Monitoring and Assessing the Impact of National Action Plans Against Racism

2. How to monitor the impact of racism on the lives of visible minorities?

Abstract

This chapter proposes indicators designed to capture bias-driven discrimination against visible minorities across the five key life areas outlined in Chapter 1, as well as in the areas of poverty risk and low life satisfaction. It examines two main categories of indicators: “core indicators”, offering broad availability across EU countries but with limited precision, and “advanced indicators”, divided into “enhanced indicators”, offering greater precision but requiring more robust data collection, and “exploratory indicators”, which achieve the highest precision but depend on original data collection. The chapter also introduces supplementary proxies for racism, racial/ethnic bias, and discrimination, based on attitudinal and perception-based measures. It concludes by suggesting options for countries to expand data collection on visible minorities to include long-established populations, enabling a more comprehensive monitoring of anti-racism efforts.

Main findings

• Chapter 2 examines two categories of indicators, detailed in Annex Tables 1 to 6, designed to monitor bias-driven discrimination against visible minorities in all five key areas discussed in Chapter 1 – education, school-to-work transition, employment, housing, and health – as well as in the areas of poverty risk and low life satisfaction, which may be impacted by the cumulative effects of bias-driven disadvantages.

• Developing precise indicators that isolate the full impact of bias-driven discrimination on visible minorities, and nothing else, remains a challenge. Overall, there is a trade‑off between how precisely bias-driven discrimination against visible minorities is measured and the availability (and cost) of the data required for such measurement. This trade‑off gives rise to two categories of indicators.

• The first category consists of “core indicators”, characterised by high availability but only moderate precision. They are derived from cross-national surveys, including the EU Labour Force Survey (EU-LFS), the Programme for International Student Assessment (PISA), the EU Statistics on Income and Living Conditions (EU-SILC), and the European Health Interview Survey (EHIS), making them computable across all EU countries (except Cyprus for PISA). These surveys allow for measuring disparities between the majority and visible minority groups across a range of life areas and offer several advantages.

  • They collect information on individuals’ and their parents’ country of birth, either in every round or within specific modules that rotate periodically. This facilitates the comparison between native‑born individuals with two native‑born parents (the majority population) and native‑born individuals with at least one foreign-born parent, distinguishing, where sample size permits, between European and non-European foreign parentage.

  • These surveys go beyond capturing final outcomes like educational attainment or achievement; they also provide standardised measures of intermediate factors that shape these outcomes, such as a sense of exclusion at school, perceived teacher hostility, limited access to quality career counselling, or exposure to bullying. By doing so, they shed light on some of the mechanisms discussed in Chapter 1 through which bias may operate.

  • These surveys measure a range of demographic factors that shape outcomes independently of discrimination. Adjusting for these factors is crucial to highlight the disparities between majority and visible minority individuals that persist after the adjustment – referred to as the “unexplained gap” and interpreted as an indication of bias. However, this gap only suggests bias; it does not directly measure it – hence, the moderate precision of core indicators in capturing bias-driven discrimination.

• The second category of indicators is composed of “advanced indicators”, which can be further divided into two subgroups: “enhanced indicators”, offering greater precision than core indicators but requiring more robust data collection, and “exploratory indicators”, which deliver the highest precision yet remain limited in availability due to their reliance on original data collection.

  • Advanced indicators of the enhanced type require more robust data collection than core indicators, as they involve calculating unexplained gaps between majority and visible minority populations across the entire population. For this to happen, it is necessary to collect information on both individuals’ and their parents’ country of birth, either through the census or national population registers, and to link these data with administrative databases in the key life areas covered in this report – a process possible in at most half of EU countries. These indicators offer greater precision in measuring bias-driven discrimination compared to core indicators for several reasons.

    • Administrative databases cover a broader range of outcomes compared to the limited set captured through surveys, with some being better suited to identify bias-driven racial/ethnic discrimination.

    • Advanced enhanced indicators are computable across the entire population rather than being restricted to cross-national survey samples of limited size. This comprehensive coverage enables a more granular analysis of visible minority subgroups. Furthermore, the removal of sample size constraints enables a full adjustment for key demographic factors, ensuring a more accurate analysis of disparities between majority and visible minority populations.

    • Administrative data are collected at a high frequency, with updates ranging from annual (e.g. education enrolment data) to monthly or quarterly (e.g. employment or health records), and even continuously in certain cases, such as births and deaths. This high-frequency data collection enables advanced enhanced indicators to support near real-time monitoring of disparities between majority and visible minority populations, offering a significant advantage over survey-based indicators, which are updated at best annually.

  • Advanced indicators of the exploratory type require the collection of original data, primarily through field experiments such as correspondence and audit studies, which offer strong potential for measuring bias-driven discrimination. Additionally, in more targeted cases, artificial intelligence can be leveraged to detect both the quantitative and qualitative underrepresentation of visible minorities in textbooks, as well as discriminatory patterns in employer searches on job portals. Despite several strengths, these methods also have limitations; they are not feasible in all settings, such as in schools or in workplaces, for instance when decisions are made about promotions.

• On top of core and advanced indicators, the chapter proposes a set of additional proxies for racism, racial/ethnic bias, and bias-driven discrimination, derived from cross-national surveys such as the European Values Survey (EVS), the European Social Survey (ESS), and the Eurobarometer survey on “Discrimination in the European Union”. These surveys allow for the measurement of two types of outcomes: attitudes of the majority population towards visible minorities, and perceptions or experiences of discrimination.

• The chapter concludes by exploring approaches for countries to enhance the identification of visible minorities within cross-national surveys and national statistical frameworks by incorporating long-established groups. Expanding data collection in this way would facilitate the adaptation of the indicators proposed in this chapter to these populations, thereby strengthening the monitoring of national anti-racism action plans aimed at protecting all segments of visible minorities.

  • A first approach involves gathering information on the country of birth across multiple generations by either adding questions about all four grandparents’ birthplaces to cross-national surveys, censuses, or large‑scale nationally representative surveys that already capture respondents’ and parents’ birthplaces, or by linking individuals to their family records in national population registers.

  • A second approach involves adding racial/ethnic identification questions to cross-national surveys, censuses, or large‑scale national surveys that already capture respondents’ and parents’ birthplaces. While this allows for identifying a broader range of visible minority subgroups beyond immigrants and their immediate descendants, it also raises concerns about potentially reinforcing racial/ethnic constructs and deepening social divisions. Strategies to enhance public acceptance of this approach are explored.

2.1. Introduction

Chapter 1 has demonstrated that racism permeates key areas of life for visible minorities. To effectively address the consequences of racism, it is crucial to first measure them. Robust monitoring allows countries to track progress, identify persistent challenges, and adapt strategies to ensure that national action plans against racism are both impactful and responsive to the needs of affected communities (European Commission, 2024[1]; European Commission, 2020[2]).

The purpose of this chapter is to support countries in evaluating the effectiveness of these action plans by proposing a set of indicators to assess the consequences of racism, hence to capture bias-driven discrimination against visible minorities across the key areas discussed in Chapter 1 – education, school-to-work transition, employment, housing, and health – as well as in the areas of poverty risk and low life satisfaction. Specifically, they seek to measure the full gaps resulting from unequal treatment of otherwise comparable individuals, driven solely by racial/ethnic bias against visible minority groups.

The proposed indicators primarily target recently arrived populations of non-European background, as they form a visible minority group common to many EU countries and for which cross-EU comparable data are most readily available. The capacity of EU-wide data collection to focus on these populations is exemplified by the 2021 addition of questions on parents’ country of birth to the core EU-Labour Force Survey (LFS) questionnaire, complementing the existing collection of data on respondents’ country of birth. That said, immigrants and their immediate descendant of non-European background only partially represents the full spectrum of visible minorities in the EU who are at risk of racialisation. Many have deeper roots, at least in some EU countries, including third-generation EU citizens of non-European descent, Black populations from overseas territories, and Roma people (see Box 2.1 for an overview). Consequently, the chapter concludes with a substantive discussion dedicated to extending the indicators proposed in its main sections to long-established visible minorities.

Even when focusing on recently arrived visible minorities for which data are most accessible, developing precise indicators that isolate the full impact of bias-driven discrimination on visible minorities, and nothing else, remains a challenge. Take the EU-LFS, the primary source for indicators measuring labour market disparities between the majority and visible minority populations due to bias-driven racial/ethnic discrimination. To ensure some level of precision, two key steps must be taken.

First, indicators should compare disparities between native‑born individuals with two native‑born parents (the majority population) and native‑born individuals with at least one foreign-born parent, distinguishing, where possible, between European and non-European backgrounds. Foreign-born individuals should be excluded from the analysis, as immigrants face unique challenges beyond discrimination. Since the EU-LFS cannot account for all these factors, including immigrants would inflate disparities without clear proof that discrimination is the driving force.

Second, since socio-economic disadvantages unrelated to discrimination can carry over to the next generation, it is crucial to account for disparities stemming from different parental backgrounds. However, as the EU-LFS lacks these data (except for youth living with their parents), education serves as a proxy.

Yet, even after these adjustments, the extent to which the measured labour market disparities reflect bias-driven racial/ethnic discrimination remains uncertain. This is because some of the variables used for adjustment, such as education, are themselves partly shaped by labour market discrimination. If visible minority students anticipate discrimination, they may invest less (or more) in education, leading to an underestimation (or overestimation) of discrimination’s impact on their labour market outcomes. These distortions are further compounded by unobserved factors that affect labour market outcomes differently for visible minorities and the majority population but are not captured in the available data. Regardless of whether these factors are themselves shaped by discrimination, their absence from the analysis introduces potential bias – either downward or upward – in the labour market gaps attributed to discrimination. One such unobserved factor is the effort individuals exert in the labour market, which can itself be influenced by discrimination. If discrimination compels visible minorities to adopt catch-up strategies, investing more effort than their majority peers, its impact will be understated. Conversely, if discrimination discourages visible minorities from fully engaging – causing them to be less active in job search when unemployed or to exert less effort once employed – its impact will be overstated.

Overall, there is a trade‑off between how precisely bias-driven discrimination against visible minorities is measured and the availability (and cost) of the data required for such measurement. This trade‑off gives rise to two categories of indicators.

The first category, composed of “core indicators”, is characterised by limited precision but high availability, as these indicators rely on EU-wide surveys. The second category, composed of “advanced indicators”, is divided into two subgroups: “enhanced indicators”, which offer greater precision than core indicators but require specific administrative data that are available in at most half of EU countries, and “exploratory indicators”, which provide the highest precision but are limited in availability due to their reliance on original data collection methods, such as field experiments, including correspondence and audit studies, and big data. A summary of these indicators is provided in Annex Table 2.A.1 through Annex Table 2.A.6 .

To enhance its relevance for policy makers committed to combating racism, Chapter 2 offers two additional insights. First, it includes a section proposing additional general proxies for racism, racial/ethnic bias, and bias-driven discrimination based on (cross-national) surveys, namely attitudes towards visible minorities or perceptions and experiences of discrimination. Second, the chapter discusses options for countries to consider in extending data collection on visible minorities to also include long-established groups.

The chapter opens with an overview of core indicators (Section 2) and advanced indicators (Section 3), followed by a discussion on additional proxies for racism, racial/ethnic bias and bias-driven discrimination (Section 4). The chapter concludes with Section 5, which addresses ways in which EU countries might improve the identification of visible minorities and the measurement of the disparities they may face within national statistics.

Before delving into the chapter’s main sections, it is important to note four caveats.

First, while the focus is on the EU, the proposed indicators, conditional on data availability, can be easily adapted to other countries or to monitor disparities affecting other groups within visible minorities, including those with longer-standing roots. Accordingly, Chapter 2 may serve as a reference for non-EU countries as well, should they aim to assess the situation of recently arrived or long-established visible minorities.

Second, when it comes to survey-based indicators – whether core indicators or broader proxies for racism, racial/ethnic bias, and bias-driven discrimination – this chapter focuses on both quantitative (e.g. income‑based) and qualitative (e.g. perception-based) measures that are nationally representative, a mixed-method approach recognised as key to comprehensively informing the fight against racism and discrimination (UNESCO, 2023[3]). These indicators are drawn from surveys that rely on probability-based sampling, ensuring statistical validity. However, some countries may wish to assess the impact of racism and anti-racism policies on specific subsets of visible minorities that are too small to be captured by national statistics, even with oversampling in general population surveys. In such cases, countries can complement the indicators proposed in this chapter with data derived from non-standard sampling techniques. Most of these approaches, however, do not produce representative results – except for certain methods like multi-stage stratified random sampling, which, for instance, the European Union Agency for Fundamental Rights uses to survey Roma populations in Europe. Other techniques, such as snowball sampling (where participants are recruited through personal networks) or social media sampling (where participants are recruited via platforms like Facebook, Twitter, LinkedIn, or Instagram), tough lacking representativeness, can still provide valuable insights – particularly for collecting qualitative information. For instance, semi-structured interviews can help uncover lived experiences and underlying mechanisms, while focus groups can shed light on how anti-racism policies and their impact have been perceived by targeted beneficiaries.

Third, a successful national action plan against racism will not necessarily result in a decline across all proposed indicators measuring racism’s harmful effects. In fact, perception-based indicators may move in the opposite direction. This pattern arises because effective anti-racism efforts are expected to raise awareness of racial/ethnic discrimination, enhancing the ability of visible minority groups – and society at large – to recognise discriminatory incidents. As a result, perceptions of how widespread discrimination is may actually increase, even as real progress is made in combating it.

Fourth, all proposed indicators are designed using methodologies that allow for intersectional analysis, making it possible to explore whether visible minorities who hold additional marginalised identities face compounded disadvantages compared to those who do not. This is a crucial step to ensure that anti-racism policies do not overlook individuals with double or triple minority statuses – whose compounded disadvantage could otherwise be exacerbated – but instead place them at the centre of policy action. For instance, the proposed indicators can be adapted to examine the intersection of visible minority status with gender or age. When looking at gender, for example, the indicators can assess whether visible minority women face greater discrimination than the sum of the disadvantages experienced by visible minority men (relative to majority men) and majority women (relative to majority men). This approach enables testing Kimberlé Crenshaw’s premise that the “intersectional experience” is greater than the sum of racism and sexism (Crenshaw, 1989[4]). Alternatively, the analysis can determine whether the disadvantage faced by visible minority women is less than this sum – while still potentially exceeding that experienced by individuals marginalised on only one dimension, hence surpassing either the separate Black or female penalties (see (McLaughlin and Neumark, 2025[5]) for an intersectional analysis of disadvantages in the labour market).

Box 2.1. Recently arrived and long-established visible minorities in the EU

Visible minorities in EU countries can be broadly categorised into two main groups.

Recently arrived visible minorities: Drawing on evidence on the geographic origin of immigrants and their immediate descendants (OECD/European Commission, 2023[6]), it is estimated that the average share of those of non-European background falls within a range of over 5% but less than 10%, although the exact value cannot be precisely measured due to data limitations.

• In 11 EU countries, the share of recently arrived visible minorities lies below 5%. They include Bulgaria Croatia, Czechia, Estonia, Greece, Hungary, Lithuania, Poland, Romania, the Slovak Republic, and Slovenia.

• In nine EU countries, the share of recently arrived visible minorities varies from 5% to 10%. They are composed of Austria, Cyprus, Finland, Germany, Ireland, Italy, Latvia, Luxembourg, and Malta.

• In seven EU countries, the share of recently arrived visible minorities exceeds 10%. They encompass Belgium, Denmark, France, the Netherlands, Portugal, Spain, and Sweden.

Long-established visible minorities which comprise at least six main groups:

• Indigenous populations, such as the Sámi people in Finland and Sweden, who are descendants of the original inhabitants now living within current national boundaries.

• Jews, who are present across all EU countries and beyond.

• Roma people and similar groups, such as Irish Travellers and Yenish, who remain predominantly concentrated in European countries, although significant numbers of Roma people migrated to North and Latin America during the 19th and 20th centuries.

• Black populations from the overseas territories of France (French Guiana, Guadeloupe, Martinique, Saint Barthélemy, Saint Martin, Mayotte, and Réunion) and the Netherlands (the Caribbean municipalities of Bonaire, Sint Eustatius, and Saba), who are primarily descendants of African slaves.

• Later generations of descendants of non-European immigrants, extending beyond immediate descendants to also include third generations (such as grandchildren of immigrants from non-European backgrounds) and potentially higher generations of EU citizens.

• Other long-standing groups whose distinctiveness benefits from official recognition in national legal frameworks to ensure their equal rights, including the preservation of their identities. These groups are officially recognised in 20 EU countries.

  • In most of them (16), these groups are explicitly recognised as minorities. They are typically referred to as “national minorities” when they identify with a nationality that has its own state. Conversely, when these groups possess a distinct language, culture, or tradition without affiliation to a nation-state, they are usually referred to as “ethnic minorities” (or linguistic or religious minorities if the focus is on their language or religion). As an illustration, in Poland, Armenians, Belarusians, Czechs, Germans, Jews, Lithuanians, Russians, Slovaks, and Ukrainians are recognised as “national minorities”, while Karaims, Lemkos, Roma, and Tatars are classified as “ethnic minorities”.

  • In contrast, in the remaining four countries, namely Belgium, Cyprus, Finland and Spain, while such groups are listed and their distinctiveness and right to preservation are officially acknowledged, their minority status is not explicitly mentioned – presumably to avoid stigmatizing integral constituents of the country, particularly in federal states where some of these groups wield significant legislative and executive powers.

2.2. Core indicators of bias-driven discrimination against visible minorities

This section introduces a set of core indicators that prioritise availability over precision, as shown in Annex Table 2.A.1 through Annex Table 2.A.6, where “+++” denotes strong availability and “+” indicates moderate precision in the columns devoted to assessing the availability of the proposed indicators across EU countries and to estimating their precision, respectively.

Box 2.2. The EU-LFS

The European Union Labour Force Survey (EU-LFS), which collects information on the respondent’s and their parents’ country of birth since 2021, is a household survey conducted across 34 countries, using probability sampling. The participating countries include the EU Member States, three EFTA countries (Iceland, Norway, Switzerland), and four Candidate Countries (Montenegro, North Macedonia, Serbia, and Türkiye).

The survey collects detailed information on all individuals aged 15 and above within selected private households, focusing on both labour market outcomes and related socio-economic factors, such as educational attainment and training. In cases where direct interviews with each eligible household member are not feasible, one member may provide responses on behalf of others. The EU-LFS consists of a core set of questions, consistently collected over time to allow for tracking labour market trends and related socio-economic factors. Additionally, it includes annually changing ad hoc modules designed to gather in-depth information on specific topics that may be of current interest or emerging relevance. These ad hoc modules vary each year and may cover topics such as work-life balance, young people in the labour market, or migration and labour market integration. Some ad hoc modules are unique to a particular year, while others may recur periodically, allowing for multi-year comparisons on certain themes.

Initially conducted annually, starting with the first wave in 1983, the EU-LFS transitioned to a continuous quarterly survey in 1998, though access to microdata for researchers remains annual. Sample sizes vary by country and are typically proportional to population size. For example, in EU countries, the annual sample sizes range from roughly 5 000 individuals in Malta to nearly 100 000 individuals or more in France, Germany and Italy (Eurostat, 2024[7]).

Core indicators are designed to measure disparities between the majority and visible minority groups across key life areas. Their key strength lies in their high availability, as they are derived from cross-national surveys, including the EU Labour Force Survey (EU-LFS), the Programme for International Student Assessment (PISA), the EU Statistics on Income and Living Conditions (EU-SILC), and the European Health Interview Survey (EHIS), making them computable across all EU countries (except Cyprus for PISA) (see Box 2.2, Box 2.3, Box 2.4, and Box 2.5 for an overview of these surveys).

Box 2.3. PISA

The Programme for International Student Assessment (PISA), which collects information on the respondent’s and their parents’ country of birth since 2018 (although some countries report significant levels of missing information), is a large‑scale international survey managed by the OECD, using probability sampling. Launched in 2000 and conducted every three years, with the next cycle scheduled for 2025, PISA has become a benchmark for assessing educational systems globally. Recent cycles have included over 80 countries and economies, encompassing OECD members and numerous partner economies. Among EU countries, all participate in PISA except Cyprus, with Bulgaria, Croatia, Malta, and Romania taking part as non-OECD members.

PISA evaluates the knowledge and skills of 15‑year‑old students nearing the end of compulsory education. It assesses competencies in mathematics, reading, and science as core areas, with additional optional assessments in fields like financial literacy and creative thinking. To further enrich the data, PISA administers several questionnaires alongside the main assessment. The student questionnaire gathers insights into students’ learning experiences, socio‑economic backgrounds, and attitudes toward learning, while the parent questionnaire (optional and thus used in selected countries) captures data on family resources and parental expectations. The teacher questionnaire, also optional, collects information on classroom practices and school resources, providing a holistic view of the learning environment.

In the 2022 cycle, sample sizes in EU countries ranged from around 3 000 students in Malta to more than 30 000 in Spain (OECD, 2023[8]).

In addition to their extensive coverage, these surveys offer several advantages.

• First, they collect information on individuals’ and their parents’ country of birth, either in every round (as in all these surveys except the EU-SILC) or within specific modules that rotate periodically, such as the EU-SILC’s ad hoc module on “intergenerational transmission of disadvantages”, conducted every six years. This facilitates the comparison between native‑born individuals with two native‑born parents (the majority population) and native‑born individuals with at least one foreign-born parent of non-European background.

• Second, these surveys go beyond capturing final outcomes like educational attainment or achievement. They also provide standardised measures of intermediate factors through which discrimination shapes final outcomes, such as a sense of exclusion at school, perceived teacher hostility, limited access to quality career counselling, or exposure to bullying. By doing so, they shed light on some of the mechanisms discussed in Chapter 1 through which bias may operate.

• Third, these surveys offer the valuable advantage of measuring a range of demographic factors that shape outcomes across the life areas examined in this report. For instance, educational outcomes are influenced by gendered societal expectations, parental education levels that impact learning environments, or disparities in access to quality resources based on urban or rural residence. Adjusting for these demographic factors is essential to isolate the differences in educational attainment between majority and visible minority youth that remain after the adjustment. This process enhances confidence that any remaining gaps – commonly referred to as adjusted, residual, or unexplained gaps – at least partially capture the impact of bias-driven discrimination.

However, it is precisely in this word, “partially”, that the main weakness of core indicators lies. While unexplained gaps suggest the presence of bias, they do not constitute definitive proof, as extensively discussed in the introduction. These indicators cannot fully isolate the impact of bias-driven discrimination on visible minorities, as they inherently carry the risk of either underestimating or overestimating its extent. This challenge is even more pronounced in countries where native‑born individuals with at least one non-European foreign-born parent make up a small share of the population, which results in small sample sizes that hinder the ability to compute adjusted gaps. This limitation is particularly likely given that maximum national sample sizes in EU cross-national surveys remain modest. As highlighted in Box 2.2, Box 2.3, Box 2.4, and Box 2.5, sample sizes exceed 100 000 individuals only in the EU-LFS, with substantially smaller samples in other surveys. For example, in PISA, only three EU countries (Finland, Italy, and Spain) had sample sizes above 10 000 participating students.

