Munich Personal RePEc Archive

Economic growth and well-being beyond

the Easterlin paradox

Sarracino, Francesco and O’Connor, Kelsey J.

STATEC Research, Laboratory for Comparative Social Research,

Higher School of Economics

13 September 2019

Online at https://mpra.ub.uni-muenchen.de/96013/

MPRA Paper No. 96013, posted 14 Sep 2019 15:56 UTC

E onomi growth and well-being beyond

the Easterlin paradox

Fran es o Sarra ino

�

and Kelsey J. O'Connor

y

September 13, 2019

Abstra t

Re ent studies suggest that e onomi growth and well-being an grow together

in the long run in presen e of generous so ial safety nets, in reasing so ial apital

and de lining in ome inequality. We put these onditions to a test in an attempt

to explain the absen e of a relation between e onomi growth and well-being in

Luxembourg. To this aim we apply an error orre tion model to a panel of 15

Western European ountries, and we use the results to predi t life satisfa tion in

Luxembourg between 1991 and 2015. We �nd that the at trend of life satisfa tion

in Luxembourg is likely the result of four for es a ting in opposite dire tions. This

suggests that the available list of moderating onditions { although not exhaustive

{ is a promising starting point to design new poli ies to durably improve well-being.

Key-words: time-series; subje tive well-being; error orre tion model; life satis-

fa tion; dynami s; in lusive growth.

JEL odes: I31; O11; E6; O21; D60.

�

Institut national de la statistique et des �etudes �e onomiques du Grand-Du h�e du Luxem-

bourg (STATEC), and LCSR National Resear h University Higher S hool of E onomi s, Russian

Federation. Email: Fran es o.Sarra ino�state .etat.lu.

y

Institut national de la statistique et des �etudes �e onomiques du Grand-Du h�e du Luxembourg

(STATEC). Email: Kelsey.OConnor�state .etat.lu.

1

1 Introdu tion

The �nding of no long-run relationship between e onomi growth and subje tive

well-being is ontroversial. This result, ontradi ting a positive ross-se tional

relation, ame to be known as the Easterlin paradox (Easterlin, 1974). Sin e, many

s holars have ontributed with diverging views. There are those who oppose (see

e.g. Easterlin et al., 2010; Bruni and Stan a, 2008; Easterlin and Angeles u, 2009;

Be hetti et al., 2011; Clark et al., 2014; Easterlin, 2017) and those who support

(see e.g. Stevenson and Wolfers, 2008; Deaton, 2008; Sa ks et al., 2012; Veenhoven

and Vergunst, 2013) e onomi growth as a way to improve well-being. Beyond

the views on erning simply whether or not growth will a ompany in reasing

well-being, some re ent eviden e suggests the relation depends on so ial, politi al,

e onomi , ultural, and institutional onditions: if e onomi growth is ompatible

with a ohesive and in lusive so iety, it is reasonable to expe t that well-being

will improve (Oishi and Kesebir, 2015; Miku ka et al., 2017; Easterlin, 2013; Ono

and Lee, 2013). In ontrast, if e onomi growth leads to loneliness and inequality,

well-being may arguably de line.

Although the quest for onditions of \in lusive growth" { a growth that bene�ts

all the members of a so iety { is still in its infan y, past eviden e provides a

preliminary explanation of how and when a positive orrelation between e onomi

growth and well-being an exist over time. This is important be ause it an suggest

ways to promote well-being. Our aim is to distill the eviden e on the onditions

a�e ting the e onomi growth - well-being gradient, to explain the at trend of

life satisfa tion in Luxembourg.

We fo us on Luxembourg as an example of the la k of orrelation between

e onomi growth and well-being in the long run. Panel 1a in Figure 1 shows that

sin e the early 1980s, Luxembourg experien ed substantial e onomi growth, at

least until the e onomi risis of 2008. Yet, the share of very satis�ed people

did not hange substantially over time.

1

The e onomi risis may explain what

happened after 2008, but prior to 2008, it is not lear why e onomi growth did

not improve people's well-being (from a traditional e onomi s perspe tive).

A possible explanation is that life satisfa tion is an unreliable measure. Yet, we

have reasons to believe that life satisfa tion is reliable based on available data and

previous literature. Panel 1b reports the share of very satis�ed people a ording to

Eurobarometer (EB), European Values Survey (EVS) and European Quality of Life

Survey (EQLS). For the years when the data are jointly available, the three surveys

1

Eurobarometer is the only dataset that provides long time-series about life satisfa tion in

Luxembourg. The answers to the question are organized on a four point s ale. The distribution

of this variable over time is remarkably stable with a onsistently fat right tail. Hen e, our

measure of life satisfa tion, the share of very satis�ed people, is onservative be ause the trend

would be even atter than observed if we fo used on the share of satis�ed people.

2

Figure 1: Share of very satis�ed people (Panel A) and real Gross National In ome

(GNI) per apita (Panel B) in Luxembourg in the period 1981-2015. The samples

onsist only of native born individuals.

40

00

06

00

00

80

00

01

00

00

01

20

00

01

40

00

0G

NI

pe

r ca

pita

(co

nsta

nt

20

10

US

$)

25

35

45

55

65

Sh

are

of

ve

ry s

atisfie

d p

eo

ple

(%

)

1980 1990 2000 2010 2020

% very satisfied GNI per capita (constant 2010 US$)

(a) Life satisfa tion (Eurobarometer data) and real Gross National

In ome per apita (World Development Indi ators).

25

35

45

55

65

Sh

are

of

ve

ry s

atisfie

d p

eo

ple

(%

)

1980 1990 2000 2010 2020

EB EVS EQLS

(b) Life satisfa tion in Luxembourg. For the years when three

di�erent data-set are jointly available, they provide omparable

information.

Note: We restri t our analysis on natives only to ensure the omparability of data on life

satisfa tion over time. Eurobarometer data olle ted before 1994 provided data on nationals

only, and after 1994 they in luded information on immigrants from other European ountries.

We dis uss this issue in more detail in se tion Data.

Sour e: Eurobarometer, European Values Study, European Quality of Life Survey data, and

World Development Indi ators, own elaboration.

3

provide a remarkably omparable pi ture. Moreover, a well-established literature

provided eviden e supporting the reliability and validity of life satisfa tion as a

measure of subje tive and obje tive well-being (Blan h ower and Oswald, 2004;

Van Reekum et al., 2007; S himma k et al., 2010; Kahneman and Krueger, 2006;

Layard, 2005).

Another possible explanation is that the trends of life satisfa tion are always at

- at least among the ri hest and most developed ountries in the world. However,

the eviden e does not support this view. Figure 2 shows the trends of the share of

very satis�ed people in Belgium, Fran e, Germany and the Netherlands, i.e. a set

of Western European ountries whi h are lose to Luxembourg. The pi ture shows

that the trends of life satisfa tion are not always at: although average levels may

di�er, the trends in Fran e and Netherlands are monotoni ally positive, whereas

the trend is rather at in Belgium and it follows a `J' urve in Germany. This is

onsistent with previous studies (Sarra ino, 2012).

Figure 2: Trends of the share of very satis�ed people in a sample of Western

European ountries.

10

20

30

40

50

10

20

30

40

50

1980 1990 2000 2010 20201980 1990 2000 2010 2020

Belgium France

Germany Netherlands

Share

of very

satisfied p

eople

(%

)

Graphs by Country

Sour e: Eurobarometer data, own elaboration.

