Exploring Wider Well-Being in the EU-15 Countries: An Empirical Application of the Stiglitz Report

Abstract

We draw on the recommendations of the Stiglitz Report to select a set of economic and social variables that can be used to make cross-country comparisons of wider well-being. Using data for the EU-15 countries for 1999 and 2005, we show how three-way analysis can be used to extract synthetic information from a large data set to determine the main latent explanatory factors. In our case, we identify one dominant factor that we term the development profile, which is positively associated with the level of education outputs, technological progress and female labour market participation and negatively associated with the level of pollution. We rank the countries according to this factor and compare these rankings with simpler GDP comparisons and find that the two rankings are only weakly correlated.

1 Introduction

The critique of GDP as an indicator of economic performance and well-being is well known, and recently received a rich and extensive restatement in the Report on the Measurement of Economic Performance and Social Progress by Stiglitz et al. (2009) (henceforth referred to as the Stiglitz Report). But this report, produced at the behest of the French President Nicolas Sarkozy and involving a large team of international experts, goes a lot further than repeating the well-known limitations of GDP as an economic and social indicator.

The Report also considers the additional information that might be required to produce “more relevant indicators of social progress¹” and comes up with eight dimensions of well-being, which should be considered together.² It also produces some Recommendations to guide the measurement of the dimensions of economic performance and social progress (henceforth termed ‘wider well-being’). As well as providing Recommendations on how to measure the material performance (or material well-being) of a country (economic production and living standards)—for example, to use income and consumption rather than GDP—the Report also provides guidance on assessing the quality of life of people living in a country.

According to the Stiglitz Report: “Quality of life is a broader concept than economic production and living standards. It includes the full range of factors that influences what we value in living, reaching beyond its material side” (p. 41). The measurement of this broad concept involves a metric of economic and non-economic resources, but also a theoretical framework that supports research on how to measure it in practice. Here the Report asserts that the measurement of the conditions of life, as well as economic resources, is a necessity for both developing and rich developed countries.

The Report endorses three main theoretical approaches, as a guide for going beyond measures of economic performance.³ The first is the Subjective Well-Being approach—popular in psychology but also applied to economics—which is based on the idea that happiness and satisfaction can only be judged by individuals. The second is the Capability Approach, which focusses instead on individuals’ capabilities to achieve higher levels of well-being, a concept originally introduced by Sen to give weight to the social aspects of development. However, as we discuss later in the paper, creating statistics fully based on this approach does not seem possible or straightforward. The third is Fair Allocation Theory, developed within the economics literature, which is also an attempt to account for non-market dimensions of the quality of life, in order to measure a wider well-being.

The research generated from the publication of the Report has not so far matched its ambition, although some authors have commented on it and related their work to some relevant aspects (see e.g. Easterlin 2010; Oswald 2010; Noll 2011; Rojas 2011). Yet there is evidently more to the agenda for research that the Report was aiming to achieve, as apparent from the following paragraph in the Executive Summary: “The Commission regards its report as opening a discussion rather than closing it. The report hints at issues that ought to be addressed in the context of more comprehensive research efforts. Other bodies, at the national and international level, should discuss the recommendations in this report, identify their limits, and see how best they can contribute to this broad agenda, each from its own perspective” (Stiglitz et al. 2009, p. 18).

In the context of our research, which compares countries’ wider well-being, our aim is to draw on the thinking in the Stiglitz Report as best we can to structure our empirical analysis. More specifically, we use the guidelines in the Stiglitz Report to select a set of indicators, which—given the statistics currently available—provide the best starting point for representing the multidimensional aspects of economic and social performance (i.e. wider well-being). Our specific empirical application consists of comparing the EU-15 countries using both this set of indicators of wider well-being and more familiar GDP measures, using three-way analysis (Lebart et al. 1995). We then explore the implications of making cross-country comparisons using measures of wider well-being. Although there is of course a larger literature on making international comparisons of economic performance, which we discuss later in the paper, as far as we are aware, our paper is the first to take the Recommendations in the Stiglitz Report seriously in selecting the relevant indicators for such analysis.

The rest of the paper is structured as follows. In Sect. 2, we review the Stiglitz Report and its contribution to the measurement of economic performance and social progress. This is followed by Sect. 3, where we survey earlier studies on comparing countries’ wider performance. In Sect. 4 we explain our approach and we discuss the results from applying the three-way analysis to the EU-15 countries. The final section provides some conclusions and policy implications.

2 The Stiglitz Report on the Measurement of Economic Performance and Social Progress

As discussed in the Introduction, the aim of the Commission associated with the Stiglitz Report was to review and highlight problems with GDP as an indicator, to produce more relevant indicators of social progress, to assess alternative methods of measuring progress and to provide guidance on the presentation of statistical data.

2.1 The Stiglitz Report’s Contents

As is apparent from these aims, the Stiglitz Report, written with the help of a large number of expert collaborators (economists and social scientists), does many things; from explaining the shortcomings of GDP as an indicator of economic and social progress, to discussing relevant theoretical frameworks used by academics to study wider aspects of well-being. It also provides a list of Recommendations for the wider audience of political leaders, policy-makers, academics, statisticians and for the general public of both developed and developing countries.

The Report does five main things:

• It attempts to integrate different disciplines, in order to obtain a wider understanding of well-being.

• It attempts to refocus the organisations in charge of producing statistics.

