Applied Research in Quality of Life

, 6:219

Does More Money Make You Happier? Why so much Debate?

Authors

    • The Brookings InstitutionUniversity of Maryland
Article

DOI: 10.1007/s11482-011-9152-8

Cite this article as:
Graham, C. Applied Research Quality Life (2011) 6: 219. doi:10.1007/s11482-011-9152-8

Abstract

Easterlin’s famous paradox questioned standard economic assumptions about a fundamental relationship in economics: that between happiness and income. In recent years there has been renewed debate about the paradox. In this essay, I highlight some of the methodological issues and challenges underlying that debate. I focus on the sensitivity of the results to the method selected, the choice of micro or macro data, and the way that happiness questions are defined and framed, all of which result in divergent conclusions. I also note the mediating role of the pace and nature of economic growth, institutional frameworks, and inequality. What is most notable is the remarkable consistency in the determinants of individual happiness – including income – within countries of diverse income levels and, at the same time, how happiness is affected by cross-country differences that are related to average per-capita income levels, such as political freedom and public goods. Income clearly plays a role in determining both individual and country level happiness. Still, assessing its role relative to other more difficult to measure factors as countries develop in new ways and at different rates will remain a challenge for the foreseeable future.

Keywords

HappinessIncomeQuality of lifeWell-beingAdaptation

“Will raising the incomes of all increase the happiness of all?” Richard Easterlin (1974)

Happiness economics has increasingly entered the mainstream. Yet rather ironically, there is much less consensus today than there was in the early stages of happiness research on the first question that it originally shed light on: the relationship between happiness and income. It has become one of the most controversial questions in the discussion of how or if the study of happiness can help economists understand how to enhance welfare or well-being. Does having more money make you happier? And if so, how much more money do you need to be just a little bit happier or very happy?

Easterlin (1974, 2003) was the first economist to systematically explore the relationship between average country happiness levels and per capita incomes over several decades. His seminal work highlighted an apparent paradox: as countries grew materially wealthier – and healthier – over time: average happiness levels did not increase. His findings are now known as the Easterlin paradox. A number of subsequent studies confirmed the general direction of his findings (e.g., that average happiness levels do not increase as average incomes increase over time). In contrast, more recent research, based on new data encompassing many more countries around the world, questions whether the paradox exists at all. Thus, there is renewed debate over Easterlin’s original question: “will raising the incomes of all increase the happiness of all?”

Easterlin’s paradox has been explained by rising aspirations and comparison effects. Once basic needs are met, aspirations rise as quickly as incomes, and individuals care as much about how they are doing in comparison with their peers as they do about absolute gains. The importance of aspirations and comparison effects to individual well-being have been demonstrated in smaller scale studies of individual attitudes across a range of contexts ranging from neighborhoods in the U.S. to cities in Latin America to regions in Russia. See Kingdon and Knight (2007); Graham and Felton (2006); (Lora and Chaparro 2011); Luttmer (2005). Easterlin (2001) also included a time dimension: most people think that they were less happy in the past and expect to be happier in the future. They judge their past living standards by their current aspirations, but fail to account for their aspirations adjusting over time as they predict their future happiness.

Within countries, across cultures and levels of development, Easterlin and a host of other economists have shown that wealthier people are, on average, happier than poorer ones, although the relationship is not necessarily linear, and the additional increases in happiness that come from extra income diminish as absolute levels of income increase (as in the case of the utility function in standard textbook economics). A number of studies1 show that the same proportional increase in income yields a lower increase in happiness at higher income levels. Differences in income, meanwhile, only account for a low proportion of the differences in happiness among persons, and other economic and non-economic factors, such as employment and health status, exert important influences on happiness. Studies in a wide range of countries demonstrate that there is the relationship between income and happiness is complex, and is mediated by a number of non-income variables, such as health, social capital, and the nature of public goods.

Individual personality differences also play a role, although they are difficult to measure. It may be that those individuals that prize material goods more highly than other things in life are less happy, and thus as ownership of material goods increases (via higher levels of income), happiness levels do not increase proportionately – as Frey (2008) suggested. Finally, there is also some evidence that happier people earn more (and are healthier) than unhappy people.2

There is now a renewed debate over whether the Easterlin paradox actually exists. A number of scholars, such as Deaton (2008), and Stevenson and Wolfers (2008), have published papers demonstrating a clear relationship between per capita incomes and average happiness levels, with no sign that the correlation weakens, either as income levels increase or over time. Indeed, the work of both sets of authors suggests that the slope may be steeper for richer countries, most likely because wealthier people are better able to enjoy higher levels of income than are poor ones (a greed effect?).3 Both of these papers rely on the newly available Gallup World Poll, which covers over 120 countries world-wide, as well as some different data sets for earlier years. In a more recent paper, updating their original work, Sacks et al. (2010), using several datasets which collectively cover 140 countries and represent nearly all of the world’s population, show that subjective well-being – as measured by Cantril’s ladder of life question – and income are positively correlated across individuals within countries, between countries in a given year, and as a country grows through time.

Inglehart et al. (2008), in a new analysis of data from the World Values Survey for 1981–2006, find that subjective well-being rose in 77% of the 52 countries for which data on the same countries over time is available. In IADB (2008) Eduardo Lora and colleagues at the Inter-American Development Bank, using Gallup data for Latin America, also find a positive relationship between per capita income levels and average happiness levels.

Other studies have more nuanced findings. Stefano Pettinato and I (see Graham and Pettinato 2002a), in the first study of happiness in a large sample of developing countries and using absolute levels of per capita GDP with data from the 1990’s, found that, on average, happiness levels are higher in the developed than in the developing countries in the sample, but that within each group of countries, there is no clear income-happiness relationship. Our work is based on the World Values survey and on the Latinobarometro poll for Latin America.

