The “damned lies” in the Sewell Report
1 April 2021
The highly divisive Sewell Report has dominated the media in part because it purports to show that children from ethnic minority groups now have better educational outcomes than White British children. This apparent “fact” goes against received wisdom; in the current media frenzy, it is functioning as sufficient proof for the idea that structural racism in the UK is a thing of the past.
Unfortunately this finding emerges from a fundamental misunderstanding about how to analyse data effectively. The more specific claim is that minority children (apart from Black Caribbean children) outperform White British children once socio-economic status (SES) is controlled. The question is whether it makes sense to control for SES in these circumstances. If we want to know the impact of race on attainment, then controlling for SES is an error.
In general, if we want to know the impact of one variable on another (call it X→Y), we need controls that are causally prior to Y and X. It is useful to think about controls using the letter W — because W comes before X. So, to restate the key point, we need controls where the relationship is W→X. If we use W as a control when W is affected by X (X→W), we cause harm to our analysis. (To keep things straight: W should not come after X.)
Here’s a simple example that shows the potential for harm. Suppose we want to know the impact of unemployment on happiness. Some researchers would control for income for this purpose — on the basis that income is “another determinant” of happiness. The question is, what is the relationship between unemployment (X) and income (W)? The best answer is that unemployment affects people’s income: get sacked and your income is reduced. So, W→X in this context, making income an inappropriate control.
Why is it an inappropriate control? Statistical control means looking at the main relationship (X→Y) while holding W constant. So, controlling for income here would mean looking at happiness differences (for employed vs. unemployed people) while holding income constant. Why should we hold income constant in this context? Loss of income is one of the reasons unemployment reduces people’s happiness. If we control it (keeping income constant), we “control away” part of the impact of unemployment on happiness. We end up severely underestimating the magnitude of that effect.
Exactly the same consequence follows from controlling SES when considering the impact of race on educational outcomes. If structural racism still exists in the UK (I say “if”, but really: it does), then it operates in part via SES: on average, people from most minority groups will not have the same opportunities as White British people to get good jobs, with incomes that enable them to live in areas with good schools. To oversimplify a bit, race affects SES, and SES (of parents) has an impact educational outcomes for children. SES is an inappropriate control when race is our X: the relationship is X→W.
The research cited in the Sewell Report leads us astray in this regard. (To be fair, a good proportion of quantitative social scientists make similar errors in their work.) Controlling for SES means “controlling away” part of the impact of race on educational outcomes. In fact, virtually any potential control variable in this context would be inappropriate. To evaluate a potential control, we can ask: what are the “determinants” of someone’s race? Answering that question perhaps requires a good understanding of race — but in practical terms, when it comes to selecting control variables the answer is “none”.
To see the impact of race on attainment, then, we need to look at unadjusted data — a straightforward comparison of the averages within each racial/ethnic group. In Table 2 of the supplementary research, we see lower average outcomes not just among Black Caribbean children — attainment is lower among children from Pakistani origins as well. Children from Black African families are on par with White British children.
Children from some groups are indeed succeeding in greater measure (Indian and Bangladeshi children, in particular). But this is a decidedly mixed picture. It is simply not true that White British children are being outpaced by minority children in general.
The authors of the Sewell report, selected by the Government no doubt for entirely political reasons, might have cherry-picked findings that helped them reach their desired conclusions. It will be difficult to undo the damage wrought by the media. But social scientists and interested observers should gain an understanding about why this particular finding was misleading, before concluding that structural racism is a thing of the past.
David Bartram is Associate Professor in Sociology at the University of Leicester. He is co-editor of the Journal of Happiness Studies and President of RC31, the section on the Sociology of Migration of the International Sociological Association. He has published two books: Key Concepts in Migration (Sage Publications) and International Labor Migration: Foreign Workers and Public Policy (Palgrave). He has held a grant (with colleagues at Leicester) from the UK Economic and Social Research Council to investigate the UK ‘citizenship process’, as well as grants from the Social Science Research Council (in the USA) and from the Hebrew University of Jerusalem. He gained a PhD from the University of Wisconsin–Madison and a BA from Kenyon College.
