Model III includes the level of corruption and economic inequality across countries. As demonstrated by previous research, these two variables have a strong negative effect on generalized trust, here measured as trust in a person ‘you meet for the first time’ (the coefficients are respectively –0.429 and –0.324). The differences in corruption and income inequality enable us to explain 45% of the variance in generalized trust levels across countries – a sizeable effect.

Table 1.

Association between education and trust in persons you meet for the first time. Multilevel linear regression: Random Intercept Models (0 to III) and Random Slope Models (IV to V). Regression coefficients for fixed effects and variance estimates for random effects. N _countries = 48; N _individuals = 62,224.

	Model 0 Empty model	Model I Bivariate	Model II +Income and health	Model III + Corrup-tion and inequality context	Model IV + Random slope (education)	Model V + Interaction
Fixed effects^A
Education		0.093***	0.027*	0.026*	0.077*	0.061*
Income (subjective decile placement)			0.148***	0.148***	0.139***	0.139***
Health			0.130***	0.130***	0.128***	0.129***
Level of corruption (World Bank index)				-0.429***	-0.438***	-0.420***
Level of inequality (Gini-coefficient)				-0.324***	-0.296**	-0.298**
Interaction: education × corruption						-0.490***
Intercept	1.012***	1.011***	1.011***	0.986***	0.973***	0.974***

Random effects. Variance estimates
Country level	0.072	0.071	0.069	0.038	0.033	0.033
Individual level	0.540	0.540	0.538	0.538	0.534	0.534
Education					0.049	0.028
Covariance: education × intercept					-0.001	-0.000
Intra class correlation	0.12	0.12	0.11	0.07

p ≥ 0.05; *p < 0.05; **p < 0.01; ***p < 0.001.

^AAll independent variables are rescaled to a minimum of 0 and a maximum of 1. See the section ‘Descriptive measures’ for the original scaling.

In Model IV, we allow the effect derived from education to vary across countries. The variance estimate of the random slope is 0.049; that is, the effect derived from education varies with a standard deviation of 0.22. So, as expected, the weak effect derived from education overall masks a more pronounced positive effect in some countries and a negative effect in other countries. The strongest negative effect is found in Rwanda, estimated at –0.33, while the strongest positive effect is found in the US, with an estimate of +0.57, both highly statistically significant. Thus, in Rwanda, highly educated indicates less trust, while the opposite is the case in the USA. This pattern of a highly varying effect on social trust derived from education is shown graphically in Figure 3; the regression lines present the estimated linear effect within each of the 48 countries.

Figure 3.

Random slopes prediction lines for effect of education (original scale 1–9) on trust in strangers (0–3): Model IV from Table 1.

Finally, Model V in Table 1 demonstrates that the interaction effect between education and the level of corruption helps us explain the varying effect derived from education on trust in various countries. There might be other explanations as to why education has a varying effect in the 48 countries analysed (see below for our study of other potential level-two variables), but the interaction effect suggested by this article is able to explain around 40% of the variation. This leads to the conclusions that: (1) education does not have a universal effect on generalized trust; and (2) corruption is a very important explanatory factor in this variation. In low-corruption contexts, education has a strong and highly statistically significant positive effect on generalized trust, while in high-corruption contexts education has a statistically significant negative effect on generalized trust. In the average country, the effect derived from education is estimated to be close to zero (0.06). The estimated interaction effect shows that when we change from the least corrupt country to the most corrupt country, the effect derived from education shrinks by approximately 0.5 (see Table 1). The estimated marginal effect on trust derived from education for different degrees of corruption is shown in Figure 4.⁴

Figure 4.

Average marginal effects of education (minimum to maximum) on trust in strangers (0–3) with 95% CIs. Model V from Table 1 with cross-level interaction between corruption and education.

