Income–happiness paradox in Australia: Testing the theories of adaptation and social comparison

Economic Modelling 30 (2013) 900–910

Contents lists available at SciVerse ScienceDirect

Economic Modelling journal homepage: www.elsevier.com/locate/ecmod

Income–happiness paradox in Australia: Testing the theories of adaptation and social comparison☆ Satya Paul ⁎, Daniel Guilbert University of Western Sydney, Penrith South DC, NSW 1797, Australia

a r t i c l e

i n f o

Article history: Accepted 6 August 2012 JEL classification: D3 D6 I3 Keywords: Life satisfaction Adaptation Comparison income Easterlin paradox

a b s t r a c t This paper investigates whether the theories of adaptation and social comparison can explain the income– happiness puzzle (Easterlin Paradox) in Australia. Alternative specifications of happiness model that incorporate adaption, comparison incomes and other relevant variables are estimated using the panel data from the five waves (2001–2005) of the Household Income and Labour Dynamics in Australia (HILDA) surveys. The statistical tests provide no support for the adaptation effect on happiness. However, we find strong support for the theory of social comparison as an explanation for the happiness paradox. An increase in peer group income hurts the poor more than the rich, suggesting that a redistribution of income is likely to enhance the overall wellbeing of society. A sensitivity analysis is conducted to check the robustness of results. © 2012 Elsevier B.V. All rights reserved.

1. Introduction Economic theory suggests that a higher income allows an insatiable consumer to reach a higher indifference curve and achieve a greater level of utility. However, the existing empirical studies of the relationship between utility (self reported happiness) and income report paradoxical results. At a point of time, people with higher levels of income are happier than those with lower income. Over the time, happiness does not increase when a country's income increases. The point of time statement is based on the cross-sectional comparison of average happiness and income within and between countries; the time series statement relates happiness with economic growth in a country. Easterlin (1974) is the first to report this paradox based on his analysis of happiness data from the yearly surveys for the United States. He reports that the rich people are happier than the poor within the US in a given year. Yet since World War II, the happiness responses are flat in the face of considerable increases in real average income. ☆ We are grateful to an anonymous referee for useful comments on an earlier draft of this paper. This paper uses unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (Melbourne Institute). The findings and views reported in this paper, however, are those of the authors and should not be attributed to either FaHCSIA or the Melbourne Institute. ⁎ Corresponding author at: School of Economics and Finance, Parramatta Campus, Room EDG 136, University of Western Sydney, Locked Bag 1797, Penrith South DC, NSW 1797, Australia. Tel.: + 61 2 96859352 (Office), + 61 404816145 (Mob). E-mail address: [email protected] (S. Paul). 0264-9993/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.econmod.2012.08.034

This happiness paradox, popularly known as the “Easterlin paradox,” is not specifically a US phenomenon. A similar picture is observed in a number of other developed countries including France, Germany, Japan and the United Kingdom at different periods of time (Easterlin, 1995; Inglehart and Klingemann, 2000; Blanchflower and Oswald, 2004; Clark et al., 2008; Easterlin and Angelescu, 2009). In Japan, despite a five-fold increase in real per capita income between 1958 and 1987, mean subjective wellbeing (happiness) has not budged. In Australia, the mean values of individual real income and happiness scores obtained from the Household Income and Labour Dynamics in Australia (HILDA) surveys reveal that while the level of income has grown, reported happiness has fallen slightly during the period 2001– 2005 (Table 1 and Fig. 1). The aim of this paper is to explain this observed income–happiness puzzle.1 To date, there appears to have been no attempt to explain the income–happiness paradox in Australia. The existing Australian studies of happiness are concerned with the effects of income, wealth and other relevant variables such as education, unemployment and age on well-being and ill-being using survey data for one or few years. For instance, Headey and Wooden (2004) used HILDA survey data for 2002 to investigate the determinant of well-being and ill-being in Australia. The well-being of an individual is measured in terms of two separate variables, life satisfaction and financial satisfaction, whereas ill-being is measured in terms of mental health and financial stress. The study reveals that the effect of wealth on life and financial 1 The terms ‘Easterlin paradox’, ‘happiness paradox’ and ‘income-happiness puzzle’ are used interchangeably throughout this paper.

S. Paul, D. Guilbert / Economic Modelling 30 (2013) 900–910

901

Table 1 Income and happiness in Australia. Source: Authors’ calculations based on data for 8530 individuals from the Household Income and Labour Dynamics in Australia (HILDA) surveys, 2001–2005. Year

Average real income

Average happiness score

2001 2002 2003 2004 2005

$24,202 $25,241 $25,218 $25,926 $27,185

7.99 7.91 7.98 7.95 7.89

satisfactions is stronger than that of income. Both income and wealth reduce financial stress. 2 The mental health is affected only by wealth whereas financial stress is affected by income as well as wealth. Females are found to be happier than their male counterparts. Dockery (2003) investigated the self reported levels of happiness of young Australians during the school-to-work transition years based on data from the Longitudinal Surveys of Australian Youths, 1967– 2002. The study reveals that young people in unemployment are less happy than those in either study or employment. The youths coming from the sole-parent households are associated with low levels of happiness, whereas those coming from the wealthier background are associated with greater levels of happiness. In a recent study based on unbalanced data from the first three waves (2001– 2003) of HILDA, Carroll (2007) reveals that unemployment has a detrimental effect on life satisfaction. In order to compensate for the effects of unemployment on unemployment men would need to be given an additional Aus$42,000 while women would need to be given AUS$86,300. We provide an empirical testing of the Easterlin paradox observed in Australia during the period of 2001–2005. The literature on happiness puts forth three theories to explain the paradox: income adaptation, social comparison and aspiration. Since the data on aspiration are not available in the Australian surveys, this paper performs empirical testing of the first two theories to explain the happiness paradox. More precisely, we specify happiness models which incorporate social comparison and adaptation incomes along with current income and many control variables such as age, gender, education, marital status, employment status and work hours. The models are estimated with panel data for 8530 individuals from the five waves (2001– 2005) of the Household Income and Labour Dynamics in Australia (HILDA) surveys. In these surveys the individuals are asked to report their happiness (life satisfaction) on a scale from 0 to 10—a standard procedure adopted in most international happiness surveys. The 0 value on the scale is labelled as ‘totally dissatisfied’ and 10 is labelled as ‘totally satisfied’. These self-reported happiness scores can be treated either as a latent variable (where comparability is assumed to be at the ordinal level) or as a cardinal variable. Most economists treat self-reported satisfaction as an ordinal concept whereas the majority of psychologists and sociologists consider it to be cardinally measurable. In our model specifications, we shall treat self-reported satisfaction as a latent variable. However, we also perform the same regressions using the cardinality assumption to check the sensitivity of results. The paper is organised as follows. We begin in Section 2 with a brief discussion of alternative theories that are used to explain happiness– income puzzle observed in developed countries. The details of happiness model and estimation details are provided in Section 3. The HILDA survey data and variables are described in Section 4. Section 5 discusses the empirical results and Section 6 presents a sensitivity analysis to check the robustness of results. A brief discussion on the impact of the structure of Australian economy on the income–happiness nexus is presented in Section 7, and Section 8 concludes the study.

2

Note that the data on wealth were collected only in the second wave of HILDA.

Fig. 1. Income and happiness in Australia.

