Social Indicators Research

, Volume 88, Issue 3, pp 497–529

Absolute Income, Relative Income, and Happiness

Authors

    • Department of EconomicsHaverford College
  • Kateryna Chernova
    • Exelon Corporation
Article

DOI: 10.1007/s11205-007-9217-0

Cite this article as:
Ball, R. & Chernova, K. Soc Indic Res (2008) 88: 497. doi:10.1007/s11205-007-9217-0

Abstract

This paper uses data from the World Values Survey to investigate how an individual’s self-reported happiness is related to (i) the level of her income in absolute terms, and (ii) the level of her income relative to other people in her country. The main findings are that (i) both absolute and relative income are positively and significantly correlated with happiness, (ii) quantitatively, changes in relative income have much larger effects on happiness than do changes in absolute income, and (iii) the effects on happiness of both absolute and relative income are small when compared to the effects several non-pecuniary factors.

Keywords

Subjective well-beingHappinessRelative income

1 Introduction

This paper examines empirical evidence on the relationship between income and happiness. The specific objective is to compare the importance of absolute income and relative income in determining happiness: how much does a person’s happiness depend on her income in absolute terms, and how much does her happiness depend on her income relative to the incomes of others? We also ask, to the extent that happiness does depend on absolute and/or relative income, how large is the effect of income on happiness compared to the effects of non-pecuniary factors such as personal relationships, employment and health?

Although the relationship between money and happiness has been of perennial interest to social scientists and philosophers, contemporary research on this issue was stimulated largely by Easterlin’s (1974) seminal paper. A prominent hypothesis presented in that paper was that people care exclusively or almost exclusively about their incomes relative to the national distributions of income in the countries in which they live, and that, given their relative income levels, they care little or not at all about their absolute income levels. Although a substantial body of subsequent literature has refuted the most extreme form of this hypothesis, Easterlin’s paper has led to the production of an important line of research on income and happiness.

In addition to being an intrinsically interesting area of academic research, questions about income and happiness have important policy implications. As Easterlin pointed out, if people care only about their relative incomes, then a distribution-neutral shift in a country’s GDP per capita will not make anyone happier. If this is true, policy-makers, donors of foreign aid, and anyone else hoping to improve the lives of people around the world should look for strategies other than simply promoting growth. In addition, a number of studies have shown how preferences over absolute versus relative income affect their behavior, and how optimal public policy will therefore depend on the weight people place on each of these measures of income. For example, Frank (1984), Frank and Hutchens (1993), Neumark and Postlewaite (1996), and Landers et al. (1996) show how concerns about relative status affect labor supply and wage profiles; Duesenberry (1949) and Kosicki (1987) study the effect of positional concerns on saving; and a number of studies have analyzed public finance and optimal tax policy when people care about their relative incomes (Boskin and Sheshinski 1978; Layard 1980; Oswald 1983; Ng 1987; Bagwell and Bernheim 1996; Ireland 1998; Frank 1985, 1996, 1997).

As one might expect, we find that there are statistically significant, positive relationships between absolute income and happiness and between relative income and happiness. By several quantitative measures, however, the effect of relative income is much larger than the effect of absolute income; and the effects on happiness of both absolute and relative income are dwarfed by the effects of the non-pecuniary factors we consider. In other words, money does buy at least some happiness, but having more than others around you matters more to happiness than simply having more. These results, however, come with an important caveat: the happiness that money (whether absolute or relative) can buy is small compared to the happiness provided by some things that money cannot usually or easily buy, like a spouse or domestic partner, a job, or good health.1 Although these results reject the strict version of Easterlin’s hypothesis—that, given their places in the income distribution, people don’t care at all about their absolute incomes—they do support the general proposition that relative concerns are very important.

2 Previous Literature

2.1 Measuring Happiness

In this paper, as in the literature upon which it builds, measures of individual happiness are based on responses to surveys that ask people to report how happy or satisfied they are with their lives, usually on some integer scale (e.g., 1 through 10) or by choosing a descriptive category (e.g., very happy, somewhat happy, somewhat unhappy, very unhappy). Because answers to these questions are based on respondents’ self-assessments, rather than on some objective measure (based, for instance, on neural activity in the brain or on observations of individual consumption bundles), they are known as measures of subjective well-being. Data is available from many surveys that have included questions about subjective well-being; prominent among these are the World Values Survey,2 the Eurobarometer,3 the General Social Survey,4 the British Household Panel Survey,5 and the Gallup International Millenium Survey.6 Numerous surveys of well-being in smaller populations have been conducted as well.

Though decreasingly so, survey measures of subjective well-being are still somewhat novel to economists. They have, however, long been a staple source of data for psychologists and sociologists, and there is a growing body of evidence for the validity and the interpersonal and intercultural comparability of these data. A number of studies have shown high correlations between self-reports of subjective well-being and other measures of well-being or happiness, such as reports of acquaintances, how often people smile, respondents’ recollections of good and bad events in their lives, and physiological measures such as neural activity, heart rate and hormone levels.7 Several studies have failed to find evidence of biases created by translation of survey questions about well-being into different languages, or by differences across societies in cultural traits such as “humility.”8 Several surveys of methodological issues associated with self-reports of subjective well-being come to generally positive conclusions about the validity of this kind of data,9 and Ng (1997) gives a general argument in favor of using cardinal measures of happiness in economic and policy analysis.

2.2 Non-monetary Determinants of Happiness

Many studies have examined the correlations between happiness and a host of potential explanatory variables, including age and sex,10 marital status and personal relationships,11 inflation and unemployment,12 political institutions and liberty,13 and income distribution.14 Recent surveys of this literature include Diener et al. (1995), Argyle (1999), Easterlin (2003) and Diener and Seligman (2004). Frey and Stutzer (2002) provide a book-length review of happiness and economics, and Veenhoven’s (2003) on-line World Database of Happiness contains comprehensive information on research related to subjective well-being.15 We discuss some of the major findings of this literature below in relation to the new results that we present.

2.3 Money and Happiness

On the relationship between income and subjective well-being, Easterlin’s paper was seminal. Easterlin (1974, p. 90) posed the broad question of whether there is “evidence that economic growth is positively associated with social welfare, i.e., human happiness.” In particular, he investigated whether (i) there is any evidence of a relationship between income and happiness within countries, and (ii) whether there is any evidence of a relationship between average levels of happiness and average incomes across countries. Easterlin based his arguments on data from a variety of surveys; most relevant to the present paper is his analysis of data collected by Cantril (1965) in 14 countries during the late 1950s and early 1960s. Along with a wide variety of demographic, social and economic questions, Cantril’s survey included an item in which respondents were asked to rate their happiness on a scale from 0 to 10.16

To assess the relationship between income and happiness within individual countries, Easterlin used Cantril’s data along with 27 other surveys conducted in a total of 20 countries. The categories used to define socioeconomic status differed among the surveys, but all of them used between two and five designations, based primarily on income, to categorize respondents. For each country, Easterlin compared the average happiness rating among members of the lowest socioeconomic group to the average happiness rating among members of the highest socioeconomic group. He found (p. 100) that “the results are clear and unequivocal. In every single survey, those in the highest status group were happier, on the average, than those in the lowest status group.”

To assess the relationship across countries between the average level of happiness and average income, Easterlin (p. 106) presented a scatterplot of the average happiness rating in each of the fourteen countries surveyed by Cantril against the natural log of income per capita. That scatterplot is reproduced here in Fig. 1, and the underlying data are presented in Table 1. Casual inspection of the scatterplot suggests that there is something of a positive relationship between happiness and income, and indeed an OLS line fitted to this data has a positive slope that is at least marginally statistically significant.17 Easterlin argued (p. 105), however, that when two influential cases and two countries with “unusual political circumstances” are deleted from the sample, any apparent positive association vanishes. And in fact, the slope of an OLS line through the remaining 10 points is not significantly different from zero.18 Particularly striking in the scatterplot is the lack of variation in average happiness levels across countries, especially when the four special cases are omitted. Easterlin acknowledged (p. 105) that such “picking and choosing among points is a dubious practice,” but concluded (pp. 105–106) that “[t]he happiness difference that one might expect on the basis of the within-country differences by economic status are not borne out by the international data.” He went on to cite several other surveys that similarly fail to provide strong evidence of a positive association between country averages of income and happiness.
https://static-content.springer.com/image/art%3A10.1007%2Fs11205-007-9217-0/MediaObjects/11205_2007_9217_Fig1_HTML.gif
Fig. 1

Easterlin/Cantril plot of average happiness on GNP per capita

Table 1

Easterlin/Cantril data on average happiness and GNP per capita

Country

GNP per capita (1961 US dollars)

Natural log of GNP per capita

Average happiness rating (scale of 0–10)

