RELATIVIST WELFARE MEASUREMENT ACCOUNTING FOR COUNTRY-SPECIFIC PREFERENCES IN INTERNATIONAL WELFARE COMPARISONS Master’s Thesis Economics Tilburg University 30 August 2010 D.C.W.M. (Dingeman) Wiertz, 209434 Supervisors: Prof.dr. J.A. Smulders Prof.dr. A.B.T.M. Van Schaik Number of words: approximately 30 000
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
RELATIVIST WELFARE MEASUREMENT
ACCOUNTING FOR COUNTRY-SPECIFIC PREFERENCES
IN INTERNATIONAL WELFARE COMPARISONS
Master’s Thesis Economics
Tilburg University
30 August 2010
D.C.W.M. (Dingeman) Wiertz, 209434
Supervisors:
Prof.dr. J.A. Smulders
Prof.dr. A.B.T.M. Van Schaik
Number of words: approximately 30 000
2
ABSTRACT
Over the past few decades, there is a renewed interest for welfare measurement methods.
Recognizing the limitations of GDP per capita for accurately representing welfare, a
number of alternative measurement methods have been proposed, ranging from
seemingly ‘objective’ indicators like the Human Development Index to subjective well-
being indicators like happiness and life satisfaction. As this thesis argues, however, most
of these proposals are far from satisfactory either: whereas the ‘objective’ indicators pay
too little attention to subjective preferences and perceptions, the subjective indices seem
rather insensitive to objective living conditions. In response to these observations, this
thesis favours a synthesis of both approaches, focusing on both objective living
conditions as well as subjective preferences. Considering recent attempts at sophisticated,
comprehensive international welfare comparisons, nevertheless, hardly any attention is
paid to potential cross-country variation in preferences towards various welfare
dimensions like inequality, leisure, health and the risk of unemployment. Therefore, this
thesis aims to investigate the opportunities for and potential implications of taking into
account country-specific preferences towards various welfare dimensions in international
welfare comparisons. For this purpose, amongst others regression analyses and principal
component analyses on the basis of the World Values Survey are conducted. Despite
facing many difficulties and suffering from several shortcomings, this thesis strongly
demonstrates the potential importance of incorporating country-specific preferences in
international welfare comparisons. Consequently, it concludes that if one aims at
sophisticated, comprehensive international welfare comparisons, one should take into
consideration potential cross-country variation in preferences towards various welfare
dimensions.
3
Contents
1 Introduction page 5
2 Approaches to welfare and its measurement page 10
2.1 Gross Domestic Product (GDP) and similar measures page 11
2.2 Alternative ‘objective’ indicators page 11
2.3 Welfarism and subjective well-being page 15
2.4 Hypothesis page 21
2.5 Equivalent income method page 23
3 Equivalent incomes and country-specific preferences for
inequality and leisure page 25
3.1 Equivalent incomes in Fleurbaey & Gaulier (2009) page 25
4.3 Comparing our results with Fleurbaey et al.’s (2009) results page 88
4.4 Final words page 92
5 Conclusion page 94
4
Appendices page 99
Appendix A Details on the variables used in the regressions
of Chapter 3 and 4 page 99
Appendix B Correlation matrix accompanying Table 3 page 104
Appendix C Life satisfaction regressions containing
regime-specific inequality effects page 105
Appendix D Average number of hours worked per country (2004) page 106
Appendix E Explanatory list of country abbreviations page 107
References page 108
Acknowledgements and contact information page 111
5
I Introduction
“The valuable capacity of the human mind to simplify a complex situation in a
compact characterization becomes dangerous when not controlled in terms of definitely
stated criteria. With quantitative measurements especially, the definiteness of the result
suggests, often misleadingly, a precision and simplicity in the outlines of the object
measured. Measurements of national income are subject to this type of illusion and
resulting abuse, especially since they deal with matters that are the center of conflict of
opposing social groups where the effectiveness of an argument is often contingent upon
oversimplification…(…)…The welfare of a nation can scarcely be inferred from a
measurement of national income.”
(Simon Kuznets, 19341)
Despite its length, this quotation represents a telling and meaningful starting-point
for this thesis, which immediately takes us to our core subject: the measurement of
welfare. The interest for welfare has almost always existed and can be traced back to
people like Jeremy Bentham, Adam Smith and even Aristotle. Public policy has always
been considered as a means to enhance a society’s welfare and well-being. However, a
strong economic and statistical foundation of such public policies has long been absent
over the course of history. During the twentieth-century Great Depression people became
really aware of the problems of such a lack of economic foundation: politicians and
policy makers found it difficult to navigate the economy without a proper compass. In
response, Simon Kuznets was commissioned by the US government to develop a system
of national accounts.
In a 1934 report to the US Congress in which he presented his proposal for the
measurement of national income (which later became the basis of our current systems of
national accounts), Kuznets made amongst others the remarks quoted above. The
quotation clearly demonstrates that Kuznets was fully aware of the limitations of his
national income measure and that he warned for inappropriate use and interpretation of
his measure. Nevertheless, these warnings did not prevent Gross Domestic Product
(GDP) of becoming world’s most dominant indicator of economic performance, often
used as a core indicator on the effectiveness of public policy and as a measure of broader
well-being.
However, critique on the GDP measure and its applications has never vanished
completely. Van den Bergh (2005, 2009), for instance, presents an ardent plea for the
abolishment of GDP as an indicator of macroeconomic policy. To underpin his position,
Van den Bergh discusses a long list of arguments. Amongst others, he criticizes GDP for
insufficiently taking into account the distribution of income as well as for neglecting the
informal economy. As a result, GDP is an inappropriate measure of welfare in his
opinion. The fact that GDP is nonetheless an important decision criterion in politics,
financial markets and international organisations implies according to Van den Bergh a
1 National Income 1929-1932. Report to the 73rd US Congress, 2nd session, Senate document 124, pp. 5-7.
6
serious information failure. A similar conclusion is drawn by the initiative of The
Declaration of Tilburg (‘De Verklaring van Tilburg’, 2008), a petition stating that GDP is
merely a speedometer reflecting how quickly we are earning money, while we actually
need altimeters containing information on the sustainability and solidarity of society.
Some recent examples further support Van den Bergh’s (2005, 2009) thesis. The recent
BP disaster in the Gulf of Mexico might, for example, very well translate into an increase
in GDP (as a consequence of all the necessary effort and expenditures to clean the mess),
whereas the environment obviously experiences a big loss, which does affect welfare, but
which is not accounted for in GDP (neither is accounted for the worsened future
prospects of the fishery sector in the region).
If we for the moment agree that the imperfections of GDP for measuring welfare
and well-being are so serious that we should abandon GDP as a welfare indicator, we
instantaneously arrive at the question of what alternative measure should be used. After
all, as the Dutch proverb says, should you throw away your old shoes before you have
new ones?
Van den Bergh (2005, 2009) seems to be in favour of a subjective well-being
approach for measuring welfare, in line with Layard (2005). It is, however, questionable
whether subjective well-being is a desirable measure for welfare and well-being. In this
context, Amartya Sen often gives the example that even though some happy slaves have
existed, slavery certainly does not mean well-being.2 Heertje (2007) is also critical
towards the use of subjective well-being measures as foundation for economic policy,
since many significant determinants of happiness are outside the domain of the
economist. Moreover, according to Heertje we should not even strive for an all-
embracing measure of well-being, as such a measure is just a utopia; we have to accept
that certain valuable things are simply not properly quantifiable.
Heertje’s work points at a tension within economic science: on the one hand,
economists should be careful to cross the borders of economic science too
enthusiastically, trying to assign a value to social issues and personal feelings that cannot
be measured that easily, while on the other hand, the dismantlement of interdisciplinary
barriers can indeed lead to valuable progress in the measurement of welfare and well-
being.
No matter what our position is in this regard, we can observe an increased interest
in broader measures of welfare and well-being over the past few years. A notable
example is the report by the Commission on the Measurement of Economic Performance
and Social Progress (Stiglitz et al., 2009), which was written by order of the French
president Nicholas Sarkozy. Furthermore, organisations like the World Bank, the OECD
and the European Union have also started projects in this field (Canoy & Lerais, 2007;
Fleurbaey, 2009). A frequent discussion at these platforms concerns the issue of
comprehensiveness versus comprehensibility. The less complex the measures are, the
easier they are to communicate and to understand, while these simple measures may give
a less accurate overview of affairs as compared with more comprehensive and more
complex measures. In a similar vein, one could also think of a dilemma of inclusiveness
versus measurability. The existence of such dilemmas implies that one always faces
tradeoffs when constructing a welfare indicator. Hence, the appropriateness of a welfare
2 See for example Sen (2001).
7
measure is dependent on its intended application: e.g. whereas GDP probably is an
insufficient indicator if one is aiming at an all-embracing, broad welfare comparison, it
might still be very useful in a number of concrete policy contexts due to its high level of
measurability.
For now we want to concentrate our attention on a recent attempt of Cnossen
(2009) at a broad welfare comparison of a group of European countries. Cnossen’s
primary goal is to address whether comprehensive welfare states like the Netherlands are
effective in achieving higher levels of welfare through their extensive public policies.
What he is actually doing, is making cross-country comparison tables for a wide array of
welfare-related subjects, ranging from basic dimensions like inequality and poverty to
factors like education, discrimination, competitiveness, social trust, sustainability et
cetera. On the basis of these tables he concludes that in terms of welfare the Netherlands
but also the Nordic countries do not suffer from their high levels of tax pressure.
Unfortunately, however, Cnossen’s study suffers from a number of shortcomings.
One first criticism is that although Cnossen’s comparison tables do not allow him to
establish any causal links, he repeatedly suggests that more extensive welfare state
policies lead to higher levels of welfare. Second, with his comparison tables it is not
possible to trade-off scores on different dimensions, since a weighting scheme of the
different dimensions is lacking. Therefore, it is in fact impossible to derive a conclusive
welfare ranking of countries on the basis of his tables. Obviously, this is a severe
drawback of Cnossen’s research and, as a result, his conclusions lose a lot of their
convincing power.
A much more promising welfare ranking of countries is provided in Fleurbaey &
Gaulier (2009). These authors have created a ranking of 24 OECD countries based on
equivalent incomes. They have corrected standard GDP per capita for amongst others
leisure, health, risk of unemployment and income inequalities. Because of their
equivalent income approach, they arrive at a one-dimensional ranking and they are able
to weight several welfare dimensions against each other, mainly by way of market prices.
Nevertheless, there is still a third criticism on Cnossen (2009), for which also
Fleurbaey & Gaulier (2009) do not provide a satisfactory solution. The central idea
behind this criticism is that welfare is a relative concept, and that countries can adhere to
different definitions of welfare. Therefore, preferences regarding different welfare
dimensions may very well differ across countries, depending on culture, social context et
cetera. Alesina et al. (2004) show for example that preferences towards inequality differ
between the United States and Europe. If one does not include such preferences in one’s
welfare analyses, the resulting rankings may create misleading impressions. Apart from
shortly mentioning that welfare is a relative concept about which people and countries
may have differences of opinion, Cnossen (2009), however, does not pay any attention at
all to this issue. Instead, at several occasions in his text it seems that he has implicitly
adopted the Dutch preferences concerning welfare and a ‘civilized’ society. Yet, as we
have just made clear, it might for example not be fair to evaluate Italy’s welfare on the
basis of Dutch preferences. Fleurbaey & Gaulier (2009) score also in this respect
somewhat better than Cnossen. By applying price-based corrections to GDP per capita,
the different welfare dimensions are valued differently across countries in their analysis.
The question remains, however, to what extent these imputed valuations correspond to a
8
country’s real preferences. Leisure, for instance, is valued against the average hourly
wage in a country in Fleurbaey & Gaulier (2009). While wages do indeed contain some
information on preferences for leisure (in accordance with the ‘reservation wage’
concept), variation in wages across countries is affected by variation in productivity and
other macroeconomic conditions across countries as well. In addition, Fleurbaey and
Gaulier use several preference-related parameters in their analysis (related to risk
aversion, inequality aversion and time preference) for which they assume universal
values for all countries they consider. So, from the perspective of country-specific
welfare preferences the work of Fleurbaey & Gaulier (2009) is not perfect either.
Since, as far as we know, there are no other studies available which succeed in
satisfactorily taking into account country-specific preferences, the aim of this thesis will
be to take up this issue and to explore the possibilities and implications of including
country-specific preferences in international welfare comparisons. The central research
question can thus be formulated as: What are the implications of including country-
specific preferences towards different welfare dimensions in cross-country welfare
comparisons?
Regarding this research question, it is relevant to remark that preferences may not
only vary across space, but also across time and individuals. Nevertheless, we have
chosen to demarcate our research problem to country-specific preferences, as we think
this is the most relevant dimension of variation from the viewpoint of international
welfare comparisons. Moreover, we would like to remark that we are not under the
illusion that this thesis can come up with some kind of superior welfare measure which
will resolve the debate on the measurement of welfare. As we noted earlier, the
appropriateness of a welfare measure strongly depends on its intended use. We recognize
that, by definition, every welfare indicator has some strong and some weak properties
and, besides, that a great leap forward is not easily attainable, were it only for reasons of
limited data availability concerning preferences. Instead, this thesis should be read as an
exploratory expedition concerning the relationship between the measurement of ‘the
wealth of nations’ and country-specific preferences towards various welfare dimensions.
