Soc Indic Res (2011) 103:481–500

DOI 10.1007/s11205-010-9713-5

Identifying Social Trust in Cross-Country Analysis:

Do We Really Measure the Same?

Lars Torpe • Henrik Lolle

Accepted: 5 September 2010 / Published online: 18 September 2010

 Springer Science+Business Media B.V. 2010

Abstract Many see trust as an important social resource for the welfare of individuals as
well as nations. It is therefore important to be able to identify trust and explain its sources.
Cross-country survey analysis has been an important tool in this respect, and often one
single variable is used to identify social trust understood as trust in strangers, namely:
‘‘Generally speaking, would you say that most people can be trusted or that you can’t be
too careful in dealing with people?’’ The question, however, is whether this variable
captures the meaning of social trust equally well in all countries. This is investigated by
comparing different measurements of trust across five clusters of countries in all parts of
the world. The analysis shows that there are considerable problems associated with the use
of the variable of ‘‘most can be trusted’’ as an indicator of trust in strangers, both in terms
of strangers understood as ‘‘people you meet for the first time’’ and in terms of strangers
understood as people of a different nationality and religion. These results question the
validity of previous investigations of social trust based on international survey material.
The analysis furthermore reveals that a new survey question about trust in people one is
meeting for the first time is better suited as indicator of social trust in comparative analysis.

Keywords Social trust
Social capital
World values survey

Cross country survey analysis

1 Introduction

To-day it has been common to see trust as an important social resource for the welfare of
both individuals and nations. At the nation level trust is understood as generalized or social
trust, which is trust in those one does not know, i.e. ‘‘trust in strangers’’, and it is precisely
this form of trust that is assumed to have beneficiary effects on the nation as a whole, as it
stimulates cooperation between citizens in general, including citizens who are divided
socially and culturally.

L. Torpe (&)
H. Lolle

Aalborg University, Aalborg, Denmark
e-mail: [email protected]

123
482 L. Torpe, H. Lolle

Given that there is some truth in that, it is important to be able to identify social trust
and explain its sources. Cross-country survey analysis is an important tool in this respect.
Based on comparisons between countries included in World Values Survey several
attempts have been done to identify the explanatory variables for social trust (Alesina and
La Ferrara 2002; Delhey and Newton 2005; Anderson and Paskeviciute 2006; Bjørnskov
2006). In these studies one single variable has been used to measure trust, namely:
‘‘Generally speaking, would you say that most people can be trusted or that you can’t be
too careful in dealing with people?’’
This variable, also called the standard measure of social trust, is assumed to capture the
underlying meaning of ‘‘trust in strangers’’. It is this form of anonymous trust. Actually,
however, we know very little about what people are thinking when answering that ‘‘most
people can be trusted’’ or ‘‘you can’t be too careful’’. How do respondents interpret ‘‘most
people’’, and do people around the world interpret the question in the same manner? Is the
trust question really able to identify ‘‘social trusters’’, both in terms of trust in strangers,
including trust in those who are different from oneself? These questions can be summa-
rized in one main question, namely whether the underlying theoretical concept of ‘‘trust in
strangers’’ is captured equally well in all countries, as it has been assumed in several
comparative national studies of trust based on the World Value Survey (see for instance
Delhey and Newton 2005 and Bjørnskov 2006).
We start the analysis by discussing the concept of social trust and how it is measured as
we attempt to elaborate on the questions asked above. Next we investigate whether the
standard measure of trust captures generalized trust equally well in different parts of the
world. This analysis is carried out by comparing the answers to the ‘‘most people can be
trusted’’ question with the answers to a new general question about trust, namely whether
one can ‘‘trust people you meet for the first time’’. Then we investigate whether those who
say they trust other people also include people who are different from oneself, which in this
context means people with a different nationality or a different religion. Finally, we
compare and discuss the validity of the two measures of social trust.

2 The Concept of Social Trust and How it is Measured in Survey Research

2.1 The Concept of Social Trust

Among researchers, there are two main perspectives on the formation of trust. From one
perspective, trust is linked to a rational evaluation of the trustworthiness of others based on
concrete experience or information. One example of this perspective is found in the
writings of Russel Harding (2002, 2006), who sees trust as a relation between two persons,
where trust emerges if one person assumes that it is in the other person’s interest to live up
to one’s trust. Whether one displays trust or not will therefore depend on the assessment of
risk, both in relation to the concrete person and the concrete circumstances. Accordingly,
Hardin rejects any concept regarding generalized trust. Positive answers to the survey
question that ‘‘most people can be trusted’’ are, in his view, at best an expression of an
optimistic perspective regarding the reliability of others and thus an indication of a certain
risk-willingness (Harding 2006: 126). Conversely, Freitag and Tranmüller (2009) are of the
opinion that it is possible to develop a concept about generalized trust on the basis of a
rational perspective, namely as a type of trust formed on the basis of generalized expe-
riences from past meetings with others.

123
Identifying Social Trust in Cross-Country Analysis 483

An alternative to this rational and thus experienced-based perspective on trust is a

