Page 1

Does intelligence foster generalized trust? An empirical test using the UK

birth cohort studies

Patrick Sturgis a,⁎, Sanna Read b, Nick Allum c

a Division of Social Statistics, University of Southampton, UK

b London School of Hygiene and Tropical Medicine, UK

c Department of Sociology, University of Essex, UK

a r t i c l e i n f o

a b s t r a c t

Article history:

Received 11 May 2009

Received in revised form 10 November 2009

Accepted 10 November 2009

Available online 30 November 2009

Social, or ‘generalized’ trust is often characterised as the ‘attitudinal dimension’ of social capital.

It has been posited as key to a host of normatively desirable outcomes at the societal and

individual levels. Yet the origins of individual variation in trust remain something of a mystery

and continue to be a source of dissensus amongst researchers across and within academic

disciplines. In this paper we use data from two British birth cohort studies to test the

hypothesis that a propensity to express generalized trust varies systematically as a function of

individual intelligence. Intelligence, we argue, fosters greater trust in one's fellow citizens

because more intelligent individuals are more accurate in their assessments of the

trustworthiness of others. This means that, over the life-course, their trust is less often

betrayed and they are able to accrue the benefits of norms of reciprocity. Our results show that

standard measures of intelligence administered when cohort members were aged 10 and 11

can explain variability in expressed trust in early middle age, net of a broad range of

theoretically related covariates.

© 2009 Elsevier Inc. All rights reserved.

Keywords:

Intelligence

Trust

Cohort study

Survey

Trust is essential to the effective functioning of social,

political, and financial systems. Without trust, many of the

everyday activities and transactions that we take as routine

and commonplace would not be possible. We trust the baby-

sitter to look after our children, the scientist to develop safe

technologies, and as recent events have brought prominently

to our attention, we trust banks to take care of our deposits.

Trust, in its broadest sense, is the belief that others will not

knowingly act in a way that is detrimental to our interests, or

better still, will act in a way that serves to maximise them

(Barber, 1983; Hardin, 2006). So, while trust can bring many

payoffs — we can be at work while someone else looks after

our children, derive the benefits and efficiencies of new

technologies, and accrue interest payments on our savings —

it is also inherently risky in the sense that, if our trust is

betrayed, we would have been better off not trusting in the

first place (Fehr & Gintis, 2007; Ostrom & Walker, 2003).

The alleged benefits of trust are now well documented. At

the country level, aggregate trust is correlated with economic

development (Fukuyama, 1995; Knack & Keefer, 1997;

Putnam, 1993), democratic governance (Inglehart, 1997),

lower levels of political corruption (La Porta, Lopez-de-

Silanes, Shleifer & Vishny, 1999; Rothstein & Uslaner, 2006),

income equality (Uslaner, 2002), and lower rates of crimi-

nality and juvenile delinquency (Halpern, 2001; Sampson,

Raudenbush & Earls, 1997). Trust also appears to be a

desirable commodity for individuals as well as societies,

with trusters concentrated amongst the better educated (Nie,

Junn & Stehlik-Barry, 1996; Paxton, 2007; Putnam, 2000),

those in professional occupations and higher income groups

(Alesina & Ferrera, 2002; Li, Pickles & Savage, 2005) and least

often by divorcees (Patterson, 1999), the unemployed

(Brehm & Rahn, 1997), ethnic minorities with a history of

discrimination (Alesina & Ferrera, 2000), and those in poorer

health (Ichiro Kawachi, 1997; Kawachi, Kennedy & Glass,

Intelligence 38 (2010) 45–54

⁎ Corresponding author. Division of Social Statistics, School of Social

Sciences, University of Southampton, Southampton SO17 1BJ, UK. Tel.: +44

23 8059 4547; fax: +44 23 8059 3846.

E-mail address: p.sturgis@soton.ac.uk (P. Sturgis).

0160-2896/$ – see front matter © 2009 Elsevier Inc. All rights reserved.

doi:10.1016/j.intell.2009.11.006

Contents lists available at ScienceDirect

Intelligence

1999). As a robust indicator of being what Newton has

called one of “society's winners” (Newton, 1999 p.185), the

origins of generalized trust has, understandably, become a

key explanandum across the social sciences.

The vast majority of this body of evidence in support of the

ameliorative effects of trust on important life outcomes is

derived from sample surveys conducted around the world

over the past fifty or so years, in which respondents are asked

some variant of what has come to be known as the

Generalized Trust Question (GTQ) (see Dekker, 2003 for a

discussion). The GTQ requires respondents to make a binary

choice between the following alternatives: 1. ‘in general,

most people can be trusted’, and 2. ‘you can't be too careful in

dealing with people’. Those selecting the first response

alternative are deemed to be ‘trusters’ and have been found,

almost without exception, to be disproportionately repre-

sented amongst the ‘healthy, wealthy, and wise’ (Delhey &

Newton, 2003, 2005). A defining characteristic of the GTQ is

that no specific referent, or ‘trustee’ is mentioned; the

respondent is asked to make a judgement about the

trustworthiness, not of specific individuals or groups, but of

people in general. And, while some regard the notion of trust

in people we do not know as inherently nonsensical (cf.

Hardin, 2001), for others it is exactly this type of ‘generalized’

trust, as opposed to the kind of ‘strategic’ trust1 that develops

between acquainted individuals, that is key to solving large-

scale collective-action problems. As Uslaner puts it “strategic

trust can only lead to cooperation among people you have

gotten to know, so it can only resolve reasonably small-scale

problems” (Uslaner, 2002 p20). When we speak of general-

ized trust, then, we are not referring to the kind of mutually

encapsulated self-interest that can develop between social

actors who are known to one another but to an essentially

indiscriminate belief in the general benevolence of one's

fellow citizens.

Yet, in conceptualizing trust in this manner, we are faced

with an apparent paradox; while this kind of generalized

trust is known to be strongly correlated with consensual

norms of socio-economic success and achievement, its

essentially indiscriminate nature bears a striking similarity

to notions of naivety, or gullibility. For, if we believe, a priori,

that anyone we happen to encounter can be trusted, will we

not be susceptible to repeated betrayal by assorted free-riders

and con-artists? Indeed, it has even been argued that to

define trust in this manner is logically inconsistent with the

apparently widespread prevalence of trusters throughout the

world, “because such trusters must be too gullible to prosper

in most societies in which studies of trust have been done”

(Hardin 2001, p35). How, then, are we to reconcile these

apparently contradictory perspectives on generalized trust;

that it is a driver of social and economic success, while

simultaneously being the hallmark of the gullible and the

credulous?

