What do we know about the determinants of SWB?

The paper cites several examples from the literature purporting to quantify factors that influence SWB. Not all of these examples are well chosen. The examples in their ‘Wellbeing policy priorities’ section are cases in point. Their figure 2 shows a number of factors that ‘explain’ variation in SWB based on the British Household Panel Survey. Several of these factors are plausibly endogenous: happier people may find it easier to find a partner (Stutzer & Frey, Reference Stutzer and Frey2006) or may be more successful in employment (Walsh et al., Reference Walsh, Boehm and Lyubomirsky2018) and they may be less inclined to commit a crime. Diagnosed depression/anxiety explains 46% of the variance, which leads to the question of whether this causally affects life satisfaction or whether it is simply a component of life satisfaction by another name (it certainly is a component of experienced wellbeing).

Their figure 3 suffers from similar problems. Social support is based on self-reports of respondents in a country. Since response styles are likely to differ between countries (Kapteyn et al., Reference Kapteyn, Smith and Vansoest2007, Reference Kapteyn, Smith, Van Soest, Diener, Helliwell and Kahneman2009), it is easy to imagine that in a country where respondents are more likely to respond positively to the question ‘If you were in trouble, do you have friends and relatives you can count on to help you whenever you need them, or not?’, they are also more likely to respond positively to a question about life satisfaction. In addition, the macro-level regressions are likely to suffer from omitted variable bias. It shows, for instance, that income differences across countries explain half of the variation that can be explained by these factors (the factors jointly explain 76% of the total variation in life satisfaction). However, average income in a country is strongly correlated with other characteristics of a country, such as life expectancy, environmental health, government quality and public safety (Helliwell & Huang, Reference Helliwell and Huang2008; Diener et al., Reference Diener, Ng, Harter and Arora2010; Kapteyn et al., Reference Kapteyn, Montgomery, Smith and Soest2017). This omitted variable bias is likely to induce substantial overestimation of the effect of income. In addition, the estimates are based on a static cross-country regression that does not take into account adaptation effects. Were one to take an image like their figure 3 as a basis for policy, then: (1) a number of potentially important policy variables would be ignored; and (2) the effect of income would be overestimated.

Policy, science and common sense

In cases where scientific research into the causes of high life satisfaction agrees with common sense, it is likely easy to implement policies that improve SWB. One might argue even that in such cases scientific research is not very important for policy-making, as the public and policy-makers are likely to agree on the best course of action. The value of scientific research for policy-making increases in cases where research yields results that are not intuitive. At the same time, these are the cases where it will be more difficult to convince policy-makers and the general public of the right course of action. The correlations listed in their ‘Wellbeing scores match our common-sense predictions’ section are good examples. Most policy-makers will agree that full employment and good health care are worthy goals, and the public is likely to agree.

Most will also agree that having a higher income will improve life satisfaction. But that is where the trouble starts. There is a fierce debate in the scientific world about the role of income in determining life satisfaction. Those in the Easterlin camp (to which I probably belong) would argue that the intuition that higher incomes buy happiness is an illusion, or at least is severely biased. At any given moment, people with higher incomes tend to be happier than those with lower incomes. However, if we raise everyone's income, then the relative position does not change and no long-term SWB gain may be had. People do not just compare themselves with others, but also with themselves in the past. From that angle, a higher income does raise satisfaction, at least temporarily. However, if norms adapt, then after a certain period SWB may return back to where it was before (e.g., Clark et al., Reference Clark, Diener, Georgellis and Lucas2008). This notion has been around for a long time and goes under names such as the ‘hedonic treadmill’ (Brickman & Campbell, Reference Brickman and Campbell1971) and ‘preference drift’ (Van Praag, Reference Van Praag1971). The point here is that these mechanisms are not intuitive. One simply observes that having more income is better, but this is likely to ignore the hedonic treadmill.

