Chapter 7. Understanding subjective well-being

National Statistical Office data collection

The availability of survey data on subjective well-being, including panel data, has increased at a rapid pace. National Statistical Offices (NSOs) are increasingly including subjective well-being questions in their surveys, and a majority of OECD countries now collect at least some subjective well-being data (Table 7.1). For example, NSOs in all but one OECD country have collected life evaluation data in recent years, and more than three-quarters of NSOs have collected some data on aspects of eudaimonia and experiential well-being.1 This represents very significant progress since 2009. The OECD Guidelines on Measuring Subjective Well-Being (OECD, 2013), which provide clear directions and proposed modules for including subjective well-being questions in surveys, have galvanised this movement. Nevertheless, in some cases different measurement approaches continue to be adopted, particularly with regard to eudaimonia and experiential well-being, where broad consensus on best practice is still lacking. To ensure greater comparability and take-up of the data, further work is needed to co-ordinate and harmonise measurement efforts across countries, and to increase the frequency with which data are collected (see Exton, Siegerink and Smith (2018), for a review of recent progress).

In 2013, the EU Survey on Income and Living Conditions (EU-SILC, Eurostat, 2013) included an ad hoc module on subjective well-being, which included a question for each of its three main elements. This has produced comparable subjective well-being data for all 28 European Member States, as well as Iceland, Norway, Switzerland and Turkey. In 2015, Eurostat (the statistical office of the European Union) also launched a publication on Quality of Life - Facts and Views (Eurostat, 2015), with explanatory pages and an interactive tool to make the data more easily accessible to a wide variety of users. This was complemented by a Eurostat analytical report on subjective well-being published in 2016. All quality of life indicators, including subjective well-being, have been evaluated by the Eurostat Expert Group.

Based on these experiences, Eurostat is now considering adding a question on life satisfaction in the core part of the EU-SILC questionnaire on a yearly basis in the near future, while every 6 years an ad hoc module with around 20 variables on the topic will supplement this information. This will provide a remarkable resource to the research and policy community: a harmonised cross-country dataset with a sufficiently large sample size to estimate the relationship of subjective well-being to a host of individual and geographic characteristics over time. Eurostat’s plan for annual measurement is an important step that will help to establish a time series for more than 30 countries.

There have also been advances in including subjective well-being in time-use surveys. This is particularly important to improve our understanding of experiential well-being, since it enables the link between people’s activities, daily circumstances, social contact, and feelings to be examined. Some countries (the United States, France, Poland, Luxembourg, the United Kingdom and Canada) have incorporated some version of experiential well-being into their time-use surveys. For example, in the United States a governmental agency, the Bureau of Labor Statistics, regularly conducts a time-use survey of over 12 000 individuals (the American Time Use Survey, ATUS). In 2010, 2012 and 2013, a well-being module was included in ATUS that sampled three time-use episodes for each person and asked a number of experiential well-being questions. Unfortunately, the module was limited to three rounds even though a strong recommendation to continue collection of these experiential well-being data was issued by a National Academy of Sciences panel (Stone and Mackie, 2015).

The investment of NSOs in measuring subjective well-being is extremely important and should continue (Box 7.3). As with other indicators such as GDP, subjective well-being data is at its most valuable when one can observe and compare trends, meaning that long periods of data are needed. Continued methodological progress would be facilitated by the collection and dissemination of long time-series in large, high-quality datasets. Collection of such data will also facilitate the generation of policy-relevant insights. Researchers can help support this process by ensuring that analyses of the data that do exist are carried out and disseminated, demonstrating their usefulness; conversely, NSOs can help by ensuring that micro-data are available in a timely manner. Nevertheless, the methodological and conceptual issues raised later in this document must be taken into account when interpreting the data collected. To take one example, Deaton (2012) and Deaton and Stone (2016) suggest that tracking of subjective well-being data can be disrupted by a change in the questionnaire design or by the provision of inappropriate cues, and that such factors do not always average out.

Global reports and tools

Several global analyses of social and economic progress beyond (or in addition to) GDP have been published and widely disseminated since 2009, including the Legatum Prosperity Index (O’Donnell et al., 2014), the UN World Happiness Report2, released annually since 2012 by UNSDSN (Helliwell, Layard and Sachs, 2018), the US National Academy of Science Report on Measuring Subjective Well-Being (Stone and Mackie, 2015), the OECD How’s Life? series and Better Life Index (OECD, 2015a). While all of these projects include sections on subjective well-being, each takes a different approach to the analysis and comparison of well-being across countries. This diversity of approaches and initiatives helps to advance our understanding of subjective well-being and how it can be used.

Two particularly important documents – the OECD Guidelines on Measuring Subjective Well-Being (OECD, 2013) and the US National Academy of Science Report on Measuring Subjective Well-Being – lay out the current experience in collecting data on subjective well-being and provide a focal point for a growing consensus around methodology. We view these documents as “required reading” for policy-makers and researchers working with subjective well-being measures, because they carefully consider many of the various conceptual and methodological issues that are only briefly touched upon below. Furthermore, supported by the US National Institute on Aging, a conference on time-use and experiential well-being was held in 2015 to take stock of progress since the publication of the Day Reconstruction Method (DRM) and to identify remaining challenges. A report of the conference proceedings (Stone and Smith, 2015) outlines many issues and questions that remain about the DRM, despite its use in dozens of research studies. Two other documents – the Legatum report and the World Happiness Report – speak more to the policy uses of subjective well-being measures and are discussed in a later section of this chapter.