In sum, the at trend of life satisfa tion in Luxembourg does not have a simple

4

explanation. We posit that e onomi growth and life satisfa tion did not grow

together be ause four fa tors a ted in opposite dire tions for well-being, namely

in reasing so ial apital and so ial expenditures { whi h are expe ted to have a

positive impa t on well-being { and in reasing in ome inequality and unemploy-

ment { whi h, on the other hand, ould have a negative impa t.

Unfortunately, we do not have mi ro data providing long time-series for Lux-

embourg: the European Value Study in ludes individual data olle ted in 1999

and 2008, the European So ial Survey was administered in 2002 and 2004, and

the European Quality of Life Survey provides data every four years sin e 2003.

Thus, we adopt a ma ro-e onomi perspe tive. Spe i� ally, we apply an error or-

re tion model to a panel of 15 Western European ountries to explain ountry-year

levels of life satisfa tion using the set of potential explanatory fa tors identi�ed in

previous literature. The results are then used to predi t life satisfa tion and to

assess whether and to what extent the explanatory fa tors explain the trend of life

satisfa tion in Luxembourg.

We build our argument in two steps: in se tion 2 we review the literature

on the Easterlin paradox and on the fa tors moderating the relationship between

e onomi growth and well-being. Subsequently, we apply a time-series approa h

(se tion 3) to a ma ro data-set onsisting of 15 Western European ountries (se -

tion 4). Se tion 5 illustrates the results of the model, whi h we use to predi t and

explain the trend of life satisfa tion in Luxembourg. The last se tion on ludes.

2 The Easterlin paradox and moderating fa tors

The debate on subje tive well-being gained spe ial attention in part be ause it on-

erns an important question: to what extent do modern so ieties bene�t from e o-

nomi growth? For years this question has divided so ial s ientists among: those

who laim that ontemporary so ieties should not expe t signi� ant improvements

in subje tive well-being from e onomi growth (Easterlin, 1974); those who argue

that e onomi growth and in reasing subje tive well-being are asso iated over time

(see e.g. Stevenson and Wolfers, 2008; Deaton, 2008; Sa ks et al., 2012; Veenhoven

and Vergunst, 2013); those who point out that whether a relationship exists de-

pends on the set of ountries onsidered (developed and developing ountries vs.

transition ountries) or the period of time, i.e. e onomi growth and the trends of

well-being orrelate in the short run, but su h orrelation disappears in the long

run (Easterlin and Angeles u, 2009; Be hetti et al., 2011; Easterlin et al., 2010;

Clark et al., 2014; De Neve et al., 2018; Bartolini and Sarra ino, 2014); and those

who laim that even if the trends of subje tive well-being and e onomi growth

are statisti ally related, the magnitude is too small for growth to have a mean-

ingful impa t (Beja, 2014). Re ently, some s holars argued that the question is

5

not whether, but when { under what onditions { e onomi growth orrelates with

in reasing subje tive well-being (Oishi and Kesebir, 2015; Miku ka et al., 2017).

The literature identi�ed three fa tors whi h plausibly a�e t the relation between

e onomi growth and well-being over time: in ome inequality (Oishi and Kesebir,

2015; Miku ka et al., 2017), so ial apital (Uhlaner, 1989; Helliwell, 2003, 2008;

Bartolini et al., 2013; Clark et al., 2014), and so ial poli y (Easterlin, 2013; Ono

and Lee, 2016).

Con erning in ome inequality, the eviden e about the ross-se tional relation-

ship with well-being is mixed (e.g. Alesina et al., 2004; Clark and D'Ambrosio,

2015). These ontradi tions may arise be ause the relationship between inequality

and well-being depends on a ountry's level of development (Jiang et al., 2012;

Iniguez-Montiel, 2014). However, previous studies found that in reasing in ome

inequality is onsistently negatively related to well-being (Bartolini and Sarra ino,

2015; Oishi and Kesebir, 2015; Miku ka et al., 2017). By widening the possibil-

ities to establish so ial omparisons, growing in ome inequality undermines the

positive e�e t of in ome growth for well-being. Raising in ome inequality an also

undermine well-being by redu ing feelings of fairness and trust in others (Oishi

et al., 2011) or by weakening so ial linkages and feelings of ooperation (Graham

and Felton, 2006; Oishi et al., 2011).

So ial apital is de�ned by the OECD (2001) as \networks together with shared

norms, values and understandings that fa ilitate o-operation within or among

groups". A well-established literature shows that so ial apital orrelates pos-

itively with subje tive well-being at both the individual (Helliwell et al., 2017;

Clark et al., 2014; Be hetti et al., 2009) and aggregate level, over time within

ountries (Bartolini et al., 2013; Bartolini and Sarra ino, 2015; Bro kmann et al.,

2009; Easterlin et al., 2012) and in ountry panels (Bartolini and Sarra ino, 2014).

Helliwell and Aknin (2018) dis uss in detail the relationship between so ial apital

and subje tive well-being.

The experien e of ountries that transitioned from ommunist e onomi sys-

tems illustrate the importan e of so ial safety nets for well-being (Ono and Lee,

2013). Survey data onsistently indi ate that people in European post- ommunist

ountries are among the least satis�ed people in Europe. Moreover, after the tran-

sition, average life satisfa tion de lined. The loss of jobs and the deterioration of

safety nets are among the auses that explain this de line. The ommunist regime

provided people with jobs, basi in ome, health insuran e, edu ation, and other

bene�ts. The transition to market apitalism was a ompanied by widespread

orruption and the ollapse of the so ial insuran e system, whi h invariably led

to greater inequality and lower well-being. In re ent years life satisfa tion re ov-

ered, but it took more than ten years and required an in rease in GDP per apita

6

averaging about 25 per ent above the 1990s value (Easterlin, 2009, p. 142).

2

In China, life satisfa tion exhibited a similar pattern of ollapse and re overy

following the transition, all the while growing at an average annual rate of more

than 9.0%. Bro kmann et al. (2009), Easterlin et al. (2012), Easterlin et al. (2017)

and Bartolini and Sarra ino (2015) dis uss possible explanations for these startling

fa ts. Ea h work partially attributes the de line in life satisfa tion to in reased

so ial omparisons, espe ially fa ilitated by rising in ome inequality. Bartolini

and Sarra ino (2015) do ument the importan e of so ial apital, estimating that

nearly 19.0% of the well-being loss in China is related to a de rease in so ial

apital. Easterlin et al. (2012) and Easterlin et al. (2017) instead emphasize the

role of rising unemployment

3

, whi h was inversely related to life satisfa tion over

the full y le from 1990 to 2010 (while inequality, in ontrast, rose throughout

the period). And, like in the European transition ountries, with unemployment

ame not only in ome losses, but also the elimination of so ial bene�ts. The loss

of these bene�ts arguably signi� antly exa erbated the e�e ts of unemployment.

So ial safety nets are positively related to life satisfa tion in general (Di Tella et al.,

2003; Rothstein, 2010; Pa ek and Rad li�, 2008; Boarini et al., 2013; Easterlin,

2013; Ono and Lee, 2016; O'Connor, 2017), not just in transition e onomies, and

the asso iation is not limited to those dire tly a�e ted (e.g., unemployed) (Carr

and Chung, 2014). In sum, the de line in Chinese well-being an be explained by

(1) in reasing in ome inequality whi h fa ilitated in reasing so ial omparisons,

(2) de lining so ial apital, and (3) in reasing unemployment a ompanied by a

severely redu ed so ial safety net. The re ent re overy appears to be driven by

improvements in trust, employment, and the so ial safety net (Easterlin et al.,

2017).