perspective according to which trust is grounded in the moral predisposition of the truster.
Eric Uslaner (2002, 2007) refers to this as ‘‘moralistic trust’’, which he understands as
‘‘a moral commandment to treat people as if they were trustworthy’’ (Uslaner 2002: 18).
Moralistic trust is based on the belief that most people share some fundamental values as to
how people should be treated. As such, moralistic trust forms the basis for generalized
trust. According to Uslaner, however, trust is also affected by one’s surroundings and to a
limited extent by life experiences (Uslaner 2002: 27–28).
Putnam (1993) originally also seemed to be in line with the rational perspective on trust,
where trust is based on an assessment of the reliability of others. Putnam writes that ‘‘trust
entails a prediction about the behaviour of an independent actor’’ based on what you know
about ‘‘his available options and their consequences, his ability and so forth’’ (Putnam
1993: 171). In smaller communities, this prediction can rest on ‘‘intimate familiarity’’. In
larger, more complex societies, an impersonal and indirect form of trust is required;
however, Putnam’s 1993 book does not clarify how such an impersonal form of trust is
established. In his next book Putnam defines generalized trust as ‘‘a standing decision to
give most people—even those whom one does not know from direct experience—the
benefit of the doubt’’ (Putnam 2000: 136). But how is it possible to trust people without
having any prior knowledge? The definition appears to be moving away from an exclu-
sively experienced based view and towards a moralistic perspective on trust.
A solution would therefore be to follow Rothstein (2003) and combine the two per-
spectives in a third perspective, where trust is based upon both experience and moral
predispositions (Rothstein 2003). Similar to Sztompka (1998), Rothstein defines trust as
‘‘hope about the future reliability of other actors’’ (Rothstein 2003: 111—own translation).
Such hope can be both rationally and morally grounded. Rationally, as it can be based on
an assessment about what is most reasonable in a given situation combined with general
knowledge about how actors will react; morally, because the hope is associated with the
perception that it is good and right to display trust to the other person. To use a concrete
example: When I choose to trust that a total stranger whom I ask for directions will not lure
me into a trap, it can be based on an objective risk assessment involving both my
knowledge of the concrete person and the place in question, but the moral belief that
I ought to display trust towards the ‘friendly’ stranger can also play in.
On that background, the question as to whether a person chooses to display trust towards
another person will depend partly on a relatively constant factor, which comprises the
person’s moral disposition to trust others, partly a variable factor based on an assessment of
the reliability of the other person in the specific situation. On the basis of the third
perspective, we can therefore define generalized trust as a standing inclination—on the
basis of experience together with moral convictions—to give the other person the benefit of
the doubt.

2.2 Measuring Social Trust

As a number of positive characteristics for society are attributed to generalized trust, social
scientists are obviously interested in mapping trust with respect to gaining knowledge
about what generates social trust. In order to map trust, however, we must be able to
measure it. There is a number of different methods for doing so, including experimental
research, anthropological studies and survey techniques. As regards measuring trust in all
of society, as is the case in comparative national studies, we are referred to the use of the
survey technique. In such cases, we are unable to map the inclination to trust strangers by

123
484 L. Torpe, H. Lolle

studying the actual actions; instead, we rely on how our selected persons respond to a
question about their inclination to trust strangers. In practice, the amount of social trust is
calculated as the proportion of population who may be called social trusters.
That which is decisive for the validity of this calculation is therefore how the ques-
tion(s) intended to capture social trust are formulated. The question that has actually been
used to capture social trust can be traced back to the ‘‘General Social Survey of the US’’
and was included in the 1981 World Value Study. The question is: ‘‘Generally speaking,
would you say that most people can be trusted or that you can’t be too careful in dealing
with people?’’ In a number of cross-country studies including countries all over the world,
this question has been used as an indicator of interpersonal trust (Inglehart 1997; Alesina
and La Ferrara 2002; Delhey and Newton 2005; Anderson and Paskeviciute 2006;
Bjørnskov 2006; Herreros and Criado 2008).
The formulation of the question has been criticized for being imprecise and asym-
metrical (Harding 2006; Nannestad 2008; Miller and Mitamura 2003), just as it has been
suggested that it is methodologically questionable to rely solely on a single question
(Reeskens and Hooghe 2008). On the other hand, it has been argued that the question
reasonably captures the underlying theoretical concept of ‘‘trust in strangers’’; that the
results have proven to be stable over time on the aggregate level; and that they are in
accordance with other measurements of trust (Uslaner 2002; Bjørnskov 2006).
We attempt to answer the main question we raise in this paper, whether the underlying
theoretical concept of ‘‘trust in strangers’’ is captured equally well in all countries by,
firstly, investigating the degree to which the standard question about trust appears to be a
precise measurement of trust in strangers across different types of societies. Secondly, to
investigate the degree to which trust is generalized as a form of trust which in principle
includes everyone, that is, persons from all groups in society.
However, we start by discussing what the respondents are thinking when answering that
‘‘most people can be trusted’’. On the one hand, the formulation of the question, ‘‘generally
speaking would you say that most people can be trusted?’’ is so broad that it is possible to
answer ‘‘yes’’ to the question, even though one is aware that there are situations in which
one does not display trust and persons that one does not trust. In that sense the question
aptly captures the inclination to trust another person given the uncertainty that will always
be present.
On the other hand, the question is formulated with such a lack of precision that it is
unclear what ‘‘most people’’ covers. Most of whom? Of all of the people in the world? Of
the residents of the nation in question? Of the people in the local area? It is conceivable
that some think of ‘‘most people’’ in more abstract terms than others, which may influence
the answer to the question. This can be part of the explanation as to why the trust question
correlates strongly with education. In other words, the phrasing of the question assumes a
common frame of reference, the existence of which we cannot be absolutely certain about
(see also Sturgis and Smith 2010).
A new general question about trust would appear to solve some of these problems. This
new question was included in the last round of the World Values Survey (2006 to 2008),
where respondents were asked whether they trust people they meet for the first time. The
possible answers are: ‘‘completely trust, trust somewhat, trust only little, do not trust at
all’’. Off-hand, the new question would appear better suited to capturing trust in strangers.
In contrast to the standard question about trust, this question asks directly about trust in
strangers, understood as persons one is meeting for the first time. Furthermore, the answer
categories are symmetrical. If one is generally inclined not to let others enjoy the benefit of
the doubt, one would answer ‘‘trust only little’’ or ‘‘do not trust at all’’. As such, the