• It asks politicians to be more in touch with what really matters to people, and not only what a country produces.

• It acknowledges the issue of sustainability.

• It aims to set an agenda for research in this area.

According to the Report, the eight dimensions of well-being are: Material living standards (income, consumption and wealth), Health, Education, Personal activities including work, Political voice and governance, Social connections and relationships, Environment and Insecurity, “of an economic as well as a physical nature”. Moreover, it suggests that “at least in principle, these dimensions should be considered simultaneously” (Stiglitz et al. 2009, p. 14).⁴ Some of these dimensions are objective measures of economic performance (e.g. income and consumption); others represent objective features of people’s quality of life (e.g. health and education).

2.2 Theoretical Frameworks Underlying the Stiglitz Report

The various indicators and recommendations in the Stiglitz Report are linked to the committee’s endorsement of three main theoretical frameworks: Subjective Well-Being, Capability Approach and Fair Allocation Theory. Somehow these three theoretical frameworks all have an important role in the search for a broader measure of well-being, though as the Stiglitz Report states the “approaches may represent opposite poles in intellectual terms” (Stiglitz et al. 2009, p. 145).

(a) Subjective Well-Being is a term associated with research in psychology, but also applied to economics,⁵ and is based on the idea that happiness and satisfaction can only be judged by each individual and thus measurement is carried out through the evaluation of subjective self-reporting.

As the Stiglitz Report states, this approach is interesting because it relies on the goals of happiness and satisfaction, which are the simple aims of human beings across culture and time. The Stiglitz Report endorses three subjective dimensions of well-being, ‘life satisfaction’, the ‘presence of positive feelings’ and the ‘absence of negative feelings’ and not the single dimension of ‘happiness’. These three dimensions (and thus subjective well-being) contain additional information to objective indicators, for example GDP. “This suggests that these measures can play a useful role in measuring quality of life of people and groups as complements of other indicators” (Stiglitz et al. 2009, p. 148).

The measurement of the dimensions of subjective well-being involves evaluative judgement for ‘life satisfaction’ and measuring ‘hedonic experiences’ for ‘positive’ and ‘negative’ feelings. With the exception of the Gallup World Poll, which allows measurement of the three dimensions of subjective well-being, data have to be collected separately for each dimension. One of the most common measurements used is the ‘ladder-of-life’ scale that uses explicit reference points to measure an individuals’ life satisfaction, with 0 being the worst possible life and 10 being the best possible life. Qualitative measures assess life satisfaction through responses such as ‘quite’ or ‘fairly’ happy with one’s life. Less frequently used due to money and time restraints is the measurement of hedonic experiences that evaluates an individual’s feelings in real-time or shortly after an event has occurred (see Stiglitz et al. 2009, pp. 146–148).

The determinants of subjective well-being have been explored in the Stiglitz Report (pp. 148–149) and include adaptation, peer effects and relative comparisons. Adaptation is a critical aspect in determining subjective well-being, especially in certain settings and cultures. Adaptation may occur when individuals living in a deprived social setting, begin to accept their conditions of life and become less dissatisfied with their subjective well-being.

The Report argues that it is possible to be less satisfied with a better life, if one has a more ambitious standard. For peer effects and relative comparisons, the hypothesis is that income gains relative to other people within a community matter more for life-evaluations than country-wide improvements in absolute income.

Studies on subjective well-being are heterogeneous in nature and methods. Results will depend on the dimensions investigated and the objective of the research.⁶

(b) The capability approach moves away from people’s perceptions of well-being and focusses instead on individuals’ capabilities to achieve higher levels of well-being. It was originally introduced by Sen as an additional approach to welfare economics to give weight to the social aspects of development. It was then developed by Sen and others and became a paradigm in development theory. This approach is centred around the idea of ‘functional capabilities’, which are associated with human freedom instead of human access to income (or, in general, resources) or the concept of utility (poverty in this context is viewed as a lack of capabilities). Lack of capabilities may be associated with individual failure or objective constraints (for example, lack of political freedom or financial resources). In this context, human well-being is associated with: free choice, people’s heterogeneity and multidimensional welfare.

The capability approach is critically integrated in the Stiglitz Report, where there is an explanation of the terminology, a discussion on the foundations of the approach and an assessment of the various steps required for its practical implementation. However, creating statistics fully based on this approach does not seem possible or straightforward. Perhaps, this is because, as Robeyns (2005) points out the capability approach is not a theory, but a conceptualisation of the phenomenon of investigation and as such is interdisciplinary in nature and can be used to explain many things, including well-being. Thus, other explanatory theories need to be used when applying it. Although this approach focuses on functionings and capabilities, an analysis of well-being within this approach would still pay attention to resources and economic growth.

The extensive use of the capabilities approach in the Stiglitz Report is an extension of its original roots in development economics and in the need to understand the multidimensional aspects of deprivation, to the economics of developed and income abundant countries where there is a need to understand what else matters for well-being (or quality of life). And the extrapolation of the capability approach tells us that according to ‘capabilities’ and ‘functionings’ human well-being is multidimensional, in a similar way that deprivation is. Thus, away from traditional economic frameworks, the capability approach provides new thinking on assessing capabilities and the complex link between resources and utility. Because of its acceptance of the capability approach, the Stiglitz Report gives a more significant weight to human beings rather than institutions and markets.