Previous studies investigating the Easterlin paradox and its moderating fa -

tors fo used mainly on ross- ountry studies or on ountries providing \negative"

examples, i.e. ountries in whi h e onomi growth and in reasing well-being do

not go together. The ase of Japan stands out as a \positive" example: a ountry

where e onomi reforms in the early 1990s shifted the ountry from a pattern of

rampant e onomi growth and stagnant well-being, to one of moderate growth and

in reasing well-being (see �gure 3). The question then be omes: what made this

hange possible?

By the end of the 1980s, Japan was in the middle of two risis: on one side, the

demographi risis; on the other, the de line in the viability of the traditional and

2

It is possible that asymmetri responses to e onomi ollapse and positive in ome growth

ould explain why life satisfa tion did not fully re over at the same time as GDP (e.g., from

loss aversion De Neve et al. (2018)), but that is insuÆ ient to explain the pattern in China as

dis ussed in the next paragraph.

3

Due to government restru turing of state-owned enterprises and large rural to urban migra-

tion asso iated with relaxed internal migration laws.

7

Figure 3: Trends of life satisfa tion and GDP per apita ( onstant 2010 US$) in

Japan between 1981 and 2010.

66

.26

.46

.66

.87

Ave

rag

e life

sa

tisfa

ctio

n

25

00

03

00

00

35

00

04

00

00

45

00

0G

DP

pe

r ca

pita

1980 1990 2000 2010Year survey

GDP per capita

life satisfaction

Note: Life satisfa tion data are from the WVS, whereas GDP �gures, presented

in real dollars with base year set to 2010, are issued from the World Development

Indi ators of the World Bank. The trends in life satisfa tion from WVS are

roughly onsistent with those issued from other sour es.

Sour e: Sarra ino et al. (2019).

8

orporate so ial safety net. Greater urbanization and industrialization, along with

e onomi stagnation and international ompetition, put pressure on the s heme of

so ial safety nets whi h traditionally relied on intergenerational support and on

generous bene�ts for the employees of large orporations. For instan e, the share of

three-generation-family households went from 54 per ent in 1975 to 13 per ent in

2013 (Ministry of Health, Labour and Welfare, 2014), whereas the share of elderly

people living alone nearly doubled. At the same time, e onomi onditions for ed

ompanies to limit the bene�ts granted to their employees, and in parti ular to

newly hired personnel. Moreover, the likelihood of lifetime employment de lined

(Ono, 2010). The share of workers in nonstandard employment more than doubled

from 15 to 38 per ent between 1984 and 2016 (Ministry of Health, Labour and

Welfare, 2014). Consequently, the population in need of so ial prote tion greatly

expanded during the 1990s, as well as in ome inequality (see �gure 4).

Figure 4: Evolution of the Gini index of in ome in Japan.

25

30

35

40

Gin

i in

dex

1980 1990 2000 2010

SWIID Post WIID

Note: Lowess smoothed urves. The two lines in the hart refer to measures of

Gini issued from two di�erent sour es of data: the Standardized World In ome

Inequality Database, and the World In ome Inequality Database. Together, the

two series of data provide onsistent eviden e that in ome inequality in Japan

in reased sin e 1980.

Sour e: Solt (2016) and UNU-WIDER (2018).

To fa e these hallenges the government introdu ed a state-sponsored so ial

support system to share so ial risk equitably by the so iety (Horioka and Kanda,

9

2010). A number of poli ies targeting elderly people, as well as work-family poli-

ies were introdu ed in the mid-1990s with the aim of improving the living and

health onditions of elderly people, alleviating the osts of having hildren, and

fa ilitating women a ess to the job-market. Figure 5 shows the trend of welfare

state generosity in Japan (S ruggs et al., 2017).

Figure 5: Average life satisfa tion and Generosity of Welfare State (Japan 1981-

2010).

6.5

6.6

6.7

6.8

6.9

Life

sa

tisfa

ctio

n

11

.02

1.0

41

.06

Ind

ex (

19

90

= 1

00

)

1980 1990 2000 2010

Gen. of Welfare State Policy Life satisfaction

Note: Lowess smoothed urves.

Sour e: Sarra ino et al. (2019).

In the years following the poli y reforms that introdu ed a state-sponsored

so ial safety net in Japan, people's satisfa tion with life in reased, and in parti ular

the satisfa tion of people in the targeted groups. By 2010, aging was asso iated

less negatively to life satisfa tion than in 1990, i.e. before the introdu tion of

the reforms; average health improved; trust in others and so ial parti ipation of

elderly people nearly doubled; single people reported higher life satisfa tion than

previously. All this happened while the e onomy grew, although at a lower pa e

ompared to the previous period.

10

2.1 Our ontribution

Available studies indi ate that: i. so ial apital, so ial safety nets, and in ome

inequality a�e t the relationship between e onomi growth and well-being over

time; ii. poli y-makers an adopt poli ies to promote well-being in the long run.

Our aim is to assess whether the fa tors dis ussed above an help explain the

at trend of life satisfa tion in Luxembourg. This test is important to evaluate

the reliability of available knowledge about the onditions to promote well-being

in the long run, and to identify possible areas of intervention for poli y-makers.

Additionally, in present work we extend the list of moderating onditions to in lude

unemployment. It is well established that unemployment is one of the major auses

of ill-being. Thus it is possible that the hanges in unemployment ontribute to

explaining the trend of life satisfa tion.

3 Method

We use an error orre tion model (ECM) to analyze the fa tors that ontribute

to life satisfa tion in the long-run. The main reason is that ECMs allows us

to estimate onsistent long-run relations between the explanatory variables and

dependent variable. Additional reasons are more te hni al. First, explanatory

variables in levels (e.g., GDP p ) often exhibit unit root properties, whi h ould

lead to the estimation of spurious relations (Engle and Granger, 1987). First-

di�eren ing the variables an be used to address su h spurious relations, but �rst-

di�eren ing limits the interpretation of the results to short-run hanges. ECMs

separately estimate the short- and long-run relations to avoid spurious relations

(under ertain onditions dis ussed below). Also, the estimated long-run relations,

referred to as long-run e�e ts in the time-series literature, are onsistent in the

presen e of short-run reverse ausality (Chudik and Pesaran, 2015; Pesaran, 2015).

Before presenting the ECM, we begin with our assumed data generating pro-

ess, represented by Equation 1. LS

it

represents life satisfa tion for ountry i

at time t, the ve tor X

i;t

in ludes the explanatory variables, and �xed ountry

hara teristi s are represented by �

i

.

LS

it

= �

i

� LS

i;t�1

+ �

0

i0

�X

i;t

+ �

0

i1

�X

i;t�1

+ �

i

+ "

it

(1)

The error orre tion form is the reparameterization of Equation 1 given by Equa-

tion 2.

�LS

it

= �

i

� (LS

i;t�1

� �

0

i

�X

i;t

) +

0

i

��X

i;t�1

+ �

i

+ "

it

(2)

where �

i

= (�

i

� 1); �

0

i

= (�

0

i0

+ �

0

i1

)=(1� �

i

) and

0

= ��

0

1i

.