2. The Easterlin paradox: alternative theories and international evidence As mentioned above, three theories, namely, adaptation, social comparison and inspiration, are put forward as the possible explanations for the Easterlin paradox. Income adaptation suggests that an increase in income will temporarily increase people's happiness, but as time goes by the effect wears off as people adapt or get used to their new income level. If there is complete adaptation to previous income level, income growth will not be accompanied by a higher level of happiness. The theory of aspiration, on the other hand, suggests that growing incomes lead to higher aspirations (expectations), which have a depressing effect on happiness. An increase in current income leads to a temporary increase in happiness but as time goes by the effect wears out as we revise the amount of income that we aspire to. Thus, the reference point for aspiration theory is the forward income; for adaptation, the backward income serves as a reference point (Clark et al., 2008). Testing of these two theories requires time series observations on individual income. These theories cannot be tested simultaneously if the time series is short. The theory of social comparison suggests that people do not assess their life in isolation from all others. Rather they compare their income and achievements with those around them, called the peer group (or reference group). If the income of an individual is constant, then an increase in the income of his peer group will have a depressing effect on him reducing his life satisfaction. This is so because rising peer group income reduces the relative position of the individual. Thus, it is one's relative income rather than one's absolute income, which determines life satisfaction.3 Easterlin (1995) argues that if economic growth raises the income of all such that their relative positions remain unchanged, the level of happiness in the society should remain stationary. A number of studies have tested these theories using sample survey data largely from the developed world. One of the most regularly cited studies of adaptation is that of Brickman et al. (1978) which shows that recent lottery winners derived less pleasure than controls in a variety of ordinary events and were not in general happier than controls due to adaptation. In other words, winners get used to new standard of living. Further studies on adaptation have produced mixed results. Di Tella et al. (2007) use panel data on the happiness of 7812 individuals living in Germany from 1984 to 2000 and report that two-thirds of the initial affect of income on happiness is lost after 4 years. Jørgensen and Herby (2004) generate much weaker conclusions on income adaptation when performing happiness regressions based on the European Community Household Panel (ECHP) survey data for Union member nations covering the period 1994–2001. These weaker results may be because these authors have looked at adaptation only within 1 year, which, if adaptation occurs over several years, will fail to pick up the true extent to which people adapt to changes in income. At the macro level, Di Tella et al. (2003) show that the happiness effect 3 The literature on the effect of relative income on utility can be dated back to Veblen (1899) and then Duesenberry (1949).

902

S. Paul, D. Guilbert / Economic Modelling 30 (2013) 900–910

of an increase in GDP per capita tends to disappear after 2 years in 12 OECD countries. These mixed results may be due to differences in data or adaptation period, or alternatively they could indicate that some countries adapt to higher incomes while others do not. Studies that have investigated the role of aspirations in explaining the happiness paradox are few. In a survey conducted in Switzerland between 1992 and 1994, people were asked to state the level of income that they consider ‘sufficient’, given their circumstances and expectations. Assuming that ‘sufficient’ income might be correlated with true aspiration level, Stutzer (2004) estimated happiness regressions incorporating ‘sufficient’ income as a proxy for aspiration income. The results revealed that people's subjective wellbeing is negatively affected by their aspiration level in Switzerland. A similar result is reported by Stutzer and Frey (2004) based on data from the German Socio-Economic Panel. This study also measures aspirations by the income which is evaluated as ‘sufficient’. The effect of social comparison on happiness is explored by introducing the mean income of the peer (reference) group, called the comparison income, into the regression model. A negative and significant coefficient for comparison income would mean that an increase in peer group income reduces the happiness of an individual. Important amongst the studies that have found support for the social comparison hypothesis using this approach, or an approach very similar, include Clark and Oswald (1996) for Britain; Neumark and Postlewaite (1998) and McBride (2001) for the US; Jørgensen and Herby (2004) for several European nations; Ferrer-i-Carbonell (2005) for Germany; and Stutzer (2004) for Switzerland. One of the few exceptions is Senik (2004), who finds that people in Russia are happier when their peers are earning higher incomes. Senik (2004) explains this unusual finding by suggesting that in Russia's unstable economy, people use others’ incomes when forming their own income expectations for the future. While the investigation into the effect of comparison income on happiness can be carried out with cross-section income data for 1 year, most of these studies have used panel data to achieve efficiency in estimation. Some researchers have used the experimental surveys to investigate the effect of social comparison on subjective wellbeing. For instance, in an experimental study conducted at the Harvard University in 1995, Solnick and Hemenway (1998) report that approximately 50% of the respondents preferred a world in which they had half the real purchasing power, as long as their relative income position is high. Similarly, the results of an experiment conducted by Miles and Rossi (2007) at the University of Vigo, between September and November 2004 reveal that changes in an individual's relative position affects his subjective wellbeing. In particular, learning that one's wage is below his reference group is negatively correlated with subjective wellbeing. The data from such experimental surveys, though useful, may not reveal how people truly respond to events and changes in their everyday lives. It is for this reason that most researchers have preferred to use large household surveys containing questions about income levels and subjective wellbeing. The identification of one or more theories that could explain happiness paradox is of importance for public policy. Social comparison implies that there exists a negative externality to income generating activities. The gain in happiness to the rich is accompanied by a loss of happiness to those who are poor. The standard economic argument would then suggest that income generating activities should be taxed to internalise such externalities. The adaption and inspiration do not have such straightforward policy implications. The idea that people may adapt to income changes raises questions for policymakers only to the extent that people fail to accurately forecast this adaptation when making important life decisions. For example, if people adapt more to an increase in income than they do to other positive life experiences (e.g. marriage, volunteer work etc.), and if they fail to anticipate this adaptation, then they are likely to devote too much effort to the pursuit of income and not enough effort to objectives in other domains of life. At its most basic level, this could imply that policymakers should adjust the marginal incentives associated with various life pursuits e.g. by taxing those activities for which

adaptation is quick and absolute (complete), and subsidising or promoting those activities for which adaptation is slow or incomplete. The relevance of aspirations would suggest that there may be some merit in policymakers giving less attention to the pursuit of economic aspirations, and more attention to other pursuit of other life aspirations such as good health, education, honesty and truthfulness. 3. The model and estimation The traditional utility theory assumes that utility is an increasing function of income, which is at odds with the Easterlin paradox. To account for the paradox, the utility function is formulated to incorporate adaptation and social comparison variables as 

hit ¼ α þ Aðλ; yi Þ þ γ ln yit þ δxit þ εit

ð1Þ

where hit is the self reported level of happiness (as a proxy for utility) of the i th individual, A(λ, yi) is the adaptation function, yit* is the mean income of the peer group which serves as a comparison income and xit is a vector of control variables such as age, education, gender, marital status, employment status, number of hours worked, volunteer work, commuting time etc. εit is an error term subsuming the effects of unquantifiable variables and inaccuracy in reporting life satisfaction (for example, my 4 could be your 5 and vice versa). Following Layard (2005, p. 252), the adaptation function can be specified as: Aðλ; yi Þ ¼ βð ln yit  λ ln yit1 Þ

ð2Þ

where λ is an adaptation parameter and yit and yit − 1 are the current and previous years’ incomes. β is a parameter expected to be positive. For complete adaptation λ = 1 which suggests that to remain at the same level of happiness, current income growth must match income growth from the previous year. For partial adaptability 0 b λ b 1, which implies that income growth can slow down without adversely affecting one's happiness level. There will be no adaptation to income if λ = 0 implying that to remain at the same level of happiness, no income growth is required. In this situation an increase in income should lead to a higher level of happiness. The adaptation function (2) assumes that people adapt to income growth from the previous year, i.e. adaptation is completed within 1 year. This may be a very restrictive and stringent assumption. Rather than adapting to income growth achieved in the previous year, people may adapt to average growth of income achieved over the previous few years. Such a possibility may not be ruled out if incomes fluctuate during these years. A generalised adaptation function that accommodates this may be specified as follows. Aðλ; yi ; K Þ ¼ β ln yit 