The question is whether the capability approach is equally suitable as framework to study both deprivation and opulence. A common perception of well-being is more likely to be found when people are deprived of housing, drinkable water etc., i.e at the bottom of the income distribution. Beyond a certain income threshold the notion of better/higher well-being can be more diverse as some people, for example, prefer to have more leisure and work less.

Economics via the welfare economics and fair allocations approaches has also attempted to account for non-market dimensions of quality of life in order to measure wider well-being. The attempts have extended consumer theory, from its conventional focus on the consumption of goods and services to consumption of other dimensions of the quality of life. In these extensions, the idea that individuals have preferences and that these can be represented by indifference sets has been preserved (see Stiglitz et al. 2009, p. 153).

The concept of ‘willingness-to-pay’ has been useful in the context of applying monetary measures to non-market dimensions of life. This means that, given people’s preferences, it is possible to trade off different dimensions of quality of life as changes in each dimension involve income changes and thus it is possible to make them equivalent. For example, provided they are willing to pay, people can achieve a higher level of health or education (see Stiglitz et al. 2009, p. 154). However, the concept of ‘willingness-to-pay’ tends to favour the richer in societies because as income increases total ‘willingness-to-pay’ increases (see Stiglitz et al. 2009, p. 154).

(c) The fair allocations approach ⁷ extends the welfare approach by introducing some fairness criteria in the allocation of resources, for example the ‘no-envy’ criteria according to which “no agent should prefer another’s bundle” (Report, p. 154).

Within the fair allocations approach, for purpose of application, the ‘equivalence approach’ aims to choose a reference vector for each aspect of quality of life (see Report 2009; Fleurbaey 2009), using a monetary measure for non-market aspects of quality of life. Thus, the application of this method involves creating a reference set to compare individuals in terms of the equivalent income (which differs only in the income but not in the chosen dimensions of the non-monetary aspects of quality of life). There is also the need to measure preferences and, as this is not a straightforward exercise, there are various suggested ways to deal with this. But there are a number of criticisms directed towards money-metric utilities and indirectly towards the equivalence approach.⁸

2.3 The Stiglitz Report’s Recommendations

To make sense of the Report’s recommendations, we need to recall some of the thinking behind them. There is an emphasis on: limiting the use of commonly used statistics only to the phenomena we are trying to investigate; monitoring country’s inequality with respect to income distribution and other economic measurements; focussing on societal well-being (both economic and social) and thus to agree on what is relevant to people; closing the gap between the measurements of important economic aggregates (often made public through announcements) and people’s perception⁹ and to distinguish between measuring current (wider) well-being¹⁰ and sustainability.¹¹

With this in mind, the first Recommendation focusses on measuring material well-being by using income and consumption rather than production. The Report thus rejects Gross Domestic Product (GDP) as an indicator of a country’s performance because it measures only the goods and services produced. As the aim is to measure the material well-being of the citizens in the country, in the context of a globalised world, the Stiglitz Report embraces the concept of Gross National Product (GNP) and various derivatives of this concept, discussed in Chapter 1 of the Report. More specifically, the Stiglitz Report suggests that “material living standards are more closely associated with measures of net national income, real household income and consumption” (Stiglitz et al. 2009, p. 13).

The second Recommendation suggests that, to assess trends in material living standards, it is important to focus on measures of household income and consumption, while the third focusses on the importance of wealth (both for households and countries) to accompany income and consumption. Then, in tackling in distributional issues, the Stiglitz Report moves away from the most used concept of average, in this context average per-capita income, consumption and wealth and it recommends combining average measures with indicators that reflect their distribution; using, for example, median disposable income, as this is more relevant for the common perception of well-being.

Recommendation 5 points to the importance of including non-market activities to obtain a broader measure of income and this brings in the importance of leisure for measuring standards of living. And although measuring leisure is not a straightforward task, the Stiglitz Report highlights the need “to take into account the amount of leisure that people enjoy” (Stiglitz et al. 2009, p. 14) in cross-country comparisons of living standards and over time.

Recommendation 6 explains the importance of improving “measures of people’s health, education, personal activities and environmental conditions” (Stiglitz et al. 2009, p. 15), as the “quality of life depends on people’s objective conditions and capabilities” (Stiglitz et al. 2009, p. 15). And here, the Stiglitz Report integrates the conceptual framework of the capability approach in its request for developing measures of social connections, political voice and insecurity, provided that they have a proved link to life satisfaction.

The Stiglitz Report also explains the need for assessing inequalities via the indicators of various dimensions of quality of life or via separate measures of inequalities (viz. Recommendation 7). The remaining Recommendations focus either on sustainability (which is beyond the scope of our cross-country comparison) or are aimed at prompting statistical organisations to create more appropriate measures of wider well-being, where both objective and subjective dimensions are accounted for (e.g. by designing special surveys).

