College Education and Social Trust: An Evidence-Based Study on the Causal Mechanisms.

This paper examines the influence of college education on social trust at the individual level. Based on the literature of trust and social trust, we hypothesize that life experience/development since adulthood and perceptions of cultural/social structures are two primary channels in the causal linkage between college education and social trust. In the first part of the empirical study econometric techniques are employed to tackle the omitted-variable problem and substantial evidence is found to confirm the positive effect of college education. In the second part contemporary information is used to examine the hypothetical mechanisms in the causal inference. That life experience is a primary channel via which college education promotes social trust fails to find support in our examination, while individual perceptions of cultural and social structures explain up to 77% of the causal effect.

Backgrounds

Trust refers to a variety of phenomena that enable individuals to accept risks in dealing with others, solve collective action problems, or act in ways that seem contrary to standard definitions of self-interest (Levi 1998, p. 78). According to Gambetta (1988, p. 217), “trust is a particular level of the subjective probability with which an agent assesses that another agent or group of agents will perform a particular action, both before he can monitor such an action (or independently of his capacity ever to be able to monitor it) and in a context in which it affects his own action”. Misztal (1996, p. 9) suggests “to trust is to believe that the results of somebody’s intended action will be appropriate from our point of view”. In line with the notion of trust discussed by Gambetta (1988, p. 78), Misztal (1996, p. 217), Hardin (1998, p. 12), Warren (1999, p. 311), and Delhey and Newton (2005), we offer a working definition of trust as the perceived likelihood by which an individual expects that another individual or group, at worst, will not knowingly or willingly do you harm, and at best, will act in your interests.

The focus of this paper is on trust in generalized others, which is known as social trust or generalized trust in the literature. Social trust denotes impersonal trust between random people and it differs fundamentally from personal trust by being extended to people on whom the trusting part has no direct information (Hardin 2003, p. 13; Delhey and Newton 2005; Paxton 2007).

Measurement of social trust is generally based on a standard survey question: “Generally speaking, would you say that most people can be trusted or that you can’t be too careful in dealing with people?” This operationalisation of social trust has been widely used for more than four decades in empirical studies and surveys around the world. Empirical studies on the benefits of social trust employ this measurement as the main indicator. These studies provide a plethora of evidence for the positive effects of social trust at the individual and societal level (see, e.g. Putnam 1993, pp. 167–71; Kollock 1994; Fukuyama 1995, pp. 23–32; La Porta et al. 1997; Knack and Keefer 1997; and Zack and Knack 2001).

Despite a large volume of literature on the social trust theory and the social trust benefits, limited efforts have been given to the empirical exploration of the formation of individual social trust from a development perspective.1 In particular, it has been commonly agreed that education is a crucial determinant of social trust. The exact magnitude of the education effect is, however, an under-studied topic. Education is generally considered as an exogenous independent variable in the equation of social trust. Few empirical studies have attempted to isolate the causal effect of education from the influences of confounding variables. Insofar little empirical evidence has been found to clarify a causal linkage between education and social trust. The theoretical rationales on the role of education remain hypothetical and untested.

In this paper we explore the role of college education in the formation of individual social trust, using the rich data of a British cohort from the National Child Development Study (NCDS). We begin with a theoretical review on the micro aspects of social trust, considering that risks are an essential element and emphasizing life experience/development and perceptions of cultural/social structures as two primary sources of expectations and beliefs for each individual in society. Based on the theoretical insights, we hypothesize two channels in the causal pathway between college education and social trust: heterogeneities in life experience/development attributable to education difference and heterogeneities in perceptions of cultural/social structures attributable to education difference.

Our empirical study proceeds in two steps. In the first step we employ econometric techniques to isolate the influence of confounding variables and quantify the causal effect (in terms of average treatment effect) of college education. Strong econometric evidence is found to support the positive role of college education in the formation of social trust. The level of social trust of college graduates exceeds that of non-college graduates by 7.5% point. This implies that a college education increases the probability of trusting generalized others by about 16% of its standard deviation.

In the second step we introduce contemporary information of individual life experience/development and contemporary information of individual perceptions of cultural/social structures to examine the theoretical insights on the role of college education. The hypothesis that college education promotes social trust via its power to shape later-life development fails to find support in the empirical analysis, while individual perceptions of cultural/social structures explain up to 77% of the effect of college education on social trust. We confirm that individual understanding of cultural/social structures is the primary channel in the causal pathway from college education to social trust.

The remainder of this paper is divided into four sections. Section 2 gives a theoretical review of the micro aspects of social trust and hypothesizes two channels in the causal linkage between college education and social trust. This section also presents a succinct illustration of the econometric techniques that are employed in our empirical examinations. The third section presents an introduction of the NCDS dataset and an illustration of an instrumental variable approach. The fourth section provides an evaluation of the causal effect of college education and offers tests on the validity of the hypotheses on the causal effect of college education. The fifth section offers concluding remarks.

Theory, Hypotheses and Evaluation Techniques

Trust is not one thing and does not have one source. It has a variety of forms and causes (Levi 1998, p. 79). A person’s trust in generalized others, likewise, has different aspects and they do not form a single syndrome (Newton 2001). As a result, some researchers argue that the standard survey question of social trust is not clear to respondents about whom to trust or under which circumstances. The ambiguity in the measurement of social trust could make it difficult for researchers to explore respondents’ perceptions of the context relating to the survey question. In order to perform a comprehensive study of the role of college education in the development of social trust, we take efforts to present a theoretical review of the micro aspects of social trust. Based on the theoretical insights, we establish two hypothetical mechanisms to explain the causal influence of college education, which we will examine in the empirical analysis.

Social Trust Theory from a Micro Perspective

Social trust reflects “a belief in the benevolence of human nature in general” (Yamagishi and Yamagishi 1994) and it implies risks to the truster of being dependent on the characteristics (competence and motives) of the trusted in social interactions (Lewis and Weigert 1985; Bluhm 1987; Coleman 1990, p. 91; Levi 1998, p. 79; Hardin 2003, p. 19). Coleman (1990, p. 91) identifies risk as an essential element in the definition of trust. Levi (1998, p. 79) proposes that the actual extent of risk and the extent to which the truster is taking a “sensible” risk are variables, and they are partially functions of the trustworthiness of not only the trustee but also those on whom the truster relies for information and sanctions against a trust-breaker.

