Adaptation, anticipation and social interaction in happiness: An integrated error-correction approach
Introduction
When a person experiences an improvement in her living conditions, she initially feels happier by that. However, what she tends not to foresee, is that after some time she will get used to it, which will push her happiness back towards its original level. This phenomenon of hedonic adaptation has been extensively investigated by psychologists (see Frederick and Loewenstein, 1999, for a survey), and more recently by economists as well (e.g., Clark et al., 2008a). In particular, in the empirical literature on the relationship between individual life satisfaction and income, it has been found that a person's life satisfaction depends not only positively on her current level of income, but also negatively on her level of income in the past (Stutzer, 2005, Weinzierl, 2006, Senik, 2009, Di Tella et al., 2010, Di Tella and MacCulloch, 2010, Layard et al., 2010, Van Landeghem, 2010, D'Ambrosio and Frick, 2012). This implies hedonic adaptation to changes in income. A second central finding on the happiness–income relation is that a person's life satisfaction depends on the level of her income relative to the average income in her social reference group rather than the absolute level of her income (McBride, 2001, Stutzer, 2005, Ferrer-i-Carbonell, 2005, Luttmer, 2005, Weinzierl, 2006, Vendrik and Woltjer, 2007, Senik, 2009, Layard et al., 2010, McBride, 2010, D'Ambrosio and Frick, 2012).1 A third, related phenomenon is that people already become happier when they expect a rise in their income in the coming year (Di Tella et al., 2010, McBride, 2010; cf. De Neve and Oswald, 2012).
However, the empirical survey literature that deals with these phenomena displays important limitations. First, happiness effects of adaptation, social comparison and anticipation with respect to income are not studied simultaneously.2 This is a drawback as past income, social-reference income and future income are correlated. For instance, social comparison income may pick up effects of income expectations if such effects are not controlled for (Senik, 2004, Clark et al., 2008b). As a result, it does not become clear in the existing survey studies what is the relative contribution of each phenomenon to the dynamics of income and happiness. Second, studies that analyze the impact of social-reference income typically assume that this impact is contemporaneous. This is not obvious as the impact may well be delayed (as in the catching-up-with-the-Joneses effect on consumption, see Ljunqvist and Uhlig, 2000). Accordingly, people may not adapt to changes in their social reference income (as assumed in explanations of the Easterlin Paradox; cf. Di Tella et al., 2010, and see below). Third, most estimation equations for adaptation to income or other variables include either only one lag of such a variable (e.g. Weinzierl, 2006, D'Ambrosio and Frick, 2012) or many lags (e.g., Di Tella et al., 2010, Di Tella and MacCulloch, 2010, Layard et al., 2010, Van Landeghem, 2010, Clark et al., 2008a).3 While the former option is less reliable, the latter option leads to a large loss of panel observations, and hence to less significant results in the face of multicollinearity than one may wish. Fourth, the existing adaptation and anticipation models do not control for similar dynamics in other variables.4 When current values of control variables are correlated with current income, by implication lags and leads of these control variables are correlated with lags and leads of income as well and should therefore be controlled for. Fifth, there is no control for the possible impact of lags of life satisfaction. Such lags can be shown to pick up the joint effect on life satisfaction of higher-order lags of independent variables that are not included in the model (see below) as well as the joint effect of lags of omitted variables. Furthermore, a person's current life satisfaction may be directly affected by its level in the past. Finally, possible endogeneity of income due to reverse causality, spurious correlations, measurement error, and unobserved costs of income generation is not taken into account in most survey studies.5
This study adds to the existing literature by providing a more complete analysis in which I simultaneously estimate short and long-run effects on life satisfaction of income and social reference income. At the same time, I control for the short and long-run dynamics of life satisfaction with respect to a large set of control variables. I propose to alleviate the pernicious trade-off between including more lags to make the estimation more reliable and minimizing the loss of panel observations by following the conventional econometric practice of assuming a distributed-lags specification with exponentially declining weights. Applying some transformations to the distributed-lags specification results in an error-correction life satisfaction equation which includes one-year-lagged life satisfaction as an additional explanatory variable and which does not depend on higher-than-first-order lags of the explanatory variables. This equation allows for a neat separation between short-run shock effects of changes in explanatory variables on life satisfaction on the one hand and adjustment of life satisfaction to a long-run equilibrium given by long-run level effects on the other hand. While the former effects indicate how, for instance, a rise in one's income leads to a temporary increase in life satisfaction, the latter adjustment models how this initial happiness boost is gradually faded out to a certain extent by the process of adaptation to a long-run level. According to the psychological set-point theory this long-run level will be the original individual-specific level before the income shock took place (implying complete adaptation). My error-correction model allows for a direct test of this theory by indicating whether in the long run still some happiness effects of initial changes in living conditions remain or whether initial changes may even be reinforced in a longer run. Moreover, such long-run effects are estimated while controlling for adjustment to long-run levels of all other variables. Thus, the error-correction model yields simultaneous estimates of the short and long-run life-satisfaction effects of all variables included in the model.