The same analyses have been conducted using the classic measure for general trust, the ‘most people’ item, and similar results were found (see Table 1 in the online appendix). Thus, on both measures of generalized trust, the result is that the educational effect depends on the level of corruption in the countries. This is in line with our Thesis 4.

Education and trust in other social relationships

In this section, we move beyond trust in ‘people you meet the first time’ and ‘most people’– that is, the normal measures for generalized trust. Applying the same models as used in the previous section, we have analysed how education influences trust in persons of another nationality, another religion, in family members, in persons known personally, and finally in persons from the neighbourhood. For each dependent variable, the models applied in Table 1 can be found in the online appendix. Table 2 summarizes our findings by showing the fixed effect coefficients for the full model (Model V + interaction).

Table 2.

Association between education and trust in persons of another nationality, religion, person from neighbourhood, person know personally and the family. Multilevel linear regression: Random Slope Models (V including interaction). Regression coefficients for fixed effects and variance estimates for random effects. N _countries = 48.

	Trust in other nationality	Trust in other religion	Neighbour- hood	Person known personally	Trust in family
Fixed effects^A:
Education	0.268***	0.216***	-0.186***	0.040^NS	0.014^NS
Income (subjective decile placement)	0.119***	0.089***	0.095***	0.120***	0.098***
Health	0.102***	0.073***	0.135***	0.188***	0.136***
Level of corruption (World Bank index)	-0.637***	-0.554***	-0.042^NS	-0.448***	0.028^NS
Level of inequality (Gini-coefficient)	-0.180^NS	-0.061^NS	-0.422**	-0.314**	-0.080^NS
Interaction: education × corruption	-0.336**	-0.265*	-0.334***	-0.205*	-0.025^NS
Intercept	1.261***	1.361***	1.875***	2.000***	2.802***
Random effects.Variance estimates:
Country level	0.065	0.054	0.051	0.037	0.009
Individual level	0.580	0.590	0.549	0.493	0.209
Education	0.048	0.045	0.031	0.029	0.007
Covariance: education × intercept	-0.016	-0.020	0.008	0.000	-0.001
N _individuals

^NSp ≥ 0.05; *p < 0.05; **p < 0.01; ***p < 0.001

^AAll independent variables are rescaled to a minimum of 0 and a maximum of 1. See the section ‘Descriptive measures’ for the original scaling.

We start out with trust in a person with a different nationality and trust in a person with a different religion. Both items refer to something more than a generalized stranger, but which is still a somewhat distant relationship. The strongest effect derived from education is found for trust in persons of a different nationality: the bivariate relationship is 0.318. The relationship is reduced after control for wealth and health, but a direct positive effect derived from education remains (0.268; see Table 2). There is also a strong effect derived from the level of corruption in the countries on the trust in persons of another nationality (–0.637). As with generalized trust, education has a different effect according to whether the respondents live in low-corruption or high-corruption environments. The interaction effect between education and corruption is statistically significant. In high-corruption countries, the higher educated indicated less trust in persons of other nationalities, and in low-corruption countries, the higher educated indicated more trust.

The effect derived from education is slightly weaker regarding trust in persons of another religion, but otherwise the results are very similar. There is a direct effect derived from education (0.216), a strong effect derived from the level of corruption (–0.554) and a significant interaction effect between these two variables (–0.265). Thus in high-corruption countries the more educated tend to have less trust in persons of another religion, whereas the opposite is the case in low-corruption countries. These findings again support the thesis (Thesis 4) that education only goes together with high trust in a low-corruption environment.

Turning to the more proximate relationships, the association between education and trust is close to absent. As expected from the descriptive analyses (see Figure 1), education proves to have no direct impact on trust in one’s family. Bivariately, there is a modest effect (0.071), but it is reduced to near zero when controlled for the wealth and health of the respondents. Thus there seems to be no direct effect derived from education on trust in one’s family; nor is trust in one’s family influenced by the level of corruption. Furthermore, there is limited cross-country variation in the effect derived from education and no interaction effect between education and corruption. This leads to the overall conclusion that trust in one’s family has very little to do with education. This gives support to Thesis 2 – the idea that education only has a direct effect on distant social relationships.