Box 2.4. EU-SILC

The European Union Statistics on Income and Living Conditions (EU-SILC) is a household survey conducted across 38 countries through probability sampling. Participants include all EU Member States, the United Kingdom, three EFTA countries (Iceland, Norway, Switzerland), six Candidate Countries (Albania, Bosnia and Herzegovina, Montenegro, North Macedonia, Serbia, and Türkiye), and Kosovo1 as a potential candidate.

EU-SILC serves as a key social policy tool for monitoring and supporting decision-making at the European level. Launched in 2003, the survey provides annual cross-sectional and longitudinal data covering a range of topics including income, housing, health, and life satisfaction. Each year, a core set of variables is consistently surveyed, while ad hoc modules, first introduced in 2005, provide additional focus on high-priority areas. As of 2021, these ad hoc modules are collected every three or six years, covering essential topics such as children, labour market and housing, and health in the 3‑year modules, and issues like over-indebtedness, access to services, intergenerational transmission of disadvantages, and quality of life in the 6‑year modules.

Household-related questions on housing and other living conditions are typically addressed by the household’s reference person, who responds on behalf of all members, while health and life satisfaction questions are answered by each individual aged 16 or older. Notably, while the EU-LFS annually collects information on respondents’ and parents’ country of birth, EU-SILC includes this detail only in the “Intergenerational transmission of disadvantages” ad hoc module, surveyed in 2011, 2019, 2023, and planned for 2029. This restricts the identification of minority households to once every six years –a minority household being defined as one in which the household reference person or their partner (if applicable) is native‑born with foreign-born parentage (allowing for distinction between European and non-European backgrounds), in contrast to a majority household where both the reference person and their partner are native‑born with two native‑born parents. However, for individuals aged 16 or older still residing with parents, it remains possible to analyse on an annual basis disparities in health and life satisfaction between those with two native‑born versus at least one foreign-born parent.

The sample size of EU-SILC varies by country, generally in proportion to population. For example, within the EU, Malta’s sample includes approximately 4 000 households and 10 000 individuals every year, while larger countries like France, Germany, and Italy exceed 10 000 households and 20 000 individuals (Wirth and Pforr, 2022[9]).

Box 2.5. EHIS

The European Health Interview Survey (EHIS), which collects information on the respondent’s and their parents’ country of birth since 2018, is a household survey conducted across 31 countries, using probability sampling (European Health Interview Survey (EHIS), 2024[10]). The participating countries include the EU Member States, two EFTA countries (Iceland and Norway), and two Candidate Countries (Serbia, and Türkiye).

The survey collects detailed information on all individuals aged 15 and above within selected private households, focusing on a wide range of health-related information, divided in four core areas which remain constant over survey cycles: health status, healthcare access and use, health determinants (such as lifestyle and environment), and socio-economic background.

The EHIS, first launched in 2006, has completed three waves, each approximately five years apart: the first wave (2006‑09), the second (2013‑15), and the third (2018‑20). The fourth wave is scheduled to start in 2025. Sample sizes differ by country, typically scaled to population size. They range from over 4 000 individuals in Malta to nearly 15 000 in France, close to 25 000 in Germany, and more than 45 000 in Italy (Eurostat, 2022[11]).

This caveat is compounded by two additional limitations affecting some proposed core indicators:

In certain cases, indicators are calculated based on a specific condition, thus relying on a subsample of the total population, which further reduces the number of observations available. For example, in health, the intermediate outcome capturing unexplained gaps in unmet medical needs is measured by the adjusted difference in the proportion of majority and visible minority individuals who, among those needing medical or dental care at least once in the past 12 months, reported unmet care needs on at least one occasion. As evident, this analysis is limited to individuals who required medical or dental care within the past year, diminishing the relevant sample size. Such reduction may even compromise the ability to further disaggregate native‑born individuals with at least one foreign-born parent by European or non-European background. To reflect these instances, which nuance the claim that core indicators are highly available, indicators with potentially limited applicability due to being based on a specific condition are flagged with the label “but small sample size” in the availability column of Annex Table 2.A.1 through Annex Table 2.A.6 .

Similarly, indicators based on the PISA parent questionnaire, which is optional, are flagged with the label “but in a selection of countries” in the availability column of Annex Table 2.A.1 through Annex Table 2.A.6.

2.1.1. Education

Chapter 1 underscores the presence of bias in children’s books and textbooks, instances of discriminatory behaviour among educators, and peer bullying targeting visible minority students, all of which can significantly impede their educational attainment and achievement. Indicators from cross-country surveys, notably the EU-LFS and PISA, offer valuable insights into these dynamics, focusing on youth aged 15 to 24.

This age range is relevant for several reasons. First, it captures the transition from post-compulsory education to further education, leading at least to a Bachelor’s degree, which students usually complete between the ages of 21 and 23 if they follow a standard educational path without interruptions. This age group is also relevant for monitoring the effects of educational policies aimed at combating bias-driven discrimination. Given that the outcomes for this group are likely to be influenced in a timely manner by such policies, it provides a meaningful window for evaluation. Furthermore, the 15‑24 age range allows for adjusting for socio-economic factors, such as parental education. While the EU-LFS does not include direct questions about parents’ education, this information can be inferred when children still live with their parents – a situation common for individuals below 24, as the average age of leaving the parental home in the EU was 26.5 years in 2021 (though this figure varies significantly across countries, ranging from 19 years in Sweden to 33.6 years in Portugal) (Eurostat, 2022[12]).

While it would be ideal to assess educational outcomes for individuals under 15, current cross-country surveys do not facilitate such analysis. The EU-LFS focuses on individuals aged 15 and above, and PISA assesses only 15‑year‑old students. That said, PISA’s student questionnaire includes retrospective questions about grade repetition, which could allow EU countries to explore unexplained gaps in the share of majority and visible minority students who repeated a grade in primary or lower secondary education, an option that is further discussed in the next section.

Regarding cross-country educational surveys targeting students under 15, such as PIRLS (Progress in International Reading Literacy Study) and TIMSS (Trends in International Mathematics and Science Study), they unfortunately do not collect detailed information on the country of birth of the respondent’s parents. PIRLS assesses the reading literacy of fourth-grade students (around 9‑10 years old) every five years, while TIMSS evaluates mathematics and science knowledge among students in fourth and eighth grades (ages 9‑10 and 13‑14) every four years. Yet, both surveys only gather data on whether the respondent’s father or mother was born abroad, which prevents distinguishing between students with European and non-European backgrounds.

Despite their limitations, the EU-LFS and PISA offer important insights into unexplained gaps in final educational outcomes likely driven by bias. Moreover, PISA sheds light on nearly all intermediate factors discussed in Chapter 1 that may contribute to shape these final outcomes, except for bias in textbooks, as no PISA questions address that specific dimension.

It is important to highlight that the insights gathered from students and their parents could be enriched by perspectives from teachers, as collected in the OECD’s Teaching and Learning International Survey (TALIS). This large‑scale international survey – covering all EU countries except Belgium, Greece, and Luxembourg – provides valuable data on teachers’ working conditions, teaching practices, and professional development, with each participating country surveying, on average, a nationally representative sample of 4 000 teachers. Notably, TALIS can shed light on trends in teachers’ capacity to teach in multicultural settings, as it includes several questions on whether teachers have received formal pre‑ or in-service training in this area – or if they feel the need for such professional development. Additionally, the survey explores how teachers engage with diversity in the classroom, asking about their involvement in practices such as reducing ethnic stereotyping among students or fostering collaboration between students with and without a migrant background. It also examines whether schools implement diversity-related initiatives, such as teaching students how to address ethnic and cultural discrimination. TALIS is conducted in five‑year cycles, with major waves in 2008, 2013, 2018, and 2024 edition (see (OECD, 2019[13]) and (OECD, 2021[14]) for an analysis of the 2018 edition from a multicultural perspective).

Final outcomes

Bias in textbooks, discriminatory behaviour by educators, and peer bullying targeting visible minority students can hinder their educational attainment and achievement, both directly and indirectly. Indirect processes include the internalisation of educators’ low expectations, as well as the stereotype threat effect – describing the heightened psychological pressure and anxiety students experience when they perceive a risk of confirming negative stereotypes associated with their racial/ethnic group.

The EU-LFS and PISA provide eight key indicators related to educational attainment and educational achievement, respectively (see Box 2.6 and Box 2.7 for a discussion of recommended adjustments when using indicators derived from these surveys).

Box 2.6. Recommended adjustments when analysing indicators derived from the EU-LFS

The EU-LFS yields core indicators covering the following three areas: education, school-to-work transition, and employment.

Education

For indicators related to educational outcomes, specifically educational attainment among youth aged 15 to 24, differences between majority and minority individuals should ideally be adjusted for gender, parental educational attainment (recorded in the EU-LFS when youth still reside with their parents), and urban or rural residence. Gender influences educational trajectories due to differing societal expectations and experiences that can shape academic choices and opportunities. Parental educational attainment is a critical determinant of youth academic success, as it often reflects the household’s overall educational environment and resources available for learning support. Additionally, urban or rural residence can affect access to quality education, by reflecting variations in school resources, teacher quality, and the availability of extracurricular programmes.

School-to-work transition

Similar control variables to those used for education should be applied when analysing the school-to-work transition, namely the difference in the share of majority and minority youth not in education, employment, or training (NEET), irrespective of whether the focus is on those aged 15‑24 or 15‑29.

Employment

For indicators aimed at capturing bias-driven discrimination at the recruitment stage and beyond, differences between majority and minority individuals aged 15 to 64 should ideally be adjusted for gender, age, highest educational attainment, and urban or rural residence. Gender significantly influences employment prospects, wage levels, and access to leadership roles, with women facing substantial barriers to career advancement. Age also plays a decisive role, with younger individuals encountering challenges related to job experience, while older workers may face age‑related discrimination. Educational attainment remains a critical determinant of labour market outcomes, as individuals with higher levels of education generally have better employment prospects and earnings potential. Lastly, access to employment opportunities often varies by location, with urban areas typically offering a higher number and diversity of jobs.

For analyses focused on post-hire outcomes, it is crucial to adjust, additionally, for industry, occupation, and whether individuals work on a full-time basis. Industries vary in job stability, wage structures, and promotion opportunities, while different occupations reflect different job responsibilities and skill requirements. Moreover, adjusting for the number of hours worked is important, as it directly impacts wage levels and may affect job security, with part-time contracts typically being more vulnerable to changes in employer needs, economic downturns, or organisational restructuring. When examining dismissal in a previous job, it is also important to control, whenever possible, for the individual’s tenure in that position.

Box 2.7. Recommended adjustments when analysing indicators derived from PISA

PISA yields core indicators covering education and school-to-work transition, as well as an indicator on racial/ethnic segregation (discussed further in the section on housing and not requiring specific adjustments).

Education

For indicators related to final educational outcomes, specifically educational achievement, differences between majority and minority individuals should ideally be adjusted for gender and socio-economic background, following the rationale outlined in Box 2.6. Notably, the OECD has developed the Index of Economic, Social, and Cultural Status (ESCS), a composite measure to capture the socio‑economic background of students participating in PISA (Awisati and Wuyts, 2024[15]). It combines three factors. First, parental education, based on the highest level of education attained by the parents. Second, parental occupation, reflecting the highest occupational status among parents. Third, home possessions, derived from student reports on various educational and cultural resources available in their homes, including items like books, a quiet study place, and access to the internet, as well as more culturally enriching possessions like artworks. Furthermore, adjusting the analysis for urban versus rural residence is necessary unless a more granular approach is feasible by introducing school fixed effects, which allows for comparisons between majority and minority students within the same school.

For indicators related to intermediate educational outcomes, the adjustment strategy may vary slightly depending on whether the data are derived from the student or parent questionnaire. If derived from the student questionnaire, the same control variables as those used for final educational outcomes should be applied. However, if substituting socio-economic background with educational achievement reduces the extent of missing data, then educational achievement should be preferred for the adjustment. If the indicators are derived from the parent questionnaire, the same set of student-level control variables should be applied. Additionally, it may be advantageous to include the gender of the responding parent or guardian, as parental perceptions of the school environment may differ by gender.

School-to-work transition

Indicators related to the school-to-work transition derived from PISA focus on intermediate outcomes and require the same adjustments as those applied to indicators associated with intermediate educational outcomes (see the paragraph above).

Educational attainment

The EU-LFS enables an exploration of unexplained gaps in the highest educational attainment among majority and visible minority individuals aged 15‑24. Specifically, this broad label could encompass the following three sub-indicators:

• The unexplained gap in the share of majority and minority individuals with at most a low educational attainment, i.e. no higher than lower secondary education, corresponding to ISCED Levels 0 (Early Childhood Education), 1 (Primary Education) or 2 (Lower Secondary Education) – see Box 2.8 for a description of ISCED levels);

• The unexplained gap in the share of majority and minority individuals with at most a medium educational attainment, corresponding to ISCED Levels 3 (Upper Secondary Education) or 4 (Post-Secondary Non-Tertiary Education). ISCED Level 4 includes short-term programmes that provide further education after upper secondary but are not at the tertiary level.

• The unexplained gap in the share of majority and minority individuals with high educational attainment, i.e. tertiary education, corresponding to ISCED Levels 5 (Short-Cycle Tertiary Education), 6 (Bachelor’s or Equivalent Level), 7 (Master’s or Equivalent Level) or 8 (Doctoral or Equivalent Level).

In the EU, more than one‑third of immigrants (35%) are low-educated – almost double the proportion among the native‑born population (20%). This significant disparity is primarily driven by non-EU migrants, 40% of whom attain only a low level of education. Similar disparities are observed among native‑born individuals of foreign-born parentage. EU-wide, young adults aged 25 to 34 with at least one foreign-born parent are less likely to be highly educated than their peers with two native‑born parents (32% versus 40%) and more likely to be low-educated (21% versus 14%). These gaps are particularly pronounced for native‑born individuals of non-EU parentage. Young native‑born women, whether their parents are native‑born or foreign-born, are more likely than their male peers to attain a high level of education in all EU countries. However, gender differences in educational attainment are less pronounced among native‑born individuals with foreign-born parents (OECD/European Commission, 2023[6]).

Box 2.8. The International Standard Classification of Education (ISCED)

The International Standard Classification of Education (ISCED), developed by UNESCO, is a framework for classifying educational programmes and qualifications by their level and field of study. The latest version, ISCED 2011, categorises education into the following eight levels:

• ISCED Level 0: Early Childhood Education, which introduces young children, typically under the age of 6, to structured learning environments before the start of compulsory education.

• ISCED Level 1: Primary Education, which provides basic instruction in fundamental subjects such as literacy, numeracy, and general knowledge, typically lasting 4 to 6 years and beginning when students are aged from 5 to 7.

• ISCED Level 2: Lower Secondary Education, which typically begins between the ages of 11 and 13 and builds upon the general education provided at the primary level, often introducing more subject-specific instruction.

• ISCED Level 3: Upper Secondary Education, which typically begins at around 15 to 16 years of age, lasts 2 to 4 years, and prepares students for either tertiary education or vocational training, offering increased specialisation.

• ISCED Level 4: Post-Secondary Non-Tertiary Education, which encompasses short-term programmes that provide further education after upper secondary, but below the tertiary level, such as vocational certificates, pre‑university courses.

• ISCED Level 5: Short-Cycle Tertiary Education, which comprises shorter tertiary programmes, typically lasting no more than two years, and is generally more practical or occupationally focused.

• ISCED Level 6: Bachelor’s or Equivalent Level, which includes academic and professional programmes that provide a broad and solid foundation in a field of study, typically lasting 3 to 4 years.

• ISCED Level 7: Master’s or Equivalent Level, which refers to advanced academic or professional education beyond the bachelor’s level, offering greater depth in a specific field. These programmes typically last 1 to 2 years.

• ISCED Level 8: Doctoral or Equivalent Level, which represents the highest level of education, emphasising original research and advanced knowledge in a field. These programmes typically last 3 to 6 years after a master’s degree.

Furthermore, the EU-LFS allows conditioning the analysis on each of these three educational levels, enabling examination of outcomes specific to each level, including:

• Conditional on having at most a low educational attainment, the unexplained gap in the share of majority and minority individuals who have dropped out of school, i.e. who are neither in education nor training. In the EU, the dropout rate among native‑born youth aged 15 to 24 with foreign-born parents stands at 11%, compared to 8% for their peers with native‑born parents. Within this group, young people with non-EU parentage are particularly vulnerable, with higher dropout rates than those whose parents are EU-born. Boys are more likely to drop out than girls, a disparity that is even more pronounced among native‑born youth of foreign-born parentage (OECD/European Commission, 2023[6]).

• Conditional on having at most a medium educational attainment, unexplained gap in the share of majority and minority individuals having enrolled in the vocational or technical track, rather than the general track, of upper secondary education.

• Conditional on high educational attainment, unexplained gap in the share of majority and minority individuals not having enrolled in fields of study that are most conducive to high labour earnings. Although high-earning fields may vary slightly across countries, they are generally concentrated within the following ISCED-F categories:2 Business, administration and law; Natural sciences, mathematics and statistics; Information and Communication Technologies; and Engineering, manufacturing and construction (OECD, 2022[16]).

Finally, as previously mentioned, PISA collects information on grade repetition, offering insights into educational attainment by capturing students’ ability to progress through the education system without delay. This survey enables the calculation of the unexplained gaps in the share of majority and minority students who report having repeated a grade at ISCED Levels 1 and/or 2.

Educational achievement

PISA measures educational achievement with standardised tests in reading, mathematics, and science.

Reading literacy is a foundational skill that is critical for personal, academic, and professional success in modern societies. PISA assesses reading literacy, which extends beyond the mechanical process of reading, i.e. converting text into sounds, to include a comprehensive set of competencies enabling individuals to effectively engage with written information presented in one or more texts for a particular purpose. PISA evaluates reading literacy by measuring how well students can understand, use, and reflect on written material. The focus is on students’ ability to derive meaning from various types of texts and to engage with them critically and constructively. This involves tasks such as extracting relevant information, understanding the relationships within and across texts, and evaluating the quality and credibility of the information provided (OECD, 2019[17]).

Likewise, a solid understanding of mathematics is essential for young people to be well-prepared for life in modern society. Many of the challenges and situations encountered every day, including those in professional settings, require a certain level of mathematical knowledge, reasoning, and the ability to use mathematical tools to fully comprehend and address them. Therefore, it is crucial to assess how effectively young people finishing compulsory schooling are equipped to apply mathematical skills to real-world issues and solve meaningful problems. In this context, PISA moves beyond the traditional approach of solving specific equations or abstract problems. Instead, it evaluates students’ ability to formulate, apply, and interpret mathematical concepts in a variety of concrete situations. This approach encompasses problem-solving, reasoning, and making judgments about quantities, shapes, patterns, and changes (OECD, 2019[18]).

Lastly, following the widely held view that a solid understanding of science is vital, PISA assesses whether students are equipped with three science‑specific competencies enabling them to understand and critically engage with scientific issues. The first competency is the ability to explain natural phenomena scientifically, which involves using scientific knowledge to describe and predict natural events. The second competency is the capacity to apply an understanding of scientific inquiry: this includes identifying questions that can be addressed through scientific investigation, proposing methods to explore these questions, and evaluating whether appropriate procedures were followed. The third competency is the ability to interpret and evaluate scientific data and evidence critically, assessing whether conclusions drawn from these data are valid and justified (OECD, 2019[18]).

Within this framework, PISA allows for the computation of unexplained gaps in reading, mathematics, and science literacy between majority and minority students, with each of these broad categories decomposable into the following three sub-indicators:

• The unexplained gap in the share of majority and minority students who are low achievers, performing below Level 2, which corresponds to scores below 335 in reading, 358 in mathematics and 335 in science (OECD, 2023[19]).

• The unexplained gap in the share of majority and minority students who are medium achievers, typically scoring at Levels 2 to 4.

• The unexplained gap in the share of majority and minority students who are high achievers, performing at Levels 5 or 6, which corresponds to score above 626 in reading, 607 in mathematics and 633 in science.

The 2018 edition of PISA reveals that, in the EU, children of native‑born parents outperform their peers with foreign-born parents in reading skills by the equivalent of one year of schooling. Even after accounting for socio-economic status, children of foreign-born parents still lag behind by approximately half a school year. This disparity extends to basic reading proficiency: 29% of native‑born pupils with foreign-born parents lack basic reading skills, compared to just 18% of those with native‑born parents (OECD/European Commission, 2023[6]). Preliminary results from the 2022 edition of PISA confirm a similar pattern, including comparable gaps in mathematics performance (OECD, 2023[20]).

Intermediate outcomes

In addition to assessing educational attainment through grade repetition and measuring educational achievement with standardised tests in reading, mathematics, and science, PISA collects rich data on students’ and their parents’ educational experiences and perceptions of the school environment. Specifically, the PISA student questionnaire enables the creation of five indicators that shed light on intermediate factors through which bias-driven racial/ethnic discrimination may hinder students’ academic progress.

Students’ and parents’ sense of exclusion at school

PISA allows devising a first indicator capturing students’ sense of exclusion within the school setting, which may be influenced by the three channels detailed in Chapter 1: lack of representation and recognition in children’s books and textbooks, unequitable treatment by educators, and low acceptance by peers. Specifically, this indicator captures the unexplained difference in the share of majority and minority students who agree or strongly agree with “I feel like an outsider (or left out of things) at school”, who disagree or strongly disagree with “I feel like I belong at school”, and who agree or strongly agree with “I feel awkward and out of place in my school”.

While the share of 15‑year‑old native‑born pupils with foreign-born parentage who report feeling awkward or out of place at school (17%) is comparable to that of their peers with two native‑born parents across EU countries, this average masks significant disparities: in the majority of EU countries, native‑born pupils of foreign-born parentage report a weaker sense of belonging at school (OECD/European Commission, 2023[6]).

Furthermore, PISA provides an additional indicator offering an opportunity to proxy for a channel – only briefly mentioned in Chapter 1 – namely, the unfair treatment of minority parents by the school system. This issue has been empirically supported primarily in the United States, with limited evidence in Europe. Specifically, the PISA questionnaire for parents enables the derivation of two sub-indicators related to parents’ sense of exclusion within the school setting.

• The first sub-indicator captures the unexplained gaps in the share of parents of majority and minority students who report that their participation in activities at their child’s school was hindered by feeling unwelcome.

