This introductory chapter presents a new measure of inequality of opportunity designed to assess the extent to which disparities in outcomes are due to circumstances beyond individuals’ control. It highlights the value that a complementary focus on opportunities adds to existing OECD analysis of inequality in outcomes and social mobility. It explains why and how a robust and comparable measure of inequality of opportunity can be developed, providing conceptual and methodological groundings for the approach used. It defines the main welfare concept (household market income) and a set of relevant circumstances for the analysis, based on data availability, comparability and accuracy. Finally, the chapter provides a stylised illustration of how the measure works and discusses its use, implications for the broader questions of equal opportunity and economic fairness, as well as the interpretation of its results. The chapter notably underlines the fact that the measure constitutes a lower-bound estimate of actual levels of inequality of opportunity.
To Have and Have Not – How to Bridge the Gap in Opportunities
Abstract
1.1.1. What does this report add to OECD analysis of inequality and social mobility?
The landmark 2018 report A Broken Social Elevator? has provided a strong empirical basis for the OECD’s work on social mobility and helped outline the key challenges for policy. A Broken Social Elevator? reviewed the trends and drivers of social mobility in OECD countries and major emerging economies (OECD, 2018[1]). In doing so, it identified different patterns of social mobility across countries, as well as the main barriers and areas for action. Overall, it also showed that the scope for social mobility tends to be relatively limited, as differences in key socio-economic outcomes – such as income, occupation and education – exhibit significant persistence over the course of an individual’s life and across generations. The main findings from OECD (2018[1]) are summarised in Box 1.1. Based on these findings, OECD (2018[1]) underlined several important conclusions.
Social mobility and economic inequality do not go hand-in-hand. First important conclusion, the empirical results confirmed that inequality in outcomes is not a necessary condition or the price to pay for ensuring all people have an equal opportunity to succeed in life. While some of the countries studied in OECD (2018[1]) exhibited low levels of both income inequality and income mobility, no country combined high levels of income inequality with high levels of intergenerational mobility. In this respect, the idea that policymakers face a trade-off between promoting greater equality of outcomes and greater equality of opportunity is not supported by the evidence.
There is a strong case for promoting social mobility and ensuring a more level playing field for all. This case does not rest on economic grounds alone. Second important conclusion, low social mobility may also have an impact on the broader political economy. This impact should matter to policymakers. The negative socio-economic consequences of low social mobility are well established (OECD, 2018[2]; 2018[3]; 2015[4]). Using survey data, OECD (2018[1]) shows that perceptions and attitudes have tended to evolve in line with changes in actual levels of social mobility, as measured by conventional statistical indicators. Moreover, where people perceive social mobility to be declining, they also express lower levels of belief in meritocracy, as well as greater concern regarding opportunities to improve their position in life and the role played by inherited circumstances in determining outcomes. This issue is made more salient by the fact that the risk of downward mobility has tended to be higher for the middle class.
Promoting opportunities and social mobility remains a high priority for governments and citizens across the OECD. Disadvantage in childhood has an outsized effect on opportunities and mobility throughout life along a wide range of well-being outcomes (OECD, 2022[5]). Recent crises stemming from the COVID-19 pandemic and the impact of high inflation have disproportionately affected vulnerable populations. In doing so, they have put added pressure on governments to intervene to address present inequalities and preserve the potential for future mobility (Caisl et al., 2023[6]; Case and Deaton, 2022[7]). Furthermore, these measures have often been framed in the broader context of the green and digital transitions. This indicates a recognition of the new risks that weigh on social mobility and the fact that people will need to be equipped with appropriate skills, resources and capacity to adapt in order to maintain a level playing field in a changing socio-economic and technological landscape. Concerns about these risks are also reflected in public views, with two-thirds of respondents to the 2022 wave of the OECD Risks that Matter cross-national survey saying that “more” or “much more” should be done to promote equal opportunities (OECD, 2023[8]). To help address these challenges, the OECD has created the Observatory on Social Mobility and Equal Opportunity in 2022.1
This report extends existing OECD analysis along two dimensions that are of high relevance to policy. First, it uses an innovative approach to develop a robust and comparable measure of inequality of opportunity. This approach allows the analysis to account for the circumstances that individuals encounter and their influence in shaping economic outcomes. This chapter presents the measure, including the rationale for its development, as well as the conceptual and methodological groundings of the approach used. Chapter 2 applies the measure in an international perspective to a large subset of OECD countries with available and comparable data. In doing so, it goes beyond the distribution and persistence of outcomes to shed light on the opportunities that are available to individuals and the way in which they shape economic outcomes throughout the life cycle. Secondly, the report provides a more detailed focus on the important geographic dimensions of opportunities. Chapter 3 looks at regional disparities in access to key drivers of social mobility – including education, employment and essential services – building on the most recent OECD research.2 Chapter 4 concludes by assessing the additional insights that can be drawn from these analytical extensions and how they can be used to inform effective policies for promoting opportunities and ensuring a more level playing field.
The 2018 report A Broken Social Elevator? (OECD[1]) constitutes both a landmark contribution to the OECD’s research on social mobility and an important reference for the policy debate on how to promote it. The report provides a comprehensive empirical review of the trends and drivers of social mobility across OECD countries and major emerging economies. It looks at social mobility for a number of key socio-economic variables (education, occupation, income and earnings, health) and from a number of different perspectives: (i) both by comparing the outcomes of parents and children (intergenerational mobility) and by comparing an individual’s own outcomes over the course of their life (intragenerational mobility); and (ii) in terms of absolute mobility (which measures overall improvements in living standards) and relative mobility (which measures changes in an individual’s position within the distribution of outcomes). Analysis in A Broken Social Elevator? focuses primarily on intergenerational mobility and relative mobility on the grounds that these measures correspond more closely to the way in which people think about social mobility.
Among its key findings, it showed that:
Overall, the scope for social mobility tends to be relatively limited: Gaps in socio-economic outcomes tend to persist over time and shape opportunities across generations. For example, on average across the countries studied, the intergenerational elasticity of earnings is 38% – meaning that 38% of the relative difference in earnings between adults in one generation is transmitted to the next generation. This ranges from below 20% in the Nordic countries to over 70% in some highly unequal emerging economies. On this basis and given current levels of inequality, OECD calculations suggest that, in a typical country, it would take 4-5 generations on average for a child born into the bottom decile to reach the mean level of income. In parallel, public perceptions and attitudes have also evolved, with survey data showing a growing sense that social mobility has fallen and a decline in the belief in meritocracy. These perceptions square somewhat with levels of actual social mobility as measured along various dimensions.
“Sticky floors”, “sticky ceilings” and pressures on the middle class pose distinct challenges for social mobility at all levels of the distribution: Mobility tends to be lower at both the bottom and the top of the distribution, with significant negative socio-economic consequences for individuals and for society as a whole. Focus must also be put on protecting the middle class from risks of downward mobility, notably linked to income shocks and a rising cost of living.
“Sticky floors” at the bottom of the distribution: Children from a disadvantaged background struggle to move up the ladder, which implies wasted potential, a misallocation of resources and unrealised opportunities. For example, across the OECD, four-in-ten people with low-educated parents have lower secondary education themselves, and only one-in-ten continue on to tertiary education – compared to two-thirds of children with high-educated parents.
“Sticky ceilings” at the top of the distribution: Similarly, lack of mobility at the top gives rise to persistent rents, reduced competition and forms of “opportunity hoarding” that are inefficient from an economic point of view and entrench advantage and disadvantage. For example, children from affluent backgrounds tend to end up in similar occupations as their parents. Across the OECD, half of children whose parents are in managerial positions become managers themselves, compared to less than a quarter of children of manual workers.
Pressures on the middle class: Opportunities and risks tend to concentrate on the middle class, where income mobility is higher. Middle-income households face a substantial risk of downward mobility: on average, one-in-seven middle-class households fell into the bottom 20% over a four-year period, with the share rising to one-in-five for lower middle-income households. In a number of countries, a divide can be seen within the middle class with the risk of downward mobility increasing to a greater extent for the bottom 40% of the distribution than for the upper-middle class.
Countries exhibit different patterns of social mobility and encounter different challenges: The main barriers to social mobility vary across countries. Some general patterns are nonetheless observed when considering mobility across generations:
Social mobility, notably in terms of earnings, occupation and education, is high in most Nordic countries and rather low in many Continental European countries, especially in terms of earnings, as well as in emerging economies.
Most Southern European countries also show relatively low mobility in terms of education or occupation, but fare somewhat better in terms of earnings mobility.
Some English-speaking countries fare relatively well in terms of earnings mobility (Canada, New Zealand) or occupation (the United Kingdom, the United States), but performance varies greatly along the other dimensions.
In Japan and Korea, educational mobility is high but earnings mobility is around average. Both sticky floors and sticky ceilings, in terms of earnings persistence over generations, are more pronounced in Germany and in the United States than in other countries.
As a result, the type of outcome that policy solutions focus on and the level at which they are applied should reflect these different patterns of social mobility and the challenges they imply.
1.1.2. Why go beyond traditional measures of social mobility to assess opportunities and the way in which they are distributed across the population?
Measures of intergenerational mobility are imperfect proxies for opportunity, on conceptual and methodological grounds as well as due to limitations relating to measurement and data. Traditional measures of intergenerational mobility typically focus on the transmission of one specific outcome. For instance, OECD (2018[1]) looks at the distribution and persistence of a range of socio-economic outcomes including education, occupation and income, as measured by so-called “intergenerational elasticities” (i.e., a measure of the persistence in outcomes across generations). This approach is useful for understanding intergenerational social mobility along these key dimensions and for identifying patterns and barriers that are specific to countries. However, it also has several limitations:
The estimation of intergenerational elasticities restricts the analysis in terms of what can be measured and for whom. Data availability issues mean that the intergenerational link in earnings can often only be modelled for fathers and sons and for full-time employees. Furthermore, direct intergenerational comparison between the income of parents and children would require long-term panel surveys spanning several generations. As these data are rarely available, intergenerational elasticities generally rely on a comparison with predicted – as opposed to actual – parental income.3
At a methodological level, measures of intergenerational mobility only cover the transmission of one specific indicator in isolation. Transmission is measured through the strength of the correlation between the outcomes of parents and children and may be applied to a range of different variables: income or earnings, occupation, educational attainment… In doing so, these measures separate the effect of other confounding factors and circumstances (such as parents’ country of birth) on outcomes. To properly measure opportunity and economic fairness, the analysis must be able to take account of a wider range of variables and the joint effect they may have on outcomes.
