Skip to content

Collaborative Learning

Pattern · All patterns
Evidence · 55 claims (23 for, 29 mixed, 3 against) · 45 studies (13 causal, 7 review, 6 associational, 6 theoretical, 5 quant-synthesis, 5 qualitative, 3 design), q1–q4 · 10 of 45 report an effect size · 53 claims rest on one study

Description and scope

This is a reusable policy for joint work: learners work on one task together and have to take up each other's reasoning. The policy treats what each member can do afterwards as the target. A shared product, balanced talk or a satisfied group is not. It instantiates the collaborative learning principle, which states when joint work of any structure is expected to raise individually measured understanding over working alone. The study configurations below are evidence. The response-dependent policy is an untested design proposal. Use it to gather better evidence about each member, not to certify that a group "collaborated".

Cooperative learning is the more structured sibling. It fixes positive interdependence and individual accountability in advance, often with a group reward computed from individual learning, and its evidence is mostly school and undergraduate syntheses of those structures. This page owns the broader and lighter-weight case: pairs or small groups on a task that needs their joint reasoning, with whatever structure the task and the learners turn out to need. When an individual check shows free-riding, domination or division of labour, moving to the cooperative configuration is one of this policy's branches. Think-pair-share, peer instruction and the jigsaw method are named configurations with their own pages. None of them defines this pattern.

Inputs: establish the design brief

Field Record or elicit
Learner and starting observation Each member's individual response to the target task before grouping, with reasoning, errors and confidence; age band; language and access needs; any known status, friendship or prior-group history. Preserve unknowns rather than labelling members "strong" or "weak".
Context and activities Setting (classroom, online, workplace), group size and how groups are formed, synchronous or asynchronous, time available, the task and whether one member could complete it alone, any roles, prompts or scripts, and how the teacher monitors. Record what was actually delivered.
Learner-valued goal Ask each member what they want from working with others and why: being heard, hearing other views, finishing the product, working with friends. Record disagreement with the designer objective rather than assuming agreement.
Designer objective What each member should be able to do afterwards and without the group, on what kind of knowledge (concept, principle, complex-skill, or collaboration itself as an outcome), and which aids are permitted.
Outcome An individual instrument aligned with the objective, a locally chosen criterion and the horizon: immediate, delayed-retention, or transfer to independent work. A group product score, the group's report of its own success and satisfaction are separate outcomes, each with its own measure, and none stands in for the individual one.

A request for "more group work" or "more engagement" does not specify these inputs. Ask for them before choosing a group size, a structure or a predicted gain.

Sequence and conditional policy

  1. Elicit individually first. Take a short individual response on the target before grouping. It is the starting observation and the comparison for every later individual check. It is also a learning exposure, so record it as one. The order of the individual and joint phases is evidenced only in single studies that disagree (see Order of individual and joint work below), so this order is a default, not a finding.
  2. Choose a task that needs joint reasoning. Prefer a task with something to explain, compare or reconcile: building a representation from scratch rather than filling in or sorting a given one. If one member could do it alone, expect parallel work or one member doing it, and either change the task or treat the session as individual practice. In one experiment, pairs who built a whole concept map themselves talked in a deeper, more reasoned way than pairs completing or sorting a given map, whose talk stayed superficial (self-constructed maps trigger deeper exploratory talk [+M]), and in three group discussions a shared multidisciplinary modelling task drew out all four processes of interdisciplinary reasoning (modelling tasks elicit interdisciplinary processes [+M], qualitative, no comparison). Both are evidence about the talk; the rule that a divisible task invites divided work remains untested here.
  3. Start with light structure and record it. Give the group a shared aim and a prompt that requires each member's view to be stated before agreement. Add roles, scripts or talk supports where members are young, new to group work or the observation calls for it. Record the configuration actually delivered, because "collaboration" names many different treatments.
  4. Observe the process, not only the product. Note who proposed each step, whether disagreements were stated and settled, and whether the work was divided. These observations concern the group. They do not show what any member understands.
  5. Check each member individually. Use a new task aligned with the objective, under the agreed aids, straight after the group work. Then use the table below to choose the next activity provisionally.
  6. Reobserve at the stated horizon. Repeat an individual check later on matched and changed wording. Retain unresolved explanations. If individual results disagree with the expectation, revise the configuration. Do not redefine the failed session as "not real collaboration" after the fact. A member may accept or reject the group's construction only after the session, through their own reflection on the differences that came up (acceptance can follow the session [~W], the authors' framing account, not a tested result), so a straight-after check alone can understate or overstate what each member keeps.