Why the discrepancy? For a number of reasons, the divergent conclusions may each be correct. One reason is that the relationship between happiness and income is mediated by a range of factors that can alter its slope and/or functional form. These include the particular questions that are used to measure happiness; the selection of countries that is included in the survey sample; the specification of the income variable (for example if absolute levels or a logarithmic specification of income are used; this is discussed in detail below); the rate of change in economic conditions in addition to absolute levels; and changing aspirations as countries go from the ranks of developing to developed economies. The objectives driving the particular study, meanwhile, dictate which happiness question is most appropriate. If the objective is to find a measure of reported well-being that has the most consistency across countries, then a more framed question – such as the ladder of life – may be more appropriate. If the objective is to see how happiness varies across countries and cultures, then a more open-ended question, such as a general happiness question, is more appropriate. Depending on which question is used, the happiness-income relation may vary.

In this essay, I review evidence from my own work with several colleagues, as well as from several other authors. My objective is as much to contribute to the method as to the substance of the debate on income and happiness.

Question Framing Issues

Diener et al. (1999) decompose subjective well-being into an affective or emotional component and a cognitive or judgmental component. The first is determined and measured by how often an individual reports experiencing positive or negative affect (such as smiling), while life satisfaction is composed of an individual’s satisfaction with various life domains (such as health and work) as well as with life in general. Affect questions typically (and not surprisingly) have less of a relationship with income than do cognitive questions. More framed life satisfaction questions, such used by Cantril (1965) as the ladder of life question – which asks respondents to compare their lives to the best possible life they can imagine on a 0 to 10 scale, have an even closer relationship with income.

The earlier surveys that Easterlin and others used, such as the World Values survey and the Eurobarometro, relied on open ended happiness or life satisfaction questions, which posed very simply “generally speaking, how happy are you with your life?” or “how satisfied are you with your life”, with possible answers on various scales, ranging from one to four to one to ten. Answers to general happiness and life satisfaction questions are highly correlated.4 In contrast, the “life satisfaction” variable that is used in the Gallup World poll – which is the basis for the Deaton (2008) and Stevenson and Wolfers (2008) papers – is Cantril’s best possible life question. The best possible life question provides much more of a reference frame than does an open ended life satisfaction question. Surely when asked to compare to the best possible life, respondents in very poor countries are aware that life is likely better in wealthier ones, not least because of how widely available information about how the rich in wealthy countries live has become, due to widespread access to the media and the internet.

Deaton (2008) makes a similar point about the Gallup World Poll findings: when asked to imagine the best and worst possible life for themselves on an 11-point scale, people use a global standard, and the Danes understand how bad life in Togo is, and the Togolese, through TV and other media, understand how good life is in high income countries. Helliwell (2008), meanwhile, compares results based on the Cantril ladder in the 2006 Gallup Poll with those on life satisfaction as a whole in the World Values survey, and finds that the correlation between income and life satisfaction is stronger with the more framed ladder of life question. At the same time, there is striking consistency in the other factors that contribute to life satisfaction across the two surveys.

As a simple test of the extent to which question framing matters, my colleagues and I compared the income and happiness relationship across several different life satisfaction questions for the Latin America sub-sample of the Gallup World Poll. Latin America is a good testing ground as the region includes countries with a wide range of income levels, with some wealthier countries such as Chile and Brazil near OECD levels, and others, such as Guatemala and Honduras, on the other extreme of the development spectrum. The questions included: the best possible life question described above; an economic satisfaction question (are you satisfied with your standard of living, all the things that you can buy and do?); a poor-rich scale question (On a scale from zero to ten, with zero the poorest people and ten the richest people, in which cell would you place yourself?); an affect question (Did you smile or laugh a lot yesterday?); a life purpose question (Do you feel your life has purpose or meaning?);5 and a freedom/opportunity question (Are you satisfied with your freedom to choose what to do with your life?).6

Given that accurately measuring incomes in a context such as Latin America is difficult, we used both income and wealth variables. A large percentage of respondents work in the informal sector or in unsteady jobs and have difficulty accurately recalling earnings, for example. And the most recent income levels which are reported may misrepresent more permanent income flows, due to seasonality, economic shocks, and/or job instability. Wealth indices, on the other hand – while adjusting better for temporary fluctuations, are less effective at capturing variability across households, particularly at the high end of the income scale, nor do they account for the quality of assets. Access to water may be irregular, or televisions and refrigerators that are owned may be functioning poorly, if at all.

For our income variable, we used log of per capita household income measured in 2005 PPP U.S. dollars.7 This specification helps control for outliers, and conforms to standard economic assumptions that an extra unit of income is more significant for those at the bottom of the distribution with less available resources than for those at the top. Some of the earlier studies of income and happiness used absolute rather than log income levels, and show a curvilinear relationship between income and happiness, suggesting the satiation point that is part of the explanation of the Easterlin paradox. Later studies, such as, Deaton (2008), use either log of average GDP per capita or the average of log income per capita. For our country level analysis, we used the average log of per capita household income (as opposed to the log of the average per capita household income).8

We constructed our wealth index, based on the list of assets in the Gallup Poll.9 We tested the cross-country relationship between income and happiness across questions, using happiness as measured by each question as the dependent variable, and then explored how the relationship varied according to average (log of) per capita income in the country and with the typical socio-demographic profile of respondents in each country.10 We find that the log-linear income and happiness relationship holds across countries for the best possible life and the poor to rich economic ladder questions, but not for affect (smiling a lot the previous day), life purpose and freedom to choose in life questions (see Table 1). Indeed, the relationship between income and the smiling and life purpose questions is negative and significant when we use our income variable, and insignificant when we use our wealth variable. The relationship of freedom to choose and income (and wealth), meanwhile, is positive, but insignificant.
Table 1

Question phrasing, income and happiness across and within Latin American countries

 