Measures of intergenerational mobility do not fully reflect the way in which people think about opportunities and economic fairness. Most notably, these measures fail to capture the distinction between circumstances within and beyond an individual’s control. As argued in the current section (see below), this distinction has a significant influence on how people perceive outcomes and whether they evaluate them as being “fair” or not. In doing so, it also influences the extent to which people believe that policy responses are needed to ensure a level-playing field and justified in reducing inequalities in outcomes.
This report starts from the premise that people care about outcomes and their distribution, but also about the process through which they are achieved. While useful for assessing levels of social mobility and identifying patterns, measures that focus on the transmission of outcomes do not say anything about the circumstances that individuals encountered, the opportunities they were provided with to succeed and the decisions they made.4 These elements matter as they play a key part in people’s evaluation of the fairness of socio-economic outcomes, of the extent to which inequalities in the distribution of these outcomes are justified or not, as well as of the need for and acceptability of policies designed to reduce inequalities.5 Consequently, there are several advantages to developing a robust and comparable measure that can capture the role played by different types of circumstances in shaping individual outcomes. Doing so would provide additional insights into individual opportunities and their distribution across the population which can complement traditional measures of social mobility. It would also provide data and evidence that are closer to people’s perceptions and evaluation of outcomes and may therefore be more effective in informing public views.
At a conceptual level, there is a crucial distinction to be made between outcomes that result from decisions and circumstances within an individual’s control and those that do not. As a formal principle of justice and human right, equal opportunity plays a fundamental role both in theory (Rawls, 1971[9]) and in law (UNDP, 2023[10]). Its objective is often presented and understood as ensuring a “level playing field” in which everyone has an equal chance to freely pursue and achieve their own goals (Ikeda, 2022[11]; Mason, 2006[12]). Defining what constitutes equal opportunity in practice requires that several challenges be addressed. First of all, the effects that different types of circumstances have on individuals’ decisions and outcomes need to be specified. Secondly, it must also be possible to distinguish between those effects that are considered to be “fair” (i.e., circumstances do not unduly constrain opportunities or decisions and individuals can be held responsible for the outcomes of their actions) and those that are considered to be “unfair” (i.e., individuals were by necessity at an unusual advantage or disadvantage that affected their outcome and should be compensated for). In this respect, the notion of equal opportunity is closely tied to a reflection on personal agency, the role of circumstances and individual responsibility.
Economics has drawn on resources from moral philosophy to operationalise this distinction and notably from the literature on “luck egalitarianism”. From an analytical perspective, specifying the full range of relevant circumstances that may influence the opportunities available to people, the choices they make and the outcomes they achieve represents a highly complex and possibly intractable task. Similarly, defining the scope of individual responsibility and distinguishing between fair and unfair circumstances would require difficult and normative judgements. At a practical level however, these problems are more easily solved. Most notably, this can be seen in the context of everyday moral evaluations of actions and outcomes, where individuals make this distinction routinely and intuitively. The literature on “luck egalitarianism” has built on this fact and formalised it as an essential distinction between what can be attributed to “luck” and what can be attributed to “effort”.6 In doing so, luck egalitarianism seeks to balance the requirements of distributive justice with common moral intuitions on individual responsibility to provide a meaningful definition of equal opportunity. This definition contains a concrete principle for assessing whether inequality of outcomes is likely to be considered as fair or not (see Box 1.2 for further detail). Moral philosophy develops other applied definitions of equal opportunity that may differ from that of luck egalitarianism (Thompson, 2022[13]; Segall, 2016[14]). Luck egalitarianism is given specific emphasis here in light of the fact that many applied economic formalisations of equal opportunity, including the method used in this report, have drawn on this theory for conceptual foundations (see Section 0.2). This emphasis also reflects the fact that, as a theory of distributive justice, luck egalitarianism proposes to define economic fairness primarily around the notion of equal opportunity.7
Having a robust measure of opportunities can help inform policy, but also public debates on inequality and social mobility. One advantage of this type of measure consists in capturing the link between individual circumstances and outcomes more broadly and effectively than traditional measures of social mobility. On this basis, more targeted policies can be designed by identifying the relevant circumstances that promote or hinder opportunities and taking account of the relative importance of these circumstances. Furthermore, measuring opportunities in this way would provide policymakers with a statistical estimate that better aligns with people’s understanding of economic fairness and can help shed light on changes in public views on inequality and social mobility, as well as their implications for policy.8 It may also provide an effective means for improving public communication on these topics. In this respect, a measure that captures inequality of opportunity may potentially have a deeper impact on public views than other relevant indicators, such as income inequality or intergenerational mobility.9 This would notably be consistent with the evidence showing that beliefs about equality of opportunity are deeply held, play a structuring role in shaping policy preferences and do not adjust to information in a straightforward manner (OECD, 2025[15]; Alesina, Stantcheva and Teso, 2018[16]).
What is luck egalitarianism?
Luck egalitarianism is a particular tradition in the theory of distributive justice. From a conceptual point of view, its main specificities are tied to the fact that it seeks to design principles for a fair repartition of resources that are sensitive to considerations of individual responsibility and merit (Arneson, 2004[17]; 1999[18]; Dworkin, 2000[19]; 1981[20]; Cohen, 1989[21]). Luck egalitarianism was developed as a response to Rawls (1971[9]) regarding the practical implications of the principle of equality of opportunity and as an attempt to solve some of the problems raised by Rawls’ solution (the Difference Principle). Compared to Rawls (1971[9]), luck egalitarianism places emphasis on the role played by circumstances and the effect they have on the distribution of opportunities, as opposed to defining the conditions under which inequalities in outcomes can be justified.
What are its main implications for economic analysis and public policy?
Luck egalitarianism draws on insights from the ethical literature on “moral luck” to establish reasonable and widely acceptable criteria for defining what constitutes equality of opportunity in practice. This implies addressing a central normative question: how to distinguish which factors of success should be viewed as appropriate and which should not. From there, it notably proceeds to determine (i) under what conditions opportunities can be considered to be equal; and (ii) when the resulting inequality in outcomes can be justified on the grounds that it was produced by a “fair” process where everyone had an equal chance. Broadly put, the literature on “moral luck” examines the way in which common moral intuitions evaluate external circumstances and the role they play in assigning responsibility to individuals for the outcomes and consequences of their voluntary actions (Nussbaum, 1986[22]; Williams, 1981[23]; Nagel, 1979[24]).
More specifically, luck egalitarianism builds on the distinction between the broad categories of “luck” and “effort”, which the ethical literature identifies as a key element in empirical moral assessments. It uses this distinction to provide a concrete definition of equality of opportunity. On this basis, luck egalitarianism defines a just society as one that seeks, as far as possible, to:
1. Reduce the influence of structural and arbitrary factors on the set of opportunities available to people (i.e., reducing the scope of “luck” – embodied by factors such as parental background and inherited circumstances, advantages linked to social capital and interpersonal connections…); and
2. Ensure that their outcomes reflect factors that are under individuals’ control and can be attributed to their choices (i.e., factors for which they can reasonably be held responsible, such as effort, risk-taking, their level of investment in their own human capital including skills and education; or that they freely adhere to, such as their personal values and goals…).
Put differently, for luck egalitarianism, a just society is one that allows each individual to freely pursue their own goals (within the limits set by fundamental rights and the respect due to others) and provides everyone with an equal opportunity to achieve these goals to the fullest extent of their ability.
In turn, economic theory has sought to operationalise key conceptual insights from luck egalitarianism. This notably includes (i) insights on the role that circumstances should play in determining the set of opportunities available to people; and (ii) the extent to which public policy is justified in addressing the effects of circumstances, either ex ante (through measures designed to expand the opportunity set of individuals who are unfairly disadvantaged by circumstances) or ex post (through compensatory measures designed to improve the outcomes of individuals who are unfairly disadvantaged by circumstances). To do so, economic theory has developed measures that seek to capture the distinction between “luck” and “effort” made in common moral assessments (Lefranc and Trannoy, 2017[25]; Roemer and Trannoy, 2016[26]; Bradbury and Triest, 2016[27]; Roemer, 1989[28]). In some cases, this has involved criticism and further specification of the conceptual foundations provided by luck egalitarianism, as for example in Fleurbaey (2001[29]). The distinction between “luck” and “effort” is also widely used in survey questionnaires designed to elicit perceptions of and attitudes towards equal opportunity (OECD, 2023[8]).
Is the theory of luck egalitarianism consistent with the available empirical evidence?
The basic assumptions of luck egalitarianism are supported by empirical evidence, including survey data. In this respect, principles of fairness tend (i) to be broadly shared, with some variation across countries; and (ii) to combine merit-based considerations with distributive concerns about excessive and unjustified inequalities, giving rise to a form of egalitarianism that is sensitive to considerations of “individual responsibility” (Almås, Hufe and Weishaar, 2023[30]; Cappelen et al., 2022[31]; European Commission / DG EMPL, 2020[32]).
Evidence on public perceptions and attitudes towards opportunities tends to confirm the importance of the distinction between “luck” and “effort”. For instance, data collected through the Opportunities module of the 2022 OECD Risks that Matter survey show that:
On average across the 27 OECD countries covered, around 60% of respondents believe that factors linked to “effort” (such as “hard work”) are essential or very important in determining one’s chances to get ahead in life. However, among this group, only a small proportion – one-fifth on average – consider that it is the sole factor of success.
Factors relating to “luck” (such as socio-economic background and individual characteristics relating to identity) are also perceived as important determinants of success by a large share of respondents. Furthermore, significant divides can be observed between different groups in terms of their beliefs about equality of opportunity. For example, younger respondents and minorities are much more likely to view traits linked to identity as key determinants of success (OECD, 2023[8]).