Improvements in methodology

The Stiglitz, Sen and Fitoussi (2009) report identified some of the methodological and interpretive issues that caused concern about using subjective well-being measures. Since the publication of the report, solutions have been presented and explored for many of those issues. A short summary of the issues is provided in Table 7.1 alongside the most promising solutions. While a deep examination of these issues is important to improving the measurement of subjective well-being, it is equally important to avoid setting a uniquely high-standard for subjective well-being in contrast to other indicators, such as income, consumption or wealth inequality, which, as shown in the other chapters in this volume, can be quite difficult to calculate or are similarly derived from self-reported measures that are equally sensitive to the survey vehicle used (for example, the length of the recall period used for expenditure diaries can have dramatic effects on consumption estimates, Beegle et al., 2011) or may have other issues related to self-reports more generally.

Some of the methodological issues detailed in Table 7.2 can be partially addressed by careful standardisation of questionnaires, which may reduce framing and potential context effects.3 For this reason, the continued collection of standardised questions across countries is needed. Following the OECD Guidelines is a good way to ensure that questions are standardised, as they represent the state of the art for question formulation and survey administration. Eurostat’s 2013 EU-SILC ad hoc module on subjective well-being followed the OECD Guidelines, and was based on its recommended questionnaire. The EU-SILC and its ad hoc modules have a legal basis, with a common list of variables, concepts, classifications and survey requirements translated in all EU languages. The legislation is accompanied by EU-SILC methodological guidelines, including the recommended questionnaire, translated in all languages. Importantly, the legislation requires that all EU countries contribute data to this effort.

To illustrate the importance of the concerns over systematically different response styles and bias, we provide a more detailed discussion here. If different groups of people show systematic differences in how they interpret subjective well-being questions or use response scales (for example, due to some common characteristic such as language or culture), then simply comparing the level of subjective well-being between these groups can yield misleading conclusions. The extent to which this is a problem will depend, in part, on the question the data are being used to answer. In some cases, this will not matter if the main focus of interest is whether the change in a variable produced a change in subjective well-being within a specific population, rather than direct comparisons between groups of individuals.

However, in simple descriptive analyses where levels of subjective well-being are compared across groups, such as gender or occupations, or across countries, then systematic differences in question interpretation or response styles between groups has the potential to cause bias. For example, if the elderly understand or respond to a subjective well-being question in a way that is systematically different to younger people, or if richer people have a response style that is somehow different to that found among the poor, then we might over- or under-estimate differences in subjective well-being between these groups. In order to have more concrete ideas about the extent to which this may be a problem, we should have a better idea of why such differences might exist in the first place, and have some theoretical justification for a concern with systematic differences in how subjective well-being questions are interpreted and answered.

Concerns about systematic biases, in particular their potential interaction with context effects, are not solved through the use of panel or longitudinal data if the goal is to compare levels of well-being between groups at a single point in time. A salient example of this is provided by Deaton (2012) on context effects, with its implications further refined in Deaton and Stone (2016) using subjective well-being data collected by the Gallup Organization. In this work, Deaton found evidence of a large impact of a set of political questions placed prior to an evaluative well-being question (the Cantril Ladder), an effect that was larger than that of the 2008 recession. This effect was driven by respondents who reported feeling that the country was going in the wrong direction, which exerted a strong downward bias on their answers to the subsequent life evaluation question. Importantly, these context effects varied by race or ethnicity. For example, the negative treatment effect of the political polling questions (compared to a control without such questions) was smaller for blacks than for whites. This meant that, while whites in the control group on average reported life evaluations almost 0.2 scale points higher than blacks (on a 0-10 scale), in the treatment group there was almost no difference between the two groups (less than 0.03 scale points). This contrasts with results for gender, age and income, where the size of the context effects remained stable across different population groups.

The finding that context effects can work differently for different populations complicates the interpretation of group differences in subjective well-being and requires more extensive study. Since NSOs are very unlikely to ask political polling questions, it will be important to understand whether other lead-in questions can also produce a significant shift in responses. For example, Lee et al. (2016) found that asking a self-rated health question immediately before a subjective well-being question prompted a stronger correlation between the answers compared to a situation when the question ordering was reversed. This effect was driven by a subsample of respondents who reported one or more chronic health conditions: among those without chronic health conditions, question ordering did not produce a significant difference in the size of the correlation.

Taken together, these studies reinforce the importance of question ordering for both survey design and data comparability. Unless explicitly tested through split-sample methods, the effects of question ordering will tend to remain hidden from view.

While systematic group differences in response styles are insufficiently addressed in the literature, several advances have been made. These include the use of vignettes (Crane et al., 2016; although see Grol-Prokopcsyk et al., 2015, and OECD, 2013, for concerns about these methods) when analysing data from migrants (Senik, 2014; Exton, Smith and Vandendriessche, 2015), and using individual fixed effects models with panel data. However, in general we view these concerns as unresolved and recommend continuing research.

