4.6 A dashboard approach requires a solid framework to select relevant dimensions and indicators. The statistical series are made available and explained using visualisation techniques (and/or a background database) rather than by reducing the information to a single composite measure.

4.7 A composite indicator is constructed by aggregating multiple individual indicators into one composite measure of a complex or multidimensional phenomenon (OECD 2008, p. 13; UNECE 2019, pp. 67-68). Well-being is an example of such a phenomenon.

4.8 In a formative framework in which well-being is the result of multiple indicators, the selection and aggregation of indicators requires a model that explains how the individual indicators are related to the complex, multidimensional phenomenon. A well-being framework is such a model: it separates the entire phenomenon into constituent dimensions, explains how each dimension is related to overall well-being and identifies the indicators that are to be aggregated into a composite measure.

4.9 Weights are required to combine individual indicators into an aggregate measure. The weights indicate how much the individual indicators contribute to the aggregate outcome. When all the components of a composite indicator are assigned equal weights, this does not mean that the composite is an unweighted index. All components are assumed to be equally important.

4.10 Different stakeholders and other users have different needs. Some users and some purposes are better served by the dashboard model, while other users and purposes benefit more from a composite indicator. Both may be useful in different contexts. A composite indicator can be an effective way to communicate the results of a complex, multi-indicator dashboard to the media, politicians and policy-makers, but this can depend heavily on individuals and their preferences. Decision makers may have different preferences and interests, and in their debates and decisions, they inevitably have to make trade-offs and understand the synergies between different aspects of well-being.

4.11 It is important to be aware of and transparent about the underlying assumptions and limitations. In general, carefully considering the assumptions and limitations involved in constructing a composite indicator facilitates responsible use of that indicator.

“The absence of an 'objective' way to determine weights and aggregation methods does not necessarily lead to rejection of the validity of composite indicators, as long as the entire process is transparent. The modeller’s objectives must be clearly stated at the outset, and the chosen model must be tested to see to what extent it fulfils the modeller’s goal.”

(OECD 2008, p. 33)

4.12 This chapter provides a general overview of the methodological steps that are involved in constructing a composite indicator of well-being. This overview builds on the guidelines of the OECD and UNECE. The aim is to ensure a degree of international comparability and methodological consistency. The explanation is deliberately not technical so that countries with less sophisticated statistical systems can also apply it. Producers who need more advanced methods and techniques can find detailed explanations in the existing technical handbooks.

4.13 The existing guidelines divide the process of constructing a composite index into ten steps (OECD 2008, pp. 20-21; UNECE 2019, pp. 70):

1. Developing a conceptual framework as the basis for the selection and combination of variables into a meaningful composite indicator.

2. Selection of data, based on the analytical soundness, measurability, country coverage, and relevance of the indicators to the phenomenon being measured and their relationship to each other.

3. Preparation of data and imputation of missing data, in order to provide a complete dataset.

4. Multivariate analysis, to study the overall structure of the dataset, assess its suitability, and guide subsequent methodological choices, such as weighting and aggregation.

5. Normalisation of data, to render the variables comparable.

6. Weighting and aggregation, along the lines of the underlying theoretical framework.

7. Uncertainty and sensitivity analysis, to assess the robustness of the composite indicator in terms of the mechanism for including or excluding an indicator, the normalisation scheme, the imputation of missing data, the choice of weights, and the aggregation method.

8. Evaluating plausibility and validity, to examine whether the composite indicator gives a good and reliable measure of the phenomenon it is intended to represent.

9. Links to other statistics, to correlate the composite indicator (or its dimensions) with existing (simple or composite) indicators, as well as to identify linkages through regressions.

10. Communication and visualisation, to help users make correct interpretations of the composite index and help avoid misunderstandings.

4.14 The use of an international framework for selection of dimensions and indicators is recommended. Individual indicators can be aggregated into composite indicators as described below, following the structure of the underlying model. Sustainable and inclusive well-being is a broad phenomenon that covers three domains: well-being ‘here and now’, ‘later’, and ‘elsewhere’. Stiglitz, Sen and Fitoussi note that measures of these domains should not be aggregated into a single composite measure attempting to cover all three domains.

