Identifying the worst off in a society: does the choice of well-being measure matter?

The first basic task in the design of anti-poverty policies is the identification of the worst off, i.e. those individuals that should benefit from redistribution through, e.g., targeted social benefit schemes. This requires a method to make interpersonal comparisons of well-being. In practice, these comparisons rely almost exclusively on income as a measure of individual well-being. The worst off are then identified as the persons with the lowest incomes. This standard approach is attractive from a practical viewpoint, as the analyst only needs information on the income position of the individuals. However, income alone has already for long been criticized as being a too narrow information basis to measure well-being appropriately (see Stiglitz et al. (2010) for an extensive survey).

Rather, in the last decades, it has been well documented that also non-monetary resources matter to well-being and that people care a lot about these other dimensions of their life, as for instance health, employment status, social life, etc. (cp., e.g., Benjamin et al. (2014)). Accordingly, various alternative approaches have been proposed to perform well-being comparisons with a broader and richer information basis. These approaches mainly differ in how the different dimensions of life are taken into account and whether individuals’ personal views or tastes are (partly) respected or not (see Fleurbaey and Blanchet (2013) for an overview).

In a recent working paper, Koen Decancq and I empirically compute several of these alternative measures using a common dataset (the German Socio Economic Panel of 2010) and compare them to income. The main question we ask is whether the different measures identify the same or different people to be at the bottom of the well-being distributions (Decancq and Neumann (2014)).

A first alternative measure we compute is a composite index, i.e. an objective measure that adds different dimensions of life together in one single index.[1] For this, different parameter choices have to be made by the analyst, amongst which the weighting of the different dimensions might be the most important one. For illustrative purposes, we consider only three life dimensions, namely health (measured via an objective health index) and employment status (being officially unemployed or not) besides household disposable income. Rather than setting equal or otherwise exogenous weights, the weighting scheme we derive is based on the underlying data, using differences in the variances of the three components. The final index is normalized such that it ranges between 0 and 1.

However, one might still object that this approach is too ‘paternalistic’[2]. Thus, a second natural alternative would be to directly use the opinions of the individuals themselves, i.e. a measure of subjective well-being – an approach that has become very popular during the last decades (cp., amongst many others, Diener (2000), Frey and Stutzer (2002), Layard (2005), Kahnemann and Krueger (2006)). Koen and I use life satisfaction, measured by the question: “To what extent are you satisfied with your life in general at the present time?” As an answer, respondents attach a score from 0 (“not satisfied at all”) to 10 (“completely satisfied”). However, one main problem with the mere use of such an ordinal measure is that the scale 0-10 will mean different things to different individuals, i.e., a “7” for Bob might be a “5” for Ann. Another important issue is possible adaptation to bad living conditions. Thus, scaling differences make it difficult to interpersonally compare such a measure of well-being.

A measure that might be an answer to this problem, while still taking into account the individuals’ opinions, is the so-called equivalent income. By respecting individual preferences over the different dimensions of life, it ensures interpersonal comparability (see, e.g., Fleurbaey (2011)). Precisely, the equivalent income of an individual is the hypothetical income that, if combined with a reference value on all non-income dimensions (i.e. health, employment status, etc.), would make the individual equally well off as compared to her initial combination of income and the non-income aspects. That is, technically, by fixing the non-income values to be the same for everyone and using everyone’s current level of preference satisfaction, individuals can be compared in terms of the associated amount of income. What is essentially needed for this measure is information about individual preferences and a choice for the reference values of the non-income dimensions. Koen and I combine the information from the two former approaches to construct an equivalent income measure. Namely, we use life satisfaction as a proxy for well-being and again assume income, health and employment status to be the dimensions from which individuals derive their well-being, and over which they thus have preferences. This enables us to estimate individual preference representation functions (cp. also Decancq et al. (frth.)). By assuming the “best” possible values as the non-monetary references (i.e. “perfect” health and not being unemployed), we finally compute equivalent incomes.

