Beyond the Joneses: Inter-country income comparisons and happiness

The Journal of Socio-Economics 45 (2013) 187–195

Contents lists available at SciVerse ScienceDirect

The Journal of Socio-Economics journal homepage: www.elsevier.com/locate/soceco

Beyond the Joneses: Inter-country income comparisons and happiness Leonardo Becchetti a,∗ , Stefano Castriota a , Luisa Corrado a,b , Elena Giachin Ricca a a b

Department of Economics, Università Tor Vergata, Via Columbia 2, 00133 Roma, Italy CreMic, CIMF University of Cambridge, United Kingdom

a r t i c l e

i n f o

Article history: Received 19 June 2012 Received in revised form 7 April 2013 Accepted 3 May 2013 JEL classification: D31 E01 I31 J61

a b s t r a c t Our paper provides novel evidence on the burgeoning literature on life satisfaction and relative comparisons by showing that in the last 30 years comparisons with the well being of top income neighboring countries have generated negative feelings on a large sample of individuals in the Euro barometer survey. The paper shows that neighboring countries, and not just our individual neighbors or peers, can be reference groups and that the above mentioned effect depends on the intensity of media exposure. © 2013 Elsevier Inc. All rights reserved.

Keywords: Life satisfaction Relative income Standard of living

1. Introduction Economists have become progressively aware of the importance of others and of relative comparisons for individual wellbeing. Such relevance has been recently confirmed by multi country experiments (Corazzini et al., 2010; White, 2013) where individuals face trade-offs between group ranking and absolute payoffs. These experiments document that many of them prefer being first, even at the cost of a lower income, and that such preference is associated to male gender, higher education and residence in a high income country. While a first traditional field in which relative preferences were taken into account was the literature of wage fairness in labor economics (see, among others, Rees, 1993 and Fehr et al., 2007), a more recent field of inquiry in which the same question has been investigated is the life satisfaction literature. The merit of this new burgeoning literature has been not just that of assuming a priori a structure of preferences which include others, but rather that of illustrating directly how objective measures of differences

∗ Corresponding author. Tel.: +39 06 36300723; fax: +39 06 2020500. E-mail addresses: [email protected] (L. Becchetti), [email protected] (S. Castriota), [email protected] (L. Corrado), [email protected] (E.G. Ricca). 1053-5357/$ – see front matter © 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.socec.2013.05.009

in performance with respect to reference groups may affect our satisfaction. From a theoretical point of view contributions from this literature (Duesenberry, 1949; Frank, 2005 and Layard, 2005) argue that positional competition with peers may generate “treadmill effects”, up to the extreme case of fully relative preferences where only relative - and not absolute - income matters. In such case it may paradoxically happen that an increase in personal income, if paralleled by an equal increase in income of all individuals in the reference group, does not affect individual life satisfaction. From an empirical perspective a starting point in this literature has been the introduction, in standard life satisfaction estimates, of the income of variously defined reference groups. Such groups have been generally created by combining geographical location, gender, age cohorts and professional characteristics (Ferrer-i-Carbonell, 2005; Dorn et al., 2008; Clark and Senik, 2010), even though systematic biases may arise from the extrapolation of information about the income distribution extracted from reference groups (Cruces et al., 2013). This literature has shown that, while relative income matters, positional effects do not fully crowd out the positive impact of personal income on individual well being (BarringtonLeigh and Helliwell, 2008). Furthermore, several studies have documented that an increase in the reference group income may become not necessarily bad news for individuals living (or perceiving to live) in socio economic environments characterized by high vertical mobility (Senik, 2004; Jiang et al., 2009; Becchetti

188

L. Becchetti et al. / The Journal of Socio-Economics 45 (2013) 187–195

and Savastano, 2009). The same literature has been extended to the role played by various inequality dimensions (income, weight) between partners and to that exerted by regional unemployment on the satisfaction of the unemployed (Clark, 2012). Summing up, the main question in the agenda of the literature on life satisfaction and relative comparisons remains: who compares with whom and with which intensity (constant or time varying)? Our paper aims to provide a novel contribution in this field. The two considerations from which we start are that: (i) individuals do not just look at others, but conventionally tend to compare the quality of life of their country with that of others and (ii) the effects of comparisons with other countries depend on media exposure of each individual. Based on these considerations we document, with an econometric analysis on the Euro barometer survey, that life satisfaction has been increasingly negatively affected by the national gross disposable per capita income (GDI) of neighboring countries. In order to obtain our findings, we follow the literature in constructing measures of average income but for different “levels” of geography (see, for example, Aslam and Corrado, 2012) by including own individual absolute income along with relative income in the same region, mean income in the same country and mean (max) income of the neighboring countries. In doing so we expect coefficients on mean income and GDI variables relative to neighboring countries to be negative, but decreasing in absolute value as one gets further away (geographically) from the individual under consideration.1 This is what we find in our results. Based on what considered above the contribution of our paper is fourfold. First, we consider that countries, and not just group of peers, may be reference groups. When doing so, we obviously control whether the country relative effect persists after controlling for various types of standard reference group effects2 . In this respect our findings provide additional insights on the well known treadmill effects and Easterlin paradox. In fact, it has been mentioned above that, under the extreme case in which only relative and not absolute income matters, an equiproportional increase in individual economic well being leaves individual life satisfaction unaffected. Our results imply that life satisfaction may even fall if this event is paralleled by a higher increase in per capita income of peer countries. Second, we show that the mean is not the only relevant moment of the distributions on which relative comparisons are drawn. More specifically, we document that the neighboring country with the maximum national gross disposable income is important since, in our case, it identifies a peak of average well being which has been achieved in some parts of the world and becomes desirable for those who have lower living standards. Third, we show that there are asymmetric responses as neighboring countries get richer. We measure this effect by estimating an Ordered Probit model with Generalized Thresholds (see Kapteyn et al., 2007 and Boes and Winkelmann, 2010). Heterogeneity in this

