Inequality and Trust Revisited - umu.se · PDF fileInequality and Trust Revisited ... 2.1 Social equality, ... According to this reading, the bulk of ‘learning’

Working paper 11/2012

Department of Sociology 901 87 Umeå, Sweden Telephone: 090-786 50 00 www.umu.se

Inequality and Trust Revisited The Impact of Female Labour Force Participation on

Generalized Trust

Jan Mewes

1

Inequality and Trust Revisited:

The Impact of Female Labour Force Participation on Generalized Trust

Abstract

It is argued that divergence in female labour force participation is a contributory factor in explaining country-level differences in generalized trust. Based on intergroup contact theory and affect theory of social exchange, it is maintained that the workplace features an excellent opportunity structure for the establishment of ‘weak’, ‘bridging’ ties between dissimilar people. These ties, in turn, are considered to promote generalized trust. Using cross-national survey data from the first five rounds of the European Social Survey (2002-2011) and applying Bayesian multilevel analysis, it is shown that country-level female labour force participation rates positively impact generalized trust. Moreover, there is a significant and positive cross-level interaction between female labour force participation and gender, indicating that it is only women’s generalized trust that is positively correlated with levels of female labour force participation at the country level. The article concludes with a general discussion and an exploration of avenues for future research.

2

1. Introduction

Generalized trust, or trust in strangers (Stolle, 2002), has been shown to be an important

remedy for many problems in modern societies (for an overview see Nannestad, 2008). While

astonishingly little is known about the sources of generalized trust (but see Glanville and

Paxton, 2007), findings regarding its geography are remarkably consistent: it is people living

in Denmark, Norway and Sweden who are, regardless of the measurement that is used,

constantly found to be the most trusting in the world (e.g. Rothstein and Uslaner, 2005;

Bjørnskov, 2006; Van Oorschot et al., 2006). In this article, it will be argued that female

labour force participation contributes to an explanation of this ‘Nordic Exceptionalism’

(Delhey and Newton, 2005) in particular and of between-country variation in generalized trust

in general.

Theoretically, it will be argued that the workplace may be considered a social context with

especially rich opportunities for forging ‘weak ties’ (Granovetter, 1973) between individuals

with dissimilar backgrounds. These ‘cross-cutting networks’ (Mutz and Mondak, 2006), in

turn, are assumed to promote tolerance of ‘otherness’ (Pattie and Johnston, 2008; Weare et al.,

2009) and, eventually, generalized trust (Stolle, 2002). This thesis will be backed up by

insights from intergroup contact theory (Allport, 1954; Pettigrew, 1998) and affect theory of

social exchange (Lawler, 2001).

Empirically, the article will make use of cross-national survey data from the first five rounds

of the European Social Survey, thereby covering 15 European countries between 2002 and

2011. Methodologically, multilevel analysis will be applied, in which respondents will be

treated as nested in countries at different points in time. Due to the rather small sample size at

the country level, a Bayesian approach with Markov chain Monte Carlo (MCMC) simulation

will be taken to increase the reliability of the quantitative findings (Hox, 2010). The article

3

will conclude with a short discussion of the most significant results and an exploration of

avenues for future research.

2. Theoretical Background

2.1 Social equality, gender and ‘Nordic Exceptionalism’ in regard to generalized trust

Generalized trust has been described as one of the most important pillars of social capital:

countries with high levels of trust are generally found to be less corrupt, more efficient in

terms of cooperation, happier and democratically better functioning (Nannestad, 2008).

Whereas only few studies address the sources of generalized trust from a more theoretical

perspective (e.g. Stolle, 2002), the relevant empirical studies unanimously find that, in

comparison, it is people in the Scandinavian countries that are most trusting in other people

(e.g. Delhey and Newton, 2005; Bjørnskov, 2006; Van Oorschot et al., 2006; Pichler and

Wallace, 2007; Reeskens and Hooghe, 2008). The exceptional status of Scandinavia is often

attributed to its universal welfare state model (Rothstein and Uslaner, 2005). More precisely,

it is the interplay of income equality (Uslaner, 2002; Rothstein and Uslaner, 2005; Bjørnskov,

2006), class equality (Albrekt Larsen, 2007), active labour market policies (Lee, forthcoming)

and ‘good government’ (Stolle and Rothstein, 2008; Freitag and Bühlmann, 2009), that is

assumed to explain the unusually high levels of trust in the Nordic countries. In the following,

I will concentrate on the dimension of social equality.

According to Rothstein and Uslaner (2005: 42), “social trust is caused by two different, yet

interrelated types of equality, namely, economic equality and equality of opportunity”.

Empirically, however, the authors entirely concentrate on the effect of income equality. In the

following, it will be argued that there might be another very important dimension of social

4

inequality in connection with generalized trust, namely gender equality in the domain of

work.1

Whereas men’s employment behaviour differs only marginally across countries, female

labour force participation (in the following: FLFP) rates are subject to much higher between-

country variation (Steiber and Haas, 2012). Indeed, one striking characteristic of the highly-

trusting Scandinavian countries is their comparably large share of gainfully employed women

(e.g. Esping-Andersen, 1990; Korpi, 2000). Yet why should inequalities in FLFP contribute to

an explanation of country differences in generalized trust? To elaborate on this question, I

will consequently start with discussing the general relationship between gender and

generalized trust.

2.2 Gender and generalized trust

Up to now, the relationship between gender and generalized trust has been relatively unclear

from a scholarly perspective. While some cross-national studies suggest that it is men who are

more trusting than women (e.g. Van Oorschot et al., 2006; Lee, forthcoming), others find the

opposite (e.g. Halman and Luijkx, 2006; Kääriäinen and Lehtonen, 2006; Hooghe et al.,

2009). Some scholars even find no gender effect at all (e.g. Kaasa and Parts, 2008). Among

these studies, there seems to be a certain pattern: scholars who measure trust through the

standard single-item question (‘Generally speaking, would you say that most people can be

trusted?’) tend to find that it is men who are more trusting (e.g. Van Oorschot et al., 2006;

Lee, forthcoming), whereas studies using the three-item misanthropy scale (Reeskens and

Hooghe, 2008), which refers to the standard question and to the perception of strangers’

helpfulness, tend to establish the opposite gender effect (e.g. Halman and Luijkx, 2006;

Hooghe et al., 2009). Hence, studies in regard to the connection between gender and trust

need to be read with caution in the light of these methodological issues.