Similarly, some experimental studies have tested and confirmed the relevance of the distinction between “luck” and “effort” for individual decisions relating to the allocation and redistribution of resources (Tinghög, Andersson and Västfjäll, 2017[33]; Möllerström, Reme and Sørensen, 2015[34]).
1.2.1. What are the main approaches and challenges for measurement?
The recent economic literature has developed robust approaches for modelling inequality of opportunity. While these approaches differ in terms of the proposed methodologies for measurement and evaluation, they have a common conceptual basis and a same goal which consists in identifying the share of the inequality of outcomes (or total inequality) that is due to circumstances beyond an individual’s control and may justifiably call for compensation. Building on the insights from luck egalitarianism (see Section 1.1.2 above), it is assumed that all determinants of an individual’s outcome can be classified as (i) structural factors for which an individual cannot reasonably be held responsible (i.e., “circumstances”); or (ii) controllable factors for which they can (i.e., personal agency and “efforts”). Box 1.3 provides a brief review of this literature.
Ex ante and ex post approaches to measuring inequality of opportunity
The existing economic literature has followed two main approaches when seeking to measure inequality of opportunity: an ex ante and an ex post approach (Fleurbaey and Peragine, 2013[35]). These approaches differ in terms of methodology and of the definition of equal opportunity they rely on.
The ex ante approach proceeds by partitioning the population into different groups, with each group consisting of individuals who share the same set of circumstances (these groups are also referred to as “types”). The group-specific distribution of outcomes is interpreted as the opportunity set for individuals with a similar background. Equality of opportunity is achieved when differences in the average outcomes of groups facing different circumstances are eliminated. In this context, focus is put on reducing inequality between groups, while inequality within groups is taken as given and a reflection of differences in factors relating to individual choice, including the level of effort. Equality of opportunity consists in ensuring that people from different backgrounds have similar prospects at their starting point (i.e., ex ante). As such, the ex ante approach views equality of opportunity primarily as a matter of “levelling the playing field” by reducing the influence of circumstances on economic outcomes.
The ex post approach starts instead from the level of effort exerted by individuals and the extent to which it is rewarded. To do so, it seeks to measure inequalities of outcomes within groups of individuals who display the same level of effort (these groups are also referred to as “tranches”). Differences between these groups are seen as resulting from a factor – the level of effort – that is under individuals’ control and do not therefore require compensation. In this context, equality of opportunity consists in a state where individuals with a similar level of effort achieve similar outcomes (i.e., ex post). As such, the ex post approach views equality of opportunity primarily as a matter of ensuring that, as far as possible, outcomes reflect individual choice and merit-based factors. As mentioned in the previous section, the notions of “effort” and “merit” do not fully coincide, with the former covering a broader range of factors than the latter (see note 6 at the end of this chapter).
Both of these approaches are valid from a conceptual point of view as they represent ways to operationalise the same distinction between “circumstances” and “effort”. However, there are significant differences between them. Ex ante approaches do not directly estimate the level of effort produced by individuals, but focus instead on different measurable aspects of their background. Conversely, ex post approaches require that all variables, including individual levels of effort, be measured and introduce additional methodological assumptions in order to do so. At an empirical level, the ex ante approach has proven easier to implement than the ex post approach. As a result, empirical applications have focused mainly on ex ante inequality of opportunity.
Methods and challenges
Researchers have proposed two methods to assess ex ante inequality of opportunity: parametric and non-parametric. While each method has respective strengths and limitations, they face a common constraint: not all of the relevant circumstances affecting individual outcomes can be observed or even specified. This results in biased estimates of inequality of opportunity. Overfitted models produce an upward bias, whereas underfitted models reinforce the downward bias caused by partial observability (Brunori, Peragine and Serlenga, 2019[36]; Ferreira and Gignoux, 2011[37]). Under some assumptions discussed in the literature, it can be shown that the sign of the bias is negative. This explains why ex ante estimates should generally be interpreted as lower-bound estimates of the “real” level of inequality of opportunity measured in a given society.
To address this challenge, recent studies have relied on the use of machine learning techniques, specifically conditional inference regression trees and forests (Brunori, Hufe and Mahler, 2023[38]). In contrast to conventional methodologies, these algorithms are capable of autonomously identifying intricate relationships within data sets without the need for pre-established assumptions regarding interaction patterns. This method has the advantage of minimising both types of bias and can thus provide more robust estimates of inequality of opportunity. Conditional inference trees offer a clear advantage in elucidating how specific circumstances shape individual opportunities and are well aligned with the theoretical frameworks used to conceptualise inequality of opportunity (Roemer, 1989[28]). Conditional inference forests enhance predictive accuracy by aggregating multiple trees, making them particularly effective for estimating counterfactual distributions in various social contexts (Athey and Imbens, 2019[39]). More information can be found in Annex 1.A.
Empirical applications of these approaches differ in terms of the welfare concept used and the set of circumstances included in the analysis. They also depend on the availability of relevant data. Inequality of opportunity has been computed for a range of relevant outcomes, such as education (Palmisano, Biagi and Peragine, 2022[40]), health (Jusot, Tubeuf and Trannoy, 2013[41]) and even subjective well-being (Kreiner and Olufsen, 2022[42]). However, most of the economic research has focused on economic outcomes, most notably income or earnings, because they offer a good proxy for standard of living and because the availability of international statistical standards facilitates cross-country comparison.10 Ideally, lifetime income would be the preferred metric, since individual income can fluctuate from year to year (OECD, 2023[43]) and follow different trajectories over the lifecycle.11 The ideal datasets for analysing inequality of opportunity are rarely available in practice. Most applications rely either on administrative records, notably for the United States and Nordic countries (Mitnik, Helsø and Bryant, 2020[44]; Eriksen and Munk, 2020[45]; Owens and Candipan, 2019[46]; Landersø and Heckman, 2016[47]; Chetty et al., 2014[48]), or on household surveys that include retrospective questions about parental status answered by the adult children (Brunori, Hufe and Mahler, 2018[49]; Fajardo-Gonzalez, 2016[50]; Jusot, Tubeuf and Trannoy, 2013[41]). While administrative data do not suffer from the same limitations as survey-based sources,12 they only include a limited set of circumstances. This can lead to a downward bias in the estimation of inequality of opportunity and limit the policy relevance of the results. Conversely, retrospective questions in household surveys offer insight on a large set of past circumstances but may be affected by (i) memory bias, whereby respondents’ ability to accurately remember and report past events may be flawed; and (ii) social desirability bias, whereby respondents may adapt the views expressed in line with what they consider to be expected or socially acceptable, notably on sensitive or personal topics.
For ex ante approaches, defining an appropriate set of circumstances is an important consideration as it will directly affect the estimation of inequality of opportunity. Under this type of approach, a counterfactual distribution of the outcome of interest (e.g., income) is derived as a means to quantify the “unfair” part of inequality for the outcome considered. The counterfactual distribution aims to reproduce only the share of inequality that is due to the measured circumstances and to leave out the share of inequality that can be attributed to factors relating to personal agency and individual choices (see Box 1.3 above). However, it is not possible to observe the entire set of relevant circumstances, as information on a large number of determinants of inequality of opportunity are rarely available in large-scale comparable datasets.13 Here, it is important to bear in mind that ex ante measures of inequality of opportunity only ever account for the role of a subset of the circumstances influencing outcomes. For this reason, they are best understood as providing a lower-bound estimate of actual levels of inequality of opportunity in a given society, as the influence of other relevant circumstances may not be accounted for.
1.2.2. What does the measure developed in this report consist in?
In line with most of the literature, this report takes an ex ante approach to measuring inequality of opportunity. The choice of this type of approach is driven by methodological considerations relating to empirical applicability and data availability (see Box 1.3). As highlighted previously, ex ante approaches are designed to provide insights on the role that circumstances play in shaping economic outcomes and how policy can help ensure a more level playing field for individuals facing different sets of initial circumstances. In order to build a measure of inequality of opportunity, a set of relevant circumstances is defined based on the available data (see below) and the population is split into non-overlapping groups, with each group being homogeneous in terms of the circumstances selected. A counterfactual distribution of outcomes that only reflects the differences between these groups – i.e., differences in outcomes that are due to the selected circumstances – is then computed (see Box 1.4 for an illustration of how the measure works in practice; see also Annex 1.A for a more in-depth technical presentation of the proposed measure and how it is designed).
Household market income is used as the main welfare concept for analysis. Throughout most of the analysis conducted in Chapter 2, inequality of opportunity is estimated for individuals between the ages of 25 and 59 and measured in terms of the child’s household equivalised market income as an adult.14 Market income is selected as the main variable of interest, instead of disposable income, in order to capture income-generating capacity and the inequalities of opportunity that arise from labour market dynamics.15 Similarly, the household is used as the unit of analysis, rather than the individual, to take account of income pooling, sharing and economies of scale within the household as doing so is likely to provide a better estimate of an individual’s standard of living. However, this household-based approach may potentially affect the measurement of intergenerational disadvantage at the individual level, as it also accounts for factors like assortative mating and fertility decisions at the household level and assumes that resources are equally shared among household members. This concern is particularly relevant when analysing inequality from a gender perspective.16
The set of circumstances has been selected based on data availability, comparability and accuracy of measurement, as well as the policy relevance of the insights that can be drawn. Based on the existing literature, a distinction is made between the factors that contribute to inequality of opportunity and those that do not (see Box 1.2). These factors will be referred to under the broad categories of “circumstances” and “effort” respectively throughout the rest of the chapter, bearing in mind the necessary caveats regarding the content, meaning and use of these categories in describing the factors analysed (see Box 1.3). A wide range of circumstances must be covered in order to account for the complex role that an individual’s background plays in shaping their opportunities and to properly understand the different channels through which advantage and disadvantage are transmitted across generations. The measure of inequality of opportunity developed in this report is based on a wider set of circumstances than is typically used in most studies.17 In addition to standard individual factors (such as gender and country of birth),18 parents’ migration status and socio-economic background (including parents’ educational level and occupation when the respondent was 14),19 the analysis also considers childhood environment factors, such as parental presence, housing tenure and the degree of urbanisation of the area of residence at age 14, to roughly differentiate between urban and rural areas.20 However, to accommodate varying levels of available information over time, trends in inequality of opportunity are based on a more limited set of circumstances (see Table 1.1).21 Annex 1.B provides more detailed information on the set of circumstances included in the analysis and the data sources used to elicit them.