K X λk ln yitk k¼1

¼ β ln yit  ln

K

λk ∏ yitk k¼1

! ð3Þ

! ¼ βð ln yit  ln Gðλ; yi ; K ÞÞ

Note that G(λ,yi,K) is the weighted geometric mean income (WGMI) for λτ = 1, where λ is a vector of adaptation parameters and τ is a vector of unit values. For λτ b 1, G(λ,yi,K) is the weighted geometric sum of K period incomes (WGSI). For λ1 = λ2 = …= λK = 0, G(·) takes unit value implying no adaptation to income. For complete adaptation λτ = 1, which implies that to remain at the same level of happiness, current income must increase at the rate at which the WGMI has grown. For partial adaptation λτ b 1, which suggests that to remain at the same level of happiness, current income must grow at the rate at which the WGSI has grown. Note that WGSI b WGMI. Hence in the case of partial adaptability, a somewhat

S. Paul, D. Guilbert / Economic Modelling 30 (2013) 900–910

lower growth of current income can attain the same level of happiness. The choice of K in the function is left to the judgement of the researcher and may be influenced largely by the availability of data. Substituting (3) into (1) we have (model A): hit ¼ α þ β ln yit þ

K X



βk ln yitk þ γ ln yit þ δxit þ εit

ð4Þ

k¼1

where βk =−βλk (k=1, 2, …, K). This equation forms the basis of testing adaptation to income in a recent NBER Working paper by Di Tella et al. (2007), though no mention is made of an underlying adaptation function. The adaptation functions (2) and (3) serve to provide an analytical support to the interpretation of adaptation implicitly built into (4). Treating self-reported life satisfaction hit as a latent variable, an efficient estimation can be conducted using an ordered probit model. The inferences on adaptation can be drawn as follows. If β is positive and statistically significant and (β + Σβk) = 0 then there is evidence of complete adaptation to income. There will be an evidence of partial adaptation if β > 0, (Σβk) b 0 and (β+ Σβk) > 0. The hypothesis of no adaptation cannot be rejected if β > 0 and β1 = β2 = …= βk = 0. 4 With no evidence to income adaptation, the model can be estimated with panel data accommodating individual random effects: 

hit ¼ α þ β ln yit þ γ ln yit þ δxit þ ηi þ uit

ð5Þ

where ηi is an individual random effect and uit is the usual error term assumed to be uncorrelated with observable variables. The individual random effect – which captures the effects of personal traits such as pessimism or optimism, depression and intelligence of individuals – may be correlated with some observable variables such as current income, work hours and commuting time. For example, a depressed person may work less leading to the loss of job and income, while a person lacking motivation may decide not to take up a lucrative job that involves long hours commuting to and from the work place. A widely used solution to this problem is suggested in Mundlak (1978). He accounts for the relationship between individual random effects and some of the observable variables by assuming the following structure of correlation (also see Hsiao, 1986; Ferrer-i-Carbonell, 2005). ηi ¼ ωi þ ∑ ϕj ln  z ji

ð6Þ

j

The individual random effect is decomposed into two components: (i) a pure random effect, ωi which is not correlated with observable explanatory variables, and (ii) a component correlated with a subset, zji of observable explanatory variables. zji is the average of zji across years. The correlation between zji and the random effect is assumed to be of the form ϕj lnz ji . As emphasised in Ferrer-i-Carbonell (2005), ϕj represents only a statistical correction, and no specific significance should be attached to its magnitude and sign. Substituting (6) into (5) we have the following equation (model B) estimable by ordered probit. 

hit ¼ α þ β ln yit þ γ ln yit þ δxit þ ∑ ϕj ln zji þωi þ uit j

ð7Þ

All the models specified above assume that comparison income effects on the happiness of poorer and richer individuals are identical. In the present context, the poorer are those whose incomes are lower than the comparison (reference) income, and the richer are those with incomes above the comparison income. It is possible that comparison income hurts the poorer individuals more than the richer individuals. 4 The hypothesis of no adaptation may not also be rejected if Σβk = 0. This is a somewhat weaker condition than β1 = β2 = … = βK = 0. It may also be noted that the testing of (β+Σβk) = 0 is equivalent to the testing of λτ = 1, the testing of (β + Σβk) > 0 is equivalent to the testing of λτ b 1, and the testing of Σβk = 0 is equivalent to the testing of λ1 = λ2 = … = λK = 0. The testing of adaptation hypothesis in terms of βs is easier than in terms of λs.

903

It is also equally possible that comparison income hurts the poorer individuals only. That is, it may not hurt the richer individuals at all. The latter possibility is consistent with Runciman's (1966) theory of relative deprivation which states that a person suffers from relative deprivation if his income is lower than his peers, but his deprivation is zero otherwise. 5 To test these possibilities (or say hypotheses), we specify a dummy variable, RICHER ¼ ð ln yit Þ if yit > yit ¼ 0 if yit ≤ yit :

ð8Þ

Eq. (7) is, thus extended to (model C): 

z ji þ ωi þ uit ð9Þ hit ¼ α þ β ln yit þ γ ln yit þ γ1 RICHER þ δxit þ ∑ ϕj  j1

Here the effect of peer group income upon the happiness of a poorer individual is captured by γ and that of a richer individual by γ + γ1. If γ b 0, and γ1 > 0 but less than |γ|, then this would imply that comparison income adversely affects the happiness of both the richer and poorer individuals but the effect is weaker for richer individuals. γ + γ1 = 0 provides support for Runciman's theory of relative deprivation, meaning that peer group income does not affect the happiness of richer individuals. 4. Data and variables This study makes use of panel data from the Household Income and Labour Dynamics in Australia (HILDA) surveys, which asks detailed questions about economic and subjective wellbeing, as well as labour market and family dynamics. We use information contained within the annual personal and household questionnaires from the years 2001–2005 inclusive. This study includes only those people who responded to each of the five available waves of the HILDA surveys. As a result, there are 9,311 individuals and 46,555 observations (i.e. 9311×5) available for analysis. On those occasions when the individual records missing data for one or more variables included in the regression, all observations for that individual during that year are dropped from the regression analysis. Hence observations have varied from 6163 to 8530 per wave depending on the model specification and variable requirements. The variables used in the estimation are measured as follows. Life satisfaction (happiness or wellbeing) is measured on a scale numbered from 0 to 10 according to each person's response to the following question: “All things considered, how satisfied are you with your life?” 6 Individual income is defined as financial year disposable personal income. The individual respondents are asked to provide extensive details about their income from all sources during the preceding financial year. The income sources include wages and salaries, business income, rental income, share dividends, private superannuation income, private transfers (e.g. child support payments) and public transfers (pensions etc.). The respondents were asked no question about taxes since most people could not answer accurately. So in order to estimate disposable income, taxes had to be computed taking into account income tax rates, the Medicare levy and applicable tax offsets and credits, see Headey and Wooden (2004). The details of tax computations are provided in Wilkins (2009). It must be noted that disposable income thus calculated may involve some errors for two reasons. First, data 5 For further details of the theory of relative deprivation, see Yitzhaki (1979), Paul (1991) and Bossert and Ambrosio (2007). 6 While the validity of self-reported happiness statistics has been a source of considerable debate in recent years, existing studies appear to suggest that there is a lot of important and reliable information contained within these figures (see Layard, 2005; Gilbert, 2006; Schimmack, 2006 among others). This paper assumes therefore that the self-reported happiness statistics used here are valid, and does not explore this issue any further.