2.4 Other Papers on the Stiglitz Report

Several papers have commented on various aspects of the Report. Easterlin (2010), for example, focusses on the Report’s message that GDP (and GNP) is deficient as a measure of social progress and also on the Report’s advocacy of subjective measures of well-being. This, he comments, is in sharp contrast to the hostile nature of economists towards questionnaires and self-descriptions, but the Report and its authors are moving away from this attitude, which has represented the economics literature and its followers for many decades. He then focusses on the multiple dimensions of well-being portrayed in the Stiglitz Report and points out how the assessment of development of a high growth country, say China, might be reassessed if other indicators come into play, for example, overall life satisfaction. Noll (2011) argues that some of the Report’s recommendations are well known from a social indicators perspective. Nevertheless, he welcomes the Report as opening a new debate on well-being and progress but offers no new analysis. Rojas (2011) focusses on quality of life and how to strengthen the impact of the Report on this subject. He argues that the Stiglitz Report is weak on the concepts of quality of life and well-being and explains that the background for this weakness is, with a few exceptions, the lack of progress by the scholars of quality of life towards a common understanding of this concept. Similarly, the Report provides Recommendations on measuring quality of life but abstains from its conceptualisation. Rojas then goes on to discuss his views on an agenda for quality of life scholars. Emotional prosperity, defined as “well-being measured in human terms…rather than in pound notes of GDP…” is the focus of Oswald (2010, p. 20). In this paper, he also refers to the Report and argues that, given that material prosperity is now remarkably high, and given the spirit of the Stiglitz Commission, we should focus on emotional prosperity. He justifies this on the grounds of Easterlin’s Paradox: that as people get richer, their life satisfaction does not change significantly. This may be associated with the changing nature of the work environment which the Stiglitz Report also acknowledges.¹²

None of these papers have attempted to draw on the Report’s recommendations to frame new empirical analysis as we do in this paper. But there is of course a large previous literature on making cross-country comparisons of economic and social performance. In the next section, we discuss the most relevant previous studies that have aimed at measuring well-being using alternative indicators beyond GDP or by using multivariate analysis.

3 Previous Studies on Measuring Countries’ Economic Performance and Well-being

The main contributions in the empirical literature on measuring economic and social performance either collect information on social and economic variables and attempt to construct a composite indicator (see e.g. UNDP 1990) or collect a large set of variables and analyse these data with statistical methods, which allow the extraction of synthetic information, viz. the multivariate approach.¹³

Many scholars address the hypothesis that GNP (or GDP) per capita cannot be used as the only indicator of the performance of a country because it does not capture the real-life conditions of the population and it does not consider the consequences of economic development on the lives of people (e.g. air, sea and water pollution, increases in certain rare diseases, congestion, cost of urbanization, etc.).¹⁴ Regarding these aspects, the challenging assertion of Kuznets is noteworthy: “The most distinctive feature of modern economic growth is the combination of a high rate of aggregate growth with disrupting effects and new problems” (Kuznets 1973, p. 257). This statement implies that the national accounting framework should be expanded so that it considers both the costs (i.e. pollution, urban concentration, commuting, etc.) and the benefits of economic growth (i.e. better health, greater longevity, more leisure, less income inequality, etc.).

More recently, Hobijn and Franses (2001) drew scholars’ attention to the need to extend the evaluation of a country’s performance to encompass relevant measures of living standards. In so doing, they readdressed the spatial convergence issue—so prominent in the economic growth literature—and presented evidence that convergence in GDP does not necessarily imply convergence in living standards, the latter being defined by daily calorie and protein intake, infant mortality, life expectancy at birth, and so forth. GDP is at best only a partial measure (or proxy) of a multi-dimensional welfare concept incorporating both the economic and the non-economic aspects of human life (see Sen 1985, 1987; Dasgupta 1990).

In the light of these studies, since the 1990s there have been some new attempts in the literature to come up with more appropriate indicators. The first was the World Bank’s Human Development Index (HDI), a composite indicator based on GDP per capita, life expectancy at birth, and the adult literacy rate (UNDP 1990). These features represent, respectively, the three main aspects of an individual’s life, viz:. access to resources; health conditions; and the opportunity to enjoy a basic education. It was inspired by Sen’s development theory, according to which a country’s development is a matter not only of long-run economic growth but also of opportunities for people, in both the high and the low growth cycle (Sen 1984).

The HDI made popular by the UNDP first Report, was criticized because of the simple weighting of each variable, and the high correlation between GDP and certain important background variables. Indeed, it has sometimes even been considered a redundant indicator that provides little additional information on inter-country development levels with respect to traditional GDP (McGillivary 1991; Desai 1991; Dasgupta and Weale 1992; Sagar and Najam 1998). Nevertheless, the framework for calculating the index has remained substantially unchanged in UNDP’s subsequent annual reports; with only a few corrections introduced to take account of gender differentials or income distribution.

The related literature in the 1990s comprised a number of critical proposals for the improvement of the HDI. For example, since the indicators of the three dimensions of HDI were closely correlated, a principal component method was proposed in order to use a linear combination of these indicators (Noorbakhsh, 1998; McGillivary 1991).

Further, Sagar and Najam (1998) proposed a more in-depth revision of HDI involving multiplication of the three component variables instead of using their arithmetic average, a logarithmic treatment of GDP, and the incorporation of an inequality measure into the index. In fact, the second Report calculated the distribution-adjusted HDI for 53 countries (UNDP 1991, pp. 17–18) and this was available until 1994, although since that year this measure has been omitted.

Notwithstanding its limitations, the HDI is particularly relevant to developing countries, where the basic dimensions of well-being depicted by the three indicators have yet to be fully achieved. By contrast, in the developed countries, a decent standard of living, longevity, and primary education have already been achieved by most people. Consequently, additional indicators that take account of different aspects of living appear to be necessary.