Another continuing area of methodological development pertinent for experiential subjective well-being research is the use of real-time and near real-time data capture, for example, with momentary data recording such as Ecological Momentary Assessment (EMA), which is based on the administration of brief questionnaires in real-time in peoples’ everyday lives (Stone and Shiffman, 1994; Shiffman, Stone and Hufford, 2008), daily diaries, and day reconstruction methods (Kahneman et al., 2004). These are important techniques because of their potential to assess experiential subjective well-being with less retrospective bias than measures using (relatively) long recall periods that ask about fluctuating levels of emotions and pain. Long recall periods conflate actual memories of experiences with broad beliefs that do not necessarily accurately reflect experience. However, from a pragmatic perspective of data collection in national surveys, momentary methods can be unwieldy, burdensome, expensive and impractical in some cases (e.g. while people are driving, or when they are engaged in activities that cannot be interrupted) yielding selection effects. As result, methods that ask about the prior day have become the standard; these include simple overall questions about yesterday (as used in the UK ONS survey and the Gallup World Poll), so-called “hybrid” measures based on the DRM (which capture some details about the day, for example, the number of hours engaged in various activities (Christodoulou, Schneider and Stone, 2014; Miret et al., 2012), and DRM surveys (addressing the entire day or sections of the day, as done by the Survey on Health, Ageing and Retirement in Europe (SHARE).

Box 7.4. Aging and subjective well-being

An intriguing line of research pursued by psychologists, sociologists and economists is whether and how subjective well-being shifts with age. There are now dozens of articles examining this question. In short, the pattern is that evaluative well-being (life satisfaction, the Cantril Ladder) is relatively high in the 20s, falls to its lowest point in the late 40s and early 50s, and then improves to the highest levels in the 70s (although there may be declines in older age, see Figure 7.1). This pattern holds for English-speaking, wealthy countries, but not for poorer, non-Western countries (Steptoe, Deaton and Stone, 2015). There is also evidence that the pattern is not attributable to cohort effects, which could have explained the pattern by different cohorts of individuals experiencing various historical events (Blanchflower and Oswald, 2008). Less is known about how experiential subjective well-being changes with people’s age, but at least in the United States patterns of different affects are not U-shaped. For example, Stone et al. (2010) found that stress is high from age 20 through to about age 50, followed by a rapid decline through the 70s (the right side of this pattern is consistent with the evaluative well-being pattern of improving outcomes in older age). What is surprising about these patterns, at least for Western countries, is that the improvement in subjective well-being occurs at an age when the prevalence of chronic disease is on the increase, and that the presence of illness is associated with lower subjective well-being. Theoretical explanations, with some empirical support, focus on a shift in priorities and in social engagements and time use in older people, resulting in higher well-being.

Figure 7.1. Life evaluations and people’s age across the world

Cantril ladder

Note: Mean life evaluations, plotted by 4-year age groups. The continuous line represents unadjusted data, while the dotted line represents data adjusted for 4 covariates (the share of the population who are women living with a partner, with a child at home, and unemployed).

Source: Stone, A.A. et al. (2010), “A snapshot of the age distribution of psychological well-being in the United States”, Proceedings of the National Academy of Sciences of the United States of America, Vol. 107(22), pp. 9985-9990.

 StatLink https://doi.org/10.1787/888933839791

Complementary approaches to survey questions have been, or are being, developed. Several approaches exist, or are being developed, that can complement survey questions. For example, big data is being used to “nowcast” subjective well-being (Box 7.5), and new approaches are being developed to elicit people’s preferences, including information on how they value trade-offs between competing goals. “Automatic data capture”, which combines data such as GPS measurement or continuous health measures with survey questions on subjective well-being, is another example of these innovative approaches.

The global picture

Setting aside the measurement issues outlined above, which may impact the substance of what follows, Gallup World Poll data show that evaluative subjective well-being is highest, on average, in the Nordic countries, Switzerland, the Netherlands, Canada, New Zealand and Australia; and lowest in the poorer countries of sub-Saharan Africa, and in countries experiencing war, such as Syria and Afghanistan (Box 7.6). Changes in average reported life evaluation across countries since the SSF 2009 report also provide evidence that subjective well-being indicators are useful measures of social progress. The biggest drop in average life evaluations between 2005-07 and 2012-14 was in Greece, followed by Egypt and Italy (Helliwell, Layard and Sachs, 2015) – although this analysis excludes five of the world’s current 10 lowest ranking countries, including Syria and Afghanistan, due to lack of data in earlier waves of the Gallup World Poll.

Box 7.6. Subjective well-being across the world

New evidence from the Gallup World Poll shows that the positive relationship between evaluative well-being and income runs throughout the range of GDP per capita, from the poorest to the richest countries. This pattern was not easily seen from the World Values Survey, which does not include the really poor countries in Africa, and had to wait until the Gallup World Poll came along to become firmly established. Prior to that, many researchers had observed that, when plotted against per capita GDP, countries’ evaluative well-being flattened out at some point so that, at income levels above those of a country like Morocco, higher average income did not lead to higher evaluative well-being. This led many researchers in the field to conclude that income did not matter once countries were no longer poor. In his early papers, Easterlin used this as evidence that income did not improve the human lot, at least once we were not dealing with real poverty anymore (Easterlin, 1974).

Deaton (2008) using the Gallup World Poll showed that this conclusion was wrong. When life evaluation is plotted against the logarithm of income, the result is very close to a straight line. There is certainly diminishing marginal utility with respect to GDP per capita, but doubling income has the same effect at the bottom as at the top of the income scale, though the absolute changes in income are much smaller at the bottom than at the top (Figure 7.2). Of course, there is a lot of scatter, so that if you do not take the whole range, the relationship is much less obvious. For example using plots limited to rich countries, an observer might conclude that there is not much of a relation there.