The results are less straightforward with regard to people the respondents know personally (implicitly non-family). As expected from Figure 1, there was no direct effect derived from education on this kind of trust, but there was a strong effect according to the level of corruption in the country (–0.448; see Table 2). Thus, in contrast to trust in family, trust in persons ‘known personally’ was not unaffected by the level of corruption in the country.⁵ A significant interaction effect can also be seen, but only on the 0.05 level, between education and the level of corruption; that is, in high-corruption environments, the more educated have less trust in persons known personally, and the opposite is the case in low-corruption environments. Thus, again, education can both have a positive and negative effect on trust, but in general, only weak effects.⁶

Finally, we analysed how education affected trust in persons from the neighbourhood. Here we found indications of a negative effect, maybe caused by disengagement as predicted by Thesis 3. The empty model indicates that there is a significant cross-country variation in trust in persons from the neighbourhood (11% of the variance is at the country level; see online appendix Table 5). In Model I, it is estimated that going from the lowest to the highest educational level reduces trust by 0.143 points on the scale 0–3. It is a modest negative effect, but it is statistically highly significant. Furthermore, the negative effect of education on trust is not reduced by controlling for the indirect effect of wealth and health; on the contrary, being wealthy and healthy increases the respondents’ trust in persons from the neighbourhood, but the negative direct effect of education increases to –0.195. Thus, education does indeed lower trust in people from one’s own neighbourhood. A little surprisingly, there is no generally significant association between the level of corruption in a country and the respondent’s trust in people from the neighbourhood (see Table 2).⁷ The models demonstrate that the effect derived from education is again dependent on the context of corruption. More highly educated people living in corrupt countries tend to have lower trust in persons from their own neighbourhood, whereas the less educated living in such countries increase their trust in persons from their neighbourhood. The most important finding is that the negative effect derived from education does not disappear even when this interaction effect is taken into account. The direct negative average effect of education is still significant and at the same levels as in previous models, which supports the argument for a detrimental effect derived from education on neighbourhood relationships.

Model specification and alternative theses

In this section, we discuss how sensitive our results are to the applied method. In order to present our results in an easily accessible manner, we have used linear regression as above, but all the conclusions presented from these multilevel linear models were tested and confirmed by multilevel logistic regression (see online appendix Tables 7, 8 and 9 for results respectively for trust in ‘strangers’ and ‘most people’). Furthermore, in all models we tested effects derived from individual-level variables for cluster confounding by including the cluster means of these variables (see Rabe-Hesketh and Skrondal, 2008). We found statistically significant effects from cluster mean variables in only two models, and in these we saw only marginal changes in the individual-level variables. As a consequence of these findings, we omitted the cluster mean variables in all the models presented above; the effects from the individual variables thus indicate within-country effects.