Brazil

375

5.93

4.6

Cuba

516

6.25

6.4

Domincan Rep

313

5.75

1.6

Egypt

225

5.42

5.5

India

140

4.94

3.7

Israel

1,027

6.93

5.3

Japan

613

6.42

5.2

Nigeria

134

4.90

4.8

Panama

371

5.92

4.8

Philippines

282

5.64

4.9

Poland

702

6.55

4.4

United States

2,790

7.93

6.6

West Germany

1,860

7.53

5.3

Yugoslavia

489

6.19

5

There is a paradox in these results: higher individual incomes are associated with higher levels of individual happiness in every country, yet higher average incomes are not associated with greater average happiness across countries. Easterlin’s principal explanation for this paradox, drawing on Duesenberry’s (1949) relative income hypothesis, was that people care not about what their incomes are in absolute terms, but about what their incomes are compared to other people in the same country. This hypothesis is consistent with the positive association Easterlin observed between absolute income and happiness within countries: given any national distribution of income, the higher is an individual’s income in absolute terms, the higher it is relative to the incomes of others in the country, and therefore the happier the individual is. The relative income hypothesis is also consistent with Easterlin’s failure to find a strong association between income and happiness across countries: if absolute incomes were identically distributed around their means in all countries, the distribution of relative incomes would be identical across countries. The distributions, and hence the means, of individual happiness ratings would therefore be identical across countries, despite the country differences in average absolute income.19

Easterlin took care not to draw conclusions that his data did not support. In particular, he did not claim to have demonstrated that people care only about their relative incomes and not at all about their absolute incomes. His general assessment (p. 116) of the evidence was that “[i]t would be premature to assert that ‘everything is relative,’ but it is hard to resist the inference that relative considerations play an important part in explaining the evidence presented here.” He was also cautious not to overstate conclusions about the relationship (or lack thereof) between economic growth and human welfare, but argued (p. 112) that if the relative income hypothesis does in fact explain the patterns he observed in the data, then it would follow that “[a]n increase in the income of any one individual would increase his happiness, but increasing the income of everyone would leave happiness unchanged.”

Easterlin’s paper, and particularly the suggestion that people may care only about their relative incomes, has been followed by many studies of the relationship between money and happiness, and in particular on the extent to which people care about absolute and/or relative income.20 One of the main findings has been that, contrary to the results presented in Easterlin (1974), there does appear to be a positive correlation between average self-reports of happiness and income per capita across countries (Veenhoven 1989; Diener et al. 1995). In time-series studies, however, there is little or no evidence of a positive relationship between averages in income and happiness (Easterlin 1995; Blanchflower and Oswald 2000; Easterlin 2001, 2005). Many studies have shown evidence that people are concerned with their relative status on a number of dimensions, including income (Clark and Oswald 1996; Zizzo and Oswald 2001). In a study of the effects of income inequality on happiness, Graham and Felton (2004) find evidence suggesting “that [relative differences] may be as if not more important than [absolute] income-based differences.” Veenhoven (1991), on the other hand, has challenged the validity of claims for the importance of relative position.

In the present paper, we analyze both within- and across-country data to gather further evidence on the relationship between income—both relative and absolute—and happiness.

3 Data Sources and Variable Definitions

The individual-level data used in this paper are taken from the World Values Survey (WVS) (Inglehart et al. 2003). This survey has been administered in face-to-face interviews with nationally representative samples of respondents in over 70 countries. It consists of over 200 questions concerning personal values and attitudes toward a wide range of social issues such as religion, family, work, and democracy. This paper uses data from the third “wave” of the WVS, which was conducted between 1995 and 1998. In this section we give brief descriptions of the variables we use in our analysis; details on their definitions and construction are presented in the data appendix.

Our measure of an individual’s happiness is her response to item 65 on the WVS, which asked people to give an integer rating from 1 (least satisfied) to 10 (most satisfied) in response to the question, “All things considered, how satisfied are you with your life as a whole these days?” We call this variable happy.

Our measures of absolute income and relative income are both derived from item 227 of the WVS. In this item, each respondent is presented with 10 income brackets, and asked to indicate the bracket in which her family income falls. The income brackets used in each country are expressed in local currency; the boundaries of the income brackets differ across countries, both in terms of their purchasing power parity equivalents and in terms of the percentiles of the country income distributions at which they fall.

To approximate each individual’s family income in absolute, internationally comparable units, we first take the midpoint of each individual’s income bracket as an approximation of her family income in local currency.21 We then use conversion factors constructed by the World Bank (2004) to convert this approximated local currency income to purchasing power parity dollars ($PPP). The variable abs, which we use to represent an individual’s absolute income, consists of these $PPP family income estimates.22

We construct two measures of an individual’s relative income, both of which are based on a comparison of the individual’s absolute income to the absolute incomes of other people in her country.23 The first measure, rel_pct, represents the percentile in the national income distribution at which the individual’s absolute income falls. The second, rel_med, is the ratio of an individual’s absolute income to her country’s median absolute income.

We use a number of additional variables from the WVS to control for factors other than income that previous studies have shown to affect happiness.

Marital status Three dummy variables indicate marital status: partner equals one for respondents who are married or living with a domestic partner (and equals zero otherwise); split equals one for respondents who are divorced or separated; and widow equals one for respondents who are widowed. The omitted category for marital status is “single.”

Sex A dummy called female equals one if the respondent is female and zero if the respondent is male.

Employment Employment status is indicated by six additional zero/one dummies: part_time, self_employed, retired, housewife,24student, and unemployed. The omitted category is “employed.”

Children The dummies kid1 and kid2plus indicate respectively whether the respondent had exactly one child or at least two children; the omitted category is “no children.”

Age The variable age represents the respondent’s age in years. Since previous studies (e.g., Oswald 1997; Blanchflower and Oswald 2000) have found a quadratic relationship between age and happiness, we also use age2, which equals the square age.

Health Respondents’ ratings of their physical health on an integer scale of 1 (least healthy) to 5 (most healthy) are recorded in the variable health.

Religion Respondents’ ratings of the importance of God in their lives on an integer scale of 1 (least important) to 10 (most important) are recorded in religion.

4 Replicating Easterlin (1974)

4.1 Within Countries

Using data from the WVS, we first attempt to replicate Easterlin’s finding that, within countries, income and happiness are positively related. Following Easterlin, we compare the average value of happy for a high economic status group to the average value of happy for a low economic status group. We define the high status group to include respondents who reported family incomes in brackets 7 through 10 (on the 1 through 10 scale of WVS item v227); we define the low status group as those respondents who reported family incomes in brackets 1 through 3. Wave three of the WVS contains adequate data on both family income and happiness for the 42 countries listed in Table 2. In all but one of these countries, the average value of happy is greater for the high status group than for the low status group, and the difference is statistically significant at a confidence level exceeding 98%. The single exception is Brazil, for which the difference between the means of happy for the high and low status groups is not statistically significant (and in fact the sample mean is higher for the low status group than it is for the high status group). Despite this one exception, we interpret these results as a validation of Easterlin’s finding that on average people at the top of a country’s economic ladder are happier than people at the bottom. We state this as our first result:
Table 2

Average happiness among low and high status groups

Country

Year

Low status group mean happiness (scale of 1–10)

High status group mean happiness (scale of 1–10)

p-value

Argentina

1995

6.59

7.05

0.0099

Armenia

1995

3.79

4.49

0.0187

Australia

1995

7.35

7.81

<0.0001

Azerbaijan

1996

4.77

6.09

<0.0001

Bangladesh

1996

6.21

6.87

<0.0001

Belarus

1996

3.75

4.86

<0.0001

Bosnia

1997

4.90

6.56

<0.0001

Brazil

1996

7.18

6.99

0.7358

Bulgaria

1998

3.53

5.49

<0.0001

Chile

1996

6.36

7.40

<0.0001

China

1995

5.47

7.94

<0.0001

Colombia

1997

8.36

8.60

0.001

Croatia

1995

5.33

6.92

<0.0001

Dominican Republic

1996

6.89

7.98

0.0001

Estonia

1996

4.32

6.40

<0.0001

Finland

1996

7.51

8.05

0.0002

Georgia

1996

4.23

5.70

<0.0001

Germany

1997

6.89

7.54

0.0009

India

1996

5.92

7.51

<0.0001

Japan

1995

6.07

7.05

<0.0001

Latvia

1996

4.27

5.53

<0.0001

Lithuania

1996

3.78

5.53

<0.0001

Macedonia

1997

5.26

5.79

0.0045

Mexico

1996

7.29

8.58

<0.0001

Moldova

1996

3.39

4.46

0.0001

Montenegro

1996

5.70

6.82

0.0009

Nigeria

1995

6.08

7.39

<0.0001

Norway

1995

7.17

7.95

<0.0001

Peru

1996

6.12

6.85

0.015

Poland

1996

5.52

7.88

<0.0001

Russia

1995

3.71

5.25

<0.0001

South Africa

1995

5.53

7.78

<0.0001

Serbia

1996

5.12

6.01

<0.0001

Spain

1996

6.34

7.01

0.0023

Sweden

1996

7.16

8.02

<0.0001

Switzerland

1996

7.56

8.31

<0.0001

Taiwan

1995

6.29

7.24

<0.0001

Turkey

1996

5.82

6.71

<0.0001

Ukraine

1996

3.63

4.88

<0.0001

Uruguay

1996

6.62

7.59

<0.0001

United States

1995

7.11

7.95

<0.0001

Venezuela

1996

6.59

7.53

0.013

Notes: Year indicates the year in which the third wave of the WVS was conducted in the country. The p-value reported for each country is for a test of the null hypothesis that the mean happiness level of people in the low status group is equal to the mean happiness level of people in the high statusgroup, against the alternative that the mean happiness in the low status group is less than the meanhappiness in the high status group

Result 1 The positive within-country association between individual income and individual happiness that Easterlin found in Cantril’s data is also present in the WVS data.

