Happiness and Income
Analysis of the relationship between happiness and income
Department of Economics,
School of Business, Economics, and Law
Daniella Ocampo & Hanna Magnusson
Abstract
The relationship between happiness and GDP per capita is relatively controversial due
to the various findings. Therefore, this paper investigates the relationship between GDP
per capita and happiness for 37 European countries between 2002 and 2018. Using data
from the European Social Survey, we test if variations in GDP per capita across countries
affect the average happiness of the countries. Controlling for socio-demographic variables
and country heterogeneity, we obtain estimates suggesting a positive and significant
relationship between GDP per capita and happiness.
Keywords: Subjective well-being, Happiness, Life satisfaction, GDP per capita
Bachelors thesis in Economics, 15 credits
Supervisor: Dick Durevall
Contents
List of Tables 3
1 Introduction 1
2 Literature review 3
2.1 Economic and happiness . . . . . . . . . . . . . . . . . . . . . . . . 3
3 Data 6
3.1 Survey data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2 Dependent variables . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.3 Independent variables . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4 Methodology 11
4.1 Linear regression model . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.1.1 Model 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.1.2 Model 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.1.3 Model 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.2 Methodological limitations . . . . . . . . . . . . . . . . . . . . . . . 13
5 Results 14
5.1 Empirical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.2 Robustness test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6 Conclusion 23
References 25
7 Appendix 29
List of Tables
1 Variable description . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 Happiness regression . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Life satisfaction regression . . . . . . . . . . . . . . . . . . . . . . . 21
3 Life satisfaction regression . . . . . . . . . . . . . . . . . . . . . . . 22
4 Happniess and Life satisfaction correlation . . . . . . . . . . . . . . 29
5 Multicollinearity (VIF test) . . . . . . . . . . . . . . . . . . . . . . 30
1 Introduction
The notion that increased income leads to higher levels of happiness is something
that is well discussed in the literature (Kahneman & Krueger 2006, Clark et al.
2008) and one would expect that increased income would lead to a higher level
of happiness due to various factors. This result is built heavily on the neoclas-
sical notion of utility functions with concave indifference curves where increased
income leads to higher levels of utility, or happiness,1 by increased consumption
possibilities.
This is indeed what is found in the literature. Easterlin (1974) was one of the
first papers empirically showing that increased household income made individuals
happier. However, Easterlin (1974) also found that on an aggregated level, the
relationship between income and happiness disappears; this finding is often referred
to as the Easterlin Paradox. More recent papers, however, have disproved the
Easterlin paradox by showing that the relationship between happiness and income
on an aggregated level is positive (Stevenson & Wolfers 2008, Tella et al. 2003,
Diener et al. 1985).
This paper builds on the results of Stevenson & Wolfers (2008), Tella et al.
(2003), Diener et al. (1985) by investigating the relationship between GDP per
capita and national happiness using more recent data from the EU. We aim to
contribute to the literature by re-visiting the Easterlin paradox providing new es-
timates for the EU region, as compared to the US previously examined by Easterlin
(1974).
Using data from the European Social Survey, we use OLS to estimate the rela-
tionship between GDP per capita and happiness. We find evidence that increased
GDP per capita increases happiness, which goes against what Easterlin predicted
1In practical terms, utility can be hard to define, so happiness is sometimes used as a proxy
for utility
1
but is consistent with more recent empirical findings.
The remainder of this thesis is organized as follows: Section 2 introduces the
literature on happiness and (income) GDP per capita. In section 3, the data used
in our analysis are presented, and section 4 describes the methodology. The results
and the interpretation of our analysis are presented in section 5. In section 6, we
summarize the empirical findings and discuss the limitations of our study.
2
2 Literature review
2.1 Economic and happiness
The study of happiness, or subjective well-being (SWB), is a relatively new field
in economics, and Richard Easterlin first studied happiness economics. By using
data from World Survey, AIPO Polls, NORIC polls, and Cantril, Easterlin (1974)
studied the United States and 11 other countries during the period 1946 to 1970, in
a cross-country comparative study. Easterlin (1974, 1995) empirical findings sug-
gest that there is an inconsistency in the relationship between subjective well-being
and income on a national level and on an individual level. On an individual level,
Easterlin’s evidence indicates that the income level affects the level of subjective
well-being. Wealthier individuals were found to have higher subjective well-being,
and on the contrary, more impoverished individuals presented to have lower sub-
jective well-being. However, the empirical evidence implies that an increase in the
national economic growth does not necessarily implicate an increase in subjective
well-being for the nation.