123
Identifying Social Trust in Cross-Country Analysis 485

question meets some of the objections raised against the standard question and may
therefore be assumed to be a more valid measurement of social trust.
The first question we examine is whether the standard question about trust is able to
capture generalized trust understood as trust in strangers in every society and thus function
as a universal measurement of social trust. In order to do so, two criteria must be fulfilled:
First, there must be a reasonably strong correlation between the standard question and the
answers to the question about trusting those one meets for the first time, as it is claimed that
the standard question measures trust in strangers. Second, the correlations must be more or
less equally strong in all countries.
The other main question we will attempt to answer is whether the trust measured is
generalized in the sense that it includes persons belonging to different groups in society,
including those who are different from oneself in important areas, such as persons of a
different ethnic origin. If this is not the case, then the positive answers that ‘‘most people
can be trusted’’ are possibly not a valid expression of generalized trust. Obviously, a
generalized truster, as indicated above, is not necessarily a person who trusts everyone in
every conceivable situation. But it would not be consistent with the definition of gen-
eralized trust to exclude persons belonging to specific groups beforehand, e.g. persons with
a different cultural background, religion etc.
The most recent values study (2006–2008) raises two questions, which makes it possible
to control the validity of the answers to the standard question about trust, namely whether
the respondents ‘‘trust people of another nationality’’ and ‘‘trust people of another reli-
gion’’. We must therefore investigate whether the answer to ‘‘most people can be trusted’’
corresponds with trust in persons with a different nationality or religion. The same study is
carried out for the new general trust question, trust in people you meet for the first time, in
order to compare the degree of accordance between the two measures of trust. If the new
trust question corresponds better than the standard question about trust, it indicates that the
new trust question has a greater validity than the standard question.
Bringing things together, we illustrate the consequences for the ranking of the countries
on the trust barometer when exchanging the standard question about trust with the new
trust question as to whether one trusts those one meets for the first time. We furthermore
illustrate the consequences for the ranking if we exclude ‘‘particularized trusters’’, i.e.
those who do not trust persons of a different nationality or religion, from the pool of
‘‘social trusters’’ as measured by the social trust questions.

3 The Standard Measure of Social Trust as a Measure of Trust in Strangers

In this section we compare the two questions of social trust. As mentioned above it is
generally assumed that the standard question, ‘‘generally speaking, do you think most
people can be trusted?’’ measures trust in ‘‘strangers’’. In cross country analysis it is
furthermore assumed that positive answers to this question are valid expressions of ‘‘trust
in strangers’’ in all countries. To test this allegation we have grouped the 52 countries that
are included in the last wave of World Value Survey (www.worldvaluesurvey.org/) in five
country clusters: Western countries, former East Bloc, Latin America, Asia and Africa.
Initially, we look at the frequency of positive answers in the five clusters (Table 1) and
observe that the Western countries deviate from the other country clusters in two respects.
Firstly, there are considerably more ‘‘social trusters’’ in the case of both questions. Sec-
ondly, there are pretty much equal numbers for both questions, i.e. a good 40%. In the
meantime, another interesting observation can also be made. While the African countries in

123
486 L. Torpe, H. Lolle

Table 1 Social trust in five

Most people Trust people you meet Number of
country clusters (percent trusters)
can be trusted for the first time countries
(somewhat or completely)

West 40.4 41.2 15

Former East 20.9 16.3 10
Bloc
Latin America 12.9 17.8 8
Only countries which have
included both questions in the Asia 32.3 18.7 10
questionnaire are included in the Africa 15.4 27.8 9
analysis. N is in all clusters above Total 25.7 25.8 52
10,000 respondents

total are at the bottom of the scale as far as generalized social trust is concerned, when
ranked on the basis of the standard question, African countries occupy a clear second place
after the Western countries for the question about trusting those one meets for the first
time. The observation indicates that the correlation between the two questions is not as
good as one might be led to expect by the theory.
This is confirmed in Fig. 1, where the correlations are shown for each country in the five
country groups (for a detailed overview, see Appendix A). We observe that the correlations
coefficient for some countries is rather low, but we also notice a fairly coherent pattern for
each country-group together with rather strong variations between the groups. While the
correlations are strongest for the Western countries, where they vary between gamma
coefficients of 0.43 (Spain) and 0.72 (Switzerland), they are weakest for Africa, where they

N = 52 (countries)
Data source: World Values Survey, wave 2005-2008

Fig. 1 Gamma correlation coefficients between ‘‘trust most people’’ and ‘‘trust people you meet for the first
time’’, split by country and country cluster

123
Identifying Social Trust in Cross-Country Analysis 487

vary between 0.08 (South Africa) and 0.41 (Egypt). It thus appears as if it affects the
strength of the correlation between the two trust questions, where in the world you are.
This impression is strengthened by an analysis of variance on the aggregate country level.
Using the countries’ gamma values as the dependent variable and country cluster as the
independent variable, an explained variance of a good 60% is achieved. In other words, the
division of countries accounts for more than 60% of the variance for the correlations.
To give a more accurate picture of the correlations Table 2 shows the distribution of
both positive and negative responses to the standard question of trust. As appears, there is a
greater deviation in relation to what should be expected for the positive than for the
negative responses. Among those who answer that ‘‘you can’t be too careful when dealing
with people’’ only a minority say they trust people they meet for the first time (‘‘some-
what’’ or ‘‘completely’’). In contrast a majority of those who say they generally trust most
people say they do not trust those they meet for the first time. In Asian countries it is a
majority of 74%, in the former East Bloc a majority of 65, in Africa a majority of 60% and
in Latin America a majority of 56%. Only among the Western countries is it a minority,
namely of 33%.
Why is there such a considerable difference between the Western countries and the
other country clusters in relation to the correlation between trusting ‘‘most people’’ and
trusting people one ‘‘meets for the first time’’? For example, why is it in Asia only

Table 2 Cross-tabulation of ‘‘most people can be trusted’’ and ‘‘trust in people you meet for the first time’’
No trust Not very Somewhat Trust N
at all much trust completely

