M PRA

Munich Personal RePEc Archive

Money, Social Capital and Materialism.

Evidence from Happiness Data

Marcin Piekalkiewicz

Department of Economics, University of Siena

23 March 2016

Online at https://mpra.ub.uni-muenchen.de/70522/
MPRA Paper No. 70522, posted 8 April 2016 05:26 UTC
 Money, Social Capital and Materialism.
Evidence from Happiness Data∗
Marcin Piekalkiewicz†
University of Siena
March 23, 2016

Abstract
Are unhappiness, high concern for money and scarcity of social capital different faces of
the same phenomenon? Economists tend to treat these variables as distinct correlates of
well-being. On the contrary, positive psychologists argue that they all relate to material-
ism, a system of personal values ascribing great importance in life to extrinsic motivations
and low priority to intrinsic motivations. Using data from two European cross-sectional
surveys and the German Socio-Economic Panel, I test the hypothesis that material inter-
ests, proxied by the effects of individual and reference income on well-being, are associated
with low levels of social capital. The results suggest that people with scarce social capital
tend to have greater material interests, whereas the negative effect of income comparisons
on well-being is eliminated for individuals exhibiting the highest levels of social capital.
The implication of such finding is that promoting social capital reduces people’s material
concerns and has positive impact on their well-being. The results from a country-level
analysis additionally show that, since social capital moderates the importance of income
for well-being on individual level, the well-being gap between income groups is signifi-
cantly smaller in countries with higher social capital.

Keywords: subjective well-being · life satisfaction · social capital · materialism · rel-

ative income · social comparisons · happiness inequality

JEL Classifications: D31 · I31 · Z13

∗
I am extremely grateful to Stefano Bartolini and Francesco Sarracino for their guidance and support. I
would also like to thank (in alphabetical order): Marianna Belloc, Samuel Bowles, Xi Chen, Luca Fumarco,
Alessio Fusco, Umut Kılınç, Malgorzata Mikucka, Chiara Peroni, Cesare Riillo, Marco Ventura, Aleksandra
Wilczyńska and Lorna Zischka. This work has also benefited from comments from seminar participants at the
University of Siena, June 2015; the Conseil économique et social du Grand-Duché de Luxembourg, July 2015;
the Polo Lionello Bonfanti, “Quality of Life in Tuscany: Theory and Policy”, October 2015; and the Institute
for Social and Economic Research, University of Essex, January 2016. The research was supported by the grant
from “Tuscany: a Global Laboratory for Quality of Life” (promoted by Tuscany Region, Toscana Promozione
and E.di C.s.p.a. - Polo Lionello Bonfanti, Prot. 2014/3014/8.4.1/30, Decreto n.135 del 28/04/2014 and Decreto
n. 325 del 15/12/2014). All errors are mine.
†
University of Siena, Department of Economics and Statistics, P.zza S.Francesco 7-8, 53100 Siena, Italy.
E-mail: [email protected], [email protected].

1
1 Introduction
Happiness economics has identified various correlates and determinants of subjective well-being1
(see Dolan et al., 2008; MacKerron, 2012, for reviews). Some of them, for example, inflation
rate and unemployment rate, being basically untrended, do not influence the long-run move-
ments in subjective well-being (SWB). Others instead, such as absolute and relative income or
social capital, play an important role in explaining the happiness-income paradox defined by
Easterlin (1974, 1995). A vast number of empirical tests show that income comparisons are
crucial for undermining the possibility of increasing the average happiness by economic growth
(Clark et al., 2008). Furthermore, the most relevant examples of the Easterlin paradox can be
additionally explained by the decline of social capital. That is to say, in the most spectacular
cases of growing economies throughout the last decades, both, in developed countries such as
the USA (Bartolini et al., 2013a) and in developing ones, for instance in China (Bartolini and
Sarracino, 2015), the negative changes in happiness are associated with the decreasing indica-
tors of trust and civicness. What those countries have in common is the fact that the SWB
trend is predicted by two social factors: the erosion of social capital and the strong role of
social comparisons. Moreover, Bartolini and Sarracino (2014) show that, in a representative
international sample, average well-being in the long run is more likely to grow in countries
where social capital grows than in countries where the economy grows.
All these factors – individual income, income comparisons and social capital – are usually
analysed separately in happiness economics; however, according to positive psychologists, they
are all related to the same phenomenon called “materialism”. Positive psychologists define it as
a system of personal values ascribing higher importance in life to extrinsic motivations and lower
to intrinsic motivations. The distinction between these two types of motivation refers to the
instrumentality, or lack thereof, of the reason for doing something. The term “extrinsic” stands
for motivations that are external to an activity (for example, money or other material rewards).
Conversely, “one is said to be intrinsically motivated to perform an activity when one receives no
apparent reward except the activity itself” (Deci, 1971, p. 105). For instance, an individual can
decide to work because he or she finds a given job interesting (intrinsic motivation) or because
it will bring him or her a certain remuneration (extrinsic motivation). In short, materialistic
persons tend to attribute an elevated priority to life goals such as money, luxury consumer’s
goods, success or high economic position, and a lower priority to human relationships, affection,
solidarity, civic engagement or – more generally – to pro-social behaviours.
The literature analysed materialism in many quantitative studies using distinct population
samples. Using various methods these works quantify the levels of materialism in individuals
in order to relate it to different psychological outcomes. The main finding of these studies is
that materialism is associated with a poorer quality of relationships with others. This negative
relation is a consequence of relational attitudes developed by individuals with materialistic
inclinations. Especially unfavourable for having decent social relationships is the tendency

1
Diener (2006, p. 399) defines ‘subjective well-being’ as “all of the various types of evaluations, both positive
and negative, that people make of their lives”. Following many benchmark studies (e.g., Easterlin, 2003; Frey
and Stutzer, 2000), I will use terms ‘life satisfaction’ and ‘happiness’ interchangeably as two most common
indicators of subjective well-being.

2
to consider people as objects. An individual who tends to “objectify” others not only lacks
generosity, empathy, and cooperative capacity, but also exhibits higher level of cynicism and
mistrust, which then becomes an obstacle for building genuine relationships. (Belk, 1985; Cohen
and Cohen, 1996; Kasser, 2002; Kasser and Sheldon, 2000; Kasser et al., 1995; McHoskey, 1999;
Sheldon et al., 2000). In this view materialism is a cause of relational poverty, however, the
opposite direction of causality is plausible as well. The scarcity of emotional relationship with
one’s parents during infancy leads to greater importance of material concerns in adult life. It
is, in fact, lack of affection that causes an individual to feel insecure, while materialism is an
answer to insecurity (Cohen and Cohen, 1996; Kasser and Sheldon, 2000; Kasser et al., 1995;
Williams et al., 2000). The message coming from positive psychologists is that materialism and
relational poverty affect each other resulting in a vicious circle.
The research question posed in this paper is then the following: is it true that high interest
for money and scarcity of social capital are associated with each other? If yes, social capital
should be a factor which decreases the importance of money in one’s life, whereas a lower
importance of individual income should also mean a lower importance of income comparisons.
I respond the question using the strength of the relationship between subjective well-being
and (i) individual income as a proxy of interest for own money; (ii) income of the reference
group as a proxy of interest for others’ money. As measures of social capital I employ survey
questions focused on the quality of social relationships (e.g., frequency of meeting friends, trust
in others). To test whether the two phenomena are related to each other, I integrate them
within interaction terms introduced into the standard happiness equation. The conclusion is
that psychologist are right: attachment to money and low social capital are two faces of the
same phenomenon. Individuals with better social relationships attach lower importance to
their own income and are less affected by income comparisons, whereas at the highest levels
of social capital the negative effect of income comparisons is eliminated and the positive effect
of individual income is maintained. An implication of this result is the following: if higher
social capital moderates the negative impact of income comparisons on subjective well-being,
it should be perceived as a condition for economic growth being followed by happiness growth.
The remainder of the paper is organised as follows. Section 2 provides background for the
study reviewing the literature. Section 3 presents the evidence from cross-section; it firsts
describes the data and methodology (section 3.1), next it presents the results from the primary
sample of individuals (main result in section 3.2.1, causality analysis in section 3.2.2), from an
alternative sample of individuals (section 3.3), and from a country-level analysis (section 3.4).
Section 4 applies panel data in order to control for individual fixed effects (section 4.1) and to
analyse the impact of social capital changes on the importance of income for well-being (section
4.2). Section 5 concludes providing discussion and potential policy implications.

2 Background
The investigation of the relationship between utility and money is as old as the economic
science, however, the importance of relative income and relative consumption for well-being
has been underlined gradually. Two early contribution were made by Veblen (1899) with his

3
conspicuous consumption and the so-called “Veblen effects”, and by Duesenberry (1949) who
proposed the relative income hypothesis explaining that individual’s consumption and saving
behaviour is determined more by his income in relation to others than by its absolute values.
In sociology, the same intuition behind the relativity of well-being components was defined by
Festinger (1954) as the “social comparison process”, and by Runciman (1966) who introduced
the notion of “relative deprivation”.
Social comparisons were proposed by Easterlin (1974) as an explanation to his happiness-
income paradox: even though income is positively correlated with happiness in a cross section,
increasing GDP per capita does not lead to an increase in average level of subjective well-
being. It is because what matters for happiness is income in relative terms, that is to say,
people compare their income to what others earn whereas increasing the income of all will
improve the relative position of no one. More generally, “as a person’s income (consumption)
increases relative to his income standard, so does his SWB. The higher the person’s income is
relative to the standard (or norm), the greater his happiness. As the economy grows, so do
income standards, and this rise in standards acts to deflate the effect of the increased income.”
(McBride, 2001, p. 254)
More recent empirical studies tend to confirm the relevance of income comparisons for in-
dividual well-being. Clark and Oswald (1996) document that, in a representative sample from
the BHPS, reported satisfaction levels of British workers are inversely related to their compar-
ison wage rates. Ferrer-i Carbonell (2005) uses the data from the German SOEP to show that
individual happiness is affected by the income of the reference group about as importantly as
by the own income, and that people are the more satisfied with their lives the higher their
income is compared to the income of the reference group. Additional evidence supporting the
hypothesis that relative income significantly affects individual assessments of SWB comes from
cross-sectional studies on the American GSS data (McBride, 2001) and international samples
from the European Social Survey (Caporale et al., 2009).
Furthermore, position on the income ladder turns out to be as important for SWB as income
expressed in absolute terms. Using a representative sample of British workers, Brown et al.
(2008) show that the level of well-being (measured with different components of job satisfaction,
including satisfaction with pay) depends on the ordinal rank of an individual’s wage within a
comparison group rather than on absolute wage or average wage in the workplace. Boyce et al.
(2010) find evidence within the BHPS data that income rank explains significantly more of
the overall variation in life satisfaction than absolute income, whereas their result holds after
introducing various reference groups. Finally, Carlsson et al. (2007) provide an experimental
proof confirming the importance of relative income and relative consumption for people’s utility,
while Fliessbach et al. (2007) show that social comparisons affect reward-related brain activity,
which constitutes a neurophysiological proof of the existence of relative income concerns.
More evidence comes from studies analysing the direct link between the importance of in-
come comparisons and individual well-being. Clark and Senik (2010), using a sample of workers
from the European Social Survey 2006, find out that there is a negative and significant cor-
relation between the “declared comparison intensity”2 and various measures of SWB: overall

2
As a measure of income comparison intensity Clark and Senik (2010) use a direct question “How important

4
life satisfaction, happiness, job satisfaction, and satisfaction with pay. What is particularly
important for the present analysis, the authors note that individuals reporting to meet socially
more often attach less importance to income comparisons; however, the link between social
capital and importance of relative income is not further investigated as it lies beyond the scope
of their study. Recently, Goerke and Pannenberg (2015) apply novel German data on self-
reported income comparison intensity to show that positional concerns are negatively related
to individual life satisfaction. Additonally, Clark et al. (2015) provide evidence from Japan
using hypothetical discrete choice experiments in which respondents choose between alterna-
tive combinations of income amounts, both for themselves and certain reference group. They
observe that individuals with strong positional concerns (i.e., those who prefer to earn less in
absolute terms but more than the reference group) report lower income satisfaction.
The second line of happiness studies proposes an alternative explanation to the happiness-
income paradox. The “negative endogenous growth” approach suggests that the economy
tends to grow faster when individuals become relatively poorer in social relationships. In
other words, the erosion of social ties can actually “feed” the economic growth as firms, facing
a decline in honesty, trust, and work ethics, have to invest more in defensive expenditures,
control mechanisms, and guard labour (Bartolini and Bonatti, 2002, 2003, 2008). Furthermore,
recent empirical findings demonstrate that the positive effect of income on happiness may be
offset by lower consumption of the so-called relational goods (Becchetti et al., 2011; Bruni and
Stanca, 2008). Becchetti et al. (2008, 2012) show that relational goods, defined as “affective
and expressive, non instrumental, side of interpersonal relationships” (measured, e.g., with
the frequency of attendance at social gatherings) have a positive effect on life satisfaction,
controlling for unobserved individual characteristics and reverse causality.
The link between SWB and the quality of social ties has been also analysed in a broader
sense showing that happiness depends on the level of “social capital”. According to the defini-
tion provided by OECD (2001, p. 41), the term should be associated with “networks together
with shared norms, values and understandings that facilitate co-operation within or among
groups”. Putnam (2000) proposes various measures of social capital: intensity of involvement
in community and organizational life; public engagement (e.g. voting) and volunteering; in-
formal sociability (e.g. visiting friends); and reported levels of interpersonal trust. Empirical
evidence shows that, no matter which measure is applied, the level of social capital is positively
related to subjective well-being (Bartolini et al., 2013b; Helliwell, 2006; Helliwell and Putnam,
2004; Ram, 2010; Sarracino, 2010, 2012). Additionally, Sarracino (2013) documents that social
capital enters positively the happiness equation indifferently in low income and high income
countries. The analysis of social capital in a dynamic perspective indicates that social capital
trends predict the long-run changes in subjective well-being much better than economic growth
(Bartolini and Sarracino, 2014).
As the above review of studies shows, income comparisons and interpersonal relationships
constitute two significant social dimensions of SWB. What is missing in the existing literature,
however, is an investigation linking the two phenomena of interest. An intuition behind a

is to you compare your income with other people’s income?”, with answers on a 0-6 scale ranging from “Not
important at all” to “Very important”.