Happiness questions versus Income per capita

Happiness questions versus Wealth Index

Individual cross-sectional analysis

Cross-country analysis

Individual cross-sectional analysis

Cross-country analysis

Subjective variable

Model 1

Model 2

Model 3

Model 4

Model 5

Model 1

Model 2

Model 3

Model 4

Model 5

Best possible life

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Satisfaction with living standard

Yes (+)

Yes (+)

Yes (+)

No (+)

No (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Position on socio-economic scale

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Yes (+)

Purpose in life

No (+)

Yes (+)

Yes (+)

Yes (−)

Yes (−)

Yes (+)

Yes (+)

Yes (+)

No (−)

Yes (−)

Smiled yesterday

Yes (+)

Yes (+)

Yes (+)

Yes (−)

No (−)

Yes (+)

Yes (+)

Yes (+)

No (−)

Yes (−)

Satusfaction with personal freedom

Yes (+)

Yes (+)

No (+)

No (+)

No (−)

Yes (+)

Yes (+)

Yes (+)

No (+)

No (−)

Yes denotes statistically significant at 10% level; No otherwise

Sign in parentheses denote positive or negative sign of the estimated coefficient.

Model 1: Probit/Ordered probit: Subjective variable = f(log percapita income)

Model 2: Probit/Ordered probit: Subjective variable = f(log percapita income, country dummies)

Model 3: Probit/Ordered probit: Subjective variable = f(log percapita income, country dummies, demographics)

Model 4: Ordered probit: Country average subjective variable = f(country average household percapita income)

Model 5: Ordered probit: Country average subjective variable = f(country average household percapita income, demographics)

Graham et al. (2010b) based on data from Gallup World Poll 2007.

At the individual level, our basic model was analogous, with happiness as the dependent variable and then exploring the effects of household wealth, the socioeconomic and demographic profiles of respondents, and controls for unobservable traits specific to each country.11 We find that income and wealth were positively correlated with most measures of happiness except for the life purpose and freedom to choose questions, which were insignificant with some specifications (see Table 1). This confirms the work of many other studies, which consistently finds a cross sectional relationship between happiness and income within countries, regardless of whether or not the Easterlin paradox holds across countries or through time.

Across the questions, we find that the highest effect size – based on the size of the coefficients – is between satisfaction with one’s standard of living and income (and/or wealth: .43 on log income/.39 on wealth). This is followed by the poor to rich scale economic ladder question (.32/.35), and the best possible life question (.27/.29). Income and wealth do a good job of explaining the distribution of responses on the ladder of life question, including when other controls are used, but they do not explain answers on smiling, life purpose, and freedom to choose questions. The first three questions provide more of an economic frame for people, while the latter are vaguer and more open ended.

Our results comparing across the questions, both across countries and within them, support our intuition that question framing can have important effects on the measured relationship between income and happiness. Questions that provide more tangible economic or status frames seem to have a closer relationship with income than do more open ended questions that capture either affect and/or life chances.

One example from my most recent work in Graham and Chattopadhyay (2009), that supports this broader point, is from a study Soumya Chattopadhyay and I conducted of happiness in Afghanistan. Afghans had relatively high mean happiness scores compared to averages for contexts such as Latin America, where objective conditions are surely better, as well as compared to the world average. Yet their answers on the best possible life question were quite a bit lower than the world average. This suggests that while Afghans may be naturally cheerful, or else have adapted their expectations downwards in the face of very adverse conditions, they are quite realistic as they assess their circumstances in relative terms. Their answers reflect their awareness of how their lives compare to a global reference point for the best possible life.

Two more recent studies, one based on world-wide data and another on detailed data for the United States, provide strong support for the basic direction of our findings. They also highlight the different dimensions of happiness. Diener et al. (2010), in a study based on the Gallup World Poll (136,000 respondents across 132 nations) find that income is strongly correlated with how people evaluate their lives, based on the ladder of life question, but only moderately correlated with day-to-day positive feelings like smiling yesterday.

Kahneman and Deaton (2010) in a study of 450,000 respondents in the Gallup-Healthways Well-Being Index, a daily survey of U.S. respondents from 2008 to 2009, also used the ladder question and asked questions about emotional experiences the prior day. They found that hedonic well-being – the emotional quality of an individual’s everyday experience – correlated less closely with income than did life evaluation – the thoughts that people have about their life when they think about it – as measured by the ladder of life question. Both questions correlated closely with income (in a log-linear manner) at the bottom end of the income ladder, but the correlation between hedonic well-being and income tapered off at about $75,000 per year, while the one between life evaluation and income did not. Thus more money does not necessarily buy more happiness, but less money is associated with emotional pain. Navigating emotional challenges like divorce, anger, and depression, is likely made even more difficult by having insufficient resources. At the same time, beyond a certain level, increases in income no longer improve people’s ability to do what matters most to their emotional well-being. Their findings highlight the importance of the distinction between the judgments people make when they think about their life and the feelings they experience as they live it. The former is sensitive to socio-economic status, the latter to circumstances that provoke positive and negative emotions, like spending time with friends or caring for a sick relative.

Country Selection Issues

Most economists agree that there is some relationship between income and happiness across countries, with wealthier countries generally showing higher levels of happiness than poorer ones. Yet different surveys sample different selections of countries, and that too seems to affect the strength of the relationship, and in part help explain the debate over how strong it is.

My above-mentioned work with Pettinato, based on a large sample of Latin American and OECD countries and an open ended life satisfaction question, finds a relationship between income levels and happiness, although within each of the poor and rich country sets there is no clear pattern. The Gallup World Poll clearly contains the largest and most diverse set of countries (in terms of region and development levels) of any of the surveys that have been used to study happiness to date. Yet that presents methodological challenges as well as analytical diversity. A large number of the new countries in the Gallup poll are small, poor countries in Sub-Saharan Africa and/or the transition economies, which have seen the dismantling of existing social welfare systems and dramatic falls in happiness. Thus the steeper sloped income and happiness relationship that appears in research based on this data may be driven as much by falling incomes and happiness in a large number of small countries at the bottom of the distribution as it is by rising incomes and happiness in the rich countries. A similar point has been made by Easterlin about the sample of countries in the World Values Survey. It is not clear, however, how to resolve the problem of different country selection, and dropping large parts of the sample – for example the transition economies – in order to retain comparability eliminates some of the most important trends in the global economy in recent decades.