The inclusion of factors relating to childhood environment reflects recent and innovative developments in the study of intergenerational inequality. For instance, there is evidence of increasing intergenerational persistence in homeownership for recent cohorts. In the United Kingdom, between 2000 and 2017, the gap in homeownership rates between those who grew up in rented accommodation compared to owner-occupied homes has doubled (Blanden, Eyles and Machin, 2023[51]). Similarly, there is a growing body of evidence suggesting that particular home environments (i.e., those with family stability and positive parental investments) are associated with higher chances of long-term success in life (Heckman and Mosso, 2014[52]). Finally, methodological advances and the use of rich administrative records have highlighted significant spatial variation in the transmission of outcomes across generations (Chetty et al., 2014[48]), partly reflecting geographical disparities in the access to quality services. This aspect of inequality of opportunity is covered in greater detail in Chapter 3 of this report.
Traditionally, the analysis of inequality has focused on measuring differences in key outcomes of interest, such as income disparities for example. The approach developed in this report takes these differences in outcomes as a starting point. From there, it seeks to assess the extent to which the observed differences in outcomes may be attributed to differences in opportunities stemming from a set of key circumstances that are beyond individuals’ control and may as a result skew the level playing field. The analysis follows an ex ante approach, as explained in Box 1.3.
To measure the extent to which opportunities are evenly distributed or not, the approach uses machine learning to divide the population into groups based on particular characteristics or circumstances. On this basis, it creates a counterfactual distribution of outcomes that only reflects the differences between these groups (i.e., differences in outcomes that are due to the selected circumstances). Doing so provides insight into the role and influence of specific external factors in shaping outcomes and the extent to which they allow for a level playing field.
Figure 1.1 below provides a visual illustration of the way in which the measure functions by applying it to a simplified example.
Panel A shows the distribution of income for the population of a fictional country, with individual incomes expressed in a given currency. While the currency is fictional and its value arbitrarily defined, the level of inequality observed (Gini coefficient of 0.34) is similar to that seen in many OECD countries.
Panel B takes the population and distribution of income shown in Panel A and considers a hypothetical case where there is only one relevant circumstance that can take two possible values (blue or yellow). Instead of examining the distribution of income among individuals (as done in Panel A), the measure calculates average incomes for the different groups defined by the set of circumstance selected. In this simple case, that means the population is now composed of two groups (“blue” and “yellow”) with respective average incomes of 87 and 50. The average incomes for these groups form a counterfactual distribution, which reflects the role played by the selected circumstances in shaping individual outcomes.
The distribution presented in Panel B, while highly stylised, is nonetheless comparable to what can be observed in an average OECD country: individuals who face "penalising" circumstances (in this case “yellow”) are mainly concentrated at the lower end of the income distribution, with some represented at higher levels but rarely at the very top.
The mean income for the overall population remains unchanged from Panel A at 76. However, the Gini coefficient for the counterfactual distribution in Panel B differs: it is now 0.10, representing absolute inequality of opportunity (IOp). Relative IOp is calculated by dividing the Gini of the counterfactual distribution (absolute IOp) by the observed Gini for individual income (total inequality). In this simple case, relative IOp represents 29% of total inequality (i.e., 0.10 / 0.34).
Measuring inequality of opportunity yields important insights that are not captured by the distribution of outcomes. A same distribution of individual outcomes may reflect significant differences in terms of opportunities and how they are distributed across the population. Panel C illustrates this by showing a situation where the income distribution in Panel A is consistent with full equality of opportunity and reflects a level playing field despite differences in individual outcomes.
Panel C once again takes the same population and distribution of income shown in Panel A and supposes instead that a slightly different set of individuals form the two groups (“blue” and “yellow”) based on the selected circumstance. In this case, the picture in terms of opportunities is quite different from the one in Panel B. In immediate and visible terms, the “yellow” group is larger and, while individuals who face that particular circumstance are still mainly concentrated at the lower end of the income distribution, they are also represented at the very top. Furthermore, for the same overall distribution at individual level, the counterfactual distribution now shows identical average incomes for all groups (i.e., absolute IOp is reduced to 0 and relative IOp is also 0%). Here, the selected circumstances seem to have no effect on potential income and no significant differences in terms of opportunities can be attributed to them.
Source: OECD Secretariat.
1.2.3. How should the measure be used and interpreted? Some key considerations
The measure developed in this report presents several advantages from an analytical point of view. Machine learning algorithms based on conditional inference regression trees and forests can help analysts minimise potential sources of bias that may be linked to discretionary decisions, such as model selection and the choice of circumstances to include in the analysis (Brunori, Hufe and Mahler, 2023[38]).22 The measure offers a rich perspective on inequality of opportunity in terms of the range of circumstances that can be covered. It can also be used to shed light on different aspects of inequality of opportunity. For example, most of the analysis in Chapter 2 focuses on relative inequality of opportunity – i.e., the share of the total inequality of outcomes that can be attributed to circumstances. However, where relevant, estimates of absolute inequality of opportunity – i.e., the level of inequality that would prevail if outcomes only reflected the influence of the selected set of circumstances, as measured by the counterfactual distribution – are also presented and discussed in order to contextualise the results for relative inequality of opportunity. This allows the analysis to reflect the fact that the level of inequality of outcomes differs across OECD countries. Finally, the measure offers flexibility in terms of its application and can be adapted to reflect the conditions of specific groups. For example, in Section 2.3, a change in the income concept is needed to properly capture the effects of gender on inequality of opportunity. The measure is then computed for individual earnings rather than household market income.
It is important to bear in mind that, despite the broader scope provided by this type of measure, the analysis does not capture the effect of all relevant circumstances. Consequently, the measure is likely to produce conservative estimates and should be viewed as a lower-bound of the actual levels of inequality of opportunity experienced by individuals, as mentioned previously (see Section 1.2.1).23 For the same reason, the remaining share of inequality that is not explained by the measure constitutes a residual variable and does not provide a direct proxy for or outcome of individual effort. While it is loosely referred to as “effort” in contrast to “circumstances” in line with part of the literature, this unexplained share of inequality is best understood as a broad category that captures the effect of different factors, including individual effort but also non-measured circumstances. In this respect, ex ante approaches provide a comparable measure of the lower bounds of inequality of opportunity. They do not provide a measure of equality of opportunity.
Similarly, while the analysis distinguishes between factors within and beyond individuals’ control, both types of factors tend to interact in practice. The distinction between “effort” and “circumstances” is meaningful at a conceptual level and plays a significant role in people’s evaluation of outcomes. However, this distinction is not always straightforward to make and its application to concrete cases may sometimes appear arbitrary or conventional. Factors that depend on individual choice are often influenced by external circumstances and background elements that are beyond the control of individuals. For instance, values, attitudes and aspirations that contribute to shape an individual’s level of effort may be transmitted through various channels – such as parental presence, the degree of parental engagement in school activities or the time a child spends on extracurricular activities (Pansacala et al., 2024[53]) – or influenced by the social context.24 Furthermore, different circumstances and prospects for success may lead individuals to adapt their preferences and expectations in ways that can affect their motivation, levels of aspiration and efforts. For example, the barriers faced by young people from disadvantaged backgrounds may give rise to a sense of having less control over their future and of relative deprivation in terms of opportunities and expected rewards. In turn, this can negatively impact on the extent to which they pursue and realise opportunities, for example through lower engagement in higher education (ONS, 2023[54]). On the other hand, those same barriers may in some cases spur people from disadvantaged backgrounds to exert more effort because they believe they will need to work harder than others to make up for unfavourable initial circumstances (Jin, 2024[55]).
The effects and role played by circumstances may be difficult to interpret, particularly over the longer-term. For example, one person’s effort can become another person’s circumstance (Fishkin, 2014[56]). This can notably be the case for parental income, as parents’ efforts contribute to provide a better start in life for their children. Furthermore, certain types of circumstances are taken as given and are not viewed as a legitimate source of advantage or disadvantage. This is the case for instance of age. While it is clearly a factor beyond people’s control, most studies do not consider it to be a circumstance whose effects should be compensated for.25 This may notably be due to the fact that ageing constitutes a natural process that affects everyone, even if at different rates, and whose effects balance out over time as individuals experience different stages of life. Reflecting these possible ambiguities, a conservative approach has been taken throughout this report when selecting the set of circumstances to include in the analysis (see Table 1.1 above). To minimise bias and avoid conflating circumstances with individual choices, the set used here focuses on variables that represent key aspects of an individual’s background and are undoubtedly exogenous, such as country of birth and parental characteristics. Disability status has been left out of the set of circumstances for this reason, though it may constitute a significant factor to include in future analysis given its importance and relevance for policy.
As a final point to consider, inequality of opportunity is an inherently dynamic concept that analysis can only capture “through the rear-view mirror”. Opportunities and their distribution are measured at a point in time, yet they reflect complex trajectories that are shaped by multiple factors over a long period of time. The analysis only observes the adult outcomes for children (and their parents) who grew up in a more or less distant past, not the events and processes that led to these outcomes. Caution should therefore be exerted when seeking to explain observed levels and trends in inequality of opportunity. First of all, it may be difficult to separate the effect of structural factors on inequality of opportunity from that of cyclical factors and short-term shocks or even one-off events such as changes in policy settings. Secondly, policies have a long-term and complex impact on the distribution of opportunities which may reduce inequalities for one generation but increase them for the next.26 Consequently, analysis of inequality of opportunity at country-level will require a more fine-grained and qualitative approach in order to properly account for national specificities, including institutional, historical and socio-cultural factors, and help determine which circumstances are most relevant in this particular context. As far as possible, the analysis in Chapter 2 seeks to take account of these challenges when examining the observed inequality of opportunity and discussing possible underlying mechanisms that can help explain the current state of opportunities within and across OECD countries.
References
[71] Aaberge, R., M. Mogstad and V. Peragine (2011), “Measuring Long-term Inequality of Opportunity”, Journal of Public Economics, Vol. 95/3-4, pp. 193-204, https://doi.org/10.1016/j.jpubeco.2010.11.023.