904

S. Paul, D. Guilbert / Economic Modelling 30 (2013) 900–910

are likely to suffer from recall bias as it is not possible to recall accurately the details of income earned during the preceding financial year. Second, the imputation of taxes may not match the actual taxes paid by an individual. It is very likely that an individual saves on taxes with the help of a smart accountant. Disposable income is therefore likely to be biased downwards and it is not possible to know the extent of this bias. This should be kept in mind while interpreting the results of this study. All incomes are converted into constant 2001 prices using consumer price indices available from the Australian Bureau of Statistics (ABS, 2007). To prevent zero income values from being treated as missing data, $1 is added to all incomes before taking the log values. The definition of peer group income is one of the arbitrary decisions involved in happiness research. While some people may compare themselves with siblings or childhood friends, others may compare themselves with colleagues at work or those with a similar level of educational attainment. Unfortunately, little information can be derived from household surveys, including HILDA, regarding the group of people against which an individual compares his income. It is therefore left to the researcher to define the peer group. Some researchers choose to define peer groups in a broad manner. For example, McBride (2001) defines peer groups according to age, while Stutzer (2004) uses region as the defining characteristic. Other researchers are far more specific when defining peer groups. For example, van de Stadt et al. (1985) define peer groups in terms of age, education and employment status, Ferrer-i-Carbonell (2005) uses age, education and region, and Jørgensen and Herby (2004) use age, education, region and gender. Neighbours serve as peer groups in Luttmer (2005). This study strikes a balance between these two extremes. We define peer groups by age and education, whereby all those who are within 15% of the individual's age and have attained the same level of education form the peer group. 7 The mean income of the peer group is called the comparison income or simply the peer income. To check the sensitivity of results, an experiment is made by defining peer group based on age, education and gender. We use binary variables to control for a wide range of individual characteristics that may influence life satisfaction. Most of these control variables are commonly used throughout the happiness literature. They include marital status, employment status, sex, education, racial status, health status and location. For marital status, a series of dummy variables is used for married, divorced, separated and widowed persons (those who have never married serve as the reference group). Binary variables are generated for females, university degree holders, indigenous Australians, those who suffer from poor health, and those who live in a major city. People are considered as suffering from poor health if they have a long-term health condition. For employment status, dummy variables are used for those who are unemployed and those who are outside the labour force (employed persons act as the reference group). The HILDA survey uses the ILO definitions to identify the employment status of individuals. A person is considered to be unemployed if he or she is not in employment but is actively looking for a job or is available for work. Employed is the one who works for pay or profits for one or more hours per week. People not in labour force are those who are neither employed nor unemployed. We also include a range of control variables that are not often used in happiness studies, but which may have a significant impact on individual happiness levels. For example, a binary variable is generated for volunteers i.e. those who perform one or more hours of volunteer or charity work on average per week. A binary variable is also created

7 This means, for instance, that a 20 year old male compares himself only with those people aged between 17-23 years, while a 50 year old male will compare himself only with those people aged between 43-57 years within his education category. A person's education level is categorised into one of two groups: those who have attained a university degree and those who have not.

for those who have to care for a disabled spouse or relative, and another for those whose parents have ever divorced or separated. Other control variables are continuous variables. For example, commuting time is measured as average hours spent commuting to and from work each week. This variable is included because the time taken to commute to and from work could be otherwise spent either earning money at work or pursuing leisure. Finally, work hours is measured as average work hours per week. Due to the likely non-linear nature of their relationships with happiness, both the commuting time and work hours variables are taken in natural log values in the regression equations. 8 5. Empirical results We begin with the estimation of model A, which includes current income and income from the previous 4 years, along with peer group income and other relevant control variables. The results are presented in column 1 of Table 2. The first point to note is that current income has a positive but statistically insignificant effect on self-reported satisfaction. This poses questions about the possibility of adaptation to income. If people do not grow significantly happier after an increase in personal income, there is little chance of capturing (or identifying) adaptation in the empirical results, as people have nothing to which they can adapt. The results for the lagged income variables confirm this. Aside from the third income lag, all other lagged income variables are positive but not statistically significant. The third income lag is negative but it too is not statistically significant. The null hypothesis of β1 = β2 = β3 = β4 = 0 is also not rejected. This, along with the fact that the coefficient of current income is not statistically significant, suggests that adaptation is not identified. 9 To test whether adaptation takes place in less than 4 years, we re-estimated the model with 3 years, 2 years and then 1 year of lagged income with random effects ordered probit. The results are presented in columns 1–3 of Appendix A Table A. When three lags are used, the coefficient of current income turns significant but the coefficients of lagged incomes remain statistically insignificant. When the third lag is dropped, the current income effect loses significance, the coefficient of the first lag becomes positive and statistically significant, and the coefficient of the second lag remains insignificant. When the second lag is dropped, coefficients of the current and 1 year lagged incomes show no changes in their magnitude and signs. 10 In another experiment, we introduced the log of the geometric mean for the incomes of the previous 4 years. The results, presented in column 2 of Table 2, again indicate that past income has no statistically significant effect on self-reported life satisfaction. Thus, adaptation could not be captured in any of these experiments. The small yet statistically insignificant positive effect of current income, as observed in columns (1) and (2) of Table 2 might be due to measurement error in data. As discussed in the previous section, individual income in the HIDLA survey is measured with errors. This is likely to lead to attenuation bias in the coefficient on current income. If incomes for other years are highly correlated with the current income, which is subject to measurement errors, then the coefficients on the lagged incomes in models A and B are also likely to be attenuated towards zero. We note from Table 3 that there is a high degree of correlation between each year's income levels. In this case, the 8 The log (x + 1) transformation is used for both commuting time and work hours. This prevents zero values from becoming missing data when logged. 9 Note that the hypothesis of β1 + β2 + β3 + β4 = 0 is also not rejected (see Table 2, Column 1). 10 To see the sensitivity of results, we included the variable ‘RICHER’ in model A and estimated with four, three, two and one lags of income. These results, presented in Columns 4 through 7 of Appendix Table A, also lend no support to the adaptation hypothesis. We also re-estimated these models assuming the cardinality of life satisfaction responses and found no support for the adaptation hypothesis. These results are not reported to save space.