11

The short-run relations are aptured by and the long-run e�e ts by �. With

a large hange in X

i;t

, the response in the LS

i;t

might overshoot the long-run

equilibrium relationship. When this happens, the error orre tion term, �, serves

to bring the relationship ba k to the long-run one.

For an error orre tion parameterization to be appropriate, (1) the error or-

re tion term should be statisti ally signi� ant, negative, but greater than negative

two, and (2) there must be a long-run ointegrating relationship between the level

variables. Condition 2 is ne essary for the term (LS

i;t�1

� �

0

i

�X

i;t

) to be station-

ary, whi h is ne essary for the error term to be stationary when � is statisti ally

signi� ant. Condition 1 is tested dire tly in the regression analysis. To he k

ondition 2 we test the regression residuals for stationarity using panel unit root

tests (Fisher type augmented Di key Fuller tests). The results indi ate that the

residuals are indeed stationary and ondition 2 is met.

We adjust Equation 2 to develop our �nal spe i� ation (presented below). No-

ti e the oeÆ ients are indexed by i in Equation 2, indi ating that they are allowed

to vary a ross ountries. We allow the short-run relations to be heterogeneous in

order to apture exibly any reverse ausality. However, in our �nal spe i� ation

we assume the long-run e�e ts (�) are ommon a ross ountries. Without this

assumption, we ould not simultaneously assess ea h of the fa tors dis ussed in

Se tion 2 that may explain the trend of life satisfa tion in Luxembourg. This

spe i� ation is referred to as a pooled mean group model (PMG).

In our �nal spe i� ation, we address two further issues: ross-se tional depen-

den e and lag order for serial orrelation. Cross-se tional dependen e o urs when

there is omitted orrelation a ross ountries. A ommonly orrelated e�e t, su h

as the impa t of European Union poli ies, an be a sour e of ross-se tional depen-

den e. To address ross-se tional dependen e we add to equation 2 ross-se tional

means of both the dependent and independent variables (as suggested by Chudik

and Pesaran (2015)). This approa h is similar to adding year dummies, but has

some advantages: adding year dummies greatly in reases the number of ontrols,

and ross-se tional means allow the ommonly orrelated e�e t to a�e t ea h oun-

try through multiple hannels and to di�erent degrees a ording to their di�erent

variable values. Cross-se tional means are also in luded among short-run variables

in the PMG model, meaning their oeÆ ients vary a ross ountries. Con erning

lags, we hose one lag in levels (as spe i�ed in equation 1) be ause the full model

did not onverge when using additional lags. However, we ran regressions using

one explanatory variable at a time in luding up to six lags (in levels). Generally

the long-run e�e ts maintain signi� an e and dire tion and the magnitudes are

larger when in luding additional lags, suggesting our estimates represent lower

bounds. The ex eption is for so ial expenditures, whi h is insigni� ant (presented

with the Results).

12

Our �nal spe i� ation is presented as Equation 3.

�LS

it

= �

i

� (LS

i;t�1

� �

0

�X

i;t

) +

0

i

��X

i;t�1

+ '

i

LS

t

+ Æ

i

X

t

+ �

i

+ "

it

(3)

4 Data

Individual life satisfa tion data are from repeated ross-se tional Eurobarometer

(EB) surveys (European Commission, 2018). Life satisfa tion is measured on a

s ale from 1 to 4 using the responses to the question, \On the whole, are you very

satis�ed, fairly satis�ed, not very satis�ed or not at all satis�ed with the life you

lead?". This question was �rst asked in 1973 in seven ountries and ontinues to-

day for more than the present 28 EU ountries. In ea h year, multiple surveys are

ondu ted that ask about life satisfa tion. Annual observations of life satisfa tion

were onstru ted in ea h ountry as the weighted proportion of native-born individ-

uals reporting the top response ategory, \very satis�ed". Foreign-born individuals

were ne essarily ex luded, be ause prior to 1994, the Eurobarometer target pop-

ulation only in luded native-born individuals, and in 1994, the target population

expanded but still ex luded individuals born in non-EU ountries (S hmitt et al.,

2009, p. 56).

The explanatory variables in lude the natural log of real Gross National In ome

per apita (GNI), unemployment rate, the Gini oeÆ ient of in ome, so ial expen-

ditures, and trust in others. GNI and unemployment data (national estimate) are

from the World Development Indi ators (WDI) (World Bank, 2018). We use the

Gini oeÆ ient of inequality in equivalent household disposable (post-tax, post-

transfer) in ome from the Standardized World In ome Inequality Database (Solt,

2016).

4

Data for so ial expenditures per apita are available every �ve years from 1980-

2015 and 2016 from the OECD So ial Expenditures Database (OECD, 2018).

The variable in ludes all publi so ial expenditures on a tive labour market pro-

grammes, family, health, housing, in apa ity related, old age, other so ial poli y

areas, survivors, and unemployment. We adjusted the variable to more losely

represent the generosity of the welfare state poli y. Con eptually subje tive well-

being relates to the generosity of poli ies not to expenditures; that is be ause so ial

4

The SWIID provides the longest, most omplete, and omparable set of data on in ome

inequality. It is based on data from the World In ome Inequality Database (WIID), but it

hinges on additional assumptions to ease ross-se tional omparability and to impute missing

data. For these reasons some s holars have expressed riti ism towards the SWIID (Jenkins,

2015). However, we �nd that in our ase, �gures from SWIID positively and signi� antly orrelate

with data from WIID and the World Inequality Database(WID) in the years and ountries when

the three data sour es are jointly available.

13

expenditures in rease me hani ally when people retire or when unemployment in-

reases. Indeed O'Connor (2017) �nds, so ial expenditures relate to subje tive

well-being, but the relation be omes statisti ally insigni� ant when ex luding a

ontrol for the old age dependen y ratio. In the present analysis, we ould also

in lude the old age dependen y ratio, but given the small number of degrees of

freedom, we hose instead to partial out the old age dependen y ratio and also the

unemployment rate. Spe i� ally, we used the residuals from a regression of so ial

expenditures on the old age dependen y ratio and unemployment rate. We then

use linear interpolation to fa ilitate annual analysis.

Trust in others is based on responses to the question, \Generally speaking,

would you say that most people an be trusted, or that you ould not be too areful

in dealing with people?" Individual responses are obtained from Eurobarometer

surveys in the years 1986, 2004, 2009, 2010, and 2014. These responses are then

aggregated at the ountry level for ea h year as the portion of people feeling most

people an be trusted. However, omparison over time is limited by di�eren es

in the response s ales. The largest hange o urred beginning in 2009, when the

s ale went from two dis rete hoi es to a s ale ranging from 1 to 10. In order to

produ e annual estimates and a ount for the hange in s ale, we impute trust in

the following steps:

1. The weighted per entage of people stating most people an be trusted is

al ulated by ountry year. For the years 2009, 2010, and 2014 responses

7-10 are re orded as most people an be trusted;

2. These s ores are de-meaned by subtra ting the average level of trust within

a ountry over the years 1986 and 2004 (the years based on the previous

response s ale);

3. The de-meaned trust values are then linearly interpolated and extrapolated

over the sample period with an ex eption { trust is not extrapolated to the

years before 2004 if it is unobserved in 1986;

4. Additional data from the World Values Survey (WVS, 2014) and European

Values Study (EVS, 2011) is used to provide additional information on trust.