The variability in the association is also important, because it shows that countries are not trapped by their level of GDP per capita. Some countries do much worse than others, some much better. One interpretation is that these deviations indicate the policy space: governments can promote high subjective well-being even with limited resources, and countries like the United States can have relatively poor outcomes in terms of subjective well-being, even when they are very rich in terms of GDP per capita. Many factors besides GDP per capita determine people’s subjective well-being levels, from employment status to health, and from environmental quality to social relationships. Nevertheless, some of the international variation may also be due to reporting styles. It is certainly plausible that people with different cultures use the scale differently. For example, countries of the former Soviet Union are way below the line, but first, there are good reasons for this, and second income measures for these countries are likely to be affected by large measurement errors, so perhaps there is no need to invoke Slavic dourness. In general, we should also remember that in these global comparisons involving per capita GDP measured in purchasing power parity (PPP) terms, there is also huge uncertainty about GDP measures; so not everything has to be attributed to oddities in life evaluation measures.

Figure 7.2. Log GDP per capita is associated with life evaluations worldwide

Cantril Ladder

Note: N = 107 countries and territories. Pooled observations, 2009-13.

Source: Gallup World Poll and World Bank World Development data; Exton, C., C. Smith and D. Vandendriessche (2015), “Comparing Happiness across the World: Does Culture Matter?”, OECD Statistics Working Papers, 2015/04, OECD Publishing, Paris. https://doi.org/10.1787/5jrqppzd9bs2-en.

Another remarkable finding from the Gallup World Poll data is that experiential well-being measures are much less associated with per capita GDP than evaluative well-being, such as the Cantril ladder. At the country average level, there is only a small positive correlation between positive emotions (the sum of smiling/laughing, enjoyment and feeling well-rested a lot “yesterday”) and log GDP per capita, while negative emotions (feeling anger, worry and sadness a lot “yesterday”) show essentially no relationship (Exton, Smith and Vandendriessche, 2015). People in some African countries report as many instances of positive emotions yesterday as do people in much richer countries. If, as argued by Benthamite hedonistic utilitarians, the purpose of policy should be to maximise experiential happiness, then Gallup World Poll data would imply that Kuwait, Trinidad and Tobago and Paraguay should be giving aid to Syria, Iraq and Armenia, at least if aid improves happiness. Stress (for which Philippines is champion, and the United States is near the top), worry (Iraq) and anger (Algeria, Iran, Iraq, Turkey) are also not strongly related to income, while pain is highest in the Middle East. So, higher national income tends to come with higher life evaluation, but does little to improve the emotional lives of citizens.

Despite these findings, the association between money income and subjective well-being is not yet settled. Stevenson and Wolfers (2012), for example, claim that the derivative of the Cantril-ladder based evaluative measure with respect to the log of income is around 0.30, pretty much no matter where you look over time.

Other findings from Gallup data show that the U-shape between life evaluation and age is not universal across countries, or at least across regions of the world (Figure 7.3). It is no puzzle that the ladder falls with age in the former Soviet Union countries, given that the elderly experienced the greatest losses from the transition away from a planned economy. But there are also areas of the world, like sub-Saharan Africa, where the relation between life-evaluation and age is flat, others like Latin America and southern Europe where life evaluations fall with age, and some, like China, that share the English speaking U-pattern. Of course, these are cross-sectional results, but it is not clear how to reconcile them with a universal U that is biological, based on evidence from primates, as has been argued (Weiss et al., 2012). The life-cycle patterns of experiential well-being are more uniform across the world, with a lot of what originally reported in Stone et al. (2010) showing up in many places. Negative emotions really do seem to become less prevalent with age around the world. This should perhaps replace the U-shape in evaluative well-being as the new stylised fact.

Figure 7.3. Life evaluations and age in four world regions

Cantril ladder, mean score

Source: Steptoe, A., A. Deaton and A.A. Stone (2015), “Subjective wellbeing, health, and ageing”, The Lancet, Vol. 385(9968), pp. 640-648.

 StatLink https://doi.org/10.1787/888933839810

Correlates and determinants of subjective well-being

At an individual level, there is a growing consensus around the factors that are correlated to higher life satisfaction: being employed, having higher income, better health and stronger relationships are among the most important factors (see for example Eurostat’s Analytical report on subjective wellbeing published in 2016). A large number of other correlates has been identified in some datasets, such as environmental conditions and pollution (Silva, de Keulenaer and Johnstone, 2012). See Box 7.2 for a partial list of the research findings on correlates of subjective well-being.

There is also new work on the importance of childhood as a critical period for later subjective well-being. Children’s emotional health is the largest predictor of adult life satisfaction, above cognitive skills (Layard et al., 2014; OECD, 2015b). This indicates that while children’s subjective well-being is important in and of itself, it matters also because it is likely to be a driver of adolescent and adult outcomes (such as adult life satisfaction, employment, or school achievement). Children’s subjective well-being and emotional health are, in turn, correlated with a variety of family characteristics such as financial difficulties, family structure, moving to a different residence, and the quality of the parent relationship. One way that this research might help policy-makers would be to better understand why some children are resilient to detrimental circumstances or events, but others are not, and what the implications for public service investments are. Eurostat is currently developing a module for EU-SILC on children’s health and material well-being that will likely be collected every 3 years in the future. When these data become available, they will provide many opportunities for deeper analysis on this topic.