4.15 A composite indicator of overall well-being ‘here and now’ can be constructed in two steps. Firstly, by weighting and aggregating the individual indicators of each dimension into composite indicators of the dimensions. Secondly, by weighting and aggregating the composite indicators of the dimensions into one composite indicator of overall well-being. Hence, constructing a composite indicator of overall well-being requires weights of the individual indicators and the dimensions.

4.16 The weights should reflect the relative importance of the components that together make up the composite indicator. To some extent, the weights may be assigned based on the conceptual framework or the intended use of the indicators. However, in most cases, weights will have to be estimated. The challenge is to find a method to expose the relative importance that people as a collective place on the components of the composite measure. There are several methods to estimate the weights that aim to avoid bias or a priori opinions by the compiler of the indicator (OECD, 2008, pp. 31-33; UNECE 2019, pp. 75-78). The most used methods are briefly presented below.

Equal weighting

4.17 Equal weighting can be applied at both aggregation steps: from individual indicators to dimensions and from dimensions to the overall composite measure. Equal weights may be applied if the components can be considered equally important, based on the conceptual framework and the purpose of the composite indicator. Equal weighting will often be a suitable solution and perform well compared to other methods. It also has the advantage of being transparent and easy to explain to users.

4.18 A composite indicator of overall well-being may include 8-10 dimensions that are different in nature and cover very different aspects of well-being. For such broad types of composite indicators, it may not be possible to estimate and apply explicit weights. Explicit weights of the components, whether estimated by statistical methods, experts’ judgements or public opinion, may be interpreted as value judgements about the relative importance of the different dimensions and raise criticism. Hence, at the level of dimensions, equal weighting might be the preferred option. However, while equal weighting avoids over- or underestimating the importance of components based on a subjective choice, the allocation of equal weights is in itself a choice that may omit information about the relative importance of the components.

Empirical methods

4.19 Empirical methods, sometimes referred to as data-driven or statistical methods, derive weights based on statistical analysis of the dataset. Methods include Principal Component Analysis (PCA) and frequency-based weights. Research on multidimensional poverty measures (Dutta, Nogales, and Yalonetzky 2021) demonstrates that the resulting indicators may violate two fundamental properties: monotonicity and subgroup consistency. Monotonicity states that if the experience of an individual worsens in any indicator while the experience of all other individuals remains unchanged, the indicator should not improve. Subgroup consistency requires that changes at subgroup level should be adequately reflected in the overall measure. For instance, if the outcome of interest improves in one region of a country but remains unchanged in all other regions, subgroup consistency implies that the overall country measure should not decrease. In short, it is possible to observe instances where, for example, poverty in one group may increase, but if the weights move by a larger factor towards groups observing constant poverty rates, this can result in the poverty metric appearing to improve.

4.20 A particular group of empirical methods builds on measures of people’s preferences through information about their actions (revealed preferences). These methods include the use of prices or time as a numeraire and the use of legislation and regulations as a proxy for social preferences. It is also possible to use a common numeraire derived for this purpose, such as a quality-adjusted life year (QALY) or well-being-adjusted life year (WELLBY) (as per Layard and De Neve 2023) and using the change in these metrics multiplied by the monetary value assigned to these as the weights.

Normative weights

4.21 Normative weights, also referred to as expert-driven or subjective weights, will generally be derived by expert-led processes. The weights for individual indicators and dimensions of composite indicators may be estimated based on experts’ judgment about the relative importance of the individual indicators or dimensions. The relevant expertise could include information from both field experts and researchers. However, for composite indicators that cover a broad set of dimensions, it may be difficult to obtain the very broad expertise required. A composite indicator of well-being, for example, may cover 8-10 dimensions, each of which requires special expertise. Weighting these together into a composite indicator would require knowledge or expertise on their relative importance. For composite indicators of particular dimensions of well-being, there may be more expertise or studies available that could be used to decide on the weights. Expert-driven weights may be discarded as being too open to criticism for being value-based or subjective.

4.22 The problem of value judgements applies particularly to the aggregation of composite indicators of dimensions of well-being into a single composite measure of overall well-being. The aggregation of different indicators within a particular dimension into a composite index for that dimension is less problematic, but still relies on the assumption that the selected indicators accurately measure the dimension of interest.