For Yamagishi, the answer lies in carefully distinguishing

between what he characterises as an individual's “default

expectation” of the trustworthiness of others and the actual

trust behaviour of that individual in specific contexts

(Yamagishi, 2001). While an individual might have a

generally optimistic attitude toward engaging in potentially

symbiotic reciprocal transactions with strangers, her decision

over whether to trust a specific ‘X to do Y’ will be highly

contingent upon the signs and signals given off by X at the

particular point at which she must make the trust decision. At

this point, trust becomes far from indiscriminate. On the

contrary, the truster is now carefully monitoring for indica-

tions of (un)trustworthiness, decoding subtle and often

highly complex situational cues relating to the intentions of

the trustee, in order to gain a strategic advantage (Gambetta,

1988). Thus, it is precisely the more intelligent social actors

who disproportionately reap the benefits of reciprocity, by

carefully endowing their trust only in those who are unlikely

to betray it. This type of intelligence can thus be viewed as a

benign form of Machiavellianism (Parales-Quenza, 2006), in

which socially astute individuals are rewarded by exerting a

high degree of control over the conditions in which they are

willing to sanction trust. The counter-side to this argument, of

course, is that the less socially astute are frequently betrayed

in trust relations and thus progressively withdraw from

potentially fruitful interactions in the future. Not only do they

then pay the opportunity costs of potential gains from ‘tit-for-

tat’ reciprocity, they also progressively come to the view that

most people cannot be trusted.

From this perspective, then, trust is at heart a problem-

solving activity, in which intelligent social actors are better

able to evaluate the trustworthiness of interaction partners,

endowing them with a progressive and accumulative advan-

tage over the life-course. A key implication of what we

refer to in the remainder of the paper as the intelligence–trust

(I–T) hypothesis is that, through repeated exposure to gainful

trust-based interactions, socially intelligent individuals de-

velop a ‘default expectation’ of trust in unknown others

while, by implication, the reverse is true for their less

intelligent counter-parts. We test this I–T hypothesis by

evaluating whether a default position of trust in adulthood is

positively correlated with intelligence measured in child-

hood, net of a range of theoretically related covariates. To do

this, we draw on unique data collected in the United Kingdom

over the past fifty years, in which every child born in a single

week in 1958 and 1970 respectively were interviewed at

regular intervals until (at the most recent sweep) early

middle-age. The paper proceeds in the following manner. We

begin by reviewing existing empirical research into the

relationship between intelligence and generalized trust. We

then describe in detail the two data sets to be used and set out

the measures upon which our analysis is based, before

presenting the results of our empirical analysis. We conclude

with a consideration of both the limitations and substantive

significance of our findings for an understanding of the

relationship between intelligence and generalized trust.

1. Intelligence and trust

What, then, does the existing empirical record tell us

about the relationship between intelligence and trust? Early

investigations in the social psychological literature generally

found either no relationship, or a negative association

1 The same distinction between generalized and strategic trust is also

drawn by Putnam, though he contrasts what he terms ‘thick’ trust in known

others with the ‘thin’ trust we have in those with whom we are not

personally acquainted.

46

P. Sturgis et al. / Intelligence 38 (2010) 45–54

between trust and cognitive ability, variously conceived. For

example, Garske (1975) found trust to be negatively corre-

lated with intelligence and, in a separate study, ‘construct

complexity’ (Garske, 1976). Similarly, Gurtman and Lion

(1982) report that high trusters were slower than low

trusters in recognizing adjectives signalling untrustworthi-

ness that were displayed to them on a tachistoscope, while

Rotter (1980) found scores on scholastic aptitude tests

among a sample of college students to be unrelated to his

Interpersonal Trust Scale (Rotter, 1967). And, although in his

own analysis of the Interpersonal Trust Scale, Rotter empha-

sised that trust and gullibility are logically and empirically

distinct, this can hardly be taken as supporting the expecta-

tion that generalized trust should be higher amongst more

intelligent individuals. On the whole, then, we find little

evidence in support of the Intelligence–Trust hypothesis

in these early investigations. Indeed, they appear to lend

considerably more weight to the opposite view, that gen-

eralized trust is coterminous with gullibility.

On closer inspection, however, all of these studies have

critical limitations with regard to assessing the Intelligence–

Trust hypothesis. Leaving to one side the problem of external

validity deriving from their exclusive use of student samples,

all of these studies employ Rotter's Interpersonal Trust Scale

as the measure of trust. However, the Interpersonal Trust

Scale covers not only trust of ‘people in general’ but also of

parents, teachers, physicians, politicians, classmates, and

friends. As such the Interpersonal Trust Scale cannot be

considered a valid measure of generalized trust, in the sense

of a default position of trust in unknown others, because the

trustees denoted in the questions refer predominantly to

individuals who would be well known to the respondent. The

measures of cognitive ability used in these studies are also

problematic in a number of ways. Garske (1975) uses a self-

report measure of abstract thinking taken from Cattell's 16PF

instrument (Cattell, Eber & Tatsouka, 1970), while Rotter's

use of scholastic aptitude tests on a sample of college students

is clearly some way from a measure of intelligence in the

general population. Similarly, spotting adjectives on a

tachistoscope and greater construct differentiation must be

considered somewhat tangential to conventional notions of

intelligence. As Yamagishi puts it, “the findings that have

been used as evidence supporting the popular image of high

trusters as naïve, credulous, and gullible individuals are thus

mostly indirect at best and often irrelevant to the claim”

(Yamagishi, 2001, p.123).

Cross-national sample surveys conducted over the past

thirty years or so provide robust, though indirect, evidence in

support of the Intelligence–Trust hypothesis. For perhaps the

strongest and most consistent predictor of trust in such

studies has been found to be education, in the vast majority of

contexts in which it has been assessed (Glaeser, Laibson,

Scheinkman, & Soutter, 1999; Helliwell & Putnam, 1999;

Inglehart, 1999). An obvious limitation of formal qualifica-

tions as a measure of cognitive ability, though, is that

attainment of educational qualifications is strongly condi-

tioned by the socio-economic circumstances of the household

into which an individual is born (Breen & Goldthorpe, 1997).

Because socio-economic status is itself strongly related to

trust, using education as a proxy for intelligence introduces a

serious confound which cannot be easily resolved. That being

said, however, the robust education effect provides, we would

argue, a degree of prima facie evidence in support for the I–T

hypothesis.

More direct evidence linking intelligence with generalized

trust can be found in a series of studies by Toshio Yamagishi

and his colleagues. In an initial investigation, Kosugi and

Yamagishi (1998) find high trusters, as measured by a six-

item generalized trust scale, to be more sensitive than low

trusters to negative contextual cues, when rating the

trustworthiness of fictional characters in stylised vignettes.

However, when no contextual information relating to

trustworthiness was provided to participants, high trusters

rated the target person as more trustworthy than did low

trusters. Their results support the view that high trusters are

not naïve and credulous in the sense of extending trust

indiscriminately. Rather, they adopt a default position of trust

in the absence of information to suggest that a specific

individual or group may not be trustworthy. When contextual

cues indicate a lack of trustworthiness, high trusters are more

adept at detecting these signals and incorporating them into

their assessments of the trustworthiness of interaction

partners. A limitation of this study, however, is that there is

no criterion by which to assess the accuracy of the

participants' evaluations of trustworthiness. High trusters

are shown to be more sensitive to the contextual information

provided but, due to the static nature of the vignettes, we

have no way of knowing whether this greater sensitivity

results in more accurate assessments of trustworthiness. As a

result, we cannot discount the possibility that the results

might simply reflect a greater proclivity amongst high

trusters to comply with (their interpretation of) the exper-

imental hypothesis (Orne, 1962).