Levels of subjective well-being are not only determined by the events that make people better or worse off, but also by the degree to which they “bounce back” after such events. Resilience is a concept closely related to that of (true) adaptation, and several papers in recent years have added to this literature. The picture that is painted by these studies is, however, mixed: there is evidence that life satisfaction adapts to some life events (such as marriage or childbirth), but less so to others (such as disability, entry into poverty, international migration, or unemployment; see, for example, Lucas, 2007; Clark et al., 2008; Oswald and Powdthavee, 2008; Frijters, Johnston and Shields, 2011; Clark and Georgellis, 2013; Clark, d’Ambrosio and Ghislandi, 2016; Helliwell, Bonikowska and Shiplett, 2016). One potential explanation is that people adapt more to positive life events than to negative life events – which may point to a relationship between loss aversion and adaptation. Again, we point out the importance of specifying which type of subjective well-being is being assessed: this is because experiential and evaluative subjective well-being are likely to have different patterns of adaptation.

Progress has also been made in analysing subjective well-being not only as an outcome, but also as a predictor or, in the framework of a production function, as an input to other life outcomes. Steptoe, Deaton and Stone (2015) have shown that, for example, there is evidence that the three components of subjective well-being predict individual morbidity and mortality even when controlling for a wide variety of individual characteristics. Similarly, a growing body of research has supported the idea that meaning and purpose in life (i.e. eudaimonic well-being) is associated with health and mortality. For example, recent findings from the Midlife in the United States study demonstrate their subjective well-being is linked to the metabolic syndrome, a group of factors that raises the risk for heart disease and other health problems such as diabetes and stroke (Boylan and Ryff, 2015). However, recent data from UK Million Women study show that ratings of “happiness” (admittedly an ambiguous construct from the multi-dimensional subjective well-being perspective advanced here – see Box 7.1) are not linked to mortality when personal and health conditions at the first assessment point were considered in the regressions (Liu et al., 2016). Although the study omitted men, could have been more refined in the subjective well-being measures employed, and may have over-controlled co-varying health status, its conclusions challenge the prevailing sentiment about this issue.

What do people mean when they say they are “satisfied with their life”?

To know what is behind these measures, and in particular to help understand how different measures relate to each other and might be combined, it is very important to understand what people mean when they say that they are satisfied with their lives, and how they weigh different well-being outcomes – that is, what matters for people’s subjective well-being.

The correlates of the three types of subjective well-being present a generally coherent picture and provide predictive evidence that the measures are performing as expected. Nevertheless, a strong case can be made for more deeply understanding the origins of the ratings – that is, how people are generating them. In this regard, most work has been done on evaluative measures like life satisfaction, where there have been investigations into the aspects of life that bear the most on judgments of life satisfaction by simply regressing overall satisfaction measures on ratings of specific domains of life such a work satisfaction, partner satisfaction, social satisfaction, etc. (Helliwell, Layard and Sachs, 2016, special section). However, simple, a-theoretical analytic approaches may result in misleading conclusions and, importantly, lead to incorrect policy inferences. More recently, an econometric approach to decomposing global life satisfaction using stated preferences has been proposed and tested by Benjamin et al. (2014). This study has shown that people’s decisions about the future are based on a complex weighting of ratings of anticipated well-being in several imagined outcomes. In another approach to the question, new work is ongoing employing traditional qualitative methods to understand the thought processes associated with making life satisfaction ratings, for example, via the use of “think aloud” techniques (Broderick et al., 2016).

Cost-benefit analysis

Some work has used subjective well-being measures in cost-benefit analysis, as a method for valuing non-market outcomes (O’Donnell et al., 2014; Fujiwara and Campbell, 2011). The principle underlying this work is that many policies have costs and benefits that are difficult to monetise. As a result, standard cost-benefit analysis, which compares monetary costs and benefits, will lead to policy decisions that underweight those non-monetary costs and benefits if these are not taken into account. In the case of valuing non-market factors for cost-benefit analysis, current methods have serious limitations, so complementing existing methods with subjective well-being-based valuations could provide additional information. The United Kingdom Treasury’s Green Book, which provides formal guidance to government agencies on the appraisal and evaluation of policy proposals, was updated in 2011 to include a section on valuation for social cost-benefit analysis, including through subjective well-being based methods. However, additional work needs to be done to make this approach more credible. In particular, there are many problems with monetising differences in levels of subjective well-being (Kahneman and Krueger, 2006) – not least a lack of datasets containing high-quality measures of both personal income and subjective well-being, as well as conceptual problems in identifying a unit of measure with which to convert subjective well-being into dollars or euros. As with other measurement issues, the difficulties in using subjective well-being for cost-benefit analysis must be seen in light of the difficulties in other methods for cost-benefit analysis – and subjective well-being-based methods should be seen as a complement to, rather than a replacement for, more traditional techniques.