Easterlin (1974) stated that ”in all societies, more money for the individual typ-
ically meant more individual happiness. However, raising the income for all does
not increase the happiness of all”. Easterlin emphasizes that this distinction is es-
sential and that it is the foundation of his research regarding the income-happiness
paradox. Furthermore, Easterlin also suggests that the relationship between in-
come and happiness may be different within countries and across countries. The
empirical evidence just presented, that individuals belonging to the higher income
group are generally more happier than impoverished individuals, are significant
within countries. Across countries the association between income and happiness
are much weaker (Easterlin 1974).
Consistently with Easterlin, Diener et al. (1985) also established a significant
3
positive relationship between income and subjective well-being, noted that this
relationship was found on an individual level, and in accordance with Easterlin,
no empirical findings were found that suggests such a relationship on a national
level.
However, there is a wide criticism against Easterlin’s empirical findings within
the happiness economic literature, especially against the Easterlin paradox. Con-
trary to Easterlin, Stevenson & Wolfers (2008) found a significant positive relation-
ship between income (GDP per capita) and subjective well-being across countries.
Stevenson and Wolfers used different and more recent data sets from countries
over decades, compared to Easterlin. The data used were from World Values Sur-
vey, Eurobarometer, Pew Global Attitudes Survey, and Gallup World poll. The
World Value Survey data is data collected from different waves during a span of
years. The Eurobarometer data is cross-country public survey data, Pew Global
Attitudes Survey data is global public survey data, and Gallup World poll data is
a collection of global and region-specific survey data.
Stevenson & Wolfers (2008) did not find evidence in line with Easterlins finding,
the evidence suggests a positive relationship between income (GDP per capita) and
subjective well-being, and this relationship is consistent through the different data
sets and within countries. They also specify that relative income is more limited
when establishing happiness than absolute income.
Likewise, Tella et al. (2003) found a significant positive relationship between
GDP per capita and happiness by using panel data.
Easterlin et al. (2010) revisited the Easterlin paradox, where they pointed out
that it is of significance to separate the long and short-term effects. Contrary
to Easterlin’s previous studies, which were limited to developed countries, this
paper emphasizes no evidence of a long-term relationship between income (GDP
per capita) and happiness for neither a smaller number of developing countries nor
4
a more significant number of developed countries. Nonetheless, they established
a positive relationship between income and happiness throughout both developed
and developing countries in the short term.
5
3 Data
3.1 Survey data
In this paper, the aim is to examine the relationship between GDP per capita and
happiness. For this purpose, we use data on subjective well-being from the Euro-
pean Social Survey (ESS). The European Social Survey is a cross-national survey
focusing on European countries, and we use data from 37 European countries. The
objective of the European Social Survey is to identify and distinguish the synergy
between institutions and individuals’ living conditions and attitudes across Eu-
rope (European Social Survey n.d.c). The data span the years 2002 to 2018, and
the survey data has been collected regularly every two years during this period
(European Social Survey n.d.a). The survey is established through face-to-face
interviews, with a broad randomized sample of residents in each country included
in the survey and each year. By using rigorous random probability methods to
select individuals, the samples represent residents aged 15 and over (European
Social Survey n.d.d).
The data is separated into nine waves, and for this thesis, we will use data
from all the waves, corresponding through the years 2002, 2004, 2006, 2008, 2010,
2014, 2016, and 2018 (European Social Survey n.d.e). It results in observations
from over a ten-year period.
3.2 Dependent variables
Subjective well-being (SWB) is a common term within psychology and is described
as an umbrella term regarding individuals’ feelings and thoughts (Kahneman et al.
1999). Subjective well-being is defined as a person’s cognitive and affective eval-
uations of his or her life and is a measurement of a person’s well-being (Sumner
1996). Subjective well-being as a measurement is regarded to be a relatively robust
6
estimation of an individual’s well-being (Dolan et al. 2008).
The European Social Survey measures subjective well-being in two ways: hap-
piness and life satisfaction. The happiness data observed in the survey comes
from the question: ”Taking all things together, how happy would you say you
are?”, where the respondents evaluate their overall happiness using an answer
range from 0 (extremely unhappy) to 10 (extremely happy). Simultaneously, the
life satisfaction data comes from the question: ”All things considered, how satis-
fied are you with your life as a whole nowadays?”, as for the happiness question,
the respondents evaluate their answers according to the same 11-point scale (Eu-
ropean Social Survey n.d.b). The answers from the happiness question form the
dependent variable in our equation. In other words, the dependent variable is the
self-reported perception of happiness. To implement a robustness test, we have
chosen to conduct further regressions isolated to life satisfaction as the dependent
variable.