During this expedition we hope to find answers to questions like: How can we include
country-specific preferences in welfare measures? Does this inclusion make a significant
difference for international welfare rankings? How can we proceed in improving our
measures of welfare? Important to note in this context is that, rather than contributing to
the vivid theoretical debate on proper welfare measures, this thesis mainly wants to add
to the empirical approximation of welfare.
Apart from the intrinsic value of gaining more sophisticated insights into the
wealth of nations, this thesis can hopefully help attenuating the information failure Van
den Bergh (2005, 2009) was pointing at.
This thesis proceeds as follows. Chapter two presents the theoretical background
of our study. In doing so, we first discuss our definition of welfare. Subsequently, we go
over some criticisms towards conventional welfare measures and finally we deal with
some alternative approaches, amongst others the subjective well-being literature. Chapter
three focuses on the earlier mentioned equivalent income measure of Fleurbaey &
9
Gaulier (2009), which is based on a welfare / utility maximization model and market-
based income corrections. In addition to explaining their method, several attempts of
more satisfactorily taking into account country-specific preferences are presented. In
these attempts, survey data from the World Values Survey play an important role. Next,
chapter four deals with another method of calculating equivalent incomes, in line with the
work of Fleurbaey et al. (2009). In this method, there is a leading role for willingness-to-
pay estimates inferred from life satisfaction regressions. Finally, chapter five concludes.
10
II Approaches to welfare and its measurement
Together with scarcity, welfare is perhaps the most central concept within
economic science. Almost everywhere around us we are confronted with scarcities of
resources, implying that we cannot have everything that we would desire. The resulting
fundamental decision problem is how to manage these scarce resources to attain welfare.
From this perspective, Mankiw & Taylor (2006) define economics as the study of how
society manages its scarce resources. In essence, most economic problems can in some
way be reduced to a welfare maximization problem under scarcity constraints.
Considering all this, it is obviously important to understand what is meant by the term
welfare.
Although the nature and definition of welfare is far from undisputed, there is a
wide consensus that it has something to do with the satisfaction of needs. However,
consensus stops already here, as a variety of opinions exists regarding questions like:
What should be considered as needs? What do we mean by the satisfaction of needs?
How should different kinds of needs be weighted? Et cetera.
In this thesis our starting-point will be a rather general notion of welfare,
following amongst others Heertje (2007). We adopt the view that welfare concerns the
satisfaction of needs and we stress that welfare should be seen as a subjective concept
without any concrete or predefined content. The subjective character of our notion
implies that welfare is dependent on one’s preferences. It is impossible to define a
universal, conclusive list of needs and to judge everyone’s welfare on this basis. Rather,
one’s needs and the degree to which they are satisfied depend on subjective preferences
and perceptions. Therefore, welfare has no concrete content: everything of which
individuals think it contributes to their satisfaction of needs is part of the welfare concept.
Thus, the answer to the question ‘what constitutes welfare?’ may differ from time to time,
from space to space and from individual to individual.
From this observation it follows that it is neither practically possible nor desirable
to aim for a single, all-inclusive measure of welfare that can act as a guideline for
policymaking. To quote the Commission on the Measurement of Economic Performance
and Social Progress (Stiglitz et al., 2009:207): “The search for an aggregate measure of
the quality of life that combines information across all its dimensions is often perceived
as the ‘holy grail’ of all efforts to go beyond conventional economic measures. This
perspective is, however, both limited and deceptive.” Nonetheless, these remarks should
mainly be interpreted as a warning against too high expectations regarding a welfare
measure as well as against ‘overinterpretation’ of such measures. Thus, these remarks do
not undermine the potential relevance of broader welfare measures in terms of richer
insights into the wealth of nations and in terms of attenuation of the information failure
related to the use of traditional welfare indicators like GDP per capita.
In the remainder of this chapter, we will discuss several (measurement)
approaches to the concept of welfare, after which we will posit the central hypothesis of
this thesis.
11
2.1 Gross Domestic Product (GDP) and similar measures
As already stated in our introductory chapter, Gross Domestic Product (GDP) per
capita has been the most widely used indicator of welfare over the past century.
According to the textbook definition, GDP is the added value of all goods and services
produced within a country in a given period of time, measured at market prices (see for
example Mankiw & Taylor, 2006:468). As such, it tries to measure the flow of economic
activity during a period. Usually, attention is mainly paid to GDP per capita, to allow for
international comparisons across differently-sized countries. Starting from GDP per
capita, one could also calculate several related indicators like Gross National Product
(GNP) per capita (to include income flows across borders) and Net National Product
(NNP) per capita (to account for depreciation of fixed assets).
The most important advantage of GDP per capita is that it is an indicator which
can relatively easily be calculated as well as communicated. Moreover, while it is widely
accepted that GDP per capita is not the ideal measure of welfare, it is assumed that there
is a large and significant correlation between GDP per capita and other welfare
dimensions (e.g. health, education, poverty). Hence, GDP per capita is often considered
to be an acceptable indicator of welfare.
Nevertheless, GDP per capita can be seriously criticized from a welfare point of
view. Its main drawback is that it only measures market transactions; only products and
services which are traded on markets are included in GDP calculations, weighted at their
market price. Obviously, however, many valuable objects exist which are not traded on
markets and therefore not captured by GDP calculations (or only imperfectly). Notable
examples entail amongst others the environment, household production, leisure, health,
government services, the distribution of income and opportunities, quality of governance,
social interaction, education, freedom, rights and risks. Since no (or only imperfect)
market prices exist for these issues, ordinary GDP measures neglect them and might,
therefore, lead to efficiency losses due to underinvestment in these issues. Furthermore,
whereas GDP is by its nature a flow measure, it is far from inconceivable that stocks also
matter for welfare, for instance via the perspective of a society’s future prospects. Yet,
GDP does not provide any insights into stocks of environmental, human and social
capital et cetera. Finally, although proponents of GDP as a welfare indicator often defend
the measure on the basis of its rather neutral and objective character, this defense cannot
be accepted, since the presumption that market prices provide a proper tool for measuring
the value of production is already a value judgment in itself. Especially when markets are
imperfectly competitive and when externalities are present, this value judgment may be
problematic, because under these circumstances market prices are likely to hide society’s
true valuation. In conclusion, notwithstanding the relevance of GDP measures in certain
contexts, it is highly debatable whether GDP per capita satisfactorily captures the welfare
within a country as defined above.
2.2 Alternative ‘objective’ indicators
Recognizing the limitations of GDP per capita for ‘truly’ measuring the welfare
within a country, there have been many initiatives over the past few decades to introduce
12
an indicator that could replace GDP per capita.3 Basically, these initiatives have sought to
come up with a measure that should not only take into account market transactions but
also other valuable aspects of life. In most of the cases, the goal was to create a
composite index that combines several domain indicators of economic, social and
environmental performance and progress.
Many different composite indexes have seen the light over the years and, by now,
it would not be difficult to present a list of several dozens of such indicators (see for
instance Decancq & Lugo (2010) for a selection). The reason for this abundance of
indicators is rather simple in our opinion: there can be as many indicators of welfare and
the quality of life as there are definitions of what constitutes welfare and quality of life.
Recall in this respect also our statement earlier in this chapter that we consider welfare as
a subjective concept without concrete content, which can be defined in many ways
depending on one’s preferences, normative views and social context. Keeping this idea in
mind, the abundance of composite welfare indexes comes as no surprise. Here one can
also clearly discover the truth behind Stiglitz et al.’s (2009) words that the search for an
all-comprising aggregate measure of welfare is both a limited as well as a deceptive
expedition: someone with other views and preferences can always prefer another index
and, therefore, it is practically impossible to ever reach consensus on a single composite
index. Does this then mean that composite welfare indicators have no value at all?
The answer to this question would still be a definite ‘no’. Although one should
always remember that every indicator just represents one normative view on welfare, all
indicators can provide interesting information and useful refinements as compared with
the standard GDP per capita measure. As Todaro & Smith (2009) observe, most
composite indexes have a strong tendency to rise with per capita income, as richer
countries simply have more financial resources to invest in things like health and
education. These authors point out, however, that despite this expected pattern, there are
still large deviations between income and broader measures of welfare. So, from this
perspective, looking at more variables than only income can indeed provide valuable
additional insights. Becker et al. (2005), for instance, show that the absence of income
convergence generally noticed in the growth literature is in stark contrast with the
reduction in cross-country inequality over time that can be observed after incorporating
the relatively large recent gains in life expectancy in the poorer countries. Moreover, in
addition to these advantages in terms of additional insights, most of the composite
indicators still have the main advantages of the GDP per capita welfare measure, as they
are in general easy to calculate as well as easy to communicate to the public.
As far as these alternative composite indicators have any theoretical basis, it is to
be found in Amartya Sen’s conceptual framework of capabilities and functionings.4 In
short, in this framework a person’s life is conceived as a combination of various ‘doings
and beings’: functionings. These functionings can be interpreted as a collection of the
3 As Stiglitz et al. (2009) note, this social indicator movement was particularly active in the 1960s and
1970s. In 1974, a special journal was even founded, Social Indicators Research, to publish research dealing
with the measurement of the quality of life. The journal still exists and is still a quite popular forum for the
exchange of ideas on social performance and progress, in particular among sociologists, but to a lesser
extent also among economists, anthropologists and political scientists. 4 A more comprehensive overview of this conceptual approach can be found in Schokkaert (2007) and
Alkire (2008).
13
observable achievements of a person, ranging from being nourished to being able to
express oneself in public without shame. Because people’s values and experiences differ
across time and space, the list of most relevant functionings depends on contextual
circumstances. What ultimately matters according to Sen, is one’s capabilities: the
opportunities one has for choosing among the various combinations of these functionings.
Thus, at the core of the capabilities approach is the idea that welfare is determined by two
things: one’s objective living conditions and one’s freedom to choose these conditions.
Although subjective feelings can also be part of one’s functionings, the capability
approach emphasizes that people are likely to adapt to their living conditions and that,
therefore, subjective feelings are inadequate as the only measure of welfare.
Regarding the measurement of welfare, the capabilities approach faces actually
one major problem, namely the difficulty to derive concrete measures from its concepts.
In the first place, there is the issue whether we should try to measure functionings or
capabilities. Obviously, functionings are far more easily observable, but on the other
hand, capabilities are what ultimately matters according to Sen. Second, aiming for a
single welfare index, the capabilities approach also entails an indexing problem: how
should the different functionings and capabilities be weighted? We will discuss this
indexing problem at length in a few moments, after presenting a more general
classification framework for composite welfare indicators.
Decancq & Lugo (2010) provide an interesting classification and conceptual
overview of composite welfare indicators. They classify the different indicator proposals
using three criteria. First, they consider the transformation functions applied to the
underlying domain indicators, looking whether the underlying variables are rescaled,
linearly transformed, exponentially transformed, et cetera. These transformation
functions are applied to transform scores on different domain indicators, which are often
measured in different measurement units, to a common basis for aggregation.
Furthermore, transformation functions can also help correcting for outliers in the original
domain indicator distributions.
The second classification criterion considered by Decancq & Lugo (2010) is the
assumed elasticity of substitution between the different welfare domains. The central
question in this respect is how, for example, one additional year of life expectancy can be
traded off against a decrease in the education level within society. If the different
components of the indicator are assumed to be perfect substitutes, a decrease in one of the
domain indicators can perfectly and rather easily be compensated for by an increase in
one of the other domain indicators. On the other hand, if no substitutability between the
components is assumed, a decrease in one of the domain indicators necessarily implies a
decrease in the value of the composite welfare indicator. Thus, the assumed elasticity of
substitution between the elements of the composite index can obviously make a huge
difference.
Thirdly, Decancq & Lugo (2010) are interested in the weights attached to the
different components of the overall index. Like an index measure clearly requires
normative assumptions regarding the first two classification criteria, this is definitely also
the case for the choice of the component weights. According to Decancq & Lugo (2010),
the employed weighting scheme reflects by definition particular value judgments on how
‘a good life’ should look like. Even if one chooses for weights that can be objectively
14
defined or derived, the choice for these weights still represents value judgments, since the
question ‘why are especially these weights the appropriate ones?’ simply cannot be
answered objectively, without any normative premises. Here we are back at the indexing
problem of the capabilities approach.
In addition to these points, it is relevant to note that weighting schemes can have a
significant impact on the scores and rankings on the obtained composite index. Becker et
al. (1987) provide a striking example in this respect. They study the quality of life in 329
metropolitan areas in the United States on the basis of standard variables like economic
performance, health and security, and find that, depending on the used weighting scheme,
there are 134 cities that could be ranked first and 150 cities that could be ranked last.
Furthermore, they find that 59 of the 329 cities could be ranked either first or last, solely
dependent on the adopted weighting scheme.
Decancq & Lugo (2010) distinguish three approaches to set the weights. On the
one hand, there is the data-driven approach, which derives the component weights
statistically, on the basis of the observed distributions of scores within society on the
different domain indicators. An example of a data-driven weighting procedure is a
frequency-based weighting method. Many multidimensional deprivation indexes impose,
for instance, an inverse relation between the frequency of deprivation in a certain
dimension and the weight of that dimension, acknowledging the idea that individuals
attach a higher importance to shortfalls in dimensions where the majority of the
population does not fall short.
