18.The Audit’s remit was to collect, process and present existing public-sector data sets held by central government. This included data collected in the course of delivering services (such as GP registration or criminal justice convictions), known as administrative data, as well as surveys and other research conducted by Government. No new statistics were produced and non-Governmental sources were excluded.
Table 1: Types of data sets captured in the Race Disparity Audit
Type of Source
Recorded by individual/third party
Crime Survey for England and Wales
Criminal Justice Statistics
Jobseekers Allowance Claimant Count
19.As the Government begins the task of analysing the results of the Audit and working to “explain or change”, it is crucial that the data contain sufficient richness to ensure that departments can account for all relevant factors, and that it can be broken down meaningfully. Data sets must be consistent and reliable if they are to be used to understand the impact of policy decisions on the ability of people from ethnic minorities to access and benefit from public services. Such an understanding is not only necessary for meeting the Government’s stated aim of developing “ambitious policy responses” to the Audit. It is also essential for fulfilling the Public Sector Equality Duty, which requires public service providers to have due regard to the need to promote equality and eliminate discrimination; this depends on being able to see how and why outcomes for different service users may differ.
20.Part of the challenge for the Race Disparity Unit was that the data sets were not standardised in their collection methodology, ethnicity classifications or the level of detail that they contained. The Audit has revealed a significant lack of consistency in data collection across government departments, which makes comparisons challenging.
21.Witnesses strongly impressed upon us the importance of having standardised classifications of ethnicity across Government bodies and beyond. Interpretation across data sets on different topics and across departments and agencies relies on using the same classifications.
22.The Office for National Statistics (ONS) publishes guidance on which ethnic classifications to use, based on the ‘18+1’ model (18 ethnicity categories plus ‘other’) employed in the 2011 census. Iain Bell of the ONS highlighted the current problem:
What became apparent through the Race Disparity Audit is something that was known, but it shone a light into this area: that many different organisations have yet to align to the latest ONS classifications for the 2011 census, and of course we are now coming up to the 2021 census.
Box 1: The 18+1 classification used by the 2011 Census
Gypsy or Irish Traveller
Mixed/Multiple ethnic group
White and Black Caribbean
White and Black African
White and Asian
Other ethnic group
Any other ethnic group
Source: Office for National Statistics
23.The Audit revealed that numerous data sets were using no more than six categories: Asian, Black, Chinese, Mixed, White and Other. In some cases, the disaggregation was restricted to White British and all Other (or BAME) groups.
Figure 2: An example of restricted disaggregation on the ethnicity facts and figures website.
24.These broad classifications do not help to identify disparities because there can be large in-group differences. Dr Debbie Weekes-Bernard of the Joseph Rowntree Foundation gave the example of education, where an ‘Other’ category would include both Chinese and Indian students and black Caribbean students—groups which have very different patterns of attainment at GCSE. Professor Steve Strand similarly argued against high-level aggregation of ethnic groups into “Asian/Black/White groups”. Gypsy, Roma and Traveller groups pointed out that they are often omitted from statistics even where more detailed categories are used. Since the launch of the website, the Race Disparity Unit has been adding detail to many of the data sets that initially had only two ethnic classifications. This is a welcome development.
25.The Government, led by the Cabinet Office, should adopt the same categories as are used in the Census as the minimum standard for data collection on ethnicity across Government departments, and work with individual departments to ensure that this happens in all official data sets and administrative data in the public services for which they are responsible. At present this means using the ‘18+1’ categories, but should the categories change for the 2021 Census, the Cabinet Office should take advice on how best to ensure comparability of data sets over time.
26.The Equality and Human Rights Commission and the Office for National Statistics should work together to provide updated guidance for public bodies, service providers and employers on how to collect consistent ethnicity data and how public sector bodies should use that data to assess their compliance with the Public Sector Equality Duty.
27.Even when detailed ethnicity categories are used, differences between or within groups are not always fully revealed. Dr Weekes-Bernard argued that factors such as country of origin and English language ability add important additional layers of understanding. Professor Shamit Saggar of the University of Essex endorsed this view:
The solution to this is to try to find a way of having both, as it were, as a way of explaining disadvantage or exclusion. Ethnicity is really important, but being able to say where people are born, for example, in addition to their ethnicity, is even better. Some bits of the data sets we use allow us to do that, but many do not.
