# Intersectional relationships between age, sex, ethnicity, nationality and experience of racism in the UK using different ethnicity categorisations: A comparative study using survey data

## TL;DR

This study shows that using more detailed ethnic categories provides better insights into how racism and other factors intersect to affect health inequalities in the UK.

## Contribution

The study demonstrates that granular ethnicity categories improve the accuracy of intersectional health inequity analysis compared to traditional broad categories.

## Key findings

- 65% of participants reported experiencing racism in their lifetime.
- The 21-category model revealed significant variations in racism experiences within broad ethnic groups.
- Coarse ethnic categories can mask meaningful differences and lead to misleading interaction effects.

## Abstract

•Careful categorisation and theorising of what ethnicity means is crucial in estimating intersectional health inequalities.•There is a need for better understanding of how different approaches of measuring and analysing ethnicity impact the interpretation of intersectional social contexts.•We compared coarse (5-category) and granular (21-category) ethnicity in their ability to explain variation of experience of racism.•More granular ethnicity enables better description of intersectional disadvantages. This has important implications for study design and analytical approaches to evaluating ethnic health inequities.

Careful categorisation and theorising of what ethnicity means is crucial in estimating intersectional health inequalities.

There is a need for better understanding of how different approaches of measuring and analysing ethnicity impact the interpretation of intersectional social contexts.

We compared coarse (5-category) and granular (21-category) ethnicity in their ability to explain variation of experience of racism.

More granular ethnicity enables better description of intersectional disadvantages. This has important implications for study design and analytical approaches to evaluating ethnic health inequities.

Traditionally, research has relied on broad ethnic categories such as "Asian," "Black," "White," "Mixed," and "Other." These categories often mask significant variations in experiences and outcomes among ethnic subgroups. The Intersectional Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (I-MAIHDA) approach has become an increasingly recognised quantitative method to study intersectional health inequities. There is a need for better understanding of how the granularity with which ethnicity is measured impacts the interpretation of I-MAIHDA. We used the Evidence from Equality National Survey: A Survey of Ethnic Minorities During the COVID-19 Pandemic, 2021 (EVENS) study, a cross-sectional survey conducted between February and November 2021, including 14,221 individuals with 21 ethnic categories. We constructed intersectional social strata using sex, age, ethnicity and UK nationality. We compared models using 21-category and 5-category ethnicity on describing predicted lifetime experience of racism. Overall, 65% of participants reported experiencing racism in their lifetime. The 5-category model has a higher interaction effect compared to 21-caterogy model due to artefacts from coarse ethnic categorisation. While the interaction effects in 21-category model are smaller, they are potentially more meaningful. The 21-cateogory models revealed significant variations within coarse ethnic groups, showing that individuals from Black Caribbean, African and mixed backgrounds had a higher likelihood of experiencing racism, regardless of UK nationality. The 5-category model failed to attribute the protective effect of not being UK nationality to lower predicted experience of racism in White other backgrounds. Our study demonstrates that using more granular ethnicity categories can lead to more accurate and specific insights in characterising inequities when applying quantitative intersectional approaches, over and above coarse ethnicity groupings used in I-MAIHDA or traditional non-interactive models.