In a subsequent investigation, this limitation is overcome

by incorporating an objective criterion of the accuracy of the

ratings of trustworthiness, via a series of experimental trust

games (Yamagishi, 2001; Yamagishi, Kikuchi & Kosugi, 1999;

Yamagishi & Kosugi, 1999). Prior to taking part in the

experiment, participants engaged in 30min small-group

discussions on the topic of garbage collection, to enable

them to familiarise themselves with their potential interac-

tion partners. The subsequent game involved 2 players

choosing either to contribute (cooperate) or take money

from (defect) their interaction partner. The rewards were set

such that DNCNS, where D is the reward for defecting while

the partner cooperates, C is the reward when both partners

cooperate, and S is the reward for cooperating when the

partner defects (see Yamagishi, 2001 for a detailed descrip-

tion of the design). It is, thus, a single-shot, anonymous trust

game in which mutual trust is Pareto-optimal but where each

player has a private incentive to defect.

The games were anonymous in the sense that each player

did not know exactly who their interaction partner was.

However, they did know that their partner was drawn from

the group of people with whom they had earlier discussed

garbage collection. After making their own decision about

whether to defect or cooperate, participants were informed of

the identity of their interaction partner. They were then asked

to predict whether their partner had defected or cooperated.

The results showed that high trusters were significantly

more accurate in their predictions of both defection and

cooperation than were low trusters. The same effect was

47

P. Sturgis et al. / Intelligence 38 (2010) 45–54

replicated in a number of subsequent studies, which varied

the degree of pre-existing familiarity of the interaction

partners (Yamagishi, 2001). In sum, then, these investigations

support the idea that trusters are both more sensitive to

contextual cues denoting the trustworthiness of others and

more accurate in their assessments of the likelihood of

cooperation and defection in cooperative situations.

Based as they are, however, on small, self-selecting

samples of Japanese university students, the results of these

investigations are rather limited in their generality. Addi-

tionally, none of these investigations employs an exogenous

measure of intelligence; the greater acumen of the high trust

participants is inferred from their behaviour in the experi-

mental games, rather than assessed via an independent

intelligence instrument. Our aim in the empirical part of this

paper is to complement the experimental work of Yamagishi

and his colleagues by testing the validity of the Intelligence–

Trust hypothesis using large-scale, representative survey data

collected in a different socio-historical context, and using an

independent measure of intelligence.

2. Data

Our data comes from two birth cohort studies; the 1958

National Child Development Study (NCDS) and the 1970

British Cohort Study (BCS70). The eligible samples in the

NCDS and BCS70 are every child born in Britain during a

particular week in 1958 and 1970 respectively. A total of

17,415 children born in the single week in March 1958 were

included in NCDS. Subsequent waves of the NCDS were

conducted in 1965, 1969, 1974, 1981, 1991 and 1999. In

BCS70, a total number of 17,196 children born in a single

week in April 1970 were included (Chamberlain, Philipp,

Howlett & Claireaux, 1975). Subsequent waves were con-

ducted in 1975, 1980, 1986, 1996, 2000 and 2004. The

analyses in this article make use of data from the four sweeps

of the surveys when the cohort members were 0,2 10, 16 and

46 of age in NCDS and 0, 5,11 and 34 years of age in BCS70.

The data were collected from the cohort member, his or her

parents, and teachers using interviews, tests and question-

naires. More information on the NCDS and BCS70 can be

found at http://www.cls.ioe.ac.uk/.

Over the study period sample attrition occurred: of the

initial sample of 17,196 participants in BCS70 in 1970, 76%

participated in the assessment at age 5 in 1975, 83% in the

assessment at age 10 in 1980, and 54% in the assessment at

age 34 in 2004. Of the initial sample of 17,415 participants in

NCDS in 1958, 84% participated in the assessment at age 11 in

1969, 79% in the assessment at age 16 in 1974 and 52% in the

assessment at age 46 in 2004. This yields 4686 cases with

complete data for the BCS70 and 6180 for the NCDS. Attrition

analysis indicates that nonresponse cannot be considered

Missing at Random (Rubin, 1987) in either study; dropping

out was more likely amongst men and those from a working

class background. At school age, a lower intelligence score,

more hostile and withdrawn behaviour, a tendency to tell lies

and not being liked by other children were all related to drop-

out.

This non-random attrition means that a complete case

analysis is not appropriate. To account for differential drop-

out from the study over time, we used the multiple

imputation procedure implemented in SPSS 17, using the

fully conditioned MCMC estimator to construct 5 complete

data sets. Our results present the pooled estimates across the

5 imputed data sets according to Rubin's procedure (Little &

Rubin, 2002; Rubin, 1987). The pooling across multiple

imputed data sets provides a correction for nonresponse

bias but also incorporates an estimate of the uncertainty

arising from the imputation in the estimates of the standard

errors. Full details of the implemented imputation procedure

are available from the corresponding author upon request.

3. Measures

Our theoretical model proposes that individuals with

greater acumen in social situations develop higher levels of

trust over the life-course, as a result of their ability to

determine when their trust is likely or unlikely to be

betrayed. We are thus invoking the notion of ‘social

intelligence’, or as Thorndike (1920) described it ‘the ability

to act wisely in human relations’. Specifying our key causal

variable as ‘social intelligence’, however, raises difficulties

with respect to measurement, particularly in the context of

large-scale observational studies like the BCS70 and NCDS.

This is because research into the social dimension of

intelligence has been dogged by problems of conceptual

and empirical differentiation since its inception (Riggio,

Messamer & Throckmorton, 1991). While the proposition

that social intelligence is “just general intelligence applied

to social situations” (Wechsler, 1958, p75) is no longer

tenable (Schneider, Ackerman & Kanfer, 1996), there remain

substantial barriers to valid measurement of this elusive

construct. These relate primarily to the fact that social

intelligence test batteries generally rely on respondent and

peer self-reports, rather than on independent observations of

verifiably socially intelligent behaviour, or on scores on test

items with objectively right or wrong answers. Where more

conventional test item batteries have been used, social

intelligence has tended to be indistinguishable from general

intelligence (Shanley, Walker & Foley, 1971). The consequent

reliance on self-reports (and, less often, peer evaluations)

makes it difficult to be confident that what is being measured

by these types of instruments is actually social intelligence as

opposed to some aspect of personality (Kihlstrom & Cantor,

2000), or simply response style (Riggio et al., 1991).

For these reasons, we use measures of general intelligence

as proxies for social intelligence in our analyses here. This

strategy has the advantage that, although indirect, our key

causal variable is undoubtedly a measure of intelligence,

rather than personality. And, although a number of studies

have found social and academic intelligence to be empirically

distinct, others have found moderate to high correlations

between general intelligence and specific dimensions of

social intelligence. For instance, Jones and Day (1997) report

a correlation of 0.79 between ‘crystallized social knowledge’

and academic problem solving, while Riggio et al. (1991) find

a correlation of 0.52 between academic intelligence and a

2 At age 0, information relates to the parents and household character-

istics of the cohort member at birth.

