Are Cognitive, Affective, and Eudaimonic Dimensions of Subjective Well-Being Differently Related to Consumption? Evidence from Japan

The paradox between income growth and a corresponding rise in reported subjective well-being (SWB) was identified nearly half a century ago in the seminal work of Easterlin (1974). Since then, there has been much debate about the income–SWB relationship. In his more recent evaluation of the topic, Easterlin (2001) concludes that people with higher incomes, on average, are happier than those with lower incomes; although there is a point of diminishing returns, he suggests that as income rises, people increase their material aspirations, which in turn undermines the potential gains of SWB.

As Clark et al.’s (2008) literature review suggests, the general consensus of previous studies is in line with Easterlin’s conclusion that income indeed raises happiness, with diminishing returns to income. Some of those studies include analyses using cross-sectional survey data from one country with or without standard demographic controls (e.g. Graham and Pettinato 2002; Blanchflower and Oswald 2007). Other studies use panel data from one country to control for unobserved individual fixed effects (e.g., Winkelmann and Winkelmann 1998; Ferrer-i-Carbonell 2005; Gardner and Oswald 2007). Yet other studies employ macroeconomic variables, including GDP, to control for country fixed effects (e.g. Di Tella et al. 2003; Helliwell 2003; Leigh and Wolfers 2006).

However, the findings in the literature seem to be highly dependent on what a given study actually measures for SWB (see, e.g. Deaton 2008; Stevenson and Wolfers 2008; Diener et al. 2010; Kahneman and Deaton 2010). According to the Organisation for Economic Co-operation and Development (OECD 2013), there are three main types of indexes that measure SWB: those related to “life evaluation,” in which an individual evaluates his or her own life experiences or life as a whole (e.g., life satisfaction and the Cantril ladder); those related to “emotional well-being” (affect balance), which measure an individual’s feelings or psychological state at or during a certain point or period in time; and those related to the concept of “eudaimonia,” which capture an individual’s goal orientation or sense of life purpose. It is important to note that there is some debate about whether these two types of life evaluation indexes—life satisfaction and the Cantril ladder—represent the same thing (Helliwell et al. 2012). For example, the Cantril ladder produces a wider distribution of responses (Office for National Statistics 2011) and is more strongly correlated with income than other life evaluation indexes. Diener et al. (2010) show that income seems to be positively correlated with cognitive measures of SWB but not with affective measures. Similarly, Kahneman and Deaton (2010) demonstrate that in the United States, there might be an income threshold for emotional well-being of around US$75,000, after which emotional well-being does not increase, while there might not be such an income threshold for life evaluation. Jebb et al. (2018) statistically identify points of income satiation for other countries. The authors apply spline regression models to global data from the Gallup World Poll and, controlling for demographic factors, find that satiation occurs at US$95,000 for life evaluation and US$60,000 to US$75,000 for emotional well-being. In addition, they find substantial variation across world regions, with satiation occurring later in wealthier regions; for some countries, incomes beyond satiation are associated with lower life evaluations. Thus, there is clear evidence that the income–SWB relationship depends on the SWB measure, but the reasons are not yet fully clear. Given that most income is used for consumption, a main reason may be related to consumption. The current study is the first to examine directly whether the consumption–SWB relationship differs between affective, cognitive, and eudaimonic SWB measures.

The broader concept of SWB has drawn much attention (Benjamin et al. 2012); however, as mentioned, our interest lies in the difference between affective, cognitive, and eudaimonic SWB measures. In this subsection, to contextualize the affective, cognitive, and eudaimonic SWB measures adopted in our study, we review the big picture of the evolving massive body of literature, chiefly drawing on a book-length treatment by Sirgy (2012).