S. Paul, D. Guilbert / Economic Modelling 30 (2013) 900–910

905

Table 2 Ordered probit estimates of happiness models. Explanatory variables

ln(Income) ln(Income-1) ln(Income-2) ln(Income-3) ln(Income-4)

With peer groups based on age and education Model A (Eq. 4)

Model A (Eq. 4 with geometric mean of past incomes)

Model B (Eq. 7)

Model C (Eq. 9)

Model C (Eq. 9)

Model C (Eq. 9) (restricted version)

(1)

(2)

(3)

(4)

(5)

(6)

0.014 (0.010) 0.003 (0.009) 0.006 (0.009) −0.010 (0.009) 0.005 (0.007)

0.017+ (0.010)

0.009 (0.006)

0.004 (0.006)

0.006 (0.006)

0.008 (0.006)

−0.967** (0.152)

−0.002 (0.005) −0.942** (0.148)

−0.407** (0.049)

0.028** (0.010) −0.0002+ (0.0001) Ref. group 0.317** (0.067) Ref. group 0.105** (0.025) Ref. group −0.280* (0.136) −0.200* (0.093) Ref. group −0.410** (0.028) Ref. group −0.124** (0.024) Ref. group 0.280* (0.116) Ref. group 0.272** (0.036) −0.226** (0.079) −0.172** (0.061) 0.073 (0.070) Ref. group −0.126** (0.043) Ref. group 0.098** (0.028) Ref. group −0.051 (0.039) −0.064* (0.027) −0.029+ (0.017)

0.027** (0.010) −0.0002** (0.0001) Ref. group 0.309** (0.066) Ref. group 0.105** (0.025) Ref. group −0.280* (0.136) −0.199* (0.093) Ref. group −0.409** (0.028) Ref. group −0.124** (0.024) Ref. group 0.282* (0.116) Ref. group 0.273** (0.036) −0.223** (0.079) −0.170** (0.061) 0.075 (0.070) Ref. group −0.126** (0.043) Ref. group 0.098** (0.028) Ref. group −0.050 (0.039) −0.064* (0.027) −0.028+ (0.017)

−0.399** (0.049) 0.007** (0.002) −0.022** (0.007) 0.0004** (0.0001) Ref. group 0.060 (0.041) Ref. group 0.143** (0.033) Ref. group −0.488** (0.082) −0.306** (0.069) Ref. group −0.306** (0.021) Ref. group −0.182** (0.028) Ref. group 0.078 (0.160) Ref. group 0.382** (0.038) −0.409** (0.064) −0.210** (0.060) −0.079 (0.072) Ref. group −0.075* (0.030) Ref. group 0.080** (0.022) Ref. group −0.140** (0.052) −0.092** (0.022) −0.001 (0.015) 0.011 (0.015) 0.024 (0.019) −0.124** (0.035)

−0.407** (0.047) 0.003++ (0.002) −0.024** (0.007) 0.0004** (0.0001) Ref. group 0.062 (0.041) Ref. group 0.044 (0.038) Ref. group −0.469** (0.082) −0.287** (0.069) Ref. group −0.306** (0.021) Ref. group −0.181** (0.028) Ref. group 0.074 (0.160) Ref. group 0.385** (0.038) −0.408** (0.064) −0.207** (0.060) −0.067 (0.072) Ref. group −0.076* (0.030) Ref. group 0.078** (0.022) Ref. group −0.144** (0.052) −0.086** (0.022) −0.001 (0.015) 0.014 (0.015) 0.027 (0.019) −0.118** (0.036)

−0.412** (0.047) 0.004+ (0.002) −0.021*** (0.007) 0.0004** (0.0001) Ref. group 0.071 (0.041) Ref. group 0.056 (0.037) Ref. group −0.451** (0.081) −0.279** (0.069) Ref. group −0.307** (0.021) Ref. group −0.189** (0.028) Ref. group 0.072 (0.159) Ref. group 0.389** (0.038) −0.407** (0.064) −0.198** (0.060) −0.057 (0.072) Ref. group −0.076* (0.030) Ref. group 0.078** (0.022) Ref. group −0.144** (0.052) −0.076** (0.021) −0.005 (0.014)

ln(Geometric mean of past incomes) ln(Peer income)

With peer groups based on age education and sex

RICHER Age Age-squared No degree Degree Male Female Employed Unemployed Not in labour force Good health Poor health Living outside city Living in city Not indigenous Indigenous Never married Married Separated Divorced Widowed Care not performed Care performed Not volunteer Volunteer Parents together Parents divorced or separated ln(Work hours) ln(Commuting) ln(Mean income) ln(Mean work hours) ln(Mean commuting) Ho: β1 = β2 = β3 = β4 = 0

−0.022** (0.007) 0.0004** (0.0001) Ref. group 0.054 (0.041) Ref. group 0.135** (0.033) Ref. group −0.468** (0.081) −0.285** (0.068) Ref. group −0.307** (0.021) Ref. group −0.179** (0.028) Ref. group 0.084 (0.160) Ref. group 0.384** (0.038) −0.405** (0.064) −0.206** (0.060) −0.073 (0.072) Ref. group −0.075* (0.030) Ref. group 0.079** (0.022) Ref. group −0.141** (0.052) −0.085** (0.022) 0.001 (0.015) 0.014 (0.015) 0.028 (0.019) −0.120** (0.036)

−0.079** (0.029)

1.64 (continued on next page)

906

S. Paul, D. Guilbert / Economic Modelling 30 (2013) 900–910

Table 2 (continued) Explanatory variables

χ2 Ho: β1+ β2 +β3+β4 = 0 χ2 Wald-statistic Likelihood ratio χ2 No. of observations

With peer groups based on age and education

With peer groups based on age education and sex

Model A (Eq. 4)

Model A (Eq. 4 with geometric mean of past incomes)

Model B (Eq. 7)

Model C (Eq. 9)

Model C (Eq. 9)

Model C (Eq. 9) (restricted version)

(1)

(2)

(3)

(4)

(5)

(6)

1238.78**

1249.23**

1246.89**

1244.02**

30,815

30,815

30,815

30,878

0.43 908.09**

8,530

905.69**

8530

Note: Standard errors are reported in parentheses. In models A, standard errors are robust to heteroskedasticity. In all remaining models, results are derived using the REOPROB command in STATA 10.0. Unfortunately, there is currently no procedure that can be used to generate robust standard errors when using the REOPROB command. For this reason, the standard errors presented in models B and C are not robust to heteroskedasticity. **, *, + and ++ denote levels of significance respectively at the 1, 5, 10 and 15%.

rejection of adaptation hypothesis might simply reflect the presence of measurement errors in the income data. Peer group income, which is calculated as the average income for a large group of people,11 is less likely to be subject to errors. With no supporting evidence for adaptation, we drop all income lags and proceed to estimate model B (Eq. 7) with the panel data set for the period 2001–2005. This model incorporates individual random effects and makes use of the Mundlak correction to address the possible correlation between unobservable personal traits and a sub-set of explanatory variables. The results are presented in column 3 of Table 2. Once again, current income has no statistically significant effect upon life satisfaction. The coefficient on peer group income is − 0.407 and this is statistically significant at the 1% level. This suggests that an increase in the average income level of the individual's peer group has a negative effect on the individual's life satisfaction. Since the coefficient of income (y) is 0.004 with a standard error of 0.006, the negative and significant coefficient on peer group income implies that a 1% rise in both personal income and peer group income would leave an individual worse off. A similar result is observed in Clark and Oswald (1996) for Britain. In their study, the coefficient of peer group income is − 0.2 with a standard error of 0.06 and that of income is 0.11 with a standard error of 0.050. The magnitude of the comparison income effect in our study is stronger than that in Clark and Oswald. We can think of two possible explanations for this. First, our method of computing comparison income (as discussed in Section 3) is different from that used in Clark and Oswald. The latter study estimates an earning equation to calculate comparison income for each individual. A comparison income is derived by predicting the typical income of someone with the individual's observable characteristics. Second, the people in Australia could be more envious than those in Britain. The higher the degree of envy, the greater would be the effect of peer group income on the individual's wellbeing. If the degree of envy is zero, then peer group income would not affect the wellbeing of an individual. Finally, we turn to the estimation results based on model C (Eq. 9) which is designed to test the differential effects of peer group income on the wellbeing of poorer and richer individuals. The comparison income effect remains negative (− 0.399) and statistically significant. The variable ‘RICHER’ has a positive coefficient (0.007) and is significant at the 1% level of confidence (Table 2, column 4). This suggests that an increase in peer group income affects the rich slightly less than it affects the poor. This is consistent with the results found in Ferrer-i-Carbonell (2005) for Germany, and could imply that there is a case for income redistribution, considering that greater income equality would raise the subjective wellbeing of the poor more than it would reduce the subjective wellbeing of the rich. This study, 11 On average, there are about 860 individuals in each peer group constructed based on age and education; some peer groups have over 2000 individuals.

