# Two-stage exams: the group part that teaches

**What is a two-stage exam, and does the group stage help students learn?** In a two-stage exam students sit a test alone, hand it in, then sit the same or a similar test in groups of three or four and agree each answer. Group scores are reliably higher. Whether individuals keep the gain weeks later is disputed, so give the group stage 15 to 25 per cent and never let it lower a mark.

Published 2026-10-10 by EdProcto. https://edprocto.com/guides/two-stage-exams-the-group-part-that-teaches

A two-stage exam turns the argument students have after a test into part of the test. Each student sits the paper alone and hands it in. Then groups of three or four sit the same paper, or a similar one, and must agree on every answer. It is also called a collaborative exam or a group exam.

The individual stage still measures each student; the group stage is feedback from peers while the questions are fresh. Every study below was read on 10 October 2026, in full or, where we say so, as its abstract in PubMed, ERIC, arXiv or on the journal's page. They disagree about whether the learning lasts.

EdProcto, which publishes this guide, runs the two stages as two quizzes: the individual one, and a copy that each group sits on one device. See its page [for universities](https://edprocto.com/solutions/universities).

## What is a two-stage exam, and where did it come from?

A two-stage exam is one test sat twice in a sitting: alone for most of the marks, then in a small group that hands in one agreed set of answers. It has no single inventor. Cortright et al. (2003, abstract read) used the design in an exercise physiology class, and by then one teaching method had built a test of the same shape into every unit.

### The team-based learning readiness test and its scratch card

That method is team-based learning, which Burgess et al. (2020) trace to Larry Michaelsen's business-school classes of the 1980s. Students prepare before class, sit an individual readiness assurance test, then a team readiness assurance test, and discuss the answers at once.

Sessions described as classic team-based learning put the team test on an immediate feedback assessment technique form (Dimeo et al., 2020, abstract read): a scratch card. As DiBattista et al. (2004, abstract read) describe it, the card shows at once whether a choice is right, lets the student go on until it is, and gives partial credit after a wrong first try. Whether trying again beats being shown the answer is in our guide to [study mode](https://edprocto.com/guides/study-mode-a-quiz-that-explains-each-answer-as-you-go).

## What happens to scores in the group stage?

Scores go up, in every study we read. In an introductory biology section of about 250 at West Virginia University, Leight et al. (2012) found group means of 74.5% against 61.9% alone on one exam and 84.1% against 63.7% on the next; 93.9% of students scored higher with their group than alone.

The rise is not only the strongest student dictating. Giuliodori et al. (2009, abstract read) had 65 veterinary physiology students sit an exam alone and then in pairs. Where partners had answered differently, 77% of the changed answers became right, and the partner who had been right prevailed about 80% of the time, whether the stronger or the weaker of the two.

A group's score describes the group, so the individual stage carries the weight.

## Do individuals keep the gain days or weeks later?

Sometimes, by a little, and the studies disagree about when. The first four below are crossover designs: each student met some material with a group stage and some without.

| Study | Class | Later test | With the group stage |
| --- | --- | --- | --- |
| Cortright et al. (2003), abstract read | 38, exercise physiology, pairs | Same questions, 4 weeks on | Retained more (P < 0.05) |
| Ives (2014), abstract read | Introductory physics | New matched questions | Ahead at 1 to 2 weeks, level at 6 to 7 |
| Cooke et al. (2019) | 125, introductory biology | New matched questions | Ahead at 23 days (p = 0.037), level at 9 |
| Leight et al. (2012) | About 250, introductory biology | Cumulative questions, 3 weeks on | Level (p = 0.166) |
| Kinnear (2021), abstract read | Undergraduate mathematics | A delayed post-test | Level with a control group |

### What differed between the studies

Three things, none of which settles it.

1. **What the comparison students did.** Cortright's met the questions once before the later test, the others twice, so repetition alone could explain the gain. Cooke et al. gave everyone a second go, alone on one topic and in a group on another; the advantage that survived was small.
2. **The later questions.** Cortright repeated the same questions; Ives and Cooke wrote new ones on the same ideas. Leight et al. note that a student can remember which answer was right without understanding why.
3. **The class.** Thirty-eight students in pairs are not 250 in self-assembled groups of two to four, and Leight's questions averaged the level of application.

