Notes

  • 1. http://hdr.undp.org/en/statistics/hdi/.
  • 2. It is important to note that these findings are based on worldwide visitors to the OECD’s Your Better Life Index website, http://oecdbetterlifeindex.org/, a sample of individuals known to be non-representative and non-random, and as such they should be interpreted with care.
  • 3. Frame-of-reference effects refer to differences in the way respondents formulate their answers to survey questions, based on their own life experiences, as well as their knowledge about the experiences of others - including both those they consider as within their “comparison group” and those outside it.
  • 4. Adaptation refers to psychological processes that may either restore or partially repair subjective well-being, and particularly affective experiences, in the face of some types of adversity. People may also show adaptation to positive life events over time (whereby the initial subjective well-being boost delivered by a positive change in life circumstances, such as marriage, reduces over time).
  • 5. Much of the critique surrounding the use of subjective well-being for public policy centres around a view that increasing positive emotions is not an appropriate goal for governments (e.g. Booth et al., 2012; McCloskey, 2012). Although this view potentially underestimates the health and well-being implications of emotional experiences (described above), and fails to distinguish between the usefulness of monitoring positive emotions, versus making them primary objectives of government policy, it is nonetheless further grounds to avoid describing subjective well-being data solely in terms of “happiness”.
  • 6. Ordinal data are those measured on scales where the intervals between scale points are not assumed to be equal, but there is an underlying sequence or rank order. For example, we assume that a 5 is lower than a 6 and a 6 is lower than a 7, but we do not assume that the distance between 5 and 6 is equivalent to the distance between 6 and 7. Linear regression relies on continuous variables, where cardinality is assumed, i.e. where the size of the number on a scale is expected to have a direct linear relationship with the amount of the variable in question. Tabachnick and Fidell (2001), however, note that in the social sciences, it is common practice to treat ordinal variables as continuous, particularly where the number of categories is large - e.g. seven or more - and the data meet other assumptions of the analysis.
  • 7. Data users are likely to want to know, for example, how good or bad a person’s experience was, not just on which side of a cut-off they fall. This is less of a problem for the reporting of headline national aggregate figures, but becomes particularly relevant when comparing responses between groups. It can be ameliorated to some extent by banding responses into several categories rather than selecting just one cut-off point.
  • 8. For example, a country with universally low levels of subjective well-being would have few individuals falling below the relative poverty line, thus masking the extent of difficulties faced.
  • 9. For example, the thresholds associated with clinically-significant mental health outcomes may be very different to the thresholds associated with different educational or income levels.
  • 10. There are some who disagree with this, arguing that cardinal interpretations of subjective well-being are possible - e.g. Ng (1997).
  • 11. It is unclear, for example, what it really means to say that the bottom 10% of the population achieves only 1% of the total subjective well-being. This can be contrasted with income, where it is easier to understand the practical implications of the bottom 10% earning just 1% of the total income across a population.
  • 12. Although Helliwell, Barrington-Leigh, Harris and Huang (2010) took the simple mean average of the Cantril Ladder and a single-item life satisfaction measure and found this was more closely correlated with predictors of subjective well-being (such as demographics, income and a set of social indicators) than either measure on its own.
  • 13. The UK’s ONS have also proposed a single-item eudaimonia question, for high-level monitoring purposes: “Overall, to what extent do you feel the things you do in your life are worthwhile?” (ONS, 2011b).
  • 14. In the case of more detailed analyses, such as group comparisons or investigation of the drivers of subjective well-being, separate estimates of the different sub-components of subjective well-being will be preferred due to the risk of information loss when summing across sub-components.
  • 15. For example, if the threshold on a 0-10 scale is set at 7, movements across that threshold will be very salient, but large-scale movements from 8 to 9, or from 2 to 5, may go undetected.
  • 16. Represented as a percentage of the population reporting higher positive than negative affect; OECD calculations based on figures from the 2010 Gallup World Poll.
  • 17. I.e. a change over a one, five or ten-year period. As discussed earlier, short-term fluctuations of this magnitude can also be detected, but may not represent meaningful societal shifts in overall levels of well-being (e.g. Deaton, 2012).