3.3 Independent variables
To estimate the relationship between GDP and happiness, we merge the European
Social Survey data with data on GDP per capita from the World Bank (World bank
n.d.). We use data from all the countries included in the waves, and data are from
the different years that the waves represent (2002 to 2018). More specifically, we
use GDP per capita based on purchasing power parity (PPP)2 in order to equalize
the currencies and the inflation. Consistent with previous studies (see Easterlin
et al. (2010); Sacks et al. (2010)), we use the logarithmic GDP per capita (see
Table 1) in our regressions to normalize the income distribution. The existing
non-linear relationship between GDP per capita and happiness is an additional
reason for including the logarithmic GDP per capita as a specification in our
2GDP per capita in international dollar, constant 2017
7
regression models. The GDP per capita is our variable of interest.
The remaining independent variables are control variables from the European
Social Survey, and they are presented in Table 1. We have chosen to use variables
closely related to the individual, which could be described as individual character-
istics or socio-demographic variables. More specifically, we use the variables age,
age squared, gender, years of full-time education completed, employment status,
marital status, subjective general health, having children living at home or not,
religiosity, political orientation, trust in the country’s parliament, trust in the le-
gal system, and social meetings. As described in Table 1 we use a dummy for
gender (1 = Female; 0 = Otherwise) and for marital status (1 = Married; 0 =
Otherwise). Religious is an ordinal variable categorized from 0 = Not religious to
10 = Very religious, likewise are political orientation, trust in the parliament, and
the legal system variables, which are categorized in this form. Whereas health is
categorized from 1 (Very good) to 5 (Very bad), social meetings are categorized
from 1 (Never) to 7 (Every day), and employment status has three categories: 1
= Employed; 2 = Self-employed; 3 = Not in paid work.
Some of these variables are affected by GDP per capita, for example, education,
employment, and health; thus, they are mediators. In contrast, other variables are
exogenous from income, like age and gender.
Earlier literature has provided evidence that happiness is likely to be affected
by health; poor health is associated with being less happier and good health with
the opposite. Simultaneously, the literature further suggests that employment sta-
tus and family circumstances (marital status and children) along with health are
important determinations of happiness (Clark & Oswald 1994, Dolan et al. 2008,
Easterlin 2006). Furthermore, Clark & Oswald (1994) find that unemployment is
negatively correlated with happiness, being married is positively correlated, and
females are more unhappy than males. On the contrary, Alesina et al. (2004) sug-
8
gest that the male variable is negatively associated with happiness. Nonetheless,
gender is associated with happiness even though the previous empirical evidence
shows different results concerning the two genders.
Previous literature shows that the relationship between age and happiness is
U-shaped (Easterlin 2006). Therefore, we have selected data on age from the
European Social Survey and the square of this variable. Adding the variable
age squared allows us to estimate the effect of a non-linear relationship with the
dependent variable.
Alesina et al. (2004) also show that education is positively correlated with
happiness and that years of education completed have a significant role. When
controlling for income, Alesina et al. find evidence that more educated individuals
tend to be happier.
Bjørnskov (2006) and Helliwell & Putnam (2004) show that vertical trust in
institutions is associated with happiness. Higher levels of trust tend to have a
positive relationship with higher happiness levels (Helliwell & Putnam 2004).
Further, Alesina et al. (2004) present evidence that political orientation is
associated with happiness and that the outcome is different depending on where
on the political spectrum the individual is located (left or right). It is suggested
that religion also is associated with happiness, and further that the relationship
between religious involvement and happiness is positive (Lelkes 2006, Lewis et al.
2005).
Additionally, we divide the countries into regional areas in Europe and create
a dummy variable for each region. Theoretically, the relationship between happi-
ness and income could be affected by each region’s unobserved characteristics. We
create dummies for each region to control for the unobserved differences. The base-
line regional dummy is Nordic, and we have defined Nordic as Denmark, Sweden,
Norway, Finland, and Iceland. Great Britain and Ireland are the second regional
9
dummy. The dummy for West Europe is defined as Belgium, France, Luxemburg,
Netherlands, Switzerland, Germany, Austria, Spain, Portugal, Italy, and Greece.
The last regional dummy is East Europe. It is defined as Russian Federation,
Ukraine, Hungary, Czechia, Poland, Romania, Slovakia, Bulgaria, Latvia, Lithua-
nia, Estonia, Slovenia, Albania, Croatia, Serbia, Montenegro, Kosovo, Cyprus, and
Turkey.
