Looking beyond population averages is essential to understanding who is doing well, who is being left behind, and where targeted policy responses may be needed. Disaggregating data by sex, age, disability, migration status or other characteristics can reveal inequalities that aggregate statistics conceal.
But looking at one characteristic at a time reveals only part of the picture. People do not experience their age, sex, disability, socio-economic position or migration status independently of one another. These characteristics overlap, and their intersections can shape people's opportunities and outcomes in ways that are not visible when each is examined separately.
Consider access to healthcare. Statistics may show differences between urban and rural populations, or between higher- and lower-income groups. But a low-income older person living in a rural area may face overlapping barriers: the cost of care or transport may be compounded by long distances and limited mobility. The extent of these barriers also depends on broader factors, such as the availability of public transport, the location of healthcare facilities and the accessibility of health services.
This illustrates the value of bringing an intersectional perspective to statistics: understanding inequalities requires considering not only how individual characteristics intersect, but also the broader social and institutional contexts in which people live.
A new OECD Statistics Working Paper, Intersectionality in Statistics, explores the statistical challenges involved in applying an intersectional perspective and provides a practical overview of the analytical approaches available.
Why greater disaggregation is only part of the answer
The first and most intuitive response might be to disaggregate statistics further: if analysing one characteristic reveals inequalities, combining two, three or more should provide an increasingly complete picture. Indeed, greater disaggregation is essential for making inequalities more visible. But an intersectional perspective goes further by asking which intersections are meaningful and how they are shaped by the broader social and institutional context.
Figure 1: From aggregate data to intersectional analysis
While greater granularity can reveal inequalities, it also creates statistical challenges. With every additional characteristic, the number of possible population groups increases, while the number of observations within each group decreases. Some intersections may contain too few observations to produce reliable estimates. Others may raise confidentiality concerns. And some population groups may be poorly represented or not represented at all in the available data.
There is also a more fundamental question: which intersections should we analyse? Not every possible combination of characteristics will be meaningful for every policy question. The choice should be informed by the issue being studied, existing knowledge and the social and institutional context, rather than by mechanically combining all available variables.
Choosing meaningful intersections is only part of the challenge; the available data must also be able to capture them. No single source necessarily captures all the characteristics, experiences and contextual information needed to understand intersectional inequalities. Surveys can provide rich information about individuals and households, including experiences and perceptions, but may have insufficient sample sizes for detailed analysis. Administrative data can provide much greater coverage and continuity, but may contain a more limited range of characteristics. Other sources, including citizen data – information collected by individuals, communities or civil society organisations – can complement official sources by providing insights into populations or experiences that may otherwise be poorly captured.
Combining or complementing data sources can therefore open new possibilities for intersectional analysis. At the same time, differences in coverage, measurement, representativeness, quality and accessibility need to be carefully considered.
The challenge is therefore not simply to produce more granular data, but to ensure that the data are sufficiently rich, representative and reliable to support meaningful intersectional analysis.
Different questions need different statistical tools
Once relevant intersections have been identified and appropriate data are available, the next question is how to analyse them. No single statistical method can answer every question.
Descriptive analysis is often the starting point. Cross-tabulations and other descriptive statistics can show whether outcomes differ across predefined intersectional groups and reveal patterns that would otherwise remain hidden.
Regression models can go further by examining associations between characteristics and outcomes, accounting for other factors and testing whether the relationship between characteristics differs depending on their combination.
Where many intersectional groups need to be considered simultaneously, multilevel modelling offers additional possibilities. Approaches such as Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) can analyse a large number of predefined intersectional groups while improving the stability of estimates for groups with relatively few observations.
But researchers do not always know beforehand which combinations of characteristics matter most. Cluster analysis, latent class analysis and machine-learning methods can help identify patterns and population groups empirically, including relationships that may be difficult to specify in advance.
Identifying patterns is one thing, but understanding why they emerge is another. Mixed methods can bring quantitative and qualitative evidence together, helping to interpret statistical findings in light of lived experiences and the contexts in which inequalities arise.
These approaches are not competing alternatives. They provide different pieces of the picture. The appropriate choice depends on the research question, the available data and the complexity of the inequalities being examined.
From more detailed data to better evidence
Bringing intersectionality into statistics therefore involves more than adding variables to a dataset or producing increasingly detailed cross-tabulations.
It requires decisions about what to measure, which intersections matter, whether the available data can adequately capture them, and which analytical methods are appropriate. It also requires careful interpretation and communication of the resulting evidence, particularly when estimates are uncertain or population groups are small.
Three messages are particularly important.
Looking at one characteristic at a time can conceal inequalities. Single-axis disaggregation is an important starting point, but examining intersecting characteristics can reveal inequalities that would otherwise remain hidden.
Greater disaggregation is essential – and intersectional analysis takes us further. This requires the right data, sufficient statistical precision, and careful consideration of which intersections are meaningful in a given context.
There is no single statistical method for intersectional analysis. Descriptive statistics, regression, multilevel modelling, data-driven approaches and mixed methods answer different questions and can provide complementary insights.
Taken together, these principles can help statistics better reflect the complexity of inequalities and strengthen the evidence available for policy. The objective is not to make statistical analysis more complicated for its own sake, but to ensure that important inequalities do not remain invisible simply because of how populations are grouped and analysed.
References
OECD (2026), "Intersectionality in statistics" OECD Statistics Working Papers, No. 2026/04, OECD Publishing, Paris