1 We could also consider how the richest European country (by year) affects individual happiness. However, this would be perfectly collinear with the year dummies and dropped from our estimations. 2 The implication that an improvement in well being occurred abroad may reduce life satisfaction of individuals in a given country is that well being innovations which historically originate in a first pioneering country may generate protests and manifestations in others where individuals feel worse off until they can catch up. This argument may be supported by several historical anecdotes. To provide an example, the eight hour working day was introduced in countries such as New Zealand before the 20th century while it became law in many others only between 1916 and 1925. Before its gradual implementation in other countries the introduction of the reform in New Zealand generated protests and manifestations outside it. A similar historical process can be observed for the introduction of the vote for women which occurred first in New Zealand in 1893 and, later, in other high income countries after demonstration and popular unrest.

generalized model enters by allowing the cut-off points associated with the different response categories for life satisfaction to be a linear function of the same set of regressors of the (latent) mean equation for life satisfaction, therefore making the vector of parameters in the threshold equation category specific. Following this approach we find that, as neighboring countries get richer, this impacts more negatively individuals who are either “Fairly satisfied” or “Very satisfied” than those who are “Not very satisfied”. Fourth, by exploiting within-country differences in accessibility to information about the income distribution in other countries we document that the salience of comparisons of domestic well being with that of other countries grows with the intensity of media use (reading newspaper, listening to radios, watching television, using internet). According to our reasoning, individuals with reduced contact with the rest of the world (i.e. never read newspapers, etc.) should be significantly less affected by changes in the income of other neighboring countries. We test this hypothesis by considering an interaction term between the “media exposure variable” (which varies across individuals) and the maximum national gross disposable per capita income of neighboring countries. In our conclusions we leave to further research the investigation of the possible missing link between our results and the determinants of migration. As it is well known several empirical contributions demonstrate that migratory flows are function of the income gap between country of origin and country of destination. Todaro (1969) and Harris and Todaro (1970) are the seminal papers claiming that migration is determined by wage differentials among geographical areas: these works document that migration is driven by expected rather than actual wage differentials3 . However, if we rule out cases of absolute necessity, the wage differential is a non sufficient condition for triggering migratory movements since the decision to move occurs if the income gap between countries has negative effects on individual well being: such negative relationship is the main result of our paper. 2. Database Our source of data is the Euro barometer Survey on Western European countries from 1973 to 2003 (except 1974 and 1996). The database contains information on individual characteristics and self-declared happiness and it is available until 2009 also for new EU members and for candidate countries. However, after 2003 personal income has not been recorded anymore4 . For this reason, we prefer to rely only on data for Western European countries. We also have some country-year gaps since data for Norway is available from 1990 to 1996, for Finland from 1993 onwards, for Sweden and Austria from 1994. Tables 1a and 1b provide a detailed description of the variables used. Data for our dependent variable, self-declared life satisfaction, is drawn from the question “On the whole, are you very satisfied, fairly satisfied, not very satisfied or not at all satisfied with the life you lead?”. The original values have been rescaled from descending to ascending order of life satisfaction intensity in order to have more intuitive

3 For later works see, among others, Mundell (1957), Borjas (1989, 1999a,b) and Venables (1999). These papers document that, beyond the gap in economic wellbeing, a number of other variables can influence migratory flows such as quality of life, differences in political stability, human rights situations, and the general rule of law which may be considered as a proxy for the level of individually perceived insecurity. 4 After 2003 we cannot rely anymore on information related to “household before taxes per month nominal income” (this variable was surveyed on a national base with categories created according to the national distribution of income). Although we deal with household income, in the paper we often refer to it as personal or individual income. This underlines the financial situation which is relevant to the individual in comparison with other groups.

L. Becchetti et al. / The Journal of Socio-Economics 45 (2013) 187–195

189

Table 1a Variable description. Name

Source

Variable

Life satisfaction

Eurobarometer

Unemployed

Eurobarometer

Employed

Eurobarometer

Retired

Eurobarometer

Student

Eurobarometer

Housewife

Eurobarometer

Male

Eurobarometer

Age Age squared Middle education

Eurobarometer Eurobarometer Eurobarometer

High education

Eurobarometer

Married

Eurobarometer

Divorced

Eurobarometer

Separated

Eurobarometer

Widowed

Eurobarometer

Income class Media Usage

Eurobarometer Eurobarometer

GDP growth

Ameco

Unemployment Inflation LGDI per capita (Country)

OECD World Bank Ameco

Income * LGDI

Eurobarometer, OECD

Relative income (Region)

Eurobarometer

Mean LGDI per capita (Neighboring Countries)

Ameco

Max LGDI per capita (Neighboring Countries)

Ameco

Self-declared life-satisfaction level from 1 (not at all satisfied) to 4 (very satisfied) DV (Dummy Variable) which takes value 1 if the respondent is unemployed, 0 otherwise DV which takes value 1 if the respondent is employed (except self-employment), 0 otherwise DV which takes value 1 if the respondent is retired, 0 otherwise DV which takes value 1 if the respondent is student, 0 otherwise DV which takes value 1 if the respondent is responsible for home and not working, 0 otherwise DV which takes value 1 if the respondent is male, 0 otherwise Age of the respondent in years Square of the respondent’s age in years DV which takes value 1 if the respondent has 15–18 years of education, 0 otherwise DV which takes value 1 if the respondent has more than 18 years of education, 0 otherwise DV which takes value 1 if the respondent is married, 0 otherwise DV which takes value 1 if the respondent is divorced, 0 otherwise DV which takes value 1 if the respondent is separated, 0 otherwise DV which takes value 1 if the respondent is widowed, 0 otherwise Income ranging from 1 (min) to 13 (max) “About how often do you watch the news on television? Read the news in the daily papers? Listen to news on the radio?” Index ranging from 0 to 9. GDP per capita growth rate (in %) in constant 2000 terms Unemployment rate (in %) Inflation rate (in %) Logarithm of gross national disposable income per capita in PPS Proxy of own (absolute) income given by the interaction dummy Income class*LGDI per capita Average income of the reference group by gender, age, education, year and region Mean per capita gross national disposable income of neighboring Countries (by year in log) Max per capita gross national disposable income of neighboring Countries (by year in log)