As expectations and beliefs are characterized by risks, social trust at the individual level involves an assessment of the trustworthiness of generalized others in which a truster makes an estimate of the competence (or reliability) and an estimate of the feeling of moral obligation (or goodwill) of the “average person” to carry out the fiduciary duties and responsibilities (Barber 1983, pp. 9–10). According to a large amount of trust and social trust literature (e.g. Hardin 1996; Sztompka 1999, pp. 65–68; Rothstein and Stolle 2002; Knight 2003, p. 358; Paxton 2007), there are two primary sources of expectations and beliefs for each individual in a society: an individual’s previous experience (or individual characteristics) and cultural and social structures (or contextual characteristics).

Firstly, personal life experience is an important aspect in an individual’s assessment of the trustworthiness of generalized others (Brehm and Rahn 1997; offe 1999; Hardin 1996; Alesina and La Ferrara 2000, 2002; Uslaner 2002; Paxton 2007). Social trust at the individual level is subject to first-hand experience of the social world and the people in it—friends, family, neighbors, colleagues, and daily contacts with others, which enable a truster to make a judgment about the trustworthiness of generalized others.

People who have matured in a disadvantaged position are more inclined to have a lower sense of the reliability or goodwill of others (Paxton 2007), because they were more likely to have the experience of being exposed to the “darker side” of society. Being victim of a crime, for example, contributes to more distrusting views (Ferraro 1995). As a prudent measure against the risk of negative social encounters or being exploited, these people are less likely to trust others with whom they do not have direct contact. People who have grown up in a well-to-do environment, on the contrary, have a higher belief in the benevolence of human nature in general. They are generally happier with how the life is going and more likely to give affirmative responses in trust surveys (Brehm and Rahn 1997; Uslaner 2002, p. 33).

Socializing activities are an important source of human information practices (Granovetter 1973). Establishing and maintaining good social relationships facilitate access to relevant information in detecting and assessing risks in an uncertain environment. Lack of socialization may prevent a person from developing communication skills and lead to misunderstanding and distrusting of others. Significant experiences in social interactions, especially traumatic experience, throughout the life course may influence the assessments of the trustworthiness of others (Boyle and Bonacich 1970; Hardin 1996). Experience of divorce, for instance, could reduce an individual’s assessment of the goodwill of others, thereby generally lowering his or her view of the trustworthiness of people in general (Alesina and La Ferrara 2000, 2002; Rahn et al. 2003; Paxton 2007).

Secondly, characteristics of cultural and social structures are also an important aspect in the assessment of the trustworthiness of generalized others (Zack and Knack 2001; Rothstein and Stolle 2002; Knight 2003, pp. 358–71; Paxton 2007). These cultural and social structures include norms and conventions, as well as formal/institutional arrangements in society,

Social norms and conventions are the informal rules that configure social life. They create an incentive structure that can place pressure on an individual to honor trust (Yamagishi and Yamagishi 1994; Buskens 2002; Knight 2003). These informal rules instantiate commonly-held behavior of others and dictate differential behavior across social groups. The more a society or community shares a common set of moral values, the greater the likelihood that a high level of collective trust will arise (Fukuyama 1995, p. 153). Knight (2003, pp. 358–71) and Hardin (2003, pp. 16–17) suggest that the trustworthiness created by norm compliance can lead to generalized trust in the community. Rapid social change that undermines the normative order is likely to produce an increase in distrust and untrustworthy behavior (Sztompka 1999, pp. 151–90; Delhey and Newton 2005).

Social heterogeneities are considered to have an adverse influence on the emergence and maintenance of social trust (Lukes 1991; Putnam 2000, p. 400; Alesina and La Ferrara 2000, 2002; Marshall and Stolle 2004; Welch et al. 2005). Heterogeneities in the identity of social norms and conventions may undermine the general beliefs about the willingness of others to cooperate (Knight 2003). The greater the asymmetry of interests reflected in social norms and conventions, the higher the distrust between members of different social, racial or ethnic groups.

Individuals also resort to formal structural arrangements, such as legal system and public institutions to complement and increase the effectiveness of informal constraints in a socially diverse environment (North 1990, pp. 46–47). Formal structural arrangements provide the truster assurance against potential risks involved in trusting others with whom there is no personal knowledge, by the (partial) presence of agencies in monitoring the conduct of the trustee, providing conduct information, and sanctioning law-breaker (Levi 1998, p. 84).

The characteristics of institutional arrangements, such as efficiency, credibility, impartiality and fairness, are influential in generating and maintaining social trust (Rothstein and Stolle 2002). In a society where legal system and public institutions enforce trustworthiness in a fair and effective manner, people perceive fewer risks of being exploited or taken advantage of in social interactions and they will therefore believe that “most people can be trusted”. As Lewis and Weigert (1985) declare, trust exists in a social system when individuals are secure in their expectations.

Some research has found the distribution of resources, in particular, economic resources, to be an important determinant of social trust (Alesina and La Ferrara 2000, 2002; Zack and Knack 2001; Delhey and Newton 2005). Obtainable resources serve as a kind of insurance for the risks involved in trusting others. In a society where substantial inequality exists in the distribution of resources, it is less likely for the disadvantaged group to draw positive inference about the attitudes of resource-favored groups and to cooperate with members of the dominant group (Knight 2003, pp. 358–60). People from disadvantaged group may have higher suspicion of the competence and willingness of institutional arrangements to act in the interest of the disadvantaged or treat them impartially and fairly. Consequently, the functions of institutional arrangements are weakened as reliable assurance against potential risks involved in social interactions.

Zack and Knack (2001) show that cultural and social structures are the most important determinants of social trust at the national level. In their study, the social environment (similarity or differences in social norms and conventions), the economic environment (income and the income distribution), and the legal environment (institutions that enforce contracts) account for 76% of the variation in trust levels across countries.

It is persons who can trust or be trusting (Levi 1998, p. 79). An individual’s assessment of the trustworthiness of generalized others is subject to this person’s understanding of the content of social norms and conventions, and perceptions of the competence and willingness of institutional arrangements in the enforcement of trustworthiness.