however, finds no support for Runciman's theory of relative deprivation, which suggests that increases in peer group income do not hurt richer individuals. The coefficients of control variables seem to be mostly in line with our expectations. Consistent with other studies, happiness is revealed to be U-shaped in age. However, unlike most studies, which find that happiness minimises at around 40 years of age, 12 this study finds that happiness reaches a minimum during the late 20s. It appears that life in Australia is most stressful for those in their late 20s, whose thoughts typically centre on concerns regarding social relationships, personal identity and uncertain career prospects. Females are significantly happier than males. While a similar result is reported in several studies including Clark and Oswald (1996) for Britain, Blanchflower and Oswald (2004) for the US and Britain, and Headey and Wooden (2004) for Australia, no explanation is provided to explain this phenomenon. One possible reason is that women do more of pleasurable activities such as leisure and looking after their children, whereas males spend more time at work and less time with families and friends. Another possible reason is that females might have very low or no aspirations, whereas males have high aspirations which if remain unfulfilled become the cause of their frustration. However, we do not have data on levels of aspirations of males and females to verify this point. Suffering from poor health has a negative (− 0.306) and significant effect upon life satisfaction. Marriage has a positive (0.385) and significant effect on life satisfaction, while separation and divorce exert negative and significant effect on life satisfaction. Married life more than neutralizes the consequences of poor health. Living in a major city has a negative and significant effect upon life satisfaction. Those who are unemployed or out of the labour force are less satisfied with life compared to those who are employed, although it is worth noting that the number of hours spent working each week has a negative impact upon life satisfaction. The finding that unemployment is a significant depressor of subjective wellbeing is consistent with the literature. 13 Indeed, the effect of unemployment appears to be stronger than the effect of divorce or separation i.e. isolation from one's spouse. Those who perform volunteer or charity work tend to be happier than those who do not. From a policy perspective this finding suggests that it could be worthwhile to put more effort into promoting volunteer work and emphasising its benefits not only to the community but also to the volunteers themselves. Conversely, those who

12 See, for example, Blanchflower and Oswald (2004) for Britain and the US, and Ferrer-i-Carbonell (2005) for Germany. 13 See, for example, Clark and Oswald (1994) for Britain, Paul (1992) for the US, Blanchflower and Oswald (2004) for Britain and the US, Stutzer (2004) for Switzerland, and Dockery (2003), Paul (2001), Headey and Wooden (2004) and Carroll (2007) for Australia.

S. Paul, D. Guilbert / Economic Modelling 30 (2013) 900–910

7. Structure of Australian economy and income–happiness nexus

Table 3 Coefficient of correlation between individual real incomes, 2001–2005. Individual real income

2001 2002 2003 2004 2005

907

Individual real income 2001

2002

2003

2004

2005

1.000 0.786 0.770 0.659 0.681

1.000 0.775 0.669 0.700

1.000 0.742 0.730

1.000 0.738

1.000

All values of coefficient of correlation are statistically significant at the 1% level. Standard errors not reported to save space.

have to care for a disabled spouse or relative tend to be less satisfied than those who do not. While this finding is not surprising, it does highlight the strong impact that the provision of care and other unpaid work can have on quality of life. Commuting time, race and widowhood do not have a significant effect upon life satisfaction. The effect of education on life satisfaction is rather ambiguous. Model A suggests that completing a university degree raises happiness, but in following models this effect is not statistically significant. Previous international studies also provide conflicting results on the matter. For example, Ferrer-i-Carbonell (2005) and Stutzer (2004) report a positive effect of education respectively in Germany and Switzerland, whereas Clark and Oswald (1996), whereas Jørgensen and Herby (2004) and Dockery (2003) report a negative effect in Britain, and European community respectively. 6. Sensitivity analysis As stated above, definitions of relevant peer groups are quite arbitrary, and it is possible that empirical results are sensitive to the manner in which relevant peer groups are defined. This study therefore provides results for an alternative definition of peer groups. Instead of forming peer groups in terms of age and education, the peer groups are defined in terms of age, education, and sex. The results of model C incorporating new comparison incomes are presented in column 5 of Table 2. The coefficient of comparison income is negative (− 0.407) and statistically significant at the 1% level of confidence. The coefficient of ‘RICHER’ is positive (0.003) and significant at the 15% level of significance. The coefficients associated with mean work hours and mean incomes used in the Mundlak correction are not statistically significant. We dropped these two variables and reestimated the model. These results are presented in column 6 of Table 2. The coefficient of comparison income has remained almost the same but the coefficient of ‘RICHER’ has improved in terms of both magnitude and the level of statistical significance. Thus the hypothesis that the comparison income hurts the poor more than the rich still holds. Some researchers have assumed cardinality of responses when performing happiness regressions (for examples, see Ferrer-i-Carbonell and Frijters, 2004). To ascertain how this assumption may influence empirical results, we repeated all estimations treating the individual responses on the 0-to-10 scale as cardinal measures of satisfaction. The results are presented in Table 4. The hypothesis of no adaptation to income is not rejected; the income coefficient is once again positive but not statistically significant. Comparison income has a negative effect on the happiness of both poorer and richer individuals, and the effect on the former is still stronger. On a close comparison of Table 4 with Table 2 one can see that the cardinality assumption does not affect the signs or significance of most of the variables included in this study. Instead of associating happiness with individuals’ incomes, a few researchers have tried to relate happiness of individuals with the household incomes they belong to (see, for example, Headey and Wooden, 2004). We carried out few regressions using per capita household income. This, however, did not alter our main conclusions: the theory of adaptation is still rejected and comparison income effects on happiness are found to be significant.

It is worthwhile to see whether the structure of the Australian economy has any impact on the income–happiness nexus. In the 19th century, Australia was oriented towards primary production with only a small manufacturing industry. Agriculture accounted for around one third of output. Over the years, the structure of Australian economy has gradually shifted away from agriculture and manufacturing towards services. Lately, Australia has experienced a mining boom leading to extremely high wages in certain industries followed through to other industries. With rising wages, the rational individuals (agents) are expected to allocate more time to the paid work. The HILDA data reveal that the time allocated to the paid work in Australia has increased from 23.4 h per week in 2002 to 24 h per week in 2005. This might have added to the stress of individuals. In an empirical study based on data from the first six waves of the HILDA surveys, Jeon and Fok (2009) report that work stress has a detrimental effect on the happiness of individuals in Australia. More importantly, an individual's decision to allocate more time to the paid job crowds out time for relational goods. This possibility cannot be ruled out because the opportunity cost of consuming relational goods tends to rise with rising wages. The relational goods are the interactions with family members, friends and relatives. An empirical study by Becchetti et al. (2011), which analyses data for more than 100,000 individuals from the representative samples in 82 countries (including Australia), reveals that a decline in the consumption of relational goods adversely affects the level of happiness. Thus, the rising incomes if accompanied by increasing stress and declining relational goods serve as an explanation for the income- happiness paradox. We hope that further research in this direction should add to our understanding of the income–happiness nexus. 8. Conclusions During the period of 2001–2005, real income has grown but happiness has remained constant or declined slightly in Australia. This paper has made an attempt to explain this happiness puzzle using theories of adaptation and social comparison. Using HILDA survey data, we estimated models of happiness incorporating these theories along with several other relevant control variables. In the happiness equations, a generalised version of Layard's adaptation function accounts for adaptation, and peer group incomes account for social comparison effect. The empirical results provide no support to the hypothesis of adaptation. Since the income data in HILDA are measured with errors, we are inclined to infer that the rejection of the adaptation theory simply reflects the presence of measurement errors in data. The results identify social comparison theory as the key explanation for the observed happiness paradox. Peer group income has a negative and significant effect on the happiness of individuals. The increase in peer group income hurts the poor more than rich. While the difference is minimal, it does suggest that income redistribution from the rich to the poor could result in a positive net effect on the level of subjective well-being in society. The number of hours spent at work has a negative impact on life satisfaction. However, the same work effort, when applied to voluntary or charitable pursuits, has a positive impact on life satisfaction and is likely to provide additional benefits to the community. For a policy perspective, this finding suggests that it could be worthwhile to put more efforts into promoting volunteer work and emphasising its benefits not only to the community but also to the volunteers themselves. Happiness is revealed to be U-shaped in age, reaching a minimum during the late 20s. Females are significantly happier than males. Suffering from poor health has a negative and significant effect upon life satisfaction. Living in a major city has a negative and significant effect