Table 1b List of neighboring countries. Country

Neighboring countries

France Belgium Netherlands Germany Italy Luxembourg Denmark Ireland United Kingdom Greece Spain Portugal Finland Sweden Austria

Andorra*, Belgium, Germany, Italy, Luxembourg, Monaco*, Spain, Switzerland* France, Germany, Luxembourg, Netherlands Belgium, Germany Austria, Belgium, Czech Republic, Denmark, France, Luxembourg, Netherlands, Poland, Switzerland Austria, France, San Marino*, Slovenia, Switzerland, Vatican City* Belgium, France, Germany Germany United Kingdom Ireland Albania*, Bulgaria, Turkey, Macedonia Andorra*, France, Gibraltar*, Portugal, Morocco* Spain Norway, Sweden, Russia Finland, Norway Czech Republic, Germany, Hungary, Italy, Liechtenstein*, Slovakia, Slovenia, Switzerland

An * denotes a country not in the sample.

190

L. Becchetti et al. / The Journal of Socio-Economics 45 (2013) 187–195

Table 2a Summary statistics of micro variables. Variable

Obs.

Mean

Std. Dev.

Min.

Max

Life satisfaction Male Age Middle education High education Married Separated Widowed Student Unemployed Retired Employed Housewife Income (class) Media usage

980,611 1,465,630 1,404,878 1,334,011 1,334,011 1,332,110 1,332,110 1,332,110 1,419,096 1,419,096 1,419,096 1,419,096 1,419,096 813,226 515,064

3.00 0.47 44.31 0.37 0.21 0.56 0.06 0.09 0.10 0.06 0.21 0.40 0.14 6.46 1.98

0.77 0.50 18.07 0.48 0.41 0.50 0.23 0.28 0.29 0.24 0.41 0.49 0.35 3.35 1.21

1 0 15 0 0 0 0 0 0 0 0 0 0 1 0

4 1 99 1 1 1 1 1 1 1 1 1 1 13 9

results (very satisfied = 4, up to not at all satisfied = 1). Personal information about respondents includes gender, age, education, civil and employment status, and personal income. This latter variable is not reported in local currency, but rather in relative terms at the country-year level and recorded from 1 to 13. We convert it into a proxy for absolute income by considering the interaction dummy between the different individual income classes (Income) and the (log) of per-capita gross disposable income in each country (LGDI).5 To allow for a deeper insight on the role exerted by interpersonal income comparisons we create an additional variable (Relative income) measuring the average income level of the reference group by age, education level,6 gender, year and region.7 Note that the number of observations for this latter variable is higher than that of personal income because, even if the respondent did not declare her wealth, it is possible to obtain the two measures mentioned above if she provided all the information required to identify her reference group. Table 2a provides summary statistics for the micro regressors used in the econometric analysis. The database on Western EU countries over the period 1973–2002 is composed by almost one and a half million people, 980.000 of whom provided a selfevaluation of their happiness level. Life satisfaction ranges from one to four with a mean of three and a standard deviation of 0.77, which ensures a good variability of the dependent variable in the regressions. 47 percent of individuals in the sample are males, 21 percent of them have a university degree and 56 percent are married. Six percent are unemployed. Macroeconomic controls include unemployment, inflation (growth rate of consumer prices), GDP growth rate and the log of gross national disposable income (LGDI) per capita in purchasing power standards to allow for a better comparability among countries. Following the standard literature on happiness, macroeconomic data is either annual or extracted as three year moving-average centered in t-1 in order to reduce possible measurement errors (see, among others, Di Tella et al., 2001 and 2003). Unemployment rates come from the OECD Center for Economic Performance dataset, inflation rates from the World Bank’s World

5 The income regressor is traditionally measured in logs in life satisfaction estimates (for a methodological discussion on this point see, among others, Stevenson and Wolfers, 2008). 6 More specifically, the sample has been divided into thirteen age classes (17–21, 22–26, 27–31, 32–36, 37–41, 42–46, 47–51, 52–56, 57–61, 62–66, 67–71, 72–76, more than 76) while the education level can be low (less than 15 years of schooling), medium (15–18 years) or high (more than 18). 7 For regional averages we consider 175 European regions.

Development Indicators and GDP growth rates and GDI per capita from Ameco, the annual macro-economic database of the European Commission’s Directorate General for Economic and Financial Affairs (DG ECFIN).8 Table 1a provides the source and the description of the variables used. According to the OECD, GDI per capita “may be derived from gross national income by adding all current transfers in cash or in kind receivable by resident institutional units from non-resident units and subtracting all current transfers in cash or in kind payable by resident institutional units to non-resident units”9 . Due to these characteristics, we regard GDI as better suited than GDP for representing the flow of economic resources which circulates in a geographical area and therefore a proxy for the standard of living. Since the main target of the paper is to analyze the impact of inter-country income comparisons on happiness, given the list of neighboring Countries in Table 1b, we create the following two variables: (i) mean of per capita gross national disposable income of neighboring Countries (by year); (ii) maximum of per capita gross national disposable income of neighboring Countries (by year). More in detail, for every year we calculate the maximum LGDIneigh max and the average LGDIneigh mean of the (log) pr capita Gross Disposable Income of neighboring countries. We start by focusing on maximum LGDIneigh max and specify the following relationship: neigh max