5

Additional insights come from a longitudinal study (Stolle and Hooghe, 2004) which

establishes no significant difference between male and female respondents in a repeatedly

measured sample of schoolchildren. Hence, gender gaps in trust are obviously not due to any

biological sex differences. This leads to the question as to whether and how experiences may

shape people’s trust in strangers. For an appropriate answer, it is first of all necessary to

address the concept of generalized trust from a more theoretical point of view.

2.3 Conceptualizing generalized trust

Where does generalized trust come from? Broadly speaking, scholars approach generalized

trust from two different theoretical angles. In a first perspective, trust is regarded as a value.

Uslaner (2002), for example, regards generalized trust as a personality trait closely related to

another stance, namely genuine optimism. According to this reading, the bulk of ‘learning’

trust in strangers should take place during people’s childhood and early adolescence (Uslaner,

2008). A second strand of literature treats generalized trust as an attitude (Stolle, 2002). From

this latter perspective, generalized trust would be far more susceptible to people’s experiences

in everyday life than according to the former. Whether generalized trust is a disposition, and

therewith stable, or whether it is (also) subject to experience is foremost an empirical

question. Probably the most comprehensive study concerning the relationship between

experiences and generalized trust was conducted by Glanville and Paxton (2007). Using

survey data from four different samples from the US and conducting a confirmatory tetrad

analysis to distinguish between causal and reflective indicators of generalized trust, they find

that experiences made in certain social contexts indeed matter for shaping trust in other

people (see also Offe, 1999; Hardin, 2002; Freitag and Traunmüller, 2009). Hence I will refer

to generalized trust as an intrapersonally variant attitude rather than a stable disposition in the

following.

6

2.4 From weak ties at work to generalized trust: a theoretical synthesis of intergroup contact theory and affect theory of social exchange

Which experiences in particular may be assumed to impact people’s trust in strangers? Stolle

(2002: 405) considers it unlikely that positive interactions with family members, friends or

any other people to whom individuals feel close should impact people’s generalized trust very

much. Rather, such interactions would be expected to promote particularized trust, that is,

trust in people they know well. Therefore, it seems fruitful to concentrate more on ‘weak ties’

(Granovetter, 1973), or ‘bridging social capital’ (Putnam, 2000), respectively. Against this

background, it is assumed that weak connections, rather than strong ties, should be more

likely to link together people with dissimilar backgrounds (Stolle, 2002: 406). In turn,

interaction with dissimilar people is considered to be especially effective for promoting

tolerance and trust with respect to other people in general (cf. Mutz and Mondak, 2006; Pattie

and Johnston, 2008; Weare et al., 2009). Unfortunately, the underlying mechanism for the

relationship between weak ties and generalized trust is theoretically rather unspecified in the

literature. Thus, I disentangle this connection by referring to insights from intergroup contact

theory (Allport, 1954; Pettigrew, 1998) and affect theory of social exchange (Lawler, 2001) in

the following.

Conceptually, trust in strangers is not very different from trust in an – as yet undefined –

outgroup: ‘Generalized trust reflects a bond that people share across a society and across

economic and ethnic groups, religions, and races’ (Rothstein and Uslaner, 2005: 45).

According to intergroup contact theory (Allport, 1954; Pettigrew, 1998), positive contact

with a member from any given outgroup may lead to a more positive re-evaluation of the

respective outgroup as a whole. Following Allport (1954), successful intergroup contact is

facilitated by meeting the following conditions: 1. equal group status; 2. common goals; 3. the

absence of competition; 4. support from authorities, norms or laws. Thus, intergroup contact

7

theory contributes to understanding why contact between members from different groups may

promote generalized trust among the involved actors. However, people’s contact with family

members and friends would be considered ingroup contact rather than outgroup contact.

Therefore, any kind of positive interaction between closely-tied individuals would be

expected to fortify particularized trust rather than generalized trust (Stolle, 2002).

If contact with familiar people is less likely to contribute in promoting generalized trust

(Freitag and Traunmüller, 2009: 790), then precisely which social domains should facilitate

the formation of weak connections between dissimilar people? In the following, I argue that it

is the domain of work that might be of great importance in this regard. More often than not,

relations at the workplace tend to be rather loose. Also, workplace relationships are likely to

fulfil at least three of the four conditions for successful intergroup contact mentioned above:

usually, there are common goals at work, one does not compete with co-workers, and working

norms help to structure communication between colleagues. This thesis is empirically

supported by findings of Glanville and Paxton (2007: 239), who establish a positive

correlation between trust in co-workers and generalized trust.

Why should workplace-induced relationships be more likely to promote generalized trust than

other ‘foci of activity’ (Feld, 1981) in which people also get to know other people? To answer

this question, it is useful to refer to affect theory of social exchange (Lawler, 2001), according

to which the formation of social ties, even weak ones, is dependent on people’s commitment

to the relationship at stake. At its core, affect theory of social exchange is in line with more

general theories of social exchange (e.g. Cook and Emerson, 1979): it proposes that repeated,

positive interaction enhances commitment to specific relationships (see also Scott Feld’s

focus theory, 1981). However, Lawler’s (2001) exchange theory goes one step further by

8

maintaining that it is the specific exchange structure which determines the likelihood and the

strength of commitment. Special attention is thereby given to the role of productive exchange,

which is made when individuals pool their resources to produce a joint good. It is argued that

it is productive exchange rather than reciprocal, negotiated or generalized social exchange

that would set free positive and diffuse emotions which are believed to underlie people’s

commitment to each other. According to Lawler, individuals fail to attribute the success of

productive social exchange to the tangible efforts of each of the involved individuals. Rather,

successful productive exchange would be attributed to the responsible social unit as a whole.