At the other side of the spectrum, one can distinguish the normative approach.
Instead of basing the weights on the actually observed distribution of domain scores, this
approach merely depends on explicit value judgments regarding the tradeoffs and the
priorities of the different underlying domains. Notable examples of this weighting
procedure include amongst others equal or arbitrary weights, weights based on expert
opinions, but also price-based weights (which rely on the normative assumption that
prices reflect an appropriate basis for the dimensional weights).
The final class distinguished, is a hybrid approach, which combines a data-driven
element with subjective value judgments concerning the importance of the different
domains. Although hybrid weights are hardly ever used as compared with the other
weighting schemes, the most obvious candidates for hybrid weights are stated preference
weights and hedonic weights. Stated preferences weights are directly based on the
expressed opinions of a (representative) group of individuals within society, thus being a
combination of a data-driven element as well as individual valuations. A hedonic
weighting procedure, on the other hand, tries to derive implicit valuations for different
dimensions from data on self-reported happiness and life satisfaction.
All of these weighting approaches have their strengths and weaknesses. The main
drawback of the data-driven methods is that they do not survive ‘Hume’s guillotine’. This
criterion, named after the famous eighteenth-century philosopher David Hume, states that
it is impossible to derive normative statements (about values, about what ‘ought to be’)
from descriptive statements (about facts, about what ‘is’). Conversely, however,
normative weights can be accused of paternalism, as the normative weighting procedure
is based on value judgments which may not be supported by the individuals for which the
composite index is calculated, who may prefer other value judgments and resulting
15
weights.5 Recognizing these drawbacks, Decancq & Lugo (2010:4) characterize the
process of setting weights for a multidimensional welfare index as “…a problem of
choosing between Scylla and Charybdis, between Hume’s guillotine and paternalism”.
The hybrid weighting procedures “…explore the narrow and dangerous strait of Messina
between Scylla and Charybdis”. This route can indeed be considered dangerous, as it
risks suffering from both Hume’s guillotine and paternalism. Nonetheless, this hybrid
approach also has the potential of avoiding (or at least attenuating) both problems.
One final remark regarding the weight-setting procedure is that one can also take
a less risky route by selecting wide ranges of acceptable weights instead of only one
weighting scheme. The necessary price to pay for this procedure is that the ranking of
individuals may become incomplete: it is possible that not every pair of cases can be
ranked anymore if one uses wide ranges of acceptable weights. Nonetheless, according to
Foster and Sen (1997), this does not have to be a huge problem, as one can still reach
agreed rankings in many situations.
Probably the most prominent example of a composite welfare index (and the most
serious ‘competitor’ of GDP per capita up until now) is the Human Development Index
(HDI), created in 1990 and thereafter annually published in the Human Development
Reports of the United Nations Development Programme (UNDP). Like most composite
indexes, the indicator has been inspired by the conceptual framework of the capabilities
approach. The index represents a composite measure of three functionings that are
generally thought to be important in almost everyone’s life: income, health (measured by
life expectancy) and education (measured by adult literacy rates and primary education
enrolment rates). A transformation function is applied to each dimension to rescale the
dimensional scores to a 0-1 scale. Subsequently, the HDI is calculated as the simple
arithmetic average of the three component indicators.
Despite the relatively high correlation that is generally found between GDP per
capita and HDI, the composite index evidently provides useful additional information:
countries like Sri Lanka, with relatively good health and educational performance, jump
up the welfare ladder when looking at HDI, whereas in contrast, countries like Brazil
experience a significant drop in their ranking.
However, HDI is still a measure with many deficiencies. Among other things, the
indicator assumes perfect substitutability between the included domains by simply taking
the average of the rescaled domain indicators.6 Obviously, one can easily criticize this
assumption. The same holds for the fact that the index implies an extremely large
difference in the monetary valuation of an extra year of life between rich and poor
countries, as observed by Ravallion (1997). In addition, life expectancy and primary
school enrolment rates, for example, do not provide any evidence on the ‘quality’ of
health or education. Moreover, since HDI is based on a combination of aggregate indices,
it can hide huge variation in welfare within a country, as large inequalities may exist.
5 Critics of normative weighting methods often describe these weighting methods as ‘playing God’,
deciding what is good for others, even if those people will never feel this to be so. 6 This implicit assumption of perfect substitutability is contradictory to the UNDP’s self-expressed vision:
“Progress in human development requires advances across a broad front: losses in human welfare linked to
life expectancy, for example, cannot be compensated for by gains in other areas such as income or
education.” (Human Development Report 2005, p. 22).
16
Furthermore, it is found that over the years movements in the HDI have tended to be
dominated by changes in its income component, in particular for countries whose
performance with respect to health and education is close to the top of the world. One can
question whether such a dominance of the income component is desirable (what is under
these circumstances the added value of HDI as compared with GDP per capita?). Finally,
and perhaps most importantly, the weighting method of the HDI can be severely
criticized for being arbitrary, as it simply assumes equal weights for every welfare
dimension.
2.3 Welfarism and subjective well-being
In the quest for broader welfare measures we can also identify another approach.
Instead of constructing composite indicators based on objective living conditions, this
approach focuses on subjective well-being. It connects to the long-standing tradition
within economics of utilitarian welfarism. The origin of this utilitarian welfarism can
probably best be attributed to the eighteenth-century philosopher Jeremy Bentham, who
was a strong advocate of utilitarianism as central moral principle. Bentham argued that an
act or a policy was right if it led to ‘the greatest good for the greatest number of people’.
This proposition is also known as the ‘greatest happiness principle’. According to
Bentham’s definition, happiness or utility is concerned with the hedonic flow of pleasure
and pain, what we nowadays tend to call experienced utility, following the economic
psychologist Daniel Kahneman (e.g. see Kahneman & Krueger, 2006).
Utilitarianism has dominated the history of economic science. Nowadays, most
economic models are still based on the maximization of a social welfare function, which
can be disaggregated to individual utility functions. An important shortcoming of this
utilitarian welfarist approach has long been that it lacked an empirical foundation. Utility
was mainly a theoretical concept, which could not be measured in reality. However,
according to certain economists recent advances in social science have changed this
situation, evoking claims about a ‘revolution in economics’ (Layard, 2005; Frey, 2008).
Though this claim is certainly not undisputed, many economists argue that we are now
able to measure utility on the basis of survey questions on subjective well-being.7 8 Since
their introduction in the 1970s, survey questions on subjective well-being have received
increasingly more attention and especially during the last decade happiness research has
culminated within economic science, and the influence of happiness research is spreading
within economics. For instance, findings from happiness research have already entered
the literature on economic growth (e.g. see Strulik, 2008; Kawamoto, 2009; Valente,
2009). The ‘measurability’ of happiness has even induced some calls for the introduction
7 The widely used World Values Survey contains for example the following questions: “Taking all things
together, would you say you are very happy, quite happy, not very happy, not at all happy?”, “All things
considered, how satisfied are you with your life as a whole these days: 1-dissatisfied,…,10-satisfied?”.
Next to the World Values Survey, there are many other surveys that ask quite similar questions, for
example the rather new Gallup World Poll, the General Social Survey for the United States, the
Eurobarometer Survey Series for Europe and the Russian Longitudinal Monitoring Survey. 8 An even more recent ‘solution’ for measuring utility can be found in the field of neuroeconomics. Being a
synthesis of economics, psychology and neuroscience, this area of study tries, among other things, to obtain
information on experienced pleasure and pain by analyzing people’s brain activity using electrodes.
17
of Bentham’s ‘greatest happiness principle’ as key directive for policy evaluations.
Amongst others, Diener (1990) and Layard (2005) have proposed the replacement of
GDP per capita by a national happiness indicator. The Kingdom of Bhutan already uses
‘Gross National Happiness’ as its core development indicator and particularly England,
Australia and New Zealand are also developing a system of national well-being accounts.
Nonetheless, we should make clear that initiatives in this direction also encounter
much critique and opposition. One of the strongest opponents is, not surprisingly,
Amartya Sen. Basically, he has two main arguments against welfarism in general and
against the use of the ‘greatest happiness principle’ as a foundation for policymaking in
particular.9
First of all, Sen raises the issue of ‘physical-condition neglect’. The idea behind
this issue is that welfarism and happiness indicators mainly focus on subjective feelings
and mental states, and that insufficient attention is paid to one’s actual, objective living
conditions. In this respect, Sen (2008:21) provides amongst others the following
example: “A person who is ill-fed, undernourished, unsheltered and ill can still be high
up in the scale of happiness or desire-fulfilment if he or she has learned to have ‘realistic’
desires and to take pleasure in small mercies”. A similar example could be provided for a
rich man who experiences an improvement in his objective living conditions, but an even
higher increase in his aspirations level; his happiness score then probably deteriorates
(perhaps even under the level of the ill-fed, undernourished, unsheltered and ill person),
while it is somewhat difficult to consider him really worse-off.
The second issue put forward by Sen is the issue of ‘valuation neglect’, which is
about the nature of valuating activities. In Sen’s perception, ‘being happy’ or ‘desiring’ is
different from ‘valuating’, which is a more reflective activity and which Sen thinks to be
more related to the notion of welfare. In this regard, there may also be a discrepancy
between preferences (what one considers valuable) and subjective well-being
(satisfaction or happiness derived from what one has). Quoting Fleurbaey (2008:25), who
paraphrases the nineteenth-century philosopher John Stuart Mill: “It is better to be a rich
dissatisfied than a poor satisfied, if the rich and the poor both prefer the former’s life.” In
other words, satisfaction welfarism does not distinguish between ‘obtaining what one
wants’ (what welfare is about) and ‘being satisfied’ according to this second criticism.
At this point in the discussion, it is relevant to make a distinction between three
aspects that matter for subjective well-being, following Diener (1984): life satisfaction,
positive affect and negative affect. Life satisfaction is a cognitive concept, representing
individuals’ overall evaluation of their lives at a particular point in time. On the other
hand, positive and negative affects represent respectively positive and negative emotions,
which flow constantly through people’s minds. Though it is quite difficult to isolate these
different aspects in survey questions, questions on ‘life satisfaction’ tend to capture
mostly the cognitive part of subjective well-being, whereas questions regarding
‘happiness’ are more sensitive to affects.
Albeit all three aspects have their influence on subjective well-being, the mutual
correlation between affects and life satisfaction is rather low, as shown by Krueger &
Schkade (2008).There is in this regard some debate on what aspect of subjective well-
9 For an extensive discussion of these arguments, see for example Sen (2008).
18
being one should preferably measure. On the one hand, it may be argued that the
cognitive evaluation behind life satisfaction data provides more information on a person’s
(perceived) welfare and standard of living. On the other hand, Kahneman & Krueger
(2006) contend that information regarding experienced affects may provide a more
accurate reflection of a person’s well-being, less prone to reporting biases as compared
with life satisfaction questions. Therefore, these authors propose the ‘U-index’ for
measuring subjective well-being: a misery index based on the time spent in a negative
mood.
To get to grips with what is behind survey answers to subjective well-being
questions, we think it is illuminating to refer to Fleurbaey et al. (2009). In these authors’
conceptual framework, an individual’s expressed life satisfaction depends on four factors:
the person’s achieved functionings, her valuation ordering of all functionings in general
(her preferences regarding the various functionings), her framework of reference and a
disturbance factor.
The inclusion of the first two factors is rather obvious. The achieved functionings
factor describes how the respondent’s life looks like. The valuation ordering factor, in
turn, tells how the ‘good life’ looks like according to the respondent. It is probably wise,
however, to discuss the third factor a bit more detailed, since one of the most consistent
findings of the literature on subjective well-being is that respondents’ answers are highly
dependent on the respondent’s reference framework. As determinants of this framework,
two issues particularly stand out.
First, one uses his personal history as reference level. According to one theory,
every individual has a genetically-established personality, with a given ‘set point’ for
subjective well-being. Under this scenario, changes in the respondent’s achieved
functionings lead only to a temporal gain in subjective well-being. After a while, the
respondent gets used to his higher level of achieved functionings and his subjective well-
being level drops back to its natural ‘set point’. More likely, however, is a scenario of
partial adaptation, under which the respondent’s subjective well-being does not
necessarily fall back completely to its initial level. Anyway, it is obvious that this
‘hedonic treadmill’, as it is often called (e.g. see Kahneman, 2008), makes subjective
well-being somewhat immune to the person’s objective living conditions (his
functionings). In a similar vein, there may also be an ‘aspirational treadmill’ (again, see
Kahneman, 2008), representing the idea that once people have attained a certain level of
objective living conditions, they instantaneously and ‘automatically’ increase their
aspiration level.
The second important determinant of one’s reference framework originates from
relative interpersonal comparisons and peer effects. The underlying reasoning is that
every individual has a natural tendency to compare his situation with the situation of
people around him (living in the same street, having the same job, being member of the
same sports club, et cetera) and to strive for status. From this perspective, then,
improvements in one’s achieved functionings especially lead to an increase in the
person’s subjective well-being if his position compared to his reference group
ameliorates. In contrast, if the achieved functionings of the members of his reference
group improve to the same extent, the person’s subjective well-being does not increase or
only slightly. In spite of the fact that the debate on the extent to which relative
19
comparisons matter is still far from closed10
, there is nonetheless a certain consensus that
‘keeping up with the Joneses’ makes people somewhat immune to their objective living
conditions.