• The second sub-indicator measures the unexplained gaps in the share of parents of majority and minority students who disagree or strongly disagree with all the following three statements: “My child’s school provides an inviting atmosphere for parents to get involved”; “My child’s school provides effective communication between the school and families”; and “My child’s school involves parents in the decision-making process”.

Biased educators

The next two indicators delve into potential bias among educators. The first examines students’ perceptions of teacher hostility, while the second explores their lack of exposure to (quality) career counselling in schools.

The first measure of perceived teacher hostility examines the unexplained gap in the share of majority and minority students reporting negative interactions with their teachers. Specifically, it considers students who disagree or strongly disagree with all of the following six statements: “The teachers at my school are respectful towards me”; “If I walked into my classes upset, my teachers would be concerned about me”; “If I came back to visit my school three years from now, my teachers would be excited to see me”; “When my teachers ask how I am doing, they are really interested in my answer”; “The teachers at my school are friendly towards me”; and “The teachers at my school are interested in students’ well-being”. Simultaneously, it includes students who agree or strongly agree with the two statements: “I feel intimidated by the teachers at my school”; and “The teachers at my school are mean towards me”.

In the EU, one‑third of 15‑year‑old native‑born children of immigrants believe that most of their teachers hold discriminatory attitudes toward other cultural groups, compared to one‑quarter of their peers with native‑born parents. The most frequently cited issue is that teachers tend to have lower academic expectations for students from different cultural backgrounds (OECD/European Commission, 2023[6]).

The second indicator focuses on career counselling and encompasses the following two sub-indicators:

• One which concentrates on the lack of exposure to career counselling at school by calculating the unexplained gaps in the share of majority and minority students who report not having spoken to a career advisor at school (responding “No” to the question, “I spoke to a career advisor at my school”).

• One which proxies for students’ dissatisfaction with career counselling by computing the unexplained gap in the share of majority and minority students who disagree or strongly disagree with the statement, “I feel well-informed about possible paths for me after [the final year of compulsory education]”.

Biased schoolmates

The last indicator focuses on the degree of biased attitudes and behaviours among schoolmates and comprises two sub-indicators:

• The first sub-indicator assesses students’ peer isolation at school by calculating the unexplained gap in the share of majority and minority students who disagree or strongly disagree with “I make friends easily at school” and “Other students seem to like me”, and who agree or strongly agree with “I feel lonely at school”.

• The second sub-indicator assesses students’ experiences of bullying by schoolmates, by computing the unexplained gap in the share of majority and minority students who report being bullied “a few times a year”, “a few times a month”, or “once a week or more”, compared to those who report “never or almost never”, across each of the following nine scenarios: “Other students left me out of things on purpose”; “Other students made fun of me”; “I was threatened by other students”; “Other students took away or destroyed things that belonged to me”; “I got hit or pushed around by other students”; “Other students spread nasty rumours about me”; “I was in a physical fight on school property”; “I stayed home from school because I felt unsafe”; and “I gave money to someone at school because they threatened me”.

While the proportion of 15‑year‑old children of immigrants reporting bullying EU-wide aligns with that of their peers with two native‑born parents, the data reveal significant disparities. In over two‑thirds of EU countries, children of immigrants are more frequently subjected to bullying than their native‑born counterparts (OECD/European Commission, 2023[6]).

Interestingly, this trend reverses depending on the socio‑economic context: in underprivileged schools, children of immigrants experience less bullying than their peers with native‑born parents, whereas in socio‑economically privileged schools, the opposite is true. This finding aligns with previous research, such as (Vitoroulis and Georgiades, 2017[21]) in the Netherlands, which highlights how the concentration of immigrant students in a school shapes the risk of bullying associated with an immigrant background. However, that study also found that non-White students faced higher odds of racial/ethnic victimisation compared to their White peers, regardless of the school’s immigrant concentration.

2.1.2. School-to-work transition

Bias-driven discrimination against visible minorities can significantly hinder their school-to-work transition, prolonging the time between leaving formal education and securing quality employment. This occurs through various mechanisms. Low educational attainment and achievement, as well as difficulties in accessing entry-level jobs – topics addressed in the previous and following sections – are two key factors. However, as outlined in Chapter 1, research demonstrates that racial/ethnic bias introduces additional barriers, such as limited access to work-based learning opportunities during formal education (e.g. internships and apprenticeships), and disproportionate exposure to disciplinary actions both inside and outside the school environment, including when interacting with law enforcement.

The EU-LFS offers a measure of the final outcome that can serve as a proxy for assessing the effectiveness of the school-to-work transition. Meanwhile, PISA helps explore some of the underlying mechanisms contributing to these disparities (see Box 2.6 and Box 2.7 for a discussion of recommended adjustments when using indicators derived from these surveys).

Final outcomes

The EU-LFS enables the measurement of unexplained gaps in the share of majority and minority youth who are NEET (Not in Education, Employment, or Training). Youth can be defined either as those aged 15‑24 or 15‑29, each definition offering distinct advantages and limitations.

Focusing on the 15‑24 age group allows better controlling for parental socio-economic background, as a significant proportion of this age group still resides with their parents. However, this narrower focus could underestimate the extent of NEETs, as many individuals in this age range may still be pursuing tertiary education.

In contrast, focusing on the 15‑29 age group offers a more comprehensive analysis of youth transitions, as most individuals who pursued tertiary education are likely to have completed their studies by their late twenties. However, this broader age range includes a greater share of youth who no longer live with their parents, making it more challenging to adjust for parental socio-economic background.

Across the EU, native‑born youth aged 15 to 34 with at least one foreign-born parent are slightly more likely than their peers with two native‑born parents to be NEET (16% vs. 15%). However, this average masks large disparities in most EU countries, with gaps reaching nearly 10 percentage points in France and Belgium. Young women and the least educated are particularly vulnerable to being NEET, regardless of whether their parents are foreign-born or native‑born. However, even highly educated individuals with at least one foreign-born parent face higher NEET rates than their counterparts with two native‑born parents (OECD/European Commission, 2023[6]).

Intermediate outcomes

PISA offers proxies for each of the two channels that hinder school-to-work transition beyond low educational attainment and achievement, and difficulties in accessing entry-level jobs.

• The PISA student questionnaire provides a basis for deriving an indicator on the lack of work-based learning experience during formal education. This indicator measures the unexplained gap in the share of majority and minority students who report not having done an internship to explore future study options or career paths. Yet, it is important to keep in mind that, while insightful, this indicator would be even more informative if gaps in internship experiences could also be measured beyond age 15, as older students often have increased access to structured work-based learning, allowing for a better understanding of disparities in this field.

• Both the PISA student and parent questionnaires provide insights into exposure to disciplinary actions within the school setting.

  • A first indicator, which relies on the student questionnaire, consists in calculating the unexplained gap in the share of majority and minority students who report having missed school for over three months in a row due to suspension for reasons such as violence, aggression, or drug-related issues, be it at ISCED Levels 1, 2 or 3. However, this measure has a notable limitation. Although unexplained gaps always require cautious interpretation – since unobserved factors may contribute to differences – this is especially true for this indicator. One critical factor that could affect results is the potential difference in behaviour between majority and minority students, a variable that, unfortunately, cannot be observed in PISA and thus not factored in when analysing disparities.

  • A second indicator, which relies on the parent questionnaire, captures parental dissatisfaction with disciplinary practices at school, allowing for the computation of the unexplained gap in the share of majority and minority parents who disagree or strongly disagree with the statement: “I am satisfied with the disciplinary atmosphere in my child’s school.” While this indicator offers valuable insight, it also has limitations. Notably, it may reflect factors unrelated to differential exposure to disciplinary actions, such as differences in cultural norms regarding how strict school rules should be and how rigorously they should be enforced.

2.1.3. Employment

In the field of employment, the literature reviewed in Chapter 1 suggests that bias-driven racial/ethnic discrimination can occur at various stages: at the recruitment stage, and after hiring, including potentially at the point of dismissal. The EU-LFS enables the measurement of unexplained differences in final outcomes across these two stages (see Box 2.6 for a discussion of recommended adjustments when using indicators derived from this survey).

To account for the possibility that state action against racism may be more effective in the public than in the private sector, it is essential to analyse some of the proposed indicators separately for each sector, in addition to examining them economy-wide. This is particularly relevant for unexplained gaps in employment rates and labour earnings.

At the recruitment stage

Hiring discrimination against visible minorities likely reduces their employment prospects, thereby increasing their risk of unemployment, particularly long-term unemployment, as well as their risk of involuntary inactivity due to discouragement. Moreover, hiring discrimination can lead to their overrepresentation in lower-quality jobs, that is jobs not related to standard employment (defined as open-ended, full-time, dependent work) and/or for which visible minority job candidates are typically overqualified, as shown by (Drydakis, Paraskevopoulou and Bozani, 2022[22]) reviewed in Chapter 1.

Indicators for each of these dimensions can be derived from the EU-LFS. The first indicator assesses individuals’ labour market status and consists of three sub-indicators that compute the unexplained gap in the proportion of majority and minority individuals across the following situations: (i) being inactive, (ii) being unemployed, and (iii) being employed.

Across the EU, 65% of immigrants are employed, compared with 69% of the native‑born, while the immigrant unemployment rate (12%) is twice that of the native‑born. Disparities persist when focusing on their immediate descendants, comparing native‑born youth of foreign-born versus native‑born parentage. In the EU, only slightly more than two‑thirds of native‑born individuals aged between 15 and 34 with foreign-born parents are in employment (67%), while this share exceeds three‑quarters (77%) among their peers with two native‑born parents. Moreover, 17% of native‑born youth of foreign-born parentage are unemployed, against 10% of their peers with native‑born parents. Young men are more likely to be unemployed than young women, and such gender gaps tend to be wider among native‑born with foreign-born parents (OECD/European Commission, 2023[6]).

The next three indicators provide a more detailed analysis of the previously mentioned labour market status categories by examining the unexplained gaps in the proportion of majority and minority individuals who:

• Conditional on being inactive, are involuntarily so, defined as individuals who wish to work but are not actively searching for employment. The EU-LFS provides additional insights into reasons for involuntary inactivity, such as “No suitable job is available” (indicating discouragement), instead of other reasons like “Currently in education or training”, “Own illness or disability”, “Care responsibilities”, or “Laid off and waiting to be called back to work”. In the EU, involuntary inactivity affects 28% of the foreign-born population compared to 18% of the native‑born. Women are generally less likely than men to experience involuntary inactivity, though the gender gap is markedly wider among immigrants than among the native‑born. Among involuntarily inactive men, discouragement in the labour market is the most common reason, while for women, family responsibilities are the primary driver – affecting migrant women nearly twice as much as their native‑born counterparts across the EU (OECD/European Commission, 2023[6]).

• Conditional on being unemployed, are in long-term rather than short-term unemployment, where long-term unemployment refers to individuals who have been unemployed for 12 months or more and have actively sought work during this period. EU-wide, the share of immigrants and native‑born who are long-term unemployed is similar. However, immigrants from non-EU countries are more likely to be unemployed than native‑born, as are immigrant women when compared with their native‑born peers (OECD/European Commission, 2023[6]).

• Lack access to quality employment, which includes the following four sub-indicators:

  • Conditional on being employed, being self-employed rather than in dependent employment, where dependent employment refers to individuals working under an employer and receiving wages or salaries under a contractual relationship, as opposed to self-employment where individuals run their own business. In the EU, immigrants are just as likely as the native‑born to be self-employed. However, the reasons behind their self-employment differ significantly: 30% of immigrants turn to self-employment out of necessity, due to a lack of alternative options, compared to only 20% of their native‑born peers (OECD/European Commission, 2023[6]). The 2024 edition of the OECD International Migration Outlook, which notably includes a focus on migrant entrepreneurship, confirms that self-employment remains a strategy for immigrants to overcome difficulties in accessing wage employment (OECD, 2024[23]).

  • Conditional on being in dependent employment, holding a fixed term (or temporary) rather than an open-ended contract. In nearly all EU countries, immigrant workers are more likely than native‑born workers to hold temporary contracts, with EU-wide shares at 17% compared to 10% (OECD/European Commission, 2023[6]).

  • Conditional on being in dependent employment, engaging in part-time rather than full-time work. The EU-LFS offers the possibility to further distinguish, among part-time workers, between voluntary and involuntary part-time work, with involuntary part-time defined as individuals who work part-time but would prefer full-time employment if available. Across the EU, immigrants are more likely than their native‑born peers to work part-time, with 22% of immigrants in part-time employment compared to 16% of native‑born workers. Additionally, immigrants are disproportionately represented in involuntary part-time work: around 30% of immigrants express a desire to work longer hours, compared to just 20% of their native‑born counterparts (OECD/European Commission, 2023[6]).

  • Conditional on being in dependent employment, being overqualified, i.e. having tertiary education (ISCED Levels 5 to 8) while being employed in jobs classified as low- or medium-skilled (ISCO Levels 4‑9; see Box 2.9 for a description of ISCO Levels). In the EU, overqualification is significantly more common among immigrants than among the native‑born: approximately one‑third (32%) of highly educated immigrants are overqualified for their jobs, compared to one‑fifth (20%) of their native‑born peers. These disparities persist among the immediate descendants of immigrants. Overqualification is a particularly acute issue for native‑born youth with non-EU parentage, compared to those with EU parentage (OECD/European Commission, 2023[6]).

After hiring

As outlined in Chapter 1, discrimination does not end at hiring but persists throughout employment, manifesting in biased managerial behaviour that affects promotion opportunities, supervision, and wage negotiations. This bias ultimately results in lower labour earnings for visible minorities – either because they face obstacles in career progression despite comparable performance, struggle to reach their full potential under discriminatory supervision, or are paid less than their peers for work of equal value. Additionally, this bias may restrict their access to lifelong learning opportunities. Finally, tentative evidence from Germany suggests that visible minorities may also face firing discrimination, being more likely to be dismissed during economic downturns, even after accounting for productive characteristics.

Three key indicators can be derived from the EU-LFS to capture the impact of bias in the workplace. The first indicator focuses on the unexplained gap in labour earnings between majority and minority employees.

Bias in managerial practices can also manifest in disparities in access to life‑long learning and training opportunities. Although research on this form of discrimination remains limited, the EU-LFS facilitates the assessment of unexplained gaps in the lack of exposure to job-related education or training programmes initiated or recommended by employers. Across the EU, immigrants are slightly less likely than native‑born individuals to participate in adult education, with participation rates of 6% for immigrants compared to 7% for the native‑born (OECD/European Commission, 2023[6]).

Such unequal access to life‑long learning could lead to faster skills depreciation among minority employees, affecting critical competencies such as literacy, numeracy, and problem-solving in technology-rich environments, essential for adaptability in evolving job markets. These competencies are measured internationally through the OECD’s Programme for the International Assessment of Adult Competencies (PIAAC). Launched in 2011, PIAAC’s first cycle evaluated skills in over 24 countries, with a second cycle initiated in 2022 across 31 countries. This representative household survey targets adults aged 16 to 65, with national sample sizes typically ranging between 5 000 and 10 000 individuals. However, the survey only gathers information on whether respondents’ parents were born abroad, without specifying their country of origin. This limitation prevents from differentiating minority adults by European versus non-European background.

Regarding proxying for firing discrimination, the EU-LFS enables the computation of the unexplained gap in the share of majority and minority respondents who, among those with prior work experience, reported leaving their previous job due to dismissal. This indicator captures individuals who mentioned that they “lost job or made redundant or laid off”, in contrast to other motives such as “contract ended or job was temporary”, “caring responsibilities”, “resuming education or training”, “illness or disability”, or “retirement”.

Box 2.9. The International Standard Classification of Occupations (ISCO)

The International Standard Classification of Occupations (ISCO), developed by the International Labour Organization (ILO), classifies jobs based on the skills and qualifications required. The latest version, ISCO‑08, organises occupations into nine major groups, ranging from the least to the most skilled:

• Low-skilled occupations, corresponding to ISCO Major Group 9, require basic education, typically aligned with ISCED Levels 0‑2.

• Semi-skilled occupations, requiring upper secondary education (ISCED Levels 3‑4), encompass several major groups:

  • ISCO Major Group 8 (Plant and machine operators and assemblers).

  • ISCO Major Group 7 (Craft and related trades workers).

  • ISCO Major Group 6 (Skilled agricultural, forestry, and fishery workers).

  • ISCO Major Group 5 (Service and sales workers).

  • ISCO Major Group 4 (Clerical support workers).

• Technicians and associate professionals (requiring ISCED Level 5 and corresponding to ISCO Major Group 3).

• Highly skilled occupations, requiring tertiary education (ISCED Levels 6‑8) include :

  • ISCO Major Group 2 (Professionals).

  • ISCO Major Group 1 (Managers).

2.1.4. Housing

As highlighted in Chapter 1, bias-driven racial/ethnic discrimination adversely affects housing outcomes for visible minorities in several ways. First, discrimination in the private sale housing market and restricted access to mortgage loans significantly limit homeownership opportunities for racial/ethnic minorities. Simultaneously, discrimination in the private rental housing market often pushes visible minorities into social housing at disproportionately higher rates.

Second, this multifaceted discrimination likely undermines the value for money of the housing accessible to visible minorities, as they may not receive priority for high-quality housing. This effect is compounded by evidence, reviewed in Chapter 1, of differential treatment extending to the prices paid for both homes for sale and rental properties, as well as to the cost of mortgages approved for minority buyers.

Third, these two channels are expected to exacerbate racial and ethnic segregation – the physical separation of racial or ethnic groups. This issue is further intensified when, as evidence suggests, bias extends beyond landlords, real estate agents, and mortgage lenders, to include neighbours, reinforcing patterns of White avoidance, where White residents tend to leave or avoid neighbourhoods once the visible minority population surpasses a certain threshold.

The EU-SILC sheds light on final housing outcomes related to the first two mechanisms (see Box 2.10 for a discussion of recommended adjustments when using indicators derived from this survey). Additionally, PISA provides an indicator that sheds light on racial/ethnic segregation.

Box 2.10. Recommended adjustments when analysing indicators derived from EU-SILC

The EU-SILC yields core indicators covering the following three areas: housing, health, and poverty risk and low life satisfaction.

• Housing

  For indicators related to housing (which are computed at the household level), differences between majority and minority households should, at a minimum, be adjusted for the gender and age of the household reference person, as well as for household disposable income, household size, and urban or rural residence. The gender and age of the household reference person are important because they often influence household financial stability. Household size and disposable income provide a basis for comparing households with similar economic resources, while accounting for urban or rural residence adjusts for the geographic disparities in housing availability and cost, which can vary significantly by location.

• Health

  For indicators related to health, differences between majority and minority individuals should, at a minimum, be adjusted for the gender, age, and highest educational attainment of the individual, as well as for household disposable income, and urban or rural residence. Both gender and age play pivotal roles in shaping health outcomes. Gender-specific roles and societal expectations drive variations in stress levels, access to resources, and health-seeking behaviours. For instance, women often encounter unique challenges to life satisfaction and mental health due to pressures around work-life balance, while men may face mental health risks, as prevailing norms can discourage them from seeking psychological support. Aging, too, influences these dimensions: while the likelihood of chronic conditions rises with age, which can impact overall well-being, increased life experience and, frequently, enhanced financial stability offer a counterbalance, contributing positively to mental health. Higher educational attainment generally promotes health literacy and healthier behaviours. Similarly, household disposable income significantly impacts access to quality healthcare and the ability to maintain a healthy lifestyle. Lastly, living in an urban or rural area affects the availability of healthcare services.

• Poverty risk and low life satisfaction

  For indicators assessing poverty risk (which is computed at the household level), the analysis should include the same control variables used for housing-related indicators, with one adjustment: instead of household disposable income (since it is used to calculate poverty risk), the highest educational attainment of the household reference person should be taken into account. This substitution serves as a valuable proxy for earnings potential, enabling a more accurate assessment of living conditions, including poverty risk, across various household types.

  For indicators assessing low life satisfaction, the analysis should include the same control variables used for health-related indicators, given their strong correlation with overall well-being. For example, living in an urban or rural area affects not only access to healthcare services but also availability of recreational amenities and social support networks, both of which have a significant influence on life satisfaction.

Housing tenure

The EU-SILC provides one indicator related to housing tenure, which can be further broken down into two sub-components. First, it facilitates the analysis of unexplained gaps in homeownership rates between majority and minority households by measuring the proportion of tenants versus homeowners within each group. Additionally, it supports the assessment of disparities among tenants by examining the share of majority and minority households renting at reduced rates, which serves as an indicator for social housing residency.

In the EU, home ownership among the native‑born population aged 16 and over is nearly twice as high as that of the foreign-born (74% vs. 42%). This disparity may reflect not only discrimination in property purchases but also other factors, such as lower financial resources, limited knowledge of the host country’s housing market, and the absence of inherited housing assets in the host country. Additionally, foreign-born renters across the EU are slightly more likely than their native‑born counterparts – by 2 percentage points – to live in dwellings rented at a reduced rate (OECD/European Commission, 2023[6]).

Since EU-SILC is a household survey collecting data from individuals living in private households, those without a fixed residence, including homeless individuals who do not reside in conventional housing or shelters, are typically excluded from the sample. Yet, discrimination in the private housing market (in access to both homeownership and rentals) combined with capacity constraints in social housing, can disproportionately push minorities towards homelessness.

Unfortunately, to our knowledge, very few countries report homelessness statistics in a manner that allows for the computation of unexplained gaps between native‑born individuals with two native‑born parents and those with at least one foreign-born parent. Moreover, none provides further disaggregation by European and non-European background. For example, Denmark has conducted a biannual survey on homelessness since 2007, collecting detailed personal information, including gender, age, immigrant background, citizenship, income, and whether the individual has children living with them in homelessness. The 2022 Danish report reveals a significant overrepresentation of second-generation Danish citizens among the homeless population, as immediate descendants of immigrants account for 8% of homeless individuals, compared to their 3% share in the total population (VIVE, 2022[24]). This lack of detailed reporting reflects a broader context where collecting comparable cross-country data on the overall size of the homeless population, let alone the share of immigrants within it, remains highly challenging (see Box 2.11).

Box 2.11. Challenges to measuring homelessness among migrants in OECD and EU countries

Comprehensive, comparable data on homelessness among migrants in OECD and EU countries are lacking. Fewer than half of OECD and EU countries report the share of migrants in their national homelessness statistics, while the remaining 20 countries do not disaggregate homelessness data by migrant status.