Trust in people you meet for the first time

West (Gamma = .70)
You can’t be too careful 28.9 46.8 23.5 .9 7,479
Most people can be trusted 4.5 28.3 63.4 3.9 5,537
Total 18.5 38.9 40.5 2.1 13,016
Former East Bloc (Gamma = .47):
You can’t be too careful 32.1 56.2 10.8 .8 6,245
Most people can be trusted 16.2 48.7 31.9 3.3 1,473
Total 29.1 54.8 15.5 1.2 7,718
Latin America (Gamma = .54):
You can’t be too careful 46.3 37.0 15.5 1.2 5,506
Most people can be trusted 18.3 37.7 38.0 6.1 909
Total 42.3 37.1 18.7 1.9 6,415
Asia (Gamma = .36)
You can’t be too careful 32.0 53.7 12.5 1.9 6,860
Most people can be trusted 15.5 58.4 22.4 3.7 3,474
Total 26.5 55.3 15.8 2.5 10,334
Africa (Gamma = .27)
You can’t be too careful 28.9 45.5 22.8 2.8 11,634
Most people can be trusted 18.7 41.4 32.9 6.9 2,156
Total 27.3 44.9 24.2 3.5 13,790
Row percent and Gamma correlation coefficients
Data source World Values Survey, wave 2005–2008. Weighted analysis

123
488 L. Torpe, H. Lolle

one-fourth of those, who say they trust ‘‘most people’’, who trust people they ‘‘meet for the
first time’’, while it is two-thirds in Europe? Is it possible to explain some of this difference
by pointing at some structural differences between Western countries and the rest of the
world? We shall examine this question in the following.
First of all it is conceivable that well-educated individuals are more inclined to think
about ‘‘most people’’ in abstract terms of ‘‘total strangers’’ than do persons with a more
limited education. Since the level of education is generally higher in Western countries
than in other parts of the world, it may help explain some of the difference between the
Western countries and the rest of the world. Secondly, it is conceivable that the same
applies for people in cities. While people in cities are more inclined to think of ‘‘most
people’’ in terms of total strangers, i.e. people you meet for the first time, people in the
country interpret ‘‘most people’’ as people ‘‘living around here’’. Since urbanization is
greater in western countries than in third world countries this may also help explain the
different strength of the correlations. Thirdly, it is conceivable that employed people are
more likely to take a broader and more abstract view on trust than unemployed people. It
therefore makes a difference on how ‘‘most can be trusted’’ is interpreted, whether one is at
the labour market, or one is a ‘‘homemaker’’. Since more women are outside the labour
market in third world countries than in western countries, such a difference will also be
associated with a difference between women and men. Appendix B presents an overview
of the variables that are included in the multilevel logistic regression analysis, presented in
Table 3.
Model 0, the so-called empty model (including no variables but only an intercept),
displays the logit variance on the national level, which is a measure of the variation between
the countries. Model 1 indicates the same as the cross-tabulated analyses in Table 1.

Table 3 Explaining trust in people you meet for the first time among respondents who answer ‘‘most can be
trusted’’ on the standard question measuring generalized social trust
Model 0 Model 1 Model 2 Model 3

Fixed factors
Constant -.28*** .53*** -.26** .49***
Education (1–9) .06*** .06***
Work or study .02 NS .02 NS
City, plus 100,000 inhabitants -.11* -.11*
Country cluster
West (ref.) – –
Former East Bloc -1.11*** -1.10***
Latin America -.88** -.80**
Asia -1.30*** -1.26***
Africa -1.08*** -.94***
Random factor/constant term variance .504 (.117) .258 (.080) .475 (.114) .251 (.063)
(standard error in brackets)

Note: Multilevel logistic regression. Regression logit coefficients and country variance for constant term.
N = 10,454
Data source World Values Survey, wave 2005–2008
Weighted analysis in MLwiN
NS not statistically significant
* p \ .05; ** p \ .01; *** p \ .001

123
Identifying Social Trust in Cross-Country Analysis 489

The assumptions as to what contributes to explaining the variations between the country
clusters are tested in models 2 and 3. As can be seen, they receive limited support from the
analysis results. As expected, there are positive and highly significant effects of education,
but there are practically no effects from being in work or study, and the effect of being a city
dweller is with an unexpected sign. More important here, however, is that these individual
factors account for very little of the variations between countries and therefore cannot
explain the effect of the country clusters. This is clearly seen in Model 3, which includes the
controlled effects of the variables at the individual level as well as the country cluster
dummies. The effects of these dummies change only marginally from Model 1 to Model 3.
In other words, these dummies are pretty much the sole source of explanation of the country
variation—roughly 50 pct.1
All together, we have no other explanation for the variations between the country
clusters other than the standard question about trust is perceived differently in different
countries, and that this varies with which part of the world one lives in. While in Western
countries most respondents understand the wording that ‘‘generally most can be trusted’’ as
‘‘trust in strangers’’, the wording has a different meaning for most people living in Asia, the
former East Bloc, Africa and Latin America. We are not able to explain difference, but
what we can conclude is that in some countries, the correlation is so low that the standard
question about trust is not able to measure generalized trust understood as trust in strangers.
This makes including such countries in comparative studies of social trust problematic.
Only within the group of Western countries do we find such strong correlations and
convergence between the two trust questions that it may justify cross-country studies.