The fourth and remaining factor influencing people’s expressed subjective well-
being distinguished by Fleurbaey et al. (2009) is a disturbance factor. This factor can
possibly represent measurement or reporting biases, resulting from the fact that people
are not given enough time to reflect, from the fact that their answers are influenced by
their mood of the day or the moment11
, from framing effects related to the ordering of the
questions asked, from misinterpretation of the questions, from a feeling of duty to give a
certain answer, et cetera. Moreover, the disturbance factor can also capture some
remaining personality-12
, culture- or country-fixed effects, e.g. if every respondent in
Spain tends to give a more ‘rosy’ answer to every question as compared with respondents
in Japan. Such fixed effects stem from the fact that heterogeneous standards are used in
answering the survey questions: different individuals and cultures may use the answering
scales differently. The resulting fixed effects are expected to play a significant role in
answers to subjective well-being questions, seriously complicating cross-country and
cross-cultural comparisons of survey answers.
Over the past few decades, happiness research has been focused on investigating
the determinants of subjective well-being. By now, there is a vast empirical literature on
this topic. Although this literature in general faces difficulties concerning the distinction
between correlation and causation, some robust and intuitively appealing results stand
out. First and foremost, there is almost unanimity about the high human costs related to
unemployment (see amongst others Frey, 2008 and Stiglitz et al., 2009). Even after
controlling for income, unemployment still has a large and significant negative impact on
subjective well-being. Similarly, there tends to be a positive and robust relationship
between perceived health and reported well-being. In addition, research has also revealed
some clear-cut demographical patterns: people who are married or live together as a
couple tend to report significantly higher subjective well-being levels, the relationship
between age and subjective well-being is generally found to display a U-shaped pattern
and having kids tends to increase reported well-being.
Yet, on one of the most interesting potential determinants of well-being (at least
from an economic point of view) the jury is still out. Even though the relationship
10
Compare for example Layard (2005) with Stevenson & Wolfers (2008): whereas Layard states that it is
all about relative incomes and that absolute incomes hardly matter for subjective well-being, Stevenson and
Wolfers contend that subjective well-being is mainly affected by absolute incomes and that relative
incomes play a significantly less prominent role than argued by Layard. 11
Famous in this context is the experiment conducted by Schwarz (1987), showing the power of
experimental context. In Schwarz’s experiment, subjects were invited to the lab to fill in a questionnaire on
life satisfaction. However, when the subjects arrived at the lab, they were first asked to make a photocopy
for the experimenter. For a randomly chosen half of the subjects a dime (a coin of hardly any value) was
placed on the copy machine. When the subjects came back to the lab room and filled in the questionnaire, it
turned out that the subjects who had found a dime on the copy machine reported significantly higher life
satisfaction, a result which of course cannot be explained by the monetary value of the coin they found. 12
Regarding the importance of personality-effects, Kahneman & Krueger (2006:8) state: “In any event,
measures of temperament and personality typically account for much more of the variance of reported life
satisfaction than do life circumstances…(…)... Apparently, a person’s subjective evaluation of his or her
own well-being is to a significant extent a personality trait.”
20
between subjective well-being and income has been studied since the moment subjective
well-being data have become available about forty years ago, the debate about this
relationship is certainly not over yet.
The main catalyst of this debate has been the seminal contribution of Easterlin
(1974). In his paper, Easterlin studies three kinds of income-happiness relationships:
within countries, among countries and across time. Whereas he finds evidence for a
positive relationship within a country at one point time, he finds insufficient evidence for
such a relationship among countries or over time. In his conclusion, Easterlin (1974:119)
concisely sets the stage for the debate that has continued since his publication: “In a
sense, these results are a testimony to the adaptability of mankind. Income and
aspirations in time and space tend to go together, and people seemingly can make
something out of what appears, in some absolute sense, to be a sorry lot. At the same
time, the conclusions raise serious questions about the goals and prospective efficacy of
much social policy.”13
Over time, there have been several updates of Easterlin’s findings, based on
newly available data. The current state of the art is, for example, well-summarized in
Stevenson & Wolfers (2008) and Clark et al. (2008). Within countries, the positive
income-well-being relationship is still standing (since people’s reference groups are still
largely restricted by national borders, income-related status effects are most clearly
observable within countries) and it has now also become clear that one can observe a
positive relationship among countries, depending on the countries included in one’s
analysis: if both developing as well as developed countries are taken into consideration,
there is a significant positive relationship, whereas within more homogeneous groups of
countries there is only a weak or even no relationship observable.
Anyway, probably the most interesting relationship remains the one over time.
Especially regarding this relationship the jury is still out. Many authors take sides with
the so-called ‘Easterlin Paradox’, which points out that average happiness has remained
constant over time despite sharp rises in income per capita. Most of the time this paradox
is explained by referring to the importance of adaptation, aspirations and relative
comparisons. For example, Clark et al. (2008) show convincingly, both graphically and
technically, how the Easterlin Paradox is consistent with the positive income-happiness
relationships within countries and across countries at a point in time, if one takes into
account status effects and internal backward- and forward-looking reference points. On
the other hand, however, Stevenson & Wolfers (2008), for instance, claim to have found
evidence which rejects the Easterlin Paradox, showing a positive relationship over time.
Nevertheless, this rejection of the Easterlin Paradox is in itself also far from undisputed.
In a comment on Stevenson & Wolfers (2008), Krueger (2008) for example provides
arguments why he is not yet willing to label the Easterlin Paradox as a ‘nonparadox’.
13
One should however not think that Easterlin (1974) was proposing a radical shift in social policy like
Layard (2005) and others did later on: “The present results do not necessarily imply that a redirection of
attention is needed from economic growth to income redistribution as a vehicle for improving welfare. The
data themselves give no indication that international differences in happiness are systematically related to
inequality. And the theoretical relationship is uncertain – if relative positions were unchanged and income
differences halved, would happiness be greater?” (Easterlin, 1974:119).
21
2.4 Hypothesis
We hope this chapter has provided the reader with some insights into the massive
amount of proposals for the measurement of welfare. In order to provide an overview as
orderly as possible, we have distinguished three general approaches to the measurement
of welfare: GDP measures based on market transactions, alternative composite indexes
based on several domain indicators and, finally, subjective well-being indexes with a
utilitarian welfarist theoretical foundation. Having discussed the pros and cons of the
different approaches, we presume it is clear that neither of the measures represents the
‘holy grail’ some people are looking for. As stated earlier in this thesis, however, it is
highly questionable whether such a ‘holy grail’ measure is ever attainable, or whether we
should even aim for it. We recognize that in the end all proposals for welfare indicators
rely on certain normative judgments and assumptions. Since people can always adhere to
other normative principles, on various grounds, an unavoidable implication is that any
proposal can be validly opposed. Moreover, the appropriateness of a certain measure is
also subject to its intended application.
Nevertheless, we do not want to abstain from presenting our own position in this
respect, and an understanding of this position is also necessary with regard to the
upcoming chapters.
To begin with, we have already taken position in the first section of this chapter,
where we discussed our preferred notion of welfare, focusing on the satisfaction of needs
and acknowledging that this satisfaction of needs has a subjective character. As a result,
we do not only have trouble with GDP per capita as welfare indicator, but also with most
composite indicators. These measures hardly take into account the subjective element of
our preferred notion of welfare. Moreover, quite many composite indicators fall prey to
Hume’s guillotine or can be accused of arbitrariness in the dimensions included, but
particularly also in the adopted dimensional weights. Although an indicator like the
Human Development Index may be very convenient for communicational purposes, it is
beyond any doubt that it is a poor measure of welfare, representing only a rather minor
improvement as compared with GDP per capita.
Considering these remarks, one might perhaps suppose we feel more comfortable
with a welfare measure with subjective well-being as its foundation. Unmistakably, such
a measure pays more attention to subjective experiences, preferences and perceptions. In
addition, one can easily construct a subjective well-being index which can conveniently
be communicated to and understood by the public. Nevertheless, we do not think an
exclusive focus on subjective well-being, in line with Layard (2005) and others, is the
right way to go. First of all, if we want to construct a measure from which we can also
derive policy recommendations, subjective well-being indicators are not really
satisfactory, as they are to a large extent influenced by psychological and emotional
aspects outside the conventional policy domain. Acknowledging the importance of these
psychological and emotional factors, perhaps the best policy prescription would then be
to develop a happiness drug, as Layard (2005) proposes. We feel, however, that such a
policy prescription is neither very realistic nor very interesting from an economic point of
view. Anyway, more importantly, it has robustly been shown that subjective well-being
indexes are hugely affected by adaptation and social comparisons. The role played by
22
these processes is even that large that we generally do not observe any changes in
average happiness over time (cf. Easterlin Paradox). In contrast, however, we think it is
hard to sustain that welfare does not change over time: it is for sure that over time an
increasing amount of our needs has been satisfied. Besides, constant average happiness
over time would imply that it hardly makes any sense to invest in things like education,
health, the environment, government services, et cetera.
So, whereas GDP and most composite indicators pay insufficient attention to the
subjective aspects of welfare, subjective well-being indicators, being heavily affected by
adaptation, rising aspirations and social comparisons, too little reflect objective living
conditions. Thus, all of these indicators are in our opinion imperfect estimates of welfare.
In conclusion, we do not really want to take sides in this debate. Instead, we
would rather propose a synthesis of composite indicators and subjective preferences,
exploring the opportunities to combine the best of both worlds. We advocate a focus on
objective living conditions in relation to preferences.
Whereas subjective well-being data in itself provide a doubtful reflection of
welfare (in particular over time), they can probably, in addition to other stated preference
data, provide useful information about people’s relative valuations for various welfare
dimensions. At each point in time, one can derive a preference ordering of welfare
dimensions for every country on the basis of subjective well-being data and stated
preferences, which is relatively invariant to adaptation and social comparison processes.
The thus obtained information on country-specific preferences can then be used for the
construction of weights for the various objective living conditions included in composite
welfare indicators. Such an approach has the potential of both getting around the
adaptation and social comparison biases of subjective well-being indicators as well as
avoiding Hume’s guillotine and the critique of being paternalistic in setting the weights
for a composite welfare index. By using hybrid weights for the calculation of a composite
indicator, we can avoid a direct confrontation with ‘Scylla’ or ‘Charybdis’. Similarly, by
concentrating on subjective well-being and preferences at one point in time, we can
largely abstract from the third factor (reference framework) in Fleurbaey et al.’s (2009)
conceptual framework, allowing us to focus only on functionings and preferences.
Hence, a synthesis of objective living conditions and subjective preferences can
lead us to empirical welfare measures which are more closely related to our preferred
notion of welfare. It is our hypothesis that such an empirical synthesis can yield valuable
insights with regard to international welfare comparisons. Although we certainly do not
want to assert that such a synthesis will be free of any flaws, it definitely deserves some
exploration.
2.5 Equivalent income method
Finally, before closing this chapter, it is relevant to spend a few words on the
general welfare measurement method that will be employed in the next chapters. This
method is based on the concept of equivalent income. Equivalent incomes can help
ranking different combinations of living conditions, or in Sen’s words functionings. For
making a ranking between situations it is of course not enough to only have information
23
on every separate living condition (recall our criticism towards Cnossen, 2009). One
needs some translation mechanism through which one can compare different
combinations of living conditions. Equivalent income is one example of such a
mechanism, just as the composite indicators we discussed earlier in this chapter all have
their own translation mechanisms.
One of the central features of the equivalent income translation mechanism is that
for every functioning (except income) a certain reference value is determined. Then,
equivalent income is the amount of income that makes an individual indifferent between
her actual bundle of functionings (including her actual income) and a bundle which
contains this equivalent income and all the other functionings at their reference value; see
also Figure 1 on the next page. Thus, equivalent income captures information on how
one’s actual functionings bundle relates to the reference functionings bundle. If one’s
actual functionings bundle exactly matches the reference bundle, then equivalent income
will equal the person’s actual income. However, if one’s state of health is, for instance,
worse than the reference value for health, the equivalent income which makes the
individual indifferent between her actual functionings bundle and the reference bundle
will be lower than her actual income. In this way, actual income is corrected for
deviations of one’s actual functionings bundle from the reference bundle to attain
equivalent income, which thus practically provides information about one’s total
functionings bundle and allows for interpersonal comparisons of functionings bundles.
It is important to notice that the resulting income corrections do not only depend
on one’s actual functionings and the corresponding reference values. Corrections depend
namely on both the deviation between these two as well as the individual’s preferences
for these functionings, which determine the shape of the individual’s indifference curve.
These preferences imply a certain ‘willingness-to-pay’ for the different functionings and
the difference between actual and equivalent income can be interpreted as an individual’s
total willingness-to-pay for moving from his actual functionings bundle to the reference
bundle. In the upcoming chapters, we will try to include subjective preferences in our
willingness-to-pay estimates.
