Not long ago, I was coaching a teacher whose math students were working in groups of three. I asked her how she decides to group students in pairs or triads for collaboration. Her answer was immediate: “Because I hear three is better.”
It’s a common response and an honest one. The idea that triads are the default unit for collaborative learning has spread through professional development circles the way a lot of instructional lore does: repeated often enough that it starts to feel like fact. Some of that comes from Peter Liljedahl’s Building Thinking Classrooms, which recommends groups of three for good reason. But the reason is specific, and it doesn’t apply to every phase of learning. That was the piece she was missing. When I shared what the research actually says about pairs and retention, her posture shifted. By the end of the conversation, she agreed to try collaboration in pairs.
That exchange is what this article is about. Not whether triads are bad, because they’re not, but whether the group size you’re using matches what you’re trying to accomplish, to match the strategy to where you are in the learning cycle.
Why Collaboration Group Size Is a Learning Variable, Not a Preference
Most conversations about collaborative learning focus on the grouping of students: pairs, triads, and larger groups (4–5). Far fewer conversations address the more fundamental question: what cognitive work is each student actually doing inside the group?
Group size is not neutral. It shapes how much individual participation is required and influences whether a student can follow along passively or must actively contribute to keep the discussion moving forward. When teachers treat group size as a logistical default rather than an instructional decision, they often build structures that feel collaborative but function as cover for passive participation.
On two points, the research on learning is unambiguous: 1. Students learn what they process, and 2. they remember what they retrieve. Both depend on students generating the thinking, not receiving it. Any group structure that allows a student to receive information without generating it is, at best, a partial learning experience. The question of pairs versus triads is ultimately a question about how reliably each structure forces every student to be the one generating.
The Science Behind the Structure
Two bodies of research are worth grounding this in before getting practical.
The first is retrieval practice. Decades of cognitive science research, including foundational work by Roediger and Karpicke, have established that the act of retrieving information from memory is itself a powerful learning event. It is not just a way to check what students know. The retrieval attempt strengthens the memory trace, making future recall faster and more reliable. This is sometimes called the "testing effect," though the mechanism applies to any low-stakes retrieval activity, not just formal tests. Think-Pair-Share works, because the "pair" phase forces retrieval. The pair structure is what makes that retrieval unavoidable.
The second is elaborative interrogation. When students are asked to explain why something is true, or to connect a new idea to something they already understand, they engage in a form of processing that goes beyond surface recognition. They are constructing meaning, not just receiving it. This generative effort is what makes an idea stick. Assigning roles in a pair, where one student explains the cause and the other predicts the effect, is not a management technique. It is a memory technique. The role assignment forces elaborative processing from both students.
Triads, by contrast, allow the cognitive work to be distributed in ways that can undermine individual encoding. A student who watches two peers work through a problem, contributes occasionally, and follows the logic to a correct conclusion has experienced something. But they have not necessarily done enough individual retrieval or elaboration to own the learning. Tomorrow, that understanding may be gone.
Not all research frames group size as a decisive variable. A 2015 study by Enu, Asominiwa, and Obeng found no statistically significant difference in mathematics achievement among students working in groups of three, four, or five. Then the question is "what cognitive behaviors does the structure require of every student?" A triad and a pair can both produce learning. The difference is in how reliably each one makes retrieval, verbalization, and individual accountability unavoidable. A pair doesn't improve retention because it has fewer people. It improves retention because it has fewer places to hide.
The Triad: Built for Discovery
None of this is an argument against triads. Liljedahl recommends groups of three for the initial problem-solving phase, and the reasons are well-supported. At vertical non-permanent surfaces, a group of three creates a social dynamic that pairs often cannot: when two students stall, a third frequently provides the divergent thought that breaks the logjam. Liljedahl's visibly random groups use this structure deliberately to disrupt the "mimicking" habit. In a randomly assigned triad, students are less likely to defer to the student they perceive as more capable, which builds the collective persistence that genuine thinking requires.
This is the right tool for the start of a learning cycle. Triads generate ideas and reduce the anxiety of not knowing. When a student is unsure, the triad provides enough social cushion to take intellectual risks. That matters, especially early in a unit or with unfamiliar content.
Research helps explain why three works. Laughlin, Hatch, Silver, and Boh (2006) gave groups of two, three, four, and five complex rule-finding problems. They measured efficiency not by clock time but by trials to solution, meaning how many hypotheses and tests a group needed to reach the correct answer. Groups of three or more outperformed the strongest individual from a same-sized set of students working alone. Pairs only matched their stronger member. Going beyond three added nothing. The authors concluded that three members are "necessary and sufficient" for a group to outperform its best individual.