Andy Shallice of the Roma Support Group spoke about how important this is in identifying groups that are hidden by the current ‘ethnicity’ classification, citing the lack of classification for migrant Roma on the census as an example. Our predecessor Committee identified this problem in its inquiry into Employment Opportunities for Muslims, where witnesses spoke of the dangers of using ethnicity as a proxy for faith groups and also the difficulties of addressing inequalities when little is known about a group other than their religion or ethnicity.
28.Also of concern is that only some of the data sets on the website are broken down by gender or age, and geographical breakdowns are inconsistent. For instance, GCSE results are broken down by local authority, gender, eligibility for free school meals and type of school, whereas students achieving three A grades at A-level is broken down by ethnicity only. This means it is not possible across the board to identify whether inequalities predominantly affect one gender, one age group or one geographical area, or whether other characteristics are affecting outcomes. For Professor Saggar, such detail was essential to data analysis:
I always stick to the rule of the three Gs: gender, generation and geography. For example, the labour market circumstances of a first-generation Pakistani woman living in Greater Manchester are often very different from a second-generation Indian male living in suburban London. […] They are completely different stories, and yet both are ethnic minorities in the UK today. One is at the top end of the labour market doing fantastically well and so on, on average, and the other is mostly not. You want to do that, otherwise you will say, ‘These are two south Asian people who have the same experiences.’ Actually, they don’t—not at all.
This was also a challenge that our predecessor Committee observed in relation to Employment Opportunities for Muslims, where women’s inactivity in the labour market showed marked variations between groups depending on their migration status, English language skills and ethnicity.
29.Nonetheless, we recognise the challenge of statistical reliability when breaking down data into subgroups, especially when the sample sizes are small or where data collected during the course of public service delivery (administrative data) does not include the necessary detail. As Dr Richard Norrie of Policy Exchange pointed out, there is also a risk of breaking down data so far that the results are no longer reliable. He explained that “the more you cut into the data, the smaller the number of observations you will have, and so the noisier the estimate will be,” meaning the greater the margin of error will be in the data. Dr Norrie used the example of the adult psychiatric morbidity survey, which showed that 17.7 per cent of black women may have a common mental health condition. However, the sample size was such that this number could actually be as low as 9.6 per cent or as high as 30.5 per cent. Such results, he argued, are of limited use to policy makers.
30.The solution, in our view, is that all data sets should allow for basic disaggregation, which can then be used by the Government in its analysis and published on the Equality Facts and Figures website. Where the data sets are not yet robust enough to be broken down by age, gender, region or other relevant factors, addressing this should be a priority.
31.We note that the Office for National Statistics (ONS) has also launched an Audit into all the data it collects across the nine protected characteristics of the Equality Act 2010. An initial report on ethnicity data was published in April 2018, and it echoed many of the concerns that we heard throughout our inquiry about ethnic classifications, geographical reach and the current limitations on intersectional analysis. We hope that this additional audit will add to the knowledge already acquired by the Government and will further assist in evidence-based policy-making.
32.The commitment of the Prime Minister to ending racial disparities in outcomes and public services, and the central role taken in this exercise by the Cabinet Office, together provide an opportunity for co-ordinated action to improve data collection and data standards on ethnicity and outcomes.
33.We recommend that the Government publish an action plan to improve the consistency and robustness of the data it collects on the basis of ethnicity, to be implemented within 12 months. In the longer term, the Government should ensure that key data can be disaggregated to allow factors such as gender, age, region, socio-economic status and religion and belief to be taken into account alongside race and ethnicity.
31 Institute for Research into Superdiversity, University of Birmingham (), para 8; Housing Learning and Improvement Network (), para 4
32 Equality and Human Rights Commission (), para 27
33 Cabinet Office (), para 55
34 Race Equality Matters ()
35 Professor Steve Strand (); Equality and Human Rights Commission (), para 22; Roma Support Group (), para 15; Joseph Rowntree Foundation (), para 3.1; Clinks and The Young Review (joint submission) (), para 27.
38 Q16 (Dr Weekes-Bernard of the Joseph Rowntree Foundation)
39 Professor Steve Strand ()
40 Friends, Families and Travellers (); Roma Support Group ().
43 Women and Equalities Committee, Second Report of session 2016–2017, , HC89, paras 36–40
44 Cabinet Office, ’, accessed 3 May 2018
45 Cabinet Office, ’, Accessed 3 May 2018
49 Office for National Statistics, , accessed 20 April 2018
50 Office for National Statistics, ’, accessed 20 April 2018
Published: 11 June 2018