Dose and progression across sessions (a labelled default, untested as a whole). In one lesson: about 3–5 minutes for the individual response, 15–25 minutes of joint work on one task, 5 minutes for the individual check. Across sessions:

  • Early sessions (first two or three). Keep groups the same, give roles and a talk prompt, and make the first joint task one of the same kind as the target task, so the group has practised working together on it before the target. Groups that had already worked together on similar tasks outperformed inexperienced groups on individual tests, including a delayed one a week later (experienced groups did better [+M]; and more efficiently at the delayed test [+M]; 240 secondary-school students in groups of three).
  • Middle sessions. Drop the roles and keep the prompt when individual checks show each member can reconstruct the group's reasoning; restore them for a group whose checks fall. Every few sessions, give the group a short structured reflection on how it worked (see the row on collaboration as the objective).
  • Late sessions. Groups set their own way of working; the individual check stays. Move on when each member passes the delayed individual check on changed wording, not when the group product is good.
  • Delayed check. One to two weeks after the unit, on matched and changed wording.

The minutes, the session counts and the fading rule are proposals; only the value of prior joint experience rests on a study.

Observation Discriminating probe Proposed action, with its uncertainty
The group's product or shared explanation is correct, but one or more members fail the individual check Ask each member to explain or redo a part they did not propose, on matched and changed wording; check the record of who proposed each step. If members cannot reconstruct others' reasoning, require each member to state and justify the shared answer in their own words before the group moves on, or move to the cooperative configuration with individual accountability. If they can explain it but not transfer it, give individual practice. Neither branch is a validated diagnosis.
Talk is balanced and on task, but individual gains are uneven Compare each member's before and after responses; probe prerequisite knowledge individually; note who explained and who accepted explanations. If a member lacks a prerequisite, give targeted instruction before the next joint task rather than more discussion. If a member only accepted others' explanations, give that member an explaining role or a solo attempt between rounds. Balanced participation alone is not evidence of each member's learning. Unequal participation goes with lower satisfaction with the group (unequal participation, lower satisfaction [~M], rho −.61 and −.72 for two tasks in one online university course, correlational at group level), so evening it out may matter for how members feel about the group even where it does not change what they learn.
Members agree quickly, on a shared misconception or on the first idea offered Compare individual answers before and after discussion; present a contrasting case or a disconfirming observation. Ask the group to settle the contrast and each member to give a revised, individually justified answer. Watch whether a confident member carries the group, and whether a member's doubting question is answered or passed over: in one recorded group a member's sceptical question was ignored and the group went ahead on partial agreement (premature shared understanding [~M], one episode). Agreement is not evidence of correctness.
The group reports a satisfying, successful collaboration, but individual results are weak Set an individual knowledge check beside the self-report; ask each member separately what problems the group met and how it dealt with them. Treat the self-report as an observation of the group's process only. Keep the task, but change the individual checks or the prompt that requires reasoning to be stated. Expect members to rate the task easier and their understanding higher after joint work whatever they learned: upper-elementary students did so after collaborative physics tasks (d ≈ 0.37 for each; difficulty rated lower [~M]; understanding rated higher [~M]; before–after ratings with no comparison group).
Work is divided ("you do part 1, I'll do part 2") or done by one member Ask whether one member could have done the task alone; ask each member about the parts they did not do. Make the task less divisible, or require integration (each member must use another's part). If the task cannot be changed, treat it as individual work with peer checking. Untested.