Nor do the delays line up: Ives found a gain at one to two weeks and none at six to seven, Cooke none at nine days and one at 23. Leight et al. advise instructors to "determine, rather than assume" that it helps their own students. One way: three weeks later, quiz two ideas from the group stage and two that only the individual stage covered, reworded, and compare. Why the first, solitary recall matters is in our guide to [retrieval practice](https://edprocto.com/guides/retrieval-practice-low-stakes-quizzing-that-teaches).

## How big should the groups be, and who should form them?

Three or four is the custom, and no study we read tested it against other sizes for learning. Connell et al. (2023) placed a biology class of about 200 in groups of three or six for a term. Sixes scored about 3.5 points higher on the group exams, probably because a larger group more often holds a high scorer, but individual scores on the course post-test did not differ. Groups kept together all term outscored reshuffled ones and liked group work better, again with no difference on the post-test.

On who chooses, Miller et al. (2022, abstract read) split one introductory physics exam at Harvard: half the groups were formed by the instructor to balance gender and earlier marks, half by the students. Women performed better in groups they chose, and students there felt more comfortable speaking up. Team-based learning advises the opposite: mixed teams formed by the instructor (Burgess et al., 2020). On video of 27 groups, Ives et al. (2016, abstract read) saw students who had been right alone take a larger part, and in mixed groups the strongest gained most in participation and the weakest least.

A defensible default: threes or fours, the groups students already work in, kept for the term.

## What do students say, and what is known for school classes?

Students like it. In Cooke et al. (2019), 96 of 123 said they enjoyed the midterm. Rempel et al. (2021, abstract read) asked 129 first-year chemistry students, who said it eased exam anxiety partly, not fully.

Nearly all of this evidence comes from universities. For schools we found no trial of a two-stage test, and one close study of group quizzes: Esmonde (2009, abstract read) followed three secondary mathematics classes for a year and found that "experts tended to dominate" them. A school teacher should borrow the shape, not the university figures: pairs or threes, nothing on the grade, and an eye on who is talking. A tutor's version is in [a quiz that runs the session](https://edprocto.com/guides/group-tutoring-a-quiz-that-runs-the-session).

## How should you weight, time and choose the questions?

Give the individual stage 75 to 85 per cent, and let the group stage only raise a mark.

- **Weight.** Rempel et al. (2021) gave the individual stage 80 to 85%; Meaders and Vega (2023) gave it 75% and used it alone whenever it was the higher score. At 85 and 15, a student on 60% in a group on 90% gets 64.5%. A student on 95% in a group on 80% would get 92.75%, so the rule keeps 95%.
- **Time.** Half the individual time or less. Meaders and Vega gave 60 minutes, then 30 for three group questions; Cooke et al. gave 40, then 10 for one.
- **Questions.** Choose those students will disagree on. Discussion helped more where partners had answered differently (Giuliodori et al., 2009), and participation peaked on questions a member had got wrong alone (Ives et al., 2016). Keep application questions with plausible wrong options, such as [multiple choice](https://edprocto.com/templates/mcq), and cut plain recall.

| Purpose | Group stage | Counts for | Answers shown |
| --- | --- | --- | --- |
| A graded midterm | The same paper or its harder half | 15 to 25%, never lowering | When both stages are in |
| A [weekly quiz](https://edprocto.com/guides/weekly-quizzes-in-a-lecture-course-low-stakes-and-cumulative) or readiness quiz | The same questions | Little or nothing | After each question |
| Revision before an exam | The same questions | Nothing | After each question |
| A school class | Pairs or threes the teacher forms | Nothing | After each question |

## How do you run it in a room and online, and what about the student who coasts?

In a room, seat the groups together from the start, take in the individual stage before anyone speaks, and give each group one paper or one device. Online it holds up: Meaders and Vega (2023) ran one term's exams online and the next in person, with the same result. In a breakout room, one member shares a screen and enters the agreed answers.

### The student who coasts

Four things limit a passenger.

1. **The weight.** At 85 and 15, a student on 50% in a group on 90% gains six points.
2. **The order.** Everyone arrives with an answer; ask each member to say theirs first.
3. **The size.** In Connell et