Table 1: Variable description
Variable Description Observations Mean Standard Dev.
Happiness 0=Extremely unhappy; 10=Extremely happy 413,299 7.194 2.029
Life satisfaction 0=Completely unsatisfied; 10=Completely satisfied 413,771 6.852 2.315
GDP per capita (Log) PPP adjusted, constant 2017 international dollar 415,960 10.528 0.408
Age Age of respondant 414,191 48.378 18.590
Female Dummy variable: 1=Female; 0=Male 415,960 0.537 0.498
Years education Years of full-time education completed 411,350 12.317 4.095
Marital status Dummy variable: 1=Married; 0=Otherwise 415,960 0.239 0.426
Children living at home or not Dummy variable: 1=Lives with children; 0=Does not 365,558 1.623 0.484
Religious 0=Not religious; 10=Very religious 412,086 4.703 3.026
Health 1=Very good; 5=Very bad 415,368 2.246 0.930
Employment status 1=Employed; 2= Self-employed; 3=Not in paid work 39,601 2.016 0.953
Placement on left right scale Placement on left right scale: 0=Left; 10=Right 354,723 5.106 2.207
Trust in parliament 0=Not at all; 10=Complete trust 404,279 4.373 2.617
Trust in legal system 0=Not at all; 10=Complete trust 404,699 6.063 2.711
Social meetings 1=Never; 7=Every day 413,971 4.853 1.604
Great Britain and Ireland Regional fixed effect 415,960 0.096 0.295
West Europe Regional fixed effect 415,960 0.379 0.485
East Europe Regional fixed effect 415,960 0.369 0.482
Nordic Regional fixed effect 415,960 0.153 0.360
10
4 Methodology
4.1 Linear regression model
To estimate the association between our chosen dependent variable and the inde-
pendent variable of interest, referenced in the latter part, we use OLS regression as
our regression technique. The following regression specification is the fundamental
equation:
yict = β0 + β1logGDPct + β2Xict + δc + ϵict (1)
where y is the dependent variable happiness (subjective well-being), x is a
vector of the explanatory variables, β is the parameters, δ is the regional fixed
effect dummy3, and ϵ is the error term. Further, i represents the individual, c
represents the country, and t is the time. Clustered standard errors will be used
continuously through all models.
4.1.1 Model 1
In our regression model, the dependent variable is the subjective well-being con-
cept, which consists of the self-reported apprehension of happiness; this will be
consistent in our models. The explanatory part in the linear regression model
consists of our variable of interest, GDP per capita, and a regional fixed effect
dummy to control for unobserved differences between regions regarding regional
specifications. More precisely, we have chosen to include the logarithmic GDP
per capita data (PPP-based). This model specification can resemble a simplistic
neoclassical model (utility) due to the absence of socio-demographic factors like
gender, age, and marital status. In neoclassical theory, income or GDP is the most
3Wave-fixed effects are not included in the specification due to collinearity causing the coef-
ficients to be omitted in stata. The variation in BNP per capita is collinear.
11
basic analysis of happiness. This first specification is a straight effect of income
on happiness. Further specifications are included in the forthcoming models to
analyze the association, given that we control for more variables.
The regression specifications to be estimated are the following:
Happyict = β0 + β1logGDPct + δc + ϵict (Model 1)
4.1.2 Model 2
In the second model, we expand the basic model by adding variables that we think
also affect happiness but are not obviously affected by income. We add the control
variables age, age squared, gender, religious, marital status, and children living at
home or not. By including these variables, the BNP per capita coefficient should
not change significantly from the estimated coefficient in model 1.
With this specification in our model, the regression that is conducted in our
analysis is:
Happyict = β0 + β1logGDPct + β2Independentict + δc + ϵict (Model 2)
4.1.3 Model 3
The last linear regression model consists of the specifications in model 2. Differen-
tially, this model also includes mediators such as years of education, employment
status, health, political orientation, trust in the country’s parliament and in the
legal system, and the continuity in social meetings. Hence, we isolate the effect of
GDP by controlling for these variables:
Happyict = β0 + β1logGDPct + β2Independentict + δc + ϵict (Model 3)
12
4.2 Methodological limitations
One methodological limitation is that we have chosen to use ordinary least squares
to estimate our regressions. This is a limitation regarding the characteristics of the
dependent variable (or variables). The dependent variable is ordinal categorical,
and it consists of an 11-point scale. In other words, OLS cannot estimate the
coefficients for each category of the scale. To solve this problem or limitation,
ordered probit could instead be used. Ordered probit analysis is used when the
dependent variable has more than two outcomes, which the dependent variable
happiness has. However, when comparing OLS and ordered probit, earlier studies
find that the results are very similar (Stevenson & Wolfers 2009). For these reasons,
we choose to use OLS.