Consider the following simple example. Let the utility function be:

$$U(y_t,h_t) = y_t-h_t$$

where y _t is log income at time t and h _t is a ‘consumption standard’ in the same period: the higher h _t is, the higher income needs to be to attain a certain utility level. I am assuming that the consumption standard is influenced by both past consumption (habit formation) and the income of others (relativity). The following simple specification captures both of these effects. Define an instantaneous consumption standard:

$$m_\tau \equiv \lambda y_{\tau -1} + (1-\lambda )\overline y _{\tau -1},{\rm} \lambda \in [{\rm 0,1]}$$

where y _τ is log income in period τ and $\overline y _\tau $ is the mean of log incomes in the agent's reference group (however defined; one possibility is that the reference group is the whole population in a country). I next specify h _t as followsFootnote ³:

$${\rm} h_t = (1-a)\sum\limits_{\tau = -\infty} ^t {a^{t-\tau} m_\tau}, {\rm} 0 \le a \le 1.$$

So the effects of past income levels (both own income levels and income levels in the reference group) decay geometrically. We can rewrite this memory specification in the usual partial adjustment form:

$$h_t-h_{t-1} = (1-a)[m_t-h_{t-1}].$$

To illustrate the effects of this specification on the acceptance of certain policies, consider two cases. The first case is where incomes have not changed in a long time. In this static equilibrium, in period 0 we have h ₀ = m ₀ and utility is equal to y ₀ − h ₀. Now assume that both the individual under consideration and everyone in his or her reference group experience a one-time increase in log income by a factor α. It is easy to see that utility in subsequent periods will be equal to:

$$U(y_t,h_t) = y_0-h_0 + a^t\alpha. $$

Thus, life satisfaction is higher than in period 0, but the difference decays over time and ultimately returns to the original equilibrium level.

A second case of interest is stationary income growth at a rate α. In that case, it is straightforward to show that utility equals:

$$U(y_t,h_t) = y_0-h_0 + \displaystyle{{a\alpha} \over {1-a}}.$$

In this case, life satisfaction is structurally higher than in the case without income growth, but it is not the case that life satisfaction increases with increasing income. It remains constant.

The point of this exercise is to show how this may affect public acceptance of policy choices that trade off income growth for other goals (e.g., stronger safety nets, more investments in health care or the environment). I am assuming that citizens are naïve (i.e., they do not account for the change in their consumption standard when considering the effect of a higher future income in comparison to other goals). This trivially biases policy goals towards income growth, since the utility benefit is systematically overestimated.Footnote ⁴

This example took the Easterlin hypothesis for granted. If, on the other hand, life satisfaction is a function of absolute income, as argued by Stevenson and Wolfers (Reference Stevenson and Wolfers2008), Sacks et al. (Reference Sacks, Stevenson and Wolfers2012) and Stevenson and Wolfers (Reference Stevenson and Wolfers2013), then no such bias would arise. But this raises a different problem. If scientific research produces such strikingly different results, how can one then base a policy recommendation on it? It seems paramount to start with findings that are robust and not subject to fierce scientific controversies. I would argue that a place to look for such findings is in the international comparison of policies and outcomes.

Macro-policies and international comparisons

Since SWB appears to vary systematically across countries, a natural (and important) question is to what extent these differences are the result of differences in policies and institutions. Countries with a higher gross domestic product (GDP) per capita tend to exhibit higher average SWB. Yet, such a correlation does not necessarily imply that money buys happiness. Several papers suggest that the positive correlation between GDP per capita and average SWB comes about because a higher GDP ‘buys’ factors that improve life in non-monetary spheres, such as longer and healthier lives. Here are some examples of studies that try to explain differences in average SWB as a result of differences in policies.

Helliwell and Huang (Reference Helliwell and Huang2008) use data from the WVS and the European Values Survey combined with data on indices of government quality (taken from Kaufmann et al., Reference Kaufmann, Kraay and Mastruzzi2009).Footnote ⁵ The ‘quality of delivery’ index is a combination of four indicators: governmental effectiveness, regulatory quality, rule of law and control of corruption. ‘Quality of the democratic process’ combines two indicators: voice and political stability. Quality of delivery appears to be particularly important for SWB in low-income countries, while quality of the democratic process gains in importance when countries get richer. The relationship between average SWB and log GDP per capita loses statistical significance in models with a sufficiently rich set of governance quality indicators. Although their paper focuses on the effect of overall government quality indicators, the authors do consider the effect of government policy on population health (measured by life expectancy and healthy life expectancy) as a channel through which government quality affects average SWB. The effect of health on SWB, in turn, is strong in poorer countries (defined as countries with per capita income of less than half the USA in 1995) and non-significant in rich countries.