This is why productive exchange is particularly likely to trigger commitment to specific

relationships. Given that work is an excellent opportunity structure for productive exchange, I

argue that work should facilitate the establishment of durable yet weak relationships between

dissimilar people.2 In turn, these weak relationships are considered to promote generalized

trust more so than ties to people known well such as friends, family members and partners

(Freitag and Traunmüller, 2009: 790). Hence, countries in which comparatively large shares

of women are economically active should, considered in the aggregate, display higher levels

of generalized trust than on average.

In sum, it can be hypothesized that FLFP at the country-level positively contributes to

generalized trust (H1). Since a direct effect of FLFP on women’s generalized trust may be

assumed, the second hypothesis maintains that there is an interaction between FLFP at the

country-level and gender at the individual level (H2).

3. Data, Measurement and Methods

3.1 Data

To test the hypotheses, the article makes use of cross-national survey data from the first five

rounds of the European Social Survey (in the following: ESS), fielded between 2002 and

9

2011 in two-year intervals. The ESS is a large-scale, cross-national survey project designed to

minimise typical measurement errors in cross-cultural survey research. Respondents were

selected by means of strict probability samples of the population aged 15 and older living in

private households within European countries. The response rates range between 33.5 per cent

(Switzerland, round 1) and 75.7 per cent (Portugal, round 4).3 A balanced data sample is used,

that is, there are complete micro- and macro-level data for each of the selected countries at

each point in time. Overall, the sample consists of survey data from 141,696 respondents from

15 European countries in the period 2002 to 2011 (see Table 1 for more details).

***Table 1 about here***

3.2 Dependent variable

To measure generalized trust, the well-established standard question (“Generally speaking,

would you say that most people can be trusted, or that you can’t be too careful in dealing with

other people?”) introduced by Rosenberg (1956) is used. In contrast to the binary measure

used in the World Values Study, the ESS features a 11-point response scale, where 0 indicates

the lowest level and 10 the highest level of trust. While the simple but efficient single-item

trust measure has been contested more recently for methodological reasons (Reeskens and

Hooghe, 2008; Sturgis and Smith, 2010; Torpe and Lolle, 2011), it still seems to be an

appropriate and valid instrument (Uslaner, 2012). One major point of criticism concerns the

single-item’s lack of cross-cultural comparability. However, testing for the radius of the

standard trust question in a sample from the World Values Survey, Delhey et al. (2011)

recently showed a consistent pattern regarding respondents from European countries.4

10

3.3 Explanatory and control variables

In terms of explanatory and control variables, variables both on the country and on the

individual level will be introduced (for a descriptive overview see Table 2). The key macro

variable, female labour force participation (FLFP), is operationalized through the country-

time-specific measures for the employment rate of women in per cent of the female

population between 15 and 64, provided by the OECD (Employment and Labour Market

Statistics database, 2011). However, women’s employment rates are relatively highly

correlated (r=0.74) with Gross Domestic Product (GDP), a variable that has been shown to be

highly correlated with generalized trust at the country level (Knack and Keefer, 1997; Delhey

and Newton, 2005; Hooghe et al., 2009; Bjørnskov, 2012). Hence, GPD per capita will be

employed as a macro-level control variable. Careful statistical diagnostics will show whether

models containing FLFP rates as well as GDP per capita will be affected by multicollinearity

problems (O'Brien, 2007). All country-time data for GDP per capita come from the Penn

World Table 7.0 (method: constant prices, Laspeyres).

Key variables at the individual level are gender and employment status at the time of

interview. With respect to the latter, it will be distinguished between the following categories:

1. economically active (part- or full-time); 2. unemployed; 3. retired; 4. disabled or sick; 5.

homemakers; 6. in education; and 7. others. In line with others (e.g. Lee, forthcoming) and

following the arguments derived from my theoretical discussion, it is assumed that it is

economically active respondents who are most trusting.

In addition, a set of personal characteristics will be used as control variables at the individual

level. Given that unemployment is constantly found to be a negative determinant of trust (e.g.

Van Oorschot and Arts, 2005), it may be concluded that previous unemployment experiences

(measured with a dummy variable) may contribute to dampen people’s current trust in

11

strangers, too. Moreover, it was shown that social class determines generalized trust in such a

way that white-collar workers are found to be more trusting than those from less privileged

strata (Whiteley, 1999; Lee, forthcoming). Class is operationalized through a simplified

version of the Ganzeboom EGP class scheme (provided by Leiulfsrud et al., 2010), in which

the following four classes are distinguished: 1. service classes (EGP I and II, acting as the

reference category in the multivariate analysis); 2. routine non-manual workers (EGP III); 3.

small self-employed (EGP IV); 4. workers (EGP V, VI and VII); and 5. not applicable. People

who were not working at the time of interview were, if possible, classified according to their

previous occupation. Since previous research established that those born abroad are less

trusting than natives (e.g. Hooghe et al., 2009), a dummy variable for migratory background

is introduced. Respondents were classified as having a migratory background if at least one of

the following criteria applied: she or he has a foreign citizenship, was born abroad, or has at

least one parent who was born abroad.

Further, several studies show that generalized trust is more widespread among the highly

educated (e.g. Freitag, 2003; Hooghe et al., 2009), hence respondents’ years of education are

measured. Moreover, generalized trust tends to be higher among those who do materially well

(Stolle and Hooghe, 2004; Hooghe et al., 2009; Hamamura, 2012), thus people’s subjective

financial household situation is used. Household size is employed as a rough proxy for

respondents’ family situation, thereby taking into account that generalized trust tends to be

less prevalent among singles (e.g. Lee, forthcoming). Occasionally, people living in big cities

are found to be less trusting than those in smaller communities, hence community size is

monitored. Finally, since religion seems to play an important role in regard to shaping

people’s trust (e.g. Hooghe et al., 2009; Lee, forthcoming), religiosity is controlled for, with

higher values denoting higher religiosity.

12

***Table 2 about here***

3.4 Methods

Methodologically, multilevel modelling techniques were considered (Hox, 2010; Snijders and

Bosker, 2011), taking into account that respondents (at level 1, ni=141696) are nested in

country-time (at level 2, nj=75) nested in countries (at level 3, nk=15). Macro-level data, such

as the important information on FLFP, were introduced on level 2. The model specification

can be expressed by:

Yijk = β0ijk + β1Xijk + … + βnZjk

with β0ijk = v0k + u0jk + e0ijk

where Yijk is generalized trust of an individual i in country k at country-time j. In the example

give above, individual-level variables are denoted as X, while country-level variables,

introduced on level-2 (‘country-time’), are denoted as Z. My model specification will allow

analyzing whether between-country variance is stably clustered between countries over time

(level 3) or whether the bulk of higher-level variation may be explained by short-term

developments irrespective of the country in question (level 2).