Many of the broader methodological challenges that hinder the measurement and cross-country comparison of homelessness also affect how accurately migrants are accounted for in official homelessness statistics. These challenges include variations in definitions and measurement of homelessness, often leading to underestimation of both general homelessness and homelessness among migrants. Notably, there is no internationally harmonised definition of homelessness, leading governments to define and measure it in different ways. The widely used ETHOS Light Typology, which facilitates cross-country comparisons, outlines six categories of homelessness, including, among others, people sleeping rough (ETHOS 1), staying in emergency accommodation (ETHOS 2) or temporary shelters (ETHOS 3), and those temporarily housed with family or friends (ETHOS 6). While some countries adopt a narrow approach, such as Japan, which only includes individuals living rough, others use broader definitions. As a result, cross-country comparisons of homelessness statistics are not always meaningful.

Moreover, there are additional country-specific approaches to data collection when it comes to migrants, particularly regarding asylum seekers and refugees. In some OECD and EU countries, official homelessness statistics explicitly include individuals staying in temporary accommodation for asylum seekers and refugees, whereas other countries exclude them from their counts.

Where disaggregated data on migrant homelessness are available, estimates indicate that migrants are overrepresented among individuals experiencing homelessness. However, due to inconsistencies in definitions and data collection, these estimates cannot be readily compared across countries.

Source: (OECD, 2024[25]), “Challenges to measuring homelessness among migrants in OECD and EU countries”, https://www.oecd.org/en/publications/challenges-to-measuring-homelessness-among-migrants-in-oecd-and-eu-countries_b9855842-en.html.

Value for money in housing

In terms of the value for money of housing accessible to visible minorities, one key indicator can be derived from EU-SILC to assess unexplained gaps between majority and minority households with respect to the following three sub-items:

• Living in an overcrowded accommodation, defined as housing where the number of rooms is insufficient according to the following criteria: one living room, plus one room for each single person or couple responsible for the household, plus one room for every two additional adults, and one room for every two children. Across the EU, more than one‑third (34%) of children under 16 in immigrant households live in overcrowded accommodation, compared to one‑fifth (20%) of their peers in native‑born households. A closer examination reveals that these disparities are largely driven by children with parents born outside the EU (OECD/European Commission, 2023[6]).

• Living in a substandard accommodation, which includes housing conditions such as being too dark, lacking exclusive access to a bathroom, or having significant issues like a leaking roof. In the EU, 26% of immigrants live in substandard housing, compared to 20% of the native‑born (OECD/European Commission, 2023[6]).

• Spending more than 40% of disposable income on housing costs (after accounting for housing allowances) which represents the housing cost overburden rate. Across the EU, nearly one‑fifth (19%) of immigrants aged 16 and over are overburdened by housing costs, compared to around one‑eighth (12%) of their native‑born counterparts (OECD/European Commission, 2023[6]).

Racial/ethnic segregation

No cross-country survey currently enables the measurement of racial/ethnic residential segregation. Accurately capturing such segregation requires exhaustive data on populations living in narrowly defined geographic areas (Verdugo, 2011[26]), including information on individuals’ and their parents’ country of birth. This level of detail is typically available only in some countries, through national population registers or census data.

However, PISA offers the possibility to compute racial/ethnic segregation at the school level, which provides two key insights. First, it serves as a proxy for residential segregation. School composition is often influenced by the segregation present in the surrounding neighbourhood as many parents prefer schools near their homes to avoid long commutes for their children (OECD, 2019[27]). Second, racial/ethnic segregation at the school level can also be driven by the flight or avoidance of majority parents, who may prefer schools with a higher concentration of similar peers. These alternative schools are often private schools, particularly in countries where residence‑based school assignment is strong. However, they are also public schools, especially in contexts where residence‑based assignment is either weak or non-existent (Givord, 2019[28]).

The detrimental impact that private schools exert on segregation within the public education system has been well-documented for France, although the focus has primarily been on social segregation due to the lack of systematic data on its racial and ethnic counterpart (Boutchenik, Givord and Monso, 2018[29]; Boutchenik, Givord and Monso, 2021[30]; Souidi, 2023[31]; Frohly, 2022[32]). It is important to recall, however, that social and racial/ethnic segregation are strongly correlated. Interestingly, the concentration of students from privileged backgrounds, and by extension, those from racial/ethnic majority backgrounds, within private schools may not solely result from parental school choice. It may also derive from discriminatory practices on the part of private schools toward visible minority students, as highlighted in Box 2.12.

Box 2.12. Discrimination against visible minority students at the entry to private schools

Only one field experiment has explored whether private schools discriminate. Specifically, in 2011, a group of researchers examined discrimination based on presumed parental national origin at the entry point of 4 269 private schools across France. They created the fictional identities of two fathers: one with a French-sounding name, and the other with a Maghreb-sounding name. A few days apart, these two fictional fathers sent short messages to each of these schools, requesting more information about enrolling their child for the upcoming school year. The researchers then compared the responses from the schools to these two inquiries.

The results demonstrate the existence of discriminatory practices. Specifically, the father with a French-sounding name was more than 40% more likely to receive a “positive” response from the private schools, including offering a meeting with the principal, sending an application form for review, providing conditional acceptance with placement on a waiting list, or giving a firm acceptance.

Source: (Brodaty, Du Parquet and Petit, 2014[33]), “La discrimination à l'entrée des établissements scolaires privés”, https://doi.org/10.3917/rfe.142.0143.

To capture these dynamics, PISA data can be used to compute a segregation (or dissimilarity) index at the school level, although the focus will be only on public schools. This index examines whether the distribution of native‑born students with two native‑born parents differs from that of foreign-born students or native‑born students with at least one foreign-born parent across schools, relative to what would be expected if students were randomly distributed. Specifically, the index reflects the proportion of students from these two groups who would need to be redistributed across schools to achieve an identical distribution.

The dissimilarity index D is formally calculated as follows:

$D= \frac{1}{2} \sum _k\left|\frac{G_k}{G} - \frac{N_k}{N}\right|$

where G represents the total number of racial/ethnic minority group members in all PISA schools and G_k is the number of racial/ethnic minority group members in PISA school k. Similarly, N represents the total number of majority group members in all PISA schools, and N_k is the number of majority group members in PISA school k. A dissimilarity index of zero indicates that both groups are equally distributed across all schools, while a value of 1 represents complete segregation.

The share of pupils either foreign-born or with at least one foreign-born parent who attend the quartile of schools with the highest concentrations of such pupils can serve as an alternative to the dissimilarity index. Using this measure highlights the significant clustering of pupils with foreign-born parentage in schools. Across the EU, over half (53%) of 15‑year‑old pupils with at least one foreign-born parent attend the quartile of schools with the highest shares of pupils with foreign-born parentage – a figure that would be 25% if these pupils were evenly distributed across schools (OECD/European Commission, 2023[6]).

It is important to stress that, as with any core indicator, this segregation index provides an imperfect measure of the impact of bias-driven racial/ethnic discrimination. Other factors, unrelated to discrimination, may also play a role. For example, residential racial/ethnic segregation could exclusively result from housing policies concentrating public housing in specific areas. Moreover, school segregation may be driven entirely by parental preferences for placing their children in environments with peers from privileged backgrounds, perceived as more likely to be high achievers, rather than by an aversion to have their children interact with peers of different racial or ethnic backgrounds. That said, research indicates that the benefits of exposure to high-achieving peers are primarily experienced by low-achieving students, while high achievers typically do not experience negative effects from being surrounded by lower-achieving peers (Fairlie et al., 2024[34]).

2.1.5. Health

Chapter 1 reveals that racism or racial/ethnic bias have a profound impact on the health of visible minorities. First, these factors have a well-established causal effect on mental health and are also associated with deteriorated physical health, at least in specific dimensions. This is largely due to the overactivation of stress pathways, known as allostatic load, which triggers three major pathologies detrimental to physical health: increased heart rate and blood pressure, raising the risk of cardiovascular diseases; elevated blood glucose levels and central fat accumulation, increasing the likelihood of diabetes; and systemic inflammation, which contributes to cancer risk. In other words, general health, mental health, and certain aspects of physical health are expected to be negatively impacted by bias-driven racial/ethnic discrimination.

These direct effects may be further exacerbated by indirect effects that operate through intermediate factors. The literature highlights two such drivers. First, individuals facing discrimination may adopt maladaptive coping strategies, such as eating disorders or substance abuse, which heighten the risk of health issues. Second, although field experiments on access to healthcare are limited and mostly conducted in the United States, findings suggest that visible minorities may face greater barriers in securing medical appointments, even when they have comparable health insurance to their majority counterparts, with tentative evidence that such barriers could extend to the quality of patient-provider interactions.

The EHIS and/or EU-SILC allow the derivation of indicators for both the aforementioned final and intermediate outcomes (see Box 2.10 and Box 2.13 for a discussion of recommended adjustments when using indicators derived from these surveys).

Box 2.13. Recommended adjustments when analysing indicators derived from EHIS

The EHIS yields core indicators covering health. For these indicators, differences between majority and minority individuals should, at a minimum, be adjusted for the gender, age, and highest educational attainment of the individual, as well as for household disposable income, and urban or rural residence, consistent with the approach detailed in the health section of Box 2.10.

Final outcomes

Final health outcomes can be divided into three categories: general health, mental health, and specific aspects of physical health.

Poor general health status

Both the EHIS and EU-SILC provide the means to compute the unexplained gap in the share of majority and minority individuals who rate their general health as bad or very bad.

As highlighted in Chapter 1, research demonstrates a positive health selection effect among immigrants, driven by the stylised fact that those who migrate are healthier on average than the rest of the population (Aldridge et al., 2018[35]; Shor and Roelfs, 2021[36]). However, the situation for children of immigrants presents a different picture. Analysis reveals significant disparities in mortality risks based on the geographic origin of immigrant parents (Wallace, Hiam and Aldridge, 2023[37]). While native‑born children of European immigrants generally mirror the mortality rates of their peers with native‑born parents, native‑born children of non-European immigrants face higher mortality risks throughout their lives.

Poor mental health status

The indicator related to mental health status builds on three sub-items: two from the EHIS and one from the EU-SILC.

From the EHIS, it is possible to compute the unexplained gap in the share of majority and minority individuals who report:

• Having suffered from depression in the past 12 months.

• Having experienced any of the following nine symptoms of poor mental health over the last two weeks: little interest or pleasure in doing things ; feeling down, depressed or hopeless ; trouble falling or staying asleep, or sleeping too much ; feeling tired or having little energy ; poor appetite or overeating ; feeling negative about yourself or that you are a failure or have let yourself or your family down ; trouble concentrating on things, such as reading the newspaper or watching television ; moving or speaking so slowly that other people could have noticed; or being so fidgety or restless that you have been moving around a lot more than usual.

With the EU-SILC, one can compute the unexplained gap in the share of majority and minority individuals who report having felt downhearted or depressed during the past four weeks.

Poor physical health status

The EHIS allows for a detailed analysis of three specific aspects of physical health, focusing on pathologies associated with:

• Increased heart rate and blood pressure, which are significant risk factors for cardiovascular diseases. This is captured through the unexplained gap in the share of majority and minority individuals who report having experienced any of the following conditions in the past 12 months: a myocardial infarction (heart attack) or chronic consequences of myocardial infarction; coronary heart disease or angina pectoris ; high blood pressure ; a stroke (cerebral haemorrhage, cerebral thrombosis) or chronic consequences of stroke.

• Elevated blood glucose levels and central fat accumulation, both of which increase the risk of diabetes. This is measured by the unexplained gap in the share of majority and minority individuals who report having had diabetes or high blood lipids in the past 12 months.

• Systemic inflammation, which can contribute to cancer risk. This is captured by the unexplained gap in the share of majority and minority individuals who report having suffered from chronic bronchitis, chronic obstructive pulmonary disease, or emphysema in the past 12 months.

Intermediate outcomes

Intermediate outcomes include maladaptive coping strategies, such as eating disorders or substance abuse, and unmet health needs, possibly due to discriminatory behaviour among healthcare professionals.

Maladaptive coping strategies

The EHIS provides a way to proxy the potentially higher prevalence of eating disorders among visible minorities. These disorders involve severe disturbances in eating behaviour and include conditions such as anorexia nervosa (characterised by extreme food restriction and an intense fear of gaining weight), bulimia nervosa (involving cycles of binge eating followed by purging to avoid weight gain), and binge eating disorder (marked by episodes of excessive eating without compensatory behaviours).

Specifically, the EHIS enables the computation of the unexplained gap in the share of majority and minority individuals who don’t have a normal weight based on their Body Mass Index (BMI). BMI can be derived from two EHIS questions: “How tall are you without shoes? (in cm)” and “How much do you weigh without clothes and shoes? (in kg)”.3 A normal weight is defined as a BMI between 18.5 and 24.9, indicating that the individual is neither underweight (BMI below 18.5), overweight (BMI between 25 and 29.9), nor obese (BMI over 30).

However, these BMI cutoffs must be applied with caution, as they were originally designed for White populations and may not accurately reflect health risks across different racial and ethnic groups. For instance, Black populations tend to have lower body fat at the same BMI, making them more likely to be misclassified as overweight, whereas the opposite is true for Asian populations, increasing the risk of misclassifying them as not overweight. Illustrating these differences, a US study found that for Black women, the threshold at which chronic illness risks increase is a BMI of 31‑33, compared to a lower cutoff of 25‑29 for White women (Stanford, Lee and Hur, 2019[38]). Similarly, an analysis of nearly 1.5 million individuals in the United Kingdom revealed that while a BMI of 30 or above was associated with a higher risk of diabetes for White individuals, the threshold for South Asian individuals was significantly lower, at 23.9 or above (Caleyachetty et al., 2021[39]).

In addition to BMI, the EHIS also allows for the derivation of a second indicator that analyses potential disparities in substance abuse. Specifically, this indicator comprises two sub-components that calculate the unexplained gap in the share of majority and minority individuals who report:

• Smoking tobacco products (excluding electronic cigarettes or similar electronic devices) on a daily basis, as opposed to occasionally or never.

• Consuming alcohol at a frequency of 3‑4 days a week or more, including “5‑6 days a week” and “every day or almost.” This contrasts with those who drink less frequently (e.g. “1‑2 days a week,” “2‑3 days in a month,” “once a month,” “less than once a month”), or who report abstaining (“not in the past 12 months, as I no longer drink alcohol” or “never, or only a few sips or tries in my whole life”).

Although abstaining from alcohol remains the safest level of consumption (Griswold et al., 2018[40]), research suggests that limiting alcohol intake to no more than three times per week may serve as a reasonable guideline for mitigating health risks. Studies show that, among alcohol consumers, those who drink 1 to 2 drinks per occasion, with an average frequency of 3.2 times per week, experience the lowest risk of alcohol-related harm (Hartz et al., 2018[41]). Exceeding this level may increase the likelihood of developing alcohol use disorder (AUD), as defined in the DSM‑5 by the American Psychiatric Association. According to the DSM‑5, AUD is characterised by a pattern of alcohol consumption leading to significant distress or impairment in key areas of life, including health, relationships, work, or legal issues. Key symptoms include drinking more than intended, an inability to cut down, and continuing to drink despite experiencing negative consequences.

It is worth emphasising that, compared to tobacco use, alcohol consumption poses greater challenges in assessing differences in maladaptive coping strategies between majority and visible minority populations. This is because, in certain religious and cultural traditions – such as those followed by individuals of Muslim background, who make up a significant segment of visible minorities in Europe – alcohol may be prohibited or strongly discouraged. As a result, unexplained differences in frequent alcohol consumption may reflect variations in cultural and religious norms rather than actual disparities in coping behaviours and should therefore be interpreted critically.

Unmet health needs

The EU-SILC provides valuable data for analysing unmet needs in medical and dental examination and treatment. Specifically, it allows for the derivation of the unexplained gap in the share of majority and minority individuals who, among those who reported needing medical or dental care at least once in the past 12 months, indicated not having received this care on at least one occasion.

2.1.6. Poverty risk and low life satisfaction

The barriers raised by bias-driven racial and ethnic discrimination in critical life areas such as education, school-to-work transition, employment, housing, and health are expected to significantly increase the risk of poverty for visible minorities and undermine their life satisfaction.

The EU-SILC provides data to capture these dimensions, allowing for the computation of two critical indicators:

• The unexplained gap in the share of majority and minority households at risk of poverty, defined as those living below the poverty threshold. According to Eurostat, the poverty threshold is set at 60% of the median equivalised disposable income in each country. Equivalised disposable income results from adjusting disposable income for household size by dividing total household income (including earning from labour and capital) by the square root of the number of household members. In the EU, 32% of children under 16 in immigrant households live in relative poverty, compared to 21% of children in native‑born households (OECD/European Commission, 2023[6]).

• The unexplained gap in the share of majority and minority individuals reporting low life satisfaction, based on the EU-SILC question: “On a scale of 0 to 10, where 0 is ‘not satisfied at all’ and 10 is ‘completely satisfied,’ how would you rate your overall satisfaction with life?” Eurostat uses a 20‑60‑20 distribution to categorise life satisfaction (Eurostat, 2017[42]): low satisfaction (scores from 0 to 5), medium satisfaction (scores from 6 to 8), and high satisfaction (scores of 9 and 10). This distribution ensures approximately 20% of respondents fall into the low and high satisfaction categories, respectively, with 60% in the medium satisfaction category.

2.2. Advanced indicators of bias-driven discrimination against visible minorities

This section introduces a set of advanced indicators, categorised into two types: “enhanced indicators”, which offer greater precision but require more robust data collection, and “exploratory indicators”, which provide the highest precision but are limited in availability due to their reliance on original data.

Advanced indicators of the enhanced type require more robust data collection than core indicators, as they involve calculating unexplained gaps between majority and minority populations across the entire population. For this to happen, it is necessary to collect information on both individuals’ and their parents’ country of birth, either through the census or national population registers, and to link these data with administrative databases in the key life areas covered in this report.

These indicators offer greater precision in measuring bias-driven discrimination compared to core indicators for several reasons.

First, administrative databases cover a broader range of outcomes compared to the limited set captured through surveys, with some being better suited to identify bias-driven racial/ethnic discrimination. Moreover, for outcomes common to both sources, administrative data are generally more reliable. This is particularly evident in income‑related measures, where misreporting – whether intentional or unintentional – is frequent in survey responses (Meyer, Mittag and Wu, 2024[43]). However, a key limitation of administrative data is their focus on objective measures, excluding perceptions, which are nonetheless crucial for a comprehensive understanding of individuals’ lived experiences. For example, relying solely on the share of majority and minority individuals diagnosed with depression provides only a partial view of mental health disparities. A more complete analysis emerges when this information is complemented by survey-based data capturing the share of individuals who report feeling depressed.

Second, advanced enhanced indicators are computable across the entire population, rather than by relying on cross-national survey samples of limited size. This comprehensive coverage allows for more granular distinctions within the minority population, such as differentiating native‑born individuals with European versus non-European foreign-born parentage. Furthermore, the removal of sample size constraints enables a full adjustment for key demographic factors, ensuring a more accurate analysis of disparities between majority and minority populations – an approach that is sometimes hindered or even rendered impossible by the smaller sample sizes of cross-national survey.

Third, administrative data are collected at a high frequency, with updates ranging from annual (e.g. education enrolment data) to monthly or quarterly (e.g. employment or health records), and even continuously in certain cases, such as births and deaths. This high-frequency data collection enables advanced enhanced indicators to support near real-time monitoring of disparities between majority and minority populations, offering a significant advantage over survey-based indicators, which are updated at best annually, and often less frequently.

Identifying EU countries that meet the aforementioned data requirements is beyond the scope of this chapter. However, preliminary findings from a 2023 OECD questionnaire on data collection for visible minorities, supplemented by desk research for non-OECD EU countries, offer initial insights. This research indicates that slightly more than half of EU countries (15) collect data on individuals’ and their parents’ country of birth in their census and/or national population registers.4 In other words, in at most 15 EU countries, it is theoretically possible to link this information to topic-specific administrative databases, such as those related to education, employment or health. To indicate that advanced indicators of the enhanced type are available in, at most, half as many countries as core indicators, but offer greater precision than core indicators, these indicators are labelled with “+/++” in the availability columns (compared to “+++” for core indicators) and “++” in the precision columns (compared to “+” for core indicators) in Annex Table 2.A.1 through Annex Table 2.A.6.

Concerning advanced indicators of the exploratory type, they require the collection of original data, primarily through field experiments, with strong potential for identifying and measuring bias-driven discrimination, and, in a more targeted set of cases, the use of artificial intelligence.

Experimental methods include correspondence studies, which entail sending fictitious applications or inquiries to detect differential treatment based on characteristics such as perceived race or ethnicity, typically signalled through the fictitious applicants’ first and last names. While commonly applied in the labour market, correspondence studies can also be used in other fields, such as housing or health. Experimental methods also include audit studies, where real individuals pose as, for example, job candidates in interviews to observe biased treatment in settings that cannot be easily explored through correspondence studies.

However, even field experiments, which yet have broader applications for capturing bias-driven discrimination compared to AI, cannot generate data in all settings. For example, they are not feasible for assessing bias in schools – where discrimination may stem from educators or peers – or in workplaces when decisions are made about promotions. To indicate their high precision but limited availability due to reliance on original data collection (or even the outright impossibility of computing them in certain settings), advanced indicators of the exploratory type are marked with “+” in the availability column and “+++” in the precision column in Annex Table 2.A.1 through Annex Table 2.A.6.

2.2.1. Education

For final educational outcomes (educational attainment and achievement), the ability to link administrative databases, including those related to education, with register-based or census data identifying individuals’ and their parents’ country of birth offers a powerful tool for analysing disparities. This linkage facilitates the examination of the same outcomes as those targeted by core indicators, but with significantly larger sample sizes and enhanced capacity to control for confounding factors, thereby achieving greater precision. Additionally, these data linkages can enrich the scope of outcomes analysed. For instance, in terms of educational achievement, unexplained gaps between majority and minority students could be calculated for each national standardised test conducted throughout the school curriculum, providing deeper insights into disparities across educational stages.

Likewise, advanced indicators focused on intermediate educational outcomes could not only enhance the measurement of existing outcomes covered by core indicators but also broaden the scope to new metrics. Specifically, to effectively capture teacher bias – and following the research methodology detailed in Chapter 1 to identify discriminatory behaviours among educators – countries with robust data collection systems might consider the following approach: comparing grading disparities between majority and minority students across two types of assessments in a specific subject. The first assessment type is national standardised examinations, which are blind-graded and thus less susceptible to teacher bias; the second is school-based, non-blindly graded tests, which may be more prone to such bias (Zanga and De Gioannis, 2023[44]). A lack of grading disparities between these two assessment types may indicate an absence of teacher bias, whereas grading disparities could suggest the presence of such bias. Yet, while this indicator potentially offers a more precise proxy for teacher bias than the core measure based on perceived teacher hostility, caution is warranted when interpreting grading disparities across blind- and non-blind assessments. Majority and minority students may react differently to each assessment type; for instance, one group might demonstrate higher (or lower) effort relative to the other under one approach, potentially creating score differentials across evaluation methods that do not necessarily reflect teacher bias.