4.2 Across Countries

To replicate Easterlin’s cross-country analysis, we plot the country means of happy (which we call MEAN_HAPPY) against the natural log of GDP per capita in purchasing power parity dollars (GDP_PC).25 The scatterplot, shown in Fig. 2, consists of 43 observations;26 the underlying data are presented in Table 3. Notice that mean happiness in the WVS sample varies substantially across countries, from a minimum of 3.73 to a maximum of 8.42; these mean happiness ratings have a mean of 6.35 and a standard deviation of 1.24. This dispersion contrasts sharply with the lack of variation in mean happiness that Easterlin observed in the Cantril sample. This greater dispersion in the data, as well as the larger number of observations, increases the chance of detecting a relationship (if one exists) between average income and average happiness.
Table 3

WVS data on average happiness and GDP per capita

Country

Year

GDP per capita (constant 1995 $PPP)

Natural log of GDP per capita

Average happiness rating (scale of 1–10)

Argentina (AG)

1995

10,346.29

9.24

6.93

Armenia (AM)

1995

1,664.83

7.42

4.32

Australia (AU)

1995

21,334.23

9.97

7.58

Azerbaijan (AZ)

1996

1,695.47

7.44

5.39

Bangladesh (BA)

1996

1,249.66

7.13

6.41

Belarus (BE)

1996

3,284.18

8.10

4.35

Bosnia (BO)

1997

4,427.01

8.40

5.46

Bulgaria (BU)

1998

5,228.10

8.56

4.66

Brazil (BZ)

1996

6,463.10

8.77

7.15

Chile (CL)

1996

7,766.94

8.96

6.92

China (CN)

1995

2,508.53

7.83

6.83

Colombia (CO)

1997

5,962.79

8.69

8.42

Croatia (CR)

1995

6,618.14

8.80

6.18

Dominican Rep. (DR)

1996

4,495.33

8.41

7.13

Estonia (ES)

1996

7,148.69

8.87

5.00

Finland (FI)

1996

19,441.03

9.88

7.78

Georgia (GA)

1996

1,515.47

7.32

4.65

Ghana (GH)

1995

1,601.60

7.38

7.93

Germany (GR)

1997

22,391.16

10.02

7.22

India (IN)

1996

1,938.52

7.57

6.53

Japan (JA)

1995

22,596.90

10.03

6.61

Latvia (LA)

1996

5,308.18

8.58

4.90

Lithuania (LI)

1996

6,475.10

8.78

4.99

Macedonia (MA)

1997

5,432.00

8.60

5.70

Mexico (ME)

1996

7,112.51

8.87

7.69

Moldova (MO)

1996

1,269.23

7.15

3.73

Nigeria (NI)

1995

780.57

6.66

6.82

Norway (NO)

1995

27,904.94

10.24

7.66

Peru (PE)

1996

4,217.66

8.35

6.36

Philippines (PH)

1996

3,512.58

8.16

6.84

Poland (PO)

1996

7,319.47

8.90

6.42

Russia (RU)

1995

5,932.60

8.69

4.45

South Africa (SA)

1995

8,542.25

9.05

6.08

Slovenia (SL)

1995

12,216.81

9.41

6.46

Spain (SP)

1996

15,935.59

9.68

6.61

Sweden (SD)

1996

19,855.22

9.90

7.77

Switzerland (SZ)

1996

25,219.17

10.14

8.02

Turkey (TU)

1996

5,387.40

8.59

6.18

Ukraine (UA)

1996

3,567.02

8.18

3.95

United Kingdom (UK)

1998

21,459.75

9.97

7.46

Uruguay (UR)

1996

7,922.16

8.98

7.13

United States (US)

1995

27,819.88

10.23

7.67

Venezuela (VE)

1996

5,506.10

8.61

6.72

Note: Year indicates the year in which the third wave of the WVS was conducted in the country

https://static-content.springer.com/image/art%3A10.1007%2Fs11205-007-9217-0/MediaObjects/11205_2007_9217_Fig2_HTML.gif
Fig. 2

Plot average happiness on GNP per capita using WVS data

And indeed, casual inspection of the scatterplot in Fig. 2 suggests that there is something of a positive relationship between average happiness and per capita income; the relationship apparent in the WVS data is certainly stronger than in the Easterlin/Cantril sample (reproduced in Fig. 1 of this paper). This impression is confirmed by a simple OLS regression of MEAN_HAPPY on ln(GDP_PC), which yields an estimated slope of 0.6458, at a significance level of 99.9% (r-squared = 0.2453).

The striking features of Fig. 2, however, have to do not just with the patterns observed in the entire sample, but with the way several groups of countries cluster together. Figure 3 is identical to Fig. 2, except that all points are labeled either eastern Europe (E), Latin America (L), rich industrial (R), and other (O). With the exception of two countries in the “other” category (South Africa and Turkey), the countries in each category fall into distinct (i.e., convex and disjoint) regions on the graph. The eastern European countries fall in the lower range of happiness and the middle to lower range in income; the rich industrial countries are high in both happiness and income; the Latin American countries are high in happiness and in the middle of the income range; and the others are high in happiness and low in income.
https://static-content.springer.com/image/art%3A10.1007%2Fs11205-007-9217-0/MediaObjects/11205_2007_9217_Fig3_HTML.gif
Fig. 3

Replica of Fig. 2 with region labels

Because of this clustering, the conclusion one draws about whether or how strongly per capita income and average happiness are associated depends heavily on which countries are included in the sample. If the eastern European countries are excluded, for instance, the dispersion in mean happiness is greatly reduced and the evidence of a positive relation is much weaker.27 On the other hand, if we consider only eastern Europe and the rich industrial nations, then the countries in the sample are mostly either poor and unhappy or rich and happy; the positive association between income and happiness is consequently very strong.28 Finally, if we look at each of the four groups of countries individually, it is only for eastern Europe that the slope of an OLS line is statistically significant at a confidence level above 90%.29

Easterlin’s comment about the dubiousness of picking and choosing among points applies with force to these observations, and we conclude only that Fig. 2 presents mixed evidence on the nature of the cross-country association between income and happiness. Result 2 is based on our failure to find strong evidence either for or against the existence of a positive association between income per capita and average happiness:

Result 2 To the extent that Easterlin’s scatterplot of countries’ average happiness levels against their per capita incomes suggested the absence of an association between these two variables, that suggestion is neither confirmed nor refuted by the analogous scatterplot constructed with the WVS data.

The following sections of the paper present a more detailed analysis using individual-level data from the WVS and controlling for a number of non-monetary factors. This analysis allows us to draw firmer conclusions about the effects of absolute and relative income on happiness.

5 Dissaggregated Analysis: Model Specification

5.1 The Sample and Descriptive Statistics

When we include the absolute and relative incomes of individual respondents in the analysis, limitations in the documentation of the WVS reduce the number of countries that can be included to just 18.30 Despite this reduction in the number of countries, the sample for which adequate data are available still consists of 20,771 observations. The countries represented, and the number of observations from each country, are listed in Table 4.
Table 4

Countries in the disaggregated sample

Country

Number of observations

Bangladesh

1,161

Bulgaria

810

Colombia

2,983

Croatia

1,109

Estonia

936

Finland

882

Georgia

490

Japan

737

Latvia

1,042

Macedonia

574

Russia

1,816

Sweden

869

Spain

863

Switzerland

894

Turkey

1,211

Ukraine

1,995

United States

1,295

Venezuela

1,104

Total

20,771

Table 5 shows basic descriptive statistics for each of the three income-related variables, abs, rel_med and rel_pct, and Fig. 4 presents their histograms. Because the distributions of abs and rel_med are highly skewed to the right, we will use logarithmic transformations of these variables in our regressions; since the distribution of rel_pct is reasonably symmetric, this variable will enter linearly in the regressions.
Table 5

Descriptive statistics for absolute and relative income variables

Variable

No. Obs.

Mean

Std. Dev.

Min.

Max.