Understanding the contents of social norms and conventions is the key to understanding their effects in rewarding compliance behavior and punishing noncompliance, which is fundamental in predicting the behavior of others. Individuals with limited knowledge of social norms and conventions are less likely to trust others within heterogeneous groups, especially in a socially diverse or a rapidly changing environment. People who are generally pessimistic about (reaching a societal or between-group consensus in) the existing normative values may have a higher preference for powerful institutional arrangements, such as strict government censorship and stiff penalties, as substitute of the informal constraints in the enforcement of trustworthiness.

Social trust is sustained by people’s confidence in the competence and commitment of formal structural arrangements in protecting the interest of generalized people (Brehm and Rahn 1997; Levi 1998, p. 81; Rothstein and Stolle 2002). In assessing whether “most people can be trusted”, one may also make an estimate of the trustworthiness of political and legal institutions. An individual who believes these institutions are effective, credible and unbiased agencies (in monitoring and enforcing trustworthy conduct) has probably a lower perceived likelihood of being exploited or taken advantage of in social interactions, and this individual is therefore more likely to express that “most people can be trusted” in the survey. An individual’s attitude toward institutional arrangements entails the perceived performance and motives of institutional arrangements in the distribution of resources. One may be more suspicious of the competence and impartiality of institutional arrangements in the enforcement of trustworthiness, when this person feels that institutional arrangements perform insufficiently in the distribution of wealth.

College Education and the Formation of Individual Social Trust

It has been commonly agreed in economic and social literature that education is a crucial determinant of social trust at the individual level. Highly-educated people are more likely to do well economically. The advantages in economic and social resources may bring highly-educated people more confidence in handling the risks involved in trusting generalized others. Knack and Keefer (1997) and Knack and Zak (2002) argue that trust is created in the education system by making individuals better informed and better at interpreting perceived information, as well as making them more conscious of the consequences of actions taken by themselves and others. Moreover, schooling might have an important socialization effect that may give young people a more positive attitude towards people in general (Bjørnskov, 2007). Yamagish (2003), p. 130) suggests that higher education makes students of elite college high trusters. According to Helliwell and Putnam (1999), higher average education levels may help to create a climate of trust that is self-reinforcing in which highly-educated people are in turn more likely to trust others.

In empirical studies, education is considered as an essential factor in the social trust equation. The estimated education effects vary across studies due to heterogeneities in survey sources and methodological or contextual variations. Huang et al. (2009) conducted a meta-analysis on the estimates of the education effects from empirical studies which have included education achievement as an independent variable and reported statistical data (t-statistics, p-value or standard error). In the meta-analysis, we synthesized 154 estimates of the education effects on social trust from 28 empirical studies and we found out that, overall, one additional year of schooling increases individual social trust by 4.6% of its standard deviation.

The correlation statistics or simple regression statistics in the existing studies do not, however, necessarily reflect a causal effect of education. Empirical studies may be exposed to the problem of omitted-variable bias (or education endogeneity) if they ignore the possibility that the choices of educational attainment and social trust formation are simultaneously affected by unobserved factors, such as cognitive ability, personality traits, as well as family backgrounds and interactions in early life.2 Coleman (1988) also argues for a reverse causal direction, as trust attributes in early-life might also lead to better educational outcomes by allowing students to gain access to the help of family and fellow students. In this case, the estimate of the education effect does not reflect a real causal effect, but a spurious relation. A rigorous empirical study should be able to disentangle the real effect of education from the potential influences of confounding variables or present evidence to demonstrate that unobserved heterogeneity across individuals does not cause a substantial bias in the estimation.

In a large amount of literature, individual past life experience and perceptions of cultural and social structures are considered as two primary aspects of social trust at the individual level. Until now, however, little (if any) empirical evidence can be presented to clarify how education contributes to the building of social trust via these two sources. The theoretical rationales of the education effect remain untested and hypothetical.

This paper attempts to provide an inclusive examination of the role of college education in the formation of social trust. We employ econometric techniques to isolate the potential influences of confounding variables and quantify the exact degree of college effect. We also examine the potential mechanisms for the education effect by testing two hypotheses that are established in line with the two primary sources of social trust at the individual level:

Hypothesis 1

College education promotes social trust via the heterogeneities (induced by education disparities) in individual life experience/development.

Highly-educated people have more chance to do well economically, and generally live in a community where there are less heterogeneities (in the identity of social norms and conventions) and lower crime rates. They are less likely to be exposed to the “darker side” of society that has an adverse impact on the formation of social trust. With the cognitive and perceptual experiences from and outside academic programs, a college education increases the individual’s capacity to communicate effectively and to socialize with others. People who have received a college education are more optimistic about controlling their own life-chances and engaging in close interaction with others, which leads to a higher belief in the benevolence of human nature in general, or at least a higher belief in the capacity of discerning and handling risks involved in trusting others.

Hypothesis 2

College education promotes social trust via the heterogeneities (induced by education disparities) in an individual’s perceptions of cultural/social structures.

In a democratic society, college institutions have a civil mission—educating their students to be effective and responsible citizens. A college education increases individual knowledge of the cultural environment, economic environment, and legal environment. High educated people are more likely to share a social consensus on normative values that create an incentive to honor trust, and they are more affirmative of the competence and willingness of social arrangements in the enforcement of trustworthiness and fairness. College institutions also offer various opportunities where students can interact and cooperate with people from different social groups (in the procedure of acquiring knowledge). The between-group interaction and cooperation may provide favorable information about the other group that would not otherwise be available (Kramer 2004, p. 152). They may enhance the perceived similarity and break down the conceptual boundaries between groups (Brewer and Kramer 1985), fostering a common knowledge that members of society share the same normative values. In general, a college experience expands the horizon of individuals on economic and social change, makes individuals more open-minded to accept otherness from heterogeneous groups, and makes individuals more affirmative of the trustworthiness of political and legal institutions.

A Brief Illustration of the Evaluation Techniques

We examine the causal effect of college education through the identification of the average treatment effect (ATE), which is an econometric measure, generally adopted in medical trials or policy evaluation, for the average causal difference in outcomes between the treatment and the control group. The expression “treatment effect” refers to the causal effect of a given treatment or policy on an outcome variable of scientific or policy interest (i.e. the health of the patients or the income of the workers). The average treatment effect is the average of the individual treatment effects across the whole population of interest. In this paper, college education is the treatment and the ATE denotes the average expected causal effect of college education relative to lower education on individual social trust.