5
possible association between the importance of income for well-being and the quality of social
ties is given by positive psychologist: high concern for money and poor social relationships
are both related to materialism, a system of personal values ascribing greater importance to
activities motivated extrinsically (Kasser, 2002).
The contribution of the present paper consists of an empirical link between two fundamental
aspects of well-being: social comparisons and social capital. Assuming that attachment to
money and scarcity of social relations are both related to materialistic system of values, I show
that more sociable and trustful people exhibit lower concern for individual and reference income.
An eventual implication of such result is that a reduction of materialism may rise the level of
social capital, and, vice-versa, an increase in social capital can decrease people’s interest for
money. Therefore, policies that reduce materialism or policies that increase social capital may
possibly trigger a self-feeding mechanism that lowers interest for money and simultaneously
increases social capital, having a positive impact on people’s well-being.

3 Evidence from cross section

3.1 Data and methodology
The primary data source used in the study is the European Union Statistics on Income and Liv-
ing Conditions (EU-SILC). In 2013 the questionnaire included an ad-hoc module for measuring
subjective well-being with a set of questions on overall experience of life, satisfaction with mate-
rial living conditions, health, as well as leisure and social interactions. In the empirical analysis
the well-being of an individual will be measured with the answer to a standard self-evaluative
question: “Overall, how satisfied are you with your life these days? Please answer on a scale of
0 to 10, where 0 means ‘Not at all satisfied’ and 10 means ‘Completely satisfied’.”3
As far as social capital is concerned, I focus on its relational aspects looking at two commonly
used measures of interpersonal relationships quality: trust in others (expressing individual
attitude) and frequency of meeting friends (expressing individual behaviour ) (OECD, 2001;
Onyx and Bullen, 2000). The trust question asks respondents to indicate how trustful they
are on a 11-step scale: “Would you say that most people can be trusted? Please answer on a
scale from 0 to 10, where 0 means that in general ‘You do not trust any other person’ and 10
that you feel ‘Most people can be trusted’.” I construct a dummy variable equal to 1 for trust
ranging from 6 to 10. The frequency of socializing is captured by the question: “Do you meet
up with friends/family for a drink/meal (at home or outside) at least once a month?”. The
second social capital dummy variable is therefore equal to 1 if an individual meets his friends or
family at least once per month. In order to capture both aspects of interpersonal relationships
(behaviour and attitude) in one variable, I introduce an “index” of social capital which simply

3
Several studies showed that direct questioning people about their recent affective experience as well as
about global evaluation of their lives are reliable measures of subjective well-being. Answers to well-being
questions correlate well with physical measures of affect such as frequency of smiling, heart rate measures,
and electrical activity in the brain (Blanchflower and Oswald, 2004; Van Reekum et al., 2007) as well as with
non-self-report measures based on evaluations of other people such friends and family (Sandvik et al., 1993).
Moreover, single item scales provide similar correlates of SWB as the multi-item scales (Krueger and Schkade,
2008).

6
adds up the two dummies. The index will be a categorical variable assuming values {0, 1, 2},
whereas its highest level means that a person is both, trustful and sociable.4
The remaining variables of interest are individual and reference income. The individual
income is defined as monthly disposable equivalised income adjusted to PPP.5 I assume that
individuals compare their incomes with others of similar socio-demographic characteristics who
live in the same geographical area (Bartolini et al., 2013b; Boyce et al., 2010; Ferrer-i Carbonell,
2005). The reference income is thus calculated as the average individual income in the reference
group defined as people of the same sex and age group living in the same region.6
The EU-SILC regression sample includes more than 320,000 observations coming from 29
European countries.7 The average life satisfaction is 6.92, whereas 73% of the respondents
meets friends at least monthly and around 58% is trustful (scoring between 6-10). Looking at
the descriptive statistics for the social capital index, one can see that 14% of the sample scores
0, meaning that approximately 1 out of 7 individuals is neither sociable nor trustful, while 45%
is both, trustful and sociable (for detailed descriptive statistics see table A12).
In order to test the hypothesis that social capital moderates the importance of income
and income comparisons for well-being, I adapt a standard “happiness equation” introducing
interaction terms between the social capital index and the income variables:

LSi = α + β1 ∗ log(Ind inc i ) + β2 ∗ log(Ref inc i ) + β3 ∗ SC index i

+ β13 ∗ SC index i ∗ log(Ind inc i )
+ β23 ∗ SC index i ∗ log(Ref inc i )
+ γ 0 X i + εi (1)

where LS is the reported life satisfaction, Ind inc and Ref inc are individual and reference
income, SC index is a categorical variable capturing the level of social capital, X is a vector
of control variables (including: sex, age group, marital status, education level, labour market
status, house owner, and country dummies)8 , and ε is the error term of standard properties.

4
The EU-SILC well-being module includes as well two variables measuring social support: “receiving help
from others” and “having anyone to discuss with personal matters”. I do not include them since the focus is
placed on individual behaviour towards others (meeting socially) and individual attitude towards others (trust),
while receiving help and having support describe rather the behaviour and attitudes of others towards the
individual, accounting for the social environment around him.
5
The equivalised disposable income is the total income of a household, after tax and other de-
ductions, that is available for spending or saving, divided by the number of household members con-
verted into equalised adults; household members are equalised or made equivalent by weighting each
according to their age, using the so-called modified OECD equivalence scale (1.0 to the first adult;
0.5 to the second and each subsequent person aged 14 and over; 0.3 to each child aged under 14);
see http://ec.europa.eu/eurostat/statistics-explained/index.php/Glossary:Equivalised disposable income. In
the PPP adjustment I use price level indices for the actual individual consumption (EU28=100); see
http://ec.europa.eu/eurostat/en/web/products-datasets/-/PRC PPP IND (last update 17.12.15, extracted
18.01.16).
6
Five age groups (under 26, 26-35, 36-45, 46-55, 55+) * two genders * 104 regions = 1040 reference groups.
The average number of individuals in a reference groups is around 312. For the list of regions see table A13.
7
AT, BE, BG, CH, CY, DE, EE, EL, ES, FI, FR, HR, HU, IE, IS, IT, LT, LU, LV, MT, NL, NO, PL, PT,
RO, RS, SE, SK and UK. Countries in which the question on meeting friends was not asked are dropped from
the sample (CZ, DK and SI).
8
I use age groups instead of age and age squared since the age variable in EU-SILC is not perfectly continuous;
it groups all individuals aged 80+ in one category. Household size is not included in the regression as it is directly

7
The interactions are introduced to check weather the impact of individual and reference income
on life satisfaction varies with the level of social capital.9 For the correct interpretation of inter-
action coefficient, the estimated model will include the main effects of the interacted variables
(see Brambor, 2005; Braumoeller, 2004). I assume that self-reported life satisfaction scores can
be treated as a cardinal variable employing an OLS regression, which allows to interpret the
coefficients of interactions more easily. In fact, it has been shown that models assuming cardi-
nality and those assuming ordinality of life satisfaction scores provide similar results (Ferrer-i
Carbonell and Frijters, 2004; Stutzer and Frey, 2006).

3.2 Results: sample of individuals

3.2.1 Main result

The coefficients and statistical significance of the variables of interest confirm the common
findings of the happiness literature (see table A1 for detailed results). Life satisfaction is
positively related to individual income and negatively to reference income. The higher the
income of an individual, the more satisfied he is, whereas his life satisfaction decreases with
rising income of the reference group. The positive impact of social capital on well-being rises
with the level of the SC index, showing that people being both, trustful and sociable are the
happiest (holding other variables constant).
The main interest is focused, however, on the sign and statistical significance of the interac-
tion terms. In fact, the signs are opposite to those of income variables, which indicates that the
impact of own and reference income on well-being is smaller for individuals with higher social
capital (table A1). Compared to people exhibiting the lowest value of social capital (index =
0), the life satisfaction of those who are either trustful or sociable (index = 1) is less affected
by individual income (the coefficient is 9% smaller) and by reference income (the coefficient
smaller by 8%, however, this difference is not significant).
In case of people being trustful and sociable (index = 2), the moderation effect for individual
income amounts to -42%, while for reference income it is equal to -101%, both differences being
statistically significant (table 1).10 This means that at the highest level of social capital own
income is less, but still important for life satisfaction, while the negative effect of reference
income is entirely eliminated (figure 1). One can therefore conclude that social capital has a
double positive effect on well-being: the direct effect, as life satisfaction rises with the level
of social capital; and the indirect effect, as the reference income is no longer harmful for life
satisfaction at the highest value of the SC index (figure 2).11

correlated with the equivalent income; still, including household size (or number of children) among the controls
does not change the results.
9
When rearranging equation 1, it may be shown that the marginal effect of Ind inc on LS is equal to the
expression β1 + β13 ∗ SC index ; the same applies for Ref inc and β2 + β23 ∗ SC index (see appendix B.1 for
details). For a broader discussion on the interpretation of interaction terms see Balli and Sørensen (2013).
10
The percentage moderation effect is calculated as the ratio of interaction coefficient to income coefficient.
For example, the coefficient of “Social capital index = 2 * Log of individual income” is equal to −0.201, while
for “Log of individual income” alone it is 0.479, meaning that for the category “SC index = 2” the income
coefficient is −0.201 + 0.479 = 0.278; the original coefficient 0.479 is therefore decreased by 0.201, which in
percentage terms gives 0.201/0.479 ' 42% (see estimates of specification 3 in table A1).
11
Figure 2 shows that the slope of reference income does not significantly differ between SC index = 0 and

8
 Table 1: Main results: moderation effects (EU-SILC 2013).

Social capital index = 1 Social capital index = 2

Log of individual income -9% -42%
Log of reference income -8% (n/s) -101%
Note: Moderation effects express by how much the income coefficient is moderated at a given
level of social capital index (compared to index = 0); the effects are calculated for specification
3 from table A1. n/s = not significant.

Figure 1: Income effects moderated by increasing social capital (EU-SILC 2013).

.6
Effects of income on life satisfaction

.4

.2

−.2
0 1 2
Social capital index (0−2)

Log of individual income Log of reference income

Note: Average marginal effects on the linear prediction of life satisfaction with 90% confidence intervals; calculated for specification
3 from table A1.

An important robustness check of the obtained results consists of splitting the sample into
two groups of countries characterized by different economic and cultural background: Western
Europe (developed countries) and Eastern Europe (economies characterized by the transition
experience).12 It has been shown in the literature that the impact of reference income on well-
being in transition countries may actually be positive due to the so-called “tunnel effect”.13

SC index = 1 (as the interaction term SC index = 1 ∗ log(Ref inc) is insignificant), whereas the slope for
SC index = 2 becomes flat showing that the negative impact of reference income is eliminated.
12
Western Europe: AT, BE, CH, CY, DE, EL, ES, FI, FR, IE, IS, IT, LU, MT, NL, NO, PT, SE and UK;
Eastern Europe: BG, EE, HR, HU, LT, LV, PL, RO, SK and RS.
13
During a transition from socialism to capitalism, when the social mobility is high, individual well-being
may be positively affected by higher incomes of others as it is not necessary a source of envy and financial
dissatisfaction, but may become a source of information creating positive expectations about own future income
(Hirschman and Rothschild, 1973). Senik (2004) follows this reasoning and shows with Russian panel data from
period 1994-2000 that reference income is positively correlated with life satisfaction on personal level, which
contradicts the standard intuition behind utility and relative income. Additionally, Senik (2008) documents
that higher reference income negatively affects individual happiness in Western Europe, but increases happiness
in the transition countries of Eastern Europe (using data from 1989-2000). In the analysis with the 2013 data I
find out the the impact of reference income is negative and significant in both groups of countries, which may
suggests the the so-called “tunnel effect” was present exclusively during the most dynamic period of transition,
that is, in the early 90-ties.

9
However, after the sample split I actually find out that the impact of reference income on life
satisfaction is negative and significant in both groups of countries. Importantly, the main result
holds showing that for the highest level of social capital, the effects of individual and reference
income on well-being are significantly moderated in both subsamples (table A1).14

Figure 2: Reference income effects eliminated for the highest social capital (EU-SILC 2013).
7.5
Life satisfaction (linear prediction)

6.5

5.5
0 2 4 6 8
Log of reference income

Social capital index = 0 Social capital index = 1 Social capital index = 2

Note: Predictive margins calculated for specification 3 from table A1.

3.2.2 Instrumenting social capital

In the estimated happiness regression the coefficient of the social capital index is significant
and strongly positive, however, the direction of causality is unclear: being sociable and trustful
may make an individual happier, or, vice versa, happiness may positively affect his sociability
and trust in others. Moreover, there may be some unobserved factors, e.g. personality traits,
which affect both, social capital and happiness. This means that the social capital index is
an endogenous variable and, in consequence, so are the interaction terms. In order to tackle
the endogeneity problem, I employ the instrumental variable approach, assuming that in a
regression including the main effect and the interaction effect it is necessary to instrument one
and the other (Balli and Sørensen, 2013).
Finding a proper instrument for social capital is not a trivial task, as most of the factors
affecting social life of an individual affect his happiness as well. I overcome this issue by using
a method proposed by Lewbel (2012), which generates the so-called heteroskedasticity-based
instruments. The Lewbel’s approach allows to identify structural parameters in models with
endogenous or mismeasured regressors; it may be used in “applications where other sources of
identification, such as instrumental variables, repeated measurements, or validation studies, are
not available. The identification comes from having regressors uncorrelated with the product of
14
I additionally run separate regressions interacting the income variables with a given social capital dummy
instead of the index, which also gives significant interaction terms of the same signs (results presented in
appendix A.1.2).