∗ = ˇ0 + ˇ1 LGDIt,c + ˇ2 LGDIt,k =/ c lifesatit,c

+ + +

4

=1

P p=1

T t=2

+ ˇ3 Relative Incomeit

ˇ4 (Incomeit ∗ LGDIt,c ) + ˛1p Ppit +

M

˛4t Yt + it

m=1

S

˛2m Mmt,c +

s=2

ˇ5s Incomesit

C c=2

˛3c Cc (1)

where life satisfaction of individual i at time t and living in country c depends on own country-year GDI (LGDIt,c ), on the neigh max maximum GDI of neighboring countries, LGDIt,k =/ c , on relative income (relincomeit ), and on the proxy for individual absolute income with a very flexible functional form using a fourthdegree polynomial for the proxy of individual absolute income,  4 ˇ (Incomeit ∗ LGDIt,c ) , plus dummies for the individual =1 4

8 We used the GDP instead of GDI growth rate because it is the economic growth indicator usually reported by the media which is expected to influence people’s expectation over the future development of the domestic economy. 9 http://stats.oecd.org/glossary/detail.asp?ID=1175.

L. Becchetti et al. / The Journal of Socio-Economics 45 (2013) 187–195

191

Table 2b Summary statistics of GDI. Variable

Obs.

Mean

Std. Dev.

Min.

Max.

LGDI per capita (Country) Mean LGDI per capita (Neighboring Countries) Max LGDI per capita (Neighboring Countries) Income * LGDI Relative income (Region)

1,386,769 1,284,005 1,284,005 776,227 1,406,272

2.73 2.71 2.88 17.25 15.22

0.53 0.58 0.66 10.08 7.39

0.76 0.00 0.00 0.76 0.76

3.95 3.45 3.95 45.6 45.61

S

income classes10 ˇ Incomesit . Life satisfaction also depends s=2 5s on a set of individual controls P, on a set of country-year macroeconomic controls M, and finally on a set of country and year dummies denoted as C and Y respectively. We estimate Eq. (1) using an Ordered Probit specification for the ∗ : (unobserved) latent variable of life satisfaction lifesatit,c ∗ lifesatit,c = j if ωj−1 < lifesatit,c < ωj

for j = 1, . . . , J

10

S = 1 denotes the lowest income class (reference category). See also Chongvilaivan and Powdthavee (2012).

3. Econometric analysis 3.1. One stage regressions

(2)

where lifesatit,c are the observed response outcomes and ωj with j = 1, . . ., J denoting the response thresholds. The specification in (2) allows us also to test for a more flexible framework (the Generalized Ordered Probit) where the effects of income in neighboring countries and of other characteristics are heterogeneous across the different levels of life satisfaction. We therefore consider a further specification where the set of thresholds is given by ωij = ωj + Xit j where X is the vector of all the explanatory variables in (1). Heterogeneity enters the generalized model given in (1) and (2) by allowing the threshold values ωij to be a linear function of the same set of regressors of the (latent) equation for life satisfaction as defined in (1) making the vector of parameters j and therefore the coefficients in Eq. (1) category specific.11 In essence, we follow the literature in constructing measures of average income but for different “levels” of geography (see, for example, Aslam and Corrado, 2012). In other words, we include own individual absolute income along with relative income in the same region, mean income in the same country and max income of the neighboring countries. We expect the coefficients on mean income and GDI variables relative to neighboring countries to be negative, but decreasing in absolute value as one gets further away (geographically) from the individual. Note that the key variable, the max GDI, varies each year since different countries have different neighbors. This allows us to get around the problem of perfect collinearity with the year effects. We further test whether the salience of comparisons of domestic well being with that of other countries grows with the intensity of media use. We create an interaction dummy between the max GDI in neighboring countries and the following individual Euro barometer index on media use which ranges from 0 to 9: “About how often do you watch the news on television? Read the news in the daily papers? Listen to news on the radio?” We introduce this interaction dummy in our estimation to assess whether people reporting higher intensity of media use are significantly more affected by changes in income of the richest neighboring country. Table 2b provides summary statistics for the GDI per capita and the related variables. The GDI per capita of the richest country is 35 percent higher than that of the average sample, the max-

11

imum gap being 3.95: differences among countries can be non negligible given the range of countries and years considered in our sample.

We start our empirical analysis by running one-stage regressions of the self-declared life satisfaction over country and year dummy variables and the set of standard controls listed in Table 1a. Given the discrete nature of the dependent variable (life satisfaction ranges from 1 to 4), we adopt a methodology similar to Di Tella and MacCulloch (2003) and run Ordered Probit regressions with standard errors adjusted for clusters at the country-year level (see Table 3). This is a severe but very important robustness check, especially when testing macro-economic variables which do not vary at the individual level. France, 1975, Low Education, Single, Self-employed and the lowest Income class are the omitted benchmarks to avoid the dummy variable trap. Since our main goal is to control for the effect of inter-country comparisons, we first run regressions without macroeconomic variables (column 1), then add yearly inflation, unemployment, yearly GDP growth (column 2), and log of country-year per capita GDI (column 3). The last two columns are the most important since we add, as further regressors, the maximum GDI per capita in neighboring countries (by year), the relative income variable calculated at the regional level (column 4) and the interaction dummy for the intensity of comparisons induced by media use (column 5). Looking at column 1 of Table 3, coefficients of microeconomic variables are in line with previous happiness research. Negative coefficients are associated with male gender, being separated, widowed, unemployed and employed. Positive coefficients are associated with higher education, being married, student, and having a high income, while age is U-shaped. Subsequent regressions include standard macroeconomic controls like GDP growth rate, inflation and unemployment. The estimates of main interest to the purpose of our inquiry are those from column 3 onwards where we add the (log) GDI per capita, the measure of relative regional income, and interaction dummy variables to control for intensity of media use. GDI per capita is always insignificant when further macro variables at the country level are added. Looking at the fourth column, the Max GDI per capita of the neighboring countries in the year considered (see equation 1 in paragraph 2) is highly significant and with a negative sign. We also find that people reporting higher intensity of media use are significantly more affected by changes in income of the richest neighboring country (column 5). These findings document that countries, and not just neighbors, are reference groups significantly affecting subjective well being.