We apply the bivariate probit (BVP) and the control functions probit (CFP) methods to handle the potentially endogenous relation between choice of college education and social trust.3 The BVP method has been widely used in medical evaluation to reduce the bias due to self-selectivity in the binary treatment choice. The BVP is a simultaneous equation model that controls for endogeneity in the likelihood of the joint sets of the treatment and outcome distribution. Bhattacharya et al. (2006) have an inclusive comparison on the performances of the probit, two-stage probit (or two-stage least squares), and BVP. They show that the BVP is the only method to produce a consistent estimator when there is an endogenous treatment.

The control functions probit (CFP) model is a special case of the control functions (CF) model, which is generally applied to evaluate the treatment effect on continuous outcome by controlling directly for the correlation between the treatment choice and the unobservable heterogeneity in the outcome variable (Heckman and Navarro-Lozano 2004; Blundell et al. 2005). Because the probit specification can be derived from a model involving a latent variable with a linear expression, the CFP model produces a good approximation of the true ATE in a binary response model.

The CFP, like the BVP, allows one to identify the real causal effect, and one can examine the presence of treatment endogeneity. In general, an exclusion restriction is required in the CFP and the BVP approaches, especially when there is inadequate variation in the observable characteristics.4

The NCDS Dataset and the Instrumental Variable Approach

To investigate the development of social trust and the impact of college education, an appropriate data set is indispensable in order to follow the respondents through multiple life stages. The rich data of a British cohort born in 1958 from the National Child Development Study (NCDS) offers an opportunity to perform such an analysis. The NCDS is a multi-disciplinary longitudinal study of all those living in the UK (England, Scotland, and Wales) who were born in the week 3–9 March, 1958 (including immigrants who were born outside the UK). The first three sweeps were carried out by the National Children’s Bureau in 1965, 1969 and 1974. The following three sweeps were carried out by the Centre for Longitudinal Studies (CLS) in 1985, 1991 and 1999–2000. The NCDS is widely used in economics, social and health sciences research to examine the patterns of human development that follow the lifespan.

Table 1 provides summary statistics of the main variables in this study. Information on social trust and information on college degree are extracted from the 1991 NCDS survey (the cohort members were 33 years old in 1991). The sample studied in this paper contains 10,441 observations.5 67.3% of the cohorts indicate that most people can be trusted and 14.8% of the respondents has a college degree.

Table 1

Descriptive statistics of the main variables

Variable	N	Mean	SD
Outcome variable (age 33)
Social trust	10,441	.672	.470
Treatment variable (age 33)
College education	10,441	.148	.355
Basic demographics (birth)
Male	10,441	.483	.500
Minority-non white	10,392	.025	.156
Social studies performance in class (age 15–16)
Excellent	7,457	.147	.354
Above average	7,457	.231	.422
Average	7,457	.334	.471
Below average	7,457	.143	.350
Lowest level	7,457	.115	.319
Math performance in class (age 15–16)
Excellent	7,999	.124	.330
Above average	7,999	.224	.417
Average	7,999	.351	.477
Below average	7,999	.167	.373
Lowest level	7,999	.130	.337
Residence region (age 15–16)
England (non-London)	9,935	.652	.476
London	9,935	.179	.383
Scotland	9,935	.103	.305
Wales	9,935	.056	.231
Father economic status (age 15–16)
Professional	7,165	.059	.236
Managerial	7,165	.212	.409
Non-manual-skilled	7,165	.101	.302
Manual-skilled	7,165	.431	.495
Non-manual-semi	7,165	.014	.116
Manual-semi	7,165	.125	.331
Unskilled	7,165	.045	.306
Mother economic status (age 15–16)
Professional	7,658	.003	.056
Managerial	7,658	.110	.313
Non-manual-skilled	7,658	.219	.414
Manual-skilled	7,658	.047	.212
Non-manual-semi	7,658	.117	.322
Manual-semi	7,658	.101	.302
Unskilled	7,658	.063	.243

All covariates are extracted from the 1973–1974 survey (except for demographic information), during which the cohort members were 15–16 years old. They were approaching the end of compulsory education (secondary education was compulsory for all pupils between the ages of 11 and 16 in the UK). At that age they are faced with O/A-level examination(s)6 as well as a choice of further education.

Parental socioeconomic covariates include indicators of parental education level and parental economic class from the 1973–1974 survey. Other covariates of family background contain information of whether parent(s) changed (as a result of divorce, death etc.), length of time in the current address, and the number of siblings of the respondent.

Academic performance and motivation in adolescence are crucial predictors for the ultimate education outcome. They are also considered to be powerful determinants of social class and income in adulthood that directly affect the development of social trust. Yamagish (2003) believes social trust is a by-product of social intelligence, the ability to detect and process signs of risks in social interactions, which depends on cognitive resources. Using the teacher’s report in the 1973–1974 survey, we collect information of respondents’ performance in social studies and math, and whether they were absent from school for trivial reasons. We also collect, from the teacher’s report, information of respondent’s socializing behaviour (withdrawn score) in school, and information of certain school characteristics, such as school enrolment, teacher/student ratio, availability of facility resources, and attendance rate.

Instrumental Variables

The BVP and CFP methods require a valid instrumental variable in their estimation procedures. The instrumental variable should have the property that its variation is associated with change in educational attainment, but does not lead to change in the social trust, aside from the indirect route via educational attainment. In other words, an exclusion restriction is imposed for the instrumental variable. We construct such an instrument from the information of the length of schooling absence due to illness (or the absence length for brevity), which is also reported in the 1973–1974 survey. From our perspective, the length of schooling absence due to illness can be decomposed into systematic components and non-systematic components. The systematic components arise from inherited health status and family background, such as living conditions, nutrition intake, parental socioeconomic status, and the roles of parents in the family. The systematic components are expected to have a lasting influence over the lifespan, impacting education achievement and possibly the level of social trust in adulthood.