Public opinion

4.23 In the public opinion (or participatory) approach, weights are based on the outcomes of polls or surveys, where people are asked to express which areas matter most. Public opinion can be used to estimate weights for both individual indicators and for dimensions, but is probably more suitable for estimation of weights of dimensions since estimation of weights for individual indicators usually will require more detailed knowledge. The survey should be of sufficient quality to ensure that the weights are of sufficient precision and reliability.

4.24 Public opinion may change, which may necessitate updating the weights of the indicators and the dimensions. There are advantages in keeping the weights constant over time. However, if public opinion about the relative importance of the components of the composite indicator changes, the weights may need to be updated to reflect this. Changing the weights implies that the development in the composite indicator will not only be the result of changes in the individual components but also their relative weights.

Compensation/substitution

4.25 Irrespective of the method used to derive weights, an important assumption to consider is that of compensability or substitutability of dimensions and indicators. In a composite time series that combines two indicators (A and B), the implicit assumption is that an increase in A can be offset by a decrease in B. For example, if A is household income and B is air pollution, the assumption is that a decline in environmental quality can be offset by an increase in material well-being. This may not necessarily be the case, or may be unacceptable to some users. The recommendation is therefore that composite indices should be accompanied by dashboards (or other visualizations or dashboards) that give information about the development in the underlying components.

4.26 This section provides examples of composite indicators compiled by Portugal, United Kingdom and Canada. The aim is to help compilers find the right approach for their purposes and identify appropriate methods and techniques.

Portugal: Well-being index of Portugal

4.27 The Well-being Index (WBI), introduced by Statistics Portugal in 2013, serves as a tool to monitor several dimensions of well-being within the country. Its inception responds to a growing international consensus recognizing the limitations of traditional economic indicators like GDP in capturing the complexities of human well-being. The index is structured around two primary perspectives: Material living conditions and Quality of life, each encompassing a range of domains reflecting various aspects of societal welfare.

4.28 Underpinning the development of the WBI are international guidelines and frameworks advocated by organizations such as the United Nations, OECD, and Eurostat. Notably, the European Commission's ‘Beyond GDP’ initiative and the Stiglitz-Sen-Fitoussi report have provided pivotal insights into complementing economic metrics with comprehensive well-being indicators. Furthermore, programmes like the OECD's Better Life Initiative have contributed valuable insights into selecting relevant domains and indicators.

4.29 The WBI comprises ten domains, incorporating 78 baseline indicators from administrative data and other sources. These indicators range from the Material living conditions perspective (Economic well-being, Economic vulnerability and Employment) to the Quality of life perspective (Health, Work-life balance, Education, Social relations, Civic participation, Personal security and Environment). To ensure the comparability and coherence of these indicators, Statistics Portugal employs normalization methods, such as the goalpost method, to account for variations in data sources and units of measurement.

4.30 Addressing the challenge of missing data, Statistics Portugal adopts imputation techniques, including linear interpolation and exponential smoothing, to fill information gaps. The aggregation of indicators within each domain is carried out using arithmetic means. The Perspective and the overall Well-being Index are computed using geometric means to limit the compensation effect. In both cases, equal weighting is assigned to each indicator and each domain. This methodological approach ensures that no single domain or indicator disproportionately influences the overall index.

4.31 By providing a holistic well-being assessment, the index facilitates a nuanced understanding of societal progress beyond purely economic metrics. Its regular annual dissemination promotes transparency and accountability in governance, empowering stakeholders to evaluate policy effectiveness and advocate for positive societal change.

4.32 Detailed information about the WBI is available on Statistics Portugal - Web Portal. Statistics Portugal also publishes the individual indicators and, in this way, provides information both to those users who are only interested in the overall development and those who want to know about the underlying indicators.

UK: money-weighted composites alongside GDP

4.33 Gross Inclusive Income (GII) and its sibling Net Inclusive Income (NII) (ONS 2023) are composite indices drawn from the same methodology as the National Accounts use to produce Gross Domestic Product (GDP) and Net National Disposable Income (NNDI). Concerning the two methods issues raised above – which variables to include and the weighting method, GII and NII utilise well-established approaches, drawing from a theoretical framework based on the National Accounts principles of the asset and production boundary. The National Accounts define a set of assets (produced capital) which lie within a specified ‘asset boundary’. Alongside these, all the resultant flows which result from human interaction with these assets deliver the sum of flows of output, income and expenditure within the ‘production boundary’.