Choosing configurations from evidence

  • Joint work against none, at programme level. Collaborative learning improves outcomes [+M]. A meta-analysis of 39 North American undergraduate STEM classroom studies found that small-group work raised achievement over instruction without group work (d = 0.51). Effects were smaller where the investigator did not teach the course (0.41) and in single-group pre/post designs (0.30), and a non-significant 0.21 at two-year colleges. A second meta-analysis, of 148 studies of early adolescents, associates cooperative goal structures with higher achievement than individualistic or competitive ones, giving direction only, from the abstract. Partly checked (1 of 2 entries pass, abstract). The undergraduate comparison is group work against no group work, not one structure against another, so it supports trying joint work. It does not say which configuration to choose, and it is a programme-level average, not a forecast for one group.
  • Joint against solo work on the same task, measured individually: one gain, one null. In a simulation-based inquiry experiment, 61 dyads scored higher than 95 individuals on the individual post-test overall (collaborative inquiry beat individual [+M], F(1,201) = 17.33, no effect size printed) and on intuitive conceptual knowledge (collaboration raised intuitive knowledge [+M]), while spending less time on task; the learners' age is not given on the claim pages. In a randomised productive-failure experiment with 137 tenth graders, students who generated solutions in groups of three before instruction did not differ in later conceptual knowledge from students who did so alone (no difference in conceptual knowledge [~M], c = 0.57, CI −0.39 to 1.53), and neither the number nor the quality of the solutions the groups produced carried any effect to knowledge (solution quantity did not mediate [~M]; solution quality did not mediate [~M]). A non-significant difference is not equivalence. Together they say the default can gain over solo work but need not, and that a better group product does not show better individual learning, which is why step 5 checks each member.
  • Order of individual and joint work. In a randomised experiment with 226 undergraduates building concept maps from a text, working alone first and then jointly gave better delayed recall than the joint-then-individual order or working alone throughout (η² = 0.12), and better comprehension than working alone (individual-then-collaborative structure gave better recall [+M]). A systematic review of virtual and remote laboratories reports one study the other way round, in which collaborating first and then doing the activity alone gave better lab grades than the reverse (collaborate first, then individually [~M], second-hand, no effect size). In a fractions tutoring system for 382 fourth and fifth graders, a condition combining individual and collaborative work gained more than either alone, but only in Grade 4 (combined condition, Grade 4 only [~M]). The page's individual → joint → individual sequence contains both orders; the wiki has no study that tests this whole sequence.
  • Affective outcomes are separate outcomes, and they move. In the same fractions study, students who collaborated reported higher situational interest (collaborators reported higher interest [+M], no effect size printed), and undergraduates' self-efficacy for quantitative biology tasks rose over group-work assignments (r = 0.58 and 0.51; self-efficacy rose after group work [~M], before–after with no comparison group). Record these beside the individual check when interest or confidence is part of the aim; neither stands in for it.
  • The product and the individual are different outcomes. Collaborative writing improves text quality [+M]. Across seven comparisons in Grades 4–12, students who planned, drafted or revised together wrote better texts than students writing alone (weighted mean ES = 0.75, 95% CI 0.54–0.97), but the outcome was the jointly produced text. One small randomised study of 28 children aged 10–11, all trained in a paired-writing system with metacognitive prompts, found that those who wrote in structured pairs for six weeks gained more in the quality of their later individual writing than those who wrote alone (no effect size in the abstract). Partly checked (1 of 2 entries pass, abstract). It supports checking individual transfer as a separate step, as step 5 does. It does not show that an unstructured pair would transfer.
  • The group's talk overstates each member's later understanding. Co-construction needs personal construction [~M]. In seven case studies of student dyads in computer-based force-and-motion lessons, conceptual change visible in the pairs' talk often did not hold when students were probed later. The authors report that it was stable when shared construction was accompanied by each student's personal sense-making. The design is qualitative with no comparison condition, and the learners' ages are not reported on the claim page. Checked by the judge (full text). It is the reason for the delayed individual check in step 6. It does not say how to bring personal construction about.
  • Balanced participation is not enough. Equal, mutual engagement did not ensure change [~M]. In the same study, in two dyads whose on-task engagement was high in equality and mutuality, one student changed substantially and the other did not. Two case dyads, no effect size, and not yet checked against its source. It supports row 2 of the table. It does not estimate how often this happens.
  • Added structure need not change achievement online. Structuring interdependence in asynchronous online groups [~M]. In a randomised experiment, triads given role, reward, role-plus-reward or no structured interdependence did not differ significantly on an individual posttest (about 71% overall; no effect size printed). Not yet checked against its source. The participants were 280 undergraduate business majors. A non-significant difference does not establish equivalence, and there is no individual-work comparison. It warns against assuming that a role or reward scheme will repair an online group; check the individual outcome either way.
  • How a group feels it worked is not what its members learned. Aligned problem perceptions in online groups [~M]. In a cross-sectional path-modelling study of 311 pre-service teachers in 112 online groups on a single case-analysis task, groups whose members saw their collaboration problems alike reported higher satisfaction, perceived regulation success and subjective learning gain (β = .31–.57). That alignment was unrelated to objectively measured knowledge. There was no pretest, and the associated outcomes are all self-reports. Not yet checked against its source. It supports row 4 of the table: read self-reports as observations of the group's process.

No claim in this wiki tests this policy as a whole, or compares light structure first with full cooperative structure from the start. No claim in this wiki establishes an optimal group size or composition, or an effect for adult workplace teams or online groups in general. Do not convert the pooled d or the writing ES into an individual's probability of benefit, rank these studies against each other, or read a group product score as evidence about any member. A precise abstention names the missing input (individual starting response, outcome, horizon) and how to get it.

Fitting the design to a situation

Ask for three facts when the brief omits them: how much experience the learners have of working in groups, whether the task could be done by one member alone, and when and how each member's own learning will be checked.