The transition economies surely have had a distinct experience. Easterlin (2008) examines happiness in Eastern Europe from 1989 to 1998 and finds that life satisfaction followed the V-shaped pattern of GDP for those same years, but failed to recover commensurately. Across domains, increased satisfaction with material standards of living occurred at the same time that satisfaction with work, health, and family life decreased. Disparities across cohorts increased, with the unhappiest respondents being the least educated and those over age 30, not surprisingly those cohorts that were least able to protect themselves from economic dislocation and take up new opportunities offered by the transition.

Since Easterlin did his work, happiness levels have recovered in at least some of the FSU countries, such as Russia (see Eggers et al. 2006). Whether or not they will recover fully in all of them is an open question. But surely their inclusion in cross country analysis during a time period when happiness levels were unusually low, as well as that of a large number of small African economies which are likely to perform poorly for the foreseeable future, will affect the slope of the cross country income and happiness relationship.

Stevenson and Wolfers (2008), meanwhile, make a contrasting point. In their work based on the World Values Survey, they note that the survey purposefully under-sampled the poor in poor countries where the surveys were non-representative, as they sought to find “like” respondents that were comparable across countries, biasing the samples towards more urban and educated respondents. They posit that this makes the mean responses in some of the poor countries happier than their income levels would predict.12 Again, this highlights how sampling issues can affect the results.

Other Issues

Some of the complexities in the income-happiness relationship are methodological and relate to question framing and other issues. Some are empirical and relate to the sample of countries and time frame chosen for study. But perhaps some of the most interesting – and still unexplained factors – relate to the nature of economic growth and the generation of income, as well as to the institutional framework that mediates that process. There are large country level differences in these trends. Many of them are difficult to observe or measure precisely. Comparing institutional frameworks and/or the effects of macro-level trends such as growth on individual welfare are rift with methodological and other challenges. Accepting those challenges, I review here what we know about the well-being effects of economic growth and of inequality.

Economic Growth: An Unhappy Paradox

The relationship between happiness and income may be affected as much by the pace and nature of income change as it is by absolute levels. Based on the Gallup World Poll in 122 countries around the world in IADB (2008), Lora et al. find that countries with higher levels of per capita GDP have, on average, higher levels of happiness. Yet controlling for levels, they find that individuals in countries with positive growth rates have lower happiness levels. In related work, Lora and I chose to call this negative correlation between economic growth and happiness the “paradox of unhappy growth”.

Simple scatter-plots, confirmed by econometric analysis, show that the relationship between per capita incomes and life satisfaction (as measured by the best possible life question) is linear (when incomes are logged), while that between growth rates and life satisfaction is negative. Lora et al., using an OLS regression with average life satisfaction in each of the 122 countries in the Gallup World Poll as the dependent variable, find that the coefficient on GDP per capita is positive, while that on economic growth – defined as the average rate of growth over the past 5 years – is negative (and significant in both cases). See Table 2. Deaton (2008), and Stevenson and Wolfers (2008), also find evidence of an unhappy growth effect based on the full sample of the Gallup World Poll. Stevenson and Wolfers find insignificant effects of growth in general, but strong negative effects for the first stages of growth in “miracle” growth economies, such as Ireland and South Korea during their take-off stages. The negative effect becomes insignificant in later stages. Deaton finds that the inclusion of region dummies make a major difference to the results, with the significance being taken up by Africa and Russia, regions which are both fast growing and very unhappy.
Table 2

The paradox of unhappy growth. The relationship among satisfaction, income per capita, and economic growth

 

122 countries

GDP per capita

Economic growth

Marginal effects (B)

Statistical significance

Marginal effects (B)

Statistical significance

Life satisfaction

0.788

***

−0.082

***

Standard of living

0.108

***

−0.018

***

Health satisfaction

0.017

*

−0.017

***

Job satisfaction

0.077

***

−0.006

 

Housing satisfaction

0.084

***

−0.006

 

Statistical significance: * = 10%, ** = 5%, *** = 1%

• GDP per capita: The coefficients are the marginal effects: how much does the satisfaction of 2 countries differ if one has 2× the income of the other.

• Economic growth: How much does an additional % point of growth affect satisfaction

• Life satisfaction is on a 0–10 scale

• All other satisfaction variables are % of people that are satisfied.

• When sample is split between those above and below median income and growth rates, the effect holds for those above but not below median incomes

Lora at al. in IADB (2008).

In Lora’s study, economic growth is also negatively correlated with perceived standard of living and with health satisfaction. When the sample is split into rich and poor countries (above and below median income), the effect holds for the rich but not for the poor countries. The only variable that is significant and negative on growth for poor countries is health satisfaction. One can imagine the factors associated with structural change and rapid growth rates in poor countries – such as long working hours and new industries without provision for worker safety or environmental externalities – that could have negative effects on health. Meanwhile growth is negatively correlated with economic, health, job, and housing satisfaction, in addition to life satisfaction for the rich countries. When they split the sample into above and below median growth rates, the unhappy growth effect only holds for those that are growing at rates above the median.

