Collaborative learning and group work
One student does the calculation and three watch
The group handed in one good page, and you have no idea which of the four students wrote it.
What it is
Group work with four things specified in advance rather than announced verbally. The fellowship’s session on this framed the design decisions as size, accountability, task structure, norms, and formation, and the unstructured version most of us run specifies none of them.
That is not a weaker version of the same practice. It is a different activity with a different distribution of who learns.
Run it
- Small groups, instructor-assigned. Three to five, so no seat is passive. Assigned rather than self-selected, and reshuffled between activities.
- A named procedural role for every student, rotating. Facilitator, recorder, reporter, equity monitor.
- Individual work alongside the group product. A pre-work memo before class, an individual submission after, or both. This is the part that does the work.
- A task worth doing together. Open-ended, more than one defensible answer. Group work is actively worse than solo work for memorization and better for judgment.
The strongest cheap addition is an isomorphic pair: the group solves one problem together, then each student solves a structurally identical one alone. The gap between the two is the finding.
Say this
Roles, posted before class
Facilitator. Keep the group on task and finish within the time allotted.
Recorder. Draft the group’s written response.
Reporter. Share the group’s key ideas with the whole class.
Equity monitor. Pay attention to who is contributing, and invite anyone who has not spoken much to do so.
Roles rotate each session, so everyone practices each one.
Individual pre-work, due before class
“Post three to five bullet points giving your initial answer and your reasoning. You are not expected to be right. You are expected to have thought about it before your group starts, so the discussion begins from four positions rather than from one confident one.”
Techniques
Jigsaw
The fellowship named jigsaw three times without supplying a procedure. Aronson’s original divides content into non-overlapping segments, puts students in home groups of that size so nobody can finish alone, has students work their segment alone, convene as expert groups to rehearse how to teach it, return to home groups to teach in turn, and then assesses everyone individually on all segments (Aronson et al., 1978).
The expert-group stage is the one people drop and the one Aronson identifies as load-bearing. Read the evidence below before adopting it.
In my courses
Adapted from EPI 553. Planned for EPI 601, odd weeks.
I will use structured pair work for in-class study critique activities. Students work in rotating pairs (assigned randomly each session) to analyze published studies. Each pair uses a structured critique template. Both students submit individual write-ups but discuss interpretations together.
EPI 553 used a driver/navigator model for paired R labs. The 601 adaptation renames the roles for critique work: a lead critic identifies methodological strengths and weaknesses, and a challenger pushes back with alternative interpretations. Pairs and roles re-randomize each session, and both students submit individually. The reflection prompt is “what did your partner notice about the study design that you initially missed?” New for 601, built on the 553 model.
Evidence
The strongest transfer on this site. In college statistics courses specifically, cooperative learning reached d = 0.60, moving the median student from the 50th to the 73rd percentile (Kalaian & Kasim, 2014). Postsecondary STEM meta-analysis puts achievement at d = 0.51 (Springer et al., 1999).
Step three rests on a mechanism rather than a classroom trial: effort drops when individual contribution is unidentifiable, and the drop shrinks when it is identifiable (Karau & Williams, 1993).
Peer assessment adds g = 0.31 and beats teacher assessment (Double et al., 2020). Note the boundary: peer assessment is supported, peer grading is a different act and the evidence does not reach it.
Jigsaw specifically does not work. Five preregistered randomized experiments, powered to detect d = 0.40, found jigsaw versus standard teaching at ES = 0.00, 95% CI [−0.10, 0.09] (Stanczak et al., 2022). A systematic review of 69 studies declined to pool an effect size and flagged research-integrity problems in the literature (Vives et al., 2025). Structured pair or group work with individual accountability has better support and costs less to set up.