Percentiles

10th

25th

50th

75th

90th

abs

20,771

12,895.01

15,830.36

64.27

165,199.64

1,181.13

2,807.90

7,611.57

17,400.41

32,695.76

rel_pct

20,771

54.3

26.86

0.09

100

17.73

31.71

59.54

74.21

88.38

rel_med

20,771

1.34

1.23

1.07

21.80

0.27

0.64

1.00

1.57

2.50

Notes: abs is measured in constant 1995 $PPP. rel_pct represents a respondent’s percentile in her national income distribution. rel_med represents a respondent’s absolute income as a proportion of her country’s median absolute income

https://static-content.springer.com/image/art%3A10.1007%2Fs11205-007-9217-0/MediaObjects/11205_2007_9217_Fig4_HTML.gif
Fig. 4

Histograms of absolute and relative income variables

5.2 Method

Since individuals’ self-ratings of their overall happiness are measured by an ordered categorical variable, our estimation is based on the ordered probit model.31 The assumption underlying the model is that, although respondents to the WVS report their happiness levels on the prescribed 1 through 10 integer scale, happiness can be measured by an unobserved (or “latent”) variable that can take on any real value. This latent happiness measure is assumed to be a linear function of a set of explanatory variables (or non-linear transformations of the explanatory variables), plus a random error. Indexing individuals by the subscript i, we write this model as
$$ H_{i} = \beta _{A} \alpha {\left( {abs_{i} } \right)} + \beta _{R} \rho {\left( {rel_{i} } \right)} + \beta _{X} \alpha {\left( {abs_{i} } \right)}\rho {\left( {rel_{i} } \right)} + {\sum\limits_{k = 1}^K {\gamma _{k} c_{{ki}} + {\sum\limits_{{j=1}\atop{j\ne \bar{j}}}^J {\lambda _{j} d_{{ji}} + \varepsilon _{i}}}}}$$
(1)
where Hi is individual i’s latent happiness level; α(abs) represents the variable abs defined above or some transformation thereof, and βA is the associated linear coefficient; rel represents either rel_pct or rel_med, ρ(rel) represents either the specified variable or a transformation thereof, and βR is the associated linear coefficient; the product α(abs) ρ(rel), with coefficient βX, is included to capture any interaction effects that exist between absolute and relative income; c1 through cK represent K individual characteristics, cki represents individual i’s value for characteristic ck, and the γk’s are the associated coefficients; d1 through dJ represent J dummies representing individual countries (or groups of countries), dji equals 1 if individual i is from country (or country group) j and 0 otherwise, the λj’s are the associated coefficients, and \( \ifmmode\expandafter\bar\else\expandafter\=\fi{j} \) is the index of the omitted country (or country group); and the εi’s are mutually independent standard normal random variables.
We do not observe the values if Hi for the individuals in our sample; our data on happiness consists only of the categorical ratings reported by survey respondents. Nevertheless, ordered probit regressions yield maximum likelihood estimates of the parameters of the latent happiness function given in equation 1. Using \( \widehat{H} \) to indicate the estimated latent happiness function, we can write
$$ \widehat{H} = \hat{\beta }_{A} \alpha {\left( {abs} \right)} +\hat{\beta }_{R} \rho {\left( {rel} \right)} + \hat{\beta }_{X}\alpha {\left( {abs} \right)}\rho {\left( {rel} \right)} +{\sum\limits_{k = 1}^K {\hat{\gamma }_{k} c_{k}}+\sum\limits_{{j=1}\atop{j\ne\bar{j}}}^J {\hat{\lambda }_{j} d_{j}}}$$
(2)
where “hats” over parameters indicate the estimates produced by the regression.

5.3 Specification

We estimate two versions of Eq. 1. In both regressions, the sixteen individual characteristics defined in Sect. 3 are included as controls. Because of the skew noted in the distribution of abs, we use the natural log of this variable in both regressions (i.e., we set α(abs_inc) = ln(abs_inc)).

The two regressions differ in which of the two measures of relative income we use. In regression 1, we use rel_pct as our measure of relative income. Because of its relatively symmetric distribution it enters linearly (i.e., ρ(rel_pct) = rel_pct). We also include country dummies for 17 of the 18 countries represented in our sample. (The omitted country is Bangladesh.)

In regression 2, we use rel_med as our measure of relative income. Because of the skew in the distribution of this variable, we use this variable in log form (i.e., ρ(rel_med) = ln(rel_med)). When both ln(abs) and ln(rel_med) are included as independent variables, however, a collinearity problem makes it impossible to include a complete set of country dummies as well (even when, as usual, the dummy variable for one country is excluded).32 To control for at least part of the country effects that would have been captured by the dummies, we include country-level variables of two types.

First, we include the log of GDP per capita (GDP_PC) and the percentage change in real per capita GDP in the previous five years (GROWTH).33 We use these variables as partial substitutes for the country dummies for the general reason that the level and rate of growth of per capita income are important broad indicators of economic conditions in a country. There are also several specific reasons for which we would expect these variables to affect individual happiness. Even holding an individual’s absolute and relative income fixed, an increase in her country’s income per capita could increase her happiness if increases in GDP per capita were associated with (i) improved infrastructure, social services and public goods, (ii) greater availability of and choice among consumer goods, or (iii) greater opportunities to pursue “post-materialist” values.34 The hypothesis that the growth rate of a country’s per capita income might affect happiness levels is related to the concept of habituation.35 Even if a person derives happiness from an increase in her income, this increase in happiness may vanish over time as she becomes accustomed to or starts to take for granted her improved material circumstances. In this case, happiness would be sensitive to recent increases in income, and we would therefore expect that on average people living in countries that had experienced greater economic growth in recent years would be happier than those in countries whose growth rates had been lower.

Second, we include two dummy variables reflecting a three-way categorization of the countries in our sample. The first category consists of the countries for which, in regression 1, the coefficients on the dummy variables were positive and significant at the 99% confidence level, namely Colombia, Finland, Sweden and Switzerland. The second category consists of the countries for which the coefficients on the dummy variables in regression 1 were negative and significant at the 99% confidence level, namely Bulgaria, Estonia, Georgia, Latvia, Russia and Ukraine. The third category consists of the remaining countries in our sample. The variable POS_DUM is defined to equal one for countries in the first category and zero for all other countries, and NEG_DUM is the analogous indicator for countries in the second category.

The results of regressions 1 and 2 are shown in Table 6.
Table 6

Regression results

Independent variables

Regression 1

Regression 2

Est. Coeff.

Std. Error

p-value

Est. Coeff.

Std. Error

p-value

Income

ln(abs)

0.1223

0.0313

<0.001

0.0430

0.0180

0.017

rel_pct

0.0125

0.0027

<0.001

   

ln(rel_med)

   

0.3051

0.0590

<0.001

Interaction

−0.0009

0.0003

<0.001

−0.0157

0.0069

0.023

Marital status

partner

0.1656

0.0239

<0.001

0.1297

0.0235

<0.001

split

−0.0526

0.0363

0.147

−0.0526

0.0360

0.143

widow

−0.0337

0.0363

0.353

−0.0380

0.0356

0.285

Sex

female

0.0147

0.0162

0.366

0.0063

0.0161

0.696

Age

age

−0.0326

0.0031

<0.001

−0.0334

0.0031

<0.001

age2

0.0004

0.0000

<0.001

0.0004

0.0000

<0.001

Children

kid1

−0.0107

0.0195

0.584

0.0080

0.0194

0.68

kid2plus

0.0639

0.0197

0.001

0.0940

0.0196

<0.001

Employment

part_time

0.0211

0.0293

0.472

0.0221

0.0290

0.446

self_employed

0.0403

0.0290

0.164

0.0495

0.0287

0.085

retired

−0.0177

0.0302

0.558

−0.0693

0.0299

0.021

housewife

0.1508

0.0276

<0.001

0.1388

0.0272

<0.001

unemployed

−0.2165

0.0283

<0.001

−0.2358

0.0280

<0.001

student

0.0845

0.0368

0.022

0.0765

0.0367

0.037

Health

health

0.3217

0.0094

<0.001

0.3179

0.0092

<0.001

Religion

religion

0.0327

0.0028

<0.001

0.0426

0.0026

<0.001

Country Dummies

Bulgaria

−0.5781

0.1182

<0.001

   

Colombia

0.9003

0.1158

<0.001

   

Croatia

−0.0141

0.1275

0.912

   

Estonia

−0.3308

0.1096

0.003

   

Finland

0.6299

0.1353

<0.001

   

Georgia

−0.7600

0.0652

<0.001

   

Japan

0.0981

0.1760

0.577

   

Latvia

−0.4422

0.1215

<0.001

   

Macedonia

−0.2418

0.1455

0.096

   

Russia

−0.5186

0.1007

<0.001

   

Sweden

0.6186

0.1602

<0.001

   

Spain

0.0598

0.1548

0.699

   

Switzerland

0.6674

0.1635

<0.001

   

Turkey

−0.2294

0.1249

0.066

   

Ukraine

−0.6513

0.0904

<0.001

   

United States

0.3907

0.1685

0.020

   

Venezuela

0.1206

0.0887

0.174

   

Country Variables

ln(GDP_PC)

   

0.1034

0.0262

<0.001

GROWTH

   

0.3359

0.0525

<0.001

NEG_DUM

   

−0.3897

0.0307

<0.001

POS_DUM

   

0.6907

0.0192

<0.001

Number of observations

 

20,771

20,771

Log likelihood of the regression

 

−41073.706

−41299.544

6 Dissaggregated Analysis: Results

6.1 Control Variables: Qualitative Results

Although this paper focuses on the effects on happiness of absolute and relative income, it is interesting first to comment briefly on the effects of the individual- and country-level variables included as controls. In this section, we simply note the signs and significance levels of the coefficients estimated in the two regressions. In later sections of the paper, we compare the magnitudes of the effects of some of these control variables to the magnitudes of the effects of the income variables.
  • On average, people who are married or living with domestic partners are happier than single people. This result is consistent with many previous studies. Argyle (1999) reviews the literature on marriage and happiness, and cites several studies to this effect.

