Guides · Teaching

Random, ability-based, or mixed grouping: which should you use?

Published: July 25, 2026·Reading time: 8 minutes

You have a class of twenty-eight students and eighteen minutes for a partner task. Do you pair by ability, pair mixed, or shuffle at random? The honest answer is that the decision matters less than the internet often suggests, and the reasons that make it matter are almost never the reasons the research on ability grouping is actually about. This guide separates the two questions — what the research says about long-term ability grouping, and how to choose a grouping method for a single lesson — and gives you a decision table and a worked example for the second one.

The important caveat, up front: nearly all of the research on ability grouping is about tracking. It measures what happens over months or years when students are placed into ability-labelled classes or reading groups. It does not measure what happens when a teacher pairs students for a twenty-minute task on a Tuesday. Those two decisions share a name but not much else, and extrapolating effect sizes from the first to the second is not warranted. This article treats them as different questions.

What the research actually says about ability grouping

The most cited synthesis in this area is Robert Slavin's Ability Grouping and Student Achievement in Elementary Schools: A Best-Evidence Synthesis, alongside his companion secondary-schools synthesis. The pattern is more differentiated than the popular summary — "ability grouping doesn't work" — makes it sound. Four buckets:

• Self-contained ability classes for the whole school day: no meaningful effect. Assigning elementary students to fully separated tracks that stay together across every subject produces achievement outcomes indistinguishable from heterogeneous classes. The most sweeping form of ability grouping is the one with the weakest evidence.
• Cross-grade grouping for a single subject: a gain on the order of two to three months. The Joplin plan and its variants — students regroup across grade levels for reading or mathematics, then return to their heterogeneous homeroom for everything else — showed positive effects across multiple studies. Translated onto a grade-equivalent scale, the median gain lands around a fifth to a third of a school year of extra progress.
• Within-class ability grouping in mathematics: also a gain of roughly two to three months. When one teacher forms small ability-based groups inside a heterogeneous class for math and re-forms them as the topic changes, the effect is positive. The evidence is strongest in mathematics and weaker in other subjects.
• Gifted-versus-struggling asymmetry. Programs for gifted students consistently show gains for that population. Separate classes for struggling students show neither gain nor harm — compressing low-achieving peers together does not accelerate them, but it also does not measurably hurt them.

Two things follow. First, "does ability grouping work" is the wrong question — the grain size matters more than the label. Whole-day tracking is not supported; narrow, subject-specific, flexible groupings are. Second, none of these studies tell you what to do when splitting your third-period class into pairs for a warm-up. The unit of analysis is a year, not a task.

Why the tracking research does not settle the Tuesday question

The mechanisms that make persistent ability tracking harmful — visible labelling of students as "low group," reduced curricular expectations for the bottom track, self-fulfilling identity effects — all require repetition to bite. A classroom observation study, Children Interactions in Ability-based Groups in a Primary Classroom, describes how quickly children notice which table they are on, discuss it with each other, and adjust their behaviour to the label. That effect is real because the groups persist across weeks. A one-off task grouping does not do the same thing.

So for the twenty-minute decision, borrow a different set of criteria. Three questions do most of the work:

1. What does the task actually need? Peer explanation benefits from a mixed pair — one explains, the other asks questions, both learn. Pace-matched drill benefits from similar-ability pairs, because otherwise one partner sits idle. Silent parallel work does not benefit from pairing at all.
2. How much social churn does the class need? A class that has hardened into cliques benefits from random pairings that force students to work with people they would not have chosen. A brand-new class benefits from the same thing for a different reason: it accelerates the class getting to know itself.
3. Is the grouping going to be visible and persistent? A one-off random pairing does not attach a label to anyone. A recurring set of coloured tables where the "blue table" is always the strongest readers does. If you plan to repeat the grouping across weeks, you are back in tracking territory.

Decision table

Which grouping method fits which task, and why:

A worked example

A ninth-grade biology teacher has forty-five minutes and wants students to work in pairs on a punnett-square problem set — twelve problems, ranging from a simple monohybrid cross to a two-trait dihybrid. Three options:

Option A: random pairs. The teacher runs the class list through the Random Pair Generator, projects the pairs, and starts the timer. The advantage is speed and social churn. The risk is that a pair of two struggling students may stall on the harder problems, and a pair of two strong students may finish early and drift off task.

Option B: similar-ability pairs. The teacher pairs students who are currently at a similar place on this specific topic — based on the last exit ticket, not a general sense of "ability." Every pair works at a pace that fits both partners, and the teacher can point struggling pairs at the earlier problems and confident pairs at the harder ones. The risk is that if the pairing repeats week after week, the labelling problem re-emerges. The mitigation is to re-form pairs from a fresh exit ticket every unit.

Option C: deliberately mixed pairs. The teacher pairs a confident student with a less-confident one. This works well if the task frames one as explainer and the other as questioner, and if the confident student is coached not to just do the problem. It works badly if the task is silent and identical for both — then the confident student finishes and the other either copies or freezes.

For this task, similar-ability pairs are the best fit: a punnett problem set is pace-sensitive and each pair can be pointed at problems that stretch them. If the task were instead a single hard cross the whole class attacks together, mixed pairs with an explainer-questioner frame would win. If the goal were a quick end-of-lesson check, random pairs would be fine.

Practical guidance

• Default to random for short tasks. It saves you time, does not attach a label, and periodically forces students out of their usual cluster.
• Sort by current topic, not by "level." A student behind on ratios may be ahead on graphing. Grouping by a general ability label ignores the shape of the class; grouping by the last exit ticket does not.
• Let deliberate groups dissolve. Persistence is what carries the labelling effect. A group that lasts one lesson does not carry the same weight as one that runs all term.
• Do not name groups by ability. If groups persist for a week or two, use neutral names — colours, numbers, animals — not "green group is the strongest readers."
• Watch for pairs that always fail. If a random pairing produces two students who cannot work together productively, quietly re-pair them next time. Random does not mean rigid.

Where the AnchorKite tools fit

The Random Pair Generator takes a class list and produces shuffled pairs in your browser. Nothing is stored, nothing is uploaded, and closing the tab clears the roster — a class list is student data, and treating it that way is the point. For a task that needs groups of three or four instead of pairs, the Random Group Generator does the same thing at a chosen group size. The full set of classroom tools lives at the Teacher Tools hub.

Sources and further reading

• Slavin, R. E. — Ability Grouping and Student Achievement in Elementary Schools: A Best-Evidence Synthesis (ERIC ED348174) — the source for the core findings: self-contained ability classes show no meaningful effect, cross-grade grouping and within-class math grouping show gains on the order of two to three months on a grade-equivalent scale.
• Slavin, R. E. — Achievement Effects of Ability Grouping in Secondary Schools: A Best-Evidence Synthesis (ERIC ED322565, full text PDF) — the companion secondary-schools synthesis, cited above for the pattern that broad tracking produces near-zero effects while narrower forms differ by subject and design.
• Children Interactions in Ability-based Groups in a Primary Classroom (ERIC EJ1287158, full text PDF) — cited above for the classroom-observation finding that children in persistent within-class ability groups notice the tiers, know which one they are in, and adjust their behaviour to the label.
• AnchorKite Random Pair Generator and Random Group Generator — the browser tools this guide supports.

FAQ

More guides at anchorkite.com/guides.