Yet, there is also a disadvantage concomitant to distinguishing between country-time level

and country level: With only 15 European countries, the sample size on the highest level is

rather small. Therefore, the commonly used ‘frequentist’ approach might lead to biased

results (Hox, 2010). Hence, more advanced, Bayesian estimation techniques were used. More

precisely, Markov chain Monte Carlo (MCMC) simulation techniques were employed,

thereby using the maximum likelihood estimators from the standard frequentist approach as

starting values for the MCMC simulation. Data preparation was done in SPSS, whereas the

actual statistical models were estimated with MLwiN 2.24 (Browne, 2009; Rasbash et al.,

2009), a program specifically tailored for multilevel regression analysis. Since MLwiN does

13

not offer the possibility to apply weights when using MCMC simulation, non-weighted data

were used.

Each of the regression coefficients was subject to careful inspection, because the Bayesian

approach yields a distribution for every single parameter of a given model. To ensure the

reliability of the results, simulation was continued until all of the stated regression

coefficients met the Brooks-Draper convergence criteria (Hox, 2010). For shortening

estimation time, re-parameterization procedures were used (hierarchical centering on level 3,

see Browne et al., 2009). A ‘burn-in’ of maximally 5,000 iterations, a Markov chain of

maximally 75,000 iterations and a thinning factor of maximally 10 were sufficient to estimate

reliable results. Even though the simulations yield contingence intervals rather than point

estimators regarding the separate regression coefficients of the models, for reasons of brevity

only the means of the confidence intervals, including the mean standard error, will be reported

in the following regression tables (complete MCMC tables, including confidence intervals

and parameter diagnostics are available by the author upon request).

To compare the models regarding their fit, the Bayesian deviance information criterion (DIC)

will be employed. Generally, the smaller the DIC value, the better the model fit. However, the

DIC used in MLwiN is rather insensitive to the inclusion of country-level variables, hence it

is predominantly used to decide whether the inclusion of additional individual-level variables

or cross-level interactions leads to a better prediction. As a rule of thumb, only a decrease in

DIC of at least 4 should be regarded as indicating a significantly better model fit

(Spiegelhalter et al., 2002). To minimise multicollinearity problems, all continuous measures

were grand-mean centred in the statistical analysis.

14

4. Empirical Analysis

4.1 Descriptive findings

In line with previous findings, data from the first five rounds of the ESS (2002-2011) show

that it is especially respondents in the Nordic countries who score particularly highly on

generalized trust (Table 3). But in cross-country comparison, also people living in the

Netherlands and Switzerland are found to be comparatively highly trusting in strangers.

***Table 3 about here***

Does employment contribute to fostering generalized trust? Approaching this question on the

basis of descriptive, individual data (Table 4) confirms this thesis: economically active

respondents score more highly on trust than homemakers, respondents on sick leave and the

unemployed. An exception is the group of those in education, who are even more trusting

than those who were in gainful employment at the time of interview. An analysis of variance

(ANOVA) confirms that these group differences are highly significant (p<0.01). Thus,

employment, next to being in education, may be considered a positive determinant of

generalized trust at the individual level, whereas being a labour market outsider negatively

impacts trust in strangers.

***Table 4 about here***

As a next step, I focused on the relation between women’s employment and generalized trust.

Figure 1 provides an overview on this correlation (see also Table 3). Given the relatively high

FLFP rates of the Netherlands and Switzerland, it may become clearer why these two

15

countries also fall into the category of the highly trusting nations, even though they do not

feature the universal welfare policies that are more characteristic of the Nordic country

cluster. In the next section, multivariate analyses will show whether the correlation between

FLFP and generalized trust will also hold when taking into account additional variables.

***Figure 1 about here***

4.2 Multivariate findings

How much of the variance regarding generalized trust is due to country differences? To

answer this question, a so-called ‘empty model’ (Table 5, Model 0) was run. Computing the

intraclass correlation coefficient (ICC) according to Snijders and Bosker (2011), one sees that

about 21 per cent of the variation are located beyond the individual level. Put differently:

about a fifth of the variation in trust cannot be explained by differences between individuals.

Rather, one needs to concentrate on differences between countries.

***Table 5 about here***

Higher-level variance in regard to generalized trust is found almost completely on level 3

(ICClevel3= 0.21), indicating that levels of variance hardly change within countries during the

period 2002-2011 (the ICC regarding level 2 is <0.00). Rather, differences in generalized trust

remain almost stably clustered between the 15 European countries over time. This result

clearly strengthens the institutionalist perspective on the formation of trust in strangers,

whereas there is hardly any empirical evidence that generalized trust might be subject to

short-term shocks like the economic crisis and its side-effects, such as suddenly rising

unemployment rates and more precarious living conditions (Uslaner, 2010).

16

Next, gender was introduced (Model 1). In line with previous cross-national studies that use

the standard single-item measure for generalized trust, it is found that men are significantly

more trusting than women. Additionally, a random intercept random slope multilevel model

(results are available upon request), in which the effect of gender was allowed to vary on level

3, was estimated. According to this model, women are, all other things being equal, still found

to be less trusting than men. However, the size of the negative effect of being a woman

significantly and substantially varies between countries. In line with the arguments derived

from the theoretical discussion, women from the Nordic countries are found to be

significantly more trusting than European women on average. Moreover, a DIC-comparison

shows that allowing for the random gender term at level 3 significantly improves the model

fit, compared to the (depicted) model with the fixed gender effect.

In the next step, a set of individual-level variables was added (Table 5, Model 2), showing

that the finding of a negative gender effect remains robust when controlling for other

important background variables. Similar to the findings of Lee (forthcoming) and in line with

my descriptive findings, the group of people inside the labour market tends to be less trusting

than the outsiders, that is, unemployed, disabled/sick and homemaking respondents. Again,

the group of respondents in education is more trusting than those who were economically

active at the time of interview.