  • 18. For example, the World Happiness Report (Figure 2.3) lists 63 countries where the mean average life evaluation between 2005 and 2011 (measured on a 0-10 Cantril Ladder scale) is lower than the scale midpoint, 5. These include India, China, Iraq, Afghanistan and Syria; and particularly low-scoring countries, with scores below 4.0, include Congo, Tanzania, Haiti, Comoros, Burundi, Sierra Leone, the Central African Republic, Benin and Togo.
  • 19. Reverse causality in this context refers to when subjective well-being drives the independent variable, rather than vice versa. For example, in a cross-sectional analysis, a significant association could be observed between income (the independent variable) and subjective well-being (the dependent variable), but this could be because subjective well-being drives income (rather than vice versa). Two-way causality is where there are reciprocal and causal relationships between two variables in both directions - i.e. income drives subjective well-being, but subjective well-being also drives income. Endogeneity refers to a situation where there is a correlation between an independent variable and the error term in a regression model. This can be due to measurement error, omitted variables, sample selection errors, and/or reverse or two-way causality.
  • 20. To quote from Kahneman and Riis (2005): “... consider theAmericans and the French. The distributions of life satisfaction in the US and France differ by about half a standard deviation. For comparison, this is also the difference of life satisfaction between the employed and the unemployed in the US, and it is almost as large as the difference between US respondents whose household income exceeds USD 75 000 and others whose household income is between USD 10 000 and USD 20 000 (in 1995). Is it possible to infer from the large differences in evaluated well-being that experienced well-being is also much lower in France than in the USA? We doubt it, because the sheer size of the difference seems implausible” (p. 297).
  • 21. Such as tendencies to use either extreme or more “moderate” scale response categories, as well as the likelihood of socially desirable responding.
  • 22. There are times when people’s subjective perceptions matter, even when they don’t reflect objective reality: “cultural differences may in some cases be relevant to policy and in some cases irrelevant. For example, people’s satisfaction with leisure opportunities might be relevant to policy deliberations, regardless of the objective conditions” (Diener, Inglehart, and Tay, 2012, p. 20). Another classic example of this is perceptions of regulatory burden, which can influence important business decisions and behaviour regardless of their accuracy (OECD, 2012). If perceptions of regulatory burden are misplaced, activity should be focused on better informing businesses and the wider public. Few would argue that the correct response would be to simply adjust the perceptions of regulatory burden so that they fit the pattern observed among more objective measures.
  • 23. This Latin American effect has been explored in depth by Graham and Lora (2009).
  • 24. See Chapter 2 for a full account of response styles.
  • 25. As Senik (2011) puts it, “If the French evaluate the happiness of some hypothetical person in a less positive manner than the Danes, perhaps it is because they would actually feel less happy in the situation of that hypothetical person” (p. 8).
  • 26. In practice, however, substituting even the highest adjusted figure for French natives (7.54) would only cause a very minor adjustment in country rankings overall, causing French natives (7.22) to exchange places with natives of Great Britain (7.38) only. Overall, mean average happiness ratings among natives in the 13 countries sampled range from 6.74 in the case of Portugal to 8.34 in the case of Denmark.
  • 27. Suppose that two individuals both have the same levels of underlying positive and negative affect. But imagine that the first has a tendency towards extreme responding - so that on a 0-10 scale this individual reports 9 on positive affect and 7 on negative affect. The second individual has a tendency towards more moderate responding, thus reporting a 7 on positive affect and a 5 on negative affect. The net affect balance for both individuals will be +2. This of course assumes that response biases operate in a similar way for both positive and negative affects, which requires further examination.
  • 28. A substantial part of the literature in this field uses country as a proxy for culture, inferring cultural differences on the basis of country differences. Whilst this is potentially problematic, the words country and culture are often used interchangeably in many studies on the subject.
  • 29. Although the implications of the subjective well-being literature are often interpreted in terms of opportunities for policy interventions, subjective well-being has just as much potential to identify areas where existing government interventions can be redesigned or stopped altogether.