Many researchers distinguish between cognitive and affective aspects of SWB (Campbell 1976; McKennell 1978; Andrews and McKennell 1980; McKennell and Andrews 1980; Organ and Near 1985; Brief and Roberson 1989; Crooker and Near 1995; Kahneman 2011). How do affective and cognitive measures of SWB relate to each other? Most studies regard life satisfaction as self-evaluation of happiness (Diener 1984; Veenhoven 1984a, 1984b; Veenhoven and Coworkers 1994). In other words, life satisfaction is viewed as a “cognitive” conceptualization of happiness or SWB and may involve judgments of fulfillment of an individual’s needs, goals, and wishes (Cantril 1965; Andrews and Withey 1976; Campbell 1976; Michalos 1980; Diener 1984). Diener et al. (1985) define life satisfaction as a cognitive judgmental process dependent upon a comparison of one’s circumstances with what is thought to be an appropriate standard. One of the most well-known cognitive measures is the Cantril ladder (Cantril 1965), comprised of steps 0 at the bottom to 10 at the top, in which respondents are asked which step of the ladder represents where each feels that he or she stands at the present time. Meanwhile, positive and negative emotions are referred to as affective measures (Ahuvia and Friedman 1998).

However, that is not the end of the story, as a large body of the literature emphasizes the relative nature of SWB. In the cognitive evaluation of happiness, an individual must surely assess his or her situation relative to some benchmark or reference, such as his or her own life in the past, the lives of surrounding people, a global standard, or a shifting “best possible life for you” standard that moves upward with rising living standards (Deaton, 2008).¹ In addition, Ahuvia and Friedman (1998) suggest that SWB includes both affective and cognitive components and that the cognitive and affective elements of SWB tend to be correlated. They state that “the cognitive judgement that one’s life is a failure is likely to result in negative affect and the reverse for more flattering cognitions.”

Another question arises: can eudaimonic well-being be treated equally with affective and cognitive measures of SWB? While Kesebir and Diener (2009) consider that eudaimonic well-being and other SWB measures are sufficiently close, Vitterso et al. (2010) argue that they are theoretically distinct, as the cybernetic principles underlying them are different. Goal attainment in hedonic well-being reflects homeostatic balance, which reflects a state of happiness. By contrast, lack of goal attainment reflects a state of disequilibrium that induces feelings of interest, curiosity, challenge, and task absorption. The latter may be reflective of eudaimonic well-being.

This points to the need to treat eudaimonic well-being separately from the dichotomy between affective and cognitive measures. Peterson et al. (2005) develop a measure based on Seligman’s theory of authentic happiness, which seems to capture the three major dimensions of happiness: hedonic well-being, life satisfaction, and eudaimonia (Sirgy 2012).²

Thus, in the current study, we adopt the following four SWB indexes: life satisfaction (cognitive), the Cantril ladder (cognitive), affect balance (affective), and eudaimonia.³ Affect balance can be classified with hedonic happiness (well-being) and affective measures, and short-term well-being (Kozma et al. 1992; Parducci 1995; Sumner 1996; Kahneman 1999). By contrast, life satisfaction, the Cantril ladder, and eudaimonia can be classified with eudaimonic well-being (happiness), cognitive measures (Phillips 2006; Kesebir and Diener 2009; Vitterso et al. 2010), and long-term well-being.⁴

In studies on SWB, in the tradition of the Easterlin paradox, income is often utilized as the yardstick (Kahneman and Deaton 2010). At a macro level, national income and GDP are frequently discussed with the implicit assumption that they are relevant to welfare. However, microeconomic theory starts with the assumption that individuals maximize utility derived from consumption, not from income. Income and consumption have equivalent impact on welfare only in a static, one-good setting in which individuals do not save at all. Moreover, recent studies stress other contributors to individual well-being, such as social capital, leisure, and inequality (e.g., Stiglitz et al. 2010; Benjamin et al. 2014; Tsurumi and Managi 2017).

This point rationalizes our focus in the current study: we consider the association of SWB with consumption instead of income. The literature on the total consumption–SWB relationship is huge, including studies by Guillen-Royo (2008), Headey et al. (2008), Zimmermann (2014), and Noll and Weick (2015). Headey et al. (2008) show somewhat contradictory results: the total consumption–life satisfaction relationship is positive for the United Kingdom but negative for Hungary. It is possible that people may become dissatisfied and insecure once their consumption level rises for a given level of income and wealth. There are other potential channels that make this relationship seem inconsistent. For example, Guillen-Royo (2008) suggests that in the Peruvian corridor, consumption has a meaning beyond mere satisfaction of basic needs, such as status concerns regarding the reference group, the pleasure of consuming, providing for household basics, and the expectation of escaping social marginalization. Likewise, Zimmermann (2014) shows total consumption in the United States is significantly associated with SWB, but only via certain consumption goods, such as experiential and conspicuous expenses.