48

P. Sturgis et al. / Intelligence 38 (2010) 45–54

measure of the ability to ‘understand the meaning of verbal

and behavioural cues in different contexts’.3 Our measure of

general intelligence was obtained at age 10 in 1980 in the

BCS70 and at age 11 in 1969 in the NCDS. The tests were

selected to measure cognitive ability, broadly conceived, and

are similar in content across the two studies. In the BCS70, the

British Ability Scales (BAS) (Elliott, Murray & Pearson, 1978)

which comprises two verbal and two non-verbal scales

alongside the Friendly Maths Test were used. BAS verbal

scales included word definitions (37 items) and word

similarities (42 items). BAS non-verbal scales included recall

of digits (34 items) and matrices (28 items). Scales were

administered at the cohort member's school by their teacher.

The subscales showed good internal consistency (Cronbach's

alpha=0.90 for word definitions, 0.86 for word similarities,

0.91 for recall of digits and 0.75 for matrices). The Friendly

Maths Test was designed specifically for the BCS70 study. It

has 72 multiple choice questions on arithmetic, number skills,

fractions, measures in a variety of forms, algebra geometry

and statistics. Cronbach's alpha was 0.94 indicating very high

internal consistency.

A z-score from a principal component analysis4 was used

in our analysis to avoid problems of multicollinearity

between the different subscales, which are very highly

correlated with each other. In the NCDS, intelligence was

measured using the General Ability Test (Douglas, 1964) with

verbal and non-verbal subscales, the Reading Comprehension

Test and Mathematics Test both constructed specifically for

use in this study by the National Foundation for Educational

Research in England and Wales. Again, a z-score of the

principle component of these subscales was used due to

excessive collinearity.5

Our dependent variable in the analyses that follow is a

generalized trust question measured in 2004 at age 34 in the

Birth Cohort Study and at age 46 in the National Child

Development Study. Cohort members were administered a

4-point question which asked “How much do you trust people

in your local area?” (1=not at all, 2=not very much, 3=a

fair amount, 4=a lot).6 This is somewhat more restrictive

than the standard generalized trust question, in that it places

an imprecise, local geographical limit on the objects of trust.

Additionally, there are 4 response alternatives rather than the

more usual 2 in the standard question. Despite these

differences, the question nonetheless elicits ratings of trust

in unspecified others and, therefore, taps the same generalized

trust construct which motivates our concerns here. Given

concerns regarding the confounding of trust and caution in the

standard version of the Generalized Trust Question (Miller &

Mitamura, 2003), this version of the question may well

represent an improvement.7

4. Covariates

Scores on intelligence batteries such as those in the cohort

studies are likely to be correlated, in some cases strongly

correlated, with many of the variables that are also believed

to influence trust. For instance, even at a very young age,

scores on standard intelligence batteries are strongly condi-

tioned by the socio-economic status of individual test-takers

(Buck, Gregg, Stavraky & Subrahmaniam, 1973). Socio-

economic disadvantage has also been shown to be a cause

of (dis)trust in adulthood (Brehm & Rahn, 1997; Li, Pickles &

Savage, 2005). Thus, if we are to be confident that any

observed relationship between our measures of intelligence

and trust is not spurious, it is important to control for such

potentially confounding variables in our models.

With some minor differences, we were able to include the

same set of control variables in the NCDS and BCS70.

Covariates were selected to represent of the primary causes

of trust that have been identified in the existing literature and

that might also be correlated with intelligence. These are:

social class position in childhood and adulthood (Delhey &

Newton, 2003, 2005; Whiteley, 1999), health status (Kawachi

et al., 1999), educational attainment (Helliwell & Putnam,

1999; Sturgis, Patulny & Allum, 2007), extent of television

watching, associational membership, and volunteering in

childhood and adulthood (Putnam, 1996; Stolle & Hooghe,

2004; Uslaner, 2002), and degree of civic and political

engagement as an adult (Paxton, 2007; Putnam, 1993, 2000).

In addition, we include a measure of life satisfaction in

adulthood, because this is also likely to be influenced by

intelligence and is argued by influential social trust theorists

to be the key driver of generalized trust (Uslaner, 2002). We

also include two indicators of neighbourhood characteristics:

whether the cohort member has been a victim of a) theft or b)

violence in the neighbourhood. We include these neighbour-

hood indicators because neighbourhoods are postulated to be

key to understanding individual level social capital and,

therefore, trust (Putnam, 2000, 2007) and also because

selection into neighbourhoods could plausibly be directly

and indirectly influenced by individual intelligence.

5. Analysis and results

We begin by considering the univariate marginal distribu-

tions of the generalized trust question in each cohort. Fig. 1

presents these as box plots.

We can see from Fig. 1 that levels of trust are quite

different across the 2 cohorts. Those born in 1958 are

substantially more trusting that those born in 1970, with

83% of the National Child Development Study reporting that

they trust people in their local area either ‘a lot’, or ‘a fair

amount’, compared with 66% in the British Cohort Study. As

we have just 2 cohorts measured at a single point in time, it is

not possible to determine whether this difference is a life-

3 As both these studies were based on student samples, it is likely that

they under-estimate the strength of the relationship in the general

population because the full ability distribution is censored at the top end

but also because correlations between cognitive ability dimensions have

been shown to be lower in high ability groups (Detterman & Daniel, 1989).

4 The loadings were 0.84 for the Friendly Maths Test, 0.81 for word

definitions, 0.79 for word similarities, 0.55 for recall of digits, and 0.74 for

matrices to the principal component.

5 The loadings of the tests to the principal component were 0.90 for the

maths test, 0.85 for the reading test, 0.92 for the verbal scale, and 0.87 to the

non-verbal scale in General Ability Test.

6 The original coding in the data sets runs in descending order. We re-

coded these items so that higher scores indicate more trust.

7 The 2008 waves of the National Child Development Study and British

Cohort Study included the standard generalized trust question, so it will be

possible to replicate our analyses once this data becomes available.

49

P. Sturgis et al. / Intelligence 38 (2010) 45–54

cycle or generational effect. It will only be possible to address

this question (and even then only to a limited degree), when

both cohorts have responded to the same item in later sweeps

of the surveys. Whatever its cause, the wide variation in trust

across the 2 cohorts presents an interesting test-bed for the

Intelligence–Trust hypothesis, in the sense that there are

clearly rather different influences operating on trust across

the 2 cohorts at the same point in time. The large inter-cohort

difference also reinforces the point that intelligence will only

ever be a partial explanation of variability in generalized

trust. As we would not expect any substantial change in the

distribution of intelligence over this 12year period,8 the

causes of the wide variability across cohorts must clearly be

found elsewhere.

It is also worth noting that these estimates of trust are

substantially higher than those generally obtained from the

more conventional binary generalized trust question (GTQ).

Although there is a degree of variability over time and across

surveys, in Britain the binary GTQ generally elicits around 4 in

10 people choosing the ‘most people can be trusted’

alternative. The combined estimate of the proportion of

trusters across the 2 cohorts is more than double this figure.