To address the final mechanism discussed in Chapter 1 – namely bias in textbooks, which is not covered by core indicators due to lack of suitable data – an advanced indicator of the exploratory type could be developed. This indicator would leverage artificial intelligence to identify the quantitative and qualitative underrepresentation of visible minorities in children’s literature and educational materials, including history curricula and textbooks, following previous research highlighted in Chapter 1 and further summarised in Box 2.14.

Box 2.14. Measuring the quantitative and qualitative underrepresentation of visible minorities using artificial intelligence

Recently, artificial intelligence has been utilised to identify the underrepresentation of visible minorities in textbooks, from both a quantitative and a qualitative perspective.

Quantitative underrepresentation of visible minorities

To measure quantitative underrepresentation in images, (Adukia, 2023[45]) leverage Google AutoML Vision’s face detection technology, which predicts a face’s gender, age, and racial or ethnic background. Recognising that children’s literature features numerous illustrations, the authors trained a custom face detection model to identify faces in both illustrations and photographs, as most existing models are optimised for photos alone. Once a face is detected, the model isolates the pixels representing skin tone, applying a machine learning algorithm to cluster all face pixels and calculate an average “representative skin colour” per face. This skin colour is then classified on a scale from 0 (lighter) to 100 (darker), providing a proxy for visible minority representation based on skin tone.

To assess quantitative underrepresentation in text, the authors utilise Google Vision Optical Character Recognition (OCR) to identify names and famous characters that can be associated with specific racial or ethnic groups.

Qualitative underrepresentation of visible minorities

While (Adukia, 2023[45]) do not explore the qualitative underrepresentation of visible minorities, this aspect is examined in other studies, such as that of (Lucy et al., 2020[46]), who analyse the portrayal of marginalised groups in 15 history textbooks widely used in Texas between 2015 and 2017. Using natural language processing (NLP), the authors extract verbs and adjectives associated with white versus non-white groups. They then rank these descriptors based on connotations of power/dominance (strong vs. weak), sentiment/valence (positive vs. negative), and agency/arousal (active vs. passive) to identify potential differences in portrayal across groups. Their findings indicate that non-white groups are more frequently described with terms scoring lower on the power/dominance and agency/arousal dimensions than those associated with white groups.

2.2.2. School-to-work transition

As for education, linking administrative databases, including those that track individuals after they leave formal education, with register-based or census data identifying individuals’ and their parents’ country of birth could enhance the analysis of outcomes covered by core indicators regarding school-to-work transition. Specifically, this linkage would enable the calculation of unexplained differences in NEET rates between majority and minority groups over a larger sample.

Beyond refining this final outcome indicator, robust data collection would also allow for a more precise measurement of unexplained disparities in an intermediate outcome, namely exposure to disciplinary actions within the school setting. For example, after adjusting for key confounding factors such as academic achievement and, where possible, the severity of student misbehaviour, this approach would facilitate comparing the share of majority and minority students experiencing in-school and out-of-school suspensions at different stages of the curriculum.

Advanced indicators of the enhanced type are not limited to providing more accurate measurements of outcomes already targeted by core indicators; they also include metrics that reveal an additional pathway identified in Chapter 1, namely exposure to disciplinary actions outside the school setting, including via interactions with law enforcement. Although research on bias-driven policing remains limited in Europe, (Søndergaard and Hussein, 2022[47]) illustrate how robust data collection can help estimate this bias by calculating the unexplained gap between majority and minority individuals in arrest rates without subsequent, withs in Denmark (see Box 2.15 for further information). If the police are more likely to arrest racial or ethnic minorities for offenses that do not lead to convictions, this may indicate bias in policing practices, as a lower threshold is applied to arrests involving minorities. However, caution is again warranted when interpreting gaps, as other factors may be involved – such as potentially higher reporting rates of racial/ethnic minorities by the general population, though this channel may itself be related to bias.

Finally, an advanced exploratory indicator can be developed to better capture discrimination in access to work-based learning (WBL) opportunities. Building on prior research (Kaas and Manger, 2012[48]; Auer et al., 2022[49]), this indicator would employ correspondence studies to compare the invitation rates to a job interview for fictitious job candidates from majority and minority groups with similar applications. For a comprehensive overview of barriers to WBL, these correspondence studies could encompass three key stages: school-mediated WBL in General Education Secondary Programs, including internships and work placements designed to introduce students to the world of work; WBL in Vocational Upper Secondary Programs, encompassing internships and apprenticeships as part of Vocational Education and Training (VET) curricula; and WBL within Higher Education, covering placements that are either formal or informal components of undergraduate or graduate programmes. To capture the full extent of potential discrimination, these correspondence studies should ideally be complemented by audit studies, where trained actors represent fictional applicants in real job interviews, thereby capturing discrimination that may arise at this final stage. Examining this stage would also allow for the collection of additional valuable information, such as whether minority candidates are more frequently offered unpaid placements rather than paid ones.

Importantly, the CVs and cover letters of fictitious candidates in correspondence and audit studies must signal strong employability and productivity to ensure that any disparities in response rates or job offers are not driven by employers’ risk assessment – commonly known as “statistical discrimination”. This practice, based on the assumption that visible minorities tend to come from lower socio‑economic backgrounds, could otherwise obscure the true extent of bias in hiring decision.

Box 2.15. Unexplained gaps in police arrest rates without subsequent convictions

Researchers from the Danish Institute for Human Rights, (Søndergaard and Hussein, 2022[47]), have linked extensive datasets from the public statistics office, including records on arrests and convictions, the population register, and education and income registers. This linkage has enabled them to compile a dataset of over 2.5 million preliminary charges, arrests, and court rulings from 2009‑19, with rich demographic and socio-economic information, including sex/gender, age, country of origin, family and housing conditions, as well as education, employment, and income data for individuals and their parents.

Leveraging this dataset, the authors calculate and compare the likelihood of arrest (without conviction) for descendants of immigrants and of native‑born individuals, controlling for key socio-economic variables. Their findings reveal that, even when these characteristics are held constant, descendants of immigrants face a 46% higher probability of being arrested without conviction compared to individuals of Danish descent. This disparity is particularly pronounced for specific groups, such as descendants of Syrian or Lebanese origin.

2.2.3. Employment

Linking administrative databases, including employment records, with census data or national population registers that identify individuals’ and their parents’ country of birth would allow generating a range of enhanced advanced indicators. These indicators enhance core measures by enabling a deeper analysis of disparities within visible minorities, leveraging population-wide data to provide greater granularity and uncover variations that might otherwise remain hidden. For example, an analysis of overqualification rates using Sweden’s total population register data not only shows that native‑born children of immigrants face up to 19% higher probabilities of overqualification compared to the majority population, but also reveals striking differences by origin. Disparities are particularly pronounced among individuals of Iranian, Middle Eastern and North African, and Other Non-Western backgrounds, with overqualification probabilities reaching up to 39% higher than those of the majority population (Kim, 2024[50]).

Furthermore, advanced indicators of the enhanced type allow for expanding the analysis to include additional dimensions. For instance, rather than focusing solely on economy-wide labour earnings gaps, robust data collection could enable the calculation of these gaps at the firm level – a critical step for effectively monitoring employers’ career management behaviour and reducing the risk of discriminatory practices in wage‑setting and promotion decisions arising in the first place. Currently, few countries or jurisdictions require such reporting from employers. In Canada, federally regulated employers must publish pay gaps across racial and ethnic groups, while recent progress has been made in the United States at the state level. For example, in July 2024, Massachusetts enacted legislation requiring employers with 100 or more employees to submit annual pay data reports disaggregated by race, ethnicity, gender, and job category.

Having this information calculated directly by a public body would represent a substantial step forward, by accelerating the adoption of pay transparency policies for visible minorities. This approach is especially timely given that over half of OECD countries already mandate gender wage gap reporting for private employers (OECD, 2024[51]) – a number set to rise in the EU following the European Parliament’s approval of the new EU Pay Transparency Directive in 2023, which member states must implement into national legislation within three years.5 Assigning a public body to calculate firm-level wage gaps between majority and minority employees would not only enhance employer buy-in by reducing administrative burdens but also ensure that the figures produced are accurate and comparable across firms. Evidence suggests that, in the absence of an independent body, employers tend to report information that underestimates wage gaps (Institut des Politiques Publiques, 2023[52]), thereby weakening incentives to address disparities.

In addition to these enhanced indicators, two advanced exploratory indicators could provide valuable insights into discrimination at the recruitment stage. The first approach would employ correspondence studies, ideally complemented by audit studies, to measure discrimination throughout the recruitment process, but also to determine whether, conditional on receiving a job offer, minority candidates are proposed a lower starting wage. Again, these field experiments should be implemented within a framework designed to minimise employer uncertainties about candidates’ employability and productivity, ensuring that observed disparities reflect bias-driven rather than statistical discrimination.

The second advanced exploratory indicator would leverage artificial intelligence tools to explore, following (Hangartner, Kopp and Siegenthaler, 2021[53]) outlined in Chapter 1, how employers navigate job portals such as public employment service websites. This would include analysing, based on millions of observations, how employers sort profiles and make decisions about whom to contact or shortlist, thereby assessing potential discriminatory behaviour against minority candidates – identified through their first and last names as well as the language(s) they speak – while controlling for a broad range of characteristics.

While correspondence and audit studies in the labour market typically focus on specific industries or occupations, recent research offers the potential to estimate hiring discrimination at the firm level (see Box 2.16 for further information). This advancement enables the use of field experiments to expand the monitoring of employer practices beyond career management, as discussed in previous paragraphs, to also encompass recruitment behaviours.

Box 2.16. Estimating firm-level hiring discrimination through correspondence studies

Researchers from the University of California (Berkeley) and the University of Chicago recently analysed callback rates from over 83 000 fictitious job applications, varied by gender and race/ethnicity, sent in response to 11 000 job postings at 108 Fortune 500 firms (Kline, Rose and Walters, 2022[54]). To estimate hiring discrimination at the firm level, the authors developed a methodology that ranks discriminatory behaviour in organisations by balancing the magnitude of callback gaps with the noisiness of these estimates. Their findings revealed that distinctly white names received higher contact rates than distinctly Black names, with the most discriminatory firms favouring white applicants over Black applicants by 24%, while the least discriminatory firms showed a 3% preference for white applicants (Kline, Rose and Walters, 2024[55]).

The grading system is designed to be adjustable, allowing for more or less strict classification. A more stringent setting assigns each firm a unique grade based on highly specific data, though it can increase the risk of false positives – incorrectly identifying a firm as discriminatory or overestimating the level of bias. This approach is advisable when sample sizes are large or the data quality is particularly robust. In contrast, a more moderate setting groups firms into broader categories, reducing specificity in favour of greater confidence in the results by minimising the risk of misinterpretation of contact rate data.

2.2.4. Housing

In housing, advanced enhanced indicators enable more precise measurement of outcomes targeted by core indicators. Additionally, by leveraging national population registers containing information on individuals’ place of residence, country of birth, and parental country of birth, it becomes possible to calculate a racial/ethnic segregation index at finely defined geographic levels. Two approaches could be considered to distinguish between majority and minority groups. As a first step, the dissimilarity index could classify the population into native‑born individuals with two native‑born parents on one hand, and foreign-born individuals or native‑born individuals with at least one foreign-born parent on the other hand. This approach mirrors the dissimilarity index computed at the school level in the core indicators section. In a second step, to better capture visible minorities, the population could be divided into the following two populations : (i) individuals of European descent, including native‑born individuals with two native‑born parents, native‑born individuals with two foreign-born parents of European background, or with one foreign-born parent of European background (in which case the other parent should be native‑born), and foreign-born individuals of European background; (ii) individuals of non-European descent, including foreign-born individuals of non-European background and native‑born individuals with one or two foreign-born parents of non-European background.

Additionally, two advanced exploratory indicators could be developed to illuminate discriminatory mechanisms that contribute to disparities in housing tenure, value for money in housing, and racial/ethnic segregation. Both indicators would rely on correspondence studies, potentially combined with audit studies, following the approach outlined in Chapter 1.

The first set of correspondence studies would focus on identifying discrimination in the private sales and rental housing markets. Extending these field experiments with audit studies would enable verification of earlier findings that minority applicants are often offered higher prices or rents for identical properties. The second set of correspondence studies would examine discrimination in access to mortgage loans, a key factor in the lower homeownership rates observed among visible minorities. Incorporating an audit study here would allow assessment of whether, all else being equal, visible minorities are indeed offered loans with higher costs, as preliminary evidence in Chapter 1 suggests. However, implementing an audit study may not always be feasible. Audit studies are inherently more complex than correspondence studies, particularly when examining mortgage access, as loan applications require the submission of detailed, publicly verifiable financial information, such as credit scores.

When conducting these field experiments, it is critical to ensure that the profiles of fictitious applicants portray equally desirable homebuyers, renters, or loan applicants, with clear and reliable indicators of financial stability. This approach minimises the risk of detecting statistical discrimination rather than bias-driven discrimination.

2.2.5. Health

Linking administrative databases, including health records, with census data or national population registers that identify individuals’ and their parents’ country of birth would enable the creation of a range of advanced enhanced indicators. These indicators would focus on final outcomes already targeted by core indicators, particularly unexplained disparities in diagnoses of mental health issues and physical conditions related to cardiovascular diseases, diabetes, or systemic inflammation. This robust data linkage would also expand the set of monitored outcomes, allowing for the calculation of unexplained gaps in mortality rates, following (Wallace, Hiam and Aldridge, 2023[56]) outlined in Chapter 1. Additionally, it would enable more precise estimation of certain intermediate outcomes, such as unexplained gaps in overweight or underweight status, using Body Mass Index as a proxy for eating disorders.

Preliminary evidence from the United States indicates that visible minorities encounter greater challenges in securing medical appointments, even when holding health insurance comparable to that of their majority counterparts. These barriers may also extend to the quality of patient-provider interactions as suggested by initial research, underscoring the importance of developing an advanced exploratory indicator based on field experiments.

In countries where appointment requests are commonly submitted via email, correspondence studies could be conducted to test for potential discriminatory treatment against minority patients, following (Fumarco et al., 2024[57]). To ensure that observed discrimination reflects bias rather than statistical discrimination, it is essential to make minority and majority profiles equally desirable by emphasising the financial solvency of minority candidates. While extending these field experiments to in-person audit studies presents challenges – given that simulating health conditions may be difficult in face‑to-face interactions – online settings provide a feasible alternative, especially as telemedicine becomes more widespread. This approach would allow researchers to assess whether fictitious patients with similar self-reported symptoms receive differential treatment based on race or ethnicity.

In countries where online platforms, such as Doctolib in France, Germany, and Italy, handle appointment scheduling by directly managing practitioners’ agendas without the need for patient interaction with the provider, the scope for discrimination during appointment requests may be limited. Consequently, conducting correspondence studies may hold little value. Nevertheless, online audit studies could still provide valuable insights by determining whether discrimination may instead arise during the consultation itself, such as through reduced attention or quality of care provided to fictitious minority patients.

2.2.6. Poverty risk and low life satisfaction

There is limited scope for improving the measurement of low life satisfaction from EU-SILC, unless countries implement larger-scale, regularly conducted, nationally representative surveys addressing well-being. In contrast, linking administrative databases, including income records, with census data or national population registers that identify individuals’ and their parents’ country of birth would enable a more precise calculation of unexplained disparities in poverty risk between majority and minority households.

2.3. Additional proxies for racism, racial/ethnic bias, and bias-driven discrimination

This section outlines additional attitudinal and perception-based measures that could serve as proxies for racism, racial/ethnic bias, and bias-driven discrimination, beyond the five life areas examined.

While racism or racial/ethnic bias do not directly measure bias-driven discrimination against visible minorities, not least because individuals may successfully suppress them when acting, monitoring these dimensions can still provide valuable insights that complement the core and advanced indicators presented in the previous sections. Yet, measuring racism and racial/ethnic bias based on self-reported attitudes presents significant challenges, as stressed in Chapter 1. First, self-reported data capture only bias of which individuals are consciously aware, while bias can be unconscious. Second, even the measurement of conscious bias is difficult, as individuals are generally reluctant to disclose socially unacceptable views, a phenomenon commonly known as social desirability bias.

Furthermore, adopting the perspectives of witnesses or potential victims to capture bias-driven discrimination may not necessarily yield an accurate assessment either. General perceptions and self-reported experiences of discrimination are heavily influenced by the population’s level of awareness regarding discrimination. Consequently, these perceptions may not accurately reflect the actual prevalence of discriminatory behaviour.

With these limitations in mind, three cross-country surveys offer additional options for measuring racism, racial/ethnic bias and bias-driven discrimination: the European Values Survey (EVS), the European Social Survey (ESS), and the Eurobarometer survey on “Discrimination in the European Union” (see Box 2.17 for a summary of their frequency, sampling methods, sample sizes, and suitability for distinguishing between majority and visible minority populations).

Specifically, these surveys allow for the measurement of two types of outcomes. First, attitudes of the majority population towards visible minorities. Second, perceptions and experiences of discrimination, which can be further disaggregated into: (i) public perceptions of the prevalence of discrimination based on race and ethnicity, and (ii) the (unexplained) gaps in experiences of discrimination between majority and visible minority populations.

The last item belongs to a broader category of social integration variables, all of which are likely to be negatively affected by experiences of bias-driven racial or ethnic discrimination. These variables encompass unexplained gaps in interpersonal trust, trust in public authorities (including the police), and levels of civic engagement –such as voting in elections, participating in political actions like signing petitions, or joining voluntary organisations. They also extend to adherence to core values of the host country, including democratic principles and gender equality (see (OECD, 2024[58]) for a discussion on measuring social integration dimensions). Many of these dimensions are captured, in one form or another, in the EVS and ESS. Countries could enhance their monitoring and assessment frameworks by incorporating these additional measures, offering deeper insights into the impact of discrimination on social cohesion.

Caution should be exercised not only when interpreting the measures proposed in this section at a specific point in time but also when analysing trends. For example, an apparent improvement in self-reported attitudes of the majority population towards visible minorities may reflect an actual decrease in racism and/or racial/ethnic bias but could also indicate an increase in social desirability, where individuals are less likely to admit to holding prejudiced views. Likewise, a decrease in the unexplained gaps in self-reported experiences of discrimination between majority and visible minority populations could signal a genuine decline in discriminatory behaviour based on race and ethnicity. However, it may also stem from a decrease in minorities’ awareness of discrimination or their ability to identify it. For instance, this could occur if efforts to normalise certain discriminatory behaviours result in lower recognition of such incidents by minority groups.

Box 2.17. The European Values Survey (EVS), the European Social Survey (ESS), and the Eurobarometer survey on “Discrimination in the European Union”

Three cross-country surveys covering EU countries provide options for measuring racism, racial/ethnic bias and bias-driven discrimination: (i) the European Values Survey (EVS), initiated in 1981 and conducted every 9 to 10 years, with five rounds to date (1981, 1990, 1999, 2008, and 2017) and the next wave scheduled for 2026‑28; (ii) the European Social Survey (ESS), launched in 2002 and conducted biennially, with the eleventh round completed in 2023 (round 12 being planned for between 2025 and 2026); and (iii) the Eurobarometer survey on “Discrimination in the European Union” (“Eurobarometer” henceforth) conducted in 2012, 2015, 2019, and 2023 (with the next round not expected before 2027).

All three surveys utilise probability sampling techniques to ensure representativeness and gather key demographic information, including gender, age, and education (with educational attainment collected in the EVS and ESS, and the number of years of education in the Eurobarometer). Furthermore, the minimum sample size generally ranges from 1 000 to 1 500 respondents, except in smaller countries where sample sizes may be lower.

The EVS and the ESS include questions regarding the country of birth of respondents and their parents, allowing for the distinction between the majority population, defined as native‑born individuals with two native‑born parents, and the minority population, defined as native‑born individuals with at least one foreign-born parent. There is also the potential to further differentiate between individuals with European and non-European foreign-born parentage. However, in some countries, this decomposition may not be feasible due to the low proportion of the latter group in the total population, compounded by the small sample sizes of the EVS and ESS.

In the Eurobarometer, it is not possible to distinguish between the majority and minority populations based on country of birth, as this information is not collected. However, a distinction can be made using a question that asks respondents whether they consider themselves part of a minority, including categories such as “a minority in terms of skin colour” and “an ethnic minority”. In this context, the majority is defined as those who do not select these categories, while the minority comprises individuals who do. Across the EU, however, the share of people who self-identify as belonging to either of these two minority groups rarely exceeds 5%, amounting to 50 observations in the Eurobarometer. As a result, in some countries, analyses may be based on very small sample sizes, which could preclude meaningful analysis altogether or, at the very least, require considerable caution in interpreting the results.

The measures proposed in this section, which are based on cross-national surveys, could be complemented by proxies drawn from nationally representative, country-specific surveys focused on perceptions of discrimination. An example is the 2021 online survey on “Racism and Ethno-racial Discrimination” conducted by the Luxembourg Institute of Socio-Economic Research (LISER) in collaboration with the Ministry of Family Affairs, Integration, and the Greater Region (Docquier et al., 2022[59]). Yet, although such surveys often involve larger samples than the cross-national surveys reviewed here – enabling enhanced measures through greater scope to disaggregate the analysis by different minority groups and adjust for confounding factors – they face similar limitations to cross-national surveys regarding the interpretability of perception-based measures of racism, racial/ethnic bias and bias-driven discrimination.

To address these limitations, one approach could be to again rely on field experiments, particularly those designed to detect differential treatment in social interactions. Such experiments are often considered a barometer of racism, racial/ethnic bias, and bias-driven discrimination within society (Crosby, Bromley and Saxe, 1980[60]; Saucier, Miller and Doucet, 2005[61]). Specifically, studies on helping behaviour offer a powerful means of examining the persistence of prejudice in everyday social settings. For example, a recent experiment in France readily replicable in other national contexts involved testers asking white pedestrians for directions at busy traffic lights (Aranguren, 2024[62]). The findings indicate that Asian and Black testers received assistance less frequently than their white counterparts, with this differential treatment becoming more pronounced when the perceived cost of helping was higher (see Box 2.18 for further information).

Box 2.18. Differential treatment in helping behaviour based on race and ethnicity, as observed in France

Sociologist Martin Aranguren conducted a field experiment in the streets of Paris from July to September 2021, at four different pedestrian crossings along busy roads. In this experiment, a pedestrian waiting for the light to turn green is approached by a trained tester – either Asian, Black, or white – who initiates the interaction with a polite “Sorry to bother you” intending to ask for directions. The pedestrian’s reaction unfolds across several possible stages: first, the pedestrian may refuse the interaction, signalling “back off” through body language, potentially influenced by the tester’s perceived race/ethnicity; or they may accept the interaction. If the interaction proceeds, the tester then requests directions, such as “Hello, I’m looking for XXX street”. At this point, the pedestrian may either respond directly or hesitate, possibly due to unfamiliarity with the location. If the pedestrian is unfamiliar with the street and is informed by the tester that they cannot search for it independently due to a phone issue, the pedestrian faces a decision: to look up the information on their own phone or not – a choice that may again be influenced by the tester’s perceived race/ethnicity.