  • The negative signs of the coefficients on split and widow suggest that on average people who are divorced or widowed are less happy than single people, but these results hold only at low levels of significance.

  • The positive, but statistically insignificant, coefficients on female are consistent with the conclusion reached by Diener et al. (1999) on the basis of a review of a number of studies on gender and happiness: “When differences [between the average happiness levels of men and women] are found, women usually report higher SWB [subjective well-being], but the differences often disappear when other demographic variables are controlled.”

  • The negative and significant coefficients on age and the positive and significant coefficients on age2 indicate that happiness initially decreases with age and then increases. In both regressions, the point estimates of the coefficients imply that minimum happiness occurs at about 41 or 42 years of age. These results are similar to the findings of Oswald (1997) and Blanchflower and Oswald (2000).36

  • The happiness level of people with exactly one child is not significantly different from the happiness level of people with no children; on average, people with two or more children are significantly happier than people with no children.

  • On average, people who are unemployed are less happy than people who are employed full time for pay. A large body of literature, reviewed by Argyle (1999), has found a similar relationship between unhappiness and unemployment; Clark and Oswald (1994) is among the papers that report such a finding.37

  • Our findings about retirement are mixed, and somewhat at odds with previous research. In regression 1, the coefficient on retired is negative, but not at all statistically significant. In regression 2, the coefficient is negative and significant at a confidence level of 97.9%. Previous studies (e.g., Campbell et al. 1976; Warr and Payne 1982), however, have found a positive relationship between retirement and happiness.

  • We find no statistically significant difference between the average happiness levels of people who are employed part-time and people who are employed full-time. Self-employed people are on average happier than people employed full-time for pay, but the statistical significance of this difference is marginal.

  • On average, students and people engaged full time in housework are happier than people who are employed full-time for pay.

  • The better is an individual’s health, the happier she is on average. The evidence accumulated in support of this hypothesis is strong, at least when, as in this paper, the measure of health is a self-report. Reviews can be found in Diener et al. (1999), and Frey and Stutzer (2002).

  • The greater is the importance of religion in a person’s life, the happier she is on average. Again, see Diener et al. (1999), and Frey and Stutzer (2002) for a discussion of numerous previous findings that are consistent with this result.

The results concerning the country dummies included in regression 1 reflect the geographical pattern observed in Fig. 3, particularly the below-average happiness levels in eastern Europe. The estimated coefficients on all eight of the eastern European countries are negative; six of them are strongly significant (with p-values less than 0.01), one is marginally significant (Macedonia, with a p-value of 0.096), and only one does not differ significantly from zero (Croatia, with a p-value of 0.912). Only one country outside of eastern Europe has a negative coefficient (Turkey, with a p-value of 0.066). Of the six rich industrial countries, four have positive and highly significant coefficients (with p-values of 0.02 or less), and two have positive but statistically insignificant coefficients (with p-values above 0.5). Both Latin American countries have positive coefficients, one marginally significant (Venezuela, with a p-value of 0.174) and one highly significant (Colombia, with a p-value less than 0.001).

Finally, with respect to the country-level variables included in regression 2 in lieu of country dummies, we find:
  • On average, the higher is the per-capita income of a person’s country, the happier she is. Since the regression controls for absolute and relative income, this result indicates that, whatever an individual’s income might be and however it compares to the incomes of others in the country, there are aspects of living in higher-income countries that are conducive to happiness. As discussed above, we speculate that these include better infrastructure and public services, greater choice of consumer goods, and greater opportunities to pursue “post-materialist” values.

  • On average, the faster a person’s country has been growing, the happier she is. Just as people may evaluate their well-being in comparison to others, they may also evaluate their well-being in comparison to themselves in previous periods.

6.2 Absolute and Relative Income: Qualitative Results

It is the effects of absolute income and relative income on happiness that are the major focus of this paper. In this section we examine just the direction of change in an individual’s happiness level when her absolute income increases or her relative income increases; we consider the magnitude of these changes in following sections.

In general, our estimate of the change in an individual’s happiness induced by a marginal change in her absolute income is given by
$$ \frac{{\partial \widehat{H}}} {{\partial abs}} = {\left[ {\hat{\beta }_{A} + \hat{\beta }_{X} \rho {\left( {rel} \right)}} \right]}\,{\left( {\frac{{\partial \alpha {\left( {abs} \right)}}} {{\partial abs}}} \right)} $$
(3)

Implicit in the computation of this derivative is the assumption that the change in absolute income being considered is not associated with any change in the individual’s relative income. This would be the case, for example, if the absolute incomes of all people in a country were to change proportionally, so that any given individual would experience a change in absolute income but no change in relative income.38 Alternatively, if we consider changes in absolute income that do in fact lead to changes in relative income, Eq. 3 can be interpreted as the part of the total change in happiness that can be attributed solely to the change in absolute income, excluding the effect of the associated change in relative income.

Using the specifications and parameter estimates of regressions 1 and 2 respectively, and using a superscript to identify the regression upon which each derivative is based, Eq. 3 takes on the particular forms:
$$ {\left( {\frac{{\partial \widehat{H}}} {{\partial abs}}} \right)}^{1} = \frac{1} {{abs}}{\left( {0.1223 - 0.0009rel\_pct} \right)} $$
(4.1)
$$ {\left( {\frac{{\partial \widehat{H}}} {{\partial abs}}} \right)}^{2} = \frac{1} {{abs}}{\left[ {0.0430 - 0.0157\hbox{ ln}{\left({rel\_med} \right)}} \right]} $$
(4.2)

Since rel_pct can not be greater than 100, expression 4.1 will be positive for any value of rel_pct between 0 and 100 (provided that abs is strictly greater than zero, which is the case for all individuals in our data set). Regression 1 therefore implies that on average an increase in an individual’s absolute income, while her relative income is held fixed, will make her happier.

Provided again that abs > 0, expression 4.2 is positive as long as rel_med < 15.47, or in other words as long as the individual’s absolute income does not exceed median absolute income in her country by a factor of 15.47 or more; this condition is satisfied for all but two of the 20,771 individuals in our data set. As in the case of regression 1, our general conclusion is that an increase in an individual’s absolute income, with her relative income fixed, increases her happiness; in the case of regression 2, however, we must add the caveat that this result may not hold for individuals whose absolute incomes are many times larger than the medians in their countries.

Result 3 summarizes the implications of both regressions about the qualitative effect of changes in absolute income on happiness:

Result 3 With relative income held constant, an increase in an individual’s absolute income increases her level of happiness (except perhaps when the individual’s relative income is extremely large compared to the median income in her country).

To evaluate the effect of a change in relative income on happiness, we compute
$$ \frac{{\partial \widehat{H}}} {{\partial rel}} = {\left[ {\hat{\beta }_{R} + \hat{\beta }_{X} \alpha {\left( {abs} \right)}} \right]}\,{\left( {\frac{{\partial \rho {\left( {rel} \right)}}} {{\partial rel}}} \right)} $$
(5)

The implicit assumption in this case is that the change in relative income being considered occurs without any change in absolute income. Expression 5 therefore represents the change in an individual’s happiness that occurs when the absolute incomes of some or all of the other people in her country fall while her absolute income remains constant. As before, if we allow for simultaneous changes in relative and absolute income, an alternative interpretation of Eq. 5 is that it represents the change in happiness due solely to the change in relative income.

Again using superscripts to denote the regression whose specification and parameter estimates we are incorporating, expression 5 can be rewritten as:
$$ {\left( {\frac{{\partial \widehat{H}}} {{\partial rel\_pct}}} \right)}^{1} = 0.0125 - 0.0009\hbox{ ln}(abs) $$
(6.1)
$$ {\left( {\frac{{\partial \widehat{H}}} {{\partial rel\_med}}} \right)}^{2} = \frac{1} {{rel\_med}}{\left[ {0.3051 - 0.0157\hbox{ ln}{\left({abs} \right)}} \right]} $$
(6.2)

Expression 6.1 is positive for any possible value of abs below $PPP 1,076,137.60. This critical value is more than six times as large as the maximum value of abs in our sample. Provided that rel_med is positive (which is true if abs is positive), expression 6.2 is positive as long as abs is less than $PPP 275,230,921.86. Both regressions therefore support result 4:

Result 4 With absolute income held constant, an increase in an individual’s relative income increases her level of happiness (except perhaps when the individual’s absolute income is extremely large).

The last qualitative result we state is based simply on the observation that the coefficient on the interaction term is negative and statistically significant in both regressions:

Result 5 The larger is an individual’s absolute income, the less her happiness is increased by an increase in her relative income; the larger is an individual’s relative income, the less her happiness is increased by an increase in her absolute income.