An important advantage of the equivalent income method is that it enables us to
combine information on objective living conditions and subjective preferences in a
theoretically sound manner. In addition, equivalent income measures can be easily
communicated, as a result of its intuitively easily understandable measurement units.
Moreover, the method largely releases us of the heavy responsibility of making
assumptions on the substitutability between functionings. Probably the most significant
disadvantage of the method is that we have to determine reference values for the
functionings involved. This determination unavoidably requires certain normative
assumptions. We think, however, that this is not too large of a problem, as we can apply
different methods for setting these values and because we can easily experiment with
different scenarios. Finally, a non-negligible reason for choosing the equivalent income
method is that by doing so, we can join with and elaborate on some promising recent
contributions to the literature on the measurement of welfare.
24
Figure 1 Illustration of the equivalent income concept
The figure above provides a simple illustration of the equivalent income concept, considering an individual whose utility is merely dependent on her income and her health status; U represents her indifference curve. Her actual functionings bundle equals (y0 , h0). If the reference value for health equals h*, the individual’s equivalent income is equal to y*: she derives the same utility from having the equivalent income together with health at its reference value as she derives from her actual functionings bundle. From this picture one can easily infer that the individual’s equivalent income is not only dependent on the difference between the individual’s actual functionings and the corresponding reference values, but also on the shape (slope) of the indifference curve, which reflects the person’s preferences for the different functionings.
25
III Equivalent incomes and country-specific preferences for
inequality and leisure
Having discussed the most common approaches towards measuring welfare in the
previous chapter, we now turn to investigating the opportunities and potential impact of
combining objective living conditions and subjective preferences for obtaining a better
understanding of welfare and differences in welfare among countries. More specifically,
we aim to explore the possibilities and consequences of including subjective preferences
in equivalent income measures based on objective living conditions. The basics of
equivalent income measures have already been explained in the previous chapter. The
present chapter and the next one each consider somewhat different methods for
calculating equivalent incomes, based respectively on a formal economic model and on
willingness-to-pay (WTP) measures derived from life satisfaction regressions.
3.1 Equivalent incomes in Fleurbaey & Gaulier (2009)
At the heart of this chapter is the paper ‘International Comparisons of Living
Standards by Equivalent Incomes’ by Marc Fleurbaey and Guillaume Gaulier (2009). In
their paper these authors present a formal economic model from which they derive
corrections of standard GDP for inter alia inequality, leisure, risk of unemployment,
health and household size. They calculate these income corrections for a sample of 24
OECD countries for the year 2004. Though not exclusively, the corrections of Fleurbaey
& Gaulier (2009) are mainly price-based. In this respect, the paper relates to Becker et al.
(2005), who compute a ‘full’ income measure encompassing income and life expectancy.
Decancq & Lugo (2010) qualify these two works as notable exceptions in the literature
on multidimensional well-being, in which price-based equivalent income measures are
not very common. Decancq & Lugo (2010) quote in this context Foster & Sen (1997),
who argue that even if implicit prices can be obtained, they are in general inappropriate
for well-being comparisons, a task for which they are not constructed according to these
authors. Indeed, although price-based weights can be calculated relatively easily, their
use is obviously debatable. Nonetheless, Fleurbaey (2009) concludes that the equivalent
income method might deserve more attention than it has received so far, since it takes an
interesting middle-ground position in the debate between the capabilities approach and
welfarism: while being based on objective living conditions, it leaves some room for
subjective preferences. Moreover, as Fleurbaey (2009) shows, the method respects the
Pareto principle14
, one of the core elements of welfarism.
Fleurbaey & Gaulier (2009) motivate their choice for their price-based equivalent
income approach by referring to other measures of social welfare, like the Human
Development Index (HDI) discussed in chapter 2. Fleurbaey and Gaulier note that many
14
This principle states that if every individual prefers a certain option to another, then so must the resulting
societal preference order (if everybody agrees on the ranking of all possible options, so should the group;
the collective ranking should coincide with the common individual ranking). See for instance Hindriks &
Myles (2006).
26
welfare indicators are based on the aggregation of various subindexes, and that the
weights used for this aggregation “have no rational basis and appear arbitrary” (Fleurbaey
& Gaulier, 2009:597). They claim that by using a price-based equivalent income method
most of their own calculations are, in contrast, not based on arbitrary weights reflecting
ethical assumptions; instead, their calculations are founded upon willingness-to-pay
figures that can be discussed on an empirical basis.
Albeit we tend to agree with Fleurbaey and Gaulier that their method comprises a
more sophisticated weighting scheme than, for example, the Human Development Index,
a critical remark is justified in this context: whereas Fleurbaey and Gaulier claim that
their corrections do not reflect any ethical assumptions, we argue that their choice for
using price-based weights for valuating the various dimensions of welfare is actually also
an ethical assumption. We do not think there is any truly objective reason for the use of
price-based weights. The reasoning that prices reflect people’s preferences and valuations
is already an ethical assumption in itself, as noted in chapter two.
As Fleurbaey & Gaulier (2009) note, however, prices can indeed sometimes
provide reasonable estimates of individuals’ valuations of various welfare dimensions.
When perfect-functioning markets exist on which these welfare dimensions are traded,
and where individuals can freely choose how much they want to ‘consume’ of the welfare
dimension concerned, prices can provide a reliable monetary estimate of the individuals’
marginal willingness-to-pay for this welfare dimension.
We should emphasize the word ‘marginal’ in this regard, as prices are the
outcome of bargaining processes on the market and merely reflect the individual’s
valuation for the last unit he bought (which may not be a good reflection of the
individual’s valuation for the other units he bought). Thus, when calculating equivalent
incomes, prices can only be used as a reliable basis for income corrections if the
differences between the individual’s actual functionings bundle and the reference bundle
are marginal. Otherwise, prices might not provide a proper reflection of how the
individual evaluates changes in his functionings bundle.
In addition, and at least as importantly, we should stress that prices only provide a
good estimate of subjective preferences if perfect-functioning markets exist. For welfare
dimensions for which only an imperfectly competitive market exists or for which no
market exists at all, prices do generally not reflect people’s preferences for these welfare
dimensions. If there is market power on the supply or demand side of the labour market,
for example, wages provide an imperfect estimate of people’s valuations for leisure time.
Similarly, prices have also little to tell about people’s valuations for air quality, since no
market for air quality (or air pollution) exists.
Keeping these issues, to which we will return in a few moments, in mind, we
think it is good to now first have a general look at Fleurbaey & Gaulier’s (2009)
calculations and their results.15
15
For more detailed information on the data and formulas used, please consult Fleurbaey & Gaulier (2009)
and / or Fleurbaey & Gaulier (2007), which is the working paper version of the 2009 article. As compared
with Fleurbaey & Gaulier (2009), Fleurbaey & Gaulier (2007) also add corrections for consumption of
fixed capital and environmental degradation. In general, most data can be obtained from OECD sources,
online retrievable via OECD.Stat (stats.oecd.org).
27
First, Fleurbaey & Gaulier (2009) make some obvious corrections to GDP, which
are conventionally also included in a country’s national accounts: apart from correcting
for population size (by looking at GDP per capita) and price levels (by using purchasing
power parities, PPPs, to translate all figures into US dollars), they correct for income
transfers paid to and received from other countries (by looking at gross national income
instead of gross domestic product) as well as for decreases in the country’s capital stock
(by subtracting the consumption / depreciation of capital from the gross domestic
product, arriving at the net domestic product).16
The latter correction is applied in order to
take into account the future prospects of the country. The authors follow in this respect
Weitzman (1976), who shows that in a competitive economy with a fixed interest rate the
discounted value of total consumption over the infinite future equals the discounted value
of constant consumption, which equals the current net domestic product (NDP).
Moreover, a correction is made for leisure, acknowledging the fact that people do
not only care about income, but also about their amount of leisure time. The reference
value of leisure time is set at the median amount of leisure time across the 24 OECD
countries in the sample and deviations are valued at each country’s average hourly wage.
In conformance with the observation that people do not like to be unemployed
(and do not like the risk of becoming unemployed), Fleurbaey and Gaulier also correct
for the risk of unemployment. In the calculation of this correction the authors include
amongst others the probability of falling into unemployment, the probability of exiting
unemployment and the income losses related to unemployment. Taking a zero risk of
unemployment as their starting-point, the authors calculate a ‘risk premium’ for the risk
of becoming unemployed, which is then subtracted from the country’s average income.
In order to correct for health, Fleurbaey and Gaulier use data on the health-
adjusted life expectancy per country. Health-adjusted life expectancy (HALE) equals the
expected number of ‘healthy years’ at a child’s birth. The authors use these HALE-data
instead of ordinary life expectancy data, since they assume that life only has value when
one is in good health. Taking the maximum of HALE observed in their sample of
countries as a reference, the authors then calculate income corrections for health on the
basis of the model of Becker et al. (2005).
Furthermore, recognizing that economies of scale occur in households comprising
more than one person (costs for certain goods like common rooms and heating can be
shared among the household members), Fleurbaey and Gaulier also correct for household
composition by computing the income that would yield the same utility if everyone was
member of a single-person household.
Fleurbaey and Gaulier’s one but last correction is for inequality. This correction
follows from the social welfare function employed by the authors, which incorporates a
certain preference for equality by giving priority to the worst-off, as the authors want to
avoid counting one dollar for the poor as equivalent to one dollar for the rich. The
inequality corrections are calculated based on the Atkinson Inequality Index, with an
assumed inequality aversion parameter of 1.5.
16
The correction for the consumption of capital is only made in Fleurbaey & Gaulier (2007), the working
paper version of the 2009 article.
28
Finally, the authors correct for environmental sustainability, which affects the
country’s future welfare prospects.17
Depletion of non-renewable resources like oil is
accounted for by subtracting current resource extraction valued at its average net price
from each country’s average income. In this respect, the authors have decided to attribute
to each country a share of the global welfare loss due to resource depletion that is
proportional to each country’s share in the global consumption of the resource. In
addition, air pollution and the cost of global warming are taken into account by
subtracting from each country’s average income its emissions of CO2, methane and
nitrous oxide, valued at a price of 25 dollars per ton of CO2 equivalents (which is,
according to the authors, not an unrealistic estimate of the shadow price of the emission
of greenhouse gases).
We have requested Fleurbaey & Gaulier’s (2009) calculations and we have
checked these with our own calculations, finding no discrepancies. The results of
Fleurbaey & Gaulier (2009), extended by the additional corrections of Fleurbaey &
Gaulier (2007), are summarized in Figure 2 and Table 1 on the next pages.18
Figure 2
plots the relative position of each country after every correction, where 100% refers to
the sample average. Table 1 shows for each country the absolute size of every correction
and provides percentages on how the final indicator relates to GDP per capita. Moreover,
country rankings are presented for both GDP per capita as well as the final indicator, and
we have added Atkinson Inequality Indexes for the cross-country distribution of
equivalent incomes after each correction.
From Figure 2 and Table 1 one can easily infer that the final indicator of
Fleurbaey & Gaulier (2009) is significantly correlated with GDP per capita. This is
however not a surprising observation, as GDP per capita has been the starting-point of the
calculations. Nevertheless, Fleurbaey & Gaulier (2009) note that none of the corrections
is significantly correlated to GDP per capita; this even holds for the health correction.
Although the final indicator and GDP per capita are clearly related, one can also
observe sharp distinctions, especially if we consider that corrections of only a few
percentage points can already translate into country rankings that are significantly
different from conventional rankings based on GDP per capita. From this perspective, the
15 percentage points decrease of the United States’ welfare measure, resulting among
other things from its high level of inequality and low amount of leisure time, is for
instance far from negligible. On the other hand, a country like France experiences an
increase in its welfare measure of almost the same magnitude, amongst others because of
its large positive correction for leisure. Similarly, Japan gains from its high level of
health-adjusted life expectancy.
In terms of cross-country inequality, it does not matter too much whether one
looks at the distribution of GDP per capita or the distribution of the final indicator.
Although the Atkinson Inequality Index experiences some significant changes during the
17
Just like the correction for the consumption of capital, the correction for environmental sustainability is
only applied in Fleurbaey & Gaulier (2007). 18
Being an outlier, Luxembourg is excluded from Figure 2. The sample average to which 100% refers, is
calculated for all 24 countries except Luxembourg. In Table 1 the Atkinson indices are also computed on
the basis of all 24 countries except Luxembourg. Atkinson indices are calculated for three different values
of the inequality aversion parameter: 0.5, 1.5 and 2.5.
29
Figure 2 Relative positions in terms of equivalent income after each correction, for the year 2004 (Source: Fleurbaey & Gaulier, 2007 and own calculations)
0%
20%
40%
60%
80%
100%
120%
140%
160%
Austra
liaAus
tria
Belgi
umC
anad
aD
enm
ark
Finla
ndFra
nce
Ger
man
yG
reec
eIc
elan
dIre
land
Italy
Japa
n
Korea
Net
herla
nds
New
Zea
land
Nor
way
Portu
gal
Spain
Swed
enSw
itzer
land
Uni
ted
Kingd
omU
nite
d Sta
tes
GDP per capita GNI per capita Leisure Unemployment Health Household size Inequality Capital consumption Sustainability
30
Table 1 Absolute corrections, comparison GDP per capita and final indicator and inequality of equivalent income distribution, for the year 2004 (Source: Fleurbaey & Gaulier, 2007 and own calculations)
Numbers in parentheses denote standard errors. Three stars indicate significance at the 1% level, two stars at the 5% level and one star at the 10% level. All regressions include the following control variables: inflation rate, unemployment rate, a dummy indicating whether the respondent is unemployed, the respondent’s perceived health status, the respondent’s sex, the age and the squared age of the respondent, dummies for the respondent’s marital status (married, living as a couple, divorced, separated, widowed) and the respondent’s number of children. Finally, wave dummies have been included.