[16] Alesina, A., S. Stantcheva and E. Teso (2018), “Intergenerational Mobility and Preferences for Redistribution”, American Economic Review, Vol. 108/2, pp. 521-554, https://doi.org/10.1257/aer.20162015.
[17] Arneson, R. (2004), “Luck Egalitarianism Interpreted and Defended”, Philosophical Topics, Vol. 32/1, pp. 1-20, https://doi.org/10.5840/philtopics2004321/217.
[18] Arneson, R. (1999), “Against Rawlsian Equality of Opportunity”, Philosophical Studies, Vol. 93/1, pp. 77-112, https://doi.org/10.1023/A:1004270811433.
[39] Athey, S. and G. Imbens (2019), “Machine Learning Methods That Economists Should Know About”, Annual Review of Economics, Vol. 11/1, pp. 685-725, https://doi.org/10.1146/annurev-economics-080217-053433.
[78] Balestra, C. and . Ciani (2022), “Current challenges to social mobility and equality of opportunity”, OECD Papers on Well-being and Inequalities, No. 10, OECD Publishing, Paris, https://doi.org/10.1787/a749ffbb-en.
[58] Biau, G. and E. Scornet (2016), “A Random Forest Guided Tour”, TEST, Vol. 25/2, pp. 197-227, https://doi.org/10.1007/s11749-016-0481-7.
[51] Blanden, J., A. Eyles and S. Machin (2023), “Intergenerational Home Ownership”, The Journal of Economic Inequality, Vol. 21, pp. 251-275, https://doi.org/10.1007/s10888-023-09563-z.
[77] Bøyum, S. (2014), “Fairness in Education: A normative analysis of OECD policy documents”, Journal of Education Policy, Vol. 29/6, pp. 856-870, https://doi.org/10.1080/02680939.2014.899396.
[27] Bradbury, K. and R. Triest (2016), “Introduction: Inequality of Economic Opportunity”, RSF: The Russell Sage Foundation Journal of the Social Sciences, Vol. 2/2, pp. 1-43, https://doi.org/10.7758/rsf.2016.2.2.01.
[59] Breiman, L. (2001), “Random Forests”, Machine Learning, Vol. 45, pp. 5-32, https://doi.org/10.1023/A:1010933404324.
[38] Brunori, P., P. Hufe and D. Mahler (2023), “The Roots of Inequality: Estimating inequality of opportunity from regression trees and forests”, The Scandinavian Journal of Economics, Vol. 125/4, pp. 900-932, https://doi.org/10.1111/sjoe.12530.
[49] Brunori, P., P. Hufe and D. Mahler (2018), The Roots of Inequality: Estimating Inequality of Opportunity from Regression Trees, https://documents1.worldbank.org/curated/en/502141519144475516/pdf/WPS8349.pdf.
[36] Brunori, P., V. Peragine and L. Serlenga (2019), “Upward and Downward Bias when Measuring Inequality of Opportunity”, Social Choice and Welfare, Vol. 52, pp. 635-661, https://doi.org/10.1007/s00355-018-1165-x.
[65] Butt, D. (2012), “Option luck, Gambling, and Fairness”, Ethical Perspectives, Vol. 19/3, pp. 417-443, https://doi.org/10.2143/EP.19.3.2172298.
[6] Caisl, J. et al. (2023), “The uneven impact of high inflation”, OECD Papers on Well-being and Inequalities, No. 18, OECD Publishing, Paris, https://doi.org/10.1787/59e2b8ae-en.
[31] Cappelen, A. et al. (2022), “A Meritocratic Origin of Egalitarian Behaviour”, The Economic Journal, Vol. 132/646, pp. 2101-2117, https://doi.org/10.1093/ej/ueac008.
[73] Carranza, R. (2023), “Upper and Lower Bound Estimates of Inequality of Opportunity: A cross-national comparison for Europe”, The Review of Income and Wealth, Vol. 69/4, pp. 838-860, https://doi.org/10.1111/roiw.12622.
[7] Case, A. and A. Deaton (2022), “The Great Divide: Education, despair, and death”, Annual Review of Economics, Vol. 14, pp. 1-21, https://doi.org/10.1146/annurev-economics-051520-015607.
[48] Chetty, R. et al. (2014), “Where is the Land of Opportunity? The geography of intergenerational mobility in the United States”, The Quarterly Journal of Economics, Vol. 129/4, pp. 1553-1623, https://doi.org/10.1093/qje/qju022.
[21] Cohen, G. (1989), “On the Currency of Egalitarian Justice”, Ethics, Vol. 99/4, pp. 906-944, https://doi.org/10.1086/293126.
[63] Devarajan, S. and E. Ianchovichina (2017), A Broken Social Contract, Not High Inequality, Led to the Arab Spring, International Monetary Fund, pp. 1-4, https://www.imf.org/External/Pubs/FT/irb/2017/02/index.pdf.
[19] Dworkin, R. (2000), Sovereign Virtue, Harvard University Press, https://www.jstor.org/stable/j.ctv1c3pd0r.
[20] Dworkin, R. (1981), “What Is Equality? Part 2: Equality of resources”, Philosophy & Public Affairs, Vol. 10/4, pp. 283-345, https://www.jstor.org/stable/2265047.
[45] Eriksen, J. and M. Munk (2020), “The Geography of Intergenerational Mobility: Danish evidence”, Economic Letters, Vol. 189, https://doi.org/10.1016/j.econlet.2020.109024.
[32] European Commission / DG EMPL (2020), Employment and Social Developments in Europe 2020: Leaving no-one behind and striving for more – Fairness and solidarity in the European social market economy, European Commission Publishing Office, https://doi.org/10.2767/117922.
[50] Fajardo-Gonzalez, J. (2016), “Inequality of Opportunity in Adult Health in Colombia”, The Journal of Economic Inequality, Vol. 14, pp. 395-416, https://doi.org/10.1007/s10888-016-9338-2.
[37] Ferreira, F. and J. Gignoux (2011), “The Measurement of Inequality of Opportunity: Theory and an application to Latin America”, Review of Income and Wealth, Vol. 57/4, pp. 622-657, https://doi.org/10.1111/j.1475-4991.2011.00467.x.
[56] Fishkin, J. (2014), Bottlenecks: A new theory of equality of opportunity, Oxford University Press, https://global.oup.com/academic/product/bottlenecks-9780199812141?cc=fr&lang=en&.
[29] Fleurbaey, M. (2001), “Egalitarian Opportunities”, Law and Philosophy, Vol. 20/5, pp. 499-530, https://doi.org/10.1023/A:1017589614382.
[35] Fleurbaey, M. and V. Peragine (2013), “Ex Ante versus Ex Post Equality of Opportunity”, Economica, Vol. 80/317, pp. 118-130, https://doi.org/10.1111/j.1468-0335.2012.00941.x.
[57] Friedman, J., R. Tibshirani and T. Hastie (2009), The Elements of Statistical Learning: Data mining, inference, and prediction (2nd Edition), Springer, https://doi.org/10.1007/978-0-387-84858-7.
[74] Gärtner, M., J. Möllerström and D. Seim (2023), “Intergenerational Transmission of Luck versus Effort Beliefs”, Economics Letters, Vol. 232, https://doi.org/10.1016/j.econlet.2023.111345.
[52] Heckman, J. and S. Mosso (2014), “The Economics of Human Development and Social Mobility”, The Annual Review of Economics, Vol. 6, pp. 689-733, https://doi.org/10.1146/annurev-economics-080213-040753.
[66] Hirose, I. (2015), Egalitarianism, Routledge, https://www.routledge.com/Egalitarianism/Hirose/p/book/9780415783194?srsltid=AfmBOorcj9UNdZ1o_BYWb5SSSqexmTDEh6NzEAvNCbei95oeRjQCp-2k.
[68] Houle, C. (2017), “Social Mobility and Political Instability”, Journal of Conflict Resolution, Vol. 63/1, pp. 85-111, https://doi.org/10.1177/0022002717723434.
[11] Ikeda, M. (2022), “What can we do to ensure a level playing field for all students?”, PISA in Focus, No. 117, OECD Publishing, Paris, https://doi.org/10.1787/e297c355-en.
[55] Jin, X. (2024), “The Role of Effort in Understanding Academic Achievements: Empirical evidence from China”, European Journal of Psychology of Education, Vol. 39, pp. 389-409, https://doi.org/10.1007/s10212-023-00694-5.
[41] Jusot, F., S. Tubeuf and A. Trannoy (2013), “Circumstances and Efforts: How important is their correlation for the measurement of inequality of opportunity in health?”, Health Economics, Vol. 22/12, pp. 1470-1495, https://doi.org/10.1002/hec.2896.
[70] Kobus, M., M. Kapera and V. Peragine (2020), “Measuring Multidimensional Inequality of Opportunity”, UNU WIDER Working Paper Series No. 19/2020, https://doi.org/10.35188/UNU-WIDER/2020/776-7.
[42] Kreiner, C. and I. Olufsen (2022), “Is Inequality in Subjective Well-being Meritocratic? Danish evidence from linked survey and administrative data”, Journal of Economic Behavior & Organization, Vol. 203, pp. 336-367, https://doi.org/10.1016/j.jebo.2022.09.004.
[47] Landersø, R. and J. Heckman (2016), “The Scandinavian Fantasy: The Sources of Intergenerational Mobility in Denmark and the US”, The Scandinavian Journal of Economics, Vol. 119/1, pp. 178-230, https://doi.org/10.1111/sjoe.12219.
[25] Lefranc, A. and A. Trannoy (2017), “Equality of Opportunity, Moral Hazard and the Timing of Luck”, Social Choice and Welfare, Vol. 49, pp. 469-497, https://doi.org/10.1007/s00355-017-1054-8.
[12] Mason, A. (2006), Levelling the Playing Field: The idea of equal opportunity and Its place in egalitarian thought, Oxford University Press, https://doi.org/10.1093/acprof:oso/9780199264414.001.0001.
[44] Mitnik, P., A. Helsø and V. Bryant (2020), Inequality of Opportunity for Income in Denmark and the United States: A comparison based on administrative data, https://doi.org/10.3386/w27835.