192

L. Becchetti et al. / The Journal of Socio-Economics 45 (2013) 187–195

Table 3 One-stage regressions (Ordered probit).

Male Age Age2 Middle education High education Married Separated Widowed Student Unemployed Retired Employed Housewife Income* LGDI

(i)

(ii)

(iii)

(iv)

(v)

−0.059*** (0.01) −0.029*** (0.00) 0.001*** (0.00) 0.066*** (0.01) 0.130*** (0.01) 0.119*** (0.01) −0.239*** (0.01) −0.141*** (0.01) 0.168*** (0.02) −0.521*** (0.02) −0.004 (0.01) −0.037*** (0.01) −0.004 (0.01) 0.088 (0.06)

−0.060*** (0.01) −0.029*** (0.00) 0.001*** (0.00) 0.067*** (0.01) 0.133*** (0.01) 0.123*** (0.01) −0.238*** (0.01) −0.140*** (0.01) 0.168*** (0.02) −0.519*** (0.02) −0.005 (0.01) −0.036*** (0.01) −0.005 (0.01) −0.023 (0.07)

−0.060*** (0.01) −0.029*** (0.00) 0.001*** (0.00) 0.067*** (0.01) 0.133*** (0.01) 0.123*** (0.01) −0.238*** (0.01) −0.140*** (0.01) 0.168*** (0.02) −0.518*** (0.02) −0.005 (0.01) −0.036*** (0.01) −0.005 (0.01) −0.016 (0.05)

−0.053*** (0.01) −0.027*** (0.00) 0.001*** (0.00) 0.070*** (0.01) 0.137*** (0.01) 0.115*** (0.01) −0.248*** (0.01) −0.155*** (0.01) 0.165*** (0.02) −0.552*** (0.02) −0.008 (0.01) −0.041*** (0.01) −0.008 (0.01) −0.018 (0.05) −0.004*** (0.00) −0.074 (0.19) −0.226** (0.12)

√ √ √ √

0.006 (0.01) −0.011*** (0.00) −0.003 (0.00) √ √ √ √

0.006 (0.01) −0.011** (0.01) −0.003 (0.00) √ √ √ √

0.007 (0.01) −0.007 (0.01) −0.003 (0.00) √ √ √ √

−0.060*** (0.01) −0.030*** (0.00) 0.001*** (0.00) 0.052*** (0.01) 0.111*** (0.02) 0.105*** (0.02) −0.232*** (0.02) −0.146*** (0.02) 0.204*** (0.03) −0.531*** (0.04) 0.021 (0.02) 0.000 (0.01) 0.025* (0.01) −0.145 (0.11) −0.002* (0.00) −0.402 (0.29) 0.039 (0.13) −0.020*** (0.00) 0.014** (0.01) −0.020*** (0.01) −0.005 (0.01) √ √ √ √

417,674 0.0962

413,985 0.0965

413,985 0.0965

348,943 0.0965

157,962 0.1050

Relative income (Region) −0.020 (0.18)

LGDI per capita (Country) Max LGDI (Neigh Countries) Max LGDI (Neigh Countries)* Media Use GDP growth Unemployment Inflation Income class dummies Fourth-degree income polynomial Year dummies Country dummies Observations Pseudo R2

***p < 0.01, **p < 0.05, *p < 0.1. The dependent variable is life satisfaction which ranges from 1 (not at all satisfied) to 4 (very satisfied). Regressions are Ordered Probitwith standard errors adjusted for clusters at the country-year level. France, 1975, Low Education, Single, Self-employed and Income class 1 are the base to avoid the dummy variable trap. For the proxy of individual income we specify a fourth degree polynomial (Income*LGDI), (Income*LGDI)2 , (Income*LGDI)3 , (Income*LGDI)4 plus twelve income class dummies (starting from the 6th income class the dummies are significant). For reasons of space we report only the first income polynomial component (Income*LGDI) and omit the 12 income class dummies. Standard errors are in parentheses.

In order to give an idea of the economic magnitude of the variables under scrutiny, we report in Table 4 the marginal fixed effects of the fourth regression of Table 3 (only statistically significant variables are reported). A 1% increase in own reference group income (computed by gender, age, education, year and region) reduces the probability to declare oneself very happy by 0.12%. Having a high education level increases the same probability by 4.38%, being married by 3.52% and student by 5.39%. On the contrary, being separated, widowed and unemployed reduces it respectively by 7.09%, 4.58% and 14.03%. Not surprisingly and consistently with general findings from this literature, the unemployment status has the most dramatic effect on people’s lives. Last, but not least, if the GDI of the richest neighboring country increases by 1 percent, the probability to declare the maximum level of life satisfaction decreases by 7.04%. Inter-country income comparisons have, therefore, not only a statistically but also an economically significant effect on human well being.