It would go beyond the scope of this article to discuss the effects of the remaining individual-

level variables in detail. Overall, however, my results point in the same direction as previous

studies: education, wealth, a privileged class position, religiosity and not living in a single

household are conducive to generalized trust, whereas having had unemployment experiences

in the past, living in urban areas, and having a migratory background contribute to distrust in

17

strangers. While the entire set of individual-level variables explains about 25 per cent of the

total variation at level 3, the bulk of between-country variation remains unexplained.

Are women’s employment rates a contributory factor in explaining between-country variation

in generalized trust? Model 3 (Table 6), in which the measure for FLFP was added,

corroborates this hypothesis (see H1): people living in countries where relatively large shares

of the female population are economically active are generally more trusting than on average.

This model accounts for 49 per cent of the variation at level 3.

Since FLFP rates are highly correlated with GDP per capita (r=.74), one may suspect that it is

a country’s economic power rather than women’s employment rates that might be significant

for generalized trust. To deal with this objection, Model 4 (Table 6) additionally controls for

the effect of GDP per capita, showing that the effect of FLFP indeed becomes insignificant

when controlling for economic output.

However, while Model 4 tests for the general effect of FLFP, there are good reasons to

assume a direct impact of FLFP on women. Therefore, an interaction term regarding women’s

employment rates and gender was added. According to the results (Model 5), it is only

women’s trust levels that are positively correlated with higher female labour force

participation, whereas there is no significant connection between FLFP and men. Therefore, it

may be concluded that the findings support H2, which maintained a direct effect of FLFP on

women’s generalized trust. The decrease of about 17 in DIC, using Model 4 as the baseline

model here, shows that adding the cross-level interaction in Model 5 significantly improves

the model fit. Overall, the final model can explain about 59 percent of variation at the

country-level (level 3).5

18

*** Table 6 about here***

5. Discussion and Conclusions

Why are inhabitants of some countries more trusting than those from other countries? This

study has provided solid evidence that female labour force participation is a contributory

factor in explaining between-country variation in generalized trust. It was assumed that being

economically active should enhance people’s opportunities to establish weak ties with people

from dissimilar background – something considered conducive to trust in strangers (Stolle,

2002). Whereas interaction with family members and friends was assumed to fortify

particularized trust, positive contact with people from different backgrounds may, from the

perspective of intergroup contact theory (Allport, 1954), lead to a more positive evaluation of

strangers and, eventually, to generalized trust. Following affect theory of social exchange

(Lawler, 2001), the formation of ties to people from different backgrounds was suggested to

be particularly likely in the domain of work, given that work brings together quite diverse

people who have common goals, interact without competition and get involved in productive

social exchange.

Empirically, survey data from the first five rounds of the European Social Survey were used,

thereby covering 15 European countries over the period 2002-2011. On the basis of results

from Bayesian multilevel regression models, it could be shown, first, that on an aggregate

level, high female labour force participation is indeed conducive to generalized trust. Second,

this effect is insignificant when controlling for GDP, a variable that is highly and positively

correlated with female labour force participation. Third, testing for the cross-level interaction

between gender and labour force participation rates, it could be shown that FLFP does

positively impact generalized trust, yet only among women.

19

While this article contributes to the literature about the important linkage between social

equality and generalized trust (Rothstein and Uslaner, 2005), there remain some open

questions. One aspect concerns the lack of conceptual distinction between different forms of

employment. Clearly, not every kind of work may be considered to bring about the positive

effects deemed necessary to make people trustful of strangers. Hence, future research shall be

dedicated to investigate which kinds of jobs are particularly conducive to generalized trust.

Moreover, it remains unclear as to whether it matters to what extent women work. For

example, female labour force participation tends to be relatively high in Switzerland, yet

almost 60 per cent of the employed Swiss women are working part-time, which is a record

among OECD countries (Booth and Van Ours, 2010). Hence, a more detailed distinction

between different kinds of working arrangements at the individual level and also more refined

variables for measuring women’s employment at the macro level would be desirable.

Finally, why are levels of generalized trust so stable within European countries over time

(Rothstein and Uslaner, 2005)? As has been shown, employment, at the micro level, is

conducive to generalized trust. So why do country-specific levels of generalized trust change

so little in the face of economic downturn? Obviously, trust in strangers has a tendency to

persist, irrespective of any short-term shocks at the societal level, such as the most recent

global financial and economic crisis (see also Uslaner, 2010). However, taking seriously the

findings regarding the interplay between work and generalized trust, once unemployment

becomes more of an enduring, collective experience (Uslaner, 2002), as is currently

happening in many parts of Europe, trust would be expected to decline in the long term.

Therefore, it will be all the more important to study time trends in generalized trust, based on

longitudinal data as well as on data from repeated cross-sectional surveys.

20

Notes

1. Even though Rothstein and Uslaner (2005: 51) acknowledge that gender equality might play an important role in shaping people’s generalized trust, this issue is not empirically addressed in their study.

2. While clubs and other associations likewise facilitate productive exchange between

their members, the current debate in the generalized trust literature suggests that it is less likely that social exchange within these environments should promote generalized trust. While Putnam (2000) started from the premise that civic engagement should lead to generalized trust, more recent studies suggest that the direction of causality is the reverse: if at all, it seems to be generalized trust that explains civic engagement, and not the other way around (Stolle, 2001; Uslaner, 2002; Delhey and Newton, 2005; Sønderskov, 2011). In addition, clubs and associations are very homogeneous regarding their composition (McPherson and Smith-Lovin, 1987), so that productive exchange within these associations should, due to the similarity of the people engaged in it, fortify particularized trust rather than generalized trust.

3. To access the ESS data and to obtain more information on the European Social Survey

in general, please visit: http://www.europeansocialsurvey.org/ 4. Additionally, I tested models featuring a three-item trust scale as the dependent

variable (yielding similar results in the multivariate analysis). While it is advantageous that this scale fulfils weak cross-national measurement equivalence (Reeskens and Hooghe, 2008), there is a problem in connection with gender: whereas women are, according to the well-established single-item trust measure, less trusting than men, they are more trusting according to the three-item trust scale. This is mainly due to the inclusion of the other two trust items which refer to the perception of strangers’ helpfulness. Hence the single-item dependent variable was preferred.