This difference can partly be explained by the fact that both

cohort studies exclude older and younger people in the

general population, who tend to be less trusting. The higher

trust estimate may also derive from the reference to ‘people

in your local area’ rather than ‘most people’ in the standard

GTQ. To the extent that people are able to exercise choice over

the area in which they live, they may choose to settle in areas

in which they feel, whether correctly or not, that people are

more trustworthy than in the general population. The

difference could also be a function of the larger number of

response alternatives on offer in the cohort studies' version of

the question. By forcing respondents to choose between the

apparent extremes of trust and distrust, the binary GTQ may

push some potential trusters into the precautious position of

‘not being too careful’. Such individuals are able to select the

more nuanced alternative of ‘a fair amount’ in the cohort

studies question, resulting in higher estimates of trust.

Next we estimate ordinal logit models to predict trust in

adulthood as a function of intelligence at age 10 (BCS70) and

11 ((NCDS). We use the logit link function because the

ordinal nature of the trust measure and its skewed empirical

distribution make it unsuitable for Ordinary Least Squares

regression. The ordered logit model treats the observed 4-

point outcome variable, trust, as a set of ordered categories

imposed on an underlying continuum (Agresti, 1986;

McCullagh, 1980). Two models are estimated for each cohort.

Model 1a includes only intelligence as an independent

variable. In Model 1b, covariates from the child and adult

sweeps of the surveys are incorporated. Estimated coeffi-

cients for both models are presented for the BCS70 in Table 1.

The coefficients in Table 1 are logits which indicate the

change in the log of the odds that an individual is in a specific

category of the ordinal outcome, or one higher, for a unit

change in the covariate. Diagnostic statistics indicate no

problems with multicollinearity for these models, with

tolerances all above 0.2.

The bivariate effect of intelligence on trust, with a logit

coefficient of 0.33, is large and significant. Before any

covariates are included in the model, a one standard deviation

unit change in intelligence results in a 40% increase in the

odds of moving to a higher response category on the trust

variable. The overall explanatory power of this model is low,

however, with a pseudo R squared of just 0.03. Covariates are

added in Model 1b and generally conform to theoretical

expectations and previous empirical investigations; trust is

higher amongst women, those in better health, with higher

educational qualifications, more interest in politics, greater

life satisfaction, and those with more associational member-

ships (in both childhood and adulthood). Trust is lower

amongst the widowed and divorced, victims of property and

violent crime. Counter to Putnam's contention that watching

television reduces trust, our measure of the extent to which

the cohort member watched television at age 10 was non-

significant. Father's social class also had no direct effect on

adult trust, after conditioning on the adult characteristics of

cohort members.

Incorporating covariates in Model 1b serves to diminish

the magnitude of the intelligence coefficient by around two

thirds, with the logit dropping to 0.14, although it remains

significant at the 99% level of confidence. However, although

the direct effect of intelligence at age 10 on trust aged 34, is

substantially reduced in Model 2, it nonetheless exerts a

robust and not insubstantial direct effect on the cohort

members' expressed trust at age 34, net of a large number of

potentially confounding variables. The explanatory power of

Model 2a is also higher, though still comparatively weak, with

a pseudo R squared of 0.1.

Table 2 presents the same coefficients for the equivalent

models fitted to the NCDS data. In general, the pattern of

coefficients is the same as that found in the British Cohort

Study analysis, although with several notable differences. The

magnitude of the coefficient for intelligence is somewhat

larger, at 0.39 in Model 2a than in Model 1a. Parental social

class is a significant predictor of adult trust using the NCDS

data, while marital status is not. The difference with regard to

paternal social class might plausibly be explained by

generational changes in the British class structure during

this period (Blanden & Gregg, 2004) and, for marital status, by

the different ages at which trust was measured across the two

cohorts. There is no equivalent variable measuring atten-

dance of clubs during childhood in the NCDS but the nearest

Fig. 1. Univariate marginal distributions for trust in BSC70 and NCDS.

8 The linear increase in IQ scores during the 20th century known as the

‘Flynn effect’ amounts to around 3 points per decade (Flynn, 2007).

50

P. Sturgis et al. / Intelligence 38 (2010) 45–54

available equivalent, extent of volunteering, is not signifi-

cantly associated with trust in Model 2b. Introducing

covariates measured in childhood and adulthood in Model

2b reduces the magnitude of the intelligence coefficient by

around a half, although as with the BCS70 this is still

significantly different from zero at the 99.9% level of

confidence.

The coefficient for intelligence in Model 2 is of very similar

magnitude to the equivalent model for the BCS70, as is the

explanatory power of the model, with a pseudo R squared of

0.11. Again, despite the inclusion of a large number of

potential confounding variables measured at different points

across the life-course, the effect of intelligence when cohort

members were aged 11 on their expressed trust at age 46 is

also significant in the NCDS.

The pseudo R squared values in MODELs 1 and 2 imply

that the effect of intelligence, though reliably different from

zero, is rather weak in terms of overall explanatory power. It

is well known, however, that pseudo R squared statistics have

a number of important limitations (Veall & Zimmerman,

1996). A more illuminating way of considering the ‘impor-

tance’ of the effect of a predictor in an ordinal regression is to

plot the model predicted probability for each response

alternative, as a function of the predictor in question. For

the sake of parsimony and clarity of exposition, Fig. 2 takes

predicted probabilities from a model which combines the ‘a

fair amount’ and ‘a lot’ into a single ‘trusting’ category and the

‘not very much’ and ‘not at all’ into a single ‘not trusting’

category.9 This is done in Fig. 2 for the BCS70, using the

coefficients from Model 2a in Table 2. Even after conditioning

on the full range of covariates, it is clear that intelligence

measured in childhood exerts a considerable influence on

trust in middle-age. In assessing the strength of the influence

of intelligence on trust, it should be noted that, given the

potentially large amount of random error in our single

indicator of trust, our models are likely to under-estimate

the magnitude of the true relationship.

6. Discussion

Our aim in this paper has been to test — on nationally

representative survey data and in a different substantive

context — Yamagishi's social interactional model of the

9 A graph for the full 4 category model is available from the corresponding

author upon request.

Table 1

Multivariate ordinal regression on trust in British Cohort Study (n= 17,196).