13
5 Results
5.1 Empirical results
Table 2 presents the main regression results. As described previously in the
methodology section, three different specifications of the model are estimated,
and the specifications are also presented in Table 2. GDP per capita and a re-
gional fixed effect dummy are the common variables for all three models, a set of
exogenous income variables are included in our second model, and the third model
includes variables which to some extent are affected by income.
The results from Model 1 indicate that GDP per capita (logarithmic) has a
positive and significant effect on happiness. The estimated coefficient for GDP per
capita shows resembling results, i.e., positive and significant effects on happiness,
for the expanded models. The estimated GDP per capita coefficient for Model 1 has
a value of 1.258 and is significant at a 0.1 % level. This means that a 1% increase
in GDP per capita increases happiness by 1.258 ·0.01 units when including regional
dummies. In Model 2 happiness increases by 1.379 ·0.01 units with a 0.1% increase
in GDP per capita, and the coefficient is positive and significant (∗∗∗p < 0.001).
However, the estimated coefficient for Model 3 is positive and significant at a 1%
level, and a 1% increase in GDP per capita yields an increase in happiness by
0.889 · 0.01 units. The results show that Model 3 is significant at ∗∗p < 0.01, which
differs from the two earlier models.
This difference may be because the observations decrease to 32611 from ap-
proximately 400 000, and this decrease in observations is most likely a cause of
the employment status variable. The employment status variable has fewer obser-
vations compared to the other variables. 4. Another possibility for the different
significant levels could be multicollinearity; hence, we test for multicollinearity by
4See Table 1, the observations on employment status is 39,601.
14
conducting a variance inflation factor (VIF) test. The VIF value5. for each vari-
able is low except for the variables age and age squared. However, age and age
squared values are expected to be high. The result of the VIF test excludes this
as a possibility for the difference in significant level. The difference between the
GDP per capita coefficients in Model 1 and Model 2 are trivial, as we expected.
Most likely, one or more variables in Model 2 are to some extent affected by GPD
per capita, hence the slightly higher coefficient.
Happiness and GDP per capita are positively associated. However, causation
is not proved - there is no consistent evidence for where a certain level of GDP
per capita causes a specific outcome of happiness. The estimated coefficients in
our analysis are consistent with previous studies (Stevenson & Wolfers 2008, Tella
et al. 2003, Diener et al. 1985)).
Further, our results are not in line with the empirical evidence suggested by
Easterlin, who do not find evidence for a relationship between happiness and GDP
per capita (Easterlin 1974). In Easterlin’s analysis, time variation is included;
therefore, Easterlin can estimate the long-term relationship between happiness
and BNP per capita (Easterlin et al. 2010). Considering the long-term aspect, our
results are not entirely comparable to Easterlin’s because our analysis does not
include time variation. Therefore we cannot estimate a long-term relationship.
However, as mentioned earlier, Easterlin finds evidence that there is a positive
relationship between happiness and GDP per capita in the short term (Easterlin
et al. 2010). Our result indicates that this positive relationship exists. Never-
theless, it must be mentioned that there is some time variation even in the short
term.
The estimations of the socio-demographic characteristics indicate that age is
negative and significant at a 0.001 level for the second-and-third model. Fur-
5See Table 5 in Appendix
15
ther, age squared is presented to be positive and significant. Earlier studies using
cross-section data have found a U-shaped relationship between age and happiness
(Blanchflower & Oswald 2004, 2008, Stone et al. 2010). Correspondingly, evidence
is found in our analysis that validates the previous findings; the estimated coeffi-
cient for age squared is positive (∗∗∗p < 0.001), which means that the relationship
between age and happiness is non-linear. Indicating that individuals are happy,
and at some point in life, the happiness of the individuals starts to decrease, and
then happiness increases again. A U-shaped relationship is found. However, we
do not analyze the cause of this relationship.
Simultaneously, the regression results show that women are happier than men,
which is the same result that is suggested in Blanchflower & Oswald (2004). Fur-
ther, the coefficients present that being married has positive and significant effects
on happiness. However, the estimated coefficient on having children living at home
is negative and statistically significant in Model 2, and insignificant in Model 3.