Graham and Chattopadhyay (2008), using Latinobarometro data, find hints of an unhappy growth effect, or at least an irrelevant growth effect. We use individual rather than average country happiness on the left hand side, with the usual socio-demographic and economic controls and clustering the standard errors at the country level. When we include the current GDP growth rate in the equation, as well as the lagged growth rate from the previous year (controlling for levels), we find that the effects of growth rates – and lagged growth rates – are, for the most part, negative but insignificant (see Tables 3, 4 and 5).
Table 3

Latin Americans unimpressed by growth

 

Dependent variable: Happiness

age

−0.0240

−0.0230

−0.0230

−0.0220

(4.40)**

(4.34)**

(4.23)**

(4.29)**

age2

0.0000

0.0000

0.0000

0.0000

(3.53)**

(3.88)**

(3.72)**

(3.76)**

gender

0.0330

0.0070

0.0070

0.0070

−1.5500

−0.4800

−0.5200

−0.4800

married

0.0790

0.0910

0.0940

0.0930

−1.7800

(2.40)*

(2.56)*

(2.60)**

edu

−0.0410

−0.0260

−0.0280

−0.0260

−1.5300

−1.1800

−1.2900

−1.2800

edu2

0.0010

0.0010

0.0010

0.0010

−0.8800

−0.7000

−0.7900

−0.7600

socecon

0.2110

0.2160

0.2150

0.2170

(5.22)**

(5.76)**

(5.77)**

(5.78)**

subinc

0.2900

0.2900

0.2940

0.2920

(8.78)**

(8.02)**

(8.36)**

(8.41)**

ceconcur

0.2340

0.2260

0.2360

0.2370

(9.04)**

(9.50)**

(7.66)**

(8.92)**

unemp

−0.1810

−0.1760

−0.1900

−0.1880

(2.05)*

(3.45)**

(3.59)**

(3.69)**

poum

0.1800

0.1890

0.1830

0.1840

(4.48)**

(5.42)**

(5.56)**

(5.59)**

domlang

0.5380

0.4810

0.4840

0.4810

(2.73)**

(2.48)*

(2.48)*

(2.48)*

vcrime

−0.1160

−0.1060

−0.1060

−0.1080

(2.30)*

(2.98)**

(2.89)**

(3.08)**

els

0.0900

   

(5.48)**

   

growth_gdp

0.0170

−0.0090

−0.0040

−0.0060

−0.5300

−1.1100

−0.6000

−0.7700

gini

−0.0170

−0.0270

−0.0240

−0.0240

−0.7000

−1.2400

−1.1200

−1.1900

gdpgrl1

  

−0.0190

−0.0180

  

−1.4000

−0.9900

gdpvol2

   

0.0030

   

−0.1400

Observations

34808

67308

67308

67308

Absolute value of z statistics in parentheses

* significant at 5%; ** significant at 1%

• Ordered logit regressions, clustered at the level of the country.

• els (economic ladder scale) question asked only in select iterations of the survey, and hence reduced observations when included in the regression.

Graham et al. (2010b) based on data from Latinobarometro

Table 4

Variable descriptions for Table 3

Variable name

Variable description

happy

Happiness/Life Satisfaction: 1=Lowest 4=Highest

age

Age

age2

Age squared

gender

Gender: 1=Male 0=Female

married

Married: 1=Yes 0=No (Single/Separated)

edu

Years of education

edu2

Years of education squared

socecon

Socioeconomic Level: 1=Very bad 5=Very good

subinc

Subjective Income, Family salary fulfil needs: 1=No, great difficulty 3=Yes, without much difficulty

ceconcur

Country Economic Situation (Current): 1=Very bad 5=Very good

unemp

Unemployed 1=Yes 0=No

poum

Prospect of upward mobility, Self: 1=Worse 2=Same 3=Better

domlang

Dominant language based on mother-tongue: 1=Spanish/Portuguese 0=Other/Minority

vcrime

Victim of crime (individual or in family) in the last 12 months: 1=Yes 0=No

els

Economic ladder scale 1–10: 1=Poorest 10=Richest

growth_gdp

GDP growth rate (WDI)

gini

Gini coefficient (WDI)

gdpgrl1

GDP growth rate, lag 1 year

gdpvol2

GDP growth rate volatility, last 2 years, based on std dev of GDP growith rate for the period

Table 5

Latin America: reference income effects. The relationship among satisfaction, own income and that of others

 

122 countries

Household per capita monthly income, US$ PPP, natural log

Reference group average per capita monthly income, US$ PPP, natural log

Marginal effects (B)

Statistical significance

Marginal effects (B)

Statistical significance

Life satisfaction

0.410

***

0.254

*

Standard of living

0.370

***

−0.217

*

Health satisfaction

0.196

***

0.003

 

Job satisfaction

0.379

***

−0.429

**

Housing satisfaction

0.261

***

−0.236

**

Statistical significance: * = 10%, ** = 5%, *** = 1%

Each individual belongs to one reference group. A reference group is composed by every individual inside a country of the same gender, within the same age range and with a similar educational level.

IADB (2008) using Gallup World Poll 2006–2007 data

Another way of interpreting these findings, as noted by both Justin Wolfers and Charles Kenny, is that past income still matters to well-being but is less important than current income.13 Thus countries that started from lower income levels in past years (remember the growth rate is an average of the past 5 years) had lower happiness in those years than those with higher incomes. So the unhappy “growth” effect may be due to the starting point rather than to the effects of growth. In short, it is better to have had high levels of income for a long time, than to start at lower levels and increase them quickly.

The difficulty in disentangling these two interpretations is that both levels and changes effects are likely at play. The unhappy growth paradox focuses on the changes effect, while the Wolfers and Kenny interpretation focuses on levels. Sacks et al. (2010) note that short term changes seem to have a more marked impact on subjective well-being than long term trends.