Age variable measured

Variable

Model 1a estimate (SE)

Model 1b estimate (SE)

Thresholds for trust

34

Not at all vs not very much

−3.344 (0.066) ⁎⁎⁎

−2.947 (0.288) ⁎⁎⁎

Not very much vs a fair amount

−0.681 (0.021) ⁎⁎⁎

−0.179 (0.265)

A fair amount vs a lot

2.659 (0.050) ⁎⁎⁎

3.327 (0.293) ⁎⁎⁎

Intelligence

0.329 (0.023) ⁎⁎⁎

0.166 (0.028) ⁎⁎⁎

1

Male

−0.141 (0.037) ⁎⁎⁎

10

Social class of father (ref=unskilled)

Professional

0.107 (0.159)

Managerial/technical

0.160 (0.131)

Skilled non-manual

−0.004 (0.130)

Skilled manual

−0.010 (0.131)

Partly skilled

−0.015 (0.131)

10

How often watches TV (ref=often),

Never or hardly ever

−0.149 (0.252)

Sometimes

−0.010 (0.056)

10

How often attends clubs (ref=often)

Never or hardly ever

−0.110 (0.051) ⁎

Sometimes

−0.069 (0.051)

34

Marital status (ref=separated/divorced/widowed)

Married/cohabiting

0.341 (0.057) ⁎⁎⁎

Single

0.156 (0.071) ⁎

34

Highest qualification (ref=higher degree)

None

−0.436 (0.277)

CSEs2-5/other Scottish equals

−0.262 (0.111) ⁎

GCE A–C/good O levels/Scottish standards

−0.172 (0.092)

AS levels/1 A level 2+ A levels/Scottish higher/6th

−0.195 (0.098) ⁎

Degree/PGCE/other degree level equal

−0.108 (0.091)

34

Self-rated health

−0.113 (0.029) ⁎⁎

34

Interest in politics

−0.133 (0.030) ⁎⁎⁎

34

Political activity (yes)

−0.051 (0.045)

34

Victim of theft (yes)

−0.305 (0.063) ⁎⁎⁎

34

Victim of violence (yes)

−0.408 (0.125) ⁎⁎

34

Number of memberships

0.098 (0.021) ⁎⁎⁎

34

Satisfaction with life

0.176 (0.013) ⁎⁎⁎

Pseudo-R2 (Cox and Snell)

0.026

0.097

Estimates are pooled logit coefficients from 5 multiply imputed data sets.

⁎ pb0.05.

⁎⁎ pb0.01.

⁎⁎⁎ pb0.001.

51

P. Sturgis et al. / Intelligence 38 (2010) 45–54

development of generalized trust as a function of individual

differences in intelligence (Kosugi & Yamagishi, 1998;

Yamagishi, 2001; Yamagishi et al., 1999; Yamagishi & Kosugi,

1999). In this account, the proposed mechanism linking

intelligence to the development of ‘thin’, or generalized trust

is that socially astute individuals are better able to accurately

detect signs of (un)trustworthiness in social and economic

interactions. This means that, through the life-course, they do

not suffer the costs of betrayal so frequently, as they are less

inclined to place their trust in those who are unlikely to

honour it. By the same token, those with less acumen in social

interactions are frequently betrayed in trust relations. This

results in a vicious cycle of distrust, as frequent experience

of misplaced trust leads to a progressive withdrawal from

potentially gainful interactions in the future. Not only do the

less socially intelligent then progressively forgo the benefits

of norms of reciprocity, they also come to develop unfavour-

able evaluations of the trustworthiness of their fellow

citizens.

To test the Intelligence–Trust hypothesis we have drawn

upon unique British cohort data, which tracks the lives of

thousands of individuals over six decades. Our results show

that, after controlling for a number of potentially confounding

Fig. 2. Predicted probabilities of trust by cognitive ability at age 10 for the

BCS70 data.

Table 2

Multivariate ordinal regression on trust in National Child Development Study (n= 17,415).

Age variable measured

Variable

Model 2a estimate (SE)

Model 2b estimate (SE)

Thresholds for trust

46

Not at all vs not very much

−4.05 (0.095) ⁎⁎⁎

−1.87 (0.342) ⁎⁎⁎

Not very much vs a fair amount

−1.61 (0.043) ⁎⁎⁎

.626 (0.336)

A fair amount vs a lot

0.93 (0.028) ⁎⁎⁎

3.31 (0.337) ⁎⁎⁎

11

Intelligence

0.39 (0.027) ⁎⁎⁎

0.21 (0.036) ⁎⁎⁎

1

Male

−0.11 (0.041) ⁎

11

Social class of father (ref=unskilled)

Professional

0.324 (0.106) ⁎⁎

Managerial/technical

0.316(0.099) ⁎

Skilled non-manual

0.323 (0.114)

Skilled manual

0.122 (0.085)

Partly skilled

0.024 (0.100)

16

How often watches TV (ref=often)

Sometimes

0.008 (0.102)

Never or hardly ever

0.011 (0.099)

16

How often does voluntary work (ref=often)

Sometimes

−0.036 (0.073)

Never or hardly ever

0.009 (0.046)

46

Marital status (ref=separated/divorced/widowed)

Married/cohabiting

0.135 (0.068)

Single

−0.082 (0.072)

46

Highest qualification (ref=higher degree)

None

−0.596 (0.230) ⁎

CSEs2-5/other Scottish equals

−0.350 (0.130) ⁎

GCE A–C/good O levels/Scottish standards

−0.259 (0.115) ⁎

AS levels/1 A level/2+ A levels/Scottish Higher/6th

−0.419 (0.113) ⁎

Degree/PGCE/other degree level equal

0.035 (0.118)

46

Self-rated health

0.181 (0.02) ⁎⁎⁎

46

Interest in politics

0.170 (0.032) ⁎⁎⁎

46

Political activity (yes)

−0.051 (0.068)

46

Victim of theft (yes)

0.505 (0.063) ⁎⁎⁎

46

Victim of violence (yes)

0.302 (0.138) ⁎

46

Number of memberships

0.095 (0.019) ⁎⁎⁎

46

Satisfaction with life

0.218 (0.017) ⁎⁎⁎

Pseudo-R2 (Cox and Snell)

0.04

0.108

Estimates are pooled logit coefficients from 5 multiply imputed data sets.

⁎ pb0.05.

⁎⁎ pb0.01.

⁎⁎⁎ pb0.001.

52

P. Sturgis et al. / Intelligence 38 (2010) 45–54

variables, intelligence measured in childhood is a robust

predictor of generalized trust, when cohort members were 34

and 46 years of age, respectively. These results cannot, of

themselves, be taken as a direct test of the Intelligence–Trust

hypothesis, because these survey measures of trust do not

provide any leverage on the question of how accurate the

respondents' assessments of the trustworthiness of people in

their local areas actually are. It is possible, for instance, that

more intelligent people simply hold more benign views of

the benevolence of others, irrespective of their actual lived

experiences of having their trust honoured and betrayed over

time.

In interpreting the substantive implications of our find-

ings, however, it is crucial to consider them alongside the

experimental evidence of Yamagishi and colleagues, rather

than in isolation. That is to say, Yamagishi's experimental

games provide strong evidence that high trusters are more

accurate in their assessments of the trustworthiness of

interaction partners but their external validity is weak as a

result of their reliance on small, non-representative samples

and their lack of an exogenous measure of intelligence. Our

survey evidence, in contrast, does not speak directly to the

question of accuracy of trust assessments but does show a

substantial and robust relationship between an independent

measure of childhood intelligence and generalized trust in

adulthood on two large random samples drawn from the

general population of Great Britain. Taken in conjunction, we

contend that these contrasting lines of evidence provide

strong support for the Intelligence–Trust hypothesis.