The two surveys provide di hotomous answers to a question asking respon-

dents whether people an be trusted. As in step 1, the answers are de-meaned

(within ountry) and extrapolated;

5. To remove the e�e t of the hange in s ale that o urred from 2004 to 2009,

EB trust from step 3 is regressed on EVS/WVS trust from step 4, a dummy

variable demarking the period post-2004, a quadrati trend, and intera -

tions between EVS/WVS trust and both the dummy and trend. Trust is

14

predi ted after ex luding the impa t from the intera tion between post-2004

and EVS/WVS trust; and

6. Last, the ountry means from step 2 were added ba k to obtain our �nal

predi tion of trust.

Our sample of ountries in ludes the �rst �fteen European Union member

states (EU15) be ause only these ountries have suitably long enough series to

be in luded. The period of analysis in ludes the years 1991 { 2016. Our sample

for regression analysis begins in 1991 to ensure there were at least ten ountries

observed in ea h year. Prior to 1991 data for fewer ountries were available when

in luding lags. It is important to use as many ountries as possible be ause the

analysis in ludes ross-se tional means in ea h period. We would prefer to begin

the sample with more than ten ountries, but data for all 15 ountries are not

available until 1995, and with lags, that would signi� antly redu e the time di-

mension. Table 1 presents the sample hara teristi s and average variable values

for ea h ountry.

Table 1: Des riptive statisti s

Country First Last Life satisfa tion Gross National In ome Gini index Unemployment Trust Adj. So ial Expenditures

year year (% very happy) ln(US$ per apita) (0 - 100) rate (%) (% an be trusted) ln(US$ per apita)

Austria 2001 2016 23.95 10.74 27.58 5.10 47.78 0.42

Belgium 1991 2016 27.35 10.62 25.75 7.99 35.10 0.16

Denmark 1991 2016 65.82 10.92 23.85 6.16 78.45 0.51

Finland 2001 2016 33.11 10.73 25.37 8.63 70.02 0.27

Fran e 1991 2015 15.90 10.57 28.69 9.83 29.14 0.15

Germany 1991 2015 17.85 10.57 27.38 7.86 45.06 0.13

Gree e 1991 2016 8.29 10.07 33.58 12.98 37.04 -0.67

Ireland 1991 2016 36.02 10.51 31.13 9.48 37.67 -0.22

Italy 1991 2015 11.41 10.45 32.99 9.83 30.93 -0.07

Luxembourg 1991 2015 43.39 11.25 26.67 3.80 35.68 0.89

Netherlands 1991 2016 47.47 10.73 26.14 4.96 64.32 0.17

Portugal 1992 2015 4.01 9.93 34.03 8.03 31.60 -0.73

Spain 1992 2016 18.17 10.24 32.93 17.38 35.53 -0.44

Sweden 2001 2016 46.83 10.86 25.20 6.77 71.07 0.40

United Kingdom 1991 2016 35.30 10.49 33.83 6.74 44.17 -0.24

Sample average 28.76 10.57 29.22 8.48 45.06 0.02

5 Results

Simple des riptive statisti s suggest that in ome inequality, unemployment, so ial

trust and so ial expenditures in reased in Luxembourg sin e the early 1980s. Panel

6a in Figure 6 shows that in ome inequality in reased by about 5 points, from 23.9

to 28.7, between 1985 and 2015. Similarly, Panel 6b indi ates that unemployment

as a per ent of total labor for e was 0.7% in 1980 and 6.7% in 2015, a nearly 9 fold

in rease in 35 years. A ording to previous literature, we should expe t that su h

in reases hindered life satisfa tion, probably over oming the positive ontribution

of e onomi growth expe ted from traditional e onomi theory.

15

Figure 6: In reasing in ome inequality (Panel A) and unemployment (Panel B) in

Luxembourg.

24

25

26

27

28

29

Gin

i in

de

x

25

35

45

55

65

Sh

are

of

ve

ry s

atisfie

d p

eo

ple

(%

)

1980 1990 2000 2010 2020

Share of very satisfied people Gini index of eqv. hh. disp. income

(a) Gini index of equivalent household

disposable in ome.

02

46

8U

ne

mp

loym

en

t (%

of

tot.

la

bo

r fo

rce

)

25

35

45

55

65

Sh

are

of

ve

ry s

atisfie

d p

eo

ple

(%

)

1980 1990 2000 2010 2020

Share of very satisfied people Unemployment (% of total labor force)

(b) Unemployment as a per entage of

total labor for e.

Sour e: authors' own elaboration.

The in reases in so ial trust and so ial expenditures, on the other hand, are

expe ted to have positively ontributed to life satisfa tion. Sin e 1980 the share of

people who feel that others an be trusted nearly doubled (see Panel 7a in Figure

7), whereas so ial expenditures rose from 8,190 US$ per apita (base year 2013) in

1980 to 23,880 in 2015, i.e. a nearly 3 fold in rease (Panel 7b). It is possible that

the e�e ts on life satisfa tion of in reasing in ome inequality and unemployment,

on one side, and in reasing so ial trust and expenditures, on the other, o�-set

ea h other. To test this hypothesis formally, we turn to the results of the error

orre tion model.

Table 2 presents the results of the error orre tion model. The �rst �ve rows

present the long-run relations orresponding to the � in equation 2; ECT orre-

sponds to the �; the middle rows present the short-run- hange relations orre-

sponding to the s, and the �nal rows the ross-se tional means of life satisfa tion

and the independent variables in levels. The �rst �ve olumns use one explana-

tory variable at a time, olumn 6 reports the results from the model in luding ea h

explanatory variable, and 7, standardized oeÆ ients of the long-run e�e ts from

the full model.

The long-run e�e ts generally orrespond with our expe tations, with the ex-

eption of the Gini oeÆ ient, whi h is statisti ally signi� ant and positive. Per-

manent in reases in GNI p , in ome inequality, so ial trust, and adjusted so ial

expenditures are positively related to life satisfa tion in the long run, and unem-

ployment, negatively.

5

The long-run e�e ts are generally onsistent between the

5

We use the word permanent to distinguish the hanges in levels that trigger the long-run

16

Figure 7: In reasing so ial trust (Panel A) and so ial expenditures (Panel B) in

Luxembourg.

20

30

40

50

% p

eo

ple

tru

stin

g o

the

rs

25

35

45

55

65

Sh

are

of

ve

ry s

atisfie

d p

eo

ple

(%

)

1980 1990 2000 2010 2020

Share of very satisfied people Pred. % People Can Be Trusted

(a) Share of people de laring that others

an be trusted.

51

01

52

02

5S

ocia

l e

xp

en

ditu

res p

.c.

(20

13

US

$/1

00

0)

25

35

45

55

65

Sh

are

of

ve

ry s

atisfie

d p

eo

ple

(%

)

1980 1990 2000 2010 2020

Share of very satisfied people Social Expend. pc Interp./1000

(b) So ial expenditures per apita.