Another potential problem is the limited number of interaction effects. In the analyses above we have used interactions with the two level-two variables (corruption and level of economic inequality) which, in previous research, have proved the most important in order to understand cross-national differences in trust level across countries (at least for the case of generalized trust). However, the included countries clearly differ on parameters other than the level of corruption and level of income inequality. Thus, the different effect of education across countries could be related to country characteristics other than the level of corruption. We therefore ran a number of additional analyses where other level-two variables were interacted with education. One alternative thesis was that the context effect was caused by the low-corruption countries being predominantly Western: it could be a matter of a distinct Western educational system or religion. However, both the Western dummy and the interaction effect between this Western dummy and education proved insignificant when introduced into our full Model IV (tested for trust in ‘most people’ and trust in ‘people you meet for the first time’). Equally, we did not find any interaction effect on generalized trust between ethnic fractionalization, another variable often used in recent trust literature, and education. We did, however, find an interesting relationship between various measures of democracy, education and generalized trust. Because low-corruption countries tend to be more democratic than high-corruption countries, part of the pattern described in the previous sections could potentially be caused by the degree of democracy. Using one of the most used measures of democracy, Vanhanen’s index of democratization,⁸ we found no direct effect between democracy and generalized trust (measured as trust in persons ‘you meet for the first time’ and ‘most people’). This is in line with previous research, which found stronger relationships between corruption and generalized trust than between democracy and generalized trust. However, we did find a significant interaction effect between the level of democracy and education. The interaction between education and Vanhanen’s index of democratization is more significant than the interaction between education and corruption in models that estimate trust in ‘people you meet for the first time’. However, the opposite was the case for models that estimate trust in ‘most people’. Most importantly, in models with both (cross-level) interaction effects included, the one between level of democracy and education did not change the one between corruption and education into insignificant. Thus, the apparent ‘democracy effect’, though clearly significant, does not rule out the ‘corruption effect’ discussed above. In any event, Vanhanen’s index of democracy and the World Bank’s corruption measure can be seen as two sides of the same coin; they are both measures of the level of procedural justice within society.

Conclusions

This article has modified the general assumption within trust research that education has a positive effect on trust. The assumption in previous research has been that education has an ability to increase both the level and the radius of trust; the highly educated are believed to trust more people more. The analyses in this article indicate that education does have the potential to increase trust. In most analyses, we found a significant positive effect of education on trust, and we found that a smaller part of this effect is mediated by wealth and health. Thus, there are some merits to the more general hypothesis (Thesis 1) that education increases trust relationships.

A universal direct effect derived from education is questionable, however. The higher educated do not simply have a general disposition to trust other people more, since trust in close relationships is similar regardless of education. Education did have a pronounced direct positive effect on trust in people of another nationality. This provides some evidence to support Thesis 2, concerning education and generalized social relationships. However, it is uncertain whether this effect is caused by a more cosmopolitan outlook. It may also simply be a matter of the higher educated having more personal experiences with other nationalities, as implicitly suggested in Thesis 3. However, it does suggest that the trust of the highly educated is disengaged, to a greater extent, from close social relationships in terms of culture and space. Whether based on experience or educational content, this does entail a more cosmopolitan, post-national outlook. For the two generalized trust items, ‘people you meet for the first time’ and ‘most people’, we found an overall weak positive effect derived from education, but with a large variation in the effect across countries. Thus, if education provides a more cosmopolitan outlook, it only does so in specific types of institutional environment. In other types of institutional environment, education apparently begets caution.

In line with previous studies, we found a strong relationship between level of corruption and generalized trust. More importantly with regard to this present study, we found a clear interaction effect between the level of corruption and education, in support of Thesis 4. In low-corruption countries, education does have a pronounced positive effect on generalized trust. This is the effect found in most previous, primarily Western, studies. However, in high-corruption countries, education has the opposite effect. Here, the more highly educated have less trust in ‘most people’ and people they ‘meet for the first time’. The same interaction effect – on a smaller scale – is found for trust in persons of other nationalities or religions, persons from the neighbourhood and persons one knows personally. Thus in a high-corruption country education not only affects the level of generalized trust negatively, but might also influence trust in more proximate relationships. Trust in the family was the only area where this interaction was not found. These findings support the argument about the detrimental effect of corruption and point to a mechanism that potentially increases (higher education) or moderates (lower education) these effects on social trust. This also demonstrates that the positive effects on generalized trust assigned to education in previous studies are caused by the fact that most researchers have worked with samples from low-corruption countries.

Finally, we found little merit in the argument that education due to urbanization could have detrimental effects on proximate relationships such as ‘family’ and ‘persons known personally’; nor do our results support the idea that the higher educated, due to disengaging, universally place more trust in distant relationships (see above). However, the trust in persons from one’s neighbourhood is an interesting exception that actually supports the thesis of a detrimental effect derived from education (Thesis 3). Disengaging is indeed a plausible explanation for this finding. Combined with the positive effect of education on trust in people from other nationalities, this suggests that at least in a spatial sense, education does lead to disengagement and a more cosmopolitan type of trust.