10
heteroscedastic errors, which is shown to be a feature of many models in which error correlations
are due to an unobserved common factor, such as unobserved ability in returns to schooling
models, or the measurement error in mismeasured regressor models.”15 This relatively new
method has been recently applied in various fields: finance (Boschi et al., 2014; Schlueter
et al., 2015), international trade (Lin, 2015), agricultural economics (Emran and Shilpi, 2012),
education economics (Denny and Oppedisano, 2013; Gao and Smyth, 2015; Mishra and Smyth,
2015), health economics (Schroeter et al., 2013; Brown, 2014), and, most importantly, happiness
economics (Tiefenbach and Holdgrün, 2015; Tiefenbach and Kohlbacher, 2015).
I use the Lewbel’s method to instrument the endogenous variables from equation 1: the main
effect of social capital index, the interaction with individual income, and the interaction with
reference income.16 I first regress each of the endogenous variables on the vector of controls
from equation 1:

Endogenous variablei = a + B0 Xi + i . (2)

A crucial assumption of the Lewbel’s approach is that there is heteroskedasticity in the error
term of the “first stage equation” (). In order to test it I run a Breusch-Pagan test indicating
that the null of constant variance is rejected, which means that the the assumption is fulfilled
and the Lewbel’s generated instruments method may be applied. Next, I generate the instru-
ments by multiplying the residuals from the “first stage equation” (equation 2) with each of
the control variables in mean-centred form:

Zj = (Xj − X j ) ∗ ˆ (3)

where j corresponds to a given control variable from vector X, and ˆ is the vector of residuals
from the “first-stage regression” of each endogenous variable on all controls from X (including a
constant vector; equation 2). For each endogenous variable the number of generated instruments
Z is therefore equal to the number of controls in vector X.17
Finally, I estimate the “second stage equation” using the 2SLS method, instrumenting the
endogenous variables with the generated instruments Z (results in table A2). The first impor-
tant observation coming from the obtained results is that the positive impact of social capital
on life satisfaction is causal in the specification with and without the interactions terms. Sec-
ondly, the estimates show that the OLS results hold after controlling for endogeneity of social
capital: the effects of individual and reference income on well-being are significantly moderated.
Again, I observe that the moderation effects for reference income are much stronger than for
the individual income, and that at the highest level of social capital the negative impact of
reference income on well-being is eliminated (table 2).

15
Lewbel, 2012, p. 67. For a detailed description of the Lewbel’s method see appendix B.3.
16
Social capital index is a categorical variable, it is therefore necessary to instrument each category (dummy)
and the interactions with each category; the list of the instrumented variables is thus the following: SC index =1 ,
SC index =2 , (SC index =1 ) ∗ log(Ind inc), (SC index =2 ) ∗ log(Ind inc), (SC index =1 ) ∗ log(Ref inc) and
(SC index =2 ) ∗ log(Ref inc).
17
For a detailed description of the procedure see appendix B.2.

11
 Table 2: Instrumenting social capital: moderation effects (EU-SILC 2013).

Social capital index = 1 (instrumented) Social capital index = 2 (instrumented)

Log of individual income -55% -81%
Log of reference income -87% -122%
Note: Moderation effects express by how much the income coefficient is moderated at a given level of social capital index (compared
to index = 0); the effects are calculated for specification 3 from table A2.

It is important to interpret the moderation effect for SC index = 2 and reference income:
it is higher than 100%, meaning that if an individual is sociable and trustful, the impact of
reference income becomes positive, in other words, the “envy effect” turns into the “tunnel
effect”. Last but not least, the 2SLS results hold also after splitting the sample into Western
and Eastern Europe (table A2).

3.3 Robustness check: alternative dataset

A further robustness check consists of implementing an alternative dataset: I employ the last
available wave (round 6) of the European Social Survey (ESS, 2012).18 As already described,
the three key concepts of the research question are: well-being, social capital, and income. The
first is again proxied by reported life satisfaction, measured with the answers to a standard
11-step self-evaluation question: “All things considered, how satisfied are you with your life as
a whole nowadays? Please answer using this card, where 0 means ‘extremely dissatisfied’ and
10 means ‘extremely satisfied’.”19 I employ the same proxies of social capital as in the main
analysis, focusing on its relational aspects: frequency of meeting friends and interpersonal trust.
The question “How often do you meet socially with friends, relatives or work colleagues?” will
be used in order to create a dummy variable equal to 1 if a respondent meets socially at least
once per week.20
The ESS questionnaire asks three questions that are considered as proxies of interpersonal
trust in a broader sense (compared to the previously used single-item question): “Generally
speaking, would you say that most people can be trusted, or that you can’t be too careful in
dealing with people?”; “Do you think that most people would try to take advantage of you if
they got the chance, or would they try to be fair?”; and “Would you say that most of the time
people try to be helpful or that they are mostly looking out for themselves?”. Each question can
be replied on a scale ranging from 0 to 10, corresponding to the lowest and the highest degree of
perceived trustworthiness, fairness, and helpfulness of others. I calculate the arithmetic mean
of the three scores in order to create a dummy variable equal to 1 if the mean ranges between
6 and 10 (the dummy will be called “social trust” to distinguish it form the previously used

18
Round 7 from 2014 has been recently released, however it is still incomplete missing the data from 7 par-
ticipating countries; see http://www.europeansocialsurvey.org/data/country index.html, accessed on February
29, 2016.
19
ESS (2014).
20
The ESS questionnaire explains that ‘Meet socially’ implies meeting by choice rather than for reasons of
either work or pure duty. Originally there are seven possible answers (from ‘Never’ to ‘Every day’), I choose
‘Once a week’ to create the dummy since it divides the sample into two groups of almost the same size (see
table A15).

12
single-item proxy of trust in others). Finally, following the approach from the previous section,
I define an index of social capital as a sum of the two dummies; it will therefore assume values
from the set {0, 1, 2}.
Contrarily to the EU-SILC database, among the ESS variables there is no information about
the exact amount of individual’s income. Instead, the ESS questionnaire asks the respondent
to choose the interval corresponding to his or her household’s total income.21 There are ten
intervals which are adjusted to the income distribution in each country so that they constitute
the income deciles. This implies that I will now use income in relative terms (measured with
income rank) as a proxy of social comparisons.22 I create a categorical variable “income rank”
which will express the individual’s position on the national income ladder. For the sake of
simplicity, it will have three levels: income rank 1-3 (for the bottom three deciles), income rank
4-7 (for the middle four deciles), and income rank 8-10 (for the top three deciles).23
Additional transformations were needed in order to obtain the measure of income in absolute
terms (disposable household monthly income from all sources). I first calculate the average from
lower and upper bounds for each income decile, next, in case of the non-Euro countries I convert
it into Euro, and then adjust for PPP.24 I am aware of the imprecision of the absolute income
proxy, still it is essential to introduce it into the analysis. To avoid further transformation
of the absolute income variable, it is calculated as logarithm of household income (instead of
equivalised income, as previously), however, I will now control for the presence of children in
household.
Since there is a high correlation between the absolute and relative income variables, I in-
troduce two separate regressions, one for interaction with household income (controlling for
income rank, equation 4) and one for interaction with income rank (controlling for household
income, equation 5):

LSi = α + β1 ∗ log(Hh inc i ) + β2 ∗ Inc ranki + β3 ∗ SC index i

+ β13 ∗ SC index i ∗ log(Hh inc i )
+ γ 0 X i + εi (4)

LSi = α + β1 ∗ log(Hh inc i ) + β2 ∗ Inc rank i + β3 ∗ SC index i

+ β23 ∗ SC index i ∗ Inc rank i
+ γ 0 X i + εi (5)

where LS is the reported life satisfaction, Hh inc is the household income, Inc rank is the

21
Defined as “after tax and compulsory deductions, from all sources”; with the possibility to indicate weekly,
monthly or annual income.
22
Theoretical (Rablen, 2008) and empirical (Boyce et al., 2010) studies demonstrate that relative income can
be modelled as the relative rank of an income in the income distribution. I adopt this approach expecting that
when a proxy of purchasing power (absolute income) is included in the regression equation, income rank reflects
solely the relative importance of wealth being rather related to individual’s economic position in the society.
23
The reference category is defined as the middle deciles, so that, after introducing the interactions term, I
can test weather the effect of being relatively poor/rich is moderated by social capital.
24
Exchange rates come from the ESS documentation (ESS, 2014). PPP figures are taken from Eurostat,
they express price level indices for the actual individual consumption (EU28=100); see footnote 5.

13
income rank variable with three categories (bottom deciles, middle deciles - base level, and top
deciles), SC index is a categorical variable measuring the level of social capital, and X includes
the control variables (sex, age, age squared, living with partner, living with children, years of
education, labour market status, country dummies).
The regression sample includes around 35,000 observations coming from 25 European coun-
tries.25 The average life satisfaction is equal to around 6.89, while as the social capital index
is concerned, 26% of the sample scores 0, 47% scores 1, and 27% scores 2 (which means being
trustful and meeting socially at least weekly; see table A15 for detailed descriptive statistics).
The results confirm the finding from the previous paragraph: higher social capital moderates
the impact of income and income comparisons on well-being. Again, the moderation effect
obtained for the proxy of income comparisons (income rank) is stronger than the one obtained
for own income (table 3).26 Moreover, at the highest value of social capital index, the impact of
absolute income is significantly weaker but still positive (figure 3), while the impact of relative
income is eliminated as it becomes insignificant (figure 4).

Table 3: Main results: moderation effects (ESS 2012).

Social capital index = 1 Social capital index = 2

Log of household income -25% -69%
Income rank 1-3 -33% (n/s) -129%
Income rank 8-10 -59% -100%
Note: Moderation effects express by how much the income coefficient is moderated at a given
level of social capital index (compared to index = 0); the effects are calculated for specifications
2 (first row) and 3 (second and third row) from table A5. n/s = not significant.

The analysis performed with the ESS data constitutes an important robustness check of the
EU-SILC results not only because it leads to the same conclusions with the use of an alternative
dataset, but also because the approach to the concept of income comparisons is different. In the
methodology applied in the previous section I used reference income as the proxy variable for
income comparisons, assuming that an individual compares his incomes with others of the same
sex, age and living in the same region, which additionally implies that he makes an evaluation
of how much the others earn.27 In the ESS questionnaire the respondent is asked to choose the
income decile corresponding to his earnings, therefore he can immediately assess his position in
the national income ladder. In this approach the “reference group assumption” is not required,
while the information about the income of others comes directly from the questionnaire (since
the lower and upper bounds of each decile are precisely specified).

25
AL, BE, BG, CH, CY, CZ, DE, DK, EE, ES, FI, FR, GB, HU, IE, IS, IT, LT, NL, NO, PL, PT, SE, SI
and SK. I had to exclude IL, RU, UA and XK since the Eurostat PPP data was missing for these 4 countries.
26
The same result is obtained when interacting the income variables with social capital dummies instead of
the index (see appendix A.1.4).
27
In fact, the psychological literature shows that an individual may choose his reference group instrumentally;
in a motivational strategy, he will compare his outcomes to those of the people above him (self-improvement),
while in order to feel more appreciated he will adapt a self-validation strategy comparing with others below him
(self-enhancement) (Diener and Fujita, 1995; Falk and Knell, 2004). Other studies demonstrate that optimistic
individuals are more likely to compare downward, whereas pessimistic individuals tend to compare with more
successful ones (Lyubomirsky and Ross, 1997).

14
 Figure 3: Absolute income effects moderated by increasing social capital (ESS 2012).
.8

Effects of absolute income on life satisfaction

.6

.4

.2

0
0 1 2
Social capital index (0−2)

Note: Average marginal effects on the linear prediction of life satisfaction with 90% confidence intervals; calculated for specification
2 from table A5.

Figure 4: Relative income effects eliminated for the highest social capital (ESS 2012).
.6
Effects of relative income on life satisfaction

.4

.2

−.2

−.4
0 1 2
Social capital index (0−2)

Income rank: 1−3 Income rank: 8−10

Note: Average marginal effects on the linear prediction of life satisfaction with 90% confidence intervals; calculated for specification
3 from table A5.

3.4 Results: sample of countries

For further cross-sectional investigation I introduce macro-level variables created by aggregating
the micro data at country level. I assume that if social capital moderates the importance of
income for well-being on individual level, one can expect that the well-being gap between income
groups should be smaller in countries with higher social capital.

15
 First, I calculate the average life satisfaction of the individuals defined as “poor” (in the ESS
sample those with household income rank between 1-3, in the EU-SILC those in the bottom
quintile of equivalised income) and for individuals defined as “rich” (income rank 8-10 in the
ESS, top quintile in the EU-SILC). The difference between the two averages will be called “life
satisfaction gap between rich and poor” and will be calculated for each country. Intuitively,
the gap is positive in all countries as the average life satisfaction in the high income group is
always greater than the average life satisfaction in the low income groups. Next, I aggregate
the trust dummy obtaining a macro variable expressing the share of people with high trust
(scores between 6-10): social trust in case of the ESS and trust in others in the EU-SILC.28

Figure 5: Well-being gap between rich and poor decreases with social trust (ESS 2012).
2.5

AL

BG
2
LS gap between rich and poor

LT
CZ
FR HU

SK
1.5 DE
SI
PT EE
PL BE
IE
NL
IT
1 ES
UA IL SE

CY IS
RU GB FI
XK CH
.5 NO
DK
0 .2 .4 .6 .8
Share of people with high social trust (6−10)

Note: Weighted averages.

Figures 5 and 6 present the results graphically.29 In both analysed samples the average score
of trust is negatively correlated with the life satisfaction gap between the rich and poor. This
finding goes in line with the results coming from the micro analysis: as social capital moderates
the individual effects of income on well-being, the life satisfaction discrepancies between the
rich and the poor are smaller in societies more endowed with social capital.
It is plausible to assume, however, that the well-being differences between income classes are
mainly driven by income inequalities: if there are high disparities in income between the rich
and the poor, most likely the life satisfaction gap between the two groups will also be elevated.
What is more, the degree of interpersonal trust is typically higher in developed countries (see
figures 5 and 6). In order to account for this, I perform a regression controlling for the Gini
index and for the level of GDP per capita. The estimated model will have the following form:

28
All the calculated averages are weighted: in the ESS with the design weight (dweight), in the EU-SILC
with the personal cross-sectional weight (RB050 ), as recommended by the data providers in the user’s manuals.
29
In the macro analysis I include the countries previously dropped due to missing data: the ESS sample will
now have 29 countries (including IL, RU, UA and XK), while the EU-SILC sample will consist of 32 countries
(including CZ, DK and SI).