4 The Inclusiveness and Exclusiveness of the Standard Measure of Social Trust

In addition to trust in strangers, generalized trust must in principle include everyone. If

persons with specific characteristics are excluded, then we are dealing with particularized
trust—not generalized trust. We can get a sense of the extent to which the standard
measure of social trust excludes representatives of certain groups by scrutinizing the
degree of correlation between the answers to the standard question and the questions as to
whether one trusts people of another nationality or religion.
Table 4 shows how ‘‘generalized social trusters’’ are distributed in relation to the
question about trust in persons with a different religion and nationality than the respondent.
The distribution of answers is roughly the same in the two subtables. We note that in the
Western countries most people who say that they trust most people also trust people of
another religion and nationality. It is quite another general situation in the Asian countries.
Here, only 42% of those indicating that they generally trust people also say that they trust
people with a different nationality or religion. We also observe smaller—though still
considerable—discrepancies in the former East Bloc, in Africa as well as to some degree in
Latin America.
Closer scrutiny of the figures reveals sizeable differences between the Western coun-
tries. In Sweden 94% of the generalized trusters also trust people of another religion and
nationality, while the corresponding figures in the former West Germany and the Nether-
lands are only 61 and 57 respectively. Also in Italy and Spain many who say they ‘‘trust
most’’ do not trust people of another religion/nationality. On the opposite side with a high
degree of convergence, together with Sweden, we find countries such as the USA,

1
(.504–.251)/.504 = .501.

123
490 L. Torpe, H. Lolle

Table 4 Trust people of another religion and nationality among ‘‘social trusters’’ in five country clusters
(row percent)
No trust Not very Somewhat Trust N (=100 pct.)
at all much trust completely

Trust people of another religion

West 2.8 15.6 70.8 10.8 6,691
Former East Bloc 12.2 36.2 46.7 4.9 2,071
Latin America 13.5 26.3 47.5 12.8 1,355
Asia 14.3 43.8 37.3 4.6 4,176
Africa 13.5 28.7 42.8 15.0 2,314
Trust people of another nationality
West 2.4 15.6 71.0 11.0 6,703
Former East Bloc 10.1 34.0 50.4 5.6 2,072
Latin America 17.4 27.7 45.0 9.8 1,355
Asia 17.9 48.7 29.0 4.3 4,117
Africa 19.9 34.0 37.0 8.1 2,306
Weighted analysis

Australia, France, Andorra, and the UK. In the other country clusters we also find countries
in which most of those claiming that they generally trust people say at the same time that
they do not trust people with a different nationality or religion, e.g. countries such as China
and Morocco. In Appendix C we have listed all countries in the data set on four different
measures of trust: (1) the standard variable of general trust in most people, (2) trust in
people one meets for the first time, (3) the standard variable adjusted for not trusting people
of another religion and/or nationality, (4) trust in people one meets for the first time
adjusted for not trusting people of another religion and/or nationality.
In other words, on the background of this relatively simple analysis, we are able to
conclude that there are persons in all countries who are classified as social or ‘‘generalized
trusters’’ but who would be more correctly classified as ‘‘particularized trusters’’. The
variation is great, however: While it is less than 10% for Sweden, it is more than 80% for
China.
The next question is whether the ‘‘new survey question’’ about trust in those one is
meeting for the first time is better able to distinguish ‘‘generalized trusters’’ from ‘‘par-
ticularized trusters’’. This seems to be the case as 75% of those responding that they trust
those they meet for the first time also trust people with a different religion or nationality,
while the corresponding figure for respondents answering that you can trust most people is
only roughly 55%. However, we cannot exclude that some of this difference is owing to
‘‘response set’’ as the questions regarding trust to those one is meeting for the first time and
trust in people with a different religion and a different nationality are lumped together in
the survey. Nevertheless, the difference is so great that this can hardly be the entire
explanation.
As illustrated in Table 5, the greater correspondence between the new trust question and
trust in people of a different religion and nationality applies to all of the country clusters
and furthermore to more or less the same degree. The result indicates that the new trust
question, ‘‘trust people you meet for the first time’’ could be a more valid measurement of

123
Identifying Social Trust in Cross-Country Analysis 491

Table 5 Trust people of another religion and nationality among respondents answering that they trust
people they meet for the first time (classified in five country clusters, row percent)
No trust Not very Somewhat Trust N (=100 pct.)
at all much trust completely

Trust people of different religion

West 1.4 6.7 78.8 13.1 6,925
Former East Bloc 3.6 15.9 71.6 8.9 1,659
Latin America 6.1 13.3 65.2 15.4 1,852
Asia 5.6 21.6 60.2 12.7 2,555
Africa 4.4 17.2 58.8 19.6 4,215
Trust people of different nationality
West 1.2 6.9 78.6 13.3 6,934
Former East Bloc 3.0 14.1 73.8 9.1 1,655
Latin America 9.4 15.5 62.7 12.4 1,847
Asia 9.5 25.4 54.6 10.6 2,520
Africa 8.7 23.5 55.1 12.7 4,197
Weighted analysis

generalized trust than the previously used measurement as to whether one ‘‘trusts most
people’’.

5 Discussion

The discrepancies presented above between the country clusters conceal some degree of
variation between the countries within the clusters. In the following discussion, we
attempt to cast light on some of these differences. We do so on the background of
Table 6, which reveals how the ranking in the top of the ‘‘trust meter’’ changes by going
from the one measure of trust to the other (for a ranking of all of the countries, see
Appendix C). The three-first columns show the score and rank for the standard question
for social trust (‘‘trust most’’) for the ten highest placed as well as the four countries that
enter the top-ten for the new measure of trust (‘‘trust first’’). The fourth column presents
the ranking for the traditional measure of trust after we have accounted for the ‘‘par-
ticularized trusters’’, i.e. excluded those indicating that they do not trust people of a
another nationality and religion. The fifth column shows the score for the new trust
question (‘‘trust first’’), and the sixth column shows the placing for the ‘‘trust first
measure’’. Finally, column 7 is a correction of the ranking for the ‘‘trust first measure’’
after having excluded ‘‘particularized trusters’’.
Table 6 reveals that four countries are pushed out of the top-ten by using ‘‘trust first’’
instead of ‘‘trust most’’: China, Vietnam, Indonesia and the Netherlands. These countries are
replaced by the UK, France, USA and Mali. It can be noted that China falls from number 5
to 52 on the corrected scale, i.e. last place on the list for the corrected measurement of ‘‘trust
first’’; conversely, Great Britain leaps from 31 all the way up to number 3.
More than the correction for ‘‘particularized truster’’, it is replacing the ‘‘trust-most
measure’’ with the ‘‘trust-first measure’’ which gives the change in the ranking although
for some countries the ranking is almost the same. This is the case for Norway, Sweden,
Finland and Switzerland. Nor do countries such as Australia and Canada have major