We anticipate that a primary objection to this conclusion

will relate to the measures of intelligence we have used in our

analyses. For, although our hypothesised mechanism speci-

fies intelligence in social contexts as the driver of trust, we

use standard IQ type measures of latent cognitive ability in

our models. Although it would certainly be preferable to have

a direct measure of our key construct, we believe that our use

of general cognitive ability as a proxy for social intelligence is

justified. For, as we noted in the earlier discussion of our key

measures, the instruments that have been developed to

measure social intelligence to date, and in particular those

that would be suitable for administration in a survey, have

been based on behavioural self-reports and attitude-type

questions (Hendricks, Guilford & Hoepfner, 1969; Silvera,

Martinussen & Dahl, 2001). If a correlation between this type

of measure and trust were observed it would be open to the

charge that we are merely tapping into the same underlying

dimension of personality; people who trust are simply more

likely to report that they behave in certain ways in social

situations, with ‘intelligence’ playing no part. Indeed, a num-

ber of scholars have pointed out that factor analytic studies

which claim to demonstrate empirical differentiation of social

from academic intelligence might, in fact, only be detecting

differences in measurement protocols, with academic intel-

ligence measured using test items and social intelligence

measured with behavioural self-reports (Kihlstrom & Cantor,

2000; Riggio et al., 1991).

Nonetheless, our use of general intelligence as a proxy

for social intelligence suggests a number of potentially fruit-

ful directions for future research. First, it is now generally

accepted that, like academic intelligence, social intelligence

is a multi-dimensional construct (Jones & Day, 1997; Riggio

et al., 1991). Schneider et al. (1996), for instance, identify

three distinct domains: social insight; social memory; and

social knowledge, each of which might be differentially

influential in the development of generalized trust. Thus, it

will be important for future research to establish which

dimensions of social intelligence are important in fostering

interpersonal trust and which, if any, are not. Second, for

those who remain sceptical about IQ as a valid proxy for social

intelligence, we hope that our findings will act as a stimulus

to establish why the robust correlation between general

intelligence and trust that we have observed does, in fact,

come about. As for our key independent variable, there will

also be concerns about the measure of social trust available to

us in the cohort studies. Because there is only a single

indicator of trust in each data set, we must acknowledge the

likelihood that these measures are contaminated by both

random and systematic errors, with the consequent implica-

tions for downward bias in our estimates of causal effects

(Bollen, 1989). Future research could usefully strengthen the

robustness of our conclusions here, by making use of a

multiple-item measure of generalized trust.

While it is important to acknowledge the potential

limitations in our research design and analysis, none of

them, we feel, are sufficient, either collectively or in isolation,

to reject our primary substantive finding; that intelligence in

childhood has a robust, independent effect on trust in early

middle-age. Following Yamagishi, we interpret this effect as

emerging from the accrual of advantage over the life-course

to intelligent individuals who are able to carefully control the

conditions under which they sanction trust. This, in turn,

enables a resolution of the apparent paradox with which this

article began; counter to commonsense intuition, a general-

ized propensity to trust people we have never even met

should not be taken as a sign of gullibility but, on the contrary,

of being an astute social operator.

Acknowledgement

We gratefully acknowledge the support of the Economic

and Social Research Council (grant number: RES-163-25-0021).

References

Agresti, A. (1986). Introduction to categorical data analysis. New York: Wiley.

Alesina, A., & Ferrera, E. L. (2000). Participation in heterogeneous commu-

nities. Quarterly Journal of Economics, 115(3), 847–904.

Alesina, A., & Ferrera, E. L. (2002). Who trusts others? Journal of Public

Economics, 85(2), 207–234.

Barber, B. (1983). The logic and limits of trust. New Brunswick, N.J.: Rutgers

University Press.

Blanden, J., & Gregg, P. (2004). Family income and educational attainment: A

review of approaches and evidence for Britain.Oxford Review of Economic

Policy, 20, 245–263 Summer.

Bollen, K. (1989). Structural equations with latent variables. New York: Wiley.

Breen, R., & Goldthorpe, J. (1997). Explaining educational differentials: Towards

a formal rational action theory. Rationality and Society, 9(3), 275–305.

Brehm, J., & Rahn, W. (1997). Individual-level evidence for the causes and

consequences of social capital. American Journal of Political Science, 41(3),

999–1023.

Buck, C., Gregg, R., Stavraky, K., & Subrahmaniam, K. (1973). Variables

associated with social class differences in the intelligence of young

children. Multivariate Behavioral Research, 8(2), 213–226.

Cattell, R., Eber, H., & Tatsouka, M. (1970). Handbook for the 16PF. Champaign:

IPAT.

Chamberlain, G., Philipp, E., Howlett, B., & Claireaux, A. (1975). British births:

The first week of life. London: Heinemann.

53

P. Sturgis et al. / Intelligence 38 (2010) 45–54

Dekker, P. (2003). ‘Generalised social trust’: Meanings and political correlates.

Paper presented at the 19th IPSA World Congress, Durban, South Africa.

Delhey, J., & Newton, K. (2003). Who trusts? The origins of social trust in

seven societies. European Societies, 5(2), 93–137.

Delhey, J., & Newton, K. (2005). Predicting cross-national levels of social

trust: Global pattern or Nordic exceptionalism? European Sociological

Review, 21(4), 311–327.

Detterman, D. K., & Daniel, M. H. (1989). Correlations of mental tests with

each other and with cognitive variables are highest for low IQ groups.

Intelligence, 13, 349–359.

Douglas, J. (1964). The home and the school. London: MacGibbon and Kee.

Elliott, C. D., Murray, D. J., & Pearson, L. S. (1978). British ability scales.

London: National Foundation for Educational Research.

Fehr, E., & Gintis, H. (2007). Human motivation and social cooperation:

Experimental and analytical foundations.Annual Review of Sociology, 33,

43–64 August.

Flynn, J. R. (2007). What is intelligence?: Beyond the Flynn effect. Cambridge:

Cambridge University Press.

Fukuyama, F. (1995). Social capital and the global economy. Foreign Affairs,

74(5), 89–98.

Gambetta, D. (1988). Can we trust trust? In D. Gambetta (Ed.), Trust: Making

and breaking cooperative relations Oxford: Blackwell.

Garske, J. (1975). Interpersonal trust and construct complexity for positively

and negatively evaluated persons. Personality and Social Psychology

Bulletin, 1(4), 616–619.

Garske, J. (1976). Personaltiy and generalized expectancies for interpersonal

trust. Psychological Report, 39(2), 649–650.

Glaeser, E. L., Laibson, D., Scheinkman, J. A., & Soutter, C. L. (1999). What is

social capital? The determinants of trust and trustworthiness (working

paper): National Bureau of Economic Research.

Gurtman, M., & Lion, C. (1982). Interpersonal trust and perceptual vigilance

for trustworthiness descriptors. Journal of Research in Personality, 16(1),

108–117.

Halpern, D. (2001). Moral values, social trust and inequality: Can values

explain crime. British Journal of Criminology, 41(2), 236–252.

Hardin, R. (2001). Conceptions and explanations of trust. In K. Cook (Ed.),

Trust and society (pp. 3–39). New York: Russell Sage Foundation.