But notice what was measured. The study tracked whether the group found the answer, not whether each member could recall or explain it later. The triad is the most efficient structure for collective discovery. That is a different claim from saying it is the best structure for individual learning, and the difference is the whole point.
The risk is what comes next. In a group of three, it is entirely possible for one student to understand a solution in the moment without performing the cognitive work required to remember it tomorrow. They followed along. They nodded. They may even have contributed. But they were not fully on the hook — and being on the hook is what learning requires.
The Dyad: Built for Retention
When the goal shifts from discovery to retention, the pair is the stronger structure. The reason is simple accountability: in a dyad, if one student disengages, the conversation stops. There is no third member to carry the load. That accountability sets up retrieval practice, which research consistently identifies as the most powerful driver of long-term memory. The act of pulling information out of the brain, rather than simply recognizing it when presented, is what strengthens it. But the pair only creates accountability. The task still has to require retrieval.
Airtime matters here too. In a pair, each student receives roughly 50% of the verbal exchange. In a triad, that drops to approximately 33%, meaning dyads provide each student more speaking time than triads do. And that assumes the talk is split evenly, which in a triad it rarely is. Since the student doing the talking does more of the learning, the math favors the pair for individual retention.
The dyad's case rests on retrieval, and the evidence there is among the strongest in education research. A 2021 meta-analysis by Yang and colleagues pooled 222 classroom studies of more than 48,000 students. Quizzing had a medium effect of achievment. A separate review of 50 classroom experiments found the benefit held across grade levels, with the largest effects in middle school (Agarwal, Nunes, & Blunt, 2021).
Every one of those studies tracked what individual students could retrieve on their own. The benefit belongs to the student who does the retrieving. When a student verbally explains a concept to a partner without notes, they aren't just chatting; they are practicing retrieval, the same cognitive act that makes quizzing work.
Groups complicate that. When people try to recall together, they interrupt each other's retrieval. Researchers call this collaborative inhibition, and a meta-analysis of 64 studies found it grows as groups get larger (Marion & Thorley, 2016). Group recall still has value. The same review found that people who recalled together remembered more on their own later. But the smaller the group, the less each student's retrieval gets disrupted. The pair is the smallest group that still gives every student a partner to check their answer.
By assigning partner roles, the Numbered Heads strategy makes this concrete. When pairs are assigned roles, such as Partner 1 explains the cause and Partner 2 predicts the effect, students have to construct the answer themselves constructing it. That effortful, generative processing is what elaborative interrogation looks like in practice, and it is a long way from the passive understanding of a student who is along for the ride in a triad.
What This Looks Like in the Classroom
Consider a middle school science class working through the concept of natural selection. A productive lesson sequence might look like this:
The teacher opens with a discovery task at the whiteboards. Random triads work through a scenario: a population of beetles, two colors, one predator. What happens over generations? The triads argue, sketch, revise. Some groups reach the right conclusion quickly. Others take longer. A few stall and restart. This is productive struggle, and the triad structure supports it well. The goal here is engagement and initial sense-making. Mastery comes later.
After the whole-class debrief, the teacher shifts the structure. Students move into assigned pairs. Partner 1 explains the mechanism of natural selection in their own words without notes. Partner 2 listens, then identifies what was accurate, what was missing, and what they would add. Then they switch. Every student retrieves. Every student generates, and no one is just watching.
That second phase is where the learning is consolidated.
This sequence is not limited to science. An English class might use triads to analyze an ambiguous passage together, then collapse into pairs for each student to independently argue a textual interpretation. A math class might use triads to work through a novel problem type, then shift to pairs for guided practice where roles alternate between solving and explaining.
What Leaders and Coaches Should Look For
For instructional coaches and principals, the dyad-triad distinction offers a practical observation lens. The question is not simply "Are students working in groups?" It is "What is each student in that group being required to produce?"
During a classroom observation, look for the passenger. In a triad, identify the student who is oriented toward the group but not generating. Are they writing? Talking? Being asked a direct question? If the answer is consistently no, the structure may be providing cover rather than accountability.