123
492 L. Torpe, H. Lolle

Table 6 Score, ranking and adjusted ranking

Trust most Rank of Adj. rank of Trust first Rank of Adj. rank
trust most trust most trust first of trust first

Norway 74 1 2 67 2 2
Sweden 68 2 1 69 1 1
Finland 59 3 3 52 3 6
Switzerland 54 4 4 52 4 4
China 52 5 41 11 49 52
Vietnam 52 6 25 22 25 34
Australia 46 7 5 49 7 7
Netherlands 45 8 9 20 32 28
Canada 43 9 6 51 5 5
Indonesia 43 10 15 20 34 31
Great Britain 31 16 8 49 6 3
France 19 30 16 45 8 8
USA 39 13 7 40 9 9
Mali 17 36 21 37 10 10

changes. The changes are primarily in the case of the Asian countries (China, Vietnam
and Indonesia) as well as the Netherlands, which scores considerably lower on the ‘‘trust-
first measure’’ than on the ‘‘trust-most measure’’; on the opposite end, we find Great
Britain and France, which had a somewhat higher score on ‘‘trust first’’ than ‘‘trust
most’’.
We also observe somewhat greater correlation between the ranking on ‘‘trust first’’ and
the corrected ranking on ‘‘trust first’’ than between ranking on ‘‘trust most’’ and the
corrected ranking for ‘‘trust most’’, which supports the perception that the ‘‘trust-first
measure’’ is a more valid measure for generalized trust than the standard measure on social
trust.
For China, applying the new measure for social trust entails a drastic change. It has
surprised many that China is categorized as ‘‘high trusting’’ country. An average cal-
culated on the basis of the first three waves of the World Value Study and based on the
standard question of social trust thus places China in fourth place after Norway, Sweden
and Finland (Rothstein and Stolle 2003:22). But as appears from this study, the vast
majority of the Chinese respondents saying that they generally trust most people do not
trust people with a different nationality or people with a different religion. Nor do they
trust people they are meeting for the first time. In the case of China it is obvious that the
standard question regarding trust does not measure generalized or social trust. But the
same would also appear to be the case for a number of other countries, particularly in
Asia and Africa.
As mentioned above, there are remarkably large internal differences in the Western
country cluster regarding the degree of correlation between the answers as to whether
one can trust people and whether one can trust those with a different nationality or
religion. It is conceivable that these differences can be accounted for to some extent with
reference to the differences in the degrees of ethnic and religious homogeneity. It can be
hypothesized that the correlation is greater in homogenous countries than in heteroge-
neous countries, as respondents with limited contact with people of a different ethnicity
or religion will primarily think of people of their own ethnicity and religion when

123
Identifying Social Trust in Cross-Country Analysis 493

confronted with the question as to whether they generally trust most people. More
respondents in homogenous countries than in heterogeneous countries will therefore be
inclined to say that they trust most people, even though they do not actually trust people
of a different nationality or religion. The diverse context can possibly contribute to
explaining the very high measurement of social trust in the Nordic countries, as these
countries have been and continue to be more homogenous then many of the other
Western societies.
However, it is not possible to confirm such a hypothesis. Sweden is actually the country
in which there is greatest correlation between the answer to the general trust measure and
the trust in people of a different nationality and religion. In Norway and Finland, the
correspondence is less, though not greater than that both countries remain at the top of the
corrected ranking (column 4).
At the same time, Sweden belongs to the countries renowned for strong or moderate
multicultural politics (Banting et al. 2009). The same is the case for Australia, Canada,
Great Britain and USA. These countries also share in common that they have a relatively
limited deviation between having general trust and trusting people of a different nationality
and religion. However, the pattern is not unambiguous. The Netherlands is also placed
within the group of countries with strong or moderate multicultural policies (Banting et al.
2009), but as already mentioned, the Netherlands is the Western country in which the lack
of correspondence is the greatest. It is tempting to attribute this discrepancy to the conflict
surrounding immigration and Islam in the Netherlands in recent years. However, the
pattern is not unambiguous here, either. France and Great Britain have undergone ethnic
and religious conflicts without corresponding effects being registered.

6 Conclusion

The comparative analysis of the standard question for trust seen in relation to the new
question about trust in those one meets for the first time and trust in persons of another
nationality and religion shows that the standard question about trust is not measuring the
same in different societies. It is not the same kind of trust that is measured in for
example Sweden and China via the standard question. But China is not the only country
where it is possible to raise questions regarding the validity of the standard measure for
trust. The same is the case for other countries in Asia and Africa, but partially also for
countries in the former East Bloc. In these country clusters, we find both a considerable
lack of correspondence between trust in most persons and trust in those one is meeting
for the first time and a considerable discrepancy between trust in most persons and trust
in people of a different nationality or religion. For many countries, it is thus not merely
the capacity of the standard question to measure social trust understood as trust in
strangers, which can be questioned but also its capacity to measure trust across cultural
divides.
We can therefore conclude that there are considerable problems associated with the use
of the measurement of ‘‘most can be trusted’’ as indicator of trust in strangers, both in
terms of strangers understood as ‘‘people you meet for the first time’’ and in terms of
strangers understood as people of a different nationality and religion. In particular one
should be careful with comparisons between countries belonging to different geographic
blocs and/or cultural settings. While it is possible to compare answers to the question of
‘‘most can be trusted’’ within the Western hemisphere, it is not equally possible to compare
between Western countries and countries in Asia and Africa. The results therefore also

123
494 L. Torpe, H. Lolle

question the validity of all previous studies of social trust based on the standard survey
question of ‘‘most can be trusted’’ and a random selection of countries included in World
Values Survey.
Finally, the analysis shows that that the new survey question about trust in people one is
meeting for the first time is better suited as indicator of social trust. It thus strengthens the
validity of the new question about social trust that the correlation between answers to this
question and answers to the questions about trust in persons belonging to different groups
of society is something greater and more uniform than is the case for the standard question.
We therefore recommend that this question is included in future survey-based comparative
research of social trust.