The overall conclusion is not that the current educational revolution has no potential to bring about better and more coherent societies. The point we make is that in order for this to happen it takes more than a change of mind of citizens facilitated by education: it takes real improvement in living conditions and societal institutions. Moreover, the findings clearly show that the influence of education on trust is in some part contextual. This suggests important future paths of investigation for trust research. An intriguing question is whether the ‘sometimes effect’ of education on trust is caused by education making citizens more aware of the character of the institutions in which they live or, as Hakhverdian and Mayne (2012) suggest, if it actually reflects that education in some settings give citizens a more cosmopolitan outlook. Another intriguing question is whether what in the literature has been labelled ‘radius’ actually reflects variation across real social relationships, or if it is more a matter of abstraction and imagination. Methodologically, the fixed ‘radius categories’ of the WVS items are not well suited to answer this question. However, our results indicate that there is no clear causal pattern that reflects the logic from proximate items to distant items. This could point to the importance of cognitive processes linked to abstractions, stereotypes and imagination. This does not make trust, or trust research, less relevant; it does, however, challenge the idea of simple universal logics and linear relationships with regard to trust.

Footnotes

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Supplemental Material

The online appendices are available at .

References

Barber

(1983) The Logic and Limits of Trust. New Brunswick, NJ: Rutgers University Press.

Bauman

(2013) Liquid Modernity. Cambridge: Polity Press.

Becker

(1996) Trust as noncognitive security about motives. Ethics 107(1): 43–61.

Bjørnskov

(2007). Determinants of generalized trust: A cross-country comparison. Public Choice 130(1): 1–21.

Bjørnskov

(2012) How does social trust affect economic growth? Southern Economic Journal 78(4): 1346–1368.

Coffé

Benny

(2007) Toward an empirical characterization of bridging and bonding social capital. Nonprofit and Voluntary Sector Quarterly 36(1): 121–139.

Coleman

(1990) Foundations of Social Theory. Cambridge, MA: Belknap Press (Harvard University Press).

Cook KS Hardin

Levi

(2005) Cooperation Without Trust? New York: Russell Sage Foundation.

Delhey

Newton

(2003) Who trusts? The origins of social trust in seven societies. European Societies 5(2): 93–137.

10.

Delhey J Newton

Welzel

(2011) How general is trust in ‘most people’? Solving the radius of trust problem. American Sociological Review 76(5): 786–807.

11.

Frederiksen

(2011) Suspending the unknown: The foundations, limits and variability of intersubjective trust. Published PhD thesis, University of Copenhagen, Department of Sociology.

12.

Frederiksen

(2012) Dimensions of trust: An empirical revisit to Simmel’s formal sociology of intersubjective trust. Current Sociology 60(6): 733–750

13.

Freitag

Traunmüller

(2009) Spheres of trust: An empirical analysis of the foundations of particularised and generalised trust. European Journal of Political Research 48(6): 782–803.

14.

Fukuyama

(1996) Trust. London: Penguin.

15.

Giddens

(1990) The Consequences of Modernity. Cambridge: Polity Press.

16.

Giddens

(1991) Modernity and Self Identity. Cambridge: Polity Press.

17.

Hakhverdian

Mayne

(2012) Institutional trust, education, and corruption: A micro–macro interactive approach. The Journal of Politics 74(3): 739–750.

18.

Hardin

(1992) The street-level epistemology of trust. Analyse & Kritik 14(2): 152–176.

19.

Hardin

(2002) Trust and Trustworthiness. New York: Russell Sage Foundation.

20.

Hooghe

(2007) Social capital and diversity generalized trust, social cohesion and regimes of diversity. Canadian Journal of Political Science/Revue Canadienne de Science Politique 40(03): 709–732.