908

S. Paul, D. Guilbert / Economic Modelling 30 (2013) 900–910

Table 4 Estimates of happiness models assuming cardinality of self reported life satisfaction responses. Explanatory variables

ln(Income) ln(Income-1) ln(Income-2) ln(Income-3) ln(Income-4)

With peer groups based on age and education Model A (Eq. 4)

Model A (Eq. 4 with geometric mean of past incomes)

Model B (Eq. 7)

Model C (Eq. 9)

Model C (Eq. 9)

(1)

(2)

(3)

(4)

(5)

0.021 (0.014) 0.002 (0.0013) 0.011 (0.013) −0.010 (0.012) 0.007 (0.009)

0.026** (0.013)

0.009 (0.006)

0.003 (0.006)

0.005 (0.006)

−1.385** (0.200)

−0.002 (0.007) −1.343** (0.194)

−0.325** (0.049)

0.042** (0.013) −0.0003* (0.0001) Ref. group 0.472** (0.088) Ref. group 0.148** (0.034) Ref. group −0.453* (0.196) −0.251* (0.122) Ref. group −0.574** (0.039) Ref. group −0.151** (0.032) Ref. group 0.328* (0.149) Ref. group 0.364** (0.051) −0.412** (0.122) −0.285** (0.089) 0.099 (0.092) Ref. group −0.167** (0.058) Ref. group 0.158** (0.036) Ref. group −0.071 (0.055) −0.061+ (0.036) −0.035 (0.022) 20.574** (1.761)

0.041** (0.013) −0.0003* (0.0001) Ref. group 0.459** (0.087) Ref. group 0.146** (0.034) Ref. group −0.451* (0.196) −0.248* (0.123) Ref. group −0.573** (0.039) Ref. group −0.151** (0.032) Ref. group 0.331* (0.149) Ref. group 0.366** (0.051) −0.408** (0.122) −0.282** (0.089) 0.103 (0.092) Ref. group −0.167** (0.058) Ref. group 0.157** (0.035) Ref. group −0.071 (0.055) −0.059+ (0.036) −0.034+ (0.022) 20.22** (1.713)

−0.314** (0.049) 0.008** (0.002) −0.025** (0.007) 0.0004** (0.0001) Ref. group 0.059* (0.036) Ref. group 0.137** (0.029) Ref. group −0.474** (0.085) −0.282** (0.067) Ref. group −0.299** (0.021) Ref. group −0.153** (0.026) Ref. group 0.081 (0.148) Ref. group 0.358** (0.037) −0.513** (0.080) −0.243** (0.065) −0.069 (0.077) Ref. group −0.071* (0.030) Ref. group 0.085** (0.020) Ref. group −0.155** (0.051) −0.078** (0.021) 0.003 (0.014) 11.336** (0.398) 0.012 (0.014) 0.037* (0.018)

−0.325** (0.046) 0.005* (0.002) −0.026** (0.007) 0.0004** (0.0001) Ref. group 0.064+ (0.035) Ref. group −0.018 (0.034) Ref. group −0.458** (0.085) −0.265** (0.067) Ref. group −0.299** (0.021) Ref. group −0.152** (0.026) Ref. group 0.081 (0.147) Ref. group 0.360** (0.037) −0.513** (0.080) −0.242** (0.065) −0.059 (0.077) Ref. group −0.071* (0.030) Ref. group 0.084** (0.020) Ref. group −0.157** (0.051) −0.072** (0.021) 0.004 (0.014) 11.487** (0.389) 0.014 (0.014) 0.040* (0.018)

ln(Geometric mean of past incomes) ln(Peer income)

With peer groups based on age education and sex

RICHER Age Age-squared No degree Degree Male Female Employed Unemployed Not in labour force Good health Poor health Living outside city Living in city Not indigenous Indigenous Never married Married Separated Divorced Widowed Care not performed Care performed Not volunteer Volunteer Parents together Parents divorced or separated ln(Work hours) ln(Commuting) Constant ln(Mean income) ln(Mean work hours)

−0.025** (0.007) 0.0004** (0.0001) Ref. group 0.055 (0.036) Ref. group 0.128** (0.029) Ref. group −0.451** (0.085) −0.258** (0.066) Ref. group −0.300** (0.021) Ref. group −0.149** (0.026) Ref. group 0.086 (0.148) Ref. group 0.360** (0.037) −0.509** (0.080) −0.239** (0.065) −0.060 (0.077) Ref. group −0.071* (0.030) Ref. group 0.085** (0.020) Ref. group −0.157** (0.051) −0.069** (0.021) −0.004 (0.014) 11.351** (0.398) 0.016 (0.014) 0.042* (0.018)

S. Paul, D. Guilbert / Economic Modelling 30 (2013) 900–910

909

Table 4 (continued) Explanatory variables

With peer groups based on age and education

With peer groups based on age education and sex

Model A (Eq. 4)

Model A (Eq. 4 with geometric mean of past incomes)

Model B (Eq. 7)

Model C (Eq. 9)

Model C (Eq. 9)

(1)

(2)

(3)

(4)

(5)

−0.099** (0.030)

−0.103** (0.030)

−0.099** (0.030)

1075.52** 30,815

1088.97** 30,815

1088.20** 30,815

ln(Mean commuting) Ho: β1 = β2 = …= βk = 0 F-value (4, 8505) R-squared Wald statistic No. of observations

0.45 0.10

0.10

8530

8530

Note: Model A is estimated by OLS and models B and C are estimated by GLS with random effects. Standard errors are reported in parentheses and are robust to heteroskedasticity. **, *, and + denote levels of significance respectively at the 1, 5 and 10%.

upon life satisfaction. Being married contributes to happiness, while becoming separated or divorced causes a reduction in life satisfaction. Those who are unemployed or out of the labour force are less satisfied with life compared to those who are employed. Those who have to care for a disabled spouse or relative tend to be less satisfied than those who do not. Finally, it is worth noting that the phenomenon of economic growth with flat or declining happiness does not necessarily mean

that economic growth should not be the goal of economic policy. After all, economic growth is likely to increases the tax yield which is often used to fund public services that may themselves enhance wellbeing in society. However, our results reveal that there is some merit in policymakers giving more attention to the pursuits of other objectives (e.g. promoting health and reducing unemployment) which are likely to provide greater benefits in terms of life satisfaction.