[34] Möllerström, J., B. Reme and E. Sørensen (2015), “Luck, Choice and Responsibility – An experimental study of fairness views”, Journal of Public Economics, Vol. 131, pp. 33-40, https://doi.org/10.1016/j.jpubeco.2015.08.010.
[24] Nagel, T. (1979), Moral Luck, Cambridge University Press, https://www.cambridge.org/core/books/abs/mortal-questions/moral-luck/A3EEA631B0CA6F1A322A56E85FB77DB4.
[72] Niehues, J. and A. Peichl (2014), “Upper Bounds of Inequality of Opportunity: Theory and evidence for Germany and the US”, Social Choice and Welfare, Vol. 43/1, pp. 73-99, https://doi.org/10.1007/s00355-013-0770-y.
[22] Nussbaum, M. (1986), The Fragility of Goodness: Luck and ethics in Greek tragedy and philosophy, Cambridge University Press, https://doi.org/10.1017/CBO9780511817915.
[75] Nybom, M. and J. Stuhler (2024), “Interpreting Trends in Intergenerational Mobility”, Journal of Political Economy, Vol. 132/8, pp. 2531-2570, https://doi.org/10.1086/729582.
[15] OECD (2025), Getting the Public on Side: How to Make Reforms Acceptable by Design, OECD Publishing, Paris, https://doi.org/10.1787/262255fd-en.
[79] OECD (2024), “Enhancing opportunities by design: Exploring people’s views of what should be done to fight inequality”, OECD Policy Insights on Well-being, Inclusion and Equal Opportunity, No. 15, OECD Publishing, Paris, https://doi.org/10.1787/f3a168ef-en.
[8] OECD (2023), “Hard work, privilege or luck? Exploring people’s views of what matters most to get ahead in life”, OECD Policy Insights on Well-being, Inclusion and Equal Opportunity, No. 12, OECD Publishing, Paris, https://doi.org/10.1787/e1903ab2-en.
[43] OECD (2023), On Shaky Ground? Income Instability and Economic Insecurity in Europe, OECD Publishing, Paris, https://doi.org/10.1787/9bffeba6-en.
[80] OECD (2023), “Working hand in hand? Exploring people’s views of the role of different actors in fighting inequality”, OECD Policy Insights on Well-being, Inclusion and Equal Opportunity, No. 13, OECD Publishing, Paris, https://doi.org/10.1787/dbd54315-en.
[5] OECD (2022), “Starting unequal: How’s life for disadvantaged children?”, OECD Policy Insights on Well-being, Inclusion and Equal Opportunity, No. 4, OECD Publishing, Paris, https://doi.org/10.1787/91fcd679-en.
[81] OECD (2021), Does Inequality Matter? : How People Perceive Economic Disparities and Social Mobility, OECD Publishing, Paris, https://doi.org/10.1787/3023ed40-en.
[1] OECD (2018), A Broken Social Elevator? How to Promote Social Mobility, OECD Publishing, Paris, https://doi.org/10.1787/9789264301085-en.
[2] OECD (2018), Opportunities for All: A Framework for Policy Action on Inclusive Growth, OECD Publishing, Paris, https://doi.org/10.1787/9789264301665-en.
[3] OECD (2018), The Productivity-Inclusiveness Nexus, OECD Publishing, Paris, https://doi.org/10.1787/9789264292932-en.
[4] OECD (2015), In It Together: Why Less Inequality Benefits All, OECD Publishing, Paris, https://doi.org/10.1787/9789264235120-en.
[54] ONS (2023), Young people from disadvantaged backgrounds feel less in control of their futures, https://www.ons.gov.uk/peoplepopulationandcommunity/educationandchildcare/articles/youngpeoplefromdisadvantagedbackgroundsfeellessincontroloftheirfutures/2023-11-06.
[46] Owens, A. and J. Candipan (2019), “Social and Spatial Inequalities of Educational Opportunity: A portrait of schools serving high- and low-income neighbourhoods in US metropolitan areas”, Urban Studies, Vol. 56/15, pp. 3178-3197, https://doi.org/10.1177/0042098018815049.
[40] Palmisano, F., F. Biagi and V. Peragine (2022), “Inequality of Opportunity in Tertiary Education: Evidence from Europe”, Research in Higher Education, Vol. 63/3, pp. 514-565, https://doi.org/10.1007/s11162-021-09658-4.
[53] Pansacala, R. et al. (2024), “Influence of Parental Presence and Learners Behavior on Their Academic Performance”, World Journal of Education and Humanities Research, https://doi.org/10.22048/wjher.4.3.
[76] PwC (2023), The Value of Age Diversity: Companies with age diverse boards are less risky and more resilient, https://www.pwc.nl/nl/actueel-publicaties/assets/pdfs/age-diversity.pdf.
[9] Rawls, J. (1971), A Theory of Justice, Harvard University Press, https://doi.org/10.2307/j.ctvjf9z6v.
[28] Roemer, J. (1989), Equality of Opportunity, Harvard University Press, https://www.hup.harvard.edu/books/9780674004221.
[26] Roemer, J. and A. Trannoy (2016), “Equality of Opportunity: Theory and measurement”, Journal of Economic Literature, Vol. 54/4, pp. 1288-1332, https://doi.org/10.1257/jel.20151206.
[67] Rozelle, S., M. Alisky and M. Whyte (2024), “Getting Ahead in Today’s China: From optimism to pessimism”, The China Journal, Vol. 92/2, https://doi.org/10.1086/733178.
[30] Sardoč, M. (ed.) (2023), Equality of Opportunity: Fairness preferences and beliefs about inequality, Springer, https://doi.org/10.1007/978-3-319-52269-2_85-1.
[14] Segall, S. (2016), Why Inequality Matters: Luck egalitarianism, its meaning and value, Cambridge University Press, https://doi.org/10.1017/CBO9781316416969.
[60] Shapley, L. (1953), A Value for n-Person Games, Princeton University Press, https://www.jstor.org/stable/j.ctt1b9x1zv.24.
[61] Shorrocks, A. (2013), “Decomposition Procedures For Distributional Analysis: A unified framework based on the Shapley Value”, The Journal of Economic Inequality, Vol. 11, pp. 99-126, https://doi.org/10.1007/s10888-011-9214-z.
[13] Thompson, C. (2022), Equality, Equity, and Distributive Justice: Background Paper to Te Tai Waiora - Wellbeing in Aotearoa New Zealand 2022, https://www.treasury.govt.nz/publications/ap/ap22-03.
[33] Tinghög, G., D. Andersson and D. Västfjäll (2017), “Are Individuals Luck Egalitarians? - An experiment on the influence of brute and option luck on social preferences”, Frontiers in Psychology, Vol. 8/460, https://doi.org/10.3389/fpsyg.2017.00460.
[62] Turek, K., M. Kalmijn and T. Leopold (2020), “Comparative Panel File: Household panel surveys from seven countries. Manual for CPF v.1.0 CPF”, Center for Open Science OSF Preprints, https://doi.org/10.31219/osf.io/7zngy.
[10] UNDP (2023), Protecting Human Rights in Constitutions, United Nations Development Programme, https://www.undp.org/publications/protecting-human-rights-constitutions.
[64] Verme, P. et al. (eds.) (2014), Facts and Perceptions of Inequality, World Bank Studies, https://doi.org/10.1596/978-1-4648-0198-3.
[23] Williams, B. (1981), Moral Luck, Cambridge University Press, https://www.cambridge.org/core_title/gb/113623.
[69] Yang, F., A. Liu and W. Li (2025), “Public Perceptions of Intergenerational Mobility in China”, The China Quarterly, pp. 1-21, https://doi.org/10.1017/S030574102500030X.
This annex outlines the methodology proposed by Brunori, Hufe and Mahler (2023[38]) for estimating inequality of opportunity and the intergenerational transmission of advantage and disadvantage using regression trees and forests. Brunori, Hufe and Mahler (2023[38]) identifies two main advantages in using regression trees and forests to study of inequality of opportunity. First, this method is well aligned with Roemer's theoretical framework for monitoring inequality of opportunity. Secondly, regression trees and forests address the issue of model selection, thereby reducing researcher-induced bias and enabling a more objective and data-driven approach to measuring inequality of opportunity.
Regression trees and forests are supervised machine learning techniques designed to make accurate out-of-sample predictions of a dependent variable based on multiple observable predictors. In the context of inequality of opportunity, the outcome, y_i, is the income received by an individual i, while the input variables, x_i=(x_(i,1),…,x_(i,n) ), represent a set of n circumstances of the individual i – factors such as sex, parental socio-economic background or place of birth – over which they have no control.
Specifically, trees make predictions by partitioning the population S= {(x_i,y_i )}_(i=1)^S into M non-overlapping groups G=(g_1,…,g_M ), where each group g_m is homogeneous with respect to a particular set of circumstances. Each resulting group can thus be interpreted as a specific circumstance type. The expected outcome y_i=f(x_i ), for each individual i is estimated by the mean outcome of the group to which they belong to:
The counterfactual distribution, represented by the vector of predicted incomes y ̂=(f ̂(x_1 ),…,f ̂(x_N )), serves as a benchmark for what individuals' incomes would be if they were determined solely by the individual’s specific circumstances. By isolating the impact of circumstances and removing the influence of individual effort, talent or choices, this distribution reflects the variation in income attributable purely to differences in circumstances. Consequently, a highly skewed counterfactual distribution indicates that circumstances play a significant role in determining income, which corresponds to a high level of inequality of opportunity. Conversely, if the counterfactual distribution is constant and equal to the average income, this indicates that circumstances play no role in determining income and therefore that there is full equality of opportunity (under the set of circumstances included in the model).
Put differently, regression trees partition the sample into M types by recursive binary splitting. Conditional inference starts by a series of univariate hypothesis tests. The circumstance that is most related to the outcome is chosen as the potential splitting variable. If the dependence between the outcome and the splitting variable is sufficiently strong, then a split is made. If not, no split is made. Whenever a circumstance can be split in several ways, the sample is split into two sub-samples such that the dependence with the outcome variable is maximised. This procedure is repeated in each of the two sub-samples until no circumstance in any sub-sample is sufficiently related to the outcome variable (Annex Figure 1.A.1).