Also presented in Table 2 is the positive and significant coefficient on the variable
regarding social meetings with family and friends.
As shown in Table 2 being religious is only associated with higher happiness
in Model 2, and having trust in the parliament and the legal system, and political
orientation and social meetings are associated with higher self-reported happiness
in Model 3. Similar to earlier studies, empirical evidence indicates that an indi-
vidual’s perceptions of his health and employment status are positively associated
with happiness. We also find evidence for a positive relationship between higher
education and happiness.
Lastly, we interpret the regional fixed effects dummies. The regional fixed
effects of Great Britain and Ireland and West Europe are negative and significant.
These regions are estimated to be less happy than our baseline region Nordic, and
the coefficient for East Europe is insignificant.
16
Table 2: Happiness regression
Model 1 Model 2 Model 3
GDP per capita (Log) 1.258*** 1.379*** 0.889**
(0.197) (0.209) (0.291)
Great Britain and Ireland -0.580** -0.570* -0.121
(0.182) (0.219) (0.133)
West Europe -0.515*** -0.546*** -0.288
(0.124) (0.128) (0.146)
East Europe -0.598** -0.576** -0.0630
(0.178) (0.187) (0.257)
Age -0.0493*** -0.0542***
(0.00456) (0.00708)
Age squared 0.000362*** 0.000602***
(0.0000355) (0.0000670)
Female -0.0253 0.160***
(0.0243) (0.0230)
Religious 0.0499*** 0.0197
(0.00859) (0.0117)
Marital status 0.319*** 0.679***
(0.0421) (0.0749)
Children living at home or not -0.163*** 0.0212
(0.0236) (0.0401)
Years of education 0.00902
(0.00654)
Employment status -0.0839***
17
Table 2 continued from previous page
Model 1 Model 2 Model 3
(0.0196)
Health -0.513***
(0.0311)
Placement on left right scale 0.0393***
(0.00768)
Trust in parliament 0.0671***
(0.00784)
Trust in legal system 0.0408***
(0.00920)
Social meetings 0.173***
(0.0216)
Constant -5.582* -5.496* -1.804
(2.122) (2.217) (3.250)
Observations 413299 358868 32611
Adjusted R2 0.096 0.126 0.210
Cluster standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
18
5.2 Robustness test
In order to control the stability of our empirical results, whether the results remain
the same or at least consistent, we operate a robustness analysis. We control the
empirical results by replacing the dependent variable, given that we use the exact
specification as in our happiness regressions.
Tabel 3 shows the estimated coefficients for our robustness analysis, and the
model specifications are6:
LSict = β0 + β1logGDPct + δc + ϵict (Model 1)
LSict = β0 + β1logGDPct + β2Independentict + δc + ϵict (Model 2)
LSict = β0 + β1logGDPct + β2Independentict + δc + ϵict (Model 3)
The differences between the models in our happiness regression and our life
satisfaction regressions are the dependent variable. We use life satisfaction, instead
of happiness, as our dependent variable in the robustness test. Table 47 show the
correlation between the two variables, and the correlation is high. Hence, the
two variables are interchangeable, and life satisfaction is plausible to perform as a
robustness test.
Tabel 3 provides the estimated coefficients for our robustness analysis. The
results presented in 3 show that the estimated coefficient on GDP per capita for
Model 1 is 1.858 and is significant at ∗∗∗p < 0.001, and an 1 % increase in GDP
per capita increases life satisfaction by 1.858 · 0.01 units. Model 2 is significant
6In the equations LS stands for Life Satisfaction.
7See Table 4 in Appendix.
19
at the same level as Model 1; however, the estimated coefficient is slightly higher
and has a value of 2.003. This increases life satisfaction by 2.003 · 0.01 units when
GDP per capita increases by 1 %. Finally, Model 3 a 1 % increase in GDP per
capita increases life satisfaction by 1.352 · 0.01 units.
The estimated results for our life satisfaction regression are more positively
significant than the estimated coefficients in our happiness regressions. Further-
more, the significant level is the same throughout the models in the happiness
and life satisfaction regressions except for Model 3 which has a higher significance
level than Model 3. Consistently, the magnitude of the effect is higher in our es-
timated coefficients for life satisfaction. Hence, GDP per capita seems to be an
essential factor when determining happiness since the results are consistent and
the robustness analysis corresponds to our primary regression.