16
Figure 6: Well-being gap between rich and poor decreases with trust in others (EU-SILC 2013).
RS

2.5 BG

HR
LS gap between rich and poor HU EE
2 LT

LV
PT DE SI

1.5 CZ
CY LU
UK
SK IT
FR EL PL BE ES
AT

MT IE RO DK
1
SE FI
IS NO
NL
CH

.5
.2 .4 .6 .8 1
Share of people with high trust in others (6−10)

Note: Weighted averages.

Table 4: Well-being gap between rich and poor: regression results (EU-SILC 2013).

(1) (2) (3) (4) (5) (6)

LS gap LS gap LS gap LS gap LS gap LS gap
Trust in others -2.080∗∗∗ -1.607∗∗∗ -1.713∗∗∗ -1.454∗∗∗
(-4.98) (-3.72) (-3.97) (-3.38)
Gini index 7.589∗∗∗ 3.880∗∗ 2.672
(3.74) (2.18) (1.53)
GDP per capita -0.0227∗∗∗ -0.0137∗ -0.0112
(-3.57) (-1.90) (-1.69)
Number of observations 32 32 32 32 32 32
Adjusted R2 0.474 0.323 0.275 0.525 0.556 0.571
Note: OLS regressions with robust standard errors. Observations = countries. Dependent variable:
the difference in average life satisfaction between the first and fifth income quintile in a given country.
“Trust in others” is the share of people declaring trust in others between 6-10. Gini index is calculated
for equivalised income. GDP per capita is expressed in purchasing power standards (in thousands). All
macro variables are derived using weights. * p< 0.1, ** p< 0.05, *** p< 0.01, t statistics in parentheses.

LSgapk = α + β ∗ Trustk + γ ∗ Ginik + δ ∗ GDPpck + εk (6)

where LSgap stands for the life satisfaction gap between the rich and the poor, Trust is the
share of people with high trust in others (scores 6-10), Gini is the Gini index calculated for
equivalised income, GDP pc is the GDP per capita adjusted to purchasing power parity30 , k
indicates the country (k = 32), and ε is an error term with standard properties.31

30
I use data from Eurostat for “Gross domestic product at market prices” in “Current
prices, PPS per capita”; see http://ec.europa.eu/eurostat/en/web/products-datasets/-/NAMA 10 PC
(last update 15.03.16, extracted 15.03.16). PPS stands for “purchasing power standard”; see
http://ec.europa.eu/eurostat/statistics-explained/index.php/Glossary:Purchasing power standard (PPS)
31
See table A14 for descriptive statistics of the EU-SILC macro variables.

17
 The results show that the negative relationship between social capital and the well-being
gap between income groups is significant also after controlling for income inequalities and the
level of economic development (table 4). Interestingly, the effects of Gini index and GDP per
capita on the life satisfaction gap become insignificant when trust in others is included in the
regression. This suggests that social capital should be considered as one of the main drivers of
happiness inequality.

4 Evidence from panel

In the second part of the empirical analysis I employ panel data in order to: (i) control for
individual unobserved heterogeneity; (ii) investigate a possible causal relationship between
changes in social capital in time and the impact of income on well-being. The sample comes
from the last available version of the German Socio-Economic Panel (GSOEP) released in 2015
(covering period 1984-2013).32 GSOEP is a household based study which reinterviews adult
household members on an annual basis generating a representative sample for the German
population of the last three decades.

4.1 Controlling for fixed effects

In order to control for time invariant unobserved characteristics, e.g. personality traits, I
introduce the individual fixed effects (fi ) within the same methodological framework as in the
EU-SILC analysis.33 The model will now take the following form:

LSi,t = α + β1 ∗ log(Ind inc i,t ) + β2 ∗ log(Ref inc i,t ) + β3 ∗ SC index i,t

+ β13 ∗ SC index i,t ∗ log(Ind inc i,t )
+ β23 ∗ SC index i,t ∗ log(Ref inc i,t )
+ γ 0 Xi,t + fi + εi,t . (7)

Life satisfaction (LS) is measured on a 0-10 scale34 , individual income (Ind inc) is defined as
monthly equivalised disposable income and is adjusted to price level in a given year, while the
vector of controls (X) includes socio-demographic characteristics as well as year and regional
dummies.35 The definition of the reference income does not change: I assume that people com-
pare their incomes with others of the same sex, age group and living in the same geographical
area.36

32
I use the data from version 30 in long format; see https://www.diw.de/en/diw 01.c.504352.en/soep v30.html.
33
In order to decide between fixed or random effects I run the Hausman test; the null hypothesis that
the individual effects are uncorrelated with the other regressors in the model is rejected. Between the two
alternatives, the fixed effects model is therefore the better choice (see Greene, 2008, pp. 301-303).
34
The GSOEP questionnaire uses the following question: “Please answer on a scale from 0 to 10, where 0
means ‘completely dissatisfied’ and 10 means ‘completely satisfied: How satisfied are you with your life, all
things considered?”.
35
Controls include: sex (omitted due to fixed effects), age, age squared, martial status, years of education,
labour market status, house owner, living in East Germany, regional dummies, year dummies.
36
Following the existing studies on income comparisons in Germany (Bartolini et al., 2013b; Ferrer-i Car-
bonell, 2005), for “geographical area” I distinguish between West and East. The variables used to construct

18
 The availability of social capital proxies in the GSOEP questionnaire will influence the
composition of the social capital index. Trust variable is no longer available, however, I will
involve two new proxies of social capital: helping and volunteering.37 The index will be now
defined as a sum of three dummy variables (capturing individual behaviour towards others):
attending social gatherings, helping friends, and performing volunteering work (each is equal to
1 if the respondent carries out a given activity at least once per month)38 . This implies that the
new social capital index will have 4 categories: ranging from 0 (for individuals not performing
any of the activities) to 3 (for those who perform all three activities).
The regression sample is limited to the 12 waves in which all the required variables are
present; it includes around 41,000 individuals interviewed at least two times between 1985-
2011, giving more than 158,000 observations.39 The average life satisfaction in the sample is
equal to 6.97; I observe the following shares of individuals scoring 0, 1, 2 or 3 in the social
capital index: 16%, 36%, 35% and 13%, respectively (for detailed descriptive statistics see
table A16).
Controlling for fixed effects and using different proxies of social capital does not change the
main results: (i) with rising social capital the impact of income and income comparisons on
well-being is significantly moderated (as all the interaction terms are significant, see table A8);
(ii) for the highest level of social capital the effect of individual income is still positive, while
the negative effect of reference income becomes insignificant (figure 7); (iii) the moderation
effects are much stronger in case of reference income (table 5). Additionally, I run separate
regressions for West and East Germany, characterized by different socio-economic background,
showing that the interactions terms are significant also in smaller, more homogeneous samples
(table A8).40

Table 5: Main results: moderation effects (GSOEP 1985-2011).

Social capital index = 1 Social capital index = 2 Social capital index = 3

Log of individual income -18% -28% -47%
Log of reference income -40% -64% -73%
Note: Moderation effects express by how much the income coefficient is moderated at a given level of social capital index (com-
pared to index = 0); the effects are calculated for specification 3 from table A8.

reference groups are the following: sex, age group (below 26, 26-35, 36-45, 46-55, 55+), living in East/West, and
year. This will give 2*5=10 reference groups per year for the 3 years before unification (which do not include
East Germany), and 2*2*5=20 reference groups per year for the 9 years after unification. Altogether it gives
10x3+20x9=210 reference groups for the whole sample. The average size of a reference groups is around 755,
the smallest one contains 140 individuals, the biggest 2852.
37
In fact, the trust question is present in years 2003, 2008 and 2013, however, in these years other proxies of
social capital are not included in the questionnaire.
38
Source variables: Attend Social Gathering (pli0094 ), Helping Relatives, Friends (pli0095 ), Perform Vol-
unteer Work (pli0096 ). The battery of questions on social participation in GSOEP includes also Participate In
Local Politics; I exclude this variable assuming that the motivation behind being involved in political activities
is not necessarily intrinsic.
39
Included years: 1985, 1986, 1988, 1992, 1994, 1996, 1997, 1999, 2005, 2007, 2009 and 2011. The social
participation questions were also asked in 1984, 1990 and 2001, but in 1984 the information on region (NUTS
1, Federal State) is missing, while in 1990 the questions were asked only in the East Germany, whereas for 2001
the volunteering variable is not present.
40
The income variables are also interacted with a given social capital dummy instead of the index, which
gives significant interaction terms of the same signs (results presented in appendix A.1.6).

19
 Figure 7: Income effects moderated by increasing social capital (GSOEP 1985-2011).

.5

Effects of income on life satisfaction

0

−.5

−1
0 1 2 3
Social capital index (0−3)

Log of individual income Log of reference income

Note: Average marginal effects on the linear prediction of life satisfaction with 90% confidence intervals; calculated for specification
3 from table A8.

Figure 8: Reference income less harmful for well-being with higher social capital (GSOEP
1985-2011).
7.2
Life satisfaction (linear prediction)

6.8

6.6

6.4
7.08 7.13 7.18 7.23 7.28 7.33 7.38 7.43 7.48 7.53 7.58 7.63 7.68
Log of reference income

Social capital index = 0 Social capital index = 1

Social capital index = 2 Social capital index = 3

Note: Predictive margins calculated for specification 3 from table A8.

Again, one can conclude that social capital positively affects well-being in two manners: it
is associated with higher well-being (direct effect), and it moderates the detrimental effects
of income comparisons on well-being (indirect effect). Figure 8 illustrates both effects; with
increasing social capital index: the predicted life satisfaction rises and the negative slope of
reference income becomes flatter.

20
 Compared to the results obtained in the cross-sectional analysis, the only difference here is
that at the highest level of social capital the income comparison effects are moderated but not
entirely eliminated (the slope for SC index = 3 is not perfectly flat, as the moderation effect
amounts to -73%). This may suggest that a necessary condition for eliminating the negative
effect of reference income (the “envy effect”) is the high degree of interpersonal trust (a variable
included in the SC index in the EU-SILC and ESS, but not present in the GSOEP analysis due
to data limitations).

4.2 Changes of social capital

In the final step of the empirical analysis I employ the GSOEP data in order to test the following
hypothesis: positive (negative) changes in social capital from yesterday to today are associated
with lower (higher) importance of income and income comparisons for well-being today. I create
a categorical variable capturing positive and negative changes in the social capital index:



 negative change, if SC index t < SC index t−1

∆SC index t = no change, if SC index t = SC index t−1 (8)



positive change, if SC index t > SC index t−1 .

∆SC index t will be implemented into the happiness regression being interacted with individual
and reference income in time t (setting no change as the reference category):

LSi,t = α + β1 ∗ log(Ind inc i,t ) + β2 ∗ log(Ref inc i,t ) + β3 ∗ ∆SC index i,t
+ β13 ∗ ∆SC index i,t ∗ log(Ind inc i,t )
+ β23 ∗ ∆SC index i,t ∗ log(Ref inc i,t )
+ γ 0 Xi,t + fi + εi,t . (9)

This means that each of the two income variables is going to be interacted with a dummy for
negative change and a dummy for positive change.
I run three separate regressions in order to check weather the impact of social capital change
on the importance of income for well-being may be lagged in time: first, I use SC index t as
defined in equation 8, then I create analogical variables for t − 2 and for t − 3.41 There results
can be summarized as follows (see table A9 for estimates):
- a negative change in social capital from yesterday to today is associated with higher
importance of individual income for well-being today (the interaction is significant in the
specifications for t − 2 and t − 3);
- a positive change in SC is associated with lower importance of individual income (signif-
icant for all three lags);

41
The sample is now limited to individuals present in two subsequent waves (t − 1 and t). I end up with
around 119,000 observations, of which ∼30,000 (25%) exhibit a negative change in social capital index, ∼61,000
(51%) exhibit no change, while ∼28,000 (24%) exhibit a positive change. For t−2 the numbers are, respectively,
∼92,000 (sample size), ∼25,000 (27%), ∼43,000 (47%), and ∼24,000 (26%). Finally, for t − 3 I have ∼68,000
(sample size), ∼19,000 (28%), ∼30,000 (44%), and ∼19,000 (28%).

21
 - a negative change in SC reinforces the negative impact of reference income on well-being
(significant for all three lags);
- a positive change in SC moderates the negative impact of reference income (significant
for t − 1).
The above findings suggest that it is plausible to assume the existence of a causal relationship
between the analysed variables: an increase in social capital is likely to cause lower importance
of income and income comparisons for well-being. Importantly, also in the case of social capital
changes, the moderation effects are higher in magnitude for reference income than for absolute
income (table 6).

Table 6: Changes of social capital: moderation effects for t-1 (GSOEP 1985-2011).

Social capital index: negative change Social capital index: positive change
Log of individual income +9% (n/s) -18%
Log of reference income +29% -30%
Note: Moderation effects express by how much the income coefficient is moderated for a given change in social capital index
(compared to Social capital index: no change); the effects are calculated for specification 1 from table A9. n/s = not significant.

