Welcome for government’s ethnic outcome audit — but a warning too
We welcome the government’s interest in keeping a close watch on ethnic outcomes in the UK—it is what we have been doing for the past three years at the Integration Hub, now based at Policy Exchange. But it is important that their new initiative, the Race Disparity Audit, pinpoints real and not imaginary problems and does not fuel division based on selective analysis of statistics.
While the Audit has pinpointed some clear areas of discrimination, there are three pitfalls in particular to avoid:
- Successes as well as failures must be acknowledged—many individuals and even whole groups (Chinese, Indian, some Africans) significantly outperform the white British average in both educational and employment outcomes. And even groups that have been regarded as lagging, such as Pakistanis and Bangladeshis, are catching up. Almost as many young Bangladeshis now go to Russell Group universities as white British.
- Not all departures from proportional representation of minorities are based on discrimination and we need the right benchmarks for minority success. Just over half of minority Britons are first generation immigrants and many of those are poorly educated and do not speak English well so they are not going to be in the running for upper professional jobs. The number of minorities coming out of Russell Group universities 25 years ago was about 9 per cent, this is the sort of proportion of non-whites we should be expecting to see in senior positions.
- We must recognise minority agency and choices. The very title,race “disparity” audit, assumes there is a majority mean that all groups should be heading towards. The lower employment rate for minorities—63 per cent compared with 75.7 per cent for whites —has been used as an example of a discriminatory disparity but a significant part of that will be because of the different choices that, say, Muslim women make about family and work (only about one third work).
Greater availability of data is always to be welcomed but all data must be handled with caution. For instance, the Audit has found that two in three white British are homeowners compared to just two in five for ethnic minorities. This finding has been widely published across the media. It is a good example of how what appears on the surface to be a problem recedes when you dig a bit deeper.
Looking at census microdata (a 5 per cent sample of the 2011 census for England and Wales) we see that the white British are more likely than all other groups to own houses (either outright or with a mortgage). For instance, the odds of a Bangladeshi owning his home are 36 per cent of those of a white British person. Those of a black Caribbean are 39 per cent. But for Indians, there is no difference.
Once we control for potentially relevant variables (being an immigrant, age, occupational class, region, English proficiency) then a completely different picture emerges: all Asian groups (including Chinese) are now more likely than the white British to own homes.
For Bangladeshis the odds are now 170 per cent greater than those of the white British. This reversal is largely accounted for by the fact that Bangladeshis tend to be clustered in London. Note also that the odds for black Caribbeans are still lesser, only not quite as low – the odds of a black Caribbean owning a home are 70 per cent of those of a white Briton.
Table 1. Odds ratios for being a home owner relative to the white British
|No controls||With controls|
|Gypsy or Irish Traveller||0.23||0.44|
|White and Black Caribbean mix||0.27||0.46|
|White and Black African mix||0.22||0.49|
|White and Asian mix||0.48||0.88|
While the white British are more likely to be homeowners, this is largely down to factors other than ethnicity. So much of the rhetoric surrounding the Race Disparity Audit has been couched in terms of “burning injustices”. Yet when we see even Pakistanis and Bangladeshis – two groups over-represented among those living in relative poverty – having a greater chance of becoming homeowners once other factors are taken into account, then it seems fair to say that they have simply done more of the kinds of things one needs to do in order to buy a house. None of this precludes the existence of discrimination or restricted opportunities in the housing market. However, it does point to a high degree of openness which we need to acknowledge too.
The Audit found differences in labour market experience. The employment rate for whites is 75.7 per cent compared to 63.9 per cent for non-white ethnic minorities. Unemployment is at 8 and 4.6 per cent respectively. It is likely that the latter difference is substantially down to discrimination and it cannot be explained away through controlling for the most obvious variables (e.g. age, region). In terms of the employment rate, for both Pakistanis and Bangladeshis the odds of being in employment are around one third of those of whites but controlling for being a Muslim and a woman brings the odds up to just 70 per cent of those for a white person, in line with other minorities.
No analysis based on cross-sectional survey data is ever exhaustive but the stubbornness of the discrepancies leaves ample room for discrimination to be playing a part. We know from ‘CV test’ experiments that people with typically minority-sounding names are less likely to get a call back in job applications. This proves the existence of discrimination although we do not have an accurate measurement of its true extent. There is also a potential source of explanation in the lack of soft skills/cultural knowledge among minority youth (something also seen for white working class youth) and the possible effect of segregation in communities that reduces access to job networks.
Another headline statistic was that more than 9 in 10 head teachers are white British. Indeed, in 2016 93.1 per cent of head teachers were white British and 96.8 per cent were white (including Irish and other white minorities), leaving just 3.2 per cent non-white. By most benchmarks this would be a shortfall but it is important to recognise a couple of things. Firstly, the direction of change is positive. Secondly, while there may be a lack of minority leadership in education, that may very well be because it is concentrated elsewhere in areas such as accountancy, law and medicine. A rush to ensure census level representation may backfire in encouraging the promotion of candidates based on race and not individual potential.
Senior ministers have said that if the disparities found by their Audit cannot be explained then they must be changed. There is a real problem of logic here. If you cannot explain something, then how are you supposed to change it? If a plumber said he could not explain why there was no hot water but nevertheless he could fix it, would you let him just proceed?
In the parliamentary debate that followed the Government’s statement, Labour MP Chris Bryant made the point that correlation is not causation and that without deeper analysis the benefits of policy formation based on these data were questionable. Regression analysis will only get you so far however – it can only isolate correlations and rule out alternative explanations, not establish causation. As we have shown in our critique of the Lammy Review, data on ethnicity can be contradictory and defy easy explanation (e.g. while minority defendants are more likely to go to prison for drug offences, this is not true in cases of sexual or acquisitive violence). There is a risk in seeing every difference as an injustice, misdiagnosing the problem and in doing so letting people down.
So far we have no direct policies arising from the Audit. The Department of Work and Pensions is to identify 20 “hotspots” where interventions can be made and also to offer mentoring to young minority people. The Department for Education is to conduct a review of school exclusions. As the First Secretary of State, Damian Green, pointed out, the data comes first followed by the analysis and then the policy. But, as we have shown, the analysis does not always flow self-evidently from the data. This is where the work must now focus.
Based on logistic regression modelling of Labour Force Survey data from 2012. These data were used instead of more recent data in order to allow for a measurement of English language difficulties to be included in the modelling. Analysis unweighted.