A possible explanation for the slight difference in the results on GDP per capita
could be the dependent variables themselves. The dependent variables are based on
two different formulated questions; ”Taking all things together, how happy would
you say you are?” and ”All things considered, how satisfied are you with your life as
a whole nowadays?”. Both questions are within the framework of subjective well-
being and are used to measure subjective well-being. However, the concepts are not
entirely interchangeable, although a high correlation is proved, demonstrated by
the questions themselves. Therefore, the respondents may interpret the questions
differently, and the answers vary, resulting in different estimated coefficients.
20
Table 3: Life satisfaction regression
Model 1 Model 2 Model 3
GDP per capita (Log) 1.858*** 2.003*** 1.352***
(0.270) (0.275) (0.342)
Great Britain and Ireland -0.882** -0.860** -0.464*
(0.267) (0.295) (0.177)
West Europe -0.746*** -0.787*** -0.408*
(0.188) (0.191) (0.180)
East Europe -0.620* -0.576* -0.0240
(0.235) (0.245) (0.293)
Age -0.0652*** -0.0684***
(0.00680) (0.00989)
Age squared 0.000543*** 0.000794***
(0.0000538) (0.000100)
Female 0.0581* -0.204***
(0.0254) (0.0237)
Religious 0.0595*** 0.0256*
(0.0109) (0.00991)
Marital status 0.280*** 0.624***
(0.0530) (0.0422)
Children living at home or not -0.0856*** 0.113**
(0.0233) (0.0296)
Years of education 0.0214
(0.0109)
Employment status -0.107**
21
Table 3: Life satisfaction regression
Model 1 Model 2 Model 3
(0.0280)
Health -0.600***
(0.0345)
Placement on left right scale 0.0647***
(0.00940)
Trust in parliament 0.0909***
(0.0109)
Trust in legal system 0.0736***
(0.0103)
Social meetings 0.147***
(0.0213)
Constant -12.12*** -12.20*** -6.923
(2.902) (2.930) (3.733)
Observations 413771 359328 32614
Adjusted R2 0.129 0.154 0.239
Clustered standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
22
6 Conclusion
This thesis investigates the relationship between happiness and income using socio-
demographic data from the European Social Survey and GDP data from the World
Bank. The empirical evidence finds a correlation between happiness and income
and that the relationship is positive and significant. Our regional analysis indicates
that the happiest region in Europe is the region we have identified as Nordic.
Income or GDP per capita appears to be a good indicator and measurement
of happiness. This thesis has also presented that happiness is also affected by
individual and personal factors. Therefore, GDP per capita is not enough to
measure happiness, and it is necessary to enhance with other variables.
This study has limitations. For instance, our specifications do not include
household net income due to missing data, and we cannot analyze any possible
effects household income has on happiness.
Our robustness test is accomplished by changing the dependent variable, and
our analysis is limited because the same estimation methodology is used through
all regressions. To further develop the robustness analysis, ordered probit analysis
can be conducted. Ordered probit is a relevant estimation due to the properties
of the variables; as previously mentioned, we have chosen to limit this study to
OLS. For further research in this field, different data sets can be used to examine
if the result is consistent through the different data sets. For example, both cross-
sectional data and panel data can be used to investigate if the result is consistent
and similar.
Further, our results indicate that GDP per capita is essential for the countries
wealth and the resident’s prosperity. Even though the magnitude of GDP per cap-
itas effect on happiness and life satisfaction is not significant, GDP per capita still
has an effect on happiness and life satisfaction which should be noted. Therefore,
our results show possible policy implications and that GDP per capita is impor-
23
tant. The empirical evidence from this thesis also indicates that other factors are
also important when determining happiness and life satisfaction. Therefore GDP
per capita along with factors should be important to policymakers.
24
References
Alesina, A., Di Tella, R. & MacCulloch, R. (2004), ‘Inequality and happiness:
are europeans and americans different?’, Journal of public economics 88(9-
10), 2009–2042.
Bjørnskov, C. (2006), ‘The multiple facets of social capital’, European journal of
political economy 22(1), 22–40.
Blanchflower, D. G. & Oswald, A. J. (2004), ‘Well-being over time in britain and
the usa’, Journal of public economics 88(7-8), 1359–1386.
Blanchflower, D. G. & Oswald, A. J. (2008), ‘Is well-being u-shaped over the life
cycle?’, Social science & medicine 66(8), 1733–1749.
Clark, A. E., Frijters, P. & 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. & Oswald, A. J. (1994), ‘Unhappiness and unemployment’, The Eco-
nomic Journal 104(424), 648–659.
Diener, E., Emmons, R. A., Larsen, R. J. & Griffin, S. (1985), ‘The satisfaction
with life scale’, Journal of personality assessment 49(1), 71–75.