It may well be that to the extent growth (e.g. changes in levels) is the cause of unhappiness, it is due its nature in rapidly changing economies, where growth is often accompanied by changes in rewards to different skill sets and increased job insecurity for some groups, and by related increases in vertical or horizontal inequality. Latin America in recent decades certainly fits this pattern, which may help explain unexpected pockets of frustration in relatively prosperous countries like Chile. Rapid growth in newly reforming economies, meanwhile, as in the case of Korea and Ireland, and in the case of many more recent examples in the emerging market economies, is typically even more uneven in terms of rewards. Cross-country analysis of the income-happiness relationship usually captures some sample of countries in this particular stage of development, and, indeed, as most of these are in the above median in the sample and also have the highest growth rates, they may be driving the results.

The Happy Peasants and Frustrated Achievers

There is an overall happiness and income relationship within countries, and wealthier people are, on average, happier than poor people. Yet the within country story is more complicated than the averages suggest, as in the case of the cross-country income and happiness relationship. It is typically not the poorest people that are most frustrated or unhappy with their conditions or the services that they have access to, for example. Stefano Pettinato and I, based on research in Peru and Russia, identified a phenomenon that is now termed the “happy peasant and frustrated achiever” problem.14 This is an apparent paradox, where very poor and destitute respondents report high or relatively high levels of well-being, while much wealthier ones with more mobility and opportunities report much lower levels of well-being and greater frustration with their economic and other situations. This may be because the poor respondents have a higher natural level of cheerfulness or because they lower expectations. The wealthier and more upwardly mobile respondents, meanwhile, have constantly rising expectations (or are naturally more curmudgeon-like). See Graham and Lora (2009b). In related research in the health arena, Graham et al. (2010a, b) find unusually high levels of satisfaction with health among very poor respondents with poor objective health conditions, while wealthier respondents with much better objective conditions are much more critical of their health situations.

A third explanation is also possible: more driven and frustrated people are more likely to seek to escape situations of static poverty (via channels such as migration). Yet even when they achieve a better situation, they remain more driven and frustrated than the average. Some combination of all three explanations could be at play.

A closer look at Pettinato’s and my frustrated achievers shows that they were more likely to have had upward mobility than the average, and they were of average incomes and education for their relative samples, of similar gender, and more likely to be living in urban than rural areas. Yet when compared with upwardly mobile respondents that did not report frustration, they had lower levels of general life satisfaction, they had higher fear of unemployment, and they were more concerned about relative income differences (as assessed by their scores on the poor to rich ladder question).

The behavioral economics literature highlights the extent to which individuals value losses disproportionately to gains. It is not a stretch of the imagination to assume that upwardly mobile respondents who managed to escape poverty or near poverty in the volatile macroeconomic context of both Peru and Russia in the late 1990’s would be loss averse, not least because of the absence of any safety net or social insurance system. Their income mobility, while having an overall positive trajectory, may have been punctuated with spells of unemployment or unstable income flows. If they were recent migrants, meanwhile, they also likely left strong family or other support networks behind, which are not readily available in crowded urban or peri-urban contexts.

The poor, some of whom rely on subsistence agriculture rather than earnings, have much less income to lose and have likely adapted to constant insecurity. Some new work on job insecurity, for example in IADB (2008), shows that reported insecurity is higher among more formal sector workers with more stable jobs than it is among informal sector workers. The latter have either adapted to higher levels of income and employment insecurity (and/or have selected into jobs with less stability but more freedom).

Knight and Gunatilaka (2007), and Whyte and Hun (2006), each find an analogous urban effect in China, where urban migrants who are materially much better off than they were in their pre-migration stage, yet they report higher levels of frustration with their material situation. Their reference norm, meanwhile, quickly becomes other urban residents rather than their previous peers in rural areas. In addition to comparison effects, there may also be new costs related to urban living which erode the positive effects of income gains.

More generally, the paradox highlights the extent to which slightly raised expectations in the context of rapid economic change may result in more frustration and risk aversion than do static poverty levels. To the extent that there are macro-level implications to this micro-level phenomenon, it supports a scenario where growth rates and economic development are associated with less rather than more happiness, at least in the short term, until higher income levels are stabilized. Over the long term, however, there does seem to be a generalized levels effect, with countries with higher levels of GNP on average happier than poorer ones, albeit with variance within the subsets of rich and poor countries.

Relative Incomes and Inequality – Part of the Paradox?

Happiness may not be all relative, but relative differences do seem to matter. While wealthier people are happier than less wealthy ones on average, people of similar income levels are less happy when the incomes of those in a relevant reference group, ranging from neighbors to professional cohorts, to towns and cities, are higher.15 These concerns for relative income differences, which in theory are greater as average income levels rise, are often cited as part of the explanation for the Easterlin paradox. The intuition is that until basic needs are met, people are not concerned about relative differences, and the relationship between happiness and income resembles a linear one. At higher levels, however, it curves off and resembles a logarithmic function.

At the same time, micro-level empirical work suggests that concerns for relative income differences arise at surprisingly low levels of income, as in work on Latin America by Andrew Felton and myself in Graham and Felton (2006), and by Lora and colleagues in IADB (2008), as well as on South Africa by Kingdon and Knight (2007). Ravallion and Lokshin (2005), meanwhile, test for relative deprivation effects in a much poorer context – Malawi – where basic needs are an issue for the average respondent, and find that they do not matter for most respondents in their sample, but do matter for those that are comparatively better off.

For the Gallup Poll for Latin America, Lora and colleagues find in IADB (2008) that reference group income – defined as similar age, income, and education cohorts – is positively correlated with life satisfaction (the Cantril best possible life question) but negatively correlated with satisfaction with one’s standard of living, job, and housing (see Table 5). It is likely that both question framing and variance across domains mediates the extent to which comparison effects matter. Indeed, one hypothesis – which could be tested in future research – is that the frames, which provide more visible or tangible reference points across jobs, housing, and education levels matter more for comparison effects than they do for absolute ones, while inherent character traits/optimism matter are more important to open ended life satisfaction questions. Alternatively, naturally less happy people might be more likely to be concerned about comparison effects.16

Helliwell et al. (2008), working with the Gallup World Poll, find that average per capita income levels are negatively correlated with life satisfaction (the ladder of life question), controlling for individual levels. The significance goes away when additional questions about basic needs, corruption, and freedom to choose are added to the model specification. When the sample is split by region, the coefficient on average per capita GDP only remains negative (and significant) for Eastern Europe and the FSU and for Africa, and negative (but not significant) for Latin America. Helliwell (2008) posits that for the sample as a whole, relative income effects are likely mediated by taxes and by the public goods that accompany particular countries’ distribution.

