The Development of Capability Indicators

Abstract

This paper is motivated by sustained interest in the capabilities approach to welfare economics combined with the paucity of economic statistics that measure capabilities at the individual level. Specifically, it takes a much discussed account of the normatively desirable capabilities constitutive of a good life, argued to be comprehensive at a high level of abstraction, and uses it to operationalize the capabilities approach by developing a survey instrument to elicit information about capabilities at the individual level. The paper explores the extent to which these capabilities are covariates of a life satisfaction measure of utility and investigates aspects of robustness and subgroup differences using standard socio‐demographic variables as well as a relatively novel control for personality. In substantial terms, we find there is some evidence of quantitative, but no qualitative, gender and age differences in the capabilities–life satisfaction relationship. Furthermore, we find that indicators from a wide range of life domains are linked to life satisfaction, a finding that supports multi‐dimensional approaches to poverty and the non‐materialist view that people do not just value financial income per se. Our most important contribution, however, is primarily methodological and derives from the demonstration that, within the conventions of household and social surveys, human capabilities can be measured with the aid of suitably designed statistical indicators.

Keywords:

• Capabilities
• Measurement
• Advantage
• Multi‐dimensional welfare indicators
• Human development
• Welfare
• Happiness
• Life satisfaction
• Personality controls
• Gender differences
• Age differences

Acknowledgements

The authors acknowledge funding from the Arts and Humanities Research Board, Grant No. 17685. In addition, we are particularly grateful to Maria Sigala, Ingrid Robeyns and Sabina Alkire for comments on question and research design and to Ron Smith for comments on econometric analysis. We also thank anonymous referees and participants at the Human Development and Capability Association conferences in UNESCO (Paris), Pavia and Delhi, a UN Wider Conference on Poverty in Helsinki, an International Health Economics Association (iHEA) conference in Copenhagen, and seminar participants at the following universities: University of Oxford, University of Cambridge, University of Norwich, University of Copenhagen, University of Pavia, Erasmus University and The Open University. Finally, the first author wishes to thank Caterina Laderchi for initial discussions about the operationalization of the capabilities approach and to Amartya Sen and Hillel Steiner for supporting this project.

Notes

1. See, for instance, surveys by Anand (1987) and Machina (1989).

2. We use the term ‘approach’ following the literature: This emphasizes the fact that the main contribution of theoretical work has been in the formation of concepts and understanding their relations. In economics (and philosophy of science to a lesser extent), the term theory is often used to refer to a narrower, mathematical formulation of concepts, and for the capabilities approach one such theory can be found in Sen (1985a).

3. Empirical applications can be found in Schokkaert and van Ootegem (1990), Qizilbash (1996, 2004), Chiappero‐Martinetti (2000), Layte et al. (2000), Laderchi (2001), Alkire (2002a, 2002b), Burchardt and Le Grand (2002), Burchardt and Zaidi (2003), Clark (2003), Kuklys (2005), and Anand et al. (2009). The theoretical literature on freedom is perhaps not as closely related to empirical work on capabilities as it might be, although more recent papers suggest modest evidence of conceptual convergence — see, for example, Gaertner and Xu (2005), van Hees (2004), Nehring and Puppe (2002) and Pattanaik and Xu (1998). (The observation is a comment on the recent economic history of development thought and does not imply that high‐income countries are more utilitarian, by necessity, than low‐income countries.)

4. Rights and freedoms enter into utilitarian calculations to the extent that people value them. However, this contingent approach to valuing so‐called de‐ontological claims is one of the aspects that causes many to object to utilitarian approaches to welfare.

5. See, for instance, Anand and Wailoo (2000).

6. It is also worth pointing out that most functions, this one included, are specific to the individual. Such functions could be measured with panel data, but as we have only one wave of data on personality and socio‐demographics, in effect we estimate an intermediate position in which the functions vary with respect to personality and socio‐demographics.

7. Nor is it possible to separate out variations in SWB due to variations in Q and those due to variations in a person's actual functionings.

8. The instrument used to measure personality is a short form developed for incorporation into research where personality is not the sole focus. Devised by Gosling and Rentfrow (2003), it consists of five pairs of questions that are responded to on a one to seven scale with agreement semantic anchors. The score is summed in each pair, thus giving five dimension scores in the range 2–14. The questions and variable names are given in Appendix 1.

9. In their survey presented to an American Economic Association conference, Kuklys and Robeyns (2005) suggest that only three studies from nearly 50 have concentrated on capabilities. Yet such exercises are vital for operationalization (Comin, 2001).

10. The value of choosing the BHPS is that it is a secondary data source with similar counterpart surveys in many countries around the world. This means that any questions that come from it are likely to be asked regularly and in similar form in other countries, which in turn implies that such questions could, in principle, be used as a basis for international monitoring and comparison.

11. The questions were devised through a process that included a workshop held at Wolfson College, Oxford in September 2004 and piloting with potential respondents. In some cases, responses are recorded in a more detailed manner than indicated in Table 1. Usually categories have been collapsed where cell responses were deemed small, although the original categories may well be of interest to future survey designers and are available from the authors on request.

12. There are a number of discussions in mainstream economics journals about the use of SWB as a measure of well‐being, although Oswald (1997) remains one of the best. Manski (2004) provides a useful complement in that he focuses on evidence that indicates the reliability of such data.

13. Respondents are from England, Scotland and Wales and will not be entirely representative of the elderly.

14. Least‐significant variables were eliminated sequentially and the model re‐run until all remaining capability variables were significant at the 5% level.

15. It has to be said that the psychological literature has tended in the past to concentrate on bivariate analyses — so multivariate analyses make a valuable addition to that literature.

16. We also estimated ordered logit and ordered probit models and find that, as one might expect, they give slightly better fits than ordinary least squares but tell a virtually identical story when it comes to identifying statistically non‐zero coefficients. It is perhaps also worth commenting on the practice of treating ordinal scales numerically. The justification is merely pragmatic and avoids regression results with hundreds of coefficients that are both difficult to read and interpret and make heavy demands on degrees of information. This amounts to imposing a linearity assumption on the functional form of the partial relations, which is innocent for truly linear relations but is likely to result in conservative estimates of relationship strength for non‐linear relations.

17. This is also consistent both with our finding that the quality of a job has a positive impact on life satisfaction, and Winkelmann and Winkelmann (1998) who also control for income and find that the non‐pecuniarary costs of unemployment are high.

18. We have shown elsewhere how the data can be used to shed light on specific topics such as the experience and consequences of violence (Anand and Santos, 2007) and the association between health and capability poverty (Anand et al., 2008).