The non-systematic components arise from haphazard events, such as accidents, illness (cold or throat) due to unexpected weather changes and other incidents. For students with poor health or chronic conditions, class cancellation/re-arrangement due to adverse weather or provisional change in school programs can also been seen as the cause of the non-systematic components, in the sense that these students might have been absent from school in the original class arrangement. The non-systematic components are not supposed to have a lasting health influence over the lifespan, and they should not have any direct impact on the level of social trust in adulthood.

Because of the timing of its occurrence, both the systematic and non-systematic components of the absence length are strongly correlated with the respondent’s grades at the A-level exams, and subsequently their chance of receiving higher education. A valid exclusion restriction is obtained for social trust if the non-systematic components can be separated from the systematic components. We achieve this design by regressing the absence length on relevant information and breaking down the dependent variable. Family background, parental socioeconomic status, and adverse health information in early life (chronic illness, low birth weight, maternal smoking during pregnancy, etc.) are included in the regression to decompose the absence length. Besides, dummy variables are created for each type of systematic illness reported for the schooling absence except for cold, throat, periods, accidents or injuries, and interacted with other adverse health factors in the regression of the absence length. The intuition is that, if an individual has certain health problems, and misses some school days because of non-accidental or chronic illness, it is highly plausible that these interactions capture some systematic health problems.

One may expect that a student might play truancy from school in the name of illness because of their distaste for schooling or poor relations with other school children, and consequently, the predicted residuals might not be excluded from the equation of social trust. We believe that this should not be a problem because in the decomposition process we will control for the teacher’s perspective of the truancy frequency of the students. Moreover, all covariates in the equation of social trust, i.e. academic performance in math and social studies, (unsociable) relations with other children, and information of school resources are included in the decomposition of the absence length. The rich information included in the decomposition process should minimize the potential influence of fabricated illness on the validity of the non-systematic components.

As relevant covariates are included in the regression of the absence length, we obtain its predicted value—ideally the systematic components, and its predicted residual value—ideally the non-systematic components and the instrumental variable. Statistical proof of the validity of the instrumental variable is presented in Table 2. Part A of Table 2 provides the test statistics for the correlations between the respondent’s contemporary health indicators (at age 33) and the predicted residual value of the absence length. For comparison, similar correlation tests are also performed for the absence length and for the predicted value of the absence length. It is straight-forward that the absence length and its predicted value are strongly correlated with contemporary health status and chronic conditions, while the instrument has no significant correlation. These statistics provide strong support for the design principle adopted in this research that the non-systematic components are not supposed to have a lasting health influence over the life span.

Table 2

Test statistics on the validity of the instrumental variable

	Absence length		Systematic term		Non-systematic term
	Coef.	p-value	Coef.	p-value	Coef.	p-value
A. Correlation with mid-life health
General health status at 32–33	−.10	.00	−0.10	.00	−.01	0.38
No. Chronics suffered at 32–33	.11	.00	0.16	.00	.01	0.40
No. Chronics ever suffered	.13	.00	0.18	.00	.01	0.23
B. Correlation with residuals of trust
Trust residuals unrelated to exams	−.03	.01	−.04	.00	−.01	0.66

Indicator of general health status has 4 categories: 0-poor, 1-fair, 2-good, 3-excellent

Part B of Table 2 provides evidence for our argument that the predicted non-systematic components of the absence length have an impact on social trust via individual’s exam grades (and thus choice of college education). We break down the trust outcome by the number of A-levels that the respondent has passed (which is an important qualification for college entrance in the UK) by age 20. Then we perform a correlation test for the instrument and the residual value of the trust outcome unrelated to the number of passed A-levels. Similar correlation tests are applied for the absence length and for the predicted systematic components of the absence length. Once again the absence length and its predicted value are strongly correlated with the residual value of the trust outcome unrelated to the number of passed A-levels, while the instrument has a negligible correlation with the residual value of the trust outcome.

Figures 1 and 2 offer additional proof of the validity of the instrumental variable. Figure 1 depicts the kernel density of the residual value of the absence length for trusters and non-trusters in the group without college degree. Figure 2 depicts the kernel density of the residual value of the absence length for trusters and non-trusters in the group with a college degree. Provided that the instrument only impacts social trust via education choice, the kernel densities of the residual value of the absence length should not be diverting for trusters and non-trusters in the same education group. It is straight-forward in Fig. 1 and 2 that the kernel densities are overlapping for the same education group. Therefore the distribution of the residual value of the absence length does not vary between trusters and non-trusters and can be regarded as an applicable exclusion restriction (or instrumental variable) for the outcome of social trust.

Fig. 1

Kernel density of the instrument for trusters and non-trusters in the group without college degree

Fig. 2

Kernel density of the instrument for trusters and non-trusters in the group with college degree

Results from the Empirical Examinations

Identification of ATE

As a first step in our empirical study, we employ the probit, bivariate probit (BVP) and the control functions probit (CFP) to quantify the causal effect, in terms of ATE, of college education on individual social trust. The findings from these evaluation methods are presented in Table 3. In the baseline probit, which includes demographic characteristics, residence region, and family composition as control variables, we observe a strongly significant estimate of the college effect. The estimated coefficient of ATE is .123 (p-value < .001) in terms of probability change. This indicates that the level of social trust of those with college degree, other conditions (in early life) being equal, exceeds that of those without a college degree by 12.3% age point. In other words, a college education increases the probability of trusting generalized others by one-fourth of its standard deviation. In the full-specification probit, where parental socioeconomic status, academic performance, report of school truancy and personality strength are added in the equation, the estimated coefficient of ATE drops from .123 to .074, although the latter remains statistically significant (p-value < .001).

Table 3

Estimates of ATE and endogeneity test

	ATE estimation		Endogeneity test
	ATE	SE	p-value
Baseline probit	.123***	.012	–
Full-specification probit	.074***	.014	–
BVP	.077	.048	.923
CFP	.077*	.047	.921

The coefficients are reported as probability change

* Significant at the 10% level

*** Significant at the 1% level

Since the choices of educational attainment and the development of social trust can be simultaneously affected by some individual characteristics that are not observable, the probit estimator may suffer an omitted-variable bias. We employ the endogeneity models, namely, the BVP and CFP methods, to identify the real causal effect through the exogenous variation in educational attainment induced by the instrumental variable—the predicted non-systematic components of the absence length. The estimated ATE coefficient is .077 in both the BVP method and the CFP method. They do not differ from the coefficient identified by the full-specification probit.