5. Diagnostics show that multicollinearity is neither an issue of concern in Model 4

(highest VIF: 2.386; lowest Tolerance: .419; Condition index: 7.720) nor in Model 5 (highest VIF: 3.365; lowest Tolerance: .297; Condition index: 7.723) (see O’Brien 2007).

21

References Albrekt Larsen, C. (2007). How Welfare Regimes Generate and Erode Social Capital: The

Impact of Underclass Phenomena, Comparative Politics, 40, 83-101. Allport, G. W. (1954). The Nature of Prejudice. Cambridge, MA: Perseus. Bjørnskov, C. (2006). Determinants of generalized trust: A cross-country comparison, Public

Choice, 130, 1-12. Bjørnskov, C. (2012). How Does Social Trust Affect Economic Growth?, Southern Economic

Journal, 78, 1346-1368. Booth, A., L. and Van Ours, J. C. (2010). Part-Time Jobs: What Women Want? Institute for

the Study of Labor (IZA): IZA Discussion Paper 4686. Browne, W. J. (2009). MCMC estimation in MLwiN: Version 2.13. Bristol: Center for

Multilevel Modelling. Browne, W. J., Steele, F., Golalizadeh, M. and Green, M. J. (2009). The use of simple

reparameterizations to improve the efficiency of Markov chain Monte Carlo estimation for multilevel models with applications to discrete time survival models, Journal of the Royal Statistical Society, 172, 579-598.

Cook, K. S. and Emerson, R. S. (1979). Power, Equity, and Commitment in Exchange Networks, American Sociological Review, 27, 721-739.

Delhey, J. and Newton, K. (2005). Predicting Cross-National Levels of Social Trust: Global Pattern or Nordic Exceptionalism?, European Sociological Review, 21, 311-327.

Delhey, J., Newton, K. and Welzel, C. (2011). How General is Trust in 'Most People'? Solving the Radius of Trust Problem, American Sociological Review, 76, 786-807.

Esping-Andersen, G. (1990). The Three Worlds of Welfare Capitalism. Princeton: Princeton University Press.

Feld, S. (1981). The Focused Organization of Social Ties, American Journal of Sociology, 86, 1015-1035.

Freitag, M. (2003). Social Capital in (Dis)Similar Democracies: The Development of Generalized Trust in Japan and Switzerland, Comparative Political Studies, 36, 936-966.

Freitag, M. and Bühlmann, M. (2009). Crafting Trust: The Role of Political Institutions in a Comparative Perspective, Comparative Political Studies, 42, 1537-1566.

Freitag, M. and Traunmüller, R. (2009). Spheres of Trust: An empirical analysis of the foundations of particularised and generalised trust, European Journal of Political Research, 48, 782-803.

Glanville, J. L. and Paxton, P. (2007). How Do We Learn to Trust? A Confirmatory Tetrad Analysis of the Sources of Generalized Trust, Social Psychology Quarterly, 70, 230-242.

Granovetter, M. S. (1973). The Strength of Weak Ties, American Journal of Sociology, 78, 1360-1380.

Halman, L. and Luijkx, R. (2006). Social Capital in Contemporary Europe: Evidence from the European Social Survey, Portugese Journal of Social Science, 5, 65-90.

Hamamura, T. (2012). Social Class Predicts Generalized Trust But Only in Wealthy Societies, Journal of Cross-Cultural Psychology, 43, 498-509.

Hardin, R. (2002). Trust and Trustworthiness. New York: Russell Sage Foundation. Hooghe, M., Reeskens, T., Stolle, D. and Trappers, A. (2009). Ethnic Diversity and

Generalized Trust in Europe: A Cross-National Multilevel Study, Comparative Political Studies, 42, 198-223.

22

Hox, J. J. (2010). Multilevel Analysis: Techniques and Applications (2nd edition). New York: Routledge.

Kääriäinen, J. and Lehtonen, H. (2006). The Variety of Social Capital in Welfare State Regimes - A Comparative Study of 21 Countries, European Societies, 8, 27-57.

Kaasa, A. and Parts, E. (2008). Individual-Level Determinants of Social Capital in Europe: Differences between Country Groups, Acta Sociologica, 51, 145-168.

Knack, S. and Keefer, P. (1997). Does Social Capital Have an Economic Pay-Off? A Cross-Country Investigation, Quarterly Journal of Economics, 112, 1251-1288.

Korpi, W. (2000). Faces of Inequality: Gender, Class and Patterns of Inequality in Different Types of Welfare States, Social Politics, 7, 127-191.

Lawler, E. J. (2001). An Affect Theory of Social Exchange, American Journal of Sociology, 107, 321-52.

Lee, C.-S. (forthcoming). Welfare States and Social Trust, Comparative Political Studies, . Leiulfsrud, H., Bison, I. and Solheim, E. (2010). Social Class in Europe II: The European

Social Survey 2002-2008, NTNU, Department of Sociology and Political Science (Discussion Paper), http://www.svt.ntnu.no/iss/ESSReport2010.pdf.

McPherson, M. and Smith-Lovin, L. (1987). Homophily in Voluntary Organizations: Status Distance and the Composition of Face-to-Face Groups, American Sociological Review, 52, 370-379.

Mutz, D. C. and Mondak, J. J. (2006). The Workplace as a Context for Cross-Cutting Political Discourse, The Journal of Politics, 68, 140-155.

Nannestad, P. (2008). What Have We Learned About Generalized Trust, If Anything?, Annual Review of Political Science, 11, 413-436.

O'Brien, R. M. (2007). A Caution Regarding Rules of Thumb for Variance Inflation Factors, Quality and Quantity, 41, 673-690.

Offe, C. (1999). Trust and Knowledge, Rules and Decisions: Exploring a Difficult Conceptual Terrain. In Warren, M. E. (Ed) Democracy and Trust. Cambridge: Cambridge University Press, pp.