  • 30. Double-blind conditions refer to scenarios where neither the respondent nor those implementing the intervention are aware of which treatment group a given respondent has been assigned to. Single-blind is where those implementing the intervention know which treatment group has been assigned to which respondent, but respondents are unaware.
  • 31. Wider applicability can be challenged where there are concerns about the extent to which a given result might generalise to other situations, beyond the experimental or quasi-experimental setting.
  • 32. For some research questions investigating international differences in subjective well-being, where the driver in question is hypothesised to operate at an aggregate country level, pooled cross-sectional time series data (i.e. international data containing repeated study waves and representative, but different, samples in each wave) may also enable some causal inferences.
  • 33. Factor analysis is a statistical procedure that is conducted to identify distinct (relatively independent) clusters or groups of related items or variables (called factors). It is based on patterns of correlations among items or variables.
  • 34. Because error is estimated and removed in the process of extracting the underlying factors. A “factor loading” is calculated for each item or variable, which reflects the variance it shares with the underlying factor - and all other variance is assumed to be error. When factors are used in the analysis (instead of measured variables), only this common variance is analysed, and thus measurement error is, in theory, purged from the data.
  • 35. One interesting exception, however, is cultural biases. Although there is currently some evidence of cultural bias in direct country comparisons of mean average levels, there is currently little to suggest that cultural biases exert a problematic influence on cross-country analyses of the drivers of subjective well-being - and drivers tend to be reasonably consistent across countries (e.g. Fleche, Smith and Sorsa, 2011; Helliwell and Barrington-Leigh, 2010). The issue of the extent to which regression solutions are replicable and can be generalised from one sample or country to another is, however, an important consideration that is discussed below.
  • 36. However, statistically, mediation and confounding are identical: both are indicated when the inclusion of the third variable in the model reduces the relationship between the independent variable and the dependent variable by a non-trivial amount.
  • 37. An example would be a significant statistical relationship between the time I spend talking to the plant in my office and how much it grows per year - both of which are related to an (unmeasured) causal variable: how often I remember to water the plant.
  • 38. An example here might be that a causal relationship between how often I remember to water my office plant and how much it grows is obscured by a third (unmeasured) variable: how much plant food my colleague gives it.
  • 39. Shared method variance refers to variance that is attributed to the measurement method, rather than the constructs of interest. In the case of subjective well-being, the main concern is that if drivers are also measured through subjective self-report data, self-report biases (including retrospective recall biases, response styles, cultural bias, etc.) could inflate observed relationships between those drivers and the subjective well-being outcomes of interest.
  • 40. I.e. a dispositional tendency towards experiencing negative affect.
  • 41. An instrumental variable is one that has a direct association with the independent variable in question (e.g. income), but not with the outcome of interest (e.g. life evaluations).
  • 42. The term function is used here to describe the overall pattern of relationships between the independent variables and the dependent variable, including the size and significance of coefficients.
  • 43. Dolan and Metcalfe conclude that “we need much more research into the extent and the sources of the differences between these valuation methods” (p. 25), particularly given that the valuation through subjective well-being approach is still in its infancy and “literally thirty years behind that of generating monetary values from revealed and stated preferences”.
  • 44. These include the “standard gamble” and “time trade-off” methods (see Dolan and Kahneman, 2008).
  • 45. See the previous section on interpreting the drivers of subjective well-being for a more detailed treatment of the generalisability of results.
  • 46. An instrumental variable is one that has a direct association with the independent variable in question (e.g. income), but that is associated with the outcome of interest (e.g. life evaluations) only via the independent variable in question.
  • 47. Based on the current state of knowledge, these authors suggest that “there would seem to be good grounds for viewing the ICs (income compensation - i.e. valuations) as a total value over a finite horizon. Clearly, the actual assumption made on how life satisfaction incorporates future expectations is crucial to the methodology of the value of the non-market good by experiences, and merits further investigation” (p. 23).
  • 48. Two different estimates were produced because different models were estimated for participants, based on which type of welfare payments they received from the government prior to their participation in the programme.
 
Source
< Prev   CONTENTS   Source   Next >