When comparing the empirical measures, the main result is that quite different people show up in the bottom decile[3] of the different well-being distributions. To highlight a few details, we find that the income poor have unsurprisingly low incomes (619 EUR per month on average compared to 1,705 EUR in the whole sample). This especially relates to a high share of unemployed individuals (33 per cent among the poor compared to 6 per cent in the entire sample), low educated (37 vs. 17.5 per cent) and single parents (12 vs 5 per cent). The composite index turns out to put a relatively large weight on unemployment so that even two thirds of the worst off are unemployed. Yet, they are better educated than the income worst off, but in worse health. The individuals with a low life satisfaction are on average almost as rich as the full sample but have a more negative attitude towards themselves than the previous two worst off groups. Yet, they are less often unemployed and better educated, but in even worse health. The equivalent income measure in particular identifies unhealthy people as worst off, which results in an overrepresentation of pensioners and people with a disability among the worst off. For the disabled, the share is almost doubled compared to the composite index (59 vs. 32 per cent) and one might wonder about this given the explicit inclusion of health in both measures. We find this to be due to the respect of preferences with the equivalent income, i.e. the people who are not in good health are also those for whom being in good health is more important. Among the different well-being measures, only equivalent incomes are able to take that into account. To bring these findings to one number, the differences in socio-demographic characteristics result in a very small overlap of the worst off according to the different measures. Namely, only 2-5 per cent of the individuals in the sample belong to both bottom deciles of any pair of measures, while only 1 per cent is identified to be worst off according to all four measures.

In sum, our illustration shows once more that measurement matters. Clearly, it remains an open question, for instance, how generalizable these …findings are. Yet, they have potentially far-reaching policy implications when it comes to the design of social policies that involve some kind of targeting or redistribution. If the consensus is that well-being is a multidimensional notion and if policy makers want to work with a single well-being measure to take account of the potential correlation between the dimensions, the crucial questions at hand are (i) which dimensions of life to take into account, and (ii) how to measure and aggregate them. Besides the need for further research to be carried out (based on the availability of high quality data), this especially entails important value judgments. And a systematic normative debate on this is yet to come.


References:

Benjamin, D. J., O. Heffetz, M. S. Kimball, and N. Szembrot (2014). Beyond happiness and satisfaction: toward well-being indices based on stated preference, American Economic Review, 104(9): 2698-2735

Decancq, K.,  M. Fleurbaey and E. Schokkaert (frth.). Happiness, equivalent incomes, and respect for individual preferences. Economica ”

Decancq, K. and Neumann, D. (2014). Does the choice of well-being measure matter empirically? An illustration with German data. CORE Discussion Paper 2014/50

Diener, E. (2000). Subjective well-being: the science of happiness and a proposal for a national index. American Psychologist, 55(1): 34-43

Fleurbaey, M. (2011). Willingness-to-pay and the equivalence approach. Revue d’Économie Politique 121(1): 35-58

Fleurbaey, M., and D. Blanchet (2013). Beyond GDP: measuring welfare and assessing sustainability. Oxford University Press

Frey, B. and A. Stutzer (2002). Happiness and economics: how the economy and institutions affect human well-being. Princeton University Press, Princeton

Kahneman, D. and A. B. Krueger (2006). Developments in the Measurement of Subjective Well-Being. Journal of Economic Perspectives, 20(1): 3-24

Layard, R. (2005). Happiness: lessons from a new science. The Penguin Press, UK

Ravallion, M. (2011). On multidimensional indices of poverty. Journal of Economic Inequality, 9(2): 235-248

Stiglitz, J. E., A. Sen and J.-P. Fitoussi (2010). Mismeasuring our lives: why GDP doesn’t add up. New Press, New York

[1] From a technical point of view, such a measure is very similar to the measures proposed in the literature on multidimensional poverty, which are also built on an objective view on how to aggregate across the different dimensions of poverty. Though, in our paper we do not provide a poverty analysis but consider mere rankings of different well-being measures, focusing on the individuals at the bottom of these rankings. Contrary to both, the literature on poverty “dashboards” favors separate measures for different dimensions without aggregating them (see Ravallion (2011) for a discussion).

[2] In economics, ‘paternalistic’ refers to a characterization of measures and policies which limit the individual’s liberty or autonomy in defining its own good. Rather, the state or government is assumed to know and define what this good is, and to design policies accordingly.

[3] The bottom decile refers to the 10 per cent of observations with the lowest values in a given distribution, e.g. of incomes.

Advertisements
This entry was posted in Poverty, Scientific Papers, Welfare. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s