After observing 4 500 such interactions, Aranguren found that white pedestrians were equally likely to initiate the interaction with Asian and Black testers as they were with white testers. Additionally, if the interaction was initiated, they were no less likely to give directions immediately. However, when the cost of helping increased – specifically, when the pedestrian did not know the street and needed to use their own phone – significant differential treatment emerged, with the likelihood of help declining as the pedestrian’s age increased. Specifically, among 10 white pedestrians who check their phones for the white tester, the number who refuse to do so when the tester is Black is 2 among younger pedestrians, 4 among middle‑aged pedestrians, and 7 among older pedestrians. For Asian testers, the corresponding figures are 1, 1.5, and 3, respectively.

Source: (Aranguren, 2024[62]), “Racial discrimination in helping situations depends on the cost of help: A large field experiment in the streets of Paris”, https://doi.org/10.1111/1468-4446.13156.

2.3.1. Attitudes of the majority population towards visible minorities

All three surveys – the EVS, the ESS, and the Eurobarometer – provide proxies for assessing majority attitudes towards visible minorities.

• In the EVS, this can be assessed by calculating the proportion of the majority population who selects “People of a different race” (and optionally, “Immigrants/foreign workers”) in response to the question: “On this list are various groups of people. Could you identify any that you would not like to have as neighbours?”.

• In the ESS, majority attitudes towards visible minorities can be proxied by comparing the share of the majority population selecting “Allow none to come” in response to the question: “To what extent should [country] allow people of a different race or ethnic group from most [country]’s people?”, with the share selecting the same response when asked: “To what extent should [country] allow people of the same race or ethnic group as most [country]’s people to come and live here?”. Calculating this difference helps to identify aversion specifically towards immigrants of different racial or ethnic backgrounds, after accounting for general aversion to migration.

• In the Eurobarometer, three proxies can be used to measure discomfort levels among the majority population with individuals from visible minority populations.

  • The first proxy is based on the question: “Using a scale from 1 to 10, how comfortable would you feel about having a person from each of the following groups in the highest elected political position in [our country]? ‘1’ means ‘not at all comfortable,’ and ‘10’ means ‘totally comfortable’.” This proxy yields the share of the majority population selecting “1” to “4” (thereby expressing discomfort) when referring to “a person with a different skin colour than the majority” or when referring to “a person from a different ethnic origin than the majority”. In 2023, 15% of EU respondents expressed discomfort with the latter hypothetical person, and 12% with the former.

  • The second proxy involves the question: “Regardless of whether you are actually working or not, how comfortable would you feel if a colleague at work, with whom you are in daily contact, belonged to each of the following groups?” This proxy compares the share of the majority population selecting “1” to “4” for “an Asian person” or “a Black person” versus “a white person”. In 2023, 8% of EU respondents reported feeling uncomfortable with an Asian or Black colleague, compared to 3% who expressed discomfort with a white colleague.

  • The third proxy is derived from the question: “Regardless of whether you have children or not, how comfortable would you feel if one of your children was in a relationship with someone from the following groups?” This last proxy consists in computing the share of the majority population selecting “1” to “4” for “an Asian person” or “a Black person” and in comparing it to the share giving a similar response for “a white person”. In 2023, 13% and 15% of EU respondents reported feeling uncomfortable with an Asian or Black son- or daughter-in-law, respectively, compared to 3% who expressed discomfort with a white son- or daughter-in-law.

Besides these visible minority populations, it is worth noting that the Eurobarometer enables analysis of majority attitudes towards a long-established visible minority group, namely the Roma people. Specifically, the Eurobarometer includes questions that assess the level of discomfort among the majority population with a Roma individual holding the highest elected position, being a colleague at work, or becoming a son- or daughter-in-law. Levels of discomfort towards Roma people remain particularly high. In 2023, 26% of EU respondents reported feeling uncomfortable with a Roma person in the highest elected political position, 16% with a Roma colleague, and 29% with a Roma son- or daughter-in-law.

2.3.2. Perceptions and experiences of discrimination

Measures of perceptions and experiences of discrimination consist of two categories of indicators: (i) public perceptions of the prevalence of discrimination based on race and ethnicity, and (ii) the (unexplained) gaps in experiences of discrimination between majority and visible minority populations.

Public perceptions of the prevalence of discrimination based on race and ethnicity

Public perceptions of the prevalence of racial/ethnic discrimination can serve as an indicator of the extent to which racism, racial/ethnic bias or bias-driven discrimination persist in society. The Eurobarometer includes a question designed to capture such perceptions: “For each of the following types of discrimination, could you please tell me whether, in your opinion, it is very widespread, fairly widespread, fairly rare, or very rare in [our country]? By discrimination, we mean when someone is treated unfavourably compared to others based on arbitrary criteria”.

The proposed indicators would calculate the share of respondents who perceive discrimination based on skin colour or ethnic origin as widespread in their country. The Eurobarometer surveys on “Discrimination in the European Union” from 2012, 2015, 2019, and 2023 consistently show that racial/ethnic discrimination is viewed as the most prevalent form of discrimination within the EU. In 2023, around 60% of respondents perceived discrimination based on skin colour and/or ethnic origin as frequent – an increase of 4 percentage points since 2012.

As highlighted above, the Eurobarometer provides an opportunity to closely examine the situation of Roma people by including a question on whether discrimination “on the basis of being Roma” is perceived as widespread by the general public. This focus on Roma people within racial and ethnic minorities reveals the highest level of perceived discrimination, with 65% of EU respondents considering it to be widespread in 2023.

(Unexplained) gaps in experiences of discrimination between majority and visible minority populations

Both the ESS and the Eurobarometer facilitate the measurement of (unexplained) gaps in experiences of discrimination between majority and visible minority populations. It is important to measure, to the extent possible, the difference between these two populations, rather than focusing solely on the share of visible minority individuals reporting discrimination. This distinction is necessary when the grounds based on which discrimination is experienced are not specified in the survey, because even members of the majority population may experience discrimination, albeit more likely on different grounds such as gender, age, sexual orientation, or disability. Where sample size allows, it is also important to account for differences in key demographic factors (see Box 2.17) across these populations that may influence both their experiences of discrimination and/or their perception of it.

• Specifically, two indicators can be derived from the ESS.

  • The question “Would you describe yourself as being a member of a group that is discriminated against in this country?” enables the calculation of differences in the likelihood of responding “yes” between majority and minority respondents, possibly after controlling for variables such as gender, age, and education.

  • This measure can be further complemented by assessing the share of minority respondents who report that their group is discriminated against specifically based on “colour or race”, “ethnic group”, or “nationality,” using the following question: “On what grounds is your group discriminated against”. In the EU, more than one in five native‑born young people with foreign-born parents feel part of a group that faces discrimination based on ethnicity, nationality, or race – a higher share than among the foreign-born population (15%) (OECD/European Commission, 2023[6]). This heightened perception of discrimination among second-generation EU citizens may stem from greater expectations of fair and equal treatment as well as a deeper awareness of discrimination processes (OECD, 2024[58]).

• Similarly, the Eurobarometer offers up to two measures for analysing disparities in experiences of discrimination between majority and visible minority populations. However, in some countries, small sample sizes may limit the feasibility of computing these gaps.

  • The first measure flows from the following question: “In the past 12 months, have you personally felt discriminated against or experienced harassment on one or more of the following reasons?”. This allows for the computation of differences in the probability of having personally felt discriminated against or harassed in the past 12 months between majority and minority respondents, if possible controlling for gender, age, and education.

  • In countries where the number of respondents identifying as a minority in terms of skin colour or ethnicity is sufficient, further analysis can be conducted using the question: “Thinking about the most recent time when you felt discriminated against or experienced harassment, under what circumstances did it take place?” This question provides insights into the unexplained gaps in experiences of discrimination among those who reported such experiences in the first place, focusing on key life areas studied in this report. These areas include education (“when attending or applying to school or university”), the labour market (“when looking for a job” and “at work”), housing (“when seeking to rent or purchase accommodation”), and health services (“when using or requiring healthcare services”). The question also enables the inclusion of additional essential life dimensions in the analysis, such as day-to-day public interactions, capturing instances of discrimination occurring in settings like “a public space”, “a café, restaurant, bar or nightclub”, or “a shop or a bank”. It also allows for the analysis of discrimination “when requesting or using social services” or “online”.

In addition to the ESS and the Eurobarometer, it is important to highlight the efforts of the Fundamental Rights Agency (FRA) in conducting cross-national surveys to capture perceptions and experiences of discrimination among both recently arrived and long-established visible minorities, using, in most cases, multistage random probability sampling to ensure representativeness (see Box 2.19 for an overview). However, because these surveys do not simultaneously include representative samples of majority populations, they do not allow for the analysis of unexplained disparities between majority and minority populations.

Box 2.19. FRA’s cross-national surveys among visible minorities

The FRA has conducted three types of cross-national surveys measuring perceptions and experiences of discrimination among visible minorities living in the EU: those that cover both recently arrived and long-established visible minorities, and those that focus on only one of these groups.

Cross-national surveys focusing on both recently arrived and long-established visible minorities

The European Union Minorities and Discrimination Surveys (EU-MIDIS I and II), conducted in 2008 and 2015‑16, respectively, gathered data from nearly 25 000 respondents across all EU countries, including the United Kingdom, which was an EU member at the time. EU-MIDIS I focused on visible minority immigrants – primarily those of African descent – and two long-established visible minorities in the EU, specifically the Roma and Russian communities residing in Estonia, Latvia and Lithuania (FRA, 2010[63]). EU-MIDIS II expanded the scope to include direct descendants of visible minority immigrants, adding both foreign-born and native‑born individuals with foreign-born parents of non-European background into the analysis, including those of Asian descent who were not covered in EU-MIDIS I (FRA, 2017[64]).

Cross-national surveys focusing exclusively on recently arrived visible minorities

In 2022, the FRA conducted a follow-up to EU-MIDIS I and II, focusing exclusively on immigrants and their immediate descendants of non-European background (FRA, 2024[65]). This survey, titled the “EU Survey on Immigrants and Descendants of Immigrants”, was administered to slightly more than 15 000 respondents across 15 EU countries: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Poland, Portugal, Spain, and Sweden.

The FRA used this last survey as a foundation to take a closer look at the experiences of Black and Muslim communities in the EU. The report “Being Black in the EU” was published in 2023 (FRA, 2023[66]), offering the first in-depth analysis of this group across Member States. In 2024, the FRA released “Being Muslim in the EU” (FRA, 2024[67]), with the first edition, based on EU-MIDIS II, having been published in 2017 (FRA, 2017[68]).

Cross-national surveys focusing exclusively on long-established visible minorities

In 2021, the FRA conducted a survey on Roma populations across eight selected EU Member States (Czechia, Greece, Spain, Croatia, Italy, Hungary, Portugal, and Romania) and two accession countries (North Macedonia and Serbia). The findings, published in 2022 (FRA, 2022[69]), build on insights from earlier reports released in 2012 (FRA, 2012[70]) and 2016 (FRA, 2016[71]), which were based on data from EU-MIDIS I and II, respectively.

Moreover, FRA conducted a cross-country survey among Jews living in the EU in 2012, 2018, and 2023 (FRA, 2013[72]; FRA, 2018[73]; FRA, 2024[74]). However, these surveys cannot be considered representative, as they rely on social sampling – using social media platforms and networks to recruit participants, typically through surveys or polls distributed via these channels. This approach is commonly employed when no comprehensive sampling frame exists for the target group, as is the case for Jewish populations. In 2023, for example, nearly 8 000 individuals aged 16 and over who self-identified as Jewish were surveyed across 13 EU countries (Austria, Belgium, Czechia, Denmark, France, Germany, Hungary, Italy, the Netherlands, Poland, Romania, Spain, and Sweden). Over 300 international, national, and regional Jewish organisations, representing a wide range of affiliations, assisted in the outreach by disseminating multiple invitations through emails, newsletters, instant messages, and social media advertisements to encourage participation.

2.4. Extending data collection on visible minorities to also include long-established groups

Visible minorities in EU countries encompass a diverse array of groups, not limited to recently arrived individuals and their immediate descendants of non-European background. As highlighted in Box 2.1, these groups also include long-established populations, such as third-generation citizens of non-European descent, alongside other distinct communities such as Indigenous peoples, Jewish populations, Roma communities, Black populations from overseas territories as well as other national and ethnic minorities whose distinctiveness benefits from official recognition in national legal frameworks.

This section explores options for EU countries to enhance the identification of visible minorities within cross-national surveys and national statistical frameworks, by incorporating these long-established groups. Expanding data collection in this way would facilitate the adaptation of the indicators proposed in this chapter to these populations, thereby strengthening the monitoring of national anti-racism action plans aimed at protecting all segments of visible minorities.

Members have two main approaches to consider for extending data collection on visible minorities to include long-established groups. The first approach is intuitive, as it builds upon the current methodology used in all cross-national surveys examined in this chapter, as well as in the national statistics of over half of EU countries. This method involves gathering information on the country of birth of grandparents, and potentially earlier generations, in addition to that of individuals and their parents. This enables analysis of third and possibly higher generations of EU citizens of non-European descent, who represent a growing share of long-established visible minorities in the EU.

The second approach is more comprehensive but represents a significant departure from current practices, as it incorporates questions on racial and ethnic identification in both cross-national and national representative surveys that already collect data on respondents’ and their parents’ country of birth. Unlike the traditional focus on ancestors’ country of birth, this approach expands the ability to identify a broader range of subgroups within visible minorities who are neither immigrants nor immediate descendants of immigrants – allowing for the recognition of those who can be characterised as “long-established”.

This section briefly discusses the strengths and limitations of each approach and, wherever possible, provides options to address some of their respective challenges.

2.4.1. Collecting information on the country of birth of grandparents and potentially earlier generations

The first approach consists in gathering information on the country of birth of grandparents, with two possible options for doing so. The first option is to expand existing cross-national surveys, national censuses, or large‑scale nationally representative surveys that already collect information on respondents’ and parents’ country of birth by adding questions on the country of birth of all four grandparents. This approach has been implemented, for example, in the second wave of France’s Trajectories and Origins survey and in Portugal’s 2023 Survey on Living Conditions, Origins, and Trajectories of the Resident Population.

However, integrating such questions comes with challenges. Respondents may hesitate to disclose detailed family background information, either because they see it as a process that risks assigning them outsider status if an ancestor is non-native, or because they may not know all four grandparents’ birthplaces. Meanwhile, the inclusion of four additional questions could significantly increase survey complexity and the perceived burden on respondents, thereby further undermining response rates. In this setting, extending data collection beyond grandparents (e.g. to great-grandparents) is impractical with the first option.

To address these challenges, a second option involves collecting data on the birthplaces of individuals and their ancestors directly from national population registers. This process enables multi-generational tracking by linking individuals with their family records. In five EU countries – Belgium, Denmark, Finland, the Netherlands, and Sweden – such data are already collected across three generations, starting with parents’ declarations at their child’s birth. However, if a country adopts this approach only now, there will be a significant delay before more than two generations of data can be observed, unless retrospective linkage of family records across generations is achievable.

It is important to note that both the first and the second options face limitations when applied to specific national contexts. For instance, in countries with overseas territories, where individuals may belong to visible minority groups despite being native born for generations, these options risk generating “false negatives”, undercounting certain populations. Likewise, in countries with a considerable population of repatriates from former colonies, individuals born abroad may still be part of the majority, leading to “false positives” in the data.

2.4.2. Relying on questions on racial/ethnic identification

The second approach involves adding questions on racial and ethnic identification to cross-national and large‑scale national surveys, including censuses, that already collect data on respondents’ and their parents’ countries of birth. This approach allows distinguishing a broader range of long-established groups, including not only the grandchildren of non-European immigrants but also other long-standing visible minorities.

Currently, only two EU countries, Estonia and Spain, have implemented or plan to implement this approach in their census.

• In Estonia, Statistics Estonia conducts the survey component of the population and housing census, which is otherwise based on registers, every ten years. This survey includes questions on respondents’, their parents’, and even their grandparents’ country of birth, along with a question on race and ethnicity, designed to identify national and ethnic minorities residing in Estonia. This question is phrased as “What is your ethnicity?” with response options such as “Estonian”, “Russian”, “Ukrainian”, “Belarusian”, “Finnish”, and “Other ethnicity” (in which case respondents are prompted to specify).

• Spain plans to incorporate a question on race and ethnicity in the 2026 wave of the Encuesta de Características Esenciales de la Población y las Viviendas (ECEPOV), or Survey of Essential Population and Housing Characteristics. This survey, first conducted in 2021 with half a million participants (representing 1% of the population), already collects information on respondents’ and their parents’ country of birth. The planned question on race/ethnicity, subject to qualitative testing and possible adjustments, reads: “Considering your family history, ancestry, background, sense of belonging, and culture, you are considered a person who is…” with multiple answers allowed. Response options include: (i) “Black, Afro, Afro-descendant, Afro Spanish, African Black”; (ii) “Gypsy, Romani, Roma”; (iii) “Arabic, Amazigh, North African non-Arab”; (iv) “White Latin American”; (v) “Native American, Indo-American, Indigenous, Abya Yala Native”; (vi) “East or Southeast Asian”; (vii) “South Central Asian”; (viii) “West Asian, Turkish”; (ix) “White, Mediterranean”; (x) “Mixed, mestizo, multi‑ancestry” (with a prompt to specify); (xi) “Other” (with a prompt to specify); (xii) “I don’t know”; (xiii) “I prefer not to answer”; (xiv) “I don’t understand this question”.

Although subject to stringent regulations, including the 2018 EU General Data Protection Regulation (GDPR), questions on sensitive characteristics are permissible if necessary for public interest and if they uphold high standards of human rights – particularly in terms of privacy, consent, and self-identification rather than third-party assignment, including interviewer assessment (European Commission, 2021[75]).

• Privacy safeguards ensure that individuals’ information is protected from unauthorised access, thereby securing the confidentiality of collected data.

• Consent is crucial to allow individuals to decide voluntarily to share their data, especially regarding sensitive characteristics such as race or ethnicity. It is essential, therefore, that individuals are fully informed of the purpose of the data collection on race/ethnicity and of their right to disclose or withhold such information. Specifically, respondents should be clearly notified that providing information on race or ethnicity, or any other sensitive characteristic, is entirely optional.

• Self-identification empowers individuals to define their own identity. While an open-text response option might initially seem ideal for capturing such self-determined information, it presents several challenges. Firstly, individuals may not readily think of themselves in racial or ethnic terms. Secondly, open-ended responses generate a broad array of answers, complicating analysis due to the need for categorisation. A more effective approach is to use pre‑coded answer options for racial and ethnic self-identification, supplemented by an “Other, please specify” option, with respondents able to select multiple categories to capture the complexity of their identities. The selection of pre‑coded categories should be informed by extensive consultation with community organisations to ensure the options provided reflect the diverse ways people perceive their racial and ethnic identities. Moreover, respondents should have the option to select multiple racial/ethnic categories to reflect the complexity of their identities.

Finally, appropriate measures should be taken to ensure broad public acceptance of racial/ethnic self-identification questions, thereby supporting a high response rate. This can be achieved through comprehensive public consultations prior to implementing these questions, helping to foster trust in the data collection process. Such consultations should serve multiple purposes: they should facilitate discussions on defining pre‑coded response options for racial and ethnic self-identification while also engaging both the majority population and visible minority groups in dialogue about the survey’s overarching goals.

Yet, fostering broad public acceptance of questions on racial and ethnic self-identification can be challenging, as concerns remain that such data collection could unintentionally reinforce racial/ethnic constructs and deepen social divisions rather than promote inclusion (Prewitt, 2013[76]). In contexts where direct self-identification of visible minorities lacks broad support, alternative approaches may be more effective.

• One option is to introduce a question on self-assessed racial/ethnic hetero-perception, which asks respondents how they believe others perceive them in terms of race and ethnicity. This approach could be more relevant than a question on self-identification, as discrimination often hinges on external perceptions rather than an individual’s own sense of identity. This perspective has influenced the design of the race/ethnicity question in the survey component of the Spanish census, which takes a middle ground between direct self-identification and self-assessed hetero-perception.

• A second option involves asking respondents if they are aware of any ancestors originating from specific world regions, based on a provided list. This approach, which captures a wide scope of family heritage without requiring direct self-identification and without asking exact details on the country of birth of ancestors that respondents may not possess, is recommended by Thomas Piketty in his 2022 book Measuring Racism, Overcoming Discriminations (Piketty, 2022[77]). He proposes a general question such as: “To your knowledge, do you have ancestors from these parts of the world?” followed by regions like North Africa, Sub-Saharan Africa, South Asia, the Middle East, East Asia, Latin America, North America, Southern Europe, and Northern Europe, with a “Yes/No” choice for each.

However, this latter option primarily identifies the segment of long-established visible minorities composed of individuals of non-European background who are neither immigrants nor direct descendants of immigrants. Yet, as already stressed, even long-established populations of European descent can be visible minorities, in the sense of being perceived as culturally distinct, such as those of Jewish or Roma heritage. Similarly, long-established populations with ancestors from the Americas may not be considered visible minorities in Europe unless they are of Indigenous origin.

To enhance accuracy, follow-up questions could be added to clarify specific ancestral backgrounds. For example, a follow-up question could be phrased as: “You indicated some of your ancestors are from Europe; could you specify if they belonged to any of these subgroups?”, with options such as “Sami”, “Jewish”, or “Roma” being proposed. Similarly, if a respondent indicates ancestry from the Americas, a subsequent question could help determine if they are of Indigenous background.

2.5. Conclusion

This chapter develops and discusses a range of indicators designed to measure bias-driven racial/ethnic discrimination across key areas, including education, school-to-work transition, employment, housing, and health. These indicators are essential for countries to track progress, identify persistent challenges, and adjust strategies to maximise the effectiveness of national action plans against racism.

These indicators can be split into three groups:

• Core indicators are derived from cross-country surveys and aim to capture “unexplained” gaps between majority and visible minority populations in critical final outcomes but also in a selection of intermediate factors that allow shedding light on some of the mechanisms highlighted in Chapter 1. Yet, although these gaps indicate potential bias-driven discrimination against visible minorities, they are not conclusive proof, as observed disparities could also be driven by differences in unobserved factors.