Programme and policy evaluation

Several policy or programme evaluations have included subjective well-being as an outcome indicator, and its inclusion can help both to assess the impact of a programme and to understand its mechanisms with more confidence. These studies not only support the idea that subjective well-being can be used to meaningfully measure policy impact; they also support the underlying construct of subjective well-being and its responsiveness to life circumstances. Another advantage of using subjective well-being measures in policy evaluations is that they may show that interventions have benefits that are not measured by conventional outcomes; or, conversely, to show that while conventional methods may show benefits, these benefits could be offset by lower subjective well-being (and, in both cases, providing a richer understanding of program impact).

For example, an in-work support programme in the United Kingdom was found to have unexpected negative impacts on the subjective well-being of people who participated in these programmes (Dorsett and Oswald, 2014). Similarly, an unconditional cash transfer programme in Kenya found positive impacts on the subjective well-being of participants, but negative spill-over effects on non-participants (Haushofer, Reisinger and Shapiro, 2015). In Morocco, while household connections to the municipal water supply showed no impact on health or income, they resulted in increased happiness (Devoto et al., 2012). Other studies suggest a positive impact of access to insurance on mental health (Finkelstein et al., 2012), of family leave policies on the life satisfaction of parents (D’Addio et al., 2014), and of national citizen service in the United Kingdom on the subjective well-being of participants (UK Cabinet Office, 2013). Finally, Ludwig, et al. (2013) found that subjective measures of mental health improved among participants in The Moving to Opportunities experiment in the United States before such results showed up in physical health. Measures of subjective well-being also have the potential to improve our understanding of people’s economic insecurity (Box 7.7), as also argued in Chapter 8 of this volume.

While work on refining the measurement and understanding of subjective well-being should continue, experimentation in policy applications should commence. As with many domains, experiments in policy applications and foundational work on measurement and understanding are likely to complement one another, in a mutually reinforcing process.

Causality

First, as with all other types of analysis, careful attention must be paid to establish credible causality. It is difficult to reach strong conclusions about causality based on much of the subjective well-being research that is currently available, which relies mainly on observational and self-reported data. In order for the field to advance, it will be important to focus less effort on exploratory and hypothesis-generating studies, and more on developing and testing theories. These theories, and the research designed to test them, should ideally take into account the complex inter-relationships among subjective well-being correlates, in order to identify which variables are acting as mediators, moderators, or causing actual shifts in subjective well-being. Panel data, especially panel data that can take advantage of a discontinuity such as a policy shifts, are likely to be beneficial in this respect. As mentioned earlier, this endeavour is facilitated, and indeed perhaps only possible, when explicit models and theories are used to inform analyses and interpretations.

Understanding the nature of sorting with respect to preferences is also a priority for future research. For example, Krueger and Schkade (2008) provide some evidence that workers sort themselves across jobs based on their preferences, with workers who are more extroverted tending to be employed in jobs that require more social interactions.

Heterogeneity

Second, and related, analyses of subjective well-being need to go beyond the average and examine heterogeneity. Focusing on the average subjective well-being of a group or nation is misleading in the same way as focusing on the average income of a group or nation. There may be large inequalities in subjective well-being, including inequalities among people with different demographic characteristics. It is also important to consider the possibility of different levels and correlations among different groups in order to more fully understand the dynamics of subjective well-being. However, any sub-group analysis requires large sample sizes, which are not always available, although there are certainly some large-scale datasets in development, and data from NSOs will have a particularly important role to play here. In addition, data quality on the dimensions of interest for disaggregation is often poor among existing data sets, in particular income. This is a substantial problem for research on the relationship between income and subjective well-being. Again, large official datasets such as EU-SILC allow this disaggregation and analysis, with high-quality information on a range of covariates collected in a standardised way across countries. Most of these indicators and analyses are published – in the datasets and Eurostat publication – with a variety of different breakdowns such as sex, age, income, education level, employment status, country (for most countries also region) of residence, and degree of urbanisation.

A theoretical model detailing the factors and processes underlying observed differences in subjective well-being would be invaluable for understanding and for designing possible interventions to remedy these inequalities. Building up a long time series of data across a wide range of countries will also be essential for testing these models, and to assess the dynamic relationships among drivers and outcomes.

Similarly, relatively little is known about the tails of the distribution of subjective well-being, particularly those who rate themselves as having extremely low levels of subjective well-being (including those who live with high levels of pain). These people are likely to represent a particular policy concern, and much may be learned from research focusing on them.

Finally, research efforts should continue to focus on adaptation and resilience, as these are among the most promising, if difficult, research avenues. Further investigation of the role of public goods and services in supporting people’s resilience might lead to findings that could be directly acted upon by policy-makers; in other terms, such research might highlight the resources that are needed to restore subjective well-being after a life challenge.

Data collection and availability

As described above, data collection on subjective well-being has expanded enormously. There are, however, two important areas where there is still a lack of data on subjective well-being, and where the inclusion of subjective well-being questions is likely be relatively low cost. The first is to expand high-quality data collection on subjective well-being to poor countries, for example, by including a life satisfaction and experiential well-being module in household surveys conducted in these countries. As well as casting important light on the societal conditions and policy environments that can influence changes in subjective well-being over time, research into level differences among countries has potential for addressing persistent methodological and conceptual questions concerning the meaning of subjective well-being responses from people in different cultural and economic settings. Have people in poor countries adapted to their circumstances, and are they using the subjective well-being scales in entirely different ways than those in wealthy countries? And if so, what do country comparisons mean?