40
satisfaction scores as compared with women, people who are married or live together as a
couple tend to report higher levels of life satisfaction than other people, and we also
observe a positive relationship between life satisfaction and a person’s number of kids.
In addition to these control variables, column 1 of the table adds inequality
(measured by the Gini coefficient, expressed on a 0-1 scale) and the natural logarithm of
income. We have used the natural logarithm of income instead of normal income,
because previous literature has consistently found that there is a logarithmic relationship
between income and subjective well-being. Pursuant to the findings of the subjective
well-being literature, the coefficient of the logarithm of income turns out to be
significantly positive in our regression. Furthermore, column 1 shows that if we do not
include country- or regime-specific effects, we find a significantly positive coefficient for
inequality as well. At first glance, the sign of this coefficient might very well be a
surprise for the reader, as one would perhaps more easily expect that inequality is a ‘bad’
thing, which therefore should lower life satisfaction scores. However, this line of
reasoning primarily approaches inequality from the perspective that people have a taste
for equality. Inequality, nevertheless, can also be interpreted in other ways. For rich
people, inequality may be a signal of their status, and thus may have a positive effect on
life satisfaction. On the other hand, if perceived social mobility is large, poor people can
interpret inequality as a signal of good opportunities for the future (the so-called ‘tunnel
effect’). Because different inequality-life satisfaction mechanisms may be at play in
different countries, the positive sign of the inequality coefficient is not really worrying.
As a matter of fact, this result for the inequality coefficient may already lend some
support to our hypothesis that preferences towards inequality may vary substantially
across countries.
The second column of Table 2 includes regime-specific inequality terms, by
including regime dummies and interaction terms of the regime dummies and the
inequality variable. The Anglo-Saxon regime is taken as the base level. Overall, all
regime-specific inequality effects seem rather significant. As can be seen from the table,
inequality has the most negative effect in the Nordic countries, followed by the Anglo-
Saxon and the Asian countries. In the Continental European and especially the
Mediterranean countries higher levels of inequality tend to correspond with higher levels
of life satisfaction. The signs, magnitudes and significance levels of the inequality
coefficients turn out to be quite robust to the inclusion of regime-specific income effects,
as has been done in the regression of column 3.27
In fact, these are quite strong results, in particular if one notes that we have also
controlled for variables like income and unemployment status. Nonetheless, for a proper
understanding of the importance of inequality for life satisfaction one should keep in
mind that the inequality variable has a 0-1 scale. This implies that, for instance, in the
Anglo-Saxon countries a decrease in the Gini coefficient from, say, 0.30 to 0.28 only
leads to a rather minor increase on the 1-10 life satisfaction scale of 0.062 (based on the
regressions in columns 2 and 3).
Besides, it is also important to remember that different forces may be at play in
different groups of countries. From the regressions presented above we cannot tell
27
Notice that income seems to have a moderately positive impact on life satisfaction in the Anglo-Saxon
countries, an even more positive impact in the Mediterranean, Continental and Asian countries, and a
roughly negligible effect in the Nordic countries.
41
whether the inequality coefficients represent intrinsic preferences towards inequality (a
‘taste’ for equality, a natural drive for status, etc.) or whether they mainly reflect context-
dependent preferences (dependent on the perceived degree of social mobility, the
perceived risks of crime and political instability, etc.). In addition, the results of Table 2
also fail to uncover potential differences in inequality preferences within groups of
countries. In response to these remarks, Table 3 on the next page presents similar
regressions as in Table 2, but now for samples restricted to several subgroups of people,
respectively people with left- and right-wing political preferences and poor and rich
people.28
Respondents are labeled left-wing it they answered 1, 2, 3 or 4 to the World
Values Survey question: “In political matters, people talk of ‘the left’ and ‘the right’.
How would place your views on this scale, generally speaking? 1-Left, 2, …, 9, 10-
Right”. Respondents that answered 7, 8, 9 or 10 are identified as being right-wing.
Regarding the rich-poor distinction, people who are in the lowest four income deciles are
labeled poor and people in the highest two income deciles are labeled rich.29
Looking at the regime-specific inequality effects presented in Table 3, we can
observe several interesting patterns. First of all, the base level inequality coefficients
show that in the Anglo-Saxon countries aversion to inequality primarily resides in the
minds of left-wing and poor people. Whereas we observe for these groups highly
significant and relatively large negative coefficients, the rich and right-wingers in the
Anglo-Saxon countries, in contrast, do not tend to worry about inequality: these latter
groups have insignificant inequality coefficients. These results indicate that there exists
no real ‘taste for equality’ in the Anglo-Saxon countries. Otherwise, we should also have
observed significant negative coefficients for the rich and right-wing people. More
specifically, under the presence of a taste for equality we should in particular have
expected a significant negative coefficient for the rich: if we assume equality to be a
normal good, demand for equality should rise with income. Therefore, instead of
supporting a ‘taste for equality’ thesis, the results for the Anglo-Saxon countries sketch a
picture of a society with low perceived social mobility30
, where inequality represents a
signal of bad future prospects for the poor and a signal of an unthreatened status position
for the rich.
As compared with the results for the Anglo-Saxon countries, the results for the
Nordic countries exhibit a remarkably different pattern. Although the results do not
indicate significantly different coefficients for the poor and the left-wingers as compared
28
Actually, the regressions in Table 3 are of the linear form instead of the ordered logit form. The main
reason for this is that our statistical package (SPSS Statistics 17.0) does not provide the option to run
ordered logit regressions for subsamples. However, as a check we have also run linear regressions based on
the regression equations in Table 2. A comparison of these linear regressions with the ordered logit
regressions in Table 2 has shown that the differences in results between the two specifications tend to be
fairly small. 29
The correlation between the variables ‘Left’ and ‘Poor’ is small, but significantly positive (0.019); the
correlation between ‘Left’ and ‘Rich’ is small, but significantly negative (-0.028); the correlation between
‘Right’ and ‘Poor’ is small, but significantly negative as well (-0.040); and the correlation between ‘Right’
and ‘Rich’ is small, but significantly positive (0.064). See also Appendix B for a more comprehensive
correlation matrix that provides further insights into the characteristics of the various subgroups. 30
This suggestion contrasts sharply with the generally assumed high degree of perceived social mobility in
the Anglo-Saxon countries and especially the United States.
42
Table 3 Life satisfaction regressions containing regime-specific inequality effects for various subgroups within society
Dependent Variable: LIFE SATISFACTION
Linear Regression Models
All things considered, how satisfied are you with your life as a whole these days?
Numbers in parentheses denote standard errors. Three stars indicate significance at the 1% level, two stars at the 5% level and one star at the 10% level. All regressions include the following control variables: inflation rate, unemployment rate, a dummy indicating whether the respondent is unemployed, the respondent’s perceived health status, the respondent’s sex, the age and the squared age of the respondent, dummies for the respondent’s marital status (married, living as a couple, divorced, separated, widowed) and the respondent’s number of children. Finally, wave dummies have been included.
43
with the Anglo-Saxon countries, they also show large and significantly negative
inequality coefficients for the rich and right-wingers in the Nordic countries. Thus, in
these countries we find more support for the existence of a widespread, culturally-
determined ‘taste for equality’. In addition, the more widespread inequality aversion in
the Nordic countries is also in line with the strong negative inequality coefficient for
these countries in Table 2.
For the Continental European countries the results point at yet another pattern.
First, these countries resemble the Anglo-Saxon countries in the sense that the rich and
the right-wingers do not care at all about inequality: for these groups we observe
insignificant inequality coefficients. Moreover, the left-wing people in the Continental
countries also seem not to care about inequality. The only subgroup for which there
appear to be significant inequality effects comprises the poor people, and here the sign of
the inequality-life satisfaction relationship tends to be positive. So, if people in these
countries care about inequality at all, it is probably not resulting from ideological beliefs
or intrinsic preferences, but more from the social context people live in: the perceived
degree of social mobility is possibly rather high, causing the poor people to interpret
inequality as a signal of potentially rosy future prospects and vast opportunities for
improvement of their situation.
Of all regimes considered, the Mediterranean and Asian countries provide the
strangest patterns, which are most difficult to explain. The Mediterranean case for the
largest part seems to display the same pattern as the Anglo-Saxon countries. However,
there is one significant difference: the inequality coefficient for the poor subgroup in the
Mediterranean countries tends to be significantly positive. This significantly positive
coefficient for the poor yields a quite surprising combination with the significantly
negative coefficient for the Mediterranean left-wingers, which is hard to explain. Is there
perhaps a tension between the ideological beliefs of the left-wingers and the context-
dependent perceptions of the poor?
Finally, it is most and for all difficult to detect any consistency in the results for
the Asian countries: for the rich we observe insignificant inequality coefficients, for the
right-wingers and the poor we find significantly negative coefficients and the coefficient
for the left-wingers seems positive. Without specific knowledge of the situation in the
Asian countries these results seem somewhat contradictory, especially as far as the left-
right distinction is concerned. Nevertheless, it is perhaps possible that the left-right
distinction in these countries is determined along different demarcation lines as compared
with the other regimes. In this case, the results for the Asian regime would already
become better understandable. Moreover, one should remember that the Asian regime is
some kind of a residual group, containing only two countries (Japan and Korea), which
besides differ in some meaningful aspects.
Concerning the discussion of the results above, it is important to note that we
want to refrain from any firm conclusions regarding the mechanisms at play in the
different groups of countries, since the results simply do not allow us to draw any
conclusions regarding the relative importance of intrinsic and context-dependent
preferences et cetera. Instead, the inferences above concern mainly suggestions derived
from the regressions and, hence, should not be interpreted as anything more than
suggestions. However, what we can conclude from Table 2 and Table 3 is that there
appear to be significant cross-regime differences in inequality aversion preferences,
44
partially also reflecting differences in the distribution of inequality aversion preferences
across various subgroups within the regimes.
One critical question that can be raised in this context is to what extent we can
derive reliable information on preferences from subjective well-being regressions. This
question certainly makes sense, not in the least because we have expressed in chapter two
that subjective well-being measures provide an unsatisfactory measure of welfare.
However, here we are dealing with a somewhat different issue. While we retain our main
criticism towards subjective well-being as a welfare indicator (that subjective well-being
over time, and sometimes also across space, is almost invariant to changes in objective
living conditions due to the dominance of adaptation, aspiration and social comparison
processes), we do not think that this criticism rejects the idea that subjective well-being
data can provide certain insights into people’s relative, ordinal valuations for different
dimensions of life at one point in time. In our regressions, the survey wave dummies for
example partly control for adaptation processes over time.
More generally, one can question whether answers to survey questions can be
cross-nationally compared at all, for instance because respondents from different
countries or cultures interpret questions differently or use different scales in answering
the questions (Heath et al., 2009). However, since our regressions include standard
regime dummies, such problems related to cross-national comparisons are substantially
diminished in our analysis.
Nevertheless, if we return to our initial goal for conducting the regression
analyses above (to explore whether we can derive inequality aversion parameter
estimates from regressions like in Alesina et al. (2004)), we are still skeptical regarding
the usefulness of the conducted regressions. Whereas it is certainly possible to think of
ways to attain inequality aversion parameter estimates on the basis of the inequality
coefficients or the relationship between the inequality and income coefficients in the
regressions above, we doubt whether this is a desirable procedure. In the first place,
whilst we were aiming for the inclusion of country-specific inequality aversion
preferences in Fleurbaey & Gaulier’s (2009) calculations, these regressions only allow us
to make a distinction between different groups of more or less similar countries.
Obviously, this is a serious drawback of the regressions. Second, we have some doubts
whether the inequality coefficients really represent what we want them to represent; in a
certain sense the regime-specific inequality coefficients are not merely a measure of
preferences towards inequality, but also a measure of the degree of similarity among the
countries within a regime. This issue has to do with the fact that we only have data on
inequality at the country level. Therefore, within each regime, we just have about three or
four possible values for the inequality variable. Thus, the degree of unity within a regime
(amongst others regarding the inequality variable) can significantly affect the slope and
significance of the inequality-life satisfaction relationship within that specific regime.31
31
We have checked the sensitivity of the results to group composition amongst others by excluding from
the regression analysis in Table 2 the countries that are difficult to assign to one of the welfare regimes:
Australia and New Zealand have been excluded from the Anglo-Saxon group, Netherlands from the Nordic
group, France from the Continental group, and the Asian group has been excluded completely. Thus, we
have repeated the regressions of Table 2 for four regimes, each consisting of four members. Results are
45
So, although the regression results are intuitively quite attractive, we do not entirely trust
them, and hence prefer to continue our exploratory expedition by investigating yet
another potential source of information on preferences towards inequality.
