Extreme bounds of subjective well-being: economic development and micro determinants of life satisfaction

ABSTRACT

This article presents a global sensitivity analysis of micro determinants of life satisfaction (LS), a subjective well-being (SWB) measure of quality of life, as it relates to economic development. We test 53 micro variables using extreme bound analysis on a pooled cross-section data from the World Value Survey representing 98 countries between 1989 and 2014. Several standard variables frequently included in SWB regressions are controlled for. The test variables are broadly categorized as demographic, personal–economic, individual traits and values, social attitudes and collectivism, social relationships and perceptions of control. We find subjective health status, household income rank, family savings, religiosity, most perceptions of control and several social attitudes and collectivism measures to be universal LS determinants. Generally, the determinants of LS vary by a country’s level of economic development. In particular, we provide suggestive evidence that as countries develop, certain needs are satisfied and stop being important contributors to LS while others take their place.

KEYWORDS:

• Subjective well-being
• life satisfaction
• extreme bound analysis
• economic development
• world values survey

JEL CLASSIFICATION:

• D010
• D120
• D630
• I310
• O120

Acknowledgements

We are very grateful to Douglas Almond, Jenny Bourne, Thomas Buchmueller, Nathan Grawe, Benjamin Jones, Mark Kanazawa, Erzo Luttmer, Sala-i-Martin, Jenny Minier, Milena Nikolova, Martha Paas, and Paolo Verme for their valuable advice and guidance. Also, a big thanks to Liz Llanes for being a wonderful research assistant and to Paula Lackie and Michael Tie for their superb computing support.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ See Veenhoven (2013).

² Establishing causality for these statistically significant SWB determinants, however, remains a work in progress (Helliwell and Barrington-Leigh 2010; Graham and Nikolova 2015).

³ There are some exceptions (e.g. Stevenson and Wolfers 2013; Kingdon and Knight 2007).

⁴ A number of theoretical models have tried to explain this phenomenon and a size-able empirical literature has tried to establish or debunk the validity of the Easterlin result.

⁵ In the baseline regression, which includes only the micro controls, variables, such as education, religious attendance and belief in superior being, are not statistically significant.

⁶ These include the countries’ openness, business climate, and the number of chambers in parliament. The authors find that splitting the dataset by gender, income, and political orientation does not change their results significantly.

⁷ To the latter point, it is worth noting that with the exception of studies exploring the relationship between SWB and income, there are not many studies, if any, that explore the relationship between micro SWB determinants and economic development.

⁸ The choice of standard variables is usually motivated by the literature.

⁹ We also weighted our regressions by the variable ‘S017’ available in the WVS.

¹⁰ There is no widely agreed upon VIF threshold among social scientists. Some of the ad hoc threshold values include 4 (Pan and Jackson 2008), 5 (Menard 2002; Rogerson 2001) and 10 (Hair et al. 1995; Menard 2002; Mason, Gunst, and Hess 1989). Changing the VIF to 5 or 10 does not change the results presented here considerably.

¹¹ Due to missing data for the dependent variable or the standard controls, we are unable to use data from Croatia, Israel and for any country from 1981, 1982 and 1984.

¹² Kahneman and Krueger (2006) point out that these simple measures do not correlate well with benchmark metrics such as time-weighted averages of daily recorded accounts of negative and positive feelings. These benchmark measures are not, however, available in cross-country surveys covering developing countries. Our article is thus limited to checking for robust determinants of the LS. Several studies find the LS measures to be reliable, stable over time, and strongly correlated with other known well-being measures (Frey and Stutzer 2002; Larsen, Diener, and Emmons 1985; Lepper 1998; Sandvik, Diener, and Seidlitz 2009; Ng 1997). The WVS also offers another measure of SWB, ‘feeling of happiness’. This measure is qualitatively different from LS, with a low correlation of about 0.3 with LS. Much of the literature that exploits the WVS dataset also uses the LS measure of SWB. Helliwell et al. (2015, chapter 2) offers a detailed discussion for why they use the LS measure in their analyses. We find their reasoning convincing.

¹³ The minimal set of standard controls may also be found by sequential data mining or some general to specific methodology. We avoid that kind of approach here since we are evaluating dominant practices in the literature.

¹⁴ Respondent are asked to subjectively rank their household’s income on a scale of 1 to 10 when compared to the income distribution of the economy.

¹⁵ See Appendix 1, Tables A2–A3 for descriptions of how the variables are coded and references to studies that use them.

¹⁶ We do not have a measure for genetic factors.

¹⁷ Excluding the country–year dummies from both the EBAs affect our findings minimally.

¹⁸ The only exception being individuals with a college degree are more satisfied in HICs.