• Advanced enhanced indicators leverage administrative databases to analyse disparities across the entire population and a broader range of outcomes – although, unfortunately, not perception-based ones – while controlling for a more extensive set of confounding factors. This robust data collection allows, in some instances, to focus on outcomes better designed to assess the impact of bias-driven discrimination against visible minorities. Such is the exploration of bias in teacher grading by comparing grading disparities between majority and minority students across two assessment types within a specific subject, one that relies on blindly graded national standardised examinations, the other which is based on non-blindly graded school-based tests. Examples of tailored outcomes also include unexplained gaps in police arrest rates without subsequent convictions. Yet, although they offer greater precision, advanced enhanced indicators require data which are currently available in at most half of EU countries.

• Advanced exploratory indicators necessitate the collection of original data, primarily through field experiments with strong potential for identifying and measuring bias-driven discrimination, and, in a more targeted set of cases, the use of artificial intelligence. For instance, artificial intelligence can detect quantitative and qualitative underrepresentation of visible minorities in textbooks, or discriminatory practices in employer searches on job portals. Experimental methods include correspondence and audit studies. Despite several strengths, these methods also have limitations, as they are impractical in certain contexts, such as schools or workplaces, for instance when promotion decisions are involved.

Despite their respective limitations, the indicators proposed in this chapter provide a valuable foundation for countries to better monitor the impact of national action plans against racism. However, two additional caveats regarding core indicators highlight the need for more countries to develop their own advanced enhanced indicators.

First, although core indicators are intended to be readily computable across all EU countries, practical implementation may be challenging. Comparing native‑born individuals with two native‑born parents (majority population) to native‑born individuals with at least one foreign-born parent – and further distinguishing those with European versus non-European ancestry – requires sufficiently large sample sizes. This may not be feasible in many EU countries with smaller populations and low shares of recently arrived visible minorities. To address this caveat, it is essential that EU countries strengthen their statistical frameworks by collecting data on individuals’ and parents’ country of birth and linking them with administrative databases in key areas such as education, employment, housing, health, and income.

Second, core indicators are derived from cross-national surveys that do not capture the situation of long-established visible minorities. Therefore, it is crucial for countries to explore options to ensure that their advanced enhanced indicators can monitor the impact of anti-racism efforts aimed at protecting not only recently arrived but also long-established visible minorities from bias-driven discrimination. It is hoped that this chapter’s guidance on extending data collection on visible minorities will assist countries in stepping up efforts to achieve this goal.

References

[45] Adukia, A. (2023), “What we teach about race and gender: Representation in images and text of children’s books”, The Quarterly Journal of Economics, 138(4), 2225-2285, https://academic.oup.com/qje/article/138/4/2225/7247000.

[35] Aldridge, R. et al. (2018), “Global patterns of mortality in international migrants: a systematic review and meta-analysis.”, The Lancet, 392(10164), 2553-2566.

[62] Aranguren, M. (2024), “Racial discrimination in helping situations depends on the cost of help: A large field experiment in the streets of Paris”, The British Journal of Sociology, https://doi.org/10.1111/1468-4446.13156.

[49] Auer, T. et al. (2022), “Same Different but not Same Same: Ethnic Discrimination in Application Procedures in the German Public Sector”, Empirical Economics Letters, 21(7).

[15] Awisati, F. and C. Wuyts (2024), The Measurement of Socio-economic Status in PISA, OECD Education Working Papers, No. 321, OECD Publishing, Paris, https://doi.org/10.1787/0c5b793c-en.

[30] Boutchenik, B., P. Givord and O. Monso (2021), “Ségrégation urbaine et choix du collège : quelles contributions à la ségrégation scolaire ?”, Revue économique, Vol. 72(5), 717-747, https://doi.org/10.3917/reco.725.0717.

[29] Boutchenik, B., P. Givord and O. Monso (2018), “Social segregation between colleges”, INSEE Analyses No 40, https://www.insee.fr/fr/statistiques/3614217.

[33] Brodaty, T., L. Du Parquet and P. Petit (2014), “La discrimination à l’entrée des établissements scolaires privés”, Revue française d’économie, Volume XXIX(2), 143-178, https://doi.org/10.3917/rfe.142.0143.

[39] Caleyachetty et al. (2021), “Ethnicity-specific BMI cutoffs for obesity based on type 2 diabetes risk in England: a population-based cohort study”, The Lancet Diabetes & Endocrinology Volume 9, Issue 7p419-426July 2021, https://www.thelancet.com/journals/landia/article/PIIS2213-8587(21)00088-7/fulltext.

[4] Crenshaw, K. (1989), “Demarginalizing the Intersection of Race and Sex: A Black Feminist Critique of Antidiscrimination Doctrine, Feminist Theory and Antiracist Politics”, U. Chi. Legal F. 139 (1989)., https://scholarship.law.columbia.edu/faculty_scholarship/3007/.

[60] Crosby, F., S. Bromley and L. Saxe (1980), “Recent unobtrusive studies of Black and White discrimination and prejudice: A literature review.”, Psychological Bulletin, 87(3), 546–563, https://psycnet.apa.org/doi/10.1037/0033-2909.87.3.546.

[59] Docquier, F. et al. (2022), “Le racisme et les discriminations ethno‐raciales au Luxembourg”, Luxembourg Institute of Socioeconomic Research, https://mfsva.gouvernement.lu/dam-assets/publications/rapport-etude-analyse/racisme/Rapport-d-etude-Enquete-Racisme.pdf.

[22] Drydakis, N., A. Paraskevopoulou and V. Bozani (2022), “A field study of age discrimination in the workplace: the importance of gender and race‒pay the gap.”, Employee Relations: The International Journal, 45(2), 304-327..

[1] European Commission (2024), Report from the Commission to the European Parliament, the Council, the Euroepan Economic and Social Committee, and the Committee of the Regions, https://commission.europa.eu/document/download/4968fa88-5350-48d9-bf36-abd3c0142aa8_en?filename=Report%20Antiracism.pdf.

[75] European Commission (2021), “Guidance note on the collection and use of equality data based on racial or ethnic origin”, Luxembourg: Publications Office of the European Union, https://commission.europa.eu/system/files/2022-02/guidance_note_on_the_collection_and_use_of_equality_data_based_on_racial_or_ethnic_origin_final.pdf.

[2] European Commission (2020), A Union of equality: EU anti-racism action plan 2020-2025, https://migrant-integration.ec.europa.eu/library-document/union-equality-eu-anti-racism-action-plan-2020-2025_en#:~:text=747.09%20KB%20%2D%20PDF)-,Download,-Action%20Plan%20factsheet.

[78] European Council (2024), “Pay transparency in the EU”, https://www.consilium.europa.eu/en/policies/pay-transparency/#:~:text=The%20Council%20has%20adopted%20new,gender%20pay%20gap%20exceeds%205%25.

[10] European Health Interview Survey (EHIS) (2024), Reference Metadata in Euro SDMX Metadata Structure (ESMS) Compiling agency: Eurostat, the statistical office of the European Union, https://ec.europa.eu/eurostat/cache/metadata/en/hlth_det_esms.htm.

[7] Eurostat (2024), Labour force survey (LFS) - sampling design, sample size and sampling errors, Eurostat Statistics Explained, https://ec.europa.eu/eurostat/statistics-explained/index.php?_sample_size_and_sampling_errors&oldid=457143#Who_decides_on_the_sampling_methodology_.3F.

[12] Eurostat (2022), Leaving home: Young Europeans spread their wings, Eurostat, https://ec.europa.eu/eurostat/web/products-eurostat-news/-/ddn-20220901-1.

[11] Eurostat (2022), Quality report of the third wave of the European Health Interview survey, Luxembourg: Publications Office of the European Union, https://ec.europa.eu/eurostat/documents/7870049/14937972/KS-FT-22-002-EN-N.pdf/40912c04-ac9d-012c-358e-111efe0a8976?t=1659613423879.

[42] Eurostat (2017), “Quality of life in Europe - facts and views - overall life satisfaction”, Eurostat Statistics Explained, https://ec.europa.eu/eurostat/statistics-explained/index.php?oldid=349426#Life_satisfaction.

[34] Fairlie, R. et al. (2024), “Estimating Peer Effects among College Students: Evidence from a Field Experiment of One-to-One Pairings in STEM”, National Bureau of Economic Research., https://www.nber.org/papers/w33060.

[67] FRA (2024), “Being Muslim in the EU - Experiences of Muslims”, European Union Agency for Fundamental Rights, https://fra.europa.eu/sites/default/files/fra_uploads/fra-2024-being-muslim-in-the-eu_en.pdf.

[74] FRA (2024), “Jewish People’s Experiences and Perceptions of Antisemitism”, European Union Agency for Fundamental Rights, https://fra.europa.eu/en/publication/2024/experiences-and-perceptions-antisemitism-third-survey.

[65] FRA (2024), “Technical and Quality Report - EU Survey on Immigrants and Descendants of Immigrants”, European Union Agency for Fundamental Rights, https://fra.europa.eu/sites/default/files/fra_uploads/fra-2024-immigrants-survey-technical-quality-report_en.pdf.

[66] FRA (2023), “Being black in the EU ― experiences of people of African descent”, Publications Office of the European Union, https://fra.europa.eu/en/publication/2023/being-black-eu.

[69] FRA (2022), “Roma Survey 2021 - Main Results”, European Union Agency for Fundamental Rights, https://fra.europa.eu/en/publication/2023/roma-survey-2021-technical-report.

[73] FRA (2018), “Experiences and perceptions of antisemitism - Second survey on discrimination and hate crime against Jews in the EU”, European Union Agency for Fundamental Rights, https://fra.europa.eu/en/publication/2018/experiences-and-perceptions-antisemitism-second-survey-discrimination-and-hate.

[64] FRA (2017), “Second European Union Minorities and Discrimination Survey - Main results”, European Union Agency for Fundamental Rights, https://fra.europa.eu/en/publication/2017/second-european-union-minorities-and-discrimination-survey-main-results.

[68] FRA (2017), “Second European Union Minorities and Discrimination Survey; Muslims - Selected findings”, European Union Agency for Fundamental Rights, https://fra.europa.eu/en/publication/2017/second-european-union-minorities-and-discrimination-survey-muslims-selected.

[71] FRA (2016), “Second European Union Minorities and Discrimination Survey: Roma - Selected findings”, European Union Agency for Fundamental Rights, https://fra.europa.eu/en/publication/2016/second-european-union-minorities-and-discrimination-survey-roma-selected-findings.

[72] FRA (2013), “Discrimination and hate crime against Jews in EU Member States: experiences and perceptions of antisemitism”, European Union Agency for Fundamental Rights, https://fra.europa.eu/en/publication/2013/discrimination-and-hate-crime-against-jews-eu-member-states-experiences-and.

[70] FRA (2012), “The situation of Roma in 11 EU Member States - Survey results at a glance”, European Union Agency for Fundamental Rights, https://fra.europa.eu/en/publication/2012/situation-roma-11-eu-member-states-survey-results-glance.

[63] FRA (2010), “EU-MIDIS: European Union Minorities and Discrimination Survey, Main Results Report”, European Union Agency for Fundamentals Rights, https://fra.europa.eu/sites/default/files/fra_uploads/663-fra-2011_eu_midis_en.pdf.

[32] Frohly, C. (2022), “Does school choice increase social segregation? Evidence from private schools opening in France between 2005 and 2019”, https://drive.google.com/file/d/1lLQxQb15ZYzFfZ0FG1vnVEEX7ZAbuMgn/view.

[57] Fumarco, L. et al. (2024), “Gender Identity-, Race-, and Ethnicity-Based Discrimination in Access to Mental Health Care: Evidence from an Audit Correspondence Field Experiment”, American Journal of Health Economics, 10(2), 182-214..

[28] Givord, P. (2019), “How are school-choice policies related to social diversity in schools?”, PISA in Focus, No. 96, OECD Publishing. Paris, https://doi.org/10.1787/2d448c77-en.

[40] Griswold, M. et al. (2018), “Alcohol use and burden for 195 countries and territories, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016”, The Lancet, 392(10152), 1015-1035, https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(18)31310-2/fulltext.

[53] Hangartner, D., D. Kopp and M. Siegenthaler (2021), “Monitoring hiring discrimination through online recruitment platforms”, Nature, 589(7843), 572-576.

[41] Hartz, S. et al. (2018), “Daily drinking is associated with increased mortality.”, Alcoholism: Clinical and Experimental Research, 42(11), 2246-2255, https://onlinelibrary.wiley.com/doi/full/10.1111/acer.13886.

[52] Institut des Politiques Publiques (2023), “Évaluation de l’Index de l’égalité professionnelle”, Rapport IPP no. 42, https://www.ipp.eu/publication/evaluation-de-lindex-degalite-professionnelle/.

[48] Kaas, L. and C. Manger (2012), “Ethnic discrimination in Germany’s labour market: A field experiment”, German economic review, 13(1), 1-20, https://docs.iza.org/dp4741.pdf.

[50] Kim, W. (2024), “Overqualification Among Second-Generation Children of Immigrants in the Swedish Labour Market”, Eur J Population 40, 34 (2024)., https://doi.org/10.1007/s10680-024-09723-5.

[55] Kline, P., E. Rose and C. Walters (2024), “A discrimination report card”, American Economic Review, 114 (8): 2472–2525, https://www.aeaweb.org/articles?id=10.1257/aer.20230700.

[54] Kline, P., E. Rose and C. Walters (2022), “Systemic discrimination among large US employers”, The Quarterly Journal of Economic, 137(4), 1963-2036.

[46] Lucy, L. et al. (2020), “Content analysis of textbooks via natural language processing: Findings on gender, race, and ethnicity in Texas US history textbooks”, AERA Open, 6(3), 2332858420940312, https://journals.sagepub.com/doi/full/10.1177/2332858420940312.

[5] McLaughlin, J. and D. Neumark (2025), “Is There Intersectional Labor Market Discrimination?”, NBER, https://www.nber.org/papers/w33388.

[43] Meyer, B., N. Mittag and D. Wu (2024), Race, Ethnicity, and Measurement Error, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/w32860.

[25] OECD (2024), “Challenges to measuring homelessness among migrants in OECD and EU countries”, https://www.oecd.org/en/publications/challenges-to-measuring-homelessness-among-migrants-in-oecd-and-eu-countries_b9855842-en.html.

[51] OECD (2024), “Communicating gender pay gap reporting rules and results”, OECD, https://www.oecd.org/en/publications/communicating-gender-pay-gap-reporting-rules-and-results_12e1dd50-en.html.

[23] OECD (2024), International Migration Outlook 2024, OECD Publishing, Paris, https://doi.org/10.1787/50b0353e-en.

[58] OECD (2024), “What can we learn from surveys on the social integration of immigrants”, OECD Publishing, Paris, https://www.oecd.org/en/publications/what-can-we-learn-from-surveys-on-the-social-integration-of-immigrants_de04dd2c-en.html.

[19] OECD (2023), “PISA 2022 Results (Volume I): The State of Learning and Equity in Education”, OECD Publishing, Paris, https://doi.org/10.1787/53f23881-en.

[20] OECD (2023), PISA 2022 Results (Volume I): The State of Learning and Equity in Education, PISA, OECD Publishing, Paris, https://doi.org/10.1787/53f23881-en.

[8] OECD (2023), The PISA target population, the PISA samples, and the definition of schools, PISA Results (Volume II): Learning During - and From - Disruption, OECD Publishing, Paris, https://doi.org/10.1787/39b3dc7e-en.

[16] OECD (2022), Relative earnings of tertiary-educated adults, by field of study (2020): 25-64 year-olds with income from employment (full-time full-year workers); upper secondary education (all fields) = 100, Education at a Glance 2022: OECD Indicators, OECD Publishing, Paris,, https://doi.org/10.1787/f483b9db-en.

[14] OECD (2021), “Teaching as a Knowledge Profession: Studying Pedagogical Knowledge across Education Systems”, OECD Publish, Paris, https://doi.org/10.1787/e823ef6e-en.

[27] OECD (2019), “Balancing School Choice and Equity: An International Perspective Based on Pisa”, PISA, OECD Publishing, Paris, https://doi.org/10.1787/2592c974-en.

[13] OECD (2019), “How education systems respond to cultural diversity in schools New measures in TALIS 2018”, OECD Publishing, Paris, https://www.oecd.org/en/publications/how-education-systems-respond-to-cultural-diversity-in-schools_1baa285c-en.html.

[18] OECD (2019), “PISA 2018 Assessment and Analytical Framework”, PISA, OECD Publishing, Paris, https://doi.org/10.1787/b25efab8-en.

[17] OECD Publishing, P. (ed.) (2019), “How does PISA assess reading?”, OECD Publishing, Paris, https://doi.org/10.1787/8eebc6cc-en.

[6] OECD/European Commission (2023), Indicators of Immigrant Integration 2023: Settling In, OECD Publishing, Paris, https://doi.org/10.1787/1d5020a6-en.

[77] Piketty, T. (2022), “Mesurer le racisme, vaincre les discriminations”, Seuil, https://www.seuil.com/ouvrage/mesurer-le-racisme-vaincre-les-discriminations-thomas-piketty/9782021505092.

[76] Prewitt, K. (2013), “What Is “Your” Race?: The Census and Our Flawed Efforts to Classify Americans”, Princeton University Press, https://press.princeton.edu/books/hardcover/9780691157030/what-is-your-race.

[61] Saucier, D., C. Miller and N. Doucet (2005), “Differences in Helping Whites and Blacks: A Meta-Analysis.”, Personality and Social Psychology Review, 9(1), 2-16., https://doi.org/10.1207/s15327957pspr0901_1.

[36] Shor, E. and D. Roelfs (2021), “A global meta-analysis of the immigrant mortality advantage.”, International Migration Review, 55(4), 999-1028..

[47] Søndergaard, J. and T. Hussein (2022), “Ethnic Profiling: Main Results from Three Studies”, The Danish Institute for Human Rights, https://menneskeret.dk/files/media/document/Ethnic%20profiling%20-%20main%20results%20from%20three%20studies%2C%20April%202022.pdf.

[31] Souidi, Y. (2023), “School Segregation: Is it All About Residential Sorting? Evidence From a Nationwide School System”, https://conference.iza.org/edu_2023/souidi_y34084.pdf.

[38] Stanford, F., M. Lee and C. Hur (2019), “Race, Ethnicity, Sex, and Obesity: Is It Time to Personalize the Scale?”, Mayo Clin Proc. 2019 Feb;94(2):362-363., https://pmc.ncbi.nlm.nih.gov/articles/PMC6818706/.

[3] UNESCO (2023), “Fighting racism and discrimination: a UNESCO toolkit”, https://unesdoc.unesco.org/ark:/48223/pf0000387454.

[26] Verdugo, G. (2011), “Public housing and residential segregation of immigrants in France, 1968-1999”, Population 66.1 (2011): 169-193.

[21] Vitoroulis, I. and K. Georgiades (2017), “Bullying among immigrant and non-immigrant early adolescents: School- and student-level effects”, Journal of AdolescenceVolume 61, Issue 1 p. 141-151, https://doi.org/10.1016/j.adolescence.2017.10.008.

[24] VIVE (2022), “Hjemløshed i Danmark 2022 - National kortlægning”, VIVE og forfatterne, https://www.vive.dk/media/pure/ozokjnvn/9732199.

[56] Wallace, M., L. Hiam and R. Aldridge (2023), “Elevated mortality among the second-generation (children of migrants) in Europe: what is going wrong? A review”, British Medical Bulletin, 148(1), 5-21.

[37] Wallace, M., L. Hiam and R. Aldridge (2023), “Elevated mortality among the second-generation (children of migrants) in Europe: what is going wrong? A review”, British Medical Bulletin, 148(1), 5-21., https://academic.oup.com/bmb/article/148/1/5/7335807.

[9] Wirth, H. and K. Pforr (2022), The European Union Statistics on Income and Living Conditions after 15 Years, European Sociological Review, Volume 38, Issue 5, October 2022, Pages 832–848, https://doi.org/10.1093/esr/jcac024.

[44] Zanga, G. and E. De Gioannis (2023), “Discrimination in grading: A scoping review of studies on teachers’ discrimination in school”, Studies in Educational Evaluation, 78, 101284., https://www.sciencedirect.com/science/article/pii/S0191491X23000500.