In an earlier work (see Graham and Felton (2006)) on Latin America based on the Latinobarometro, we find that average country level incomes do not matter to individual happiness, but relative income differences – measured as distance from the mean for the average income in one’s country – do matter. We looked at the effects of relative differences both across countries in the region and across city sizes. The relative income effects hold for both, with the only difference being that when the reference group was cities rather than countries, average wealth mattered (negatively) for happiness in addition to relative distance from the average. With a smaller scale reference group, it is likely that individuals are more influenced by comparison effects (which they can make more easily at the city than at the country level). See Table 6.
Table 6

Happiness and inequality in Latin America. Ordered logit estimation of a 1–4 scale of happiness

 

Average wealth calculated by:

Country

Country

Country

Country

Country

Country

City Size

City Size

City Size

City Size

Individual wealth

0.1117583

 

0.1121746

 

0.0968018

 

5.44**

−6.9**

7.96**

Average wealth

−0.0523256

0.0594327

0.0543354

0.0578392

−0.0805081

0.0162937

−0.70

0.78

−0.92

0.99

−2.19*

0.42

Relative wealth

 

0.1117583

 

0.1121746

 

0.0968018

5.44**

6.9**

7.96**

Country dummies*

N

N

N

N

Y

Y

City dummies*

Y

Y

Y

Y

Y

Y

Cluster by:

Country

Country

Country

Country

Country

Country

City Size

City Size

City Size

City Size

Demographic variables in all regressions: age, age squared, years education, married, male, health, unemp, selfemp, retired, and student

*When calculating average wealth at the country level, country dummies cannot be included in the regression due to multicollinearity. When we run split sample regressions, by city size, average wealth is positive and significant for small cities.

* t-statistics underneath coefficients

Our findings depart from those of Alesina et al. (2004) for the Untied States and Europe, where the effects of inequality (albeit measured very differently) on individual happiness are very modest. The starkest contrast is the United States, where the only group that is made less happy by inequality is left leaning rich people! Meanwhile, in Brauer and Graham (2010) based on the U.S. GSS we also find that inequality matters most to wealthier socio-economic cohorts and to those that select as upper (as opposed to middle and lower) class. In the United States, inequality remains for many respondents a sign of future opportunities and mobility, even though the data on mobility rates no longer support that perception.17 We posit that in Latin America, in contrast, inequality is still a sign of persistent advantage for the rich and disadvantage for the poor, even though the data show more mobility than public perceptions suggest.

Country level aggregations may not be the most relevant ones for studying concerns for relative income differences; the average person may be more concerned with reference groups such as neighbors or the workplace, where comparisons are more visible. We find that the relative income effect holds and is indeed more notable across cities of different sizes. It is stronger for large cities where there is more income variance and smaller for small cities, where average income levels are positively and significantly correlated with happiness, while relative incomes are still negatively correlated with happiness in the small cities.18 These findings are in keeping with those of Erzo Luttmer for the U.S. Luttmer (2005) looks across PUMAs in the U.S. census tract and finds that higher average income levels are associated with lower levels of happiness and financial satisfaction, once the effects of individual incomes are controlled for. The effects on financial satisfaction, meanwhile, are much stronger than those for life satisfaction.

The relevant reference group likely also varies across cohorts and countries and even that may still be subject to change. Pettinato and I find that our frustrated achievers assess their living standards favorably in comparison to others in their community, but much more negatively when the reference group is expanded beyond the community to others in their country, a reference group that became more relevant as information and technology such as the internet became more widely available. The Kingdon and Knight work on China shows that recent migrants quickly change their reference group to their new urban counterparts rather than the living standards in their towns of origin.

The evidence suggests that concerns for relative income differences matter and can erode the positive effects of higher absolute income levels on happiness. It also suggests that they hold at surprisingly low levels of income, as is suggested by the lack of a clear income-happiness relationship within some LDC samples. Yet it is difficult to be conclusive about how relative differences matter. One reason for this is that different reference groups matter to different cohorts or cultures, and country level incomes may not be the most relevant comparator group in many instances. In addition, concerns for relative income differences are mediated by perceptions about what inequality signals as well as the availability (or not) of public goods.

Conclusion

My aim in this essay was to help disentangle the debate on the Easterlin paradox and, more generally, the income-happiness relationship, both within and across countries. In doing so, I highlighted some of the methodological issues and challenges that are germane to both this debate and to the happiness literature more broadly.

My review, based on some of our work and that of many others, finds that while in general rich countries are happier than poor ones, there is a great deal of variance among the countries within the rich and poor clusters, as well as in the slope of the relationship. The results are quite sensitive to the method selected, the choice of micro or macro data, and the way that happiness questions are framed, thus supporting divergent conclusions about the importance of the paradox.

Question framing, for example, makes a major difference to the relationship, both in terms of direction and slope. Analysis based on questions that are framed in economic or status terms, for example, seems much more likely to yield a positive and linear relationship between income and happiness, across and within countries, than are open ended happiness or affect questions.

Country selection also matters. Respondents in poorer countries, who are still struggling to meet basic needs, display a stronger income-well-being link than do those in wealthy countries, where that relationship is mediated by factors such as relative differences and rising aspirations. There is some evidence, based on the Gallup/Cantril ladder of life question, suggesting that the slope of the income-happiness relationship is steepest at the top of the country wealth distribution, where respondents are either better positioned to enjoy wealth and/or are more aware of how their lives compare to those of others in poor countries. Several new studies suggest that the same steep slope would not hold with a more open ended life satisfaction or affect question.