Pattie, C. J. and Johnston, R. J. (2008). It's Good To Talk: Talk, Disagreement and Tolerance, British Journal of Political Sciences, 38, 677-698.

Pettigrew, T. F. (1998). Intergroup Contact Theory, Annual Review of Psychology, 49, 65-85. Pichler, F. and Wallace, C. (2007). Patterns of Formal and Informal Social Capital in Europe,

European Sociological Review, 23, 423-435. Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community. New

York: Simon & Schuster. Rasbash, J., Steele, F., Browne, W. J. and Goldstein, H. (2009). A User's Guide to MLwiN:

Version 2.10. Reeskens, T. and Hooghe, M. (2008). Cross-cultural measurement equivalence of generalized

trust: Evidence from the European Social Survey (2002 and 2004), Social Indicators Research, 85, 515-532.

Rosenberg, M. (1956). Misanthropy and political ideology, American Sociological Review, 21, 695.

Rothstein, B. and Uslaner, E. M. (2005). All for All: Equality, Corruption, and Social Trust, World Politics, 58, 41-72.

Snijders, T. A. B. and Bosker, R. J. (2011). Multilevel Analysis: An introduction to basic and advanced multilevel modeling. London/Thousand Oaks/New Delhi: Sage (2nd revised edition).

Sønderskov, K. M. (2011). Does Generalized Social Trust Lead to Associational Membership? Unravelling a Bowl of Well-Tossed Spaghetti, European Sociological Review, 27, 419-434.

23

Spiegelhalter, D., Best, N. B., Carlin, B. P. and van der Linde, A. (2002). Bayesian measures of model complexity and fit, Journal of the Royal Statistical Society, 64, 583-639.

Steiber, N. and Haas, B. (2012). Advances in explaining women's employment patterns, Socio-Economic Review, 10, 343-367.

Stolle, D. (2001). Clubs and Congregations: The Benefits of Joining an Association. In Cook, K. S. (Ed) Trust in Society. New York: Russel Sage Foundation, pp. 202-244.

Stolle, D. (2002). Trusting Strangers - The Concept of Generalized Trust in Perspective, Österreichische Zeitschrift fuer Politikwissenschaft, 31, 397-412.

Stolle, D. and Hooghe, M. (2004). The Roots of Social Capital: Attitudinal and Network Mechanisms in the Relation between Youth and Adult Indicators, Acta Politica, 39, 422-441.

Stolle, D. and Rothstein, B. (2008). The State and Social Capital: An Institutional Theory of Generalized Trust, Comparative Politics, 40, 441-459.

Sturgis, P. and Smith, P. (2010). Assessing the validity of generalized trust questions: What kind of trust are we measuring?, International Journal of Public Opinion Research, 22, 74-92.

Torpe, L. and Lolle, H. (2011). Identifying Social Trust in Cross-Country Analysis: Do We Really Measure The Same?, Social Indicators Research, 103, 481-500.

Uslaner, E. M. (2002). The Moral Foundations of Trust. Cambridge: Cambridge University Press.

Uslaner, E. M. (2008). Where you stand depends on where your grandparents sat: the inheritability of generalized trust, Public Opinion Quarterly, 72, 725-740.

Uslaner, E. M. (2010). Trust and the Economic Crisis of 2008, Corporate Reputation Review, 13, 110-123.

Uslaner, E. M. (2012). Measuring Generalized Trust: In Defense of the Standard Question. In Lyon, F., Möllering, G., Sanders, M. and Hatzakis, T. (Eds) Handbook of Research Methods on Trust. London: Edward Elgar, pp. 72-82.

Van Oorschot, W. and Arts, W. (2005). The social capital of European Welfare States: the crowding out hypothesis revisited, Journal of European Social Policy, 15, 5-26.

Van Oorschot, W., Arts, W. and Gelissen, J. (2006). Social Capital in Europe: Measurement and Social and Regional Distribution of a Multifaceted Phenomenon, Acta Sociologica, 49, 149-167.

Weare, C., Musso, J. and Jun, K.-N. (2009). Cross-Talk: The Role of Homophily and Elite-Bias in Civic Associations, Social Forces, 88, 147-173.

Whiteley, P. (1999). The Origins of Social Capital. In Van Deth, J. W., Maraffi, M., Newton, K. and Whiteley, P. (Eds) Social Capital and European Democracy. London: Routledge, pp. 25-44.

24

Table 1: Sample size (by country and ESS survey round)

round 1

(2002-2003)

round 2

(2004-2005)

round 3

(2006-2007)

round 4

(2008-2009)

round 5

(2010-2011)

Belgium 1891 1776 1798 1759 1704

Denmark 1498 1478 1496 1609 1573

Finland 1998 2015 1895 2193 1877

France 1498 1805 1985 2070 1727

Germany 2913 2866 2913 2746 3027

Hungary 1676 1490 1510 1538 1557

Netherlands 2363 1880 1885 1775 1825

Norway 2036 1758 1749 1549 1547

Poland 2097 1707 1713 1615 1748

Portugal 1506 2049 2180 2359 2137

Slovenia 1510 1437 1472 1278 1397

Spain 1704 1661 1868 2565 1881

Sweden 1991 1941 1923 1829 1492

Switzerland 2037 2141 1801 1815 1505

UK 2049 1893 2384 2348 2415

Source: European Social Survey (2002-2011)

25

Table 2: Descriptives

variable mean SD min max

generalized trust 5.20 2.398 0 10

female .527 .499 0 1

economically active

.501 .500 0 1

unemployed .049 .216 0 1

retired .231 .421 0 1

disabled/sick .026 .158 0 1

homemaker .088 .282 0 1

in education .088 .284 0 1

labour force status: other/NA

.017 .112 0 1

previous unemployment

experience

.256 .437 0 1

service classes .287 .452 0 1

routine non-manual workers

(ref.: service class)

.228 .419 0 1

small self-employed

.076 .264 0 1

working class .307 .461 0 1

class not applicable

.102 .303 0 1

age 47.566 18.504 14 123

migratory background

0.835 .277 0 1

years of education 12.057 4.232 0 56

household size 2.718 1.406 1 22

community size 3.018 1.177 1 5

religiosity 4.615 2.926 0 10

female labour force participation

rate (OECD)