To allow for deviations from the uniform lag structure for particular independent6 variables I add one and two-years lagged shock terms of all independent variables to the error-correction equation. As my individual-fixed-effects estimate of the lagged-life-satisfaction coefficient suffers from Nickell bias and an additional small-T bias I apply a bias-corrected least squares dummy variables estimator to correct for the Nickell bias (Bruno, 2005) and replace the bias-corrected estimate of the coefficient in my unbalanced panel by the bias-corrected estimate in a corresponding balanced panel with high T (21). Constraining the lagged-life-satisfaction coefficient to be equal to this estimate, the resulting error-correction equation is estimated by ordinary least squares (OLS) as well as by the generalized method of moments (GMM) using instruments for the future, current and past shocks and past level of income. These instruments are based on a predictor of household income like that used by Luechinger (2009) and Luttmer (2005). I use data from the Socio-Economic Panel study (SOEP) for West Germans over the years 1984–2007. As adaptation, anticipation and social-reference effects on life satisfaction seem more important for consumption than for income per se (Frank, 1990, Frank, 2008, Vendrik and Woltjer, 2007), I construct a measure of real household-equivalent income by dividing each household's real income by an appropriate equivalence scale. Social reference income is defined as the average household-equivalent income in individual-specific peer groups of same sex and similar age and education (cf. Ferrer-i-Carbonell, 2005, Vendrik and Woltjer, 2007, Weinzierl, 2006, Layard et al., 2010).
The main results are as follows. First, the GMM estimates for the instrumented income variables reveal a strongly significant,7 positive and large current income effect on life satisfaction and a future income effect that is virtually zero. The long-run income effect is just insignificant and the income adaptation becomes significant in a model in which insignificant lags and leads of the income variables have been dropped. As the long-run income effect is just insignificant, complete adaptation cannot be rejected. Second, social reference income has a significant, negative and strong effect in the long run, but not in the short run. This implies a significant positive relative income effect of the same size8 in the long, but not in the short run, whereas the absolute income effects are significant and positive in the short run, but insignificant in the long run. Third, the bias-corrected coefficient of lagged life satisfaction implies that adaptation to an income shock and reinforcement of a reference income shock take place for more than 90% within three years.
Instrumenting income turns out to be important for my main results as the OLS estimates of the income effects are quite different. These estimates indicate a significant and positive current income effect that is about one third as large as the GMM estimate and a significant, positive and sizable future income effect. The positive long-run income effect is now strongly significant and larger than its GMM counterpart, and there is no statistically measurable adaptation to income. I also investigate the robustness of my results to replacing household-equivalent income and reference income by household income and reference income not adjusted for household size, to changes in the estimation period, and to modifications in the definition of social reference income. Since the OLS results in this study deviate from findings in the literature, I analyze where the differences come from. This comparison strongly suggests that the findings of significant adaptation to household income in previous studies (Di Tella et al., 2010, Di Tella and MacCulloch, 2010, Layard et al., 2010, Van Landeghem, 2010) are severely biased due to the omission of lags of control variables. On the other hand, the finding of significant adaptation to instrumented income in the present study may also be biased to a certain extent. Nevertheless, the insignificance of the long-run absolute income effect turns out to be a robust result. One important implication of this and other results is that they allow for a more comprehensive explanation of the Easterlin Paradox than those given in the happiness literature so far. An implication of the significant long-run effects of social reference income and a number of control variables is a direct rejection of the psychological set-point theory (see above).
The organization of this paper is as follows. Section 2 presents the basic life satisfaction equation, its transformation in an error-correction equation, the extended estimation equation, and the estimation strategy. In Section 3 the data and variables used are described. Section 4 discusses the main estimation results for several variants of the life satisfaction equation and their implications in explaining the Easterlin Paradox for West Germany. In Section 5 the robustness of the estimation results is investigated and Section 6 compares them with the findings in the literature. Finally, Section 7 concludes.