Result 5 implies that a person with a large absolute income would care less about changes in her relative status induced by changes in other people’s absolute incomes than would a person with a smaller absolute income; a person near the top of her country’s income distribution would care less about changes in her absolute income than would a person at a lower rung of the income distribution.

The finding that people are made happier both by increases in their absolute incomes and by increases in their relative incomes is perhaps unsurprising. But it is important to state, because previous research has sometimes asserted or implied that people must care exclusively about one of these measures of income or the other. The qualitative evidence from our regressions supports Frank’s (1985, p. 35) more moderate view: “the conclusion that absolute income does not matter at all appears just as spurious as the notion that absolute income is the only income concept that matters.” The question that naturally follows is how large, compared to each other and to other explanatory variables, are the effects of absolute and relative income on happiness?

6.3 Quantitative Results: Absolute and Relative Income

The first measure we use to compare the magnitude of the effects of absolute and relative income is the ratio of two elasticities:
$$ \psi {\left( {abs,rel} \right)} \equiv \frac{{{\left( {\frac{{\partial \widehat{H}}} {{\partial rel}}} \right)}{\left( {\frac{{rel}} {{\widehat{H}}}} \right)}}} {{{\left( {\frac{{\partial \widehat{H}}} {{\partial abs}}} \right)}{\left( {\frac{{abs}} {{\widehat{H}}}} \right)}}} $$
(7)

The numerator of this expression represents the elasticity of the latent happiness measure with respect to relative income; the denominator represents the elasticity of latent happiness with respect to absolute income. The value of ψ indicates the percentage change in happiness induced by a one percent change in relative income as a proportion of the percentage change in happiness induced by a 1% change in absolute income.

When we use the specifications and parameter estimates of regressions 1 and 2, respectively, Eq. 7 takes on the special forms:
$$ \begin{aligned}{} & \psi ^{1} {\left( {abs,\,rel\_pct} \right)} = \\ & \quad \frac{{{\left[ {\hat{\beta }_{R} + \hat{\beta }_{X} \hbox{ ln}{\left({abs} \right)}} \right]}\frac{{rel\_pct}} {{\widehat{H}}}}} {{{\left( {\frac{1} {{abs}}} \right)}{\left[ {\hat{\beta }_{A} + \hat{\beta }_{X} rel\_pct} \right]}\frac{{abs}} {{\widehat{H}}}}} = {\left( {rel\_pct} \right)}\frac{{.0125 - 0.0009\hbox{ ln}{\left({abs} \right)}}} {{0.1223 - 0.0009rel\_pct}} \\ \end{aligned} $$
(8.1)
$$ \begin{aligned}{} & \psi ^{2} {\left( {abs,\,rel\_med} \right)} = \\ & \quad \frac{{{\left( {\frac{1} {{rel\_med}}} \right)}{\left[ {\hat{\beta }_{R} + \hat{\beta }_{X} \hbox{ ln}{\left({abs} \right)}} \right]}\frac{{rel\_med}} {{\widehat{H}}}}} {{{\left( {\frac{1} {{abs}}} \right)}{\left[ {\hat{\beta }_{A} + \hat{\beta }_{X} \hbox{ ln}{\left( {rel\_med} \right)}} \right]}\frac{{abs}} {{\widehat{H}}}}} = \frac{{0.3051 - 0.0157\hbox{ ln}{\left( {abs} \right)}}} {{0.0430 - 0.0157\hbox{ ln}{\left( {rel\_med} \right)}}} \\ \end{aligned} $$
(8.2)

We first evaluate these expressions for someone we call a “median individual,” by which we mean a hypothetical person whose absolute income is equal to the median of abs in our sample ($PPP 7,611.59), and whose relative income can be expressed either as rel_pct = 50 or as rel_med = 1.0. (These values of the relative income variables imply that median income in the median individual’s country would also be $PPP 7,611.59.) For this median individual we calculate ψ1(7,611.59, 50) = 2.84 and ψ2(7,611.59, 1) = 3.84. Hence we have result 6:

Result 6 For a median individual, we find that the effect of a marginal change in relative income on happiness is several times larger than the effect of a marginal change in absolute income: by regression 1 the effect of a change in relative income is 2.84 times as large as the effect of a change in absolute income, and by regression 2 the effect of a change in relative income is 3.84 times as large as the effect of a change in absolute income.

To compare the weights that people other than a median individual place on absolute and relative income, we calculate ψ1 and ψ2 for all of the individuals in our sample—i.e., we evaluate these expressions at each of the pairs of (abs, rel_pct) and (abs, rel_med) observed in the data. Table 7 presents summary statistics and categorical frequency distributions for the values of ψ1 and ψ2 thus obtained. The sample medians of both ψ1 and ψ2 are between three and four: by either version of the measure, the effect of a change in relative income is more than three times as large as the effect of a change in absolute income for at least half of the people in the sample. The sample means of ψ1 and ψ2 are both above four. The most notable difference in the two distributions is that although all the observed values of ψ2 are greater than two—indicating that for all the individuals in the sample the effect of a change in relative income would be more than twice as large as the effect of a change in absolute income—the observed values of ψ1 are less than two for almost 32% of the sample. Even using the distribution of ψ1, however, we find that a change in relative income has a larger effect on happiness than a change in absolute income for almost 88% of the sample; moreover, the effect of an absolute income change is twice or more as large as the effect of a relative income change for less than 5% of the sample.
Table 7

Sample distributions of ψ1 and ψ2

Descriptive statistics

Variable

No. Obs.

Mean

Std. Dev.

Percentiles

10th

25th

Median

75th

90th

ψ1

20,771

4.08

3.10

0.89

1.51

3.34

5.83

8.74

ψ2

20,769

4.27

1.95

2.91

3.28

3.84

4.68

5.46

Frequency distributions

Distribution of ψ1

 

Distribution of ψ2

Interval

Freq.

Percent

Cumul.

 

Interval

Freq.

Percent

Cumul.

0 < ψ1 < 0.25

212

1.02

1.02

 

2 < ψ2 < 3

2,585

12.45

12.45

0.25 ψ1 < 0.5

796

3.83

4.85

 

3 ψ2 < 4

9,636

46.40

58.84

0.50 ψ1 < 0.75

717

3.45

8.30

 

4 ψ2 < 5

4,955

23.86

82.70

0.75 ψ1 < 1

808

3.89

12.19

 

5 ψ2< 6

1,994

9.60

92.30

1 ψ1 < 2

4,068

19.58

31.78

 

6 ψ2 < 7

286

1.38

93.68

2 ψ1 < 3

3,214

15.47

47.25

 

7 ψ2 < 8

689

3.32

97.00

3 ψ1 < 4

1,424

6.86

54.11

 

8 ψ2 < 9

95

0.46

97.45

4 ψ1 < 5

3,091

14.88

68.99

 

9 ψ2 < 10

258

1.24

98.70

5 ψ1 < 6

1,568

7.55

76.54

 

10 ψ2 < 15

169

0.81

99.51

6 ψ1 < 7

1,509

7.26

83.80

 

15 ψ2 < 20

47

0.23

99.74

7 ψ1 < 8

715

3.44

87.25

 

20 ψ2 < 25

14

0.07

99.80

8 ψ1 < 9

1,529

7.36

94.61

 

25 ψ2 < 30

36

0.17

99.98

9 ψ1 < 10

303

1.46

96.07

 

30 ψ2 < 35

2

0.01

99.99

10 ψ1 < 11

249

1.20

97.27

 

ψ2 = 54.48

3

0.01

100.00

11 ψ1 < 12

0

0.00

97.27

     

12 ψ1 < 13

211

1.02

98.28

     

13 ψ1 < 14

0

0.00

98.28

     

14 ψ1 < 15

253

1.22

99.50

     

15 ψ1 < 16

5

0.02

99.52

     

16 ψ1 < 17

99

0.48

100.00

     

Total

20,771

100

   

20,769

100

 

Notes: ψ1 and ψ2 both measure the ratio of the elasticity of latent happiness with respect to relative income to the elasticity of happiness with respect to absolute income. ψ1 is based on regression 1, in which relative income is measured by the respondent’s percentile in her national income distribution; ψ2 is based on regression 2, in which relative income is measured by the ratio of the respondent’sabsolute income to median absolute income in her country. For two observations, the values of abs are large enough that the latent happiness function estimated in regression 2 is decreasing in rel_med. (See discussion in text preceding result 3) These two observations, for which our definition of ψ2 would yield negative values, are dropped from the distribution of ψ2 described in this table

The preponderance of the evidence in Table 7 therefore supports the claim that changes in relative income tend to have greater effects on how happy people are than do changes in absolute income. Result 7 presents specific evidence for this general claim:

Result 7 For the joint distribution of absolute and relative income observed in our sample, the effect on happiness of a change in relative income is typically larger, and often much larger, than the effect of a change in absolute income. Relative income changes have a greater effect than absolute income changes for either 87.81% (regression 1) or 100% (regression 2) of the sample. The effect of a change in relative income is at least twice as large as the effect of a change in absolute income for either 68.22% (regression 1) or 100% (regression 2); at least three times as large for 52.75% (regression 1) or 87.55% (regression 2); and at least four times as large for 45.89% (regression 1) or 41.16% (regression 2).