Listen for the ratio of teacher talk to student talk, but go further: listen for the ratio of one student's talk to another's. In a functional dyad engaged in retrieval practice, both students should be speaking in roughly equal measure. If one student dominates consistently, the pair has drifted into a tutoring dynamic, which has its own value, but is a different thing.
In post-observation conversations, coaches can ask teachers a reflection question: "At that point in the lesson, was the goal thinking or remembering?" That question often opens a productive conversation about whether the group structure matched the instructional intention. Most teachers have a strong instinct for when to build thinking. The dyad-for-retention move is less intuitive and often underdeveloped.
Common Objections
"My students work better in threes. Pairs create conflict."
Some conflict in a pair is social, and that needs to be managed. But much of it is a signal that retrieval is actually happening. When two students disagree about what the answer is, that disagreement is where thinking happens. The discomfort of being wrong, and the effort of figuring out who is wrong and why, are productive. Pairs that never conflict can be pairs where one student is simply agreeing with the other. That is social harmony, not learning.
"Random grouping works for Liljedahl. Why would I ever assign pairs intentionally?"
Randomness serves a specific purpose in Building Thinking Classrooms: it breaks fixed social hierarchies that prevent thinking. That purpose is real and important. Liljedahl cautions against strategic grouping, but his concern is the thinking phase. Once the goal shifts to retention, intentional pairing can be more powerful. A teacher who knows that a particular student needs to be the one doing the summarizing, not watching someone else summarize, can build in that accountability on purpose. Both have a place.
"This adds another transition to manage."
It does. The question is whether the cognitive payoff justifies the management cost. In most cases, it does, and the transition cost decreases as students internalize the routine. When a class learns that the whiteboard phase always ends with a pair phase, the shift becomes predictable and fast. Over time, the routine tells students that this is where we make sure the learning sticks.
Does Composition Matter?
There is a legitimate tension between two views on how to form pairs.
The Building Thinking Classrooms case for randomness: strategic groupings create fixed expectations, even when the groups change. A student who is routinely paired with a high performer learns to wait rather than retrieve. Random pairs disrupt that pattern and prevent the social stratification that kills thinking culture.
The Learning-Focused case for intentionality: strategic pairing allows for targeted depth.The students in a high-middle pair push each other toward abstraction. A middle-low pair makes it more likely that the lower-performing student is the one doing the summarizing, not observing someone else do it. When a teacher knows a student has a specific gap in understanding, an intentional pairing can put that student in the role that requires them to address it directly.
Both approaches have merit. The deciding factor should be the specific instructional goal and what you know about the students in front of you. Randomness is a strong default. Intentionality is the right override when you have information that randomness would waste.
The Practical Takeaway
Use the triad to drive discovery and build thinking culture, especially early in a lesson or unit. Then, when it is time for students to own the content (to practice, process, and prepare to remember), collapse those groups into pairs. Put students on the hook with each other. Make the retrieval non-negotiable.
The research on how memory works is clear: students remember what they are required to retrieve. And they retrieve most reliably when there is nowhere to hide. The dyad provides that condition in a way the triad cannot consistently guarantee.
This is not a critique of Building Thinking Classrooms. It is a completion of it. Liljedahl's framework addresses consolidation, but it does its most distinctive work at the front end of learning: generating ideas, building persistence, making thinking visible. The pair structure adds a specific tool for the back end: making sure that what was discovered is actually remembered.
The triad sparks the thinking. The dyad does the remembering.
Works Cited
- Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students' learning with effective learning techniques: Promising directions from cognitive and educational psychology. Psychological Science in the Public Interest, 14(1), 4–58.
- Enu, J., Asominiwa, L., & Obeng, P. (2015). Effects of group size on students' mathematics achievement in small group settings. British Journal of Education, 3(4), 58–64.
- Kagan, S. (1994). Cooperative learning. Kagan Publishing.
- Latané, B., Williams, K., & Harkins, S. (1979). Many hands make light the work: The causes and consequences of social loafing. Journal of Personality and Social Psychology, 37(6), 822–832.
- Liljedahl, P. (2021). Building thinking classrooms in mathematics, grades K–12: 14 teaching practices for enhancing learning. Corwin Press.
- Lyman, F. (1981). The responsive classroom discussion. In A. S. Anderson (Ed.), Mainstreaming digest. University of Maryland College of Education.
- Palincsar, A. S., & Brown, A. L. (1984). Reciprocal teaching of comprehension-fostering and comprehension-monitoring activities. Cognition and Instruction, 1(2), 117–175.
- Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.