In a study of seven OECD countries over the period 1980–1996, Osberg and Sharpe (2002) propose an Index of Economic Well-Being based on four dimensions: consumption flows; the stock of wealth, including physical capital and natural resource; income distribution; and economic security. This index is compared to GDP per capita. Their key finding is that economic well-being has increased over time much less than real GDP for most of the countries (Norway is an exception as, for this country, both GDP per capita and the index of economic well-being have a similar trend).

In 2006, Marchante and Ortega used an alternative augmented composite indicator (AHDI) to the HDI, in order to measure the quality of life and economic convergence across Spanish regions. In particular, they considered three different per capita income measures (total personal income minus grants, GVA, and total disposable income) and six quality of life indicators (life expectancy at birth, the infant survival rate, the probability at birth of surviving to the age of 60, the adult literacy rate, the mean years of schooling of the working age population, and the long-term unemployment rate). Moreover, they used an averaged arithmetic mean scheme with (arbitrary) weights, for the variables transformed, to obtain the AHDI.

To overcome the problem of giving arbitrary weights to the variables in composite index, Berenger and Verdier-Chouchane (2007) propose two composite indices, Standard of Living (SL) and Quality of Life (QL) as an alternative to HDI using two methodologies, Totally Fuzzy Analysis (TFA) and Factor Analysis of Correspondence (FAC). The weights given to the component indicators are different for the two methods: TFA derives weights from ‘frequency of symptoms of under-development’, FAC uses a weighting system similar to principal component analysis where the weights are calculated from various concepts of variance. Their results, for the African countries, suggest deficiencies in many aspects of well-being with exception of the quality of the environment and point to the importance of education as a key indicator for development. The standard of living (SL) is measured by nine variables on education, health and material well-being. These variables reflect the ‘commodities’ and are a measure of goods and services in the economy and the services produced by GDP. Quality of life (QL) is measured by nine variables on education, health and environment focussing on the quality of these domains.

Since the 1980s—after the creation of the United Nations World Commission on Environment and Development—some economists have highlighted that the environment, like the social aspects of life, is an essential element of well-being or country performance.

In 1989 Daly and Cobb, proposed the first Index of Sustainable Economic Welfare (ISEW). This was an attempt to integrate the economic aspects of an economy, as depicted by the conventional national accounting, with the social (i.e. income distribution inequality) and the environmental (i.e. air and water pollution) aspects. ISEW was criticized (see, e.g. Neumayer 1999, 2000) for the arbitrary selection of its component variables and for the method of aggregation and construction. After that, various indices, such as the Living Planet index, the Ecological Footprint, the Environmental Performance Index and so forth were proposed (see, e.g. Bohringer and Jochem 2007).

The appropriate way to define a multidimensional index of sustainability, combining the economic, social and environmental aspects of human life, is still subject to debate among ecological economists (see, e.g. Pulselli et al. 2006; Distaso 2007).

The assessment of the environmental aspects is potentially very important in developed countries where growth and technological progress may become ‘uneconomic’, worsening the life of citizens, through, for example, air and water pollution. In developing countries, policies towards the environment constitute a potential plus point for those governments. Therefore, the introduction of measures of environmental aspects in the context of a multidimensional assessment could produce a very different country ranking.

Although a number of efforts have been made to obtain a more comprehensive index of multidimensional well-being or country performance, many methodological issues still need to be explored more deeply, concerning how to integrate the above-mentioned different aspects in a unique measure (i.e. a composite indicator). Furthermore, as discussed previously, the main problem related to the construction of a composite indicator concerns the weights to be given to the variables composing the well-being indicator. An alternative approach is to perform a multi-dimensional analysis as, for example, in Cuffaro et al. (2008). In this paper the authors argue that the analysis of a region’s performance cannot be limited solely to either economic or social aspects, so they attempt to combine both material (i.e. economic) and non material (i.e. social) aspects of welfare and well-being. Using Principal Component Analysis applied to data on the Italian regions they show that a high level of economic welfare does not necessarily translate into a high level of social well-being. Cracolici et al. (2010) propose instead a simultaneous equation system approach that captures the relationships between the economic and non-economic aspects of country well-being. They estimate the model using data on 64 countries over the period 1980–1999. Their findings suggest that most countries are unable to convert higher educational skills into higher economic performance and only a few countries have been able to combine favourable economic and environmental outcomes.

As discussed, several previous studies have attempted to measure economic performance and well-being, but few have focused at the EU level. Our paper differs in that we draw on the Stiglitz Report’s Recommendations in framing our empirical analysis of the EU-15 countries.

4 An Application of Three-Way Analysis to the EU-15 Countries

In this paper, we use three-way analysis to explore wider well-being in the EU-15 countries. Three-way analysis is a relatively novel statistical approach to economics (most applications have been in chemistry and psychometrics)¹⁵ that allows us to examine data across three dimensions. In our context these dimensions represent time, different countries and different economic, social and environmental variables.

4.1 An Interpretation of the Indicators Associated with the Recommendations in Stiglitz Report

As the Stiglitz Report reviews and captures different views, putting forward a list of variables is not an easy and objective task. Unfortunately, some of the recommendations refer to improvements in measurement systems, which will only lead to improved statistics in the future. We focus on the Eurostat dataset (1999, 2005) available on the EU countries and choose a list of key variables, guided by our judgement of the recommendations, discussed in Sect. 2, given the currently available statistics for the EU-15.¹⁶

To evaluate the evolution of socio-economic performance of the EU-15 countries, we consider the years 1999 and 2005. This choice of years reflects both economic and statistical reasons. The first year coincides with the introduction of the euro, which might have been expected to represent a structural break, and therefore provides a natural starting point for the analysis. The end date is dictated by the availability of data: 2005 is the most recent year for which consistent data are available in the disaggregation we require.