Finally, we show that, as other countries get richer, this will have different distributional impacts on “happiness”. For instance, a standard Ordered Probit model may underestimate the effects of an increase in the income of neighboring countries on the “highly” satisfied individuals, whilst overestimating the incidence on the “very dissatisfied” individuals. We account for scale heterogeneity between income in neighboring countries and happiness by estimating an Ordered Probit model with Generalized Thresholds (Table 5) and find that as neighboring countries get richer, this impacts more negatively, individuals who are “Fairly Satisfied” (−0.29) and “Very Satisfied” (−0.28) than those who are “Not Very Satisfied” (−0.24). 3.2. Robustness checks In order to test the robustness of our results we adopt five additional econometric strategies: (i) one stage regressions where

L. Becchetti et al. / The Journal of Socio-Economics 45 (2013) 187–195 Table 4 Marginal fixed effects. Variable

dy/dx

Male Middle education High education Married Separated Widowed Student Unemployed Employed Relative income (Region) Max LGDI per capita (Neighboring Countries)

−1.65% 2.19% 4.38% 3.52% −7.09% −4.58% 5.39% −14.03% −1.26% −0.12% −7.04%

Marginal fixed effects refer to the fourth regression in Table 3 and show the effect of the variables on the probability to declare the maximum level of life satisfaction (very satisfied = 4). Only variables with statistically significant coefficients are reported. All the variables in the table are dummies except relative income (average reference group income defined on the basis of age, education, gender, year and region) and per-capita GDI of the richest neighboring country. For these two variables the marginal effects refer to a 1% increase either in relative income or in GDI since both are measured in logarithm.

the macro variables are three-year moving averages; (ii) one stage regressions where the variable of interest is the average per capita GDI of neighboring countries, rather than its peak; (iii) twostage regressions similar to Di Tella et al. (2001); (iv) two-stage regressions à la Donald and Lang (2007); (v) DF-beta test to control whether the results are driven by one or more specific countries. The use of three-year moving averages and the average per capita GDI of neighboring countries provides similar results (omitted for reasons of space and available upon request). The third robustness check consists in running two-stage regressions similar to Di Tella et al. (2001): in the first stage the happiness level is regressed on a standard set of microeconomic controls, while in the second we regress the average country-year error term (the “unexplained” component of the first stage regression) on the macroeconomic variables of interest. The reason for exploring the fourth procedure proposed by Donald and Lang (2007) is well explained by the authors in their paper. In panel datasets in which the dependent variable differs across individuals, but at least some explanatory variables (in our case the macroeconomic ones) are constant among all members of a group, standard asymptotic tests provide poor estimation of the final sample distribution12 . Following the approach set forth by the authors, the happiness level is regressed on a standard set of microeconomic controls and a set of countryyear dummy variables. In the second stage the coefficients of the joint country-year dummy variables obtained in the first stage (308 coefficients) are regressed on the macroeconomic variables of interest.13 Table 6 illustrates findings from these two methodologies: on the left hand side (columns 1 and 2) we show the results of the

12 “Under standard restrictions, the efficient estimator can be implemented by a simple two-step procedure, and the resulting t-statistic may have, under restrictions on the distribution of the group-level error, an asymptotic t-distribution as the number of observations per group goes to infinity. In addition, under more restrictive assumptions, when the same procedure is used in finite samples, the t-statistics have a t-distribution”, Donald and Lang (2007), p. 221. 13 With the two latter methodologies both the first and second stage are implemented with an OLS procedure without need of country-year clustering. In fact, in Di Tella et al. (2001) the coefficients of the first stage are irrelevant since the object of interest is the average error term. This latter variable is regressed in the second stage on a set of macroeconomic variables: this methodology provides one data for every country-year observation, therefore clusterisation at the country-year level is useless. Similarly, in Donald and Lang (2007) the target of the first stage are the coefficients of the country-year dummy variables, which are regressed in the second stage on a set of macroeconomic variables whose data are available for every country-year. Again, clusterization does not change in the second stage estimates.

193

second stage where the dependent variable is the average error term of the first stage (à la Di Tella et al., 2001), while, on the right hand side (columns 3 and 4), the results of the second stage where the dependent variable is the coefficient of the country-year dummy variables of the first stage (à la Donald and Lang, 2007). Results are consistent with those in Table 3. After having run a two-stage regression with the full sample (1973–2002, not shown in the table), as a robustness check we repeat the procedure (i) by dropping one year each time starting from 1973 (columns 1 and 3), thereby reducing at every iteration the size of the sample which becomes progressively more concentrated around 2002, and (ii) by considering 5 year moving windows from 1975–1979 until 1995–1999 (columns 2 and 4). Both procedures allow to verify the evolution over time of the weight people attach to the GDI of the richest neighboring country. However, with the latter procedure we have a similar number of observations in each regression. For reasons of space, results in Table 6 are summarized for a subset of iterations and exclude overlapping time windows. The first two columns show the coefficient of GDI of the richest neighboring country over time obtained in the second stage with the average error term as a dependent variable. In the first and second columns the coefficient is always statistically significant, although its magnitude changes over time. A similar path is observed in the third and fourth columns with the Donald and Lang (2007) methodology. Results are consistent with those in Table 3: individuals are sensitive to the GDI of the richest neighboring country. Our final robustness check (Table 7) is the DF-beta test performed over the fourth regression of Table 3 following the approach adopted by Frey and Stutzer (2000). The rationale for this check is that we have a limited number of countries and want to control whether our results are sensitive to the inclusion/exclusion of one of them. More in detail, in the first step we estimate our fully augmented model with country-year dummies but without macroeconomic variables. In the second step we build a dependent variable represented by coefficients of country dummies from the previous estimate and then regress it on variables of our base estimate (Table 3, column 4) which vary at country-year level. We then repeat this estimate by excluding any time one of the sample countries. For each repeated estimate the coefficient of interest (Max per capita LGDI of neighboring countries) is subtracted from the one obtained in the regression with all the countries and divided by the estimated standard error. The obtained ratio has a tdistribution and, if bigger than 1.96 in absolute terms, implies that the country excluded from the second regression drives the result of the first one with the full sample (the null of independence of our result from a country outlier is rejected). Table 7 reports in the first column the coefficients of the variable of interest obtained by excluding a certain country while, in the second, the DF-beta test, which is well below the critical value of 1.96 except for France. We have tested whether our results are sensitive to the inclusion of France re-estimating our base model (Table 3, column 4) by excluding France and find that the coefficient of interest (max per capita LGDI of neighboring countries) is still significant although changes its magnitude (from −0.22 to −0.36).14 All the five proposed methodologies confirm that inter-country income comparisons, especially among neighboring countries, are important determinants of individual well being.