Dolan, P., Peasgood, T. & 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.
Easterlin, R. A. (1974), Does economic growth improve the human lot? some
empirical evidence, in ‘Nations and households in economic growth’, Elsevier,
pp. 89–125.
25
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. (2006), ‘Life cycle happiness and its sources: Intersections
of psychology, economics, and demography’, Journal of economic psychology
27(4), 463–482.
Easterlin, R. A., McVey, L. A., Switek, M., Sawangfa, O. & Zweig, J. S. (2010),
‘The happiness–income paradox revisited’, Proceedings of the National Academy
of Sciences 107(52), 22463–22468.
European Social Survey, . (n.d.a), ‘About ess — european social survey’, http:
www.europeansocialsurvey.org/about/. (Accessed on 11/11/2021).
European Social Survey, . (n.d.b), ‘Data by theme — european social sur-
vey (ess)’, https://www.europeansocialsurvey.org/data/themes.html?t=
wellbeing. (Accessed on 11/12/2021).
European Social Survey, . (n.d.c), ‘Om ess — european social survey’, https:
//www.europeansocialsurvey.org/about/country/sweden/. (Accessed on
11/14/2021).
European Social Survey, . (n.d.d), ‘Sampling — european social survey’,
https:www.europeansocialsurvey.org/methodology/ess_methodology/
sampling.html. (Accessed on 11/11/2021).
European Social Survey, . (n.d.e), ‘Sampling — european social survey (ess)’,
https://www.europeansocialsurvey.org/methodology/ess_methodology/
sampling.html. (Accessed on 11/11/2021).
Helliwell, J. F. & Putnam, R. D. (2004), ‘The social context of well–being’, Philo-
26
sophical Transactions of the Royal Society of London. Series B: Biological Sci-
ences 359(1449), 1435–1446.
Kahneman, D., Diener, E. & Schwarz, N. (1999), Well-being: Foundations of
hedonic psychology, Russell Sage Foundation.
Kahneman, D. & Krueger, A. B. (2006), ‘Developments in the measurement of
subjective well-being’, Journal of Economic perspectives 20(1), 3–24.
Lelkes, O. (2006), ‘Tasting freedom: Happiness, religion and economic transition’,
Journal of Economic Behavior & Organization 59(2), 173–194.
Lewis, C. A., Maltby, J. & Day, L. (2005), ‘Religious orientation, religious cop-
ing and happiness among uk adults’, Personality and Individual differences
38(5), 1193–1202.
Sacks, D. W., Stevenson, B. & Wolfers, J. (2010), Subjective well-being, income,
economic development and growth, Technical report, National Bureau of Eco-
nomic Research.
Stevenson, B. & Wolfers, J. (2008), Economic growth and subjective well-being:
Reassessing the easterlin paradox, Technical report, National Bureau of Eco-
nomic Research.
Stevenson, B. & Wolfers, J. (2009), ‘The paradox of declining female happiness’,
American Economic Journal: Economic Policy 1(2), 190–225.
Stone, A. A., Schwartz, J. E., Broderick, J. E. & Deaton, A. (2010), ‘A snapshot of
the age distribution of psychological well-being in the united states’, Proceedings
of the National Academy of sciences 107(22), 9985–9990.
Sumner, L. W. (1996), Welfare, happiness, and ethics, Clarendon Press.
27
Tella, R. D., MacCulloch, R. J. & Oswald, A. J. (2003), ‘The macroeconomics of
happiness’, Review of Economics and Statistics 85(4), 809–827.
World bank, . (n.d.), ‘Gdp per capita, ppp (constant 2017 international $) — data’,
https:data.worldbank.org/indicator/NY.GDP.PCAP.PP.KD. (Accessed on
14/11/2022).
28
7 Appendix
Table 4: Happniess and Life satisfaction correlation
Variable Happiness Life satisfaction
Happniess 1.000
Life satisfaction 0.708 1.000
29
Table 5: Multicollinearity (VIF test)
Variable VIF
Age squared 40.80
Age 39.35
East Europe 3.18
GDP per capita (Log) 2.53
West Europe 1.76
Trust in legal the system 1.53
Trust in country’s parliament 1.51
Employment status 1.47
Marital status 1.40
Great Britain and Ireland 1.39
Children living at home or not 1.37
Health 1.26
Years of education 1.24
Social meetings 1.18
Religious 1.16
Female 1.06
Placement on left right scale 1.05
Mean VIF 6.07
30