Ability grouping

Why can sorting students by ability raise one group's results while lowering another's?

The case for sorting students by attainment is almost too obvious to argue: a lesson pitched at the middle of a wide class is too fast for some and too slow for others, and narrowing the spread should let a teacher aim properly. If that were the whole mechanism, sorting would help everyone somewhat and harm nobody. Yet the research does not show that. It shows something lopsided — and it is worth working out why an intervention that sounds symmetric produces asymmetric results.

Begin with the one thing sorting definitely does: it reduces the spread of prior attainment inside a room. What follows from that alone? A teacher can pitch closer to where the students actually are. Call this the instructional match, and note that it is the entire theory of benefit. If nothing else changed, both groups would gain, the lower group perhaps most, since it is furthest from where a mixed class is usually pitched.

But something else does change, and this is the step people skip. Sorting does not only regroup students — it creates a label, and the label lands on a class that will now be taught for a year. Ask what a teacher rationally does with a class labelled low: reduce pace, simplify material, cover less ground. Each move is defensible for the class in front of them. Their cumulative effect is that the two groups are no longer covering the same curriculum, so the difference between them at the end is partly a difference in what they were offered.

Now add expectations. Teachers form beliefs about what a class can do, and those beliefs shape how much is demanded of it; students form beliefs about themselves from where they were placed. Neither effect is enormous — the classic expectancy studies are much contested and the honest summary is "real but modest" — but they push in the same direction as the curriculum difference rather than against it.

Then close the loop. Next year's placement is made from this year's attainment. A student taught less has learned less, is measured as lower, and is placed lower again. The sorting is therefore not a one-time measurement of ability but a process that partly manufactures the thing it measures — which is why the gap between tracks widens over time in the systems that track early and hard.

So what does the evidence actually say? Roughly this. The average effect of rigid between-class tracking on achievement is small — close to nothing in several syntheses, including Slavin's reviews of elementary and secondary tracking. The contested part is the distribution: several bodies of work find the top group gains modestly while the lower group loses ground, and cross-country comparisons suggest that systems selecting students very early increase inequality of outcomes without raising average performance. Meanwhile, arrangements that keep the flexibility and drop the label do better: within-class grouping for a specific subject, regrouping for one subject across classes, and above all acceleration for highly able students, which has among the more consistently positive results in education research.

Does that pattern make sense on the mechanism above? It does, and that is the useful part. Those arrangements deliver the instructional match while limiting the curriculum divergence, the permanence, and the label — they are short-term, subject-specific, and reviewed often. Which suggests the benefit and the harm come from different features of the same policy, and can partly be separated.

Three refinements, and one warning.

First, the mechanism is not "grouping is bad." It is that grouping has two consequences with opposite signs — better instructional match, and differentiated curriculum plus expectation effects — and that the balance depends on design choices, not on the sorting itself. Flexible, subject-specific, frequently reviewed grouping keeps more of the first. Rigid, whole-timetable, early tracking accumulates more of the second.

Second, a small average effect is not a small effect. An average over two groups moving in opposite directions can be near zero while both groups are substantially affected. This is why "tracking makes no difference" and "tracking matters a great deal" are both quoted from the same literature, and the second is closer to what the studies actually contain.

Third, the topic is politically loaded, and I should be plain about which parts are contested. The near-zero average is fairly robust. The claim that the top gains is reasonably supported. The size of the loss to the lower group, and how much of it is caused by grouping rather than by pre-existing differences, is genuinely argued over, because the students are not randomly assigned and placement correlates with family background — which makes the equity question, in many countries, the real dispute rather than the achievement question.

• Whether peer effects — what classmates contribute directly to each other's learning — add to this or are largely captured by the teaching differences.
• How school systems that select at age ten differ in outcomes from those that select at sixteen.

Every term the collection defines is gathered in the glossary.