Appendix A Table A Ordered probit estimates of happiness model A (Eq. 4) with alternative lags of income and with and without ‘RICHER’ variable. Variables

ln(Income) ln(Income-1) ln(Income-2) ln(Income-3)

Model A (Eq. 4) (without ‘RICHER’ variable)

Model A (Eq. 4) (with ‘RICHER’ variable)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

0.028** (0.008) 0.010 (0.009) 0.007 (0.008) −0.001 (0.008)

0.012 (0.007) 0.019* (0.007) 0.002 (0.006)

0.009 (0.006) 0.011* (0.005)

0.005 (0.010) 0.001 (0.009) 0.004 (0.009) −0.011 (0.009) 0.004 (0.007)

0.020 (0.009) 0.008 (0.008) 0.005 (0.008) −0.007 (0.007)

0.005 (0.007) 0.017* (0.007) 0.001 (0.006)

0.003 (0.006) 0.009+ (0.005)

1171.4** 16030

1373.6** 23746

1514.7** 51562

ln(Income-4)

All other variables included in estimation. Results not presented to save space Wald statistics Likelihood ratio χ2 1162.9** 1363.0** No. of observations 16030 23746

920.91** 1501.0** 31562

8530

Note: **, * and + denote respectively significant at the 1, 5 and 10%.

References Australian Bureau of Statistics, 2007. Consumer Price Index, Australia. Catalogue No. 6401.0. Becchetti, L., Trovato, G., Bedoya, D.A., 2011. Income, relational goods and happiness. Applied Economics 43, 273–290. Blanchflower, D.G., Oswald, A.J., 2004. Well-being over time in Britain and the USA. Journal of Public Economics 88 (7–8), 1359–1386. Bossert, W., Ambrosio, C., 2007. Dynamic measures of individual deprivation. Social Choice and Welfare 28, 77–88. Brickman, P., Coates, D., Janoff-Bulman, R., 1978. Lottery winners and accident victims: is happiness relative? Journal of Personality and Social Psychology 36 (8), 917–927. Carroll, N., 2007. Unemployment and psychological well-being. The Economic Record 83 (262), 287–302.

Clark, A.E., Oswald, A.J., 1994. Happiness and unemployment. The Economic Journal 104 (424), 648–659. Clark, A.E., Oswald, A.J., 1996. Satisfaction and comparison income. Journal of Public Economics 61 (3), 359–381. Clark, A.E., Frijters, P., Shields, M., 2008. Relative income, happiness and utility: an explanation for the Easterlin paradox and other puzzles. Journal of Economic Literature 46 (1), 95–144. Di Tella, R., MacCulloch, R., Oswald, A.J., 2003. The macroeconomics of happiness. The Review of Economics and Statistics 85 (1), 809–827. Di Tella, R., Haiskens-De New, J., MacCulloch, R., 2007. Happiness adaptation to income and to status in an individual panel. (NBER Working Paper # W13159, June). Dockery, A.M., 2003. Happiness, life satisfaction and the role of work: evidence from two Australian surveys. Curtin University of Technology . (Working paper # 03/10).

910

S. Paul, D. Guilbert / Economic Modelling 30 (2013) 900–910

Duesenberry, J.S., 1949. Income, saving and the theory of consumer behaviour. Harvard University Press, Cambridge, MA. Easterlin, R., 1974. Does economic growth improve the human lot? In: David, Paul A., Reder, Melvin W. (Eds.), Nations and households in economic growth: essays in the honor of Moses Abramovitz. Academic Press, New York. Easterlin, R., 1995. Will raising the incomes of all increase the happiness of all? Journal of Economic Behavior and Organization 27 (1), 35–47. Easterlin, R., Angelescu, L., 2009. Happiness and growth the world over: time series evidence on the happiness-income paradox . (IZA Discussion Paper # 460, March). Ferrer-i-Carbonell, A., 2005. Income and well-being: an empirical analysis of the comparison income effect. Journal of Public Economics 89 (5–6), 997–1019. Ferrer-i-Carbonell, A., Frijters, P., 2004. How important is methodology for the estimates of the determinants of happiness? The Economic Journal 114 (497), 641–659. Gilbert, D., 2006. Stumbling on happiness. Harper Press, London. Headey, B., Wooden, M., 2004. The effects of wealth and income on subjective well-being and ill-being. Economic Record 80, September, Special Issue, pp. S24–S33. Hsiao, C., 1986. Analysis of panel data. Cambridge University Press, New York. Inglehart, R., Klingemann, H.D., 2000. Genes, culture, democracy, and happiness. In: Diener, E., Suh, E.M. (Eds.), Culture and subjective well-being. MIT Press, Cambridge. Jeon, S., Fok, Y.K., 2009. Work and mental well-being: work stress and mental wellbeing of the Australian working population. (Project # 5/08) Melbourne Institute of Applied Economic and Social Research. (June). Jørgensen, C.B., Herby, J., 2004. Do people care about relative income? Centre for Economic and Business Research, University of Copenhagen. (Student Paper 2004-01). Layard, R., 2005. Happiness: lessons from a new science. Penguin Books, London. Luttmer, E., 2005. Neighbours as negatives: relative earnings and wellbeing. Quarterly Journal of Economics 120 (3), 963–1002. McBride, M., 2001. Relative-income effects on subjective well-being in the cross-section. Journal of Economic Behavior and Organization 45 (3), 251–278.

Miles, D., Rossi, M., 2007. Learning about one's relative position and subjective well-being. Applied Economics 39 (13), 1711–1718. Mundlak, Y., 1978. On the pooling of time series and cross section data. Econometrica 46 (1), 69–85. Neumark, D., Postlewaite, A., 1998. Relative income concerns and the rise in married women's employment. Journal of Public Economics 70 (1), 157–183. Paul, S., 1991. An index of relative deprivation. Economics Letters 36, 567–573. Paul, S., 1992. An illfare approach to the measurement of unemployment. Applied Economics 24, 737–743. Paul, S., 2001. A welfare loss measure of unemployment with an empirical illustration. The Manchester School 69 (2), 148–163. Runciman, W., 1966. Relative deprivation and social justice. Routledge and Kegan Paul, London. Schimmack, U., 2006. Internal and external determinants of subjective well-being: review and policy implications. In: Ng, Y.K., Ho, L.S. (Eds.), Happiness and public policy: theory, case studies and implications. Palgrave Macmillan, New York. Senik, C., 2004. When information dominates comparison: learning from Russian subjective panel data”. Journal of Public Economics 88 (8–9), 2099–2123. Solnick, S.J., Hemenway, D., 1998. Is more always better? A survey on positional concerns. Journal of Economic Behavior and Organization 37 (3), 373–383. Stutzer, A., 2004. The role of income aspirations in individual happiness. Journal of Economic Behavior and Organization 54 (1), 89–109. Stutzer, A., Frey, Buno S., 2004. Reported subjective well-being: a challenge for economic theory and economic policy. Schmollers Jahrbuch 124 (2), 191–231. van de Stadt, H., Kapteyn, A., van de Geer, S., 1985. The relativity of utility: evidence from panel data. The Review of Economics and Statistics 67 (2), 179–187. Wilkins, R., 2009. Updates and revisions to estimates of income tax and government benefits. (Hilda Project Discussion Paper Series # 1/09, February) Melbourne Institute of Applied Economic and Social Research, Melbourne University. Yitzhaki, S., 1979. Relative deprivation and the Gini coefficient. Quarterly Journal of Economics 93, 321–324.