Sour e: authors' own elaboration.

redu ed models ( ols. 1-5) and the full model ( ol. 6). The magnitudes and signif-

i an e of GNI p and so ial expenditures are redu ed. Indeed so ial expenditures

are no longer statisti ally signi� ant. This �nding is surprising in light of the

positive relations found in ross-se tional eviden e; however, insigni� an e ould

be due to multi ollinearity and low statisti al power (re all adjusted so ial expen-

ditures are positive and signi� ant in olumn 5). The magnitudes of the other

variables (Gini, unemployment rate, and trust) in rease in size. A ross variables,

trust has the largest standardized oeÆ ient. The oeÆ ient of trust in others is

more than two times bigger than the one of GNI and nearly three times larger

than the one of inequality or unemployment (in absolute terms). This indi ates

that trust in others is the strongest orrelate of the hanges of life satisfa tion in

the long run among the onsidered variables. The magnitude, however, is small: a

one standard deviation di�eren e in trust is related to 0.6 per entage point greater

life satisfa tion. The standard deviation of life satisfa tion over the full sample is

17.5 per entage points.

It is surprising that the Gini oeÆ ient is positively related to life satisfa tion,

however, as mentioned in the literature se tion, positive relations have been ob-

tained in ross-se tional studies. The Hirs hman tunnel e�e t ould explain the

relation { in reasing in ome of a few, leading to greater inequality, may signal

that the in omes of everyone are in reasing, thereby raising subje tive well-being.

Future resear h should fo us further on inequality and revaluate it in a time-series

ontext.

e�e ts from annual deviations asso iated with short-run di�eren es in life satisfa tion.

17

Among the ross-se tional means, the most important impa t is for GNI. While

a permanent in rease in GNI in a parti ular ountry and year is asso iated with

a long-run in rease in life satisfa tion, when GNI in reases in all ountries in the

same year, the impa t is negative. A three per ent growth rate in ea h ountry

is asso iated with a de rease in life satisfa tion of 0.5 per entage points per year

(based on olumn 6 in luding the long run e�e t and ross-se tional mean but not

the insigni� ant short-run relations: -0.05 = 0.03 * 10.6 + 0.03 * -27.4). This

suggests that GNI positively a�e ts the life satisfa tion in a parti ular ountry if

it grows at a signi� antly greater rate than in other ountries: to break even, the

GNI hange in a ountry needs to be 27.4/10.6 = 2.58 times the average hange

a ross ountries. For the other four fa tors, the impa ts of the signi� ant ross-

se tional means are also in the opposite dire tion but they are smaller than the

long-run e�e ts. Con erning mean life satisfa tion, the impa t of ross-se tional

life satisfa tion is positive indi ating positive spill-over e�e ts. If life satisfa tion

in reases in one ountry, then life satisfa tion will in rease in the other ountries

by approximately one �fteenth of that in rease (the oeÆ ient rounds to 1.0 in

olumn 6).

Short-run variation in various fa tors has theoreti ally distin t impa ts on life

satisfa tion ompared to permanent in reases, espe ially for in ome as illustrated

in Bartolini and Sarra ino (2014). For instan e, we would expe t short run hanges

in unemployment to be signi� antly (and negatively) orrelated to the hanges

of life satisfa tion. However, the present study annot omment on the short-

run relations. The short-run variation is not independent and the relations are

generally statisti ally insigni� ant.

Perhaps the most intuitive way to illustrate our results is to use the model's

predi tion of life satisfa tion, whi h in ludes the impa ts of ea h variable and

their interdependen ies. Figure 8 presents the predi tion and observed share of

very satis�ed people in Luxembourg, based on the estimates presented in table 1,

olumn 6. From this �gure, it is lear the model has high predi tive power. The

model does not get the level of life satisfa tion right, but the short-run hanges

and long-run trend mat h well. Indeed, predi ted and observed life satisfa tion

are strongly orrelated at 84% (signi� ant at 1%). The di�eren e in level is due

to a strong error orre tion term being applied to a relatively high level of life

satisfa tion (Luxembourg averages 43.4 ompared to 28.8 in all ountries), whi h

brings the predi ted level of life satisfa tion in Luxembourg toward the average.

Although we in luded ountry �xed e�e ts in the model, they are treated as short-

run deviations that are ounterbalan ed by the error orre tion term. The results

indi ate that the at trend of life satisfa tion in Luxembourg is due, at least

in part, to o�setting in u es of in reases in: GNI per apita, in ome inequaltiy,

unemployment, so ial trust, and so ial expenditures.

18

Table 2: Results from the ECM model applied to the panel of 15 Western European

ountries (1991 - 2016).

� Life satisfa tion � oeÆ ients

(1) (2) (3) (4) (5) (6) (7)

GNI 38.274

���

10.648

�

0.259

�

(3.990) (6.004)

Gini 0.526

��

0.734

���

0.190

���

(0.249) (0.231)

Unemployment rate -0.437

���

-0.572

���

-0.181

���

(0.249) (0.134)

Trust in others 0.175

��

0.494

���

0.570

���

(0.082) (0.046)

So ial expenditures 19.689

���

0.695 0.023

(3.449) (3.101)

ECT -0.745

���

-0.741

���

-0.693

���

-0.610

���

-0.770

���

-1.036

���

(0.049) (0.062) (0.061) (0.075) (0.043) (0.090)

� GNI -6.238 8.324

(9.132) (12.123)

� Gini -0.424 0.441

(0.538) (1.183)

� Unemployment rate 0.085 0.202

(0.212) (0.309)

� Trust in others 0.162

���

-0.170

(0.060) (0.171)

� So ial expenditures p ) 5.943 -15.673

(16.668) (28.448)

Mean Life Satisfa tion 0.955

���

0.869

���

0.860

���

0.868

���

0.926

���

0.984

���

(0.139) (0.140) (0.137) (0.149) (0.123) (0.169)

Mean GNI -30.443

���

-27.442

�

(5.410) (16.434)

Mean Gini -0.475 -2.541

���

(0.923) (0.877)

Mean Unemployment rate 0.352

���

0.319

(0.131) (0.291)

Mean Trust in others -0.118 -0.399

�

(0.124) (0.232)

Mean So ial expenditures -15.231

���

15.418

(4.621) (11.047)

Constant 14.287 -2.074 -5.714

��

-5.191 -5.112 222.303

(63.727) (25.398) (2.846) (6.063) (3.342) (173.954)

N 353 353 353 353 353 353

Note: So ial expenditures are per apita, adjusted and transformed in logarithm. GNI is per apita

and transformed in logarithm. For more details, please, refer to se tion 4.

Standard errors in parenthesis.

�

p < 0:10,

��

p < 0:05,

���

p < 0:001

19

Figure 8: Predi ted life satisfa tion Vs. observed data.

−20

020

40

60

80

Perc

ent R

espondin

g V

ery

Satisfied

1990 1995 2000 2005 2010 2015

Pred. % Very Satisfied % Very Satisfied

Sour e: authors' own elaboration.

20

6 Con lusions

Previous literature suggests that the relationship between well-being and e onomi

growth depends on a set of onditions: if e onomi growth is a ompanied by ex-

tensive so ial safety nets, high so ial apital, and low in ome inequality, than it

is likely to be asso iated with in reasing well-being. In this arti le we test this

view. In parti ular, we he k whether the at trend of life satisfa tion in Luxem-

bourg, despite a growing e onomy, an be explained by the onditions identi�ed

in previous literature.