5 Conclusions and policy implications

Using data from two European cross-sectional surveys and from the German Socio-Economic
Panel, I show that:
(i) individual income is positively correlated with life satisfaction and this relationship is
significantly weaker for individuals with higher social capital, meaning that they have
lower material interests;
(ii) reference income is negatively correlated with life satisfaction and this relationship is
significantly weaker for individuals with higher social capital, meaning that they are less
affected by income comparisons;
(iii) social capital moderates the importance of income for well-being, while the moderation
effects for reference income are stronger than those for individual income.
Importantly, the above statements hold after accounting for the endogeneity of social capital
and after controlling for the individual unobserved heterogeneity. Additionally, as the panel
data analysis demonstrated, it is reasonable to expect that positive changes in social capital
are likely to reduce the material interests of individuals.
These findings provide an empirical evidence that the two components of SWB usually
analysed separately in happiness economics, the importance of income and the level of social
capital, actually constitute two faces of the same phenomenon, that is, materialism. People
who exhibit higher values of social capital, as they are less materialistic, tend to attach less
importance to money expressed in absolute and relative terms. For individuals who are trustful
and often meet socially, the effects of individual income on life satisfaction are significantly
moderated, while the negative effects of reference income are practically eliminated and become
insignificant. This outcome indicates that in the societies highly endowed with social capital,
economic growth may be followed by happiness growth: if for high levels of trust and sociability

22
the concern for relative income is eliminated, while the absolute income still positively affects
happiness, social capital becomes a necessary condition to make economic growth compatible
with happiness growth. Similar result has been recently obtained by Mikucka and Sarracino
(2014), who demonstrate that economic growth has a positive effect on subjective well-being
in presence of increasing social trust and decreasing income inequality.
Therefore, social capital should be taken into consideration in future studies on the happiness-
income paradox defined by Easterlin (1974, 1995). This additional social dimension of SWB
is crucial for determining the long-run relationship between economic growth and society’s
well-being. Income comparisons play a significantly more important role in life satisfaction
assessments in case of individuals with lower social capital. In consequence, such individuals
are less likely to benefit from economic growth bringing higher standards of living to all. As the
effects of relative income are virtually eliminated at high levels of trust and sociability, while
the positive relationship between SWB and absolute income is maintained, it is plausible to
assume that social capital is a necessary condition for making economic growth consistent with
happiness growth.
The intuition behind the results on micro level is supported by the macro analysis. Social
capital is significantly and negatively correlated with the life satisfaction gap between income
groups. In countries characterized by high degree of interpersonal trust, the differences in av-
erage SWB between the rich and the poor is smaller. That is to say, societies highly endowed
with social capital are more egalitarian in terms of happiness distribution among income groups.
This finding constitutes an important contribution to the growing literature on happiness in-
equality, which, so far, has not considered the role of social capital (see, e.g., Becchetti et al.,
2014; Delhey and Kohler, 2011; Dutta and Foster, 2013).
Clark et al. (2012) argues that in countries which have exhibited positive income growth,
the happiness inequality has decreased, adding that “if raising the income of all does not raise
the happiness of all, it will at least harmonize the happiness of all”. The results obtained in
the macro analysis suggest, instead, that SWB inequality may be reduced by increasing social
capital. Raising social capital of all can raise the happiness of all and it can harmonize the
happiness of the rich and poor. This conclusion, however, should be taken with care as it comes
from cross-sectional data. Further investigation of international happiness times-series should
therefore focus on the relationship between the trends of social capital and the trends of SWB
inequality.
Finally, increases in the level of social capital, apart from having a direct positive impact on
well-being, should be expected to build a society that is less dependent on material goods and
more attached to commonly shared values such as solidarity, equality and civic engagement.
Policies aimed to increase social capital or to reduce materialism could stimulate a mechanism
that lowers people’s interest for positional competition and, at the same time, increases trust
and cooperation among individuals.

23
References
Balli, H. O. and Sørensen, B. E. (2013). Interaction effects in econometrics. Empirical Economics, 45(1):583–603.

Bartolini, S., Bilancini, E., and Pugno, M. (2013a). Did the decline in social connections depress Americans’
happiness? Social Indicators Research, 110(3):1033–1059.

Bartolini, S., Bilancini, E., and Sarracino, F. (2013b). Predicting the trend of well-being in Germany: How
much do comparisons, adaptation and sociability matter? Social Indicators Research, 114(2):169–191.

Bartolini, S. and Bonatti, L. (2002). Environmental and social degradation as the engine of economic growth.
Ecological Economics, 43(1):1–16.

Bartolini, S. and Bonatti, L. (2003). Endogenous growth and negative externalities. Journal of economics,
79(2):123–144.

Bartolini, S. and Bonatti, L. (2008). Endogenous growth, decline in social capital and expansion of market
activities. Journal of Economic Behavior & Organization, 67(3):917–926.

Bartolini, S. and Sarracino, F. (2014). Happy for how long? How social capital and economic growth relate to
happiness over time. Ecological Economics, 108:242–256.

Bartolini, S. and Sarracino, F. (2015). The dark side of Chinese growth: Declining social capital and well-being
in times of economic boom. World Development, 74:333–351.

Baum, C. F., Lewbel, A., Schaffer, M. E., and Talavera, O. (2012). Instrumental variables estimation using
heteroskedasticity-based instruments. In United Kingdom Stata User’s Group Meetings, volume 7.

Becchetti, L., Massari, R., and Naticchioni, P. (2014). The drivers of happiness inequality: Suggestions for
promoting social cohesion. Oxford Economic Papers, 66:419–442.

Becchetti, L., Pelloni, A., and Rossetti, F. (2008). Relational goods, sociability, and happiness. Kyklos,
61(3):343–363.

Becchetti, L., Ricca, E. G., and Pelloni, A. (2012). The relationship between social leisure and life satisfaction:
Causality and policy implications. Social Indicators Research, 108(3):453–490.

Becchetti, L., Trovato, G., and Londono Bedoya, D. A. (2011). Income, relational goods and happiness. Applied
Economics, 43(3):273–290.

Belk, R. W. (1985). Materialism: Trait aspects of living in the material world. Journal of Consumer research,
12:265–280.

Blanchflower, D. G. and Oswald, A. J. (2004). Well-being over time in Britain and the USA. Journal of public
economics, 88(7):1359–1386.

Boschi, M., Girardi, A., and Ventura, M. (2014). Partial credit guarantees and smes financing. Journal of
Financial Stability, 15:182–194.

Boyce, C. J., Brown, G. D., and Moore, S. C. (2010). Money and happiness rank of income, not income, affects
life satisfaction. Psychological Science, 21(4):471–475.

Brambor, T. (2005). Understanding interaction models: Improving empirical analyses. Political Analysis,
14(1):63–82.

Braumoeller, B. (2004). Hypothesis testing and multiplicative interaction terms. International Organization,
58(4):807–820.

Brown, G. D., Gardner, J., Oswald, A. J., and Qian, J. (2008). Does wage rank affect employees well-being?
Industrial Relations: A Journal of Economy and Society, 47(3):355–389.

Brown, T. T. (2014). How effective are public health departments at preventing mortality? Economics &
Human Biology, 13:34–45.

Bruni, L. and Stanca, L. (2008). Watching alone: Relational goods, television and happiness. Journal of
Economic Behavior & Organization, 65(3):506–528.

24
Caporale, G. M., Georgellis, Y., Tsitsianis, N., and Yin, Y. P. (2009). Income and happiness across Europe:
Do reference values matter? Journal of Economic Psychology, 30(1):42–51.

Carlsson, F., Johansson-Stenman, O., and Martinsson, P. (2007). Do you enjoy having more than others?
Survey evidence of positional goods. Economica, 74(296):586–598.

Clark, A., Flèche, S., and Senik, C. (2012). The great happiness moderation. IZA DP No. 6761.

Clark, A., Senik, C., and Yamada, K. (2015). When experienced and decision utility concur: The case of income
comparisons. IZA DP No. 9189.

Clark, A. E., Frijters, P., and Shields, M. A. (2008). Relative income, happiness, and utility: An explanation
for the Easterlin paradox and other puzzles. Journal of Economic Literature, 46(1):95–144.

Clark, A. E. and Oswald, A. J. (1996). Satisfaction and comparison income. Journal of Public Economics,
61(3):359–381.

Clark, A. E. and Senik, C. (2010). Who compares to whom? The anatomy of income comparisons in Europe.
The Economic Journal, 120(544):573–594.

Cohen, P. and Cohen, J. (1996). Life Values and Adolescent Mental Health. Psychology Press, Erlbaum,
Mahwah, NJ.

Deci, E. L. (1971). Effects of externally mediated rewards on intrinsic motivation. Journal of Personality and
Social Psychology, 18(1):105.

Delhey, J. and Kohler, U. (2011). Is happiness inequality immune to income inequality? New evidence through
instrument-effect-corrected standard deviations. Social Science Research, 40(3):742–756.

Denny, K. and Oppedisano, V. (2013). The surprising effect of larger class sizes: Evidence using two identification
strategies. Labour Economics, 23:57–65.

Diener, E. (2006). Guidelines for national indicators of subjective well-being and ill-being. Journal of Happiness
Studies, 7(4):397–404.

Diener, E. and Fujita, F. (1995). Resources, personal strivings, and subjective well-being: A nomothetic and
idiographic approach. Journal of Personality and Social Psychology, 68(5):926.

Dolan, P., Peasgood, T., and White, M. (2008). Do we really know what makes us happy? A review of the
economic literature on the factors associated with subjective well-being. Journal of economic psychology,
29(1):94–122.

Duesenberry, J. S. (1949). Income, Saving, and the Theory of Consumer Behavior. Harvard University Press,
Cambridge and London.

Dutta, I. and Foster, J. (2013). Inequality of happiness in the US: 1972–2010. Review of Income and Wealth,
59(3):393–415.

Easterlin, R. A. (1974). Does economic growth improve the human lot? Some empirical evidence. Nations and
Households in Economic Growth, 89:89–125.

Easterlin, R. A. (1995). Will raising the incomes of all increase the happiness of all? Journal of Economic
Behavior & Organization, 27(1):35–47.

Easterlin, R. A. (2003). Explaining happiness. Proceedings of the National Academy of Sciences, 100(19):11176–
11183.

Emran, M. S. and Shilpi, F. (2012). The extent of the market and stages of agricultural specialization. Canadian
Journal of Economics/Revue canadienne d’économique, 45(3):1125–1153.

ESS (2012). ESS Round 6: European Social Survey Round 6 Data. Data file edition 2.1. Norwegian Social
Science Data Services, Norway - Data archive and distributor of ESS data.

ESS (2014). ESS Round 6: European Social Survey. ESS-6 2012 Documentation Report. Edition 2.1. Bergen,
European Social Survey Data Archive, Norwegian Social Science Data Services.

25
Falk, A. and Knell, M. (2004). Choosing the Joneses: Endogenous goals and reference standards. The Scandi-
navian Journal of Economics, 106(3):417–435.

Ferrer-i Carbonell, A. (2005). Income and well-being: An empirical analysis of the comparison income effect.
Journal of Public Economics, 89(5):997–1019.

Ferrer-i Carbonell, A. and Frijters, P. (2004). How important is methodology for the estimates of the determi-
nants of happiness? The Economic Journal, 114(497):641–659.

Festinger, L. (1954). A theory of social comparison processes. Human relations, 7(2):117–140.

Fliessbach, K., Weber, B., Trautner, P., Dohmen, T., Sunde, U., Elger, C. E., and Falk, A. (2007). Social
comparison affects reward-related brain activity in the human ventral striatum. Science, 318(5854):1305–
1308.

Frey, B. S. and Stutzer, A. (2000). Happiness, economy and institutions. The Economic Journal, 110(466):918–
938.

Gao, W. and Smyth, R. (2015). Education expansion and returns to schooling in urban China, 2001–2010:
Evidence from three waves of the China Urban Labor Survey. Journal of the Asia Pacific Economy, 20(2):178–
201.

Goerke, L. and Pannenberg, M. (2015). Direct evidence for income comparisons and subjective well-being across
reference groups. Economics Letters, 137:95–101.

Greene, W. H. (2008). Econometric Analysis. 5th Edition. Prentice Hall, Upper Saddle River, NJ.

Helliwell, J. F. (2006). Well-being, social capital and public policy: What’s new? The Economic Journal,
116(510):C34–C45.

Helliwell, J. F. and Putnam, R. D. (2004). The social context of well-being. Philosophical Transactions of the
Royal Society B: Biological Sciences, 359:1435–1446.

Hirschman, A. O. and Rothschild, M. (1973). The changing tolerance for income inequality in the course of
economic development. The Quarterly Journal of Economics, 87(4):544–566.

Kasser, T. (2002). The High Price of Materialism. MIT Press, Cambridge, MA.

Kasser, T., Ryan, R. M., Zax, M., and Sameroff, A. J. (1995). The relations of maternal and social environments
to late adolescents’ materialistic and prosocial values. Developmental Psychology, 31(6):907–914.

Kasser, T. and Sheldon, K. M. (2000). Of wealth and death: Materialism, mortality salience, and consumption
behavior. Psychological Science, 11(4):348–351.

Krueger, A. B. and Schkade, D. A. (2008). The reliability of subjective well-being measures. Journal of Public
Economics, 92(8-9):1833–1845. Special Issue: Happiness and Public Economics.

Lewbel, A. (2012). Using heteroscedasticity to identify and estimate mismeasured and endogenous regressor
models. Journal of Business & Economic Statistics, 30(1):67–80.

Lin, F. (2015). Estimating the effect of the internet on international trade. The Journal of International Trade
& Economic Development, 24(3):409–428.

Lyubomirsky, S. and Ross, L. (1997). Hedonic consequences of social comparison: A contrast of happy and
unhappy people. Journal of Personality and Social Psychology, 73(6):1141–1157.

MacKerron, G. (2012). Happiness economics from 35 000 feet. Journal of Economic Surveys, 26(4):705–735.

McBride, M. (2001). Relative-income effects on subjective well-being in the cross-section. Journal of Economic
Behavior & Organization, 45(3):251–278.

McHoskey, J. W. (1999). Machiavellianism, intrinsic versus extrinsic goals, and social interest: A self-
determination theory analysis. Motivation and Emotion, 23(4):267–283.

Mikucka, M. and Sarracino, F. (2014). Making economic growth and well-being compatible: The role of social
capital and income inequality. MPRA Paper No. 59695.

26
Mishra, V. and Smyth, R. (2015). Estimating returns to schooling in urban China using conventional and
heteroskedasticity-based instruments. Economic Modelling, 47:166–173.