21.

Huang

van den Brink

Groot

(2009) A meta-analysis of the effect of education on social capital. Economics of Education Review 28(4): 454–464.

22.

Inglehart

(1970) Cognitive mobilization and European identity. Comparative Politics 3(1): 45–70.

23.

Jones

(1996) Trust as an affective attitude. Ethics 107(1): 4–25.

24.

Knack

Keefer

(1997) Does social capital have an economic payoff? A cross-country investigation. The Quarterly Journal of Economics 112(4): 1251–1288.

25.

Lewis

Weigert

(1985) Trust as a social reality. Social Forces 63(4): 967–985.

26.

Luhmann

(1979) Trust and Power: Two Works by Niklas Luhmann. New York: John Wiley.

27.

Miszal

(1996) Trust in Modern Societies. Cambridge: Polity Press.

28.

Morris

Klesner

(2010) Corruption and trust: Theoretical considerations and evidence from Mexico. Comparative Political Studies 43(10):1258–1285.

29.

Newton

(2004) Social trust: Individual and cross-national approaches. Portuguese Journal of Social Science 3(1): 15–37.

30.

Nooteboom

(2007) Social capital, institutions and trust. Review of Social Economy 65(1): 29–53.

31.

Offe

(1999) How can we trust our fellow citizens? In: Warren

(ed.) Democracy and Trust. Cambridge: Cambridge University Press, pp. 42–87.

32.

Oorschot

Arts

Gelissen

(2006) Social capital in Europe: Measurement and social and regional distribution of a multifaceted phenomenon. Acta Sociologica 49(2): 149–167.

33.

Parsons

Bales

(1956) Family, Socialization and Interaction Process. London: Routledge.

34.

Portes

(1998) Immigration theory for a new century: Some problems and opportunities. International Migration Review 32(1): 290–290.

35.

Putnam

(2000) Bowling Alone: The Collapse and Revival of American Community. New York: Simon and Schuster.

36.

Putnam

Leonardi

Nanetti

(1993) Making Democracy Work. Princeton, NJ: Princeton University Press.

37.

Rabe-Hesketh

Skrondal

(2008) Multilevel and Longitudinal Modeling Using Stata. College Station: Stata Press.

38.

Rothstein

Uslaner

(2005) All for all: Equality, corruption, and social trust World Politics 58 (1): 41–72.

39.

Simmel

(1910) How is society possible? American Journal of Sociology 16(3): 372–391.

40.

Simmel

(1971) On Individuality and Social Forms. Chicago, IL: University of Chicago Press.

41.

Stolle

(1998) Bowling together, bowling alone: The development of generalized trust in voluntary associations. Political Psychology 19(3): 497–525.

42.

Stolle

Soroka

Johnston

(2008) When does diversity erode trust? Neighborhood diversity, interpersonal trust and the mediating effect of social interactions. Political Studies 56(1): 57–75.

43.

Sztompka

(1999) Trust: A Sociological Theory. Cambridge: Cambridge University Press.

44.

Uslaner

(2002) The Moral Foundations of Trust. Cambridge: Cambridge University Press.

45.

Uslaner

(2008) Corruption, Inequality, and the Rule of Law: The Bulging Pocket Makes the Easy Life. Cambridge: Cambridge University Press.

46.

Van Hoorn

(2014) Trust radius versus trust level: Trust radius as a distinct trust construct. American Sociological Review 79(6): 1256–1259.

47.

Yamagishi

(2001) Trust as a form of social intelligence. In: Cook

(ed.) Trust in Society. New York: The Russel Sage Foundation, pp. 121–145.

48.

Zmerli

Newton

(2011) Winners, losers and three types of trust. In: Hooghe

Zmerli

(eds) Political Trust: Why Context Matters. Colchester: ECPR Press, pp. 67–94.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.47 MB