The paradox of unhappy growth, meanwhile, suggests that the rate of change matters as much to happiness as do per capita income levels, and that rapid growth with the accompanying dislocation may undermine the positive effects of higher income levels, at least in the short term. The number of countries experiencing these kinds of changes at the time a survey is conducted could surely affect results. A mirror image of this paradox at the micro level – the happy peasant and frustrated achiever phenomenon – again suggests that the nature and pattern of economic growth, and in particular instability and inequality issues – can counter-balance the positive effects of higher income levels for a significant number of respondents.

The income-happiness relationship is also mediated by factors such as inequality levels and institutional arrangements, particularly as countries get beyond the basic needs level. The complexity of the relationship – and the range of other mediating factors – seems to increase as countries go up the development ladder. Rising aspirations and increasing knowledge and awareness interact with pre-existing cultural and normative differences, as well as the extent and quality of public goods, which are in turn endogenous to the cultural and normative differences. At the same time, because global information and access to a range of technologies is now available to countries at much lower levels of per capita income than was previously the case, they have access to the benefits associated with higher income levels, such as better health care, quite early on in the development process.

This essay highlighted the methodological and empirical complexities underlying the debate over happiness and income, as well as some of the deeper questions underlying that debate. Indeed, one of the fundamental contributions of happiness economics is to expose the complexities of the determinants of human welfare and the limits to which they can be captured simply via a specification of the income variable. The extensive debate on the matter at the least suggests that the answer is surely not as simple as happiness=income, log or linear. What is most notable is the remarkable consistency in the determinants of individual happiness within countries of diverse income levels and, at the same time, how happiness is affected by cross-country differences that are indirectly related to income levels, such as political freedom, the distribution of public goods, and social capital. Income surely plays a role in determining both individual and country level happiness. Still, assessing its role relative to other more difficult to measure factors as countries develop in new ways and at different rates will remain a challenge for the foreseeable future.

Footnotes
1

See Easterlin (1974and 2003); Blanchflower and Oswald (2004); Frey and Stutzer (2002a); and Graham and Pettinato (2002b).

 
2

This finding holds for people who are, on average, happier, but not necessarily for those that are the happiest in every sample. See (Diener and Biswas-Diener 2008); and Graham et al. (2004).

 
3

Deaton gets a positive and significant coefficient on a squared specification of the income variable. Stevenson and Wolfers split their sample into those countries above and below $15,000 per capita (in year 2000 U.S. dollars), they get a slightly steeper slope for the rich countries than for the poor ones.

 
4

Blanchflower and Oswald find a correlation coefficient of .50 for the two questions in Europe and the United States; Graham and Pettinato find one of .55 for Latin America, where the questions were used inter-changeably in various years of the Latinobarometro poll. See Blanchflower and Oswald (2004); and Graham and Pettinato (2002a, b).

 
5

This question is limited, at least in econometric terms, as 96% answer yes to the yes-no question.

 
6

For detail on the exact question phrasing and distribution of responses, see Graham (2010), Chapter 3.

 
7

In each country, Gallup includes a categorical question on total monthly household income, with respondents given choices within brackets expressed in local currency. The number of brackets is different in each country, and in Latin America it ranges from four brackets to 20. We relied on an adjustment to the Gallup income variable in which each household was assigned a normalized random value within the bracket that they self-reported. Income was transformed to U.S. PPP dollars and then divided by household size, resulting in a monthly per capita household income variable which is normally distributed across the sample. See Gasparini et al. (2009). While the most common adaptation for scale is to divide total household income by the square root of the number of people in the household. The Gasparini et al. adaptation divided income by the number of household members. As a check, we adjusted the same income variable by the square root of household size and got essentially the same results.

 
8

In theory, these two should be identical. In practice, with substantial misreporting at the top and with a very fat left tail (with far fewer few observations on the right/top) the log of the average may place higher relative weight on the households at the bottom of the distribution and smooth out the effects of the outliers on the right.

 
9

We used both the simple, un-weighted scale of asset ownership, and then a principle-components-analysis (PCA) based index in which the assets that are more unequally distributed across households are weighted more. Our results are essentially identical using alternative methods; the results on wealth reported in the tables are those based on the PCA approach. For more detail on the particular assets in the index and its construction, see Graham et al. (2010b).

 
10

We use the following model: average happiness (as measured by each separate question) in countryi= f (average log of per capita income or wealth in countryi+characteristics of the average individual in countryi).

 
11

The model is: individual happiness = f (household log income or wealth +personal controls +country dummies). We ran the model sequentially, first looking at just happiness and income or wealth, then adding the country dummies, and finally adding the personal controls.

 
12

See the appendix to Stevenson and Wolfers (2008).

 
13

I thank both Justin Wolfers and Charles Kenny for thoughtful conversations on this point.

 
14

For more detail, see Graham and Pettinato (2002b).

 
15

See Graham and Felton (2006); Luttmer (2005); Luttmer’s work is based on U.S. PUMA’s, geographic units which are established in census data, which proxy for neighborhoods; and Kingdon and Knight (2007).

 
16

In related research, Graham et al. (2011) show that better reference group health is positive for health satisfaction, controlling for individual levels of health. Signaling effects seem to dominate over comparison effects, most likely because there are positive externalities related to being surrounded by healthier people.

 
17

For additional details see Alesina et al. (2004). Benabou and Ok (1998); and Graham and Young (2003).

 
18

Because there is not a good income variable in the Latinobarometro, the authors use an index of assets to proxy for wealth/income. See Graham and Felton (2006).

 

Copyright information

© Springer Science+Business Media B.V./The International Society for Quality-of-Life Studies (ISQOLS) 2011