Annex 2.A. Tables

Annex Table 2.A.1. Thirteen core indicators, along with additional advanced indicators, can be developed to measure bias-driven discrimination in education

Overview of core and advanced indicators to measure bias-driven discrimination in education

Indicator	Source	Target group	Frequency	Type of indicator	Availability	Precision
FINAL OUTCOMES: Eight core indicators complemented by a range of advanced enhanced indicators
Educational attainment: Unexplained gap in highest educational attainment, as captured by the following three sub-indicators: (i) Having at most a low educational attainment, defined as ISCED Level 2 or below; (ii) Having at most a medium educational attainment, defined as higher than ISCED Level 2 but below ISCED Level 5; (iii) Having a high educational attainment, defined as ISCED Level 5 or above.	EU-LFS	Individuals aged 15‑24	Yearly	CORE	+++	+
Educational attainment: Conditional on having at most a low educational attainment, unexplained gap in being classified as having dropped out of school, i.e. being neither in education nor training.	EU-LFS	Individuals aged 15‑24	Yearly	CORE	+++ but small sample size	+
Educational attainment: Conditional on having at most a medium educational attainment, unexplained gap in having enrolled in the vocational or technical track, rather than the general track, of upper secondary education.	EU-LFS	Individuals aged 15‑24	Yearly	CORE	+++ but small sample size	+
Educational attainment: Conditional on having a high educational attainment, unexplained gap in not having enrolled in fields of study that are most conducive to high labour earnings.	EU-LFS	Individuals aged 15‑24	Yearly	CORE	+++ but small sample size	+
Educational attainment: Unexplained gap in having repeated a grade at least once at ISCED Level 1 (during primary education), or at ISCED Level 2 (during lower secondary education).	PISA student questionnaire	Students aged 15	Every three years (next round in 2025)	CORE	+++	+
Educational achievement: Unexplained gap in reading literacy score, as captured by the following three sub-indicators: (i) Being a low achiever; (ii) Being a medium achiever; (iii) Being a high achiever.	PISA tests	Students aged 15	Every three years (next round in 2025)	CORE	+++	+
Educational achievement: Unexplained gap in mathematics literacy score, as captured by the following three sub-indicators: (i) Being a low achiever; (ii) Being a medium achiever; (iii) Being a high achiever.	PISA tests	Students aged 15	Every three years (next round in 2025)	CORE	+++	+
Educational achievement: Unexplained gap in science literacy score, as captured by the following three sub-indicators: (i) Being a low achiever; (ii) Being a medium achiever; (iii) Being a high achiever.	PISA tests	Students aged 15	Every three years (next round in 2025)	CORE	+++	+
FINAL OUTCOMES (continued)
A range of additional indicators, targeting the same outcomes as core indicators as well as new ones. For instance, in terms of educational achievement, unexplained gaps between majority and minority students could be calculated for each national standardised test conducted throughout the school curriculum, providing deeper insights into disparities across educational stages.	Administrative databases, including education records, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	Individuals below 24	Continuous	ADVANCED (enhanced)	+/++	++
INTERMEDIATE OUTCOMES: Five core indicators and three advanced indicators (two enhanced and one exploratory)
Sense of exclusion at school (among students): Unexplained gap in students’ reactions to the following three statements: “I feel like an outsider (or left out of things) at school”; “I feel like I belong at school”; “I feel awkward and out of place in my school”.	PISA student questionnaire	Students aged 15	Every three years (next round in 2025)	CORE	+++	+
Sense of exclusion at school (among parents): Unexplained gap in the following two sub-indicators measured among parents of 15‑year‑old students: (i) Reporting that their participation in activities at their child’s school was hindered by feeling unwelcome; (ii) Disagreeing or strongly disagreeing with all of the following three statements: “My child’s school provides an inviting atmosphere for parents to get involved”; “My child’s school provides effective communication between the school and families”; and “My child’s school involves parents in the decision-making process”.	PISA parent questionnaire	Parents of 15‑year‑old students	Every three years (next round in 2025)	CORE	+++ but in a selection of countries	+
Biased textbooks: Identifying the quantitative and qualitative underrepresentation of visible minorities using artificial intelligence.	Artificial intelligence tools	Children’s literature and educational materials, including history curricula and textbooks	Flexible, depending on resources available to perform this analysis	ADVANCED (exploratory)	+	+++
Biased educators (teachers): Unexplained gap in perceived teacher hostility based on students’ reactions to the following eight statements: “The teachers at my school are respectful towards me”; “If I walked into my classes upset, my teachers would be concerned about me”; “If I came back to visit my school three years from now, my teachers would be excited to see me”; “When my teachers ask how I am doing, they are really interested in my answer”; “The teachers at my school are friendly towards me”; “The teachers at my school are interested in students’ well-being”; I feel intimidated by the teachers at my school”; “The teachers at my school are mean towards me”.	PISA student questionnaire	Students aged 15	Every three years (next round in 2025)	CORE	+++	+
INTERMEDIATE OUTCOMES (continued)
Biased educators (teachers): Bias in teacher grading by comparing grading disparities between majority and minority students across two assessment types within a specific subject, one that relies on blindly graded national standardised examinations, the other which is based on non-blindly graded school-based tests.	Administrative databases, including education records, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	Individuals below 24	Continuous	ADVANCED (enhanced)	+/++	++
Biased educators (career counsellors): Unexplained gap in the lack of students’ exposure to (quality) career counselling at school based on the following two sub-indicators: (i) Reporting not having spoken to a career advisor at school; (ii) Disagreeing or strongly disagreeing with the following statement: “I feel well-informed about possible paths for me after [the final year of compulsory education]”.	PISA student questionnaire	Students aged 15	Every three years (next round in 2025)	CORE	+++	+
Biased educators (career counsellors): Unexplained gap in students’ access to career counselling services throughout the school curriculum, provided education records are comprehensive enough to support analysis of this dimension.	Administrative databases, including education records, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	Individuals below 24	Continuous	ADVANCED (enhanced)	+/++	++
Biased schoolmates: Unexplained gap in student’s perception of biased attitudes and behaviours among schoolmates, based on the following two sub-indicators: (i) Students’ peer isolation at school which is computed based on students’ reactions to the following three statements: “I make friends easily at school”; “Other students seem to like me”; and “I feel lonely at school”; (ii) Students’ experiences of bullying by schoolmates based on their rating of the frequency of the following nine incidents: “Other students left me out of things on purpose”; “Other students made fun of me”; “I was threatened by other students”; “Other students took away or destroyed things that belonged to me”; “I got hit or pushed around by other students”; “Other students spread nasty rumors about me”; “I was in a physical fight on school property”; “I stayed home from school because I felt unsafe”; and “I gave money to someone at school because they threatened me”.	PISA student questionnaire	Students aged 15	Every three years (next round in 2025)	CORE	+++	+

Note: Core indicators are defined by limited precision (+) but high availability (+++) as they rely on EU-wide cross-country surveys. Advanced indicators are divided into two subgroups: enhanced indicators, which provide greater precision than core indicators (++) but require more extensive data collection and are therefore available in at most half of EU countries (+/++), and exploratory indicators, which achieve the highest precision (+++) but have limited availability (+) due to their reliance on original data collection (or even the outright impossibility of computing them in certain settings). In some cases, core indicators are calculated based on specific conditions and therefore rely on a subset of the total population, which reduces the number of available observations. Indicators that may be limited by sample size are flagged with the label “but small sample size” in the availability column. Similarly, core indicators that rely on the PISA parent questionnaire, which is optional, are flagged with the label “but in a selection of countries” in the availability column.

Source: OECD.

Annex Table 2.A.2. Four core indicators, along with four advanced indicators, can be developed to measure bias-driven discrimination in school-to-work transition

Overview of core and advanced indicators to measure bias-driven discrimination in school-to-work transition

Indicator	Source	Target group	Frequency	Type of indicator	Availability	Precision
FINAL OUTCOMES: One core indicator and one advanced enhanced indicator
NEET rate: unexplained gap in being NEET (neither in employment, nor in education or training).	EU-LFS	Individuals aged 15‑24 or 15‑29	Yearly	CORE	+++	+
NEET rate: unexplained gap in being NEET (neither in employment, nor in education or training).	Administrative databases, including those that track school-to-work transition, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	Individuals aged 15‑24 or 15‑29	Continuous	ADVANCED (enhanced)	+/++	++
INTERMEDIATE OUTCOMES: Three core indicators and three advanced indicators (two enhanced and one exploratory)
Lack of work-based learning experience during formal education: Unexplained gap in not having done an internship to explore future study options or career paths.	PISA student questionnaire	Students aged 15	Every three years (next round in 2025)	CORE	+++	+
Discrimination in access to work-based learning opportunities during formal education: As assessed through correspondence studies and, where feasible, audit studies.	Field experiments	Fictitious candidates playing the role of individuals below 24	Flexible, depending on resources available to perform this analysis	ADVANCED (exploratory)	+	+++
Exposure to disciplinary actions within the school setting (as reported by students): Unexplained gap in reporting having missed school for over three months in a row due to suspension for reasons such as violence, aggression, or drug-related issues, be it at ISCED Levels 1, 2 or 3.	PISA student questionnaire	Students aged 15	Every three years (next round in 2025)	CORE	+++	+
Exposure to disciplinary actions within the school setting (as reported by parents): Unexplained gap in parents’ dissatisfaction with disciplinary practices in their child’s school.	PISA parent questionnaire	Parents of 15‑year‑old students	Every three years (next round in 2025)	CORE	+++ but in a selection of countries	+
INTERMEDIATE OUTCOMES (continued)
Exposure to disciplinary actions within the school setting (objective measure): Unexplained gap in being subjected to in-school and out-of-school suspensions at different stages of the curriculum, provided education records are comprehensive enough to support analysis of this dimension.	Administrative databases, including education records, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	Individuals below 24	Continuous	ADVANCED (enhanced)	+/++	++
Exposure to disciplinary actions outside the school setting (law enforcement): Unexplained gap in police arrest rates without subsequent convictions.	Administrative databases, including records on arrests and convictions, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	Individuals below 24 although the analysis could be performed on the entire population	Continuous	ADVANCED (enhanced)	+/++	++

Note: Core indicators are defined by limited precision (+) but high availability (+++) as they rely on EU-wide cross-country surveys. Advanced indicators are divided into two subgroups: enhanced indicators, which provide greater precision than core indicators (++) but require more extensive data collection and are therefore available in at most half of EU countries (+/++), and exploratory indicators, which achieve the highest precision (+++) but have limited availability (+) due to their reliance on original data collection (or even the outright impossibility of computing them in certain settings). In some cases, core indicators are calculated based on specific conditions and therefore rely on a subset of the total population, which reduces the number of available observations. Indicators that may be limited by sample size are flagged with the label “but small sample size” in the availability column. Similarly, core indicators that rely on the PISA parent questionnaire, which is optional, are flagged with the label “but in a selection of countries” in the availability column.

Source: OECD.

Annex Table 2.A.3. Seven core indicators, along with additional advanced indicators, can be developed to measure bias-driven discrimination in employment

Overview of core and advanced indicators to measure bias-driven discrimination in employment

Indicator	Source	Target group	Frequency	Type of indicator	Availability	Precision
FINAL OUTCOMES: Seven core indicators complemented by a range of advanced enhanced indicators
Labour market status: Unexplained gap in labour market status is captured by the following three sub-indicators: (i) Being inactive; (ii) Being unemployed; (iii) Being employed.	EU-LFS	Individuals aged 15‑64	Yearly	CORE	+++	+
Involuntary inactivity: Conditional on being inactive, unexplained gap in being involuntarily so.	EU-LFS	Individuals aged 15‑64	Yearly	CORE	+++ but small sample size	+
Long-term unemployment: Conditional on being unemployed, unexplained gap in being in long-term rather than short-term unemployment.	EU-LFS	Individuals aged 15‑64	Yearly	CORE	+++ but small sample size	+
Low quality employment: Unexplained gap in low-quality employment is captured by the following three sub-indicators: (i) Conditional on being employed, being self-employed rather than in dependent employment; (ii) Conditional on being in dependent employment, holding a fixed-term (or temporary) rather than an open-ended contract; (iii) Conditional on being in dependent employment, engaging in part-time rather than full-time work; (iv) Conditional on being in dependent employment, being overqualified for the job held.	EU-LFS	Individuals aged 15‑64	Yearly	CORE	+++ but small sample size	+
Wage disparities: Unexplained gap in labour earnings, calculated conditional on being in dependent employment.	EU-LFS	Individuals aged 15‑64	Yearly	CORE	+++ but small sample size	+
Lack of access to life‑long learning: Conditional on being in dependent employment, unexplained gap in not having followed a job-related education or training initiated or recommended by employers.	EU-LFS	Individuals aged 15‑64	Yearly	CORE	+++ but small sample size	+
Firing disparities: Conditional on prior work experience, unexplained gap in having left one’s previous job due to dismissal.	EU-LFS	Individuals aged 15‑64	Yearly	CORE	+++ but small sample size	+
A range of additional indicators, targeting the same outcomes as core indicators as well as new ones. For instance, rather than focusing solely on economy-wide labour earnings gaps, robust data collection could enable the calculation of these gaps at the firm level.	Administrative databases, including employment records, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	Individuals aged 15‑64	Continuous	ADVANCED (enhanced)	+/++	++
INTERMEDIATE OUTCOMES: Two advanced exploratory indicators
Hiring discrimination: As assessed through correspondence studies and, where feasible, audit studies.	Field experiments	Fictitious candidates playing the role of individuals aged 15‑64	Flexible, depending on resources available to perform this analysis	ADVANCED (exploratory)	+	+++
Discrimination in employers’ candidate search behaviour: As assessed through the public employment service website.	Artificial intelligence tools	Employers and job seekers registered on the public employment service website	Flexible, depending on resources available to perform this analysis	ADVANCED (exploratory)	+	+++

Note: Core indicators are defined by limited precision (+) but high availability (+++) as they rely on EU-wide cross-country surveys. Advanced indicators are divided into two subgroups: enhanced indicators, which provide greater precision than core indicators (++) but require more extensive data collection and are therefore available in at most half of EU countries (+/++), and exploratory indicators, which achieve the highest precision (+++) but have limited availability (+) due to their reliance on original data collection (or even the outright impossibility of computing them in certain settings). In some cases, core indicators are calculated based on specific conditions and therefore rely on a subset of the total population, which reduces the number of available observations. Indicators that may be limited by sample size are flagged with the label “but small sample size” in the availability column. Similarly, core indicators that rely on the PISA parent questionnaire, which is optional, are flagged with the label “but in a selection of countries” in the availability column.

Source: OECD.

Annex Table 2.A.4. Three core indicators, along with additional advanced indicators, can be developed to measure bias-driven discrimination in housing

Overview of core and advanced indicators to measure bias-driven discrimination in housing

Indicator	Source	Target group	Frequency	Type of indicator	Availability	Precision
FINAL OUTCOMES: Three core indicators complemented by a range of advanced enhanced indicators
Housing tenure: Unexplained gap in housing tenure is captured by the following two sub-indicators: (i) Being a tenant rather than homeowner household; (ii) Conditional on being a tenant household, residing in social housing.	EU-SILC	Private households (information on the indicator provided by the household reference person)	Every six years as part of the ad hoc module “Intergenerational transmission of disadvantages” (next round in 2029)	CORE	+++ but small sample size for the second sub-indicator	+
Value for money in housing: Unexplained gap in value for money in housing is captured by the following three sub-indicators: (i) Living in an overcrowded accommodation; (ii) Living in a substandard accommodation; (iii) Spending more than 40% of disposable income on housing costs, after accounting for housing allowances.	EU-SILC	Private households (information provided by the household reference person)	Every six years as part of the ad hoc module “Intergenerational transmission of disadvantages” (next round in 2029)	CORE	+++	+
Racial/ethnic segregation: as measured with a dissimilarity index at the school level, comparing two groups, namely native‑born students with two native‑born parents on one hand, and foreign-born students or native‑born students with at least one foreign-born parent on the other hand.	PISA student questionnaire	PISA schools	Every three years (next round in 2025)	CORE	+++	+
A range of additional indicators, targeting the same outcomes as core indicators as well as new ones. For instance, robust data collection could facilitate the calculation of a dissimilarity index not only within schools but also outside schools, focusing on finely defined geographic levels.	Administrative databases, including housing records, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	Private households	Continuous	ADVANCED (enhanced)	+/++	++
INTERMEDIATE OUTCOMES: Two advanced exploratory indicators
Discrimination in the private sales and rental housing markets: As assessed through correspondence studies and, where feasible, audit studies.	Field experiments	Fictitious private households	Flexible, depending on resources available to perform this analysis	ADVANCED (exploratory)	+	+++
Discrimination in access to mortgage loans: As assessed through correspondence studies and, where feasible, audit studies.	Field experiments	Fictitious private households	Flexible, depending on resources available to perform this analysis	ADVANCED (exploratory)	+	+++

Note: Core indicators are defined by limited precision (+) but high availability (+++) as they rely on EU-wide cross-country surveys. Advanced indicators are divided into two subgroups: enhanced indicators, which provide greater precision than core indicators (++) but require more extensive data collection and are therefore available in at most half of EU countries (+/++), and exploratory indicators, which achieve the highest precision (+++) but have limited availability (+) due to their reliance on original data collection (or even the outright impossibility of computing them in certain settings). In some cases, core indicators are calculated based on specific conditions and therefore rely on a subset of the total population, which reduces the number of available observations. Indicators that may be limited by sample size are flagged with the label “but small sample size” in the availability column. Similarly, core indicators that rely on the PISA parent questionnaire, which is optional, are flagged with the label “but in a selection of countries” in the availability column.

Source: OECD.

Annex Table 2.A.5. Six core indicators, along with additional advanced indicators, can be developed to measure bias-driven discrimination in health

Overview of core and advanced indicators to measure bias-driven discrimination in health

Indicator	Source	Target group	Frequency	Type of indicator	Availability	Precision
FINAL OUTCOMES: Three core indicators complemented by a range of advanced enhanced indicators
Poor general health status: Unexplained gap in rating one’s general health as bad or very bad.	EU-SILC or EHIS	Individuals aged 15 (EHIS) or 16 (EU-SILC) and older	Every six years as part of the ad hoc module “Intergenerational transmission of disadvantages” (next round in 2029) for EU-SILC, and every five years as part of the EHIS (next round starting in 2025)	CORE	+++	+
Poor mental health status: Unexplained gap in poor mental health status is captured by the following three sub-indicators: (i) Having suffered from depression in the past 12 months; (ii) Having experienced any of nine symptoms of poor mental health over the last two weeks; (iii) Having felt downhearted or depressed during the past four weeks.	EU-SILC for the third sub-indicator and EHIS for the first two indicators	Individuals aged 15 (EHIS) or 16 (EU-SILC) and older	Every six years as part of the ad hoc module “Intergenerational transmission of disadvantages” (next round in 2029) for EU-SILC, and every five years as part of the EHIS (next round starting in 2025)	CORE	+++	+
Poor physical health status: Unexplained gap in poor physical health status is captured by the following three sub-indicators: (i) Having experienced any of four conditions associated with increased heart rate and blood pressure in the past 12 months; (ii) Having experienced diabetes or high blood lipids in the past 12 months; (iii) Having experienced symptoms of systemic inflammation.	EHIS	Individuals aged 15 and older	Every five years (next round starting in 2025)	CORE	+++	+
A range of additional indicators, targeting the same outcomes as core indicators as well as new ones. For instance, robust data collection could facilitate the calculation of mortality gaps between majority and minority populations.	Administrative databases, including health records, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	The universe of individuals	Continuous	ADVANCED (enhanced)	+/++	++
INTERMEDIATE OUTCOMES: Three core indicators and two advanced indicators (one enhanced and one exploratory)
Maladaptive coping strategies (eating disorder): Unexplained gap in being underweight or overweight based on the Body Mass Index.	EHIS	Individuals aged 15 and older	Every five years (next round starting in 2025)	CORE	+++	+
Maladaptive coping strategies (eating disorder): Unexplained gap in being underweight or overweight based on the Body Mass Index.	Administrative databases, including health records, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	The universe of individuals	Continuous	ADVANCED (enhanced)	+/++	++
Maladaptive coping strategies (substance abuse): Unexplained gap in substance abuse is captured by the following two sub-indicators: (i) Smoking tobacco products (excluding electronic cigarettes or similar electronic devices) on a daily basis; (ii) Consuming alcohol at a frequency of 3‑4 days a week or more.	EHIS	Individuals aged 15 and older	Every five years (next round starting in 2025)	CORE	+++	+
Unmet health needs: Conditional on having needed medical or dental care at least once in the past 12 months, unexplained gap in reporting at least one instance of not receiving the necessary care.	EHIS	Individuals aged 15 and older	Every five years (next round starting in 2025)	CORE	+++ but small sample size	+
Discrimination in access to medical appointments and in the quality of patient-provider interactions: As assessed through correspondence and audit studies.	Field experiments	Fictitious patients across all age groups	Flexible, depending on resources available to perform this analysis	ADVANCED (exploratory)	+	+++

Note: Core indicators are defined by limited precision (+) but high availability (+++) as they rely on EU-wide cross-country surveys. Advanced indicators are divided into two subgroups: enhanced indicators, which provide greater precision than core indicators (++) but require more extensive data collection and are therefore available in at most half of EU countries (+/++), and exploratory indicators, which achieve the highest precision (+++) but have limited availability (+) due to their reliance on original data collection (or even the outright impossibility of computing them in certain settings). In some cases, core indicators are calculated based on specific conditions and therefore rely on a subset of the total population, which reduces the number of available observations. Indicators that may be limited by sample size are flagged with the label “but small sample size” in the availability column. Similarly, core indicators that rely on the PISA parent questionnaire, which is optional, are flagged with the label “but in a selection of countries” in the availability column.

Source: OECD

Annex Table 2.A.6. Two core indicators, along with one advanced indicator, can be developed to measure the overall impact of bias-driven discrimination, specifically in terms of poverty risk and low life satisfaction

Overview of core and advanced indicators to measure the overall impact of bias-driven discrimination, specifically in terms of poverty risk and low life satisfaction

Indicator	Source	Target group	Frequency	Type of indicator	Availability	Precision
FINAL OUTCOMES: Two core indicators and one advanced enhanced indicator
Poverty risk: Unexplained gap in being a household living below the poverty threshold.	EU-SILC	Private households (information on the indicator provided by the household reference person)	Every six years as part of the ad hoc module “Intergenerational transmission of disadvantages” (next round in 2029)	CORE	+++	+
Poverty risk: Unexplained gap in being a household living below the poverty threshold.	Administrative databases, including income records, linked with census data or national population registers identifying individuals’ and their parents’ country of birth	Private households	Continuous	ADVANCED (enhanced)	+/++	++
Low life satisfaction: Unexplained gap in rating one’s life satisfaction below 5 on a 0‑to‑10 life satisfaction scale	EU-SILC	Individuals aged 16 and older	Every six years as part of the ad hoc module “Intergenerational transmission of disadvantages” (next round in 2029)	CORE	+++	+

Note: Core indicators are defined by limited precision (+) but high availability (+++) as they rely on EU-wide cross-country surveys. Advanced indicators are divided into two subgroups: enhanced indicators, which provide greater precision than core indicators (++) but require more extensive data collection and are therefore available in at most half of EU countries (+/++), and exploratory indicators, which achieve the highest precision (+++) but have limited availability (+) due to their reliance on original data collection (or even the outright impossibility of computing them in certain settings). In some cases, core indicators are calculated based on specific conditions and therefore rely on a subset of the total population, which reduces the number of available observations. Indicators that may be limited by sample size are flagged with the label “but small sample size” in the availability column. Similarly, core indicators that rely on the PISA parent questionnaire, which is optional, are flagged with the label “but in a selection of countries” in the availability column.

Source: OECD.

Notes

← 1. * This designation is without prejudice to positions on status, and is in line with United Nations Security Council Resolution 1244/99 and the Advisory Opinion of the International Court of Justice on Kosovo’s declaration of independence.

← 2.  ISCED-F (International Standard Classification of Education – Fields of Education and Training) is a detailed framework introduced by UNESCO in 2013 to classify educational programmes and qualifications by fields of study.

← 3.  The Body Mass Index is calculated by dividing a person’s weight in kilograms by the square of their height in metres.

← 4.  These 15 EU countries are Austria, Belgium, Cyprus, Czechia, Denmark, Estonia, Finland, France, Germany, Italy, Luxembourg, the Netherlands, Slovenia, Spain and Sweden.

← 5.  Specifically, the 2023 EU Pay Transparency Directive introduces rules in three areas: access to information, reporting, and access to justice. Regarding access to information, employers must disclose starting salaries or pay ranges to job applicants and are prohibited from asking about pay history. Moreover, employees are entitled to request information on average pay levels for categories of employees doing the same work or work of equal value, broken down by sex, as well as on the criteria used for pay and career progression, which must be objective and gender-neutral. Regarding reporting, companies with over 250 employees must report annually on gender pay gaps, while smaller firms (100‑249 employees) must report every three years; those with less than 100 employees are exempt. If a pay gap over 5% is found without objective justification, a joint pay assessment with worker representatives is required. Regarding access to justice, workers facing pay discrimination can claim compensation, while the burden of proof, which has traditionally fallen on the employee, now rests on employers, with penalties for non-compliance set to be effective, proportionate and dissuasive (European Council, 2024[78]).