We can also compare the effects on happiness of discrete changes in absolute and relative income. For an individual with any levels of abs and rel, we let ϖ(abs, rel, Δ) represent the amount by which her relative income would have to be raised (with her absolute income held constant) to increase her happiness by the same amount that it would be increased if her absolute income were raised by $PPP Δ (with her relative income held constant). Formally, ϖ (abs, rel, Δ) is defined by the identity:39
$$ \widehat{H}{\left({abs + \Updelta,rel} \right)} \equiv \widehat{H}{\left( {abs,rel + \varpi {\left( {abs,rel,\Updelta} \right)}} \right)}$$
(9)
With the specifications and parameter estimates of regressions 1 and 2 respectively, Eq. 9 has the special forms:
$$ \varpi ^{1} {\left( {abs,rel\_pct,\Updelta} \right)} = \frac{{{\left( {\hat{\beta }_{A} + \hat{\beta }_{X} rel\_pct} \right)}\hbox{ ln}{\left({\frac{{abs + \Updelta}} {{abs}}} \right)}}} {{\hat{\beta }_{R} + \hat{\beta }_{X} \hbox{ ln}{\left( {abs} \right)}}} $$
(10.1)
$$ \begin{aligned}{} \varpi ^{2} {\left( {abs,rel\_med,\Updelta} \right)} & = - rel\_med + \\ & \quad \exp {\left\{ {\frac{{\hat{\beta}_{A} {\left[ {\hbox{ln}{\left( {\frac{{abs + \Updelta}} {{abs}}} \right)}} \right]} +\hbox{ ln}{\left( {rel\_med} \right)}{\left[ {\hat{\beta }_{R} + \hat{\beta}_{X} \hbox{ ln}{\left( {abs + \Updelta} \right)}} \right]}}} {{\hat{\beta }_{R} + \hat{\beta }_{X} \hbox{ ln}{\left( {abs} \right)}}}} \right\}} \\ \end{aligned} $$
$$ \begin{aligned}{} \varpi ^{2} {\left( {abs,rel\_med,\Updelta} \right)} & = - rel\_med + \\ & \quad \exp {\left\{ {\frac{{\hat{\beta }_{A} {\left[ {\hbox{ln}{\left( {abs + \Updelta} \right)} - \hbox{ln}{\left({abs} \right)}} \right]}\hat{\beta }_{R} \hbox{ ln}{\left( {rel\_med} \right)} + \hat{\beta }_{X} \hbox{ ln}{\left( {abs + \Updelta} \right)}\hbox{ ln}{\left({rel\_med} \right)}}} {{\hat{\beta }_{R} + \hat{\beta }_{X} \hbox{ ln}{\left( {abs} \right)}}}} \right\}} \\ \end{aligned} $$
(10.2)
Several illustrative calculations of ϖ1 and ϖ2, all for a median individual, are shown in Table 8. These calculations, which indicate that modest increases in relative income increase happiness as much as comparatively large increases in absolute income, are the basis of results 8.1 and 8.2:
Table 8

Changes in relative income equivalent to specified changes in absolute income

Δ($PPP)

ϖ1(7611.59,50,Δ)

ϖ2(7611.59,1.0,Δ)

2,000.00

4.10

0.0626

7,611.59

12.19

0.1978

15,223.17

19.31

0.3316

Notes: For an individual with any levels of abs and rel_pct, ϖ1(abs,rel_pct,Δ) represents the number of points her percentile in the national income distribution would have to be raised to increase her happiness by as much as in increase in her absolute income of $PPP Δ. For an individual with any levels of abs and rel_med, ϖ2 (abs,rel_med,Δ) represents the amount by which the ratio of her absolute income to her country’s median absolute income would have to be increased to increase her happiness by as much as in increase in her absolute income of $PPP Δ. The figures in this table represent a median individual, for whom abs = 7611.59, rel_pct = 50, and rel_med = 1.0

Result 8.1 On the basis of regression 1, we find that a median individual would be indifferent between an increase in her absolute income from $PPP 7,611.59 to $PPP 9,611,59 and an increase in her position in the national income distribution from the 50th to the 54.10th percentile. A doubling of her absolute income would increase her happiness as much as a move up to the 62.19th percentile, and a tripling of her absolute income would be equivalent to a move up to the 69.31st percentile.

Result 8.2 On the basis of regression 2, we find that a median individual would be indifferent between an increase in her absolute income from $PPP 7,611.59 to $PPP 9,611.59 and an increase in the ratio of her absolute income to her country’s median absolute income from 1 to 1.0626. A doubling of her absolute income would increase her happiness by as much as an increase in this ratio from 1 to 1.1978, and a tripling of her absolute income would be equivalent to an increase in this ratio from 1 to 1.3316.

Because the measure of relative income we use in regression 2, rel_med, is defined as the ratio of an individual’s absolute income to her country’s median income, the value of ϖ2 has a special interpretation that allows us to gain some further intuition about the trade-offs people would be willing to make between absolute and relative income. Finding that ϖ2 (abs, rel_med, Δ) = θ implies that an individual with the specified levels of abs and rel_med would be indifferent between (i) having her absolute income and the absolute incomes of all the other people in her country increased by $PPP Δ and (ii) having her absolute income increased by θ times her country’s median income while the incomes of everyone else were held constant. The values of ϖ2 reported in Table 8 therefore indicate that a median individual would be as happy to have her absolute income increase by 6.26% ($PPP 476.49), provided that no one else’s absolute income went up, as she would be to have her absolute income increase by $PPP 2,000.00 if everyone else’s absolute income were to go up by the same amount. Similarly, she would be indifferent between having her absolute income alone increase by 19.78% and having her absolute income and everyone else’s doubled; and between having her absolute income alone increase by 33.16% and having hers and everyone else’s tripled.

6.4 Quantitative Results: Non-monetary Factors

Although the focus of this paper has been on the importance of absolute and relative income in determining happiness, it is also important to compare the importance of these income measures to non-pecuniary factors. We make these comparisons on the basis of discrete changes in the variables, using identities analogous to Eq. 9. The results below illustrate the importance of absolute and relative income compared to three important non-pecuniary factors included in the regressions: marital status, employment status, and health. (In this section, we consider only the case of a median individual.)
  • For a median individual who is single, getting married or finding a domestic partner would increase happiness as much as an increase in her absolute income of 767% (regression 1) or 1,948% (regression 2).

  • For a median individual who is single, getting married or finding a domestic partner would increase happiness as much as an increase in her relative income from the 50th to the 88th percentile (regression 1), or from 100% to 219% of her country’s median income (regression 2).

  • For a median individual who is unemployed, finding a full-time job for pay would increase happiness as much as an increase in her absolute income of 1,583% (regression 1) or 24,118% (regression 2).

  • For a median individual who is unemployed, finding a full-time job for pay would increase happiness as much as an increase in her relative income from the 50th to the 99.6th percentile (regression 1), or from 100% to 418% of her country’s median income (regression 2).

  • For a median individual who (on a 1–5 integer scale) gives her health a rating of 3 (the 25th percentile of our sample), an improvement in her health that increased her rating to 4 (the median of our sample) would increase happiness by as much as an increase in her absolute income of 6,531% (regression 1) or 163,650% (regression 2).

  • For a median individual who (on a 1–5 integer scale) gives her health a rating of 3 (the 25th percentile of our sample), an improvement in her health that increased her rating to 4 (the median of our sample) would increase happiness more than an increase in her relative income from the 50th to the 100th percentile (regression 1), or as much as an increase in her relative income from 100% to 687% of her country’s median income (regression 2).

These findings present an important caveat to all the preceding analysis is this paper. Although we have found strong evidence that, ceteris paribus, larger absolute incomes and larger relative incomes both tend to make people happier, and that changes in relative income tend to have a greater effect on happiness than do changes in absolute income, the results presented in this section show that the effects on happiness of several non-pecuniary factors are greater by many orders of magnitude than the effects of either income measure.40 Money can buy some happiness, but compared to the happiness people derive from personal relationships, employment and good health, it can not buy much.

7 Conclusions

On the question of whether a person’s happiness is affected by her absolute income or by her relative income, we have found that both matter. There is certainly no logical reason that people could not like to have incomes that are high absolutely and relatively, and empirically our results indicate that they do. We reject both a strict interpretation of Easterlin’s (1974) hypothesis that only relative income matters and a narrowly “economic” model in which only absolute income enters the utility function.

The finding that people do care about their absolute incomes implies that a distribution-neutral increase in average income will make everyone happier. In that sense, economic growth can increase human welfare. But the finding that people also care about their relative incomes adds a note of caution: if some people’s incomes grow more slowly than others, the relative losers could end up feeling worse off, despite the increases in their absolute incomes. And any individuals whose absolute incomes remain constant or fall during a period of general economic growth lose unambiguously.

We have also found strong evidence that changes in relative income tend to have larger effects on happiness than do comparable changes in absolute income. This result suggests that, even in an episode of economic growth during which everyone’s absolute income increases, the welfare losses of relative losers could be large. In terms of aggregate happiness, any adverse distributional consequences of growth could swamp the material benefits.