Between 1980 and 2008 { the year of the e onomi risis { the Gross National

In ome per apita in Luxembourg grew by 6.35% yearly, while the share of very

satis�ed people remained onstant at about 35%. If the eviden e from previous

studies is orre t, we should expe t that the onditions mentioned above have a

zero net e�e t on life satisfa tion. Unfortunately, the la k of long time-series of

mi ro data prevents a mi ro-e onomi analysis in Luxembourg. We thus adopt a

ma ro-e onomi perspe tive, and we apply an error orre tion model to a panel

of 15 Western European ountries to predi t life satisfa tion in Luxembourg on

the basis of a known set of explanatory fa tors. These are: in ome inequality,

unemployment, so ial expenditures, and trust in others { a ommonly used proxy

of so ial apital.

We �nd the fa tors explain the at trend of life satisfa tion in Luxembourg rea-

sonably well and broadly onsistently with expe tations. In reases in unemploy-

ment o�set the positive in uen es of in reasing so ial trust and e onomi growth.

On the ontrary, in reases in in ome inequality apparently positively a�e ted life

satisfa tion, and so ial expenditures, did not have the statisti ally robust impa t

on well-being that we expe ted. A ross fa tors, the single most impa tful is trust

in others. Standardized oeÆ ients indi ate that the long-run e�e t of trust is

nearly twi e the e�e t of e onomi growth. We also found that the ross-se tional

average of GNI per apita attra ts a signi� ant and negative oeÆ ient. Usu-

ally ross-se tional averages are in luded in the regression to ontrol bias due to

ross- ountry orrelations, su h as the impa t of European Union poli ies, but in

this ase, average GNI plays an unexpe ted role. E onomi growth in a parti ular

ountry has a weak, positive, impa t on well-being, but when ea h ountry grows

at a similar pa e the total e�e t on well-being is negative; this is reminis ent of a

\so ial omparisons" e�e t, but this time a ross ountries. As far as we know, this

is the �rst time that su h eviden e has been do umented in a panel of ountries.

Our �ndings should be viewed with aution. They are based on the best avail-

able data, but several assumptions were ne essary to develop the long time series.

In parti ular, so ial trust is adjusted to improve omparability over time and so ial

expenditures is adjusted to obtain a better proxy for so ial safety nets. Moreover,

the sample size limits the degrees of freedom and our ability to in lude additional

21

ontrol variables and time lags. Our results re e t a preliminary assessment that

ould hange with new data or methods. Indeed, the availability of a new wave of

data from the European Value Study will soon allow us to perform a mi ro analy-

sis overing the period 1999-2019, and therefore to gain a more re�ned knowledge

about what happened to life satisfa tion in Luxembourg.

Nonetheless, we believe our results are en ouraging. They support the view

that the quality of growth matters for well-being. The quest to determine the

onditions that hara terize this \quality" is still in its infan y, but we have a

promising starting point. Further assessment is ne essary, but it is plausible that

jointly onsidering e onomi growth, so ial safety nets, so ial apital, unemploy-

ment, and in ome inequality is the best route to promote a lasting well-being.

22

Referen es

Alesina, A., Di Tella, R., and Ma Cullo h, R. (2004). Inequality and happiness: are

Europeans and Ameri ans di�erent? Journal of Publi E onomi s, 88(9):2009{

2042.

Bartolini, S., Bilan ini, E., and Pugno, M. (2013). Did the de line in so ial onne -

tions depress Ameri ans' happiness? So ial Indi ators Resear h, 110(3):1033{

1059.

Bartolini, S. and Sarra ino, F. (2014). Happy for how long? How so ial apital and

e onomi growth relate to happiness over time. E ologi al E onomi s, 108:242{

256.

Bartolini, S. and Sarra ino, F. (2015). The dark side of Chinese growth: de lining

so ial apital and well-being in times of e onomi boom. World Development,

74:333{351.

Be hetti, L., Gia hin Ri a, E., and Pelloni, A. (2009). The 60es turnaround as a

test on the ausal relationship between so iability and happiness. E onometi a

Working Papers wp07, E onometi a.

Be hetti, L., Trovato, G., and Bedoya, D. (2011). In ome, relational goods and

happiness. Applied E onomi s, 43(3):273{290.

Beja, E. L. (2014). In ome growth and happiness: Reassessment of the Easterlin

Paradox. International Review of E onomi s, 61(4):329{346.

Blan h ower, D. and Oswald, A. (2004). Money, sex and happiness: An empiri al

study. The S andinavian Journal of E onomi s, 106(3):393{415.

Boarini, R., Comola, M., de Keulenaer, F., Man hin, R., and Smith, C. (2013).

Can governments boost people's sense of well-being? The impa t of sele ted

labour market and health poli ies on life satisfa tion. So ial indi ators resear h,

114(1):105{120.

Bro kmann, H., Delhey, J., Welzel, C., and Yuan, H. (2009). The China puzzle:

falling happiness in a rising e onomy. Journal of Happiness Studies, 10:387{405.

Bruni, L. and Stan a, L. (2008). Wat hing alone: relational goods, television and

happiness. Journal of E onomi Behavior and Organization, 65 (3-4):pp. 506 {

528.

23

Carr, E. and Chung, H. (2014). Employment inse urity and life satisfa tion: The

moderating in uen e of labour market poli ies a ross Europe. Journal of Euro-

pean So ial Poli y, 24(4):383{399.

Chudik, A. and Pesaran, M. H. (2015). Common orrelated e�e ts estimation of

heterogeneous dynami panel data models with weakly exogenous regressors.

Journal of E onometri s, 188(2):393{420.

Clark, A. E. and D'Ambrosio, C. (2015). Attitudes to in ome inequality: exper-

imental and survey eviden e. In Handbook of in ome distribution, volume 2,

hapter 13, pages 1147{1208. Elsevier.

Clark, A. E., Fl�e he, S., Senik, C., et al. (2014). The great happiness moderation.

In Clark, A. and Senik, C., editors, Happiness and E onomi Growth: Lessons

from Developing Countries. Oxford University Press, Oxford.

De Neve, J.-E., Ward, G., De Keulenaer, F., Van Landeghem, B., Kavetsos, G.,

and Norton, M. I. (2018). The asymmetri experien e of positive and negative

e onomi growth: Global eviden e using subje tive well-being data. Review of

E onomi s and Statisti s, 100(2):362{375.

Deaton, A. (2008). In ome, health, and well-being around the world: Eviden e

from the GallupWorld Poll. The Journal of E onomi Perspe tives, 22(2):53{72.

Di Tella, R., Ma Cullo h, R., and Oswald, A. (2003). The ma roe onomi s of

happiness. The Review of E onomi s and Statisti s, 85(4):809{827.

Easterlin, R. (1974). Does e onomi growth improve the human lot? Some empir-

i al eviden e. In David, P. and Melvin, W., editors, Nations and households in

e onomi growth, pages 98 { 125. CA: Stanford University Press, Palo Alto.

Easterlin, R. (2009). Lost in transition: Life satisfa tion on the road to apitalism.

Journal of E onomi Behavior & Organization, 71(2):130{145.

Easterlin, R. (2017). Paradox lost? Review of Behavioral E onomi s, 4(4):311{339.

Easterlin, R. and Angeles u, L. (2009). Happiness and growth the world over:

Time series eviden e on the happiness-in ome paradox. IZA Dis ussion Paper,

4060.

Easterlin, R. A. (2013). Happiness, growth, and publi poli y. E onomi Inquiry,

51(1):1{15.

24

Easterlin, R. A., Angeles u, L., Switek, M., Sawangfa, O., and Zweig, J. S. (2010).