OECD (2001). The Well-being of Nations: The Role of Human and Social Capital, chapter 3: The evidence on
social capital, pages 39–63. OECD, Paris.

Onyx, J. and Bullen, P. (2000). Measuring social capital in five communities. The Journal of Applied Behavioral
Science, 36(1):23–42.

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

Rablen, M. D. (2008). Relativity, rank and the utility of income. The Economic Journal, 118(528):801–821.

Ram, R. (2010). Social capital and happiness: Additional cross-country evidence. Journal of Happiness Studies,
11(4):409–418.

Runciman, W. G. (1966). Relative Deprivation and Social Justice: A Study of Attitudes to Social Inequality in
Twentieth-century England. University of California Press.

Sandvik, E., Diener, E., and Seidlitz, L. (1993). Subjective well-being: The convergence and stability of self-
report and non-self-report measures. Journal of Personality, 61(3):317–342.

Sarracino, F. (2010). Social capital and subjective well-being trends: Comparing 11 western European countries.
The Journal of Socio-Economics, 39(4):482–517.

Sarracino, F. (2012). Money, sociability and happiness: Are developed countries doomed to social erosion and
unhappiness? Social Indicators Research, 109(2):135–188.

Sarracino, F. (2013). Determinants of subjective well-being in high and low income countries: Do happiness
equations differ across countries? The Journal of Socio-Economics, 42:51–66.

Schlueter, T., Sievers, S., and Hartmann-Wendels, T. (2015). Bank funding stability, pricing strategies and the
guidance of depositors. Journal of Banking & Finance, 51:43–61.

Schroeter, C., Anders, S., and Carlson, A. (2013). The economics of health and vitamin consumption. Applied
Economic Perspectives and Policy, 35(1):125–149.

Senik, C. (2004). When information dominates comparison: Learning from Russian subjective panel data.
Journal of Public Economics, 88(9):2099–2123.

Senik, C. (2008). Ambition and jealousy: Income interactions in the ‘Old’ Europe versus the ‘New’ Europe and
the United States. Economica, 75(299):495–513.

Sheldon, K. M., Sheldon, M. S., and Osbaldiston, R. (2000). Prosocial values and group assortation. Human
Nature, 11(4):387–404.

Stutzer, A. and Frey, B. S. (2006). Does marriage make people happy, or do happy people get married? The
Journal of Socio-Economics, 35(2):326–347.

Tiefenbach, T. and Holdgrün, P. S. (2015). Happiness through participation in neighborhood associations in

Japan? The impact of loneliness and voluntariness. VOLUNTAS: International Journal of Voluntary and
Nonprofit Organizations, 26(1):69–97.

Tiefenbach, T. and Kohlbacher, F. (2015). Disasters, donations, and tax law changes: Disentangling effects on
subjective well-being by exploiting a natural experiment. Journal of Economic Psychology, 50:94–112.

Van Reekum, C. M., Urry, H. L., Johnstone, T., Thurow, M. E., Frye, C. J., Jackson, C. A., Schaefer, H. S.,
Alexander, A. L., and Davidson, R. J. (2007). Individual differences in amygdala and ventromedial prefrontal
cortex activity are associated with evaluation speed and psychological well-being. Journal of Cognitive
Neuroscience, 19(2):237–248.

Veblen, T. (1899). The Theory of the Leisure Class. Macmillan, George Allen and Unwin.

Williams, G. C., Frankel, R. M., Campbell, T. L., and Deci, E. L. (2000). Research on relationship-centered
care and healthcare outcomes from the rochester biopsychosocial program: A self-determination theory
integration. Families, Systems, & Health, 18(1):79–90.

27
A Appendix: regressions results and descriptive statis-
tics
A.1 Regressions results
A.1.1 EU-SILC: social capital index

Table A1: Main results (EU-SILC 2013).

(1) (2) (3) (4) (5)

All All All West East
Log of individual income 0.375∗∗∗ 0.430∗∗∗ 0.479∗∗∗ 0.405∗∗∗ 0.585∗∗∗
(55.91) (53.34) (25.04) (15.99) (21.27)
Log of reference income -0.0656∗∗ -0.0600∗∗ -0.113∗∗∗ -0.342∗∗∗ -0.427∗∗∗
(-2.18) (-1.99) (-3.22) (-6.55) (-6.06)
Social capital index = 1 * Log of individual income -0.0311∗∗∗ -0.0411∗ -0.0201 -0.0627∗∗
(-6.06) (-1.91) (-0.71) (-1.98)
Social capital index = 2 * Log of individual income -0.0984∗∗∗ -0.201∗∗∗ -0.176∗∗∗ -0.198∗∗∗
(-18.32) (-9.79) (-6.55) (-6.39)
Social capital index = 1 * Log of reference income 0.00933 0.0520 0.0422
(0.43) (1.18) (1.34)
Social capital index = 2 * Log of reference income 0.114∗∗∗ 0.177∗∗∗ 0.175∗∗∗
(5.46) (4.20) (5.64)
Social capital index = 1 1.060∗∗∗ 1.209∗∗∗ 1.203∗∗∗ 0.746∗∗∗ 1.123∗∗∗
(88.77) (40.20) (37.66) (3.12) (31.95)
Social capital index = 2 1.724∗∗∗ 2.307∗∗∗ 2.215∗∗∗ 1.549∗∗∗ 1.914∗∗∗
(143.80) (71.14) (64.30) (6.71) (49.34)
Controls (socio-demographic, country) Yes Yes Yes Yes Yes
Number of observations 324059 324059 324059 207614 116445
Adjusted R2 0.302 0.303 0.303 0.238 0.349
Note: OLS with robust standard errors. Dependent variable: Life satisfaction. Omitted categories: “Social capital index = 0”, “So-
cial capital index = 0 * Log of individual income” and “Social capital index = 0 * Log reference income”. Controls: sex, age group,
marital status, education level, labour market status, house owner, country dummies. * p< 0.1, ** p< 0.05, *** p< 0.01, t statistics
in parentheses.

Table A2: Instrumenting social capital: 2SLS (EU-SILC 2013).

(1) (2) (3) (4) (5)

All All All West East
Log of individual income 0.416∗∗∗ 0.421∗∗∗ 0.878∗∗∗ 0.874∗∗∗ 0.898∗∗∗
(64.87) (60.10) (30.63) (20.88) (22.46)
Log of reference income -0.0531∗ -0.0612∗∗ -0.502∗∗∗ -0.842∗∗∗ -0.727∗∗∗
(-1.77) (-2.06) (-12.39) (-14.33) (-9.46)
Social capital index = 1 (inst.) * Log of individual income -0.0289∗∗∗ -0.480∗∗∗ -0.513∗∗∗ -0.453∗∗∗
(-6.17) (-13.80) (-10.59) (-8.86)
Social capital index = 2 (inst.) * Log of individual income -0.0812∗∗∗ -0.709∗∗∗ -0.761∗∗∗ -0.588∗∗∗
(-15.34) (-20.68) (-15.16) (-12.00)
Social capital index = 1 (inst.) * Log of reference income 0.436∗∗∗ 0.578∗∗∗ 0.422∗∗∗
(12.80) (10.45) (8.48)
Social capital index = 2 (inst.) * Log of reference income 0.610∗∗∗ 0.784∗∗∗ 0.558∗∗∗
(18.14) (13.89) (11.64)
Social capital index = 1 (inst.) 1.244∗∗∗ 1.178∗∗∗ 1.105∗∗∗ 0.269 1.051∗∗∗
(65.41) (43.53) (38.55) (1.48) (31.96)
Social capital index = 2 (inst.) 1.302∗∗∗ 2.208∗∗∗ 2.117∗∗∗ 1.133∗∗∗ 1.848∗∗∗
(42.66) (69.11) (64.16) (6.46) (47.64)
Controls (socio-demographic, country) Yes Yes Yes Yes Yes
Number of observations 324059 324059 324059 207614 116445
Adjusted R2 0.286 0.303 0.301 0.234 0.348
Note: 2SLS regression with heteroskedasticity-based instruments (instrumented variables: “Log of individual income (inst.)”, “Log of refer-
ence income (inst.)”, “Social capital index (inst.) = 1”, “Social capital index (inst.) = 2”, “Social capital index (inst.) = 1 * Log of individual
income”, “Social capital index (inst.) = 1 * Log of reference income”, “Social capital index (inst.) = 2 * Log of individual income”, “Social
capital index (inst.) = 2 * Log of reference income”). Dependent variable: Life satisfaction. Omitted categories: “Social capital index (inst.)
= 0”, “Social capital index (inst.) = 0 * Log of individual income” and “Social capital index (inst.) = 0 * Log reference income”. Controls:
sex, age group, marital status, education level, labour market status, house owner, country dummies. * p< 0.1, ** p< 0.05, *** p< 0.01, t
statistics in parentheses.

28
A.1.2 EU-SILC: social capital dummies

Table A3: Interacting social capital dummies (EU-SILC 2013).

(1) (2)
SC = Getting together with friends SC = Trust in others
Log of individual income 0.477∗∗∗ (33.92) 0.619∗∗∗ (53.17)
Log of reference income -0.0749∗∗ (-2.25) -0.143∗∗∗ (-4.48)
∗∗∗
SC * Log of individual income -0.0940 (-6.05) -0.270∗∗∗ (-20.14)
SC * Log of reference income 0.0301∗ (1.89) 0.208∗∗∗ (14.88)
SC (main effect) Yes Yes
Controls (socio-demographic, country) Yes Yes
Observations 324059 324059
Adjusted R2 0.270 0.279
Note: OLS with robust standard errors. Dependent variable: Life satisfaction. Controls: sex, age group, marital status,
education level, labour market status, house owner, country dummies. * p< 0.1, ** p< 0.05, *** p< 0.01, t statistics in
parentheses.

Table A4: Interacting social capital dummies: moderation effects (EU-SILC 2013).

Getting together with friends Trust in others

Log of individual income -20% -44%
Log of reference income -40% -145%
Note: Moderation effects express by how much the income coefficient is moderated for a
given social capital dummy; the effects are calculated for specifications 1 and 2 from table A3.

Figure A1: Income effects moderated by higher social capital (EU-SILC 2013).

.6 .6

.4 .4
Effects of income on life satisfaction

Effects of income on life satisfaction

.2 .2

0 0

−.2 −.2
otherwise at least once per month low (0−5) high (6−10)
Getting together with friends Trust in others

Log of individual income Log of reference income Log of individual income Log of reference income

Note: Average marginal effects on the linear prediction of life satisfaction with 90% confidence intervals; calculated for specifications
1 and 2 from table A3.

29
A.1.3 ESS: social capital index

Table A5: Main results (ESS 2012).

(1) (2) (3)

Log of household income 0.460∗∗∗ 0.642∗∗∗ 0.458∗∗∗
(11.87) (13.83) (11.84)
Income rank 1-3 -0.157∗∗∗ -0.156∗∗∗ -0.294∗∗∗
(-3.77) (-3.75) (-4.62)
Income rank 8-10 0.148∗∗∗ 0.180∗∗∗ 0.392∗∗∗
(4.48) (5.45) (6.43)
Social capital index = 1 * Log of household income -0.163∗∗∗
(-4.49)
Social capital index = 2 * Log of household income -0.440∗∗∗
(-10.64)
Social capital index = 1 * Income rank 1-3 0.0966
(1.47)
Social capital index = 1 * Income rank 8-10 -0.232∗∗∗
(-3.46)
Social capital index = 2 * Income rank 1-3 0.380∗∗∗
(5.42)
Social capital index = 2 * Income rank 8-10 -0.392∗∗∗
(-5.94)
Social capital index = 1 0.637∗∗∗ 1.779∗∗∗ 0.641∗∗∗
(21.96) (6.67) (15.12)
Social capital index = 2 1.140∗∗∗ 4.372∗∗∗ 1.109∗∗∗
(35.91) (14.00) (24.91)
Controls (socio-demographic, country) Yes Yes Yes
Number of observations 35556 35556 35556
Adjusted R2 0.299 0.302 0.301
Note: OLS with robust standard errors. Dependent variable: Life satisfaction. Omitted categories: “In-
come rank 4-7”, “Social capital index = 0”, “Social capital index = 0 * Log of household income” and
“Social capital index = 0 * Income rank 4-7”. Controls: sex, age, age squared, living with partner, liv-
ing with children, years of education, labour market status, country dummies. * p< 0.1, ** p< 0.05, ***
p< 0.01, t statistics in parentheses.

A.1.4 ESS: social capital dummies

Table A6: Interacting social capital dummies (ESS 2012).

(1) (2) (3) (4)

SC = Meeting socially SC = Meeting socially SC = Social trust SC = Social trust
Log of household income 0.623∗∗∗ (14.52) 0.488∗∗∗ (12.46) 0.570∗∗∗ (13.88) 0.475∗∗∗ (12.19)
Income rank 1-3 -0.160∗∗∗ (-3.80) -0.288∗∗∗ (-5.34) -0.146∗∗∗ (-3.49) -0.226∗∗∗ (-4.61)
Income rank 8-10 0.171∗∗∗ (5.13) 0.291∗∗∗ (6.42) 0.181∗∗∗ (5.45) 0.278∗∗∗ (6.31)
∗∗∗ ∗∗∗
SC * Log of household income -0.237 (-8.10) -0.249 (-8.28)
∗∗∗
SC * Income rank 1-3 0.221 (4.12) 0.213∗∗∗ (4.10)
SC * Income rank 8-10 -0.207∗∗∗ (-4.22) -0.231∗∗∗ (-4.92)
SC (main effect) Yes Yes Yes Yes
Controls (socio-demographic, country) Yes Yes Yes Yes
Number of observations 35556 35556 35556 35556
Adjusted R2 0.283 0.282 0.295 0.294
Note: OLS with robust standard errors. Dependent variable: Life satisfaction. Controls: sex, age, age squared, living with partner, living with children,
years of education, labour market status, country dummies. * p< 0.1, ** p< 0.05, *** p< 0.01, t statistics in parentheses.