14

Results of this additional estimation are available upon request.

194

L. Becchetti et al. / The Journal of Socio-Economics 45 (2013) 187–195

Table 5 Generalized ordered probit. Thresholds equations

Not very satisfied

Fairly satisfied

Very satisfied

Male

−0.059*** (0.01) −0.022*** (0.00) 0.001*** (0.00) 0.084*** (0.01) 0.180*** (0.02) 0.100*** (0.01) −0.232*** (0.02) −0.088*** (0.02) 0.227*** (0.03) −0.597*** (0.03) −0.052** (0.03) 0.005 (0.02) −0.068*** (0.02) 0.131 (0.10) −0.011*** (0.00) 0.129 (0.26) −0.241* (0.13) 0.005 (0.01) −0.024*** (0.01) −0.007** (0.00) √ √ √ √

−0.039*** (0.01) −0.027*** (0.00) 0.001*** (0.00) 0.082*** (0.01) 0.170*** (0.02) 0.120*** (0.01) −0.275*** (0.02) −0.137*** (0.01) 0.249*** (0.03) −0.620*** (0.03) −0.016 (0.02) −0.008 (0.01) −0.009 (0.01) 0.112** (0.06) −0.008*** (0.00) −0.184 (0.22) −0.292** (0.13) 0.012 (0.01) −0.012* (0.01) −0.005 (0.00) √ √ √ √

−0.045*** (0.01) −0.024*** (0.00) 0.001*** (0.00) 0.079*** (0.01) 0.168*** (0.01) 0.110*** (0.01) −0.201*** (0.02) −0.171*** (0.01) 0.134*** (0.02) −0.378*** (0.02) 0.018 (0.01) −0.069*** (0.01) 0.011 (0.01) −0.056 (0.06) −0.011*** (0.00) −0.152 (0.18) −0.278** (0.12) 0.010* (0.01) −0.008 (0.01) −0.001 (0.00) √ √ √ √

Age Age2 Middle education High education Married Separated Widowed Student Unemployed Retired Employed Housewife Income* LGDI Relative income (Region) LGDI per capita (Country) Max LGDI per capita (Neigh Countries) GDP growth Unemployment Inflation Income class dummies Fourth-degree income polynomial Year dummies Country dummies

***p < 0.01, **p < 0.05, *p < 0.1. The dependent variable is life satisfaction which ranges from 1 (not at all satisfied) to 4 (very satisfied). Regressions are Generalized Ordered Probit with standard errors adjusted for clusters at the country-year level. France, 1975, Low Education, Single, Self-employed and Income class 1 are the base to avoid the dummy variable trap. For the proxy of individual income we specify a fourth degree polynomial (Income* LGDI), (Income* LGDI)2 , (Income* LGDI)3 , (Income* LGDI)4 plus twelve income class dummies (starting from the 6th income class the dummies are significant). For reasons of space we report only the first income polynomial component (Income* LGDI) and omit the 12 income class dummies. Standard errors are in parentheses.

Table 6 Robustness checks with two-stage procedures. Y

1977 1982 1987 1992 1997

2nd stage on the error term

2nd stage on the country-year DV coefficient

(i) Years > Y

(ii) 5y moving windows

(iii) Years > Y

(iv) 5y moving windows

−0.369*** (0.036) −0.352*** (0.036) −0.346*** (0.044) −0.318*** (0.056) −0.253*** (0.064)

−0.592*** (0.175) −0.413*** (0.087) −0.298*** (0.053) −0.356*** (0.070) −0.324*** (0.063)

−0.620*** (0.066) −0.601*** (0.067) −0.582*** (0.083) −0.548*** (0.106) −0.465*** (0.127)

−0.902*** (0.284) −0.624*** (0.139) −0.484*** (0.091) −0.590*** (0.128) −0.555*** (0.116)

Robustness check on Max LGDI per capita (Neighboring Countries). Results refer to the second stage and come from OLS regressions. In columns 1 and 3 the sample is restricted by considering only the years after that shown in the first column (Y). In columns 2 and 4 the sample is restricted by using 5-year moving windows centered around the year shown in the first column (Y). Standard errors are in parenthesis.

L. Becchetti et al. / The Journal of Socio-Economics 45 (2013) 187–195 Table 7 DF-beta test.

195

References

Omitted country

Coefficient of GDI distance from max after 1990

DF-beta

France Belgium Holland Germany Italy Luxemburg Denmark Ireland Great Britain Greece Spain Portugal Finland Sweden Austria

−0.42 −0.72 −0.56 −0.55 −0.68 −0.50 −0.51 −0.65 −0.68 −0.59 −0.61 −0.46 −0.59 −0.59 −0.57

2.23 −1.91 0.46 0.57 −1.25 1.11 1.22 −0.92 −1.03 0.01 −0.41 1.90 0.01 0.01 0.30

The second column shows the coefficient of the variable “Max LGDI per capita of neighboring Countries” obtained when dropping from the fourth regression of Table 3 the country listed in the first column. The third column shows the DF-beta test which has a t-distribution.

4. Conclusions Our paper contributes to the literature on the relationship between life satisfaction and relative comparisons by showing that individuals are influenced not just by traditional reference groups but also by the well being levels of neighboring countries, and this occurs proportionally to their media exposure. More specifically, we document that the difference between their own and the richest neighboring country gross disposable income has a significant effect, net of the impact of traditional relative income measures in which the reference group at the local level is built by looking at age, education level, gender, year and region. We also document that the previous effect arises in proportion to media use. Since our main variable of interest (GDI) varies only at countryyear level we provide several robustness checks such as two stage procedures usually adopted in this case and ad hoc diagnostics aimed to verify whether our results are robust to the inclusion/exclusion of individual countries. We think that our findings open the way to several interesting considerations and potential directions of further research by showing that (i) reference groups may be nations, (ii) the relevance of comparisons depends on the intensity of media exposure and (iii) individuals do not look just at averages but also at the best performer in terms of standards of living. A suggestion for further research would be to relate our results to the stylized facts from the empirical analysis of migratory flows in order to check whether dissatisfaction arising from country well being gaps may be one of the hidden forces which pushes individuals to migrate abroad. Acknowledgement we thank Stefano Bartolini, Luigino Bruni, Andrew Clark, Rafael Di Tella, Benedetto Gui, Maurizio Pugno and all participants to the 2010 GSOEP conference for comments and suggestions on the issues raised in this paper. We also thank an anonymous referee for his comments on the work. The usual disclaimer applies.