To perform the statistical analysis, we use 35 variables which allow us to take into account several aspects of a country’s performance, including economic, social and environmental factors. To represent material well-being, we use disposable income and 12 variables related to consumption. In addition, we use the Gini coefficient to capture income inequality within each country. We also consider variables linked to health, environmental conditions, education, security and safety, and gender differences with respect to education and the labour market.

We also use other variables linked indirectly to material well-being and representing directly the competitiveness of a country. Higher competitiveness is associated with higher standards of living and higher levels of people’s well-being. In other words, a highly competitive country allows citizens to benefit from better to more stable jobs, higher incomes and consumption, better services, good health, etc. We choose to represent competitiveness by real labour productivity, real unit labour costs, purchasing power parities and the number of high-tech patents.

Summary statistics of the variables are presented in Table 1. For details on the list of variables and countries see Appendix 1.

Table 1 Descriptive statistics

4.2 Three-Way Analysis

Multi-way analysis allows us to explore and interpret multidimensional datasets; for example, in the case where there are a large number of variables and these relate to more than one statistical unit (e.g. countries, regions, individuals etc.). It is an extension of factor analysis (FA) or principal component analysis (PCA), which can only be used to analyze two-dimensional matrices.

Multi-way becomes three-way analysis if, as in our case, we have a data set with three dimensions, viz. countries, variables and time. To implement our three-way analysis we use the STATIS method proposed by L’Hermier des Plantes (1976) and Escoufier (1980).¹⁷ The STATIS method allows us to analyze multiple data tables, each containing information from the same set of individuals (in our case EU-countries). The differences and similarities between these tables are analyzed in terms of a common structure called the compromise (for details see Appendix 2).

In our case we want to analyze a large number of selected variables for the EU-15 countries over the 1999 and 2005 period. As with FA or PCA in a lower dimensional context, the aim is to extract a limited number of unobserved or latent components (or factors) from the observed variables chosen to capture most of the information contained in the original variables, in the sense of explaining the maximum amount of variance in the data. And, as with FA and PCA, it is important to ensure that the identified components (or factors) are interpretable, in order to reach some insights on the phenomenon being investigated.

In general and the method is applied by carrying out the following three steps (Appendix 2 contains more details):

• The first step involves comparing the structure of two-dimensional matrices in the different time periods, in order to discover if a relationship exists between them. This involves calculating the RV coefficient, which shows how similar the variance–covariance matrices are in different periods. The RV coefficient ranges from 0 to 1. The closer the RV coefficient is to 1 the greater similarity between the variance–covariance matrices across time. In our case, the RV coefficient allows us to assess whether or not there are significant changes in the relative position of the European countries over 1999 and 2005, thus an RV coefficient close to 1 would indicate little change. PCA is then applied to the matrix of RV coefficients, to calculate the weights used to compute the compromise space.

• The second step involves applying PCA to the compromise space in order to obtain the position of units in this space. Similarly, we can also obtain the position of variables in this space. In our case, STATIS allows us to look at the position of each EU country in each time period. In order to choose the number of latent factors, there are two general criteria: the first one looks at eigenvalues greater than 1; the second one is based on the change in the explained variance between two successive eigenvectors, where the size of the change guides the choice of factors (see, for example, Lebart et al. 1995).

• The third step involves plotting of the coordinates of the different units (i.e. in our case countries or variables) in the compromise space, in order to visualize the path of units across time periods. We neglect this step in our analysis, as we only have data for two years.

4.3 Results for the EU-15 Countries

Here we proceed to discuss our empirical results. As the RV relation coefficient regarding the two years is equal to 0.94, we can say that the structure of the variables representing the countries in the years 1999 and 2005 does not change and the mutual position of the countries is stable. On the face of it, this is surprising as it suggests that various policy initiatives over the period, including the introduction of the euro, have not lead to material changes in measures of wider well-being.

The analysis shows that the first five latent factors associated with the compromise matrix explain about the 74.0% of the variance in the data. In order to decide on how many factors we need to interpret, we use the criteria mentioned before. Since the first factor has a large eigenvalue, equal to 1.98, and explains 32.0% of total variance, this suggests it plays a relevant role in explaining the variance in both 1999 and 2005. In contrast, the second factor and successive ones have smaller eigenvalues less than one and contribute much less to explaining the overall variance (see Table 2). We therefore focus solely on the first factor.

Table 2 Eigenvalues and explained variance

The relationship between the latent factor and the variables shows the first factor is positively correlated to variables linked to a country’s competitiveness in terms of high-tech development (i.e. HTP, ICTP, BTP) and quality of human capital (PISA_LET, PISA_MAT), and it is negatively linked to shares of consumption expenditures, like food, clothing and restaurants and hotels (FNAB, CFW, RH), but positively to recreation and culture consumption (RC). We call this factor the ‘development profile’. This interpretation is also confirmed from the negative sign and the magnitude of factor loadings relating to female unemployment rate (YUF), the total level of pollution (APOL) and the Gini coefficient (GINI). It is also consistent with the positive sign on the number of crimes recorded by the police (CRIME)—it seems plausible that the level of crime and the level of development are positively related—the male and female participation rate in education (MPRE and FPRE), and total public expenditure on education as share of GDP (TPEE) (see Table 3).