If the brief says… Then change… Basis
The task could be split into parts that members do separately Make it less divisible, or require each member to use another's part; groups of strong students still fail when they do not actually work together High-performing groups fail without collaboration [~W]
Student group leaders will run the discussion Brief leaders to collect each member's view before settling an answer; leaders who rushed and imposed their view prevented discussion Poor group leaders undermined discussion [~W]
The objective is measured at a delayed test Make step 6 a delayed individual check on matched and changed wording; agreement in the pair's talk often did not hold later Co-construction needs personal construction for stable change [~W]
The goal is writing Pair learners to plan and revise, then have each write a piece alone as the individual check in step 5 Collaborative writing improves text quality [+M]
The work is asynchronous online, in small groups Start with a shared prompt and response deadlines; have each member (or pair) post their own proposal before anyone builds on others', and work in a medium that keeps what was written in view (a shared document or wiki rather than calls). If individual checks are weak, change the task or the prompt, since roles or rewards alone did not change achievement Structuring interdependence did not change achievement online [~M]; initial proposals, thinking-together preparation and a shared artefact supported joint work in a primary-school wiki [+M] (the authors' reading of their own study); persistent text and drawings let students explain their postings to each other [+W] (design-based case, no comparison)
Online groups whose participation is uneven Show members how much each has contributed (a group awareness display) with prompts when one member falls silent, and still judge learning by the individual check Groups given the support more often reached near-equal participation [+M] (tentative, descriptive counts, no test statistic); unequal participation went with lower satisfaction [~M] (satisfaction, not learning)
Groups of four or five share one lab station or procedure, or a discussion forum has many participants Use three or four per group, or rotate roles so each member carries out each step; in forums, keep discussion groups small Large groups limited individual participation [~W]; forum learning needs small groups [~W] (an editorial's summary of a study)
A computer-supported collaboration script is used Script one or two named activities, plan the teacher's rounds of the groups as part of it, and judge it by the individual check; a script that visibly changes the talk may leave learning unchanged (more in the principle's script row) Teacher orchestration matters as much as script design [~W]; no direct script effect on learning gains [-M] (d = 0.08, CI −0.39 to 0.56, two scripts compared)
Collaboration itself is the objective (how to work well in a group) After a joint task, have the group reflect together, with prompts, on how it worked, rather than relying only on a script that steers the interaction; choose tasks that require the interactions you want learned, since knowledge gained concentrated on the ones the task required Collaborative reflection beat a script for knowledge about collaboration [+M] (150 university students, laboratory); gains concentrated on information-pooling interactions [~M] (exploratory)
The group will report how well it worked Keep the self-report, but choose the next activity from the individual check, not from it Aligned problem perceptions predict self-reported, not measured, outcomes [~M]
Generative AI is available to the group Set limits on AI use in the joint task and make the individual check AI-free Open collaboration with AI, offloading and lower independent thinking [-W]
Pair work might be done with an AI assistant instead of a human partner Keep a human partner where the aim includes collaboration or social presence; if an AI assistant is used, still check each learner's own performance Human pairs reported the most collaboration and social presence [+W] (self-reported perceptions, one programming quasi-experiment)
A reading programme asks pupils to read independently Add regular peer discussion of what each has read, and check that members are reading before relying on it Extensive reading with peer interaction gave larger gains [+W]; group activities are no guarantee when students have not read [~W] (the author's caution)
Learners are too young to write, and the goal is composing a text Let the group compose aloud with the teacher as scribe, leaving the content and the judgement of when it is finished to the children Teacher-as-scribe co-construction gave five-year-olds authorship and long focused engagement [+W] (one recorded session; engagement, not learning)
The task is building a concept map or diagram from a text Have each member build a whole map alone first, then build one together, rather than filling in or sorting a given map Individual-then-joint gave better delayed recall [+M]; building a whole map gave better recall, most of all in that order [+M] (226 undergraduates, one experiment)
Learners include neurodivergent students and group size is open Offer a choice between pairs and teams of three to five, and ask each learner; this is evidence about comfort, not learning Neurodivergent and neurotypical students prefer different team sizes [~W]
Learners are young or new to group work, or groups are newly formed for a complex task Give roles and a talk prompt from the first session rather than starting light; give the group a practice task of the same kind before the target task, and keep groups together across sessions Roles and prompts: untested proposal. Prior joint experience: groups experienced with similar tasks outperformed inexperienced groups on individual and delayed tests [+M]

The evidence by question

Further evidence, not yet read against this model

Claims this page cited before it was rewritten as a conditional model. They are evidence about the relationship, but none has yet been re-read against the model above; each says how far its own sources have been checked.