Hardin, R. (2006). Trust. Cambridge: Polity.

Helliwell, J. F., & Putnam, R. D. (1999). Education and social capital (no. 7121) :

National Bureau of Economic Research.

Hendricks, M., Guilford, J. P., & Hoepfner, R. (1969). Measuring creative social

intelligence.Reports from the Psychological Laboratory : University of

Southern California No. 42.

Inglehart, R. (1997). Modernisation and postmodernisation. Princeton:

Princeton University Press.

Inglehart, R. (1999). Trust, well-being and democracy. In M. E. Warren (Ed.),

Democracy and trust (pp. 88–120). Cambridge: Cambridge Uni Press.

Jones, K., & Day, J. (1997). Discrimination of two aspects of cognitive-social

intelligence from academic intelligence. journal of educational psychol-

ogy, 89(3), 486–497.

Kawachi, I. (1997). Long live community: Social capital as public health.

American Prospect, 35, 56–59.

Kawachi, I., Kennedy, B., & Glass, R. (1999). Social capital and self-rated health: A

contextual analysis. American Journal of Public Health, 8, 1187–1193.

Kihlstrom, J., & Cantor, N. (2000). Social intelligence. In R. J. Sternberg (Ed.),

Handbook of intelligence, 2nd edition Cambridge, UK: Cambridge University

Press.

Knack, S., & Keefer, P. (1997). Does social capital have an economic payoff? A

cross-country investigation. Quarterly Journal of Economics, 112(4),

1251–1288.

Kosugi, M., & Yamagishi, T. (1998). Generalized trust and judgements of

trustworthiness. Japanese Journal of Psychology, 69, 349–357.

La Porta, R., Lopez-de-Silanes, F., Shleifer, A., & Vishny, R. (1999). The quality

of government. Journal of Law, Economics and Organization, 1, 222–279.

Li, Y., Pickles, A., & Savage, M. (2005). Social capital and social trust in Britain.

European Sociological Review, 21(2), 109–123.

Little, R. J., & Rubin, D. B. (2002). Statistical analysis with missing data, Second

edition New York: John Wiley & Sons.

McCullagh, P. (1980). Regression models for ordinal data (with discussion),

series B. Journal of the Royal Statistical Society, 42, 109–142.

Miller, A., & Mitamura, T. (2003). Are surveys on trust trustworthy? Social

Psychology Quarterly, 66, 62–70.

Newton, K. (1999). Social and political trust in established democracies. In P.

Norris (Ed.), Critical citizens: Global support for democratic government

Oxford: Oxford University Press.

Nie, N., Junn, J., & Stehlik-Barry, K. (1996). Education and democratic

citizenship in America. Chicago: University of Chicago Press.

Orne, M. (1962). On the social psychology of the psychology experiment:

With particular reference to demand characteristics and their implica-

tions. American Psychologist, 17(11), 776–783.

Trust and reciprocity. Ostrom, E., & Walker, J. (Eds.). (2003). Interdisciplinary

lessons from experimental research.

Parales-Quenza, C. (2006). Astutenss, trust, and social intelligence. Journal

for the Theory of Social Behaviour, 36(1), 39–56.

Patterson, O. (1999). Liberty against the democratic state: On the historical

and contemporary sources of American distrust. In M. E. Warren (Ed.),

Democracy and trust (pp. 151–207). Cambridge: Cambridge University

Press.

Paxton, P. (2007). Not all association memberships increase trust: A model of

generalized trust in thirty-one countries. Social Forces, 86, 47–76.

Putnam, R. (1996). The strange disappearance of civic America. The American

Prospect, 24 Winter.

Putnam, R. D. (1993). Making democracy work: Civic traditions in modern Italy.

Princeton: Princeton University Press.

Putnam, R. D. (2000). Bowling alone: The collapse and revival of American

community. New York: Simon and Schuster.

Putnam, R. D. (2007). E Pluribus Unum: Diversity and community in the

twenty-first century. The 2006 Johan Skytte prize lecture. Scandinavian

Political Studies, 30(2), 137–174.

Riggio, R. E., Messamer, J., & Throckmorton, B. (1991). Social and academic

intelligence: Conceptually distinct but overlapping constructs. Person-

ality and Individual Differences, 12, 695–702.

Rothstein, B., & Uslaner, E. M. (2006). All for all: Equality and social trust.

World Politics, 58(3).

Rotter, J. (1967). A new scale for the measurement of interpersonal trust.

Journal of Personality, 35(4), 651–665.

Rotter, J. (1980). Interpersonal trust, trustworthiness, and gullibility. Amer-

ican Psychologist, 35(1), 1–7.

Rubin, D. B. (1987). Multiple imputation for nonresponse in surveys. New York:

J. Wiley & Sons.

Sampson, R. J., Raudenbush, Stephen, & Earls, F. (1997). Neighborhoods

and violent crime: A multilevel study of collective efficacy. Science, 277,

918–924.

Schneider, R., Ackerman, P., & Kanfer, R. (1996). To “act wisely in human

relations”: Exploring the dimensions of social competence. Personality

and Individual Differences, 21(4), 469–481.

Shanley, L. A., Walker, R. E., & Foley, J. M. (1971). Social intelligence: A

concept in search of data. Psychological Reports, 29, 1123–1132.

Silvera, D. H., Martinussen, M., & Dahl, T. I. (2001). The Tromso Social

Intelligence Scale, a self-report measure of social intelligence. Scandi-

navian Journal of Psychology, 42, 313–319.

Stolle, D., & Hooghe, M. (2004). The roots of social capital: Attitudinal and

network mechanisms in the relation between youth and adult indicators

of social capital. Acta Politica, 39, 422–441.

Sturgis, P., Patulny, R., & Allum, N. (2007). What makes trusters trust? Paper

presented at the ‘Reciprocity: Theories and Facts’ Conference, Verbania,

Italy February 22–25 2007.

Thorndike, L. (1920). Intelligence and its uses. Harper's Magazine, 140, 227–235.

Uslaner, E. (2002). The moral foundations of trust. Cambridge: Cambridge

University Press.

Veall, M., & Zimmerman, K. (1996). Psuedo-R2 measures for some common

limited dependent variable models. Journal of Economic Surveys, 10(3),

241–259.

Wechsler, D. (1958). The measurement and appraisal of adult intelligence, 4th

ed. Baltimore: Williams & Wilkins.

Whiteley, P. (1999). The origins of social capital. In Jan van Deth, Marco

Maraffi, Kenneth Newton, & P. Whiteley (Eds.), Social capital and

European democracy London: Routledge.

Yamagishi, T. (2001). Trust as a form of social intelligence. In K. Cook (Ed.),

Trust in society New York: Russel Sage Foundation.

Yamagishi, T., Kikuchi, M., & Kosugi, M. (1999). Trust, gullibility, and social

intelligence. Asian Journal of Social Psychology, 2(1), 145–161.

Yamagishi, T., & Kosugi, M. (1999). Cheater detection in social exchange.

Cognitive Studies, 6, 179–190.

54

P. Sturgis et al. / Intelligence 38 (2010) 45–54