Table 3 Correlations between 1st factor and variables

We interpret high values of this factor as indicating a favourable position. It indicates the ability of a country to increase the quality of its development by means of investment in technology, innovation and human capital and by policies aiming at favouring female labour market participation, and reducing income inequalities.

In short, the first factor represents the ability of a country to achieve a higher development profile, indicated by a competitive and dynamic knowledge-based economy, capable of sustainable economic growth. In principle, this can be facilitated by policies aimed at enhancing the information society and research and development, in order to increase competitiveness, innovation, employment, economic growth and social cohesion. So the insights coming out from the analysis are consistent with the strategic goals defined by the Lisbon European Council in 2000; viz. to build a competitive and dynamic knowledge-based economy capable of sustainable economic growth.

The empirical results highlight that the European countries have not uniformly taken up the challenge of the knowledge society of the second millennium. Some of them have engaged in improving the quality of human capital in terms of both mathematical and literacy skills of pupils, and raising the education participation rate. Moreover, they have also increased public expenditure on education and research and innovation in information technology, as showed by the competitiveness and high-tech variables. But some other countries have done less to improve their development profile.

We show these results graphically by plotting the rankings of countries in the two years according to both their loadings on the first factor and their real GDP per capita. In doing this exercise, we aim to verify if there is a strong (low) correlation in the two rankings; for example, if a country has a high ranking both on GDP (i.e. a high level of GDP) and on the first factor we could conclude that GDP has been efficiently used to reach a high development profile.

Figure 1 shows the rankings related to the year 1999; generally, we find a positive weak correlation between the two rankings. To get a better understanding of this, we divide the plane into four quadrants. In the top-right corner, we find countries characterized by both a high level of factor 1 and a high GDP per capita (i.e. HH countries). Whilst in the bottom-left corner of Fig. 1, we find countries with a low level of both variables (i.e. LL countries).

Fig. 1

Development Profile and GDP Rankings (1999)

The top-right corner includes countries that have the best relative performance, like Sweden, Belgium, Austria, and Netherlands and Denmark. The last two countries are positioned on the bisecting line indicating the same position on both rankings. These countries represent best practice in terms of ability to allocate economic resources to improve their development profile.

On the bottom-left corner of Fig. 1, we find countries that have lower rankings in both their development profile and per capita GDP; this quadrant includes Italy, Spain, Greece, Portugal and, and to a lesser extent, France.

In summary, the countries positioned in the top-right and bottom-left quadrants have a balance between economic performance and quality of development; in other words high levels of GDP per capita combine with high levels of development and vice versa.

The most virtuous country is Finland, which has a similar development profile to Sweden, The Netherlands and Denmark, despite having a lower level of GDP. Finland therefore shows a good ability to convert its resources into a high level of educational attainment (e.g. PISA scores), a high education participation rate, low levels of youth unemployment rate, etc. The countries with the worst performance are Ireland and Luxemburg. Although, they have a high level of GDP per capita, they have a low development profile.

Turning to the rankings for 2005, Fig. 2 shows a generally similar position of the countries, with only some exceptions; furthermore, it shows the weak correlation between the two rankings as in 1999. In particular, Finland and Germany reach a higher rank score on the development profile, while Belgium has a similar position to 1999. Although its GDP rank score is lower, its position in the development profile rankings remains the same.

Fig. 2

Development Profile and GDP Rankings (2005)

Thus the countries belonging to the best practice group maintain the same level of development profile, despite their deterioration in the GDP ranking. The best practice group is composed of the same countries and generally these countries show a high development profile; despite their deterioration in the GDP ranking. Similar considerations apply to the countries with low rankings with regard to their development profiles and GDP per capita, though the number of countries placed on the bisecting line increased, indicating a more balanced position. Finally, the composition of the worst performing countries is unchanged, though Ireland improves its position in both rankings, but the gains are smaller in the development profile ranking (Table 4).

Table 4 Development profile and GDP rankings (1999 and 2005)

A non-parametric test (i.e. the Spearman rank order correlation) relating to the two rankings of both years confirms the weak correlation of GDP and the development profile;¹⁸ but we cannot infer anything about the direction of the relationship, although we can imagine the existence of a two-way relationship. This is because GDP per capita is only a necessary condition to reach a basic threshold of economic development; after this threshold the ability of a country to transform its economic resources into social progress or development profile becomes more important. Vice versa, a high development profile increases GDP per capita through higher levels of productivity promoted by technological progress, ICT usage and human capital.

The empirical findings indicate that most EU countries, notwithstanding their levels of GDP, are still some distance away from the strategic goals of the Lisbon European Council; policy makers of these countries have to work hard in order to produce suitable policies directed at achieving the knowledge society and a good level of development for their citizens.

5 Conclusion and Policy Implications

In this paper we drew on the recommendations of the Stiglitz Report to select a set of variables of wider well-being that could be used to make cross-country comparisons. Using data for the EU-15 countries for 1999 and 2005, we showed how three-way analysis can be used to synthesize the information associated with a large number of variables, in order to determine the main latent explanatory factors. We then ranked countries according to the first latent factor and compared these rankings with simpler GDP comparisons.