Read 2026-10-19 (wave 2): the claims linked since the conversion were read from their evidence entries. Those that bear on the model now sit in it (the sequence, the observation table, the configurations and the situation rows); these remain, each with why.

Illustrative design instance and observation record

Invented illustration, not a tested lesson. In a secondary-school physics class, pairs use a motion simulation. A student wants to agree with her partner and hand in an answer the teacher accepts. The teacher wants each student to predict, alone and two weeks later, how a moving object behaves when the force on it changes. Before pairing, each student writes one prediction with a reason. The pair runs the simulation with a prompt: "each of you states a prediction and why before you run it; if you disagree, say what result would settle it." The teacher notes who proposed each prediction and whether disagreements were settled. Straight after, each student answers a changed prediction item alone. Where a student's individual answer has reverted, she gets a solo attempt with a contrasting case before the next paired task. A matched individual item is repeated two weeks later. The pair's agreed answer is recorded, but it is not used as evidence of either student's understanding.

Record: individual starting response → group, task and structure as delivered → who proposed what, and how disagreements were handled → each member's individual response → candidate interpretations and confidence basis → discriminating probe → chosen next activity and justification → individual response at the stated horizon. Preserve each learner's stated purpose and any change in it. An agent may summarise this record, but must not fabricate missing inputs or assign numerical state probabilities without a calibrated model.

Elements and limits

Collaboration, group work, collaborative problem solving, peer explanation, group roles, individual accountability and group processing.

This pattern is scoped to pairs and small groups where each member's individual learning is the target. Collaboration skill as an outcome (beyond the one situation row above), belonging and peer relationships, long-running project teams, adult workplace teams and open-ended creative group work each need their own outcome measures and evidence. The policy supports observation and design reasoning; its predictive accuracy and the role of learners' valued goals remain to be tested with actual learners.

Claims

Key Sources

  • Johnson, D. W., Johnson, R. T., & Smith, K. A. (2014). Cooperative learning. Journal on Excellence in College Teaching, 25(3-4), 85-118.
  • Kolloffel, B., Eysink, T. H. S., & de Jong, T. (2011). Comparing the effects of representational tools in collaborative and individual inquiry learning. International Journal of Computer-Supported Collaborative Learning, 6, 223–251. doi:10.1007/s11412-011-9110-3
  • Brand, C., Hartmann, C., Loibl, K., & Rummel, N. (2023). Do students learn more from failing alone or in groups? Insights into the effects of collaborative versus individual problem solving in productive failure. Instructional Science. doi:10.1007/s11251-023-09619-7
  • Amante, C., Lucero, M., & Montanero, M. (2026). Learning with concept maps: the effect of activity structure and the type of task. Instructional Science. doi:10.1007/s11251-025-09764-1
  • Olsen, J. K., Rummel, N., & Aleven, V. (2019). It is not either or: An initial investigation into combining collaborative and individual learning using an ITS. International Journal of Computer-Supported Collaborative Learning, 14, 353–381. doi:10.1007/s11412-019-09307-0
  • Zambrano, J., Kirschner, F., Sweller, J., & Kirschner, P. A. (2019). Effects of group experience and information distribution on collaborative learning. Instructional Science. doi:10.1007/s11251-019-09495-0
  • Strauß, S., & Rummel, N. (2021). Promoting regulation of equal participation in online collaboration by combining a group awareness tool and adaptive prompts. But does it even matter? International Journal of Computer-Supported Collaborative Learning, 16, 67–104. doi:10.1007/s11412-021-09340-y
  • Strauß, S., Tunnigkeit, I., Eberle, J., Avdullahu, A., & Rummel, N. (2025). Comparing the effects of a collaboration script and collaborative reflection on promoting knowledge about good collaboration and effective interaction. International Journal of Computer-Supported Collaborative Learning. doi:10.1007/s11412-024-09430-7
  • Malmberg, J., Haataja, E., & Järvelä, S. (2022). Exploring the connection between task difficulty, task perceptions, physiological arousal and learning outcomes in collaborative learning situations. Metacognition and Learning. doi:10.1007/s11409-022-09320-z
Page metadata
Field Value
type pattern
id collaborative-learning
title Collaborative Learning
description A reusable policy for joint work that elicits each member's starting response, chooses a task that needs the members' reasoning, and uses individual checks rather than the group's product or self-report to choose the next activity.
status review
grain_size lesson
canonical True
generated by claude/unspecified on 2026-10-19
trust unverified — no one has confirmed this page's content

Frontmatter is the page's machine-readable record — see the Schema & Guide.