Note: Hypothetical example of a regression tree. The values in the white boxes show the predicted market income associated with each type.
Source: Adapted from Brunori, Hufe and Mahler (2018[49]), The Roots of Inequality: Estimating Inequality of Opportunity from Regression Trees, https://documents1.worldbank.org/curated/en/502141519144475516/pdf/WPS8349.pdf.
Regression trees offer a simple and reliable method for segmenting a population into distinct types, effectively addressing the challenge of model selection. However, it should be noted that they do have limitations, with Brunori, Hufe and Mahler (2018[49]) underlining the fact that:
The structure of regression trees, and thus the counterfactual distribution of income, can be highly sensitive to variation in the data sample. This sensitivity is particularly pronounced when multiple circumstances are competing to define the initial splits (Friedman, Tibshirani and Hastie, 2009[57]).
Regression trees assume a non-linear data-generating process, which emphasises interactions between variables while neglecting any potential linear effects of circumstances.
Regression trees may be sub-optimal in their use of available information by overlooking specific circumstances. This can become an issue if two or more circumstances are highly correlated. Once a split is made based on one of these circumstances, the others are unlikely to provide sufficient additional information to justify further splits.
Random forests address these issues by creating many trees on random sub-samples and by using only a random subset of circumstances before averaging over all of these when making predictions. This approach is more robust and comprehensive (Biau and Scornet, 2016[58]; Breiman, 2001[59]). In our empirical application, the forests are made of 500 trees.
In addition to measuring inequality of opportunity, the analysis conducted in Chapter 2 also examines the relative importance of the different observed circumstances in contributing to total inequality of opportunity. While the measures of inequality of opportunity and their decompositions cannot be interpreted causally as they omit various factors beyond a few key circumstances such as parental background, the observed circumstances do contribute differently to the overall estimate of inequality. The quantification of these contributions offers valuable descriptive insights, thereby facilitating the identification of the factors that exert the most significant influence on inequality of opportunity.
The Shapley-Shorrocks decomposition, as outlined by Brunori, Hufe and Mahler (2023[38]), provides a robust framework for quantifying the individual contribution of each observed circumstance to total inequality of opportunity. The concept was initially proposed by Shapley (1953[60]) and subsequently refined by Shorrocks (2013[61]). The Shapley-Shorrocks decomposition is the only decomposition that satisfies two crucial properties. First, the decomposition is exact under the addition, whereby the estimated sub-components can be interpreted as the proportion of total inequality of opportunity that can be attributed to a specific factor. Second, the decomposition is symmetric with respect to the order of the arguments. Put differently, the Shapley decomposition calculates the contribution of each variable by assessing the impact on the outcome function when that variable is excluded from all possible combinations of other variables. This is achieved by averaging the marginal contributions of the variable across all possible sequences of exclusion, thereby ensuring that contributions are fairly distributed and accounting for interactions between variables. The R program used to compute the Shapley decomposition was kindly provided by Paolo Brunori and Pedro Salas-Rojo (International Inequalities Institute, LSE).
The analysis presented in this chapter is based on a comprehensive set of data sources that, in addition to providing detailed information on income and demographic characteristics at the national level, also include retrospective questions on parental background and living arrangements during the respondents' childhood and formative teenage years.27 These questions (see Annex Table 1.B.1) enable the definition of a set of individual-level factors that are relevant for understanding their opportunities, decisions and outcomes – i.e., the set of so-called “circumstances”. In particular, Chapter 2 draws upon the following four sources of data, which were harmonised ex post to generate comparable estimates across countries:
The European Statistics on Incomes and Living Conditions Survey (EU-SILC)
The Panel Survey of Income Dynamics (PSID)
The Household, Income and Labour Dynamics in Australia Survey (HILDA)
The British Household Panel Survey and UK Household Longitudinal Study (BHPS-UKHLS)
The National Socioeconomic Characterization Survey (CASEN)
The European Union Statistics on Incomes and Living Conditions (EU-SILC) Survey is a comprehensive data collection and harmonisation initiative that encompasses information on income, social exclusion and living conditions collected across individuals and households in all EU Member States, as well as in Iceland, Norway, Switzerland, and Türkiye. This chapter draws upon data from the 2005, 2011 and 2019 waves of the EU-SILC survey which included an ad-hoc module on the intergenerational transmission of disadvantages. The modules focused on respondents aged 25 to 59 and included a series of retrospective questions that gathered information on parental background and family circumstances when respondents were 14 years old.
The Panel Survey of Income Dynamics (PSID) conducted by the University of Michigan is a longitudinal household survey that has provided insights into the economic, social and demographic conditions of United States families for over five decades. The PSID was originally constituted with an initial sample of over 18 000 individuals in 5 000 families and has since collected extensive data on these individuals and their descendants. This chapter makes use of two principal PSID data files, covering the period from 1968 to 2021: the Family Files and the Cross-Year Individual Files. The Family Files contain the majority of PSID variables, including family-level data on income, working hours, wages, wealth and consumption, as well as comprehensive information about the reference person and their spouse or partner. The Cross-Year Individual Files provide a record for each individual present in an interviewed family in a given survey year, including both respondents and non-respondents. PSID only collects retrospective questions for household heads and their partners. For the purpose of the analysis, individuals classified as "other family unit members" have therefore been excluded from the sample.
The ex-post harmonisation was based on the Comparative Panel File (CPF) (Turek, Kalmijn and Leopold, 2020[62]). The CPF is an open-source project that provides a blueprint for harmonising seven of the world's largest panel surveys, including three of the five datasets used to produce the estimates in Chapter 2.
The Household, Income and Labour Dynamics in Australia (HILDA) Survey conducted by the Melbourne Institute is an ongoing household-based panel study that annually collects detailed data from over 17 000 Australians. Initiated in 2001, the HILDA survey offers a comprehensive longitudinal dataset covering a diverse range of topics, including education, family background, employment, income, health and life satisfaction. Over time, the survey has introduced questions on various special topics. For the purposes of the analysis in this chapter, the HILDA Waves 5, 8, 11, 12, 15, 19 and 21 were used. These modules provide insights into the socio-economic backgrounds of respondents through retrospective questions on respondents' parental history and status, capturing key variables such as parental educational level, occupation and living conditions during childhood.
The Understanding Society UK Household Longitudinal Study (UKHLS) is a longitudinal study conducted by the Institute for Social and Economic Research at the University of Essex. It builds upon the British Household Panel Survey (BHPS), which started in 1991 and collected data on UK households until 2008. The UKHLS, which was initiated in 2009, shares numerous similarities with the BHPS in terms of design, content and the type of data collected. Chapter 2 primarily draws upon more recent UKHLS data as some key income variables used in the chapter are only available in this survey. Yet, it also incorporates retrospective information from the BHPS. A harmonised BHPS-UKHLS dataset, developed by the Institute for Social and Economic Research, allows for seamless integration, as the vast majority of BHPS participants continued with the Understanding Society survey.
The National Socioeconomic Characterization Survey (Encuesta de Caracterización Socioeconómica Nacional, CASEN) is a multi-wave cross-sectional survey conducted by the Ministry of Social Development and Family (Ministerio de Desarrollo Social y Familia) to collect information that allows for a regular assessment of the socio-economic conditions of the population and evaluation of the effectiveness of social policies. The survey targets households residing in occupied private dwellings across the national territory, excluding certain municipalities or segments of municipalities classified as special areas by the National Statistics Institute (INE). Since its inception in 1990, the CASEN survey has been conducted on a biannual or triennial basis. Chapter 2 draws on data from the 2006, 2009, and 2011 waves, which offer detailed information on demographics, education, health, housing, employment and income. Moreover, retrospective questions allow for the capture of key variables such as parental educational level, occupation and living conditions during childhood. To construct the outcome variable, micro-simulation techniques were used to generate respondent-level estimates of individual wages, household market income and household disposable income. These models simulate the effects of taxes and social contributions on various income sources – both gross and net – for employees, self-employed, pensioners and capital income recipients. In most of the analysis in Chapter 2, the 2009 wave of CASEN is used, as it includes a larger set of circumstances than the 2011 wave.
Notes: 1 See: https://www.ons.gov.uk/methodology/classificationsandstandards/standardoccupationalclassificationsoc/soc2010; 2 Although the variable is included in the dataset, it could not be used in the analysis because the information is only available for younger generations. This limitation stems from the variable’s retrospective nature and short time series.
← 1. The activities of the OECD Observatory on Social Mobility and Equal Opportunity are organised around three main objectives: (i) collecting new data and improving the measurement of social mobility and equal opportunity to better understand their drivers; (ii) providing insight on the challenges to social mobility and equal opportunity and the policies that can effectively address them; and (iii) analysing the role played by civil society and the private sector in fostering equal opportunity and how to effectively align it with policy action. Under the Observatory, the OECD has also deepened the analysis of the political economy dimensions of social mobility and equal opportunity highlighted in OECD (2018[1]). In doing so, it has collected and analysed data on public perceptions, attitudes and preferences relating to equal opportunity, notably through the Opportunities module in the 2022 wave of the OECD Risks that Matter cross-national survey (OECD, 2023[8]; 2023[80]; 2024[79]). Further OECD analysis has confirmed the important role that public perceptions and attitudes towards inequality play in shaping policy preferences. It has also underlined the valuable insights that can be drawn from the comparison between “objective” measures of actual inequality and “subjective” measures of perceived inequality (OECD, 2025[15]; 2021[81]). More information on the OECD Observatory on Social Mobility and Equal Opportunity can be found here: https://www.oecd.org/en/about/programmes/observatory-on-social-mobility-and-equal-opportunity.html
Important innovations in data collection and analysis have also taken place outside the OECD. For example, the Global Estimates of Opportunity and Mobility (GEOM) project provides comparable cross-country evidence and data visualisation on inequality of opportunity and its drivers. As such, it constitutes a useful resource that can contribute to inform policy debates and public perceptions. More information on the GEOM project can be found here: https://geom.ecineq.org/
← 2. Both of these analytical extensions correspond to priority areas identified in the programme of work of the OECD Observatory on Social Mobility and Equal Opportunity (Balestra and Ciani, 2022[78]).