The happiness-in ome paradox revisited. Pro eedings of the National A ademy

of S ien es, 107(52):1 { 6.

Easterlin, R. A., Morgan, R., Switek, M., and Wang, F. (2012). China's life

satisfa tion, 1990{2010. Pro eedings of the National A ademy of S ien es,

109(25):9775{9780.

Easterlin, R. A., Wang, F., and Wang, S. (2017). Growth and happiness in China,

1990-2015. In Helliwell, J. F., Layard, R., and Sa hs, J. D., editors, World

Happiness Report 2017, pages 48{83. New York.

Engle, R. F. and Granger, C. W. (1987). Co-integration and error orre tion: rep-

resentation, estimation, and testing. E onometri a: journal of the E onometri

So iety, (55):251{276.

European Commission (2018). Eurobarometer. Te hni al report, GESIS Data

Ar hive for the So ial S ien es.

EVS (2011). European values study 1981-2008, longitudinal data �le. Te hni al

report, GESIS Data Ar hive, Cologne. ZA4804 Data File Version 2.0.0.

Graham, C. and Felton, A. (2006). Inequality and happiness: insights from Latin

Ameri a. Journal of E onomi Inequality, 4(1):107{122.

Helliwell, J. (2008). Life satisfa tion and quality of development. Working Paper

14507, National Bureau of E onomi Resear h.

Helliwell, J. F. (2003). How's life? Combining individual and national variables

to explain subje tive well-being. E onomi Modelling, 20(2):331{360.

Helliwell, J. F. and Aknin, L. B. (2018). Expanding the so ial s ien e of happiness.

Nature Human Behaviour, (2):248{252.

Helliwell, J. F., Aknin, L. B., Shiplett, H., Huang, H., and Wang, S. (2017). So ial

apital and proso ial behaviour as sour es of well-being. Working Paper 23761,

National Bureau of E onomi Resear h.

Horioka, C. Y. and Kanda, R. (2010). Revitalizing the Japanese e onomy by

so ializing risk. Japanese E onomy, 37(3):3{36.

Iniguez-Montiel, A. J. (2014). Growth with equity for the development of Mex-

i o: Poverty, inequality, and e onomi growth (1992{2008). World development,

59:313{326.

25

Jenkins, S. P. (2015). World in ome inequality databases: an assessment of WIID

and SWIID. The Journal of E onomi Inequality, 13(4):629{671.

Jiang, S., Lu, M., and Sato, H. (2012). Identity, inequality, and happiness: Evi-

den e from urban China. World Development, 40(6):1190{1200.

Kahneman, D. and Krueger, A. (2006). Developments in the measurement of

subje tive well-being. Journal of E onomi Perspe tives, 20:3 { 24.

Layard, R. (2005). Happiness: lessons from a new s ien e. The Penguin Press,

New York.

Miku ka, M., Sarra ino, F., and Dubrow, J. K. (2017). When does e onomi

growth improve life satisfa tion? Multilevel analysis of the roles of so ial trust

and in ome inequality in 46 ountries, 1981{2012. World Development, 93:447{

459.

Ministry of Health, Labour and Welfare (2014). Graphi al Review of Japanese

Household. Ministry of Health, Labour and Welfare, Tokyo.

O'Connor, K. J. (2017). Happiness and welfare state poli y around the world.

Review of Behavioral E onomi s, 4(4):397{420.

OECD (2001). The eviden e on so ial apital. In The well-being of nations: the

role of human and so ial apital, pages 39 { 63. OECD, Paris.

Oishi, S. and Kesebir, S. (2015). In ome inequality explains why e onomi growth

does not always translate to an in rease in happiness. Psy hologi al s ien e,

26(10):1630{1638.

Oishi, S., Kesebir, S., and Diener, E. (2011). In ome inequality and happiness.

Psy hologi al s ien e, 22(9):1095{1100.

Ono, H. (2010). Lifetime employment in Japan: Con epts and measurements.

Journal of the Japanese and international e onomies, 24(1):1{27.

Ono, H. and Lee, K. S. (2013). Welfare states and the redistribution of happiness.

So ial For es, 92(2):789{814.

Ono, H. and Lee, K. S. (2016). Redistributing Happiness: How So ial Poli ies

Shape Life Satisfa tion. Praeger, Santa Barbara.

Pa ek, A. C. and Rad li�, B. (2008). Welfare poli y and subje tive well-being

a ross nations: An individual-level assessment. So ial Indi ators Resear h,

89(1):179{191.

26

Pesaran, M. H. (2015). Time series and panel data e onometri s. Oxford University

Press.

Rothstein, B. (2010). Happiness and the welfare state. So ial Resear h: An Inter-

national Quarterly, 77(2):441{468.

Sa ks, D. W., Stevenson, B., and Wolfers, J. (2012). Subje tive well-being, in-

ome, e onomi development and growth. In Booth, P., editor, The pursuit

of happiness, hapter 3, pages pp. 59 { 98. The Institute of E onomi A�airs,

London.

Sarra ino, F. (2012). Money, so iability and happiness: Are developed oun-

tries doomed to so ial erosion and unhappiness? So ial Indi ators Resear h,

109(2):135{188.

Sarra ino, F., O' Connor, K. J., and Ono, H. (2019). Making e onomi growth

and well-being ompatible: eviden e from Japan. Working paper, MPRA paper

n. 93010.

S himma k, U., Krause, P., Wagner, G., and S hupp, J. (2010). Stability and

hange of well-being: an experimentally enhan ed latent state-trait-error anal-

ysis. So ial Indi ators Resear h, 95(1):19 { 31.

S hmitt, H., S holz, E., Leim, I., and Mos hner, M. (2009). The Mannheim Euro-

barometer Trend File Codebook and Unweighted Frequen y Distributions.

S ruggs, L., Detlef, J., and Kati, K. (2017). Comparative welfare entitlements

dataset 2. Te hni al report, University of Conne ti ut & University of Greif-

swald. Version 2017-09.

Solt, F. (2016). The standardized world in ome inequality database. So ial S ien e

Quarterly, 97(5):1267{1281. SWIID Version 7.0, July 2018.

Stevenson, B. and Wolfers, J. (2008). E onomi growth and subje tive well-being:

reassessing the Easterlin Paradox. Brookings Papers on E onomi A tivity, (1):1

{ 87.

Uhlaner, C. (1989). Relational goods and parti ipation: in orporating so iability

into a theory of rationl a tion. Publi Choi e, 62:253 { 285.

UNU-WIDER (2018). World in ome inequality database.

Van Reekum, C., Urry, H., Johnstone, T., Thurow, M., Frye, C., Ja kson, C.,

S haefer, H., Alexander, A., and Davidson, R. (2007). Individual di�eren es in

27

amygdala and ventromedial prefrontal ortex a tivity are asso iated with eval-

uation speed and psy hologi al well-being. Journal of Cognitive Neuros ien e,

19(2):237{248.

Veenhoven, R. and Vergunst, F. (2013). The Easterlin illusion: e onomi growth

does go with greater happiness. EHERO working paper 2013/1.

World Bank (2018). World development indi ators. Te hni al report, World Bank.

A essed May 2018.

WVS (2014). World values survey 1981-2014 longitudinal aggregate v.20150418.

Te hni al report, World Values Survey Asso iation. Aggregate File Produ er:

JDSystems, Madrid.

28