Moreover, we treat both total consumption and different subtypes of consumption, as the latter may yield different marginal utilities. One category worth studying is relational consumption, through which people feel connected to others and which often entails both leisure and social capital investment.⁵ Examples include consumption involving interactions with family members, friends, and acquaintances, such as sending a gift, having dinner, or traveling in social groups. DeLeire and Kalil (2010), Noll and Weick (2015), and Zhang and Xiong (2015) study the effect of relational consumption of leisure on SWB. However, relational consumption has a deeper meaning than that of a mere leisure activity. Like economic conditions, social capital exerts a significantly positive influence on SWB relative to other factors (Frey and Stutzer 2002; Bjørnskov 2003; Di Tella et al. 2003; Becchetti et al. 2008; Bruni and Stanca 2008). Relatedly, Kasser and Sheldon (2002) examine consumption practices during Christmas and consumer well-being. They show that consumers report higher levels of happiness when they consider family and religious experiences as important but lower levels of happiness when they consider spending money and buying gifts as important. In addition, by utilizing social capital indicators that measure trust in others and citizen participation, Bjørnskov (2003) shows that social capital affects happiness to a greater degree than income, at least in developed countries. Although our concept of relational consumption in this study does not necessarily entail social capital investment, we can conjecture that relational consumption may exert a correspondingly powerful influence on SWB.⁶

A related but separate class of consumption worth clarifying here is experiential consumption, such as vacations and concert tickets, as opposed to material consumption or the possession of goods (Van Boven and Gilovich 2003; Dunn and Weidman 2015; Gilovich et al. 2015). We focus on consumption involving interactions with other people and not experiential consumption per se. Relational and experiential consumption share common traits: dining out with family, for example, is an activity that falls under both categories. Eating out alone, by contrast, is an example of experiential consumption but not relational consumption.⁷

Based on literature review, Dolan et al. (2008) suggest that the effect of material well-being on SWB is not linear but rather diminishes with higher levels of material well-being; this is similar to the economic concept of diminishing marginal utility. Diminishing marginal effects are also revealed in empirical studies. For example, Gokdemir (2015) reports that, in Turkey, only the consumption of durable goods is correlated with life satisfaction. DeLeire and Kalil (2010) analyze US data for nine consumption categories and show that, of all the categories measured, only one, leisure, is positively correlated with SWB. Zhang and Xiong (2015) employ 77 consumption categories and 13 SWB indicators to examine the relationship between consumption and SWB in Japan. A notable finding of their study is the particularly strong correlation between relational consumption and SWB. Dumludag (2015) analyzes countries at different stages of development in Europe and shows that the relationship between life satisfaction and each consumption category varies in accordance with the development stage. Relatedly, Pandelaere (2016) reviews studies on experiential versus material purchases and suggests that even though many studies find an advantage for experiential purchases, this effect does not occur for materialists. He states that materialists do not benefit more from material than from experiential consumption owing to unrealistic expectations.

The literature outlined in this section has not clarified how the choice of SWB indexes affects the consumption–SWB relationship, which is not sufficiently discussed in the extant literature. In addition, on the consumption side, we consider both material and relational consumption.

The rest of the paper is organized as follows. Section 2 describes the theoretical model; Sect. 3 presents the empirical models and data; Sect. 4 describes the estimation results; and Sect. 5 discusses the findings, presents the conclusion, and indicates opportunities for further research.