Aslam, A., Corrado, L., 2012. The geography of well-being. Journal of Economic Geography 12, 627–649. Barrington-Leigh, C.P., Helliwell, J.F., 2008. Empathy and Emulation. Life Satisfaction and the Urban Geography of Comparison Groups. NBER Working Paper Series 14593. National Bureau of Economic Research, Cambridge. Becchetti, L., Savastano, S., 2009. The money-happiness relationship in transition countries: evidence from Albania. Transition Studies Review 17, 39–62. Boes, S., Winkelmann, R., 2010. The Effect of income on general life satisfaction and dissatisfaction. Social Indicators Research 95, 111–128. Borjas, G.J., 1989. Economic theory and international migration. International Migration Review 23, 457–485. Borjas, G.J., 1999a. The economic analysis of immigration. In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor Economics, vol. 3A. North-Holland, Amsterdam. Borjas, G.J., 1999b. Economic research on the determinants of immigration: lessons for the European union. World Bank technical paper 438, 18–22. Clark, A.E., 2012. Happiness, Habits and High Rank Comparisons in Economic and Social Life. SOEPpapers on Multidisciplinary Panel Data Research 452. The German Socio-Economic Panel (SOEP), DIW Berlin. Clark, E., Senik, C., 2010. Who compares to whom? the anatomy of income comparisons in europe. Econ. J. 120, 573–594. Chongvilaivan, A., Powdthavee, N., 2012. Do different work characteristics have different distributional impacts on job satisfaction? a study of slope heterogeneity in workers’ well-being. British Journal of Industrial Relations, http://dx.doi.org/10.1111/j.1467-8543.2012.00904.x. A prize to give for: an Corazzini, L., Faravelli, M., Stanca, L., 2010. experiment on public good funding mechanisms. Economic Journal 120, 944–967. Cruces, G., Perez-Truglia, R.N., Tetaz, M., 2013. Biased perceptions of income distribution and preferences for redistribution: evidence from a survey experiment. Journal of Public Economics 98, 100–112. Di Tella, R., MacCulloch, R., 2003. The Macroeconomics of happiness. Review of Economics and Statistics 85, 809–827. Di Tella, R., MacCulloch, R., Oswald, A., 2001. Preferences over inflation and unemployment: evidence from surveys of happiness. American Economic Review 91, 335–341. Donald, S.G., Lang, K., 2007. Inference with difference-in-differences and other panel data. Review of Economics and Statistics 89, 221–233. Dorn, D., Fischer, J., Kirchgässner, G., Sousa-Poza, A., 2008. Direct democracy and life satisfaction revisited: new evidence for Switzerland. Journal of Happiness Studies 9, 227–255. Duesenberry, J., 1949. Income, Savings, and the Theory of Human Behavior. Harvard University Press, Cambridge, MA. Fehr, E., Klein, A., Schmidt, K.M., 2007. Fairness and contract design. Econometrica 75, 121–154. Ferrer-i-Carbonell, A., 2005. Income and well-being: an empirical analysis of the comparison income effect. Journal of Public Economics 89, 997–1019. Frank, R., 2005. Does absolute income matter? In: Bruni, L., Porta, P.L. (Eds.), Economics and Happiness. Oxford University Press, Oxford. Frey, S., Stutzer, A., 2000. Happiness, economy and institutions. Economic Journal 110, 918–938. Harris, J., Todaro, M.P., 1970. Migration unemployment and development a two sectors analysis. American Economic Review 60, 126–142. Jiang, S., Lu, M., Sato, H., 2009. Happiness in the dual society of urban China: Hukou Identity, horizontal inequality and heterogeneous reference. Global COE Hi-Stat Discussion Paper Series 20. Kapteyn, A., Smith, J.P., Van Soest, A., 2007. Vignettes and self-reports of work disability in the United States and the Netherlands. American Economic Review 97, 461–472. Layard, R., 2005. Rethinking public economics: the implications of rivalry and habit. In: Bruni, L., Porta, P.L. (Eds.), Economics and Happiness. Oxford University Press, Oxford. Mundell, R.M., 1957. International trade and factor mobility. American Economic Review 47, 321–325. Rees, A., 1993. The role of fairness in wage determination. Journal of Labor Economics 11, 243–252. Senik, C., 2004. When information dominates comparison: a panel data analysis using Russian subjective data. Journal of Public Economics 88, 2099–2123. Stevenson, B., Wolfers, J., 2008. Economic growth subjective well-being. Reassessing the easterlin paradox, brookings papers on economic activity. Economic studies program. The Brookings Institution 39, 1–102. Todaro, M.P., 1969. A model of labor migration and urban unemployment in less developed countries. American Economic Review 69, 486–499. Venables, T., 1999. Trade Liberalisation and Factor Mobility: an Overview. In: Faini, R., de Melo, J., Zimmermann, K. (Eds.), Migration: The Controversies and the Evidence. Cambridge University Press, pp. 23–48. White, J., 2013. Parallel Lives Social comparison across national boundaries. In: Niilo Kauppi (Ed.), A Political Sociology of Transnational Europe. ECPR, Press, Colchester, pp. 223–242.