30
 Table A7: Interacting social capital dummies: moderation effects (ESS 2012).

Meeting socially Social trust

Log of household income -38% -43%
Income rank 1-3 -77% -94%
Income rank 8-10 -71% -83%
Note: Moderation effects express by how much the income coefficient is
moderated for a given social capital dummy; the effects are calculated
for specifications 1-4 from table A6.

Figure A2: Absolute income effects moderated by higher social capital (ESS 2012).
.7 .7

.6 .6
Effects of absolute income on life satisfaction

Effects of absolute income on life satisfaction

.5 .5

.4 .4

.3 .3

.2 .2
otherwise at least once per week low (0−5) high (6−10)
Meeting socially Social trust

Note: Average marginal effects on the linear prediction of life satisfaction with 90% confidence intervals; calculated for specifications
1 and 3 from table A6.

Figure A3: Relative income effects moderated by higher social capital (ESS 2012).
.4 .4

.2 .2
Effects of relative income on life satisfaction

Effects of relative income on life satisfaction

0 0

−.2 −.2

−.4 −.4
otherwise at least once per week low (0−5) high (6−10)
Meeting socially Social trust

Income rank: 1−3 Income rank: 8−10 Income rank: 1−3 Income rank: 8−10

Note: Average marginal effects on the linear prediction of life satisfaction with 90% confidence intervals; calculated for specifications
2 and 4 from table A6.

31
A.1.5 GSOEP: social capital index

Table A8: Main results (GSOEP 1985-2011).

(1) (2) (3) (4) (5)

All All All West East
Log of individual income 0.389∗∗∗ 0.483∗∗∗ 0.500∗∗∗ 0.460∗∗∗ 0.622∗∗∗
(20.15) (13.52) (13.79) (10.90) (8.98)
Log of reference income -0.358∗∗∗ -0.360∗∗∗ -0.651∗∗∗ -0.726∗∗∗ -1.302∗∗∗
(-3.45) (-3.48) (-4.72) (-4.14) (-4.42)
Social capital index = 1 * Log of individual income -0.0776∗∗ -0.0917∗∗ -0.0787∗ -0.151∗∗
(-2.19) (-2.52) (-1.89) (-1.97)
Social capital index = 2 * Log of individual income -0.116∗∗∗ -0.142∗∗∗ -0.134∗∗∗ -0.159∗
(-3.12) (-3.71) (-3.09) (-1.86)
Social capital index = 3 * Log of individual income -0.205∗∗∗ -0.235∗∗∗ -0.233∗∗∗ -0.235∗∗
(-4.77) (-5.34) (-4.65) (-2.53)
Social capital index = 1 * Log of reference income 0.260∗∗ 0.250∗ 0.287
(2.44) (1.85) (0.97)
Social capital index = 2 * Log of reference income 0.419∗∗∗ 0.390∗∗∗ 0.988∗∗∗
(3.74) (2.78) (3.17)
Social capital index = 3 * Log of reference income 0.477∗∗∗ 0.561∗∗∗ 0.901∗∗
(3.51) (3.37) (2.19)
Social capital index (main effect) Yes Yes Yes Yes Yes
Controls (socio-demographic, region, year) Yes Yes Yes Yes Yes
Individual fixed effects Yes Yes Yes Yes Yes
Number of observations 158587 158587 158587 123876 34711
Number of individuals 40919 40919 40919 32264 9158
R2 within 0.0422 0.0425 0.0427 0.0479 0.0386
R2 between 0.0490 0.0495 0.0500 0.0231 0.0535
R2 overall 0.0458 0.0462 0.0466 0.0266 0.0511
Note: OLS with individual fixed effects (robust standard errors). Dependent variable: Life satisfaction. Omitted categories: “Social
capital index = 0 * Log of individual income” and “Social capital index = 0 * Log reference income”. Each equation includes the
main effect of social capital index (three dummies). Controls: sex (omitted due to fixed effects), age, age squared, marital status,
years of education, labour market status, house owner, living in East Germany, regional dummies, year dummies. * p< 0.1, ** p<
0.05, *** p< 0.01, t statistics in parentheses.

Table A9: Changes of social capital (GSOEP 1985-2011).

(1) (2) (3)

t-1 t-2 t-3
Log of individual income 0.391∗∗∗ (15.96) 0.368∗∗∗ (12.85) 0.352∗∗∗ (10.32)
Log of reference income -0.585∗∗∗ (-4.82) -0.505∗∗∗ (-3.72) -0.531∗∗∗ (-3.45)
∗∗∗
SC index: negative change * Log of individual income 0.0342 (1.23) 0.0863 (2.76) 0.0684∗ (1.78)
SC index: positive change * Log of individual income -0.0702∗∗ (-2.57) -0.0812∗∗ (-2.52) -0.0996∗∗∗ (-2.69)
∗∗ ∗∗∗
SC index: negative change * Log of reference income -0.172 (-2.09) -0.314 (-3.05) -0.287∗∗ (-2.35)
SC index: positive change * Log of reference income 0.175∗∗ (2.13) -0.0199 (-0.20) -0.00645 (-0.05)
Social capital index change (main effect) Yes Yes Yes
Controls (socio-demographic, region, year) Yes Yes Yes
Individual fixed effects Yes Yes Yes
Number of observations 119202 91738 68145
Number of individuals 30723 26136 20503
R2 within 0.0391 0.0399 0.0452
R2 between 0.0392 0.0385 0.0276
R2 overall 0.0369 0.0414 0.0289
Note: OLS with individual fixed effects (robust standard errors). Dependent variable: Life satisfaction. Omitted categories: “Social cap-
ital index: no change * Log of individual income” and “Social capital index: no change * Log reference income”. Each equation includes
the main effect of social capital index changes (two dummies). Controls: sex (omitted due to fixed effects), age, age squared, marital
status, years of education, labour market status, house owner, living in East Germany, regional dummies, year dummies. * p< 0.1, **
p< 0.05, *** p< 0.01, t statistics in parentheses.

32
A.1.6 GSOEP: social capital dummies

Table A10: Interacting social capital dummies (GSOEP 1985-2011).

(1) (2) (3)

SC = Social gathering SC = Helping friends SC = Performing volunteer work
Log of individual income 0.481∗∗∗ (14.90) 0.433∗∗∗ (19.93) 0.413∗∗∗ (19.06)
∗∗∗ ∗∗∗
Log of reference income -0.588 (-4.67) -0.440 (-4.08) -0.451∗∗∗ (-4.16)
∗∗∗ ∗∗∗
SC * Log of individual income -0.122 (-3.99) -0.103 (-4.38) -0.0836∗∗∗ (-3.18)
∗∗∗ ∗∗∗
SC * Log of reference income 0.285 (3.22) 0.195 (2.82) 0.224∗∗∗ (2.79)
SC (main effect) Yes Yes Yes
Controls (socio-demographic, region, year) Yes Yes Yes
Individual fixed effects Yes Yes Yes
Number of observations 158587 158587 158587
Number of individuals 40919 40919 40919
R2 within 0.0424 0.0391 0.0378
R2 between 0.0490 0.0407 0.0385
R2 overall 0.0459 0.0386 0.0369
Note: OLS with individual fixed effects (robust standard errors). Dependent variable: Life satisfaction. Controls: sex (omitted due to fixed effects), age,
age squared, marital status, years of education, labour market status, house owner, living in East Germany, regional dummies, year dummies. * p< 0.1,
** p< 0.05, *** p< 0.01, t statistics in parentheses.

Table A11: Interacting social capital dummies: moderation effects (GSOEP 1985-2011).

Social gathering Helping friends Performing volunteer work

Log of individual income -25% -24% -20%
Log of reference income -48% -44% -50%
Note: Moderation effects express by how much the income coefficient is moderated for a given social capital
dummy; the effects are calculated for specifications 1-3 from table A10.

Figure A4: Income effects moderated by higher social capital (GSOEP 1985-2011).

.5 .5 .5

0 0 0
Effects of income on life satisfaction

Effects of income on life satisfaction

Effects of income on life satisfaction

−.5 −.5 −.5

−1 −1 −1
otherwise at least once per month otherwise at least once per month otherwise at least once per month
Social gathering Helping friends Performing volunteer work

Log of individual income Log of reference income Log of individual income Log of reference income Log of individual income Log of reference income

Note: Average marginal effects on the linear prediction of life satisfaction with 90% confidence intervals; calculated for specifications
1-3 from table A10.

33
A.2 Descriptive statistics
A.2.1 EU-SILC

Table A12: Descriptive statistics (EU-SILC 2013).

Variable Obs Mean Std. Dev. Min Max
Life satisfaction 324059 6.923 2.14 0 10
Individual income (EUR, EU28=100) 324059 1137.251 1256.401 0 110438.8
Log of individual income 324059 6.159 1.905 0 11.612
Reference income (EUR, EU28=100) 324059 1132.845 765.403 1.983 3120.429
Log of reference income 324059 6.309 1.822 1.093 8.046
Getting together with friends 324059 .731 .443 0 1
Trust in others 324059 .575 .494 0 1
Social capital index (0-2) 324059 1.306 .708 0 2
Social capital index = 0 324059 .144 .351 0 1
Social capital index = 1 324059 .405 .491 0 1
Social capital index = 2 324059 .451 .498 0 1
Female 324059 .55 .497 0 1
Under 26 324059 .096 .294 0 1
26-35 324059 .125 .331 0 1
36-45 324059 .168 .374 0 1
46-55 324059 .189 .392 0 1
Above 55 324059 .422 .494 0 1
Single 324059 .245 .43 0 1
Married 324059 .567 .495 0 1
Widowed 324059 .098 .297 0 1
Divorced or separated 324059 .089 .285 0 1
Primary education or no education 324059 .126 .332 0 1
Secondary education 324059 .629 .483 0 1
Tertiary education 324059 .245 .43 0 1
Working 324059 .475 .499 0 1
Unemployed 324059 .076 .264 0 1
Student 324059 .054 .227 0 1
Retired 324059 .279 .449 0 1
Not working 324059 .116 .321 0 1
House owner 324059 .561 .496 0 1
AT 324059 .03 .171 0 1
BE 324059 .03 .171 0 1
BG 324059 .027 .161 0 1
CH 324059 .038 .19 0 1
CY 324059 .031 .173 0 1
DR 324059 .058 .233 0 1
EE 324059 .03 .171 0 1
EL 324059 .042 .2 0 1
ES 324059 .064 .245 0 1
FI 324059 .031 .172 0 1
FR 324059 .045 .207 0 1
HR 324059 .021 .142 0 1
HU 324059 .054 .226 0 1
IE 324059 .018 .134 0 1
IS 324059 .009 .094 0 1
IT 324059 .076 .265 0 1
LT 324059 .024 .154 0 1
LU 324059 .017 .129 0 1
LV 324059 .03 .171 0 1
MT 324059 .021 .142 0 1
NL 324059 .03 .171 0 1
NO 324059 .018 .133 0 1
PL 324059 .067 .25 0 1
PT 324059 .027 .161 0 1
RO 324059 .037 .189 0 1
RS 324059 .035 .185 0 1
SE 324059 .018 .133 0 1
SK 324059 .034 .182 0 1
UK 324059 .039 .193 0 1

34
Table A13: Regions used to construct reference groups (EU-SILC 2013).
Variable Obs Mean Std. Dev. Min Max
AT1 324059 .013 .112 0 1
AT2 324059 .006 .078 0 1
AT3 324059 .011 .106 0 1
BE 324059 .03 .171 0 1
BG3 324059 .014 .117 0 1
BG4 324059 .013 .112 0 1
CH0 324059 .038 .19 0 1
CY0 324059 .031 .173 0 1
DE 324059 .058 .233 0 1
EE0 324059 .03 .171 0 1
EL1 324059 .014 .117 0 1
EL2 324059 .011 .105 0 1
EL3 324059 .013 .112 0 1
EL4 324059 .004 .065 0 1
ES11 324059 .005 .069 0 1
ES12 324059 .002 .045 0 1
ES13 324059 .002 .042 0 1
ES21 324059 .004 .06 0 1
ES22 324059 .002 .048 0 1
ES23 324059 .002 .048 0 1
ES24 324059 .003 .056 0 1
ES30 324059 .006 .075 0 1
ES41 324059 .004 .066 0 1
ES42 324059 .003 .059 0 1
ES43 324059 .003 .052 0 1
ES51 324059 .007 .08 0 1
ES52 324059 .005 .07 0 1
ES53 324059 .002 .046 0 1
ES61 324059 .008 .088 0 1
ES62 324059 .003 .05 0 1
ES63 324059 .001 .024 0 1
ES64 324059 .001 .026 0 1
ES70 324059 .003 .052 0 1
FI19 324059 .008 .089 0 1
FI1B 324059 .008 .091 0 1
FI1C 324059 .006 .08 0 1
FI1D 324059 .008 .087 0 1
FR10 324059 .005 .069 0 1
FR21 324059 .001 .036 0 1
FR22 324059 .002 .041 0 1
FR23 324059 .001 .036 0 1
FR24 324059 .002 .044 0 1
FR25 324059 .001 .034 0 1
FR26 324059 .001 .038 0 1
FR30 324059 .004 .06 0 1
FR41 324059 .002 .045 0 1
FR42 324059 .001 .037 0 1
FR43 324059 .001 .028 0 1
FR51 324059 .003 .058 0 1
FR52 324059 .003 .052 0 1
FR53 324059 .002 .04 0 1
FR61 324059 .003 .053 0 1
FR62 324059 .002 .045 0 1
FR63 324059 .001 .026 0 1
FR71 324059 .004 .066 0 1
FR72 324059 .001 .033 0 1
FR81 324059 .002 .044 0 1
FR82 324059 .003 .053 0 1
FR83 324059 0 .011 0 1
HR0 324059 .021 .142 0 1
HU1 324059 .011 .106 0 1
HU2 324059 .014 .116 0 1
HU3 324059 .029 .168 0 1
IE0 324059 .018 .134 0 1
IS 324059 .009 .094 0 1
ITC 324059 .018 .134 0 1
ITF 324059 .016 .126 0 1
ITG 324059 .005 .073 0 1
ITH 324059 .019 .137 0 1
ITI 324059 .017 .131 0 1
LT0 324059 .024 .154 0 1
LU0 324059 .017 .129 0 1
LV0 324059 .03 .171 0 1
MT0 324059 .021 .142 0 1
NL 324059 .03 .171 0 1
NO0 324059 .018 .133 0 1
PL1 324059 .014 .119 0 1
PL2 324059 .012 .11 0 1
PL3 324059 .014 .118 0 1
PL4 324059 .01 .101 0 1
PL5 324059 .007 .081 0 1
PL6 324059 .009 .096 0 1
PT 324059 .027 .161 0 1
RO1 324059 .01 .101 0 1
RO2 324059 .01 .101 0 1
RO3 324059 .008 .091 0 1
RO4 324059 .008 .088 0 1
RS 324059 .035 .185 0 1
SE1 324059 .007 .084 0 1
SE2 324059 .008 .088 0 1
SE3 324059 .003 .056 0 1
SK0 324059 .034 .182 0 1
UKC 324059 .001 .038 0 1
UKD 324059 .004 .062 0 1
UKE 324059 .003 .052 0 1
UKF 324059 .002 .049 0 1
UKG 324059 .003 .055 0 1
UKH 324059 .003 .057 0 1
UKI 324059 .003 .055 0 1
UKJ 324059 .005 .07 0 1
UKK 324059 .003 .054 0 1
UKL 324059 .002 .042 0 1
UKM 324059 .006 .075 0 1
UKN 324059 .004 .061 0 1