As we attempt to examine the consumption–SWB relationship, we are essentially interested in the partial derivatives of instantaneous or intertemporal utility. For this empirical purpose, a reduced form of \(U\) or \(V\) from the prior section would be as follows:where i represents the individual and \(SWB_{i}^{k}\) represents the SWB indicator of type k, with k being any of the following four types: the Cantril Ladder (cognitive SWB); life satisfaction (cognitive SWB); affect balance (affective SWB); and eudaimonia. \({\gamma }_{1}\) is a constant. The main independent variable, \({consumption}_{i}\), represents the monthly total consumption of individual i. \(X_{ij}\) are control variables that have been used in prior studies of SWB (i.e., age, gender, level of subjective health, education, marriage, having children, having grandchildren, and personality).⁹,¹⁰ Furthermore, since consumption expenditure is reported per household, we also control for the number of people in the household.¹¹ Finally, \(\alpha_{j}\) are coefficients and \({\epsilon }_{1}\) is the uncorrelated error term.

Because we are concerned about the different effects of material and relational consumption, our empirical model is extended as follows:where material consumption (\({C}_{1}\)) and relational consumption (\({C}_{2}\)) are included on the right-hand side. Like in Eq. (1), \({\gamma }_{1}\) is a constant, \({\epsilon }_{2}\) is an uncorrelated error term, and \({Y}_{ij}\) stands for other control variables with \({\beta }_{j}\) as the coefficient. To ensure that residual consumption does not affect our results, we also control for other consumption \(\left( {consumption_{i} - \left( {C_{1i} + C_{2i} } \right)} \right)\).

Using a reduced form of our theoretical construct, this study aims to clearly explore the different functional forms of consumption. Thus, we utilize nonparametric functions for consumption. Regarding the other explanatory variables, we account for parametric functions and apply semiparametric regression, as in prior studies. Previous studies of SWB using nonparametric methods include Blanchflower and Oswald (2008), Ferreira and Moro (2010), and Stevenson and Wolfers (2013).

There are two obstacles to the usual nonparametric multiple regressions. First, as the number of explanatory variables increases, the sparseness of the data inflates the variance in the estimates, and massive amounts of data are required for accurate estimation. Second, because a nonparametric regression does not provide an equation relating the average response to the independent variables, we need to display the response surface graphically by slicing the surface. When the number of explanatory variables exceeds three, the shape of the sliced surface becomes difficult to interpret (see Hastie and Tibshirani 1990).

These problems have spurred the development of additive regression models (Stone 1985). In this study, we use generalized additive models. We use cubic spline smoothing¹² iteratively to minimize the partial residuals that remain after removing the influence of the other variables in the model. In this model, a Bayesian approach is employed to derive standard errors and confidence intervals. Our estimation technique follows Wood (2004, 2008).

This study relies on data from 9,635 subjects who submitted valid responses from a total of 13,690 respondents to our original Internet survey conducted in March 2017 across Japan.¹³ The main dependent variables for SWB are summarized in Table 1, while Table 2 shows the control variables.

The survey selected respondents based on prefectural demographics, such as population, gender, and age ratios, to ensure that the sample represented Japan’s population. The survey was conducted as follows. First, an invitation was emailed from the internet survey company to relevant cohorts from their 1 million-plus registered monitors in Japan. To obtain a random sample, e-mails were sent to selected monitors based on prefectural demographics, such as population, gender, and age, and basic attributes, such as income level, and education. About 67% of the individuals who were contacted completed the survey.

Consumption is of utmost interest among the independent variables. In particular, we ask the questions listed in Table 3 to differentiate material consumption from relational consumption. Material consumption represents the consumption of goods, both consumer durables and others, but not services. Our measure of relational consumption pertains to consumption involving interactions with family, relatives, friends, and acquaintances. While monthly household consumption expenditure includes expenditures for housing, cars, and bikes, both material consumption and relational consumption do not. In addition, monthly household consumption expenditure includes medical expenses, insurance, and educational expenses, which are not included in either material or relational consumption.

In Table 4, we show the descriptive statistics of both the dependent and explanatory variables, while the frequencies of each type of consumption expenditure are shown in Fig. 1. Among SWB indexes, we find that affect balance and the Cantril ladder have the smallest standard deviations and that subjective happiness has the highest mean.