62.639 7.853 46.2 74.4

GDP per capita (in 1000 US-

Dollar) (Eurostat)

30.954 8.590 11.89 51.10

Sources: European Social Survey 2002-2011 (n=141696); Eurostat; OECD

26

Table 3: Generalized trust and female labour force participation in Europe (2002-2011)

2002-2003 2004-2005 2006-2007 2008-2009 2010-2011

Belgium Trust 4.81 4.79 4.98 5.13 5.04

FLFP 51.8 52.6 54.0 56.2 56.5

Denmark Trust 6.99 6.76 7.02 6.92 6.84

FLFP 70.5 71.6 73.4 73.9 71.1

Finland Trust 6.46 6.52 6.57 6.45 6.50

FLFP 65.7 65.5 67.3 69.0 66.9

France Trust 4.47 4.52 4.43 4.45 4.33

FLFP 58.2 58.2 58.6 60.4 59.9

Germany Trust 4.61 4.75 4.71 4.84 4.65

FLFP 58.7 59.2 61.4 64.3 66.1

Hungary Trust 4.08 4.01 4.28 4.15 4.48

FLFP 50.9 50.7 51.2 50.6 50.6

Netherlands Trust 5.71 5.80 5.73 5.89 6.00

FLFP 63.5 63.5 65.4 69.3 69.4

Norway Trust 6.60 6.63 6.82 6.62 6.68

FLFP 72.7 72.7 72.3 75.4 73.3

Poland Trust 3.72 3.60 4.07 4.17 4.39

FLFP 46.2 46.4 48.2 52.4 53.0

Portugal Trust 4.00 3.88 4.09 3.65 3.71

FLFP 61.4 61.7 62.00 62.5 61.1

Slovenia Trust 3.98 4.13 4.06 4.32 3.94

FLFP 57.6 60.5 61.8 64.2 62.6

Spain Trust 4.86 4.89 5.12 4.90 5.14

FLFP 46.8 49.0 54.0 55.7 53.0

Sweden Trust 6.09 6.05 6.28 6.35 6.34

FLFP 72.8 71.8 72.1 73.2 70.3

Switzerland Trust 5.65 5.69 5.72 5.70 5.64

FLFP 70.7 70.3 71.1 73.5 72.3

UK Trust 5.05 5.14 5.40 5.27 5.36

FLFP 66.4 66.6 66.8 66.9 65.3

Notes: Trust= Aggregate level of generalized trust; FLFP= female labour force participation rate. Source: European Social Survey (2002-2011); OECD (FLFP data)

27

Table 4: Generalized trust by individual labour force status Main activity at the time of

interview Generalized trust (mean)

economically active 5.43

unemployed 4.52

disabled/sick 4.65

housework 4.84

retired 4.96

in education 5.50

don’t know/no answer/not

applicable

4.70

Source: European Social Survey 2002-2011; n= 141696.

28

Table 5: Determinants of generalized trust (results from linear multilevel regressions) Model 0 S.E. Model 1 S.E. Model 2 S.E.

constant 5.214 ** 0.285 5.270 ** 0.288 5.863 ** 0.210

female (ref.: male) -0.092 ** 0.012 -0.093 ** 0.013

unemployed (ref.: economically active)

-0.111 ** 0.029

retired -0.177 ** 0.020 disabled/sick -0.410 ** 0.037 homemaker -0.140 ** 0.023 in education 0.362 ** 0.026

other -0.018 0.051

unemployment experience > 3 months

(ref.: no)

-0.132 ** 0.014

routine non-manual

workers (ref.: service class)

-0.213 ** 0.017

small self-employed -0.295 ** 0.025 working class -0.410 ** 0.017

class not applicable -0.266 ** 0.025

migratory background (ref.: no)

-0.189 ** 0.021

education (years) 0.060 ** 0.002

age (metrical) 0.005 ** 0.001

coping on present

income (ref.: living comfortably on present

income)

-0.324 ** 0.014

difficult on present income

-0.590 ** 0.020

very difficult on present income

-0.901 ** 0.032

don’t know/no answer -0.429 ** 0.058

household size 0.022 ** 0.005

community size -0.010 * 0.005

religiosity 0.038 ** 0.002

level-3 variance (countries)

1.225 0.554 1.226 0.567 0.914 0.400

level-2 variance (country-time)

0.020 0.004 0.020 0.004 0.017 0.004

level-1 variance 4.810 0.018 4.808 0.018 4.563 0.017

DIC 624739 624679 617304 Notes: **p< 0.01; *p< 0.05. Results from MCMC simulations. Only the means of the MCMC confidence intervals are stated. Source: European Social Survey (2002-2011), ni= 141696; nj= 75; nk= 15.

29

Table 6: Macro-level determinants of generalized trust (results from linear multilevel models) Model 3 S.E. Model 4 S.E. Model 5 S.E.

constant 5.863 ** 0.210 5.867 ** 0.186 5.874 ** 0.187

individual-level variables

female (ref.: male) -0.093 ** 0.012 -0.093 ** 0.013 -0.492 ** 0.093

individual-level control

variables see Model 3 (coefficients are almost identical,

results available upon request) country-level variables

female labour force

participation (FLFP) rate

0.027 ** 0.008 0.009 0.010 0.005 0.011

GDP per capita 0.032 ** 0.013 0.033 ** 0.013

cross-level interaction

FLFP rate*female 0.006 ** 0.001 FLFP rate*male Ref.

level-3 variance (countries)

0.627 0.284 0.503 0.234 0.511 0.244

level-2 variance (country-time)

0.016 0.004 0.015 0.004 0.015 0.004

level-1 variance (individuals)

4.563 0.017 4.563 0.017 4.563 0.017

DIC 617303 617304 617287

Notes: **p< 0.01; *p< 0.05. FLFP: female labour force participation. Results from MCMC simulations. Only the means of the MCMC confidence intervals are stated. Models 3 to 5 contain all individual-level variables depicted in Model 2, Table 3. The relevant coefficients are almost identical to those in Model 2. For reasons of brevity, they are not shown here (results are available upon request). Source: European Social Survey (2002-2011), ni= 141696; nj= 75; nk= 15.