Appendix A

See Table 7.

Table 7 Correlation between

Country Cluster Country Gamma
‘‘trust most people’’ and’’ trust
people you meet for the first
time’’ West
1 Norway .72
2 Switzerland .71
3 Andorra .70
4 USA .70
5 Sweden .69
6 France .69
7 Italy .68
8 Netherlands .63
9 Canada .61
10 West Germany .59
11 Finland .58
12 Britain .57
13 Australia .57
14 Cyprus .54
15 Spain .43
Total N 15 15
Old east block
1 Slovenia .71
2 East Germany .67
3 Poland .61
4 Serbia .52
5 Georgia .47
6 Ukraine .47
7 Russia .45
8 Romania .44
9 Bulgaria .41
10 Moldova .39
Total N 10 10

123
Identifying Social Trust in Cross-Country Analysis 495

Table 7 continued
Country Cluster Country Gamma

Latin America
1 Argentina .63
2 Uruguay .60
3 Peru .58
4 Mexico .57
5 Colombia .46
6 Brazil .46
7 Trinidad and Tobago .34
8 Chile .19
Total N 8 8
Asia
1 Indonesia .55
2 Taiwan .46
3 Turkey .43
4 S Korea .33
5 Vietnam .32
6 Malaysia .31
7 India .28
8 Thailand .27
9 Jordan .26
10 China .23
Total N 10 10
Africa
1 Egypt .42
2 Morocco .37
3 Zambia .36
4 Mali .35
5 Ethiopia .19
6 Rwanda .17
7 Burkina Faso .17
8 S Africa .13
9 Ghana .12
Total N 9 9
Total N 52 52

Appendix B

See Table 8.

123
 Table 8 Variable list
496

Variables (labels from tables and Scale Comments and origin in WVS, rnd. four
text)

123
Trust variables
Most can be trusted Ordinal (0–1) Question formulation:
0: Need to be very careful Generally speaking, would you say that most people can be trusted
1: Most people can be trusted or that you need to be very careful in dealing with people?
Original variable: v23 (original values was 1 and 2).
Trust people you meet for the first Ordinal (1–4) Question formulation:
time 1: Do not trust at all I’d like to ask you how much you trust people from various groups.
2: Do not trust very much Could you tell me for each whether you trust people from this
3: Trust somewhat group completely, somewhat, not very much or not at all? …
4: Trust completely People you meet for the first time
Original variable: v128 (The coding is reversed from the original
variable)
Trust people you meet for the first Dummy Recoding of the above variable v128
time (dummy) 0: Do not trust (1 or 2 in the above variable)
1: Trust (3 or 4 in the above variable)
Trust people of another religion Ordinal (1–4) Question formulation:
1: Do not trust at all I’d like to ask you how much you trust people from various groups.
2: Do not trust very much Could you tell me for each whether you trust people from this
3: Trust somewhat group completely, somewhat, not very much or not at all? …
4: Trust completely People of another religion
Original variable: v129 (The coding is reversed from the original
variable)
Trust people of another nationality Ordinal (1–4) Question formulation:
1: Do not trust at all I’d like to ask you how much you trust people from various groups.
2: Do not trust very much Could you tell me for each whether you trust people from this
3: Trust somewhat group completely, somewhat, not very much or not at all? …
4: Trust completely People of another nationality
Original variable: v130 (The coding is reversed from the original
variable)
L. Torpe, H. Lolle
 Table 8 continued

Variables (labels from tables and Scale Comments and origin in WVS, rnd. four
text)

Independent variables in Table 3

Education Ordinal (1–9), treated as an interval scaled variable Original variable: v238
Work or study Dummy Recoded from v241: 1, 2, 3 and 6 ? 1 (other values equal to zero).
0: Not in work or studying
1. In work or studying
City, plus 100,000 inhabitants Dummy Several countries did not include the common variable for
0: Up to 100,000 inhabitants urbanization. In some of these there was instead included a
1: 100,000 or more inhabitants country specific variable. Our measure of urbanization is a
recoding from original variable v255 and several country specific
variables. In one of the countries, Japan, the value 1 indicate a
city of 150,000 inhabitants or more.
Country cluster 1: West Recoded from original variable v2A
Identifying Social Trust in Cross-Country Analysis

2: Former East Bloc

3: Latin America
4: Africa
497

123
498 L. Torpe, H. Lolle

Appendix C

See Table 9.

Table 9 Trust in most people (trust most) and trust in people one meets for the first time (trust first)
adjusted for not trusting people of another religion and/or nationality and a new ranking of the countries
Country Trust Trust Adj. trust Adj. trust Rank of Rank of Rank of Rank of
most first most first trust most trust first adj. trust adj. trust
most first