Although the question of how economic growth affects income distribution has been the subject of a great deal of research—including the vast literature on the Kuznets curve, and a recent spate of work following an influential report by Dollar and Kraay (2002)—no clear consensus has emerged. The effects of economic growth on income distribution are surely neither simple nor uniform across time and countries, but further research that sheds light on this issue—if not generally, at least in particular contexts—will be an important step toward a better understanding of how economic growth affects happiness.

The welfare benefits of economic growth are also called into question by our finding that the effects of income (measured absolutely or relatively) on happiness are very small compared to some non-monetary factors. It is important to note, however, that that we have measured the effects of income on happiness holding many other factors constant. But income growth might eventually lead to improvements in some of the factors that have a big impact on happiness, such as health and employment. So to the extent that growth in income leads to improvements in other, non-pecuniary, factors that people value, it may have a large, albeit indirect, effect on happiness. Of course, as people’s incomes increase they may, as a result of some form of group or even individual irrationality, use their new resources in ways that do not make them happier.41 It is not growth in income per se that matters, but what people do with the opportunities that greater material prosperity creates.

Footnotes
1

Although there are cases in which spouses or jobs can in some sense be purchased, and to some extent good health can be “bought” through expenditures on health care.

 
2

See http://www.worldvaluessurvey.org. This survey is the main source of data for this paper.

 
3

Conducted by the Public Opinion Analysis sector of the European Commission. See http://europa.eu.int/comm/public_opinion/description_en.htm.

 
4

Conducted by the National Opinion Research Center at the University of Chicago. See http://www.norc.uchicago.edu.

 
5

Conducted by the Institute for Social and Economic Research at the University of Essex. See http://www.iser.essex.ac.uk/bhps/index.php.

 
7

See Diener (1984), Eckman et al. (1990), Pavot et al. (1991), Shedler et al. (1993), Sandvik et al. (1993), Balatsky and Diener (1993), Sutton and Davidson (1997), and Siedlitz et al. (1997).

 
8

See Ouweneel and Veenhoven (1991), and Diener et al. (1995).

 
9

See Veenhoven (1993, 1996), Diener et al. (1999), Diener and Suh (2000), and Frey and Stutzer (2002).

 
10

See Horley and Lavery (1995), Diener and Suh (1997), Nolen-Hoeksema and Rustig (1999), and Lucas and Gohm (2000).

 
11

See Lee et al. (1991) and Myers (1999).

 
12

See Clark and Oswald (1994), and Di Tella et al. (2001).

 
13

See Frey and Stutzer (2000) and Veenhoven (2000).

 
14

See Alesina et al. (2001) and Graham and Felton (2004).

 
15
 
16

In eliciting these ratings of subjective well-being, Cantril used a survey methodology that was somewhat more subtle than simply asking people to choose a number. Details on Cantril’s “self-anchoring” methodology can be found in Easterlin (1974) and in the original study (Cantril 1965).

 
17

The estimated slope is 0.6486, with a p-value of 0.086. The r-squared for the regression is 0.2257.

 
18

The influential cases are the US and India; the countries with “unusual political circumstances” are Cuba and the Dominican Republic. With these four cases deleted, an OLS line has a slope of 0.1014, with a p-value of 0.535. The r-squared is 0.0498.

 
19

Even if absolute incomes were not distributed identically around their means across countries, this would be true as long as there was no systematic relationship between mean income and income distribution.

 
20

Surveys that focus on the relationship between money and happiness can be found in Argyle (1999), Diener and Oishi (2000), Frey and Stutzer (2000), and Easterlin (2001).

 
21

For individuals in the highest income category (for which no upper limit is specified), we use a figure equal to 120% of the lower bound of the category as an approximation of family income in local currency.

 
22

Ideally, we might also adjust family income for the number of people in the respondent’s family, but the WVS contains no explicit question on household or family size, nor does it contain other questions from which family size could be inferred with any reasonable confidence.

 
23

We are therefore implicitly assuming, as did Easterlin (1974), that the reference group to which a person compares herself is the population of the country in which she lives. An interesting area for further research would be to investigate the extent to which people’s happiness depends on their status relative to comparison groups defined by other criteria, such as ethnicity, religion or occupation. Praag and Ferrer-i-Carbonell (2004, Chap. 8) discuss recent research along these lines.

 
24

The term “housewife” is one of the employment categories listed in the WVS codebook. There is no category for men who work full time on domestic tasks.

 
25

Data on purchasing power parity GDP per capita are taken from World Bank (2004). Wave 3 of the WVS was conducted in different years in different countries; for each country, we use the value of PPP GDP per capita corresponding to the year in which wave 3 of the WVS was conducted there (see Tables 2 and 3).

 
26

These 43 countries are the same as those listed in Table 2, except that (i) Montengro, Serbia and Taiwan are dropped because World Bank (2004) does not have GDP data for them, and (ii) Ghana, the Philippines, Slovenia and the UK, which were excluded from Table 2 because of missing data on family income in the WVS, are added.

 
27

An OLS line through the non-eastern European data still has a positive slope, but the significance level falls to 94.9%, the estimated slope is just 0.2260, and the r-squared falls to 0.1440.

 
28

With an estimated coefficient of 1.2549, significant at above the 99.9% confidence level, and an r-squared of 0.7944.

 
29

An OLS line through just the eastern European data has an estimated slope of 0.7977, significant at the 99.2% confidence level, with an r-squared of 0.4009.

 
30

The limitation in the WVS documentation is that for many countries no information is available on what local currency values were used to define the boundaries of the family income brackets used in item 227. Without that information, it is not possible to approximate respondents’ family incomes, which, as described in Sect. 3, is necessary for the construction of abs, rel_pct, and rel_med.

 
31

For a detailed exposition of the ordered probit regression model, see Long (1997) and Long and Freese (2001).

 
32

To see the collinearity problem formally, let med(j) represent the median absolute income of country j; the median absolute income of the country whose dummy variable is excluded from the regression is then denoted \( med{\left( {\ifmmode\expandafter\bar\else\expandafter\=\fi{j}} \right)} \). The following + 2 vectors then add up to a vector of all zeros: a vector of constants, each equal to \(\hbox{ln}{\left({med{\left({\ifmmode\expandafter\bar\else\expandafter\=\fi{j}} \right)}} \right)} \); one vector for each of the − 1 countries whose dummies are included, each vector of the form ln(med (j))dj; the vector of observed values of ln(abs); and negative one times the vector of observed values of ln(rel_med). Intuitively, the source of the collinearity is that, because rel_med is defined as the ratio of abs to med(j), the difference between ln(abs) and ln(rel_med) is constant within countries. (For all individuals in any country j, that difference is equal to ln(med (j))).

 
33

Data for both of these variables is from World Bank (2004).

 
34

As described in Inglehart (1997).

 
35

See Easterlin (1995, 2001, 2003).

 
36

As Easterlin (2004) has pointed out, however, this U-shaped relationship between age and happiness appears in analyses that hold constant factors such as health and whether one is widowed. Since people’s health and probability of being widowed in fact do not remain constant over their life-spans, this U-shaped relationship is not inconsistent with the possibility that people’s experienced happiness levels tend to fall in the later periods of their lives.

 
37

Because the regressions estimate the effect of employment status on happiness with absolute and relative income held constant, the increases in happiness attributed to finding a full-time job are the consequence of factors other than changes in the individual’s income associated with finding a job. Our analysis does not tell us exactly what those other factors are, but it seems reasonable to speculate that they have to do with increased self-esteem and respect from others, and perhaps an increased feeling of purpose or productivity (that may come from simply having a job, even if the purpose or product of the job are not inspiring to the person holding it).

 
38

There are also less restrictive conditions under which a change in an individual’s absolute income would not be accompanied by a change in her relative income. If we measure relative income by rel_pct (as in regression 1), this would be the case as long as the change in the individual’s absolute income did not change her rank in the national income distribution (perhaps because of simultaneous changes in the absolute incomes of some or all of the other people in her country). If we measure relative income by rel_med (as in regression 2), it would be the case as long as median income in the individual’s country increased proportionately with her absolute income.

 
39

We assume that, other than absolute and relative income, all of the variables upon which \( \widehat{H} \) depends are held constant, and so write \( \widehat{H} \) as a function of just abs and rel.

 
40

Blanchflower and Oswald (2000) present similar results showing that people value important non-pecuniary qualities of life as much as very large amounts of money.

 
41

Frank (1999) presents extensive evidence of this phenomenon.

 

Acknowledgements

For helpful discussion and comments, we would like to thank Gabriela Catterberg, Ellsworth Dägg, Picard Janné, Christopher Kilby, Vladimir Kontorovich, and Anne Preston; participants in the Behavioral Research Council’s symposium on Behavioral Economics and Neoclassical Economics, July 2002, Great Barrington, MA; and participants in the 6th International Conference of the International Society for Quality of Life Studies, November 2004, Philadelphia, PA.

Copyright information

© Springer Science+Business Media B.V. 2007