The trivial correlation between college education and the unobservable heterogeneity in social trust, according to the endogeneity test in the BVP and CFP models (p-value > .9), suggest that omitting-variable bias is not a severe problem in the estimation of the college education effect. The rich information of early-life development, especially parental socioeconomic status, academic performance, and unsociable behavior in adolescence, is the key for reducing the probability of unobserved characteristics in childhood and adolescence simultaneously impacting educational attainment and social trust.7

The results from the ATE identification indicate that the level of social trust of college graduates exceeds that of non-college graduates by 7.5% point. This implies that a college education increases the probability of trusting others by about 16% of its standard deviation. The regression statistics in the first step of our empirical study confirm the significance of college education in the formation of social trust.8 However, these statistics cannot provide any clear mechanisms to explain how college education contributes to the building of social trust. The theoretical expectations about the role of education remain hypothetical and untested

The Investigation of the Hypothetical Mechanisms

The findings from the first step of our empirical study do not clarify how college education fosters social trust. In the second step we use contemporary information to examine whether individual experience/development and individual perceptions of cultural/social structures are primary channels in the causal linkage between college education and social trust.

The contemporary variables applicable to the hypothetical channels are drawn from the 1991 survey, the same survey from which information on social trust is collected. As presented in Table 4, these contemporary variables of the NCDS observations are classified into 3 categories: Category i—characteristics of life experience/development since adulthood; Category ii—views on the status of social norms and conventions; and Category iii—views on the competence and motives of formal/institutional arrangements in the enforcement of trustworthiness. Contemporary variables in category ii and category iii represent the heterogeneities in individual perceptions of cultural and social structures. We classify them into two categories to distinguish the informal contextual characteristics from the formal contextual characteristics.

Table 4

Contemporary variables for the examination of the hypothetical mechanisms

i. Indicators of Life experience and development since of adulthood	a. Economic class of the first job;
	b. Economic class of current/last job;
	c. Respondent and his/her partner have debt;
	d. Happiness, all things considered;
	e. Currently married and remained in the first marriage*;
	f. Capacity in handling/avoiding argument.
ii. Views of the status of social norms and conventions (Optimistic vs. pessimistic)	a. The young are losing respect on traditional values;
	b. Would not mind if the neighbor is from other race;
	c. Law breaker should be given stiffer sentences;
	d. Death penalty is the most appropriate sentence for some crimes.
iii. Views of the competence and willingness of social arrangement in the enforcement of trustworthiness (Optimistic vs. pessimistic)	a. Ordinary people have no say in what government does;
	b. One law exists for the rich and one law exists for the poor;
	c. No political party would benefit people like the respondent;
	d. Government not doing enough in redistributing wealth.

* Respondents also reported in the 1991 survey their satisfaction of current relationship. Since this variable has an identical say as (and it can be used to substitute) the indicator of respondents’ marital status in the investigation of the hypothetical mechanisms, it is not included as one of the indicators in category i

The category of individual life experience/development since adulthood includes 6 indicators: economic class of the first job after schooling; economic class of the current job or economic class of the last job for those currently not in the job; debt status of the respondents and their partner; sense of happiness with all things being considered; marital status of the respondents (whether they have been married and remained in the first marriage); and sense of the capacity in avoiding arguments. The information transmitted by these variables enables us to explore whether college graduates have a higher belief in the benevolence of human nature in general, or at least a higher belief in the capacity of discerning, handling and taking risks in social interactions, for the reason that they are more likely to stay in a well-to-do environment, more effective in integrating into close interaction with others, and more optimistic of controlling their own life-chances.

The category of individual views of the status of social norms and conventions includes 4 indicators: concern about whether the young are losing respect for traditional values; concern of whether the neighbor is from another ethnic origin; concern about whether law breaker should be given stiffer sentences; concern about whether the death penalty is the most appropriate sentence for some crimes. These indicators offer information about the respondents’ confidence of social consensus on normative values, tolerance of otherness from heterogeneous group, and their dependence of formal/institutional arrangements as substitute of the informal constraints in the enforcement of trustworthiness.

The category of individual views of the competence and willingness of formal/institutional arrangements includes 4 indicators: sense of people like me (the respondent) having no say in what government does; sense of the existence of two different laws for the rich and for the poor; and sense of no political party being beneficial to people like me (the respondent); and sense of government doing enough to ensure fairness in the wealth distribution. These indicators offer information on the respondents’ confidence of the reliability and goodwill of formal structural or institutional arrangements in protecting the interest of generalized people and ensuring fairness in the enforcement procedure.

The information contained in category ii and category iii enables us to investigate whether a college experience expands people’s horizon on economic and social change, improves their understanding of the existing social values, makes them more open-minded to accept otherness from heterogeneous groups and more affirmative of the competence and fairness of institutional arrangements, and subsequently leading to a higher trust in people in general.

In the investigation of the hypothetical channels, each contemporary variable is introduced separately into the original probit regression of social trust, with all early-life covariates remaining in the equation.9 We examine the change in the estimated ATE of college education due to the introduction of the contemporary variables. If, for example, college education is a key determinant of life happiness in adulthood and the variation of life happiness, on account of education differences, is strongly and positive associated with the variation in social trust, the introduction of life happiness in the original full-specification probit regression should pick up, at least partially, the causality from college education to social trust. We should then expect a substantial drop in the estimate of the college effect.

Table 5 presents the ATE estimates after we introduce each contemporary variable separately into the full-specification probit model.10 Controlling for the heterogeneities in contemporary life development does not have any noticeable impact on the estimates. These six estimates range from .072 to .074, and three of them are precisely .074, which are exactly the same as the estimate identified by the full-specification probit model (see Table 3).