3.2.3 Estimating inequality aversion on the basis of direct survey questions
In fact, this section will not describe an entirely new source of information. Just as
in the previous section, we will namely rely on information from large-scale opinion
surveys. In contrast to the previous section, however, we will not try to (hedonically)
derive preferences from subjective well-being regressions, but instead rely on stated
preferences: respondents’ direct answers to questions related to inequality.
Nevertheless, we do not have any single question at our disposal that measures
inequality aversion fully satisfactorily. Due to the framing of most inequality-related
survey questions, answers to these questions are never purely an expression of one’s
preferences towards inequality. These answers always depend on certain contextual
reference levels or involve preferences towards other issues. Consider for example the
following question from the US General Social Survey: “Do you agree or disagree?
Differences in income in America are too large. (1-Strongly agree, …, 5-Strongly
disagree)” Obviously, this question does not only capture preferences towards inequality,
but is also strongly dependent on the contextual situation. Furthermore, the question may
also implicitly evoke answers that are partly based on opinions on the desired size of the
government apparatus. Therefore, answers to this question are not purely a measure of
intrinsic inequality aversion preferences.
However, the lack of availability of a single perfect question regarding
preferences towards inequality is, as a matter of fact, not such a large problem as it may
seem. By combining various questions that are somehow related to inequality aversion
preferences, it is still possible to obtain a fairly good impression of people’s attitudes
towards inequality. Indeed, such a combination method even has some benefits as
compared with looking at only one single question. By using a variety of questions, we
actually conduct a little robustness test, smoothing the influence of possible social-
desirability biases or biases due to the ordering of questions and so on, thus arriving at a
rather reliable estimator of preferences towards inequality.
The procedure that is generally followed for creating such a composite variable is
principal component analysis. The central idea of principal components analysis is to
reduce the multidimensionality of a group of interrelated variables by combining the
variables into a component variable, while still retaining as much as possible of the
reported in Appendix C. These results show that the Nordic group still tends to exhibit the highest level of
inequality aversion, followed by the Anglo-Saxon countries. However, whereas in the regressions of Table
2 the Mediterranean countries seem to be the least inequality-averse, this is no longer the case for the
regressions presented in Appendix C, where the Continental group seems to be the least inequality-averse.
Moreover, comparing the regressions in Table 2 with those presented in the appendix, the coefficients of
the inequality terms experience serious changes in their absolute value. Finally, it should be noted that the
coefficients of the control variables hardly differ between the two tables, both with respect to their
significance levels as well as with respect to their values.
46
variation in the underlying variables. The newly created variable is a linear combination
of the underlying variables, with the weights in this combination being represented by the
eigenvectors of the covariance or correlation matrix of the underlying variables.
Principal component analysis has often been used for deriving certain index
variables from a range of survey questions; see for instance Inglehart & Baker (2000),
who use survey answers from the World Values Survey to create various modernization
indexes, measuring amongst others ‘Traditional vs. Secular-Rational Values’ and
‘Survival vs. Self-Expression Values’. Here we base our principal components analysis
on six variables from the World Values Survey, videlicet:
- Income equality: Now I'd like you to tell me your views on various issues. How would
you place your views on this scale? 1 means you agree completely with the statement on
the left; 10 means you agree completely with the statement on the right; and if your views
fall somewhere in between, you can choose any number in between. Sentences: Incomes
should be made more equal vs. We need larger income differences as incentives.
- Government responsibility: Now I'd like you to tell me your views on various issues.
How would you place your views on this scale? 1 means you agree completely with the
statement on the left; 10 means you agree completely with the statement on the right; and
if your views fall somewhere in between, you can choose any number in between.
Sentences: People should take more responsibility to provide for themselves vs. The
government should take more responsibility to ensure that everyone is provided for.
- Freedom or equality: Which of these two statements comes closest to your own
opinion? A. I find that both freedom and equality are important. But if I were to choose
one or the other, I would consider personal freedom more important, that is, everyone can
live in freedom and develop without hindrance (coded 1) B. Certainly both freedom and
equality are important. But if I were to choose one or the other, I would consider equality
more important, that is, that nobody is underprivileged and that social class differences
are not so strong (coded 2).
- Self-positioning in political scale: In political matters, people talk of “the left” and “the
right.” How would you place your views on this scale, generally speaking? (1=Left, ….,
10=Right)
- Most people can be trusted: Generally speaking, would you say that most people can be
trusted (coded 1) or that you need to be very careful in dealing with people (coded 2)?
- Post-materialist index: General index indicating to what extent people adhere to either
traditional or post-materialist values, ranging from 0-Materialist to 5-Post-materialist.
The variable is a cluster of variables involving materialist values (like maintaining order,
fighting inflation, having a safe job) versus post-materialist values (freedom, tolerance,
self-expression). See also Inglehart (1997) for more detailed information about this index.
47
Clearly, the first three variables all contain directly an element of preferences for
equality or inequality aversion. The last three variables, on the other hand, have a more
general character, but are all related to preferences towards inequality as well: people on
the left side of the political spectrum tend to be more inequality-averse, generalized trust
is an indicator of social cohesion and can through this channel also be related to
preferences towards inequality, and finally, people with more post-materialist values are
likely to be more concerned with social issues like inequality.
Table 4 below shows the component loadings of the first principle component
calculated on the basis of the six variables mentioned above. The principle component
analysis includes a total number of 20,478 observations. The Kaiser-Meyer-Olkin (KMO)
Measure of Sampling Adequacy equals 0.62 and the p-value of Bartlett’s Test of
Sphericity equals 0, indicating that we have conducted an appropriate analysis. The
created first principal component accounts for almost 27 percent of the variation in the
underlying variables. Although this percentage might seem quite low, it is not really
worrying: since we have conducted our principal component analysis on the basis of
individual-level data, there is likely quite some noise caused by measurement error et
cetera. If we had conducted our principal component analysis for country-level data, the
resulting first principal component would certainly have explained a larger fraction of the
variation in the underlying variables. In the same manner, the component loadings in
Table 4 would probably also have been higher if we had used country-level data.
Nevertheless, for our purposes we prefer individual-level data, as we think it is most
correct to include as much variation as possible at the individual level when estimating
country-specific preferences.
Table 4 Component matrix
Variable Component loading
Income equality -0.57
Government responsibility 0.55
Freedom or equality 0.49
Self-positioning in political scale -0.69
Most people can be trusted -0.02
Post-materialist index 0.50
The component loadings represent the correlations between the first principal
component and the underlying variables. As can be deduced from the signs of the
component loadings, we can label the obtained first principal component as ‘Degree of
inequality aversion’ or ‘Preference for equality’. Interpreting the principal component
variable in this way, all of the correlations are in conformance with our expectations (also
recall the above mentioned coding of the underlying variables). In addition, except for the
trust variable, all component loadings are rather large.32
32
We have experimented with omitting some of the underlying variables from the principal component
analysis, particularly the trust variable and the post-materialist index. Nonetheless, this did not lead to
48
Finally, we have calculated the principal component scores for the 20,478
individuals in our sample that have been included in the principal components analysis.33
Subsequently, we have used these component scores for calculating the average
component score per country, which can be interpreted as a measure of the country’s
inequality aversion. These country-specific averages are shown in Table 5 below.34 35
Table 5 Average country scores on the principal component ‘Inequality Aversion’
Country Average score
Austria -0.15
Belgium 0.00
Canada -0.19
Denmark -0.20
Finland -0.24
France 0.26
Germany -0.14
Iceland 0.11
Ireland -0.10
Italy 0.38
Japan 0.16
Korea -0.40
Netherlands 0.13
Norway -0.16
Portugal 0.24
Spain 0.59
Sweden -0.31
United Kingdom -0.01
United States -0.44
One first interesting observation from this table is that we can distinguish several
groups of countries with quite similar average component scores. In fact, we can observe
spectacular changes in the results: the component loadings of the other variables stayed about the same, and
the Kaiser-Meyer-Olkin (KMO) Measure of Sampling Adequacy and the percentage of variance in the
underlying variables explained by the first principal component rose moderately, as one would expect when
reducing the number of variables included in a principal component analysis. 33
Due to a lack of data availability, it was not possible to include all of the individuals in our sample in the
principal components analysis. Actually, all 20,478 individuals included in the principal component
analysis were part of survey wave 2 (1989-1994). Although this is obviously not an ideal situation, it does
not have to be a large problem either: Alesina & Giuliano (2009) have for example shown that preferences
towards redistribution, which are strongly related to inequality aversion preferences, are highly persistent
over time. More generally, other literature on values and preferences has also consistently shown that
preferences and values tend to change only very slowly over time. 34
The first column shows the standard principal component scores (PC) per country, as depicted in Table 5. In the second column these component scores have been rescaled to a 0-1 scale, with 0 and 1 respectively referring to the lowest and highest principal component score within the sample. The remaining columns of the table show the inequality aversion parameter estimates calculated on the basis of different conversion formulas. Variant (0) is the Fleurbaey & Gaulier (2009) scenario, where a uniform parameter value of 1.5 is assumed. The formulas corresponding to the other scenarios are:
The first column shows the standard principal component scores per country (PC), as depicted in Table 9. In the second column these component scores have been rescaled to a 0-1 scale, with 0 and 1 respectively referring to the lowest and highest principal component score within the sample. The column PC* presents the first principal component scores created on the basis of a principal component analysis including the hourly wage variable and the original PC scores from Table 9. The remaining columns of the table show the adjusted hourly wages per country calculated on the basis of various adjustment formulas. Variant (0) represents the standard hourly wage data, as have been used in Fleurbaey & Gaulier (2009). The formulas corresponding to the other scenarios are (with W referring to the standard wage):
- Variant (1): adjusted wage = W + 10 * PC - Variant (2): adjusted wage = W + 20 * PC - Variant (3): adjusted wage = Wmin + PC(0-1) * ( Wmax - Wmin ) - Variant (4): adjusted wage = Wmin + PC*(0-1) * ( Wmax - Wmin ) - Variant (5): adjusted wage = ( 1 + PC / 2 ) * W - Variant (6): adjusted wage = ( 1 + PC ) * W
65
Table 11 Leisure corrections for various wage adjustment scenarios (as % of Gross National Income (GNI) per capita)
Numbers in parentheses denote standard errors. Three stars indicate significance at the 1% level, two stars at the 5% level and one star at the 10% level. An explanatory list of the country abbreviations can be found in Appendix E.
In any case, we can conclude from Table 13 that significant differences in
preferences apparently exist among countries and that it therefore makes sense to
estimate country-specific willingness-to-pay measures. These WTP estimates, calculated
in accordance with the formulas above, are presented in Table 14.55
In this table, columns
3 and 4 present the WTP equivalence factors for respectively a positive and a negative
deviation of size 0.2 between the country’s average health score and the health reference
score. Likewise, columns 5 and 6 represent the WTP equivalence factors for respectively
a positive and a negative deviation of 2 percentage points between the country’s actual
55
For interaction terms for which the country-specific coefficient is insignificant, we have set the value of λ
or µ equal to zero in our calculations. Thus, these countries have the same willingness-to-pay estimates as
the United States.
76
unemployment rate and the reference unemployment rate.56
Recall for the interpretation
of the results in Table 14 that equivalent incomes can be obtained by multiplying
people’s actual income by these WTP equivalence factors.
Table 14 Country-specific willingness-to-pay estimates as derived from Table 13
D_US base LogY*US base Health*US base Unem*US base
Dummy Wave1 0.189***
(0.037)
Dummy Wave2 0.245***
(0.034)
Dummy Wave3 0.000
(0.037)
Age -0.038***
(0.003)
Age squared 0.052***
(0.003)
Dummy Male -0.125***
(0.015)
Dummy Married 0.471***
(0.024)
Dummy As Married 0.319***
(0.036)
Dummy Divorced -0.17***
(0.044)
Dummy Separated -0.518***
(0.063)
Dummy Widowed -0.127***
(0.040)
Observations 55266
Pseudo R-square 0.18
-2 Log Likelihood 189824
Numbers in parentheses denote standard errors. Three stars indicate significance at the 1% level, two stars at the 5% level and one star at the 10% level. An explanatory list of the country abbreviations can be found in Appendix E.
Table 15 in fact dissects some part of the general country fixed-effects captured
by the standard country dummies: a comparison of Table 13 and Table 15 shows that the
decrease in each country’s coefficient of the standard country dummy caused by the
inclusion of country-specific income effects is almost exactly equal to the country-
specific income coefficient times the country’s average value of the logarithm of income.
So, some part of the variation among countries that was previously accounted for by the
standard country dummies is now captured by the country-specific income effects.
Except for these changes, the values and significance levels of all other variables remain
80
nearly perfectly the same after the inclusion of the country-specific income effects. This
can be interpreted as a signal of the robustness of the regression results in Table 13.
As Table 15 shows, the American income coefficient is exactly equal to the
general income coefficient in Table 13. Furthermore, although it is difficult to distinguish
any particular patterns in the country-specific income effects, there indeed seems to be
some variation in income preferences across countries: compare, for instance, the
Netherlands or Australia with Spain or France. This observation supports our hypothesis
that it is relevant to allow for country-specific preferences for income when constructing
equivalent incomes. Table 16, which is similar to Table 14, next presents the willingness-
to-pay estimates based on the regressions results in Table 15.