Second, in order to increase our understanding of experiential well-being, the inclusion of subjective well-being measures in official time-use surveys is important, as recommended in the NAS report mentioned above. Such efforts could be supported by more guidance on which approach is best for this purpose, and by increasing the research output linked to existing efforts. The Guidelines on Harmonising Time Use Surveys (UNECE, 2013) have been useful in this area but additional analysis of different approaches to collecting experiential well-being data are necessary in order to provide empirical guidance on best practice in this area. As mentioned above, the inclusion of subjective well-being items in on-going time use surveys, such as the American Time Use Survey, is an efficient way to achieve this goal. Collecting time-use data and experiential well-being in poor countries would also be particularly useful, as little is known about the daily activities or experiential well-being of rural populations.

Finally, timely access to these data is critical, as well as responsiveness by researchers to new data availability. Measurement initiatives will push forward our knowledge and understanding of subjective well-being, but researchers must have access to these data in order to achieve this outcome. In turn, researchers must demonstrate the usefulness of these measures, or the measures risk being dropped. Increased co-operation between actors would improve the quality and usage of data on subjective well-being, and networks can play an important role in this respect.

Data analysis and interpretation

As discussed above, one of the most important issues inadequately addressed by current research is that of systematic differences in question interpretation and response styles between population groups. Is there conclusive evidence that this is a problem? And, if so, are there ways to adjust for it? Information is needed about which types of group comparisons are affected, about the magnitude of the problem, and about the psychological mechanisms underlying these systematic differences.

While some of the methodological issues that have been associated with subjective well-being measures are, to varying degrees, minimised through the use of standardised questionnaires, this issue is not resolved through standardisation alone. So far, only limited analysis has been performed for the assessment of the 2013 EU-SILC ad hoc module regarding the viability of cross-national comparisons (Eurostat, 2016). The analysis that exists has been mostly undertaken through internal and external validation (correlations with certain variables of interest), and the results are encouraging.

Some progress has also been made on using vignettes to address this issue but this approach has limitations. For example, people’s responses to the circumstances described in vignettes are very likely to be shaped by the policy environment in which they live – because the implications of, for example, living with a chronic health condition depend on factors such as health care cost and availability, as well as disability benefits. This means that “correcting” subjective well-being self-reports of their own conditions according to international differences in how people rate the same vignettes could ultimately remove the most policy-relevant part of the international variation in well-being, effectively throwing the baby out with the bathwater.

The U-index, which calculates the share of time that individuals spend in an unpleasant state – defined as a period when the strongest reported emotion is a negative one (Kahneman and Krueger, 2006) – is a promising method for neutralising differences in the extent to which response scales are used differentially across countries or groups of people. While some work is available supporting its utility, its application so far has been limited to measures of experiential well-being. More work is needed to see how large a problem inter-personal differences in the use of response scale actually is, and to find new cost-effective solutions.

Subjective well-being questions are generic, but there are times when they should be tailored to the application at hand. For example, in the US Moving to Opportunity project, feelings of security and anxiety were especially relevant for programme-participants and were specifically targeted. In many medical studies, pain may be a particularly relevant type of experiential well-being, whereas misery may be especially relevant in studies of refugees.

More population-level work is needed on subjective well-being beyond life satisfaction, i.e. extending measurements to eudaimonic and experiential well-being. At present, a variety of different approaches are being used across OECD statistical offices to collect this type of information (see Exton, Siegerink and Smith, forthcoming, for further details). As described in Box 7.1, measures of subjective well-being belong to three different categories. Life satisfaction is generally the most widely used measure of subjective well-being, and the one for which there has been the most research, for example, on adaptation. A focus on a single indicator is, on the one hand, beneficial and pragmatic, as it ensures a wider evidence base for at least one indicator. On the other hand, there are surely missed opportunities, as there is likely much to be learned from data on the other dimensions of subjective well-being.

We concur with the conclusion of the NAS report (Stone and Mackie, 2015) that research on subjective well-being should be explicit about the types of measure used, and should ideally include more than one type of measure. More effort should be given to understanding eudaimonia and experiential well-being, to describing the relationship between the different measures, and to ensuring that these outcomes are not being neglected.

Work on the other aspects of subjective well-being needs to be balanced with the need for clarity in communication with those outside the research community. While a synthetic indicator that combines different dimensions of subjective well-being is probably not a strategy that is well-grounded in current understanding of the different constructs, there is often some demand for combining information in summary form. Effort should be made to ensure that these measures, if they are to gain traction in a policy setting, are reasonably easy to understand and compare.

Related to this question is an issue that studies on subjective well-being share with other topics in this volume: how to think about, and communicate, what constitutes a meaningful change or difference in the measure considered. Especially in large samples, a very small difference may be statistically significant but not very important for policy-making. This can be particularly problematic when considering differences on a scale that has no inherent meaning. One option is to phrase differences in terms of some calibration (i.e. a difference of magnitude “x” corresponds to the increase in subjective well-being associated with an increase in personal income of “y”, although much more research is needed to establish a universal unit of measurement). It should also be noted that the original work on the Day Reconstruction Method presented differences in experiential well-being across groups by aligning those differences with the affect associated with typical daily activities, providing an informative ruler for interpreting affective differences.