Table 5

Estimates of the college effect after controlling for contemporary variable

	ATE	SE	N
Category i: individual life experience/development since adulthood
Economic class of the first job (including housework)	.074***	.014	10,441
Economic class of current/last job	.072***	.014	10,441
Respondent and his/her partner have debt	.073***	.014	10,280
Happiness, all things considered	.074***	.014	10,283
Currently married and remained in the first marriage	.074***	.014	10,298
Capacity in handling/avoiding argument	.075***	.014	10,411
Category ii: Views of the status of social norms and conventions
Would not mind if the neighbor family are from other race	.059***	.014	10,405
The young are losing respect on traditional values	.052***	.015	10,359
Death penalty is appropriate for some crimes	.046**	.015	10,410
Law breaker should be given stiffer sentences	.055***	.015	10,363
Category iii: Views of the competence and motives of formal and institutional arrangement
Ordinary people have no say in government	.064***	.014	10,332
Different laws exist for the rich and for the poor	.070***	.014	10,388
No political party would benefit me	.063***	.014	10,353
Government not doing enough in redistributing wealth	.072***	.014	10,378

** Significant at the 5% level

*** Significant at the 1% level

We observe a substantial drop in the estimates following the introduction of contemporary variable in category ii (views of social norms and conventions). These four estimates range from .046 to .059, which are 20–40% smaller than the estimate identified by the full-specification probit model. It turns out that lower-educated people are more reluctant to accept in the neighborhood a family from another ethnic origin, they are more pessimistic about the significance of the existing social values, and they have a higher preference for stiffer laws, or even the death penalty as a protective assurance against the perceived risks of noncompliance of social norms and extreme non-cooperative behavior.

Controlling for the contemporary variables in category iii (views of formal or institutional arrangements) causes a moderate drop in the estimates. The views of the impartiality and fairness in legal system and the views of the performance of government in distributing wealth have a trivial effect on the estimates of ATE, while the views of the government or political parties (in terms of representing the interest of people in general) lead to 12–15% drop in the estimates.

For a more perceptual comparison, we present in Fig. 3 bar graphs for the estimate of ATE obtained from the original probit model in which no contemporary variable is included, and from the augmented-probit model in which the entire set of contemporary variables in category i, category ii, and category iii are included separately as covariates. It is straight-forward that controlling for the entire set of category i indicators only causes a negligible change in the estimated ATE, while controlling for the entire set of category ii indicators and category iii indicators reduces the estimate by 60% (from .074 to .029) and 30% (from .074 to .051), respectively. Information contained in category ii and category iii depict individual perceptions of cultural and social structures. Controlling for both category ii and category iii in the regression reduces the estimate of ATE by 77% (from .074 to .017, presented as the fifth bar in the figure). These findings suggest that a college education experience enhances the perceived similarity and breaks down the conceptual boundaries between social groups. College education is an effective approach in fostering a social consensus on normative values that create an incentive to honor trust (60% of the estimated causal effect acts through the heterogeneities in social norms and conventions). Our investigation provides substantial evidence to support the hypothesis that individual perceptions of cultural and social structures are the key channel in the causal linkage between college education and social trust.

Fig. 3

Bar graph for the college effect

Early-Life Backgrounds Versus Contemporary-Life Development

Our investigation based on contemporary variables indicates that individual life experience/development since adulthood is not a key channel in the causal connection between college education and social trust. We take efforts to give an illustration on the raison d’être underlying this somewhat unexpected finding.

Individual life experience is strongly associated with social trust in theoretical and empirical literature on social trust (Brehm and Rahn 1997; Offe 1999; Hardin 2003; Alesina and La Ferrara 2000, 2002; Paxton 2007). Our study also has similar findings. We use the probit model to regress social trust exclusively on the entire set of category i indicators. We repeat this procedure for category ii indicators and for category iii indicators. The estimates of each set of these indicators have strong statistical significance. The pseudo-R squares obtained from the probit estimation are also similar for these three categories, which indicates that contemporary variables in category i have the same explanatory power as those in category ii or category iii.

Endogeneity models are employed to evaluate the causal effects of college education on the heterogeneities in contemporary-life development. We observe a substantial and positive influence of college education on economic condition and personal happiness. College education only has a trivial effect on marital status (or satisfaction with marriage) and capacity in handling arguments (detailed findings are presented in Appendix B).

At first glance, we are confronted with a paradox: individual life experience and development since adulthood have a negligible role in the causal pathway from college education to social trust, although education disparities are making a large difference in socioeconomic status and perceived happiness, and these factors are strongly associated with social trust.

We obtain some insight in this paradox by an exploration of the development of social trust in early life. Trust can be learned in early life and persists as part of a core personality trait, unless challenged by trauma (Uslaner 2002, pp. 160–90; Stolle 2002). Family background and environmental influences are considered to have a direct and lasting impact on social trust at the individual level. It is also widely believed and well documented that the heterogeneities in family and environmental parameters are a major source of the variations in the socioeconomic outcomes of children. Therefore, we claim that the association between contemporary-life success (economic conditions, social class and life happiness) and social trust reflects, at least partially, the lasting effect of early-life factors on individual development and social trust in the life course. When a large proportion of the correlation between contemporary-life success and social trust is attributable to the lasting effect of childhood and adolescent experience, college education may not have much causal influence on social trust via its power in promoting individual development in adulthood.

To support our argument, we break down the outcome variable of social trust with early-life information by a probit model.11 The residual variable of social trust denotes the variations of social trust that cannot be explained by early-life information. Provided that current-life status has a direct and strong effect on social trust, which is not attributable to the lasting effect of childhood experience and adolescence experience, we should expect a strong correlation between the residual variable and the contemporary variables of current-life success (economic conditions, social class and life happiness). As shown in Table 6, it turns out that economic class and conditions have no statistically meaningful correlation with the residual variable of social trust that cannot be explained by the early-life variables, albeit that they have a very strong correlation with the outcome variable of social trust. Current sense of happiness remains strongly associated with the residual variable, but the correlation coefficient drop by 1/3.

Table 6

Evidence of the lasting effects of early-life backgrounds on the development of social trust

Development/success since adulthood	Residual variable of social trust		Outcome variable of social trust
	Coef.	p-value	Coef.	p-value
Economic class of the first job	.00	.88	.06	.00
Economic class of current/last job	.01	.57	.06	.00
Respondent and his/her partner have no debt	.01	.18	.02	.03
Happiness, all things being considered	.06	.00	.09	.00

Table 6 indicates that a large proportion of the correlations between indicators of development/success since adulthood (especially for past and current economic class) and social trust are indeed attributable to the lasting effect of experiences formed in childhood and adolescence. It suggests that college education has little causal influence on social trust via its power in promoting development in adulthood.