Oishi et al. (Reference Oishi, Schimmack and Diener2012) used data for 54 countries represented in the 2007 Gallup World Poll for which information is available about the progressivity of the income tax system. They found that, in countries with a more progressive tax system, residents are on average more satisfied with their lives. Moreover, in countries with more progressive taxation, residents are more satisfied with public goods.Footnote ⁶ In countries with a less progressive tax system, there appears to be a stronger relationship between personal income and life satisfaction. This is consistent with the results from Kapteyn et al. (Reference Kapteyn, Smith and Van Soest2013), who found that satisfaction with income has a much stronger relationship with own income in the USA than in The Netherlands. One interpretation of this is that in a country with a less progressive tax system, marginal tax rates are lower and hence private consumption possibilities increase more sharply with income; an alternative interpretation is that if citizens derive more satisfaction from higher income, they are less willing to tolerate higher marginal tax rates.

Di Tella et al. (Reference DI Tella, Macculloch and Oswald2003) used the Eurobarometer survey series and the US General Social Survey (GSS) to study the effects of macroeconomic changes on life satisfaction. The data represent cross-sections of 13 countries over several years (18 years for the Eurobarometer and 23 years for the GSS). Their regressions included several individual-level variables, including employment status and income (in quartiles), all of which had the expected sign (income had a positive effect and being unemployed had a negative effect on SWB). More strikingly, the national unemployment rate displayed a strong negative relationship with life satisfaction, while the benefit replacement rate was strongly positively related to life satisfaction. The authors interpreted this finding as a possible ‘fear of unemployment’ effect. When the national unemployment rate goes up, one's personal unemployment risk rises. High benefit-replacement rates, however, mitigate this threat. Blanchflower et al. (Reference Blanchflower, Bell, Montagnoli and Moro2013) obtained similar results. Several other papers have also found a positive association between SWB and the existence and generosity of social protection programs (Di Tella et al., Reference DI Tella, Macculloch and Oswald2003; Helliwell & Huang, Reference Helliwell and Huang2008; Pacek & Radcliff, Reference Pacek and Radcliff2008a, Reference Pacek and Radcliff2008b; Boarini et al., Reference Boarini, Comola, Smith, Manchin and DE Keulenaer2012; Boarini et al., Reference Boarini, Comola, DE Keulenaer, Manchin and Smith2013; Easterlin, Reference Easterlin2013; Carr & Chung, Reference Carr and Chung2014; Wulfgramm, Reference Wulfgramm2014; Fonseca et al., Reference Fonseca, Kapteyn, Lee and Zamarro2015). Similarly, in an analysis of life satisfaction across the 50 US states, Alvarez-Diaz et al. (Reference Alvarez-Diaz, Gonzalez and Radcliff2010) found a strong positive effect of the size of transfer payments on life satisfaction in a state.

It appears, therefore, that policies that mitigate the consequences of adverse financial events, such as job loss or catastrophic health events, are likely to be SWB enhancing. There is indeed considerable evidence that this is the case (Morduch & Schneider, Reference Morduch and Schneider2017). For instance, when asked whether it is more important to them to have financial stability or to move up the income ladder, 92% of Americans choose security (Trusts, Reference Trusts2015). Income volatility may negatively affect psychological wellbeing if one is uncertain about whether one will make enough to make ends meet. The psychology literature shows that uncertainty causes anxiety (Carleton et al., Reference Carleton, Norton and Asmundson2007), worry (MacLeod et al., Reference Macleod, Williams and Bekerian1991) and stress (De Berker et al., Reference De Berker, Rutledge, Mathys, Marshall, Cross, Dolan and Bestmann2016). Although this is only meant as an example, it appears that a policy rooted in known psychological mechanisms, and with strong comparative macro-evidence, would be an attractive candidate for policy intervention.