1 Norway .74 .67 .63 .60 1 2 2 2

2 Sweden .68 .69 .64 .67 2 1 1 1
3 Finland .59 .52 .46 .45 3 3 3 6
4 Switzerland .54 .52 .44 .48 4 4 4 4
5 China .52 .11 .06 .04 5 49 41 52
6 Vietnam .52 .22 .11 .12 6 25 25 34
7 Australia .46 .49 .39 .43 7 7 5 7
8 Netherlands .45 .20 .24 .17 8 32 9 28
9 Canada .43 .51 .38 .47 9 5 6 5
10 Indonesia .43 .20 .17 .15 10 34 15 31
11 West Germany .42 .31 .23 .25 11 14 10 14
12 Thailand .42 .17 .13 .09 12 36 18 41
13 USA .39 .40 .34 .38 13 9 7 9
14 East Germany .32 .22 .18 .19 14 26 13 19
15 Jordan .31 .24 .10 .15 15 22 29 30
16 Britain .31 .49 .28 .51 16 6 8 3
17 Italy .29 .19 .19 .18 17 35 11 24
18 Uruguay .28 .36 .19 .29 18 11 12 11
19 S Korea .28 .15 .11 .12 19 40 26 35
20 Ukraine .28 .20 .15 .18 20 31 17 25
21 Russia .26 .14 .12 .12 21 44 23 36
22 Ethiopia .24 .27 .09 .19 22 18 33 20
23 Taiwan .24 .21 .10 .14 23 27 27 32
24 India .23 .26 .09 .17 24 21 32 27
25 Bulgaria .22 .20 .12 .17 25 33 22 26
26 Romania .20 .13 .08 .11 26 48 36 37
27 Andorra .20 .26 .18 .25 27 20 14 15
28 Spain .20 .31 .13 .28 28 13 19 12
29 Poland .19 .23 .11 .19 29 24 24 22
30 France .19 .45 .16 .38 30 8 16 8
31 S Africa .19 .31 .09 .22 31 15 34 18
32 Egypt .19 .27 .06 .09 32 19 38 42
33 Georgia .18 .14 .10 .11 33 42 31 38
34 Slovenia .18 .10 .10 .09 34 50 28 46
35 Moldova .18 .14 .06 .08 35 45 40 48
36 Mali .17 .37 .12 .32 36 10 21 10
37 Argentina .17 .30 .13 .27 37 16 20 13

123
Identifying Social Trust in Cross-Country Analysis 499

Table 9 continued

Country Trust Trust Adj. trust Adj. trust Rank of Rank of Rank of Rank of
most first most first trust most trust first adj. trust adj. trust
most first

38 Mexico .16 .16 .06 .10 38 39 37 40

39 Serbia .15 .21 .10 .19 39 28 30 21
40 Burkina Faso .15 .29 .08 .24 40 17 35 17
41 Colombia .14 .14 .06 .08 41 47 39 49
42 Morocco .13 .21 .02 .12 42 30 52 33
43 Chile .13 .14 .05 .09 43 46 42 43
44 Zambia .12 .16 .05 .09 44 38 44 44
45 Cyprus .10 .09 .05 .07 45 51 43 50
46 Brazil .09 .17 .03 .11 46 37 46 39
47 Malaysia .09 .14 .03 .08 47 43 47 47
48 Ghana .09 .23 .04 .17 48 23 45 29
49 Peru .06 .09 .02 .05 49 52 49 51
50 Rwanda .05 .35 .02 .25 50 12 50 16
51 Turkey .05 .15 .02 .09 51 41 51 45
52 Trinidad and Tobago .04 .21 .03 .18 52 29 48 23

References

Alesina, A., & La Ferrara, E. (2002). Who trust others? Journal of Public Economics, 85, 207–234.
Anderson, C. J., & Paskeviciute, A. (2006). How ethnic and linguistic heterogeneity influence theprospect
for civil society: A comparative study of citizenship behavior. The Journal of Politics, 68, 783–802.
Banting, K., Johnston, R., Kymlicka, W., & Soroka, S. (2009). Do multiculturalism policies erode the
welfare state? An empirical analysis. In K. Banting & W. Kymlicka (Eds.), Multiculturalism and the
welfare state: Recognition and redistribution in contemporaty democracies (pp. 49–91). Oxford:
Oxford University Press.
Bjørnskov, C. (2006). Determinant of generalized trust: A cross-country comparison. Public Choice, 130,
1–21.
Delhey, J., & Newton, K. (2005). Predicting cross-national level of social trust: Global pattern or nordic
exceptionalism? European Sociological Review, 21(3), 311–327.
Freitag, M., & Tranmüller, R. (2009). Spheres of trust: An empirical analysis of the foundation of partic-
ularized and generalized trust. European Journal of Political Research, 48(6), 782–803.
Harding, R. (2002). Trust and trustworthiness. New York: Russel Sage Foundation.
Harding, R. (2006). Trust. Cambridge: Polity Press.
Herreros, F., & Criado, H. (2008). The state and the devopment of social trust. International Political
Science Review, 29(1), 53–71.
Inglehart, R. (1997). Modernization and Postmodernization. Cultural, economic and political change in 43
societies. Prindeton: Princeton University Press.
Miller, A. S., & Mitamura, T. (2003). Are surveys on trust trustworthy? Social Psychology Quarterly, 66(1),
62–70.
Nannestad, P. (2008). What have we learned about generalized trust, if anything? The Annual Review of
Political Science, 11, 413–436.
Putnam, R. D. (1993). Making democracy work. Princeton: Princeton University Press.
Putnam, R. D. (2000). Bowling alone. The collaps and revival of American community. New York: Simon &
Schuster.
Reeskens, T., & Hooghe, M. (2008). Cross-cultural measurement equivalence of generalized trust. Evidence
from European Social Survey (2002 and 2004). Social Indicator Research, 85, 515–532.

123
500 L. Torpe, H. Lolle

Rothstein, B. (2003). Sociala fällor och tillitens problem. Stockholm: SNS-förlag.

Rothstein, B., & Stolle, D. (2003). Social capital in Scandinavia. Scandinavian Political Studies, 26(1),
1–26.
Sturgis, P., & Smith, P. (2010). Assessing the validity of generalized trust questions: What kind of trust are
we measuring? International Journal of Public Opinion Research, 22(1), 74–92.
Sztompka, P. (1998). Trust, distrust and two paradoxes of democracy. European Journal of Social Theory, 1,
19–38.
Uslaner, E. M. (2002). The moral foundation of trust. New York: Cambridge University Press.

123