The three-way analysis suggests that countries’ wider well-being is associated with the dimension of material well-being, both negatively (consumption expenditure on items like food, clothing, restaurants and hotels) and positively (recreation and culture consumptions, and ICT levels). It is also positively associated with the education and the insecurity dimensions of well-being. It is negatively associated with the environment dimension and the total female unemployment rate, a multidimensional variable. More specifically, our analysis suggests that the quality of education, measured by the PISA scores and the level of technological progress play a large role in the determination of the first factor, which we term the development profile.

In summary, the multivariate analysis suggests that the relative ranking of the EU-15 countries’ wider well-being changed little between the two years, despite various policy initiatives over the period, including the introduction of the euro. We also find that the GDP ranking of the EU-15 countries is only weakly correlated with the ranking associated with the development profile. So the GDP and development profile comparisons show that if we relied only on the GDP measurement we would achieve misleading conclusions. In the spirit of the Stiglitz Report, comparisons of economic performance and social progress have to extend to societal factors (e.g. education, unemployment, and innovation) and beyond simple GDP measures.

The detailed comparisons show that the best performing countries are Finland, Sweden, The Netherlands and Denmark. These countries are better able to allocate economic resources, contributing to the knowledge society and their development profile. Among the so-called rich countries, there are some countries—Italy, Spain and to a lesser extent France—that have significant deficiencies in their development profile, irrespective of their level of GDP. Rectifying these deficiencies will require policies aimed at enhancing the information society and research and development, in order to increase competitiveness, innovation, employment, economic growth and social cohesion. These policy implications are in line with the strategic goals defined by the Lisbon European Council in 2000; viz. to build a competitive and dynamic knowledge-based economy capable of sustainable economic growth.

Our analysis has been dictated by data availability. The available variables for the measurement of the multi-dimensions of wider well-being require further development and further progress will be possible as new statistics associated with the Stiglitz Report become available. However, the statistical analysis in our paper provides an illustration of how the concepts of the Stiglitz Report might be made operational in applied research on country comparisons, in ways that go beyond simple GDP criteria.

Appendices

Appendix 1: Countries and Variables

See Tables 5 and 6.

Table 5 List of countries

Table 6 List of variables

Appendix 2: STATIS

STATIS (Structuration des Tableaux a Trois Indices de la Statistique)—attributed to Escoufier (1980) and L’Hermier des Plantes (1976)—allows us to analyse a dataset represented by a matrix, X, composed of a three dimensions: time (for k = 1, 2,…K time periods), variables (for j = 1,2,…J variables) and units (for i = 1,2,…I units). Thus, each matrix I × J _k of data (X _k ), defines the structure for all units in the kth time period, and this depends on their respective position as defined by the distances between each pair of units:

$$ d_{k}^{2} (i,i^{\prime } ) = \sum\limits_{j = 1}^{J} {\left( {x_{ijk} - x_{{i^{\prime } jk}} } \right)^{2} } $$

The goals of STATIS are the following: (a) to compare and analyze the relationship between the different data matrices; (b) to combine the matrices into an average structure (compromise) on which PCA is applied; (c) to project the trajectories of units and variables in a compromise space.

In order to derive the compromise structure, STATIS generates some weights that are applied to each of the data matrices. Applying PCA to the compromise structure one derives the position of the observations in the compromise space.

STATIS is composed of three main steps: the inter-structure analysis, the intra-structure one and the project the trajectories of units and variables in a compromise space. The first one involves the analysis of correlation among variables or time periods (objects). This is done by comparing the structure of K matrices, each one to the others. To do this, each matrix is considered as a unique statistical object and the variance–covariance matrix W _k —reflecting the similarities between objects within each matrix—is computed:

$$ W_{k} = X_{k} Q_{k} X_{k}^{{^{\prime } }} $$

where X ^′ _k is the transpose of data matrix X _k , and Q _k is a matrix of dimension J _k  × J _k with the diagonal elements equal to 1/J _k . In our case, as we have the same number of variables in each time period, Q _k is an identity matrix of dimension J × J.

The similarities between two variance–covariance matrices W _k and W _k’ can be computed as follows:

$$ \Upomega_{{kk^{\prime } }} = (W_{k} ,W_{{k^{\prime } }} ) = {\text{trace (}}W_{k} DW_{{k^{\prime } }} D) $$

where D is a matrix of weights. A measure of closeness between the variance–covariance matrices is provided by the RV coefficient defined as:

$$ RV(W_{k} ,W_{{k^{\prime } }} ) = \frac{{W_{k} W_{{k^{\prime } }} }}{{\sqrt {(W_{k} ,W_{k} )(W_{{k^{\prime } }} ,W_{{k^{\prime } }} )} }}. $$

The RV coefficients range from 0 to 1. If the RV coefficients are close to 1 the variance–covariance matrices are very similar. Applying PCA to the RV matrix, we can represent in the space of principal components the similarities among data matrices concerning different time periods.

The elements of the first eigenvector that come out from PCA are normalized in such way that their sum is equal to one. In the intra-structure analysis, they are used as weights (v _k ) to define the so called ‘compromise’ among k time periods, that is:

$$ W = \sum\limits_{k = 1}^{K} {v_{k} W_{k} } . $$

The compromise matrix is then analysed to explore the common structure in the data.

Finally, plotting the coordinates of the different units in the compromise space allows us to visualize the path of units across time periods.