Value Added v2.2 candidates

Pupils’ ethnicity and neighbourhood deprivation, tested for every group of pupils and every year: what they add, what they change, and the case for and against.

Candidates, not changes

Value Added v2.1 is what this site publishes. This page tests two candidate terms for a v2.2, the way the v2.1 candidates were tested: on identical school-years, for every group, with each year held out in turn. Where the terms came from is set out in the neighbourhood covariates experiment.

In short

Neighbourhood deprivation helps every group, in every year. It improves predictions out of sample for all five groups and takes about 7 to 9% off the local authority variance (22% for disadvantaged pupils). It changes little else. Recommended for v2.2.
Pupils’ ethnicity matters almost only for disadvantaged pupils. There it is a large gain (AIC -388). For non-disadvantaged pupils it is zero. A third of schools would change fifth on disadvantaged value added, with London losing most. Whether to adjust for it is a question of what value added is for. Recommended beside v2.2, not in it, as v2.1 publishes the kind of absence.

The candidates

Term	What it is	Source	One standard deviation
Pupils’ ethnicity	The school’s share of White British pupils, each January	School census, 2022 to 2025	29 percentage points
Neighbourhood deprivation	Share of households deprived in two or more of four dimensions (employment, education, health, housing), over the 150 Output Areas nearest the school	Census 2021, TS011	5.2 percentage points

v2.1 has neither. It has English as an additional language for pupils. For the neighbourhood it has adult qualifications only. Ethnicity is matched to 99.9% of school-years, by year.

Fit and prediction

Table 1: Each candidate against v2.1, by group: the effect of one standard deviation (% of Attainment 8), the change in AIC (negative is better), and the change in out-of-sample error (each year predicted from the other three, averaged).

Group	Candidate	Ethnicity	Deprivation	AIC change	Out-of-sample error
All pupils	+ pupils' ethnicity	-1.7%		-87	-0.26%
	+ neighbourhood deprivation		+1.0%	-45	-0.38%
	+ both	-1.5%	+0.6%	-102	-0.46%
Disadvantaged	+ pupils' ethnicity	-5.7%		-388	-1.13%
	+ neighbourhood deprivation		+2.7%	-119	-0.74%
	+ both	-5.2%	+1.3%	-414	-1.39%
Non-disadvantaged	+ pupils' ethnicity	-0.4%		-2	-0.01%
	+ neighbourhood deprivation		+1.1%	-48	-0.41%
	+ both	+0.0%	+1.1%	-46	-0.41%
Girls	+ pupils' ethnicity	-2.0%		-98	-0.30%
	+ neighbourhood deprivation		+0.9%	-30	-0.30%
	+ both	-1.8%	+0.5%	-103	-0.41%
Boys	+ pupils' ethnicity	-1.4%		-36	-0.12%
	+ neighbourhood deprivation		+1.2%	-43	-0.37%
	+ both	-1.0%	+1.0%	-59	-0.37%

Both terms improve fit and prediction for every group, with one exception: ethnicity does nothing for non-disadvantaged pupils (-0.4%, AIC -2). For disadvantaged pupils, a school with one standard deviation more White British pupils gets 5.7% lower Attainment 8 for its disadvantaged pupils, all else in v2.1 equal. That matches what is well known at pupil level: disadvantaged White British pupils do less well at GCSE than disadvantaged pupils of most other ethnic groups. Neighbourhood deprivation goes the other way, about +1% per standard deviation for most groups. At the same share of graduates, a school among more deprived households does a little better than v2.1 expects.

Year by year

Figure 1: Each term’s effect fitted on each year alone (both terms together, with v2.1’s terms and region/local authority effects), % of Attainment 8 per standard deviation, with 95% intervals.

Both are steady. Ethnicity is between −4% and −6.5% for disadvantaged pupils every year and near zero for non-disadvantaged pupils every year. Deprivation is positive in every year for every group, largest in 2024/25.

What would move

Table 2: What each candidate changes against v2.1: the correlation of schools’ value added, the share that change fifth, the average change in a school’s value added (Attainment 8 points), the correlation of local authority effects, the change in local authority variance, and Brighton and Hove.

Group	Candidate	Schools: r	Change fifth	Average change	Authorities: r	Authority variance	Brighton and Hove
All pupils	+ pupils' ethnicity	0.989	17%	0.33	0.983	-1%	+4.1% (5th)
	+ neighbourhood deprivation	0.994	12%	0.24	0.987	-7%	+3.1% (11th)
	+ both	0.985	19%	0.39	0.974	-5%	+3.6% (6th)
Disadvantaged	+ pupils' ethnicity	0.937	33%	0.94	0.915	-17%	+4.5% (11th)
	+ neighbourhood deprivation	0.981	19%	0.52	0.964	-22%	+2.0% (33rd)
	+ both	0.926	35%	1.01	0.892	-24%	+3.5% (16th)
Non-disadvantaged	+ pupils' ethnicity	0.999	3%	0.08	0.999	+0%	+3.8% (3rd)
	+ neighbourhood deprivation	0.994	13%	0.27	0.985	-9%	+3.0% (9th)
	+ both	0.994	12%	0.27	0.985	-9%	+3.0% (9th)
Girls	+ pupils' ethnicity	0.986	19%	0.39	0.978	-2%	+3.2% (9th)
	+ neighbourhood deprivation	0.995	10%	0.23	0.990	-7%	+2.3% (19th)
	+ both	0.983	20%	0.42	0.972	-5%	+2.9% (14th)
Boys	+ pupils' ethnicity	0.994	11%	0.25	0.991	-1%	+4.8% (4th)
	+ neighbourhood deprivation	0.994	11%	0.27	0.986	-8%	+3.8% (6th)
	+ both	0.990	14%	0.34	0.979	-6%	+4.1% (6th)

Neighbourhood deprivation is a light touch. Schools’ value added correlates at 0.994 with v2.1 for all pupils, and the average school moves by 0.24 points. Brighton and Hove goes from +3.8% to +3.1% for all pupils.

Ethnicity is not a light touch for disadvantaged pupils. 33% of schools change fifth, and the average school’s disadvantaged value added moves by 0.9 points.

Who would gain and who would lose

Figure 2: The average change in schools’ value added (Attainment 8 points) with both terms added, by fifth of schools on their share of White British pupils, for each group.

With both terms, disadvantaged value added falls by 2.3 points on average in the fifth of schools with the fewest White British pupils (about 13%), and rises by up to 0.7 in the middle and upper fifths. By region, London’s schools lose 2.0 points of disadvantaged value added on average; every other region gains a little, except the West Midlands (-0.4).

The case for and against ethnicity

For. Value added is meant to compare schools fairly, on what they do rather than whom they teach. Disadvantaged White British pupils do less well wherever they are taught. A school that serves many of them is judged against schools whose disadvantaged pupils were always more likely to do well. Adjusting for it is what DfE’s contextual value added did until 2011. It is also what the evidence here supports, for this group, in every year.

Against. An expectation that varies by ethnicity builds a lower expectation into the measure for some pupils. That was the main criticism of contextual value added and part of why it was dropped. It would also take out of view a real gap that policy should see: the low attainment of disadvantaged White British pupils. And the term is a school share, not each pupil’s own ethnicity, so it adjusts schools, not pupils.

What we would do. Publish it beside v2.2 rather than in it, as v2.1 publishes value added allowing for the kind of absence. The headline measure stays free of ethnicity. Anyone asking “how does this school do with its particular disadvantaged pupils?” can see the answer next to it.