Section snippets
Basic life satisfaction equation
This section presents the basic life satisfaction equation that models adaptation, anticipation and social-reference effects on life satisfaction. This equation specifies life satisfaction in terms of future, current and lagged values of income, social reference income, and the other explanatory variables. Hereby I impose a certain structure on the weights of the lagged variables to facilitate the derivation of tractable estimation equations in the next section. This life satisfaction equation
Data and variables
The database used for all estimations is the German Socio-Economic Panel Study (SOEP). SOEP is a yearly survey that follows about 11,000 households and 20,000 individuals in Germany. I focus on West Germany because of the longer time series (1984–2007) and different background and experiences of people in West as compared to East Germany. I further restrict the sample to persons from 27 to 59 years old as the corresponding age brackets in my social-reference-income measure (see below) then
Static and dynamic equations
This section and the next one present estimation results for variants of Eq. (5) that are rewritten in terms of current, lagged and leaded levels of the variables. I will start with the static variant of Eq. (5) with no lags and leads and will then successively add three lags of income, one lead of income, three lags and one lead of all other independent variables, and lagged life satisfaction to the estimated equation. This will allow me to investigate how the estimates of the income and
Robustness results
This section presents robustness results for the baseline ECM with instrumented household-equivalent income in column (3) of Table 2. First, I investigate the sensitivity of the estimated income and reference income effects to replacing household-equivalent income and reference income by household income and reference income not adjusted for household size. The resulting estimates are presented in column (1) of Table 3. The income coefficients and effects are very similar to those for
Adaptation and anticipation
The finding of (marginally) significant hedonic adaptation to household-equivalent income and household income in the ECMs in columns (3) and (4) of Table 2 and column (1) of Table 3 agrees with similar findings in the literature for SOEP data for (West) Germany (Weinzierl, 2006, Di Tella et al., 2010, Di Tella and MacCulloch, 2010, Layard et al., 2010, Van Landeghem, 2010, D'Ambrosio and Frick, 2012). However, an essential difference is that I only obtained this result in GMM estimations with
Conclusions
To the best of my knowledge, this study is the first in the happiness literature to employ an integrated error-correction approach to model and estimate the dynamics of life satisfaction in a fixed-effects context. I consider this as an improvement because an error-correction model seems the most adequate model to simultaneously estimate short and long-run effects on life satisfaction of a large set of variables. More specifically, it allows for a simultaneous identification of adaptation,
Acknowledgements
I thank Lex Borghans, Dennis de Crombrugghe, Thomas Dohmen, Rob Euwals, Bruno Frey, Marcus Klemm, Andreas Knabe, Bert van Landeghem, Erzo Luttmer, Raymond Montizaan, Arno Riedl, Nicholas Salamancas, two anonymous referees, and participants to the 2008 GSOEP conference in Berlin, the 2008 IIPF conference in Maastricht, the 2009 and 2011 HEIRS conferences in Venice and Milano, the 2010 European Econometric Society conference in Amsterdam, and seminars in Maastricht, Reading and Den Haag (2010)
References (53)
- et al.
On the diminishing returns of higher order terms in asymptotic expansions of bias
Economics Letters
(2003) - et al.
Happiness adaptation to income and to status in an individual panel
Journal of Economic Behavior and Organization
(2010) - et al.
Relative versus absolute income, joy of riding, and gender: brain imaging evidence
Journal of Public Economics
(2011) Does economic growth improve the human lot? Some empirical evidence
Will increasing the incomes of all increase the happiness of all?
Journal of Economic Behavior and Organization
(1995)Income and well-being: an empirical analysis of the comparison income effect
Journal of Public Economics
(2005)Should public policy respond to positional externalities?
Journal of Public Economics
(2008)On bias, inconsistency, and efficiency of various estimators in dynamic panel data models
Journal of Econometrics
(1995)Relative-income effects on subjective well-being in the cross-section
Journal of Economic Behavior and Organization
(2001)Money, happiness, and aspirations: an experimental study
Journal of Economic Behavior and Organization
(2010)
Does happiness adapt? A longitudinal study of disability with implications for economists and judges
Journal of Public Economics
When information dominates comparison: learning from Russian subjective panel data
Journal of Public Economics
Direct evidence on income comparisons and their welfare effects
Journal of Economic Behavior and Organisation
Happiness and loss aversion: is utility concave or convex in relative income?
Journal of Public Economics
Positional concerns through the life cycle: evidence from subjective well-being data and survey experiments.
Enhanced routines for instrumental variables/generalized method of moments estimation and testing
The Stata Journal
Happiness and the persistence of income shocks
Bias-corrected estimation in dynamic panel data models
Journal of Business & Economic Statistics
Approximating the bias of the LSDV estimator for dynamic unbalanced panel data models
Economics Letters
Lags and leads in life satisfaction: a test of the baseline hypothesis
The Economic Journal
Relative income, happiness and utility: an explanation for the Easterlin Paradox and other puzzles
Journal of Economic Literature
Individual wellbeing in a dynamic perspective
Economica
Estimating the influence of life satisfaction and positive affect on later income using sibling fixed effects
Proceedings of the National Academy of Sciences
Some uses of happiness data in economics
Journal of Economic Perspectives
Happiness adaptation to income beyond basic needs
Building a better theory of well-being
Cited by (41)
All we want is a healthy baby – well, and one that is the opposite sex to what we have already<sup>1</sup>
2023, Journal of Behavioral and Experimental EconomicsSharing energy poverty: The nexus between social interaction-oriented gift expenditure and energy poverty in rural China
2023, Energy Research and Social ScienceUsing memories to assess the intrapersonal comparability of wellbeing reports
2022, Journal of Economic Behavior and OrganizationPeople do not adapt. New analyses of the dynamic effects of own and reference income on life satisfaction: People do not adapt
2020, Journal of Economic Behavior and OrganizationCitation Excerpt :Second, we observe less adaptation to own income when including lagged terms of controls (Ferrer-i Carbonell and Van Praag, 2008; Vendrik, 2013). Third, with respect to adaptation to reference income and from today’s viewpoint, the only (to my knowledge) viable study is the one of Vendrik (2013). Fourth, there seem to be almost no studies doing cross-country comparative work.
Global terror, well-being and political attitudes
2020, European Economic Review