Experimentation and innovation

The application of subjective well-being to policy is at an early stage, and there is still much to learn. At this point, there is sufficient understanding of methodological issues and consensus on the best way to address most of those issues to cautiously move forward. Experimental initiatives will, in turn, generate new questions and more progress on the methodological issues. However, many policy applications will have to wait until a sufficiently large and long cross-country dataset has been built up, which will take time. In the meantime, there are a variety of ways that progress can be made.

Subjective well-being can complement existing policy analysis methods, and we recommend taking steps to consider all three dimensions of subjective well-being in order to obtain a complete picture. However, we realise that this recommendation is much less specific than we would like it to be, and the reason for this follows. Evaluative measures of subjective well-being (life satisfaction, Cantril Ladder, Diener scale) seem to us to have a conceptual advantage over the eudaimonic and experiential subjective well-being measures in that they target peoples’ summary evaluations of their current lives, which at face value appears more consistent with their choices and therefore with the economic concept of “utility”.

Experiential measures may be less ideal because of concerns about true adaptation to changing environmental situations (see Sen’s “happy peasant” arguments), though experiential measures do capture an essential aspect of well-being – how one feels. Eudaimonia addresses broader meaning and aspirations, which are undoubtedly an important aspect of life, yet it appears less directly applicable as a measure of utility. An example of this approach tracked evaluative and experiential well-being for people who migrated from Tonga to New Zealand, and showed how these measures illuminated their complex psychological transition (Stillman et al., 2015).

There are drawbacks, though, with exclusively adopting evaluative well-being as the primary measure for policy. First, as mentioned above, we believe that experience is an essential part of subjective well-being that should not be omitted. Second, as reviewed in the methodological sections of this chapter, evaluative measures are prone to being perturbed and, possibly, biased by a number of irrelevant factors, which raises questions about recommending their use for policy-making – particularly if the data are to be used as a proxy for utility, rather than as one of a variety of subjective well-being measures that might inform policy decisions.

Additional work should be done on cost-benefit analysis, to understand whether and how results from analyses based on subjective well-being valuation can complement existing methods. Early efforts have produced some extreme results, potentially due to income measurement issues (i.e. if the estimated partial effect of income on subjective well-being is small, and does not represent the “true” value of income, scaling other effects by this coefficient will lead to implausibly large monetary effects); and because models and theories were not sufficiently developed to allow sound interpretation. In addition, some other measures are particularly amenable to links with subjective well-being research, for example, the well-being adjusted “quality adjusted life years” (QALY), the so-called WELBY – well-being adjusted life year.

Some fairly low-cost initiatives would be to routinely collect, and report on, subjective well-being indicators in programme evaluation questionnaires, and to routinely add subjective well-being questions to questionnaires such as labour force surveys or surveys carried out in schools. In the European context this will be done using the EU-SILC as a vehicle, as 2-3 indicators (including life satisfaction) will be collected on a yearly basis. As shown in Box 7.2, data on subjective well-being is an important outcome not only in itself, but also as an input or driver of other outcomes of interest, and can help researchers provide a richer data analysis.

Researchers, in turn, need to spur applications to policy, in part to demonstrate to NSOs that investment in subjective well-being data is worthwhile. Much of the current research on subjective well-being is difficult to apply to policy, even experimentally. A greater focus on policy applications in the literature (e.g. on policy-amenable drivers of subjective well-being) would be helpful. Co-ordination between researchers, policy-makers and NSOs may be very valuable, a role that the OECD may be well-placed to play.

Cautions in using subjective well-being data for policy

Well-meaning but naïve policy changes may make people worse off due to the complex inter-relationships between choices, prices and heterogeneity in subjective well-being, as well as general equilibrium effects. Airport noise is one example: people live near airports for a reason, typically because of lower housing prices, or because they do not care too much about noise, and well-meaning policy (e.g. to reduce traffic at night) could actually make them worse off because of the effect of lower noise in raising house prices (this example is taken from a personal communication with Angus Deaton). We need to understand why people live where they do, and build models of how location could change in response to policy changes. Hedonic models of sorting are well established in labour economics (e.g. Rosen, 1986) and urban economics (e.g. Roback, 1982), and this type of work can be extended to subjective well-being.

A less ambitious but still important goal for policy is simply to provide information on subjective well-being and let people and businesses use it as they see fit. To use a term from David Halpern, “de-shrouding” subjective well-being means giving the public information on the correlates of subjective well-being – informing them that priests, or people without children, or people who live in Denmark are happier than others (clearly, some of this has already occurred in the news media). To some extent, this is simply providing information, which might in principle be useful to someone considering a move to Copenhagen.

However, such information may also be misleading, partly because of the considerable challenges in identifying causality in these studies. The types of people who become priests are very special, and the standard battery of variables used to control for differences between groups (e.g. sex, age, educational background, income, etc.) is unlikely to fully control for the difference between people who become priests and those who select other occupations. In this regard, subjective well-being data are no different from data on average income: while it is informative to know that doctors earn more than the average worker, it is the case that many individuals do not have the training, aptitude, or temperament to work as doctors. Therefore, to continue with the subjective well-being example, entering the clergy may not yield the expected gains in subjective well-being that were anticipated on the basis of the de-shrouded well-being averages.

In addition, results from observational studies are averages, even if they are averages within groups, and as such may not apply to a given individual. So, the application of subjective well-being data must be done in a considered way, given that the potential for unintended consequences are far from academic.