← 3. The prediction of parental income rests in turn on a series of hypotheses and assumptions which may potentially affect accuracy. One of the advantages of the measure presented in this chapter is that it does not require any modelling of parental income and avoids thereby one potential source of bias in the results.
← 4. This holds whether the analysis focuses on comparing outcomes from an intergenerational perspective (i.e., comparison between an individual’s outcome and that of their parents) or from an intragenerational perspective (i.e., comparison of an individual’s outcome over their life course).
← 5. Recent OECD research has sought to provide a more realistic understanding of the role that perceptions of and attitudes towards inequality play in the formation of policy preferences (OECD, 2025[15]; 2021[81]). In doing so, it has highlighted (i) the potential gaps that may emerge between actual inequality as measured by outcome-based indicators and perceived inequality as measured by survey data; and (ii) the crucial importance of the latter in driving public support for policy and political behaviour. The literature on political discontent has notably underlined the case of the Arab Spring of 2010-2012 as a recent topical example illustrating these points. In many of the countries affected, the uprisings took place in a context where income inequality was moderate and declining, but dissatisfaction with a perceived lack of economic opportunities and lack of fairness of public institutions was growing (Devarajan and Ianchovichina, 2017[63]; Verme, 2014[64]).
← 6. “Luck” can be broadly understood as covering circumstances and factors that are not chosen by individuals, but affect their prospects for success in a way that leads to differences in outcomes which are not compatible with equal opportunity. This includes, for example, inherited traits or resources, such as parental wealth and education, and essential characteristics that are not chosen by individuals but may affect their opportunities and outcomes, such as their ethnic and racial origin or place of birth. “Effort” can be broadly understood as covering circumstances and factors that are attributable to individuals’ freely-made and responsible choices, do not imply differential prospects for success and allow for equal opportunity though they may lead to unequal outcomes. This includes elements over which individuals can be deemed to have direct control, such as, for example, hard work and their level of effort per se. It also includes “accidental” outcomes which may nonetheless be considered fair because they result from an individual’s freely-made and responsible choices, such as, for example, the outcomes of deliberate gambles and their level of risk-taking. Further moral distinctions apply to these different types of “effort”. In the former case, “effort” (strictly understood) is generally assessed in terms of merit. In the latter case, “effort” (broadly understood) is generally assessed in terms of fortune. For a more in-depth discussion of the distinction between “luck” and “effort”, see for example Hirose (2015[66]) and Butt (2012[65]).
← 7. In a review of its work on education policy, Bøyum (2014[77]) finds that the OECD’s approach to “fairness in education” has been consistently underpinned by a similar and often implicit principle centred around equal opportunity. Bøyum (2014[77]) argues that this approach is notably visible in the OECD’s analysis of the relation between socio-economic background and educational outcomes, as well as the role of education policy in addressing gaps in these outcomes. Bøyum (2014[77]) also calls for greater and more explicit links to be made between the OECD’s approach to “fairness in education” and its analysis of other forms of inequality. The methodology and conceptual background presented in this report can contribute to do so.
← 8. The value of this type of analysis and comparison between objective and subjective measures has notably been demonstrated in the case of income inequality (OECD, 2021[81]) and of intergenerational income mobility (Alesina, Stantcheva and Teso, 2018[16]). It can notably contribute to shed light on (i) the extent to which public perceptions are aligned or not with actual measures of social mobility; and (ii) the impact that changing patterns of social mobility are having on public attitudes, such as belief in meritocracy, and on the broader political economy. Reflecting this, the robustness of the proposed measure is tested against some commonly used perceptual indicators in Chapter 2.
← 9. The current debate on the policy implications of inequality of opportunity in China, spurred by recent survey data showing significant changes in public perceptions and attitudes, provides a topical illustration of this (Yang, Liu and Li, 2025[69]; Rozelle, Alisky and Whyte, 2024[67]). For evidence on the broader empirical link between social mobility and socio-political stability, see Houle (2017[68]).
← 10. Despite the large set of outcomes considered by the literature and recent attempts to account for the multi-dimensional nature of inequality of opportunity, most research treats each dimension independently, neglecting interdependencies (Kobus, Kapera and Peragine, 2020[70]).
← 11. The ideal dataset would contain several years of income, both for individuals and their parents, preferably observed at mid-career. However, these data are not easily obtainable and most studies have therefore used single-year measures as proxies for lifetime income instead. While estimates of inequality of opportunity based on current income may be potentially biased, Aaberge, Mogstad and Peragine (2011[71]) show that analyses drawing on snapshots of income can approximate results based on lifetime income by using panel data from Norway on individuals’ incomes over their working life span.
← 12. Such as under-reporting, small sample sizes and declining response rates.
← 13. These unobserved “circumstances” may include, for instance, IQ and genetic endowments, parenting styles, the extent and quality personal networks and social connections.
← 14. That is, income from market sources (i.e., the wage and salary income of the household members, excluding employers’ contributions to social security, but including publicly-funded sick pay, self-employment income, as well as capital and property income streams) net of public cash transfers and household taxes and adjusted by the square root of the household size. Negative or nil market incomes are set to 1.
← 15. Section 4.2.2 in Chapter 4 compares inequality of opportunity for market income and disposable income as a way to assess the effectiveness of tax and benefit systems in reducing inequality of opportunity and ensuring a more level playing field.
← 16. Section 2.3 in Chapter 2 complements this analysis by offering estimates of inequality of opportunity based on individual earnings when looking at gender dimensions.
← 17. Issues of data availability and comparability remain a significant constraint for the cross-country analysis of inequality of opportunity and impose trade-offs. For example, detailed information on country of birth may be available for some countries and in some years. However, one of the main sources of data used in this report – the Eurostat European Statistics on Incomes and Living Conditions Survey (EU-SILC) – uses broader and highly specific categories (i.e., by asking whether an individual was born in their current country of residence, in another EU country or in a country outside the EU), limiting comparability to a simple born inside the country/born outside the country dummy (See Annex Table 1.B.1). Furthermore, accuracy of measurement may require that certain variables be collapsed at the expense of finer-grained detail. For instance, the United States has used different job classifications over time. Collapsing job categories and using an aggregate classification of occupation is likely to improve confidence in the results obtained by avoiding problems and potential biases that may occur in translating these classifications.
← 18. Although country of birth is treated as a circumstance in this analysis, the nature of the choices involved implies a more nuanced and less dichotomous understanding of this factor. In many cases, it can safely be assumed that, for a significant proportion of individuals born outside the country of residence, their status reflects a conscious decision to relocate (for instance, for work-related or personal reasons) and therefore a degree of agency. However, this decision itself may frequently be constrained by factors such as limited opportunities or social inequality in the country of birth. As a result, the decision to migrate or change country can be more accurately described as a response to external pressures, rather than as a purely free choice. In the country of destination, foreign-born individuals may still encounter systemic barriers such as discrimination or restricted access to services, which impact their opportunities. The case of forced migration, driven by factors such as conflict, economic crises or environmental disasters, underlines to an even greater extent the fact that the country of residence is not always a matter of voluntary choice or a factor over which individuals have significant control. The age at which an individual moves to a new country introduces additional nuances. Migrating as a child, often as a result of a household decision, presents different challenges and opportunities compared to moving as an adult, when an individual is more likely to have a say and may have already established certain skills or networks. These factors further complicate the analysis of the role played by country of birth in determining life outcomes. Overall, while it recognises the complexities attached to this factor and differences between cases, for the purpose of the analysis this report considers country of birth as a circumstance when measuring inequality of opportunity, based on the fact that in many cases it reflects conditions beyond an individual’s control.
← 19. In most countries, retrospective information refers to the respondent’s situation at age 14. However, in some cases, the reference age is 15 or 16. For further details, see Annex Table 1.B.1.
← 20. In line with most research, age is not included in the set of circumstances. Instead, it is examined separately in Section 2.3, with a focus on the level of inequality of opportunity across birth cohorts and throughout the life-course.
← 21. Not all the circumstances listed for 2019 are available for every country included in the analysis. For details on country-specific availability see Annex Table 1.B.1 and the note to Figure 2.1 in Chapter 2.
← 22. Tree construction involves recursive binary splitting based on the most influential circumstance variables, chosen via permutation tests. While effective, single trees may suffer from sensitivity to data variations, non-linear assumptions, and underutilisation of circumstances not selected for splits. The use of random forests addresses these issues by averaging predictions across multiple trees and taking random subsets of the total population (and by limiting the splits to a random subset of circumstances), which enhances the robustness and predictive power of the model. The fact that machine learning techniques can help address certain sources of bias does not imply however that they are always preferable to other methods or that researchers should seek to avoid making choices regarding the modelling and design of the analysis. For example, applied knowledge of national contexts and their specificities may be needed in order to identify and select the relevant circumstances that may affect opportunities and help explain differences in outcomes.
← 23. See Niehues and Peichl (2014[72]) and Carranza (2023[73]) for an attempt to estimate the upper bounds of inequality of opportunity, using fixed effects models applied to panel data.
← 24. Some empirical evidence from Sweden on the intergenerational transmission of beliefs suggests that parents tend to emphasise the value of effort when teaching their children about the relative importance of luck and effort in determining life outcomes. Interestingly, this tendency is largely independent of parents’ own beliefs (in a “bootstrapping effect”) and widely shared, with only limited differences in terms of parents’ gender and level of income and education (Gärtner, Möllerström and Seim, 2023[74]).
← 25. Roemer and Trannoy (2016[26]) state for example that males should not be considered “disadvantaged with respect to females if, due to innate biological factors, their life expectancy is shorter [on average]”. Evidence suggests that a similar specific view of age may also be prevalent in public attitudes and corporate practices. In this respect, a recent study covering 5 European countries found that only 8% of companies surveyed included age among the grounds covered by their diversity strategies (PwC, 2023[76]).
← 26. Nybom and Stuhler (2024[75]) provides an illustration of this through the study of the long-term and differentiated impact on mobility trends of an education reform in Sweden.
← 27. In some cases, retrospective questions may not be available, but the panel component of some of the surveys included in the analysis allows for the collection of information on specific childhood environment factors.