Table 16 Country-specific willingness-to-pay estimates as derived from Table 15
United Kingdom 1.99 -2.67 27,627 23% 0% -21% -40% -93%
United States 1.85 -2.67 35,733 78% 45% 14% -12% -90%
The value of the health variable in column 1 has been calculated as the country’s average answer over wave 2, 3 and 4 (the period 1989-2004) to the question “All in all, how would you describe your state of health these days? 1-Very good, …, 5-Very poor”. The GNI per capita column represents each country’s average GNI per capita over the period 2000-2007. The corrections in the subsequent columns use the following reference values for the subjective health variable:
(1) the mean value of the health variable across the 21 countries (2.0662) (2) the median value of the health variable across the 21 countries (1.99) (3) the arbitrarily chosen value of 1.9 (4) the arbitrarily chosen value of 1.8 (5) the optimal health value of 1.0
reference values. This pattern is in particular exemplified by countries like Germany,
Italy and Portugal.
Finally, whereas the first four reference values underlying the corrections in Table
19 are quite close to each other, the last column presents the corrections based on the
ethically appealing reference value of perfect health for everyone within the country. For
this scenario every country experiences an extremely large negative health correction,
reflecting both the strong universal preferences for health as well as the relatively large
distance between the countries’ actual health scores and this maximum reference value.
One of the conclusions that can be drawn from Table 20 is that the unemployment
corrections are significantly smaller than the health corrections, irrespectively of how
ambitiously one sets the reference values. Interestingly, however, the percentage
unemployment corrections are still much more pronounced than the corrections for the
risk of unemployment in Fleurbaey & Gaulier (2009), presented in the previous chapter.
86
Table 20 Equivalent income corrections for unemployment, based on various reference values
Correction as % of GNI per capita
Unemployment
variable General WTP
factor GNI per capita (1) (2) (3) (4) (5) (6)
Australia 3.6% -1.508 26,769 -0.7% -1.7% -2.4% -3.2% -3.9% -5.3%
Austria 2.6% -5.432 26,752 3.4% -0.4% -3.1% -5.7% -8.2% -13.0%
United States 3.2% -1.508 35,733 0.0% -1.0% -1.8% -2.5% -3.2% -4.7%
Because the unemployment variable is a dummy variable, with a 0-1 scale, we can interpret the scores on this variable as a kind of unemployment rates (see also footnote 56). For a more convenient interpretation of the reference values, we therefore have expressed the values of the unemployment variable in this table as percentages. These numbers represent the average total unemployment as percentage of the total population aged 15 or older for the period 2000-2007. The GNI per capita column represents each country’s average GNI per capita over the period 2000-2007. The corrections in the subsequent use the following reference values for the unemployment variable:
(1) the median value of the unemployment variable across the 21 countries (3.19 %) (2) the arbitrarily chosen value of 2.5 % (3) the arbitrarily chosen value of 2 % (4) the arbitrarily chosen value of 1.5 % (5) the arbitrarily chosen value of 1 % (6) no unemployment at all, value 0 %
From this perspective, Table 20 may represent another indication that the market-based
corrections in Fleurbaey & Gaulier (2009) do not fully capture people’s preferences.
Anyway, for societies as a whole the existence of some unemployment seems to be
accepted as a fact of life. Though being unemployed is undoubtedly a serious adversity
for those who are directly confronted with it, for society as a whole the existence of
unemployment is apparently not perceived as a dramatic setback. Even for the most
ambitious reference values, the unemployment corrections exceed the 20 percent level for
hardly any country.
The most negative unemployment corrections are observed for Germany and
France, which suffer from the combination of relatively high unemployment levels and a
relatively high degree of unemployment aversion. Also some other Continental and
Mediterranean countries experience quite large negative corrections, in the case of the
87
Continental countries mainly due to their high level of unemployment aversion and for
the Mediterranean countries primarily the result of their relatively high unemployment
rates. Furthermore, whereas the Nordic countries have reasonably high levels of
unemployment aversion as well, these high levels are somewhat compensated for by the
lower levels of unemployment, thus moderating these countries’ unemployment
corrections (this holds especially for Denmark and Norway). Finally, the influence of
preferences can also clearly be seen in Table 20 by looking at the countries that
experience the least negative unemployment corrections. All of these countries share with
the United States a fairly low degree of unemployment aversion.
Tables 21 and 22 next present information on the ranking of the 21 countries
based on the various calculated corrections. In this context, the equivalent incomes can be
interpreted as each country’s average equivalent income over the period 2000-2007.
Since space constraints prevent us from elaborating extensively on these tables, we just
would like to mention the core messages conveyed by these tables. First, we can observe
significant differences between the country rankings in terms of uncorrected GNI per
capita and the rankings in terms of corrected GNI per capita. This holds in particular for
the health-corrected incomes. Moreover, the tables also show that, although the sample
range of equivalent incomes strongly depends on the chosen reference values, the
absolute country ranks with respect to equivalent income are quite robust to the choice of
different reference values.
Table 21 Country rankings in terms of health-corrected GNI per capita, based on the
health corrections in Table 19 (as % of sample mean and absolute ranks)
United States 137% 2 179% 2 187% 2 198% 2 209% 2 301% 1
88
Table 22 Country rankings in terms of unemployment-corrected GNI per capita, based on the unemployment corrections in Table 20 (as % of sample mean and absolute ranks)
Two stars denote significance at the 1% level; one star denotes significance at the 5% level. LEFT: dummy indicating whether the respondent is classified as having left-wing political preferences. RIGHT: dummy indicating whether the respondent is classified as having right-wing political preferences. RICH: dummy indicating whether the respondent is classified as being rich (i.e. belonging to the top two income deciles). POOR: dummy indicating whether the respondent is classified as being poor (i.e. belonging to the bottom four income deciles). LIFE SAT: the respondent’s answer to the WVS question: “All things considered, how satisfied are you with your life as a whole these days? (1-dissatisfied, …, 10-satisfied)”. UNEM: dummy indicating whether the respondent is unemployed. HEALTH: the respondent’s answer to the WVS question: “All in all, how would you describe your state of health these days? Would you say it is: 1-Very good, …, 5-Very poor?” AGE: the respondent’s age. MALE: dummy indicating whether the respondent is male. MARRIED: dummy indicating whether the respondent is currently married. AS MARRIED: dummy indicating whether the respondent is currently living together as married. DIVORCED: dummy indicating whether the respondent is currently divorced. SEPARATED: dummy indicating whether the respondent is currently separated. WIDOWED: dummy indicating whether the respondent is currently widowed. NKIDS: the respondent’s number of children.
104
Appendix C Life satisfaction regressions containing regime-specific inequality effects
The table below investigates the sensitivity of the regression results presented in Table 2.
In the regressions in the table below, four groups of countries have been included: an
Anglo-Saxon group (United States, Canada, United Kingdom, Ireland), a Nordic group
(Sweden, Norway, Finland, Denmark), a Continental group (Germany, Austria, Belgium,
Switzerland) and a Mediterranean group (Italy, Spain, Portugal, Greece).
Dependent Variable: LIFE SATISFACTION
Ordered Logit Regression Model
All things considered, how satisfied are you with your life as a whole these days?
Numbers in parentheses denote standard errors. Three stars indicate significance at the 1% level, two stars at the 5% level and one star at the 10% level. All regressions include the following control variables: inflation rate, unemployment rate, a dummy indicating whether the respondent is unemployed, the respondent’s perceived health status, the respondent’s sex, the age and the squared age of the respondent, dummies for the respondent’s marital status (married, living as a couple, divorced, separated, widowed) and the respondent’s number of children. Finally, wave dummies have been included.
105
Appendix D Average number of hours worked per country (2004)
Country Number of hours worked per capita
Belgium 609
Canada 875
Denmark 752
Finland 788
France 584
Germany 677
Greece 688
Iceland 752
Ireland 744
Italy 656
Japan 917
Korea 1147
Luxembourg 1005
Netherlands 700
Portugal 822
Spain 712
Sweden 773
United Kingdom 785
United States 879
Median 752
Source: OECD.Stat (stats.oecd.org)
106
Appendix E Explanatory list of country abbreviations
Table 13 and 15 of chapter four use abbreviations for the names of the countries included
in the regression analyses. The abbreviations refer the following countries:
AUSL - Australia
AUS - Austria
BEL - Belgium
CAN - Canada
DEN - Denmark
FIN - Finland
FRA - France
GER - Germany
IRE - Ireland
ITA - Italy
JAP - Japan
KOR - Korea
NET - Netherlands
NEW - New Zealand
NOR - Norway
POR - Portugal
SPA - Spain
SWE - Sweden
SWI - Switzerland
UK - United Kingdom
US - United States
107
References
Alesina, A., Di Tella, R., MacCulloch, R. (2004). Inequality and happiness: are
Europeans and Americans different? Journal of Public Economics, 88, 2009-2042.
Alesina, A.F., Giuliano, P. (2009). Preferences for Redistribution. NBER Working Paper,
no. 14825.
Alkire, S., (2008). The Capability Approach as a Development Paradigm? Working
Paper.
Amiel, Y., Creedy, J., Hurn, S. (1999). Measuring attitudes towards inequality.
Scandinavian Journal of Economics, 101, 83-96.
Atkinson, A.B. (1970). On the measurement of inequality. Journal of Economic Theory,
2, 244-263.
Atkinson, A.B. (1998). The Distribution of Income in Industrialized Countries. In:
Federal Reserve Bank of Kansas City, Income Inequality: Issues and Policy Options.
Kansas City: Federal Reserve Bank of Kansas City, 11-32.
Aristei, D., Perugini, C. (2010). Preferences for redistribution and inequality in well-
being across Europe. Journal of Policy Modeling, 32, 176-195.
Becker, G.S., Philipson, T.J., Soares, R.R. (2005). The Quantity and Quality of Life and
the Evolution of World Inequality. American Economic Review, 95(1), 277-291.
Becker, R.A., Denby, L., McGill, R., Wilks, A.R. (1987). Analysis of data from the
Places Rated Almanac. The American Statistician, 41(3), 169-186.
Canoy, M., Lerais, F. (2007). Beyond GDP. Overview paper for the Beyond GDP
conference, Bureau of European Policy Advisers (BEPA), European Commission.
Carlsson, F., Daruvala, D., Johansson-Stenman, O. (2005). Are People Inequality-Averse,
or Just Risk-Averse? Economica, 72(287), 375-396.
Clark, A.E., Frijters, P., Shields, M.A. (2008). Relative Income, Happiness, and Utility:
An Explanation for the Easterlin Paradox and Other Puzzles. Journal of Economic
Literature, 46(1), 95-144.
Cnossen, S. (2009). Hoe beschaafd is Nederland? Een fiscale kosten-batenanalyse. CPB
Bijzondere Publicaties, no. 83.
Decanq, K., Decoster, A., Schokkaert, E. (2009). The Evolution of World Inequality in
Well-Being. World Development, 37(1), 11-25.
108
Decancq, K., Lugo, M.A. (2010). Weights in Multidimensional Indices of Well-Being:
An Overview. Working paper, March 4, 2010.
Diener, E. (1984). Subjective Well-Being. Psychological Bulletin, 93, 542-575.
Easterlin, R.A. (1974). Does Economic Growth Improve the Human Lot? Some
Empirical Evidence. In: David, R., Reder, M. (ed), Nations and Households in Economic
Growth: Essays in Honor of Moses Abramovitz. New York: Academic Press, 89-125.
Esping-Andersen, G. (1990). Three Worlds of Welfare Capitalism. Princeton: Princeton
University Press.
Fleurbaey, M., Gaulier, G. (2009). International Comparisons of Living Standards by
Equivalent Incomes. Scandinavian Journal of Economics, 111(3), 597-624.
Fleurbaey, M., Schokkaert, E., Decancq, K. (2009). What good is happiness? Core
Discussion Paper, no. 2009-17.
Fleurbaey, M. (2009). Beyond GDP: The Quest for a Measure of Social Welfare. Journal
of Economic Literature, 47(4), 1029-1075.
Foster, J., Sen, A. (1997). On Economic Inequality. Oxford: Oxford University Press.
Frey, B. (2008). Happiness: A Revolution in Economics. Cambridge, Massachusetts: The
MIT Press.
Heath, A., Martin, J.P., Spreckelsen, T. (2009). Cross-National Comparability of Survey
Attitude Measures. International Journal of Public Opinion Research, 21(3), 293-315.
Heertje, A. (2007). Echte Economie (Translation: Real Economics). Nijmegen: Uitgeverij
Valkhof Pers.
Hindriks, J., Myles, G.D. (2006). Intermediate Public Economics. Cambridge,
Massachusetts: The MIT Press.
Inglehart, R. (1997). Modernization and Postmodernization: cultural, economic and
political change in 43 societies. Princeton: Princeton University Press. s
Inglehart, R., Baker, W.E. (2000). Modernization, Cultural Change, and the Persistence
of Traditional Values. American Sociological Review, 65, 19-51.
Kahneman, D. (2008). The sad tale of the aspiration treadmill. Edge – World Question