35
 Table A14: Descriptive statistics: macro variables (EU-SILC 2013).
Variable Obs Mean Std. Dev. Min Max
LS gap between rich and poor 32 1.41 .498 .652 2.657
Trust in others (share of people with scores 6-10) 32 .59 .168 .275 .899
Gini index (for equivalised income) 32 .294 .039 .23 .367
GDP per capita (current prices, PPS per capita, in thousands) 32 27.05 12.005 10.1 70.5

A.2.2 ESS

Table A15: Descriptive statistics (ESS 2012).

Variable Obs Mean Std. Dev. Min Max
Life satisfaction 35556 6.885 2.369 0 10
Household income (EUR, EU28=100) 35556 1974.286 1490.118 140.523 11781.15
Log of household income 35556 7.287 .836 4.945 9.374
Income rank 1-3 35556 .355 .478 0 1
Income rank 4-7 35556 .399 .49 0 1
Income rank 8-10 35556 .247 .431 0 1
Social trust 35556 .43 .495 0 1
Meeting socially 35556 .579 .494 0 1
Social capital index (0-2) 35556 1.009 .729 0 2
Social capital index = 0 35556 .261 .439 0 1
Social capital index = 1 35556 .469 .499 0 1
Social capital index = 2 35556 .27 .444 0 1
Female 35556 .527 .499 0 1
Age 35556 49.322 18.056 15 103
Age squared (divided by 100) 35556 27.587 18.328 2.25 106.09
Years of education 35556 12.654 4.126 0 51
Working 35556 .492 .5 0 1
Unemployed 35556 .075 .264 0 1
Student 35556 .069 .254 0 1
Retired 35556 .241 .428 0 1
Not working 35556 .122 .327 0 1
Living with partner 35556 .605 .489 0 1
Living with children 35556 .378 .485 0 1
AL 35556 .031 .172 0 1
BE 35556 .048 .213 0 1
BG 35556 .054 .227 0 1
CH 35556 .034 .182 0 1
CY 35556 .025 .156 0 1
CZ 35556 .037 .188 0 1
DE 35556 .071 .257 0 1
DK 35556 .033 .179 0 1
EE 35556 .055 .228 0 1
ES 35556 .044 .205 0 1
FI 35556 .058 .233 0 1
FR 35556 .05 .218 0 1
GB 35556 .05 .217 0 1
HU 35556 .039 .194 0 1
IE 35556 .054 .226 0 1
IS 35556 .017 .13 0 1
IT 35556 .015 .122 0 1
LT 35556 .048 .213 0 1
NL 35556 .044 .205 0 1
NO 35556 .044 .205 0 1
PL 35556 .041 .199 0 1
PT 35556 .028 .166 0 1
SE 35556 .047 .211 0 1
SI 35556 .026 .158 0 1
SK 35556 .003 .053 0 1

36
A.2.3 GSOEP

Table A16: Descriptive statistics (GSOEP 1985-2011).

Variable Obs Mean Std. Dev. Min Max
Life satisfaction 158587 6.967 1.826 0 10
Individual income (2011 EUR) 158587 1682.892 1017.249 0 44728.43
Log of individual income 158587 7.311 .479 0 10.708
Reference income (2011 EUR) 158587 1677.146 251.864 1192.22 2243.969
Log of reference income 158587 7.414 .147 7.084 7.716
Social gathering 158587 .776 .417 0 1
Helping friends 158587 .399 .49 0 1
Performing volunteer work 158587 .277 .447 0 1
Social capital index (0-3) 158587 1.451 .913 0 3
Social capital index = 0 158587 .16 .366 0 1
Social capital index = 1 158587 .362 .481 0 1
Social capital index = 2 158587 .345 .475 0 1
Social capital index = 3 158587 .133 .34 0 1
Female 158587 .517 .5 0 1
Age 158587 46.891 17.219 16 101
Age squared (divided by 100) 158587 24.953 17.176 2.56 102.01
Single 158587 .22 .414 0 1
Married 158587 .64 .48 0 1
Widowed 158587 .066 .247 0 1
Divorced or separated 158587 .075 .264 0 1
Years of education 158587 11.592 2.638 7 18
Working 158587 .585 .493 0 1
Unemployed 158587 .054 .225 0 1
Student 158587 .029 .168 0 1
Retired 158587 .161 .367 0 1
Not working 158587 .172 .377 0 1
House owner 158587 .474 .499 0 1
East Germany 158587 .219 .413 0 1
Baden-Wuerttemberg 158587 .136 .342 0 1
Bavaria 158587 .142 .349 0 1
Berlin 158587 .038 .19 0 1
Brandenburg 158587 .037 .189 0 1
Bremen 158587 .008 .087 0 1
Hamburg 158587 .015 .122 0 1
Hesse 158587 .075 .264 0 1
Mecklenburg-Western Pomeran 158587 .022 .148 0 1
Lower Saxony 158587 .089 .284 0 1
North Rhine-Westphalia 158587 .209 .407 0 1
Rhineland-Palatinate 158587 .053 .225 0 1
Saarland 158587 .005 .072 0 1
Saxony 158587 .066 .248 0 1
Saxony-Anhalt 158587 .039 .194 0 1
Schleswig-Holstein 158587 .027 .162 0 1
Thuringia 158587 .04 .196 0 1
1985 158587 .065 .246 0 1
1986 158587 .057 .232 0 1
1988 158587 .055 .227 0 1
1992 158587 .073 .261 0 1
1994 158587 .07 .255 0 1
1996 158587 .07 .256 0 1
1997 158587 .069 .254 0 1
1999 158587 .081 .273 0 1
2005 158587 .119 .323 0 1
2007 158587 .119 .323 0 1
2009 158587 .117 .321 0 1
2011 158587 .106 .307 0 1

37
B Appendix: methodology
B.1 Interpreting the interaction term between a categorical and a
continuous variable (EU-SILC specification)
Interacting a categorical variable with a continuous variable means that each category of the
categorical one (treated as a dummy) is interacted with the continuous one. In case of equation
1 from section 3.1 (where SC index has three categories, each treated as a dummy variable:
SC index =0 - reference category, SC index =1 and SC index =2 ), the estimated model takes
the following form:

LSi = α0 + α1 ∗ log(Ind inc i ) + α2 ∗ log(Ref inc i ) + α3 ∗ (SC index =1) i + α4 ∗ (SC index =2) i
+ α13 ∗ (SC index =1) i ∗ log(Ind inc i ) + α14 ∗ (SC index =2) i ∗ log(Ind inc i )
+ α23 ∗ (SC index =1) i ∗ log(Ref inc i ) + α24 ∗ (SC index =2) i ∗ log(Ref inc i )
+ γ 0 X i + εi . (10)

The equation may be rearranged into:

LSi = α0 + [α1 + α13 ∗ (SC index =1) i + α14 ∗ (SC index =2) i ] ∗ log(Ind inc i )
+ [α2 + α23 ∗ (SC index =1) i + α24 ∗ (SC index =2) i ] ∗ log(Ref inc i )
+ α3 ∗ (SC index =1) i + α4 ∗ (SC index =2) i
+ γ 0 X i + εi . (11)

The marginal effect of log(Ind inc) on LS changes with the level of SC index and will be equal
to the expression:

∂LS
= α1 + α13 ∗ (SC index =1) + α14 ∗ (SC index =2)
∂log(Ind inc)

α1 ,

 for (SC index =0 ) = 1

= α1 + α13 , for (SC index =1 ) = 1 (12)



α + α , for (SC index =2 ) = 1.
1 14

Analogically, the marginal effect of log(Ref inc) on LS will be equal to:

∂LS
= α2 + α23 ∗ (SC index =1) + α24 ∗ (SC index =2)
∂log(Ref inc)

α, for (SC index =0 ) = 1
 2


= α2 + α23 , for (SC index =1 ) = 1 (13)



α + α , for (SC index =2 ) = 1.
2 24

38
B.2 Instrumental variables estimation using heteroskedasticity-based
instruments (EU-SILC specification)
Social capital index is a categorical variable, thus it is necessary to instrument each category
(dummy) and the interactions with each category; the list of the instrumented variables is thus
the following:
- SC index =1 ,
- SC index =2 ,
- (SC index =1 ) ∗ log(Ind inc),
- (SC index =2 ) ∗ log(Ind inc),
- (SC index =1 ) ∗ log(Ref inc),
- (SC index =2 ) ∗ log(Ref inc).
First, I regress each of the above mentioned endogenous variables on the vector of controls X
from equation 1:

(SC index =1 )i = a1 + B01 Xi + 1,i (14)

(SC index =2 )i = a2 + B02 Xi + 2,i (15)
(SC index =1 )i ∗ log(Ind inc)i = a3 + B03 Xi + 3,i (16)
(SC index =2 )i ∗ log(Ind inc)i = a4 + B04 Xi + 4,i (17)
(SC index =1 )i ∗ log(Ref inc)i = a5 + B05 Xi + 5,i (18)
(SC index =2 )i ∗ log(Ref inc)i = a6 + B06 Xi + 6,i . (19)

Second, I generate the instruments by multiplying the residuals from the “first stage equation”
with each of the control variables in mean-centred form:

Z1,j = (Xj − X j ) ∗ ˆ1 (20)

Z2,j = (Xj − X j ) ∗ ˆ2 (21)
Z3,j = (Xj − X j ) ∗ ˆ3 (22)
Z4,j = (Xj − X j ) ∗ ˆ4 (23)
Z5,j = (Xj − X j ) ∗ ˆ5 (24)
Z6,j = (Xj − X j ) ∗ ˆ6 (25)

where j stands for the given control variable from vector X. There are 41 control dummy
variables (base levels not counted, see table A12), therefore j = 1, 2, ..., 41. Each of the 6
endogenous variables will thus have 41 instruments generated in the above described manner.

39
B.3 Instrumental variables estimation using heteroskedasticity-based
instruments (Lewbel, 2012)
“Consider Y1 , Y2 as observed endogenous variables, X a vector of observed exogenous regressors,
and ε = (ε1 , ε2 ) as unobserved error processes. Consider a structural model of the form:

Y1 = X 0 β1 + Y2 γ1 + ε1 (26)
Y2 = X 0 β2 + Y1 γ2 + ε2 . (27)

This system is triangular when γ2 = 0 (or, with renumbering, when γ1 = 0). Otherwise, it
is fully simultaneous. The errors ε1 , ε2 may be correlated with each other. If the exogeneity
assumption, E(εX) = 0 holds, the reduced form is identified, but in the absence of identifying
restrictions, the structural parameters are not identified. These restrictions often involve set-
ting certain elements of β1 or β2 to zero, which makes instruments available. In many applied
contexts, the third assumption made for the validity of an instrument - that it only indirectly
affects the response variable - is difficult to establish. The zero restriction on its coefficient
may not be plausible. The assumption is readily testable, but if it does not hold, IV estimates
will be inconsistent. Identification in Lewbel’s approach is achieved by restricting correlations
of εε0 X with X. This relies upon higher moments, and is likely to be less reliable than identi-
fication based on coefficient zero restrictions. However, in the absence of plausible identifying
restrictions, this approach may be the only reasonable strategy. The parameters of the struc-
tural model will remain unidentified under the standard homoskedasticity assumption: that
E(εε0 | X) is a matrix of constants. However, in the presence of heteroskedasticity related to
at least some elements of X, identification can be achieved. In a fully simultaneous system,
assuming that cov(X, ε2j ) 6= 0, j = 1, 2 and cov(Z, ε1 ε2 ) = 0 for observed Z will identify the
structural parameters. Note that Z may be a subset of X, so no information outside the model
specified above is required. The key assumption that cov(Z, ε1 ε2 ) = 0 will automatically be
satisfied if the mean zero error processes are conditionally independent: ε1 ⊥ ε2 | Z = 0.
However, this independence is not strictly necessary. In the most straightforward context, we
want to apply the instrumental variables approach to a single equation, but lack appropriate
instruments or identifying restrictions. The auxiliary equation or ‘first-stage’ regression may be
used to provide the necessary components for Lewbel’s method. In the simplest version of this
approach, generated instruments can be constructed from the auxiliary equations’ residuals,
multiplied by each of the included exogenous variables in mean-centered form:

Zj = (Xj − X) ∗  (28)

where  is the vector of residuals from the ‘first-stage regression’ of each endogenous regressor
on all exogenous regressors, including a constant vector.”42

42
Source: Baum et al. (2012).

40