Conclusion

In this paper we have explored the role of college education in the formation of social trust, using the rich data of a British cohort. The theoretical review on the micro aspects of social trust considers risks as an essential element and emphasizes life experience/development and perceptions of cultural/social structures as two primary sources of social trust at the individual level. In the empirical study, we quantify the causal effect of college education and we test the validity of the hypothetical mechanisms in the causal pathway from college education to social trust.

Isolating the influences of confounding variables is the key to quantify the causal effect of college education. Empirical studies may be exposed to the problem of omitted-variable bias (or education endogeneity) if they ignore the possibility that the choices of educational attainment and social trust formation are simultaneously affected by unobserved factors. To identify the overall causal effect of college education, we should not include contemporary indicators of individual development or individual perceptions of cultural/social structures as the explanatory variables. These contemporary indicators are subject to the influences of educational attainment and conditioning on such variables would block the part of the causal effect of education experience that acts through these variables.

In the first part of the empirical analysis we identify the average treatment effect (ATE) of college education, with an emphasis on tackling education endogeneity. Strong evidence is found to support the substantial role of college education in promoting individual social trust. We show that the rich information of adolescent development and experience is crucial in reducing omitted-variable bias. The estimate obtained from the baseline probit, in which we do not controlled for information of parental socioeconomic status, cognitive capability, academic motivation, and personality strength in adolescence, is 66% larger than the estimates obtained from the full-specification probit and endogeneity models. Lack of information of adolescence development and experience indeed causes a upward omitted-variable bias in the estimation.

Using the measurement adopted in the meta-analysis (Huang et al. 2009), the base-line probit in this study indicates that one additional year of schooling increases individual social trust by 4.4% of its standard deviation. The full-specification probit and endogeneity models indicate that one additional year of schooling increases individual social trust by merely 2.6% of its standard deviation. The base-line probit model produces an estimate that is equal to the synthesized education effect (which is 4.6% of the standard deviation in our meta-analysis) on the existing studies of social trust. Since most of the existing studies do not consider education as an endogenous variable and cannot control for information of early-life development, their estimations are not different from our base-line probit estimation and the estimates, as a consequence, are exposed to a substantial upward bias.

In the second part of the empirical analysis we examine the change in the estimate of ATE due to the introduction of individual information of contemporary-life experience/development and contemporary perceptions of cultural/social structures. The hypothesis that individual experience is a key channel from college education to social trust fails to find support in the investigation. There is a negligible change in the estimates of the ATE due to the introduction of economic conditions, personal happiness, marriage status, and capacity in effective interactions with others. Our further analysis shows that a large proportion of the associations between contemporary-life success and social trust are indeed attributable to the lasting effect of experiences formed in childhood and adolescence.

Individual understanding of cultural and social structures explains 77% of the college effect on social trust. College education promotes individual social trust because it plays a positive and fundamental role in expanding the horizon of individuals on economic and social change, making individuals more open-minded to accept otherness from heterogeneous groups, and inspiring consensus on normative values and affirmative attitude toward institutional arrangements.

This paper contribute to the research of social trust in four aspects:

Firstly, our study provides strong econometric evidence to support the positive role of college education in the building of social trust. Although education has been commonly believed to be one of the most important determinants of social trust at the individual level, little evidence has been presented to quantify the causal effect of college education. Our study shows that a college education increases one’s probability of trusting generalized others by about 16% of its standard deviation.

Secondly, previous studies have not, in general, provided an estimate of the true causal effect of education. The majority of these studies consider education as an exogenous independent variable and they were not able to control for information on early-life development. Combining results from this study and our previous meta-analysis, we show that there is a roughly 70% upward bias in the estimate of education when empirical studies fail to control for early-life information and education endogeneity in the estimation.

Thirdly, our study verifies that individual understanding of cultural and social structures is the primary channel in the causal linkage between college education and social trust. The theoretical rationale that college education promotes social trust via its effect on later-life experience/development fails to find support in our study.

Fourthly, our study indicates that the strong association between contemporary-life success or happiness and contemporary level of social trust might largely reflect the lasting effect of early-life factors on later life development. A development perspective on social trust enables researchers to gain a better understanding of the development of social trust at the individual level.

Table 7

Regression statistics of key covariates in the estimations of social trust

	Probit		BVP		CFP
	Coef.	SE	Coef.	SE	Coef.	SE
Male	−.056***	.010	−.059***	.010	−.056***	.010
Non-white	−.146***	.034	−.155***	.034	−.146***	.034
Father age on leaving school	.011**	.004	.008*	.004	.011**	.004
Self-rated relation with mother	−.028***	.007	−.027***	.007	−.028***	.007
Parent change since birth	−.057***	.022	−.057***	.022	−.057***	.022
Withdrawn (unsociable) score	−.013**	.005	−.013**	.005	−.013**	.005
Math rating in class	.023***	.007	.019**	.008	.023***	.007
N	10,441		10,441		10,441

The coefficients are reported as probability change

* Significant at the 10% level

** Significant at the 5% level

*** Significant at the 1% level

Table 8

Estimates of the college effect on contemporary variable

	Coef.	SE	Endogeneity (p-value)	N
A. Simple model (OLS or probit model)
Economic class of the first job (including housework)	.523***	0.045	–	8,724
Economic class of current/last job	.774***	0.037	–	9,846
Respondent and his/her partner have debt	−0.022	0.015	–	10,280
Happiness, all things being considered	−0.006	0.015	–	10,283
Currently married and remained in the first marriage	−0.007	0.015	–	10,298
Capacity in handling/avoiding argument	−0.018	0.012	–	10,411
B. Endogeneity model (CF or CFP model)
Economic class of the first job (including housework)	0.244	0.156	0.061	8,724
Economic class of current/last job	.317**	0.015	0	9,846
Respondent and his/her partner have debt	−.118**	0.047	0.042	10,280
Happiness, all things considered	.128**	0.056	0.011	10,283
Currently married and remained in the first marriage	0.066	0.05	0.139	10,298
Capacity in handling/avoiding argument	0.036	0.043	0.649	10,411

** Significant at the 5% level

*** Significant at the 1% level