4.34 Using this method, the Inclusive Income indices expand the asset boundary to account for human and natural capital, and commensurately expand the production boundary to align with the output arising from this wider class of assets. As these flows of benefits are not all strictly ‘output’, the framework is re-conceptualised to align with the concept of consumption and proxied by equivalent income flows commensurate with the flow of consumption from the results of these assets' interaction with human activity or as part of natural actions.

4.35 Importantly these indices, under this regime, forms a strong theoretical match for the concept of economic welfare, that is the absolute total of the flows of benefits received from humanity’s interaction with the economy and the other structures and ecosystems, primarily unpaid work and the environment which deliver flows of benefits which can be conceptualised in such a framework. Factors such as subjective well-being and purely social factors may not be best considered within such a framework, and as such are excluded.

4.36 The Inclusive Income indices are weighted using market price money metrics. This is not to say that alternative money price metrics could not be considered. As proposed by Dasgupta (2021), accounting prices, loosely defined as shadow prices taking account of externalities, would potentially be a superior alternative numeraire in the Inclusive Income framework, and indeed, Dasgupta would argue a more appropriate one. ONS research on how inclusive income/wealth could be measured in accounting prices is available from this Discussion paper.

4.37 While dashboards deliver value by providing a broad set of indicators, a composite indicator summarises this range of information into a single measure. If appropriately weighted, it is possible to derive trade-offs between components of the composite indicator. So, if GDP increases but at a cost to the environment and the services it delivers to citizens, this can be seen by the growth rate being lower than that of GDP, and maybe even negative. Equally, it can reveal the importance of the market economy within a wider measure of economic welfare.

Canada: thematic outcome indices

4.38 As distinct from composite indices that seek to combine dimensions to provide summary measures of overall well-being, Canada’s Quality of Life framework includes composite indices as indicators for specific outcome areas, such as health and crime.

4.39 Health-adjusted life expectancy (HALE) is defined as the number of years an individual is expected to live given current morbidity and mortality conditions. This measure incorporates information on mortality (such as life expectancy) and health status (such as morbidity) into a single estimate that can be considered not only a measure of quantity of life, but also a measure of quality of life from a health standpoint. Population-level estimates of HALE assess expected future trends if current patterns of mortality and health status persist. This summary measure can be useful for analytical purposes, such as to understand health inequities and the relative impact of different risk factors.

4.40 From a technical standpoint, HALE combines the Health Utilities Index Mark 3 (HUI3) as a measure of health status with estimates of life expectancy. HUI3 is itself a composite index of health-related quality of life based on eight attributes of self-reported health status: vision, hearing, speech, ambulation, dexterity, emotion, cognition, and pain. The value of HUI3 can range from -0.36 (state worse than death; death represented by 0) to 1.00 (best possible health state).

4.41 The Crime Severity Index (CSI) tracks changes in the severity of police-reported crime by accounting for both the amount of crime reported by police in a given jurisdiction and the relative seriousness of the crimes. It is useful to measure change in time and differences between communities, and complements the volume-based reported crime rate (per 100,000 population).

4.42 From a technical standpoint, this composite index assigns a weight to all Criminal Code violations (including traffic and drug violations) and assigns weights based on their seriousness. These weights are determined based on actual sentences handed down by the courts. More serious crimes are assigned higher weights and thus have a greater impact on changes in the index. The weights are updated every five years.

4.43 In both cases, these composite indices provide meaningful insight about broad trends and distributional differences that span the relevant outcome dimension.

4.44 There are various examples of composite indexes for global comparison. Prominent examples are the global Multidimensional Poverty Index (global MPI), the Human Development Index (HDI) and the inequality-adjusted Human Development Index (IHDI). The global MPI measures acute multidimensional poverty in more than 100 developing countries. It captures acute, overlapping deprivations in health, education, and living standards that people in poverty face. The HDI measures human development as a composite index of data on education, life expectancy and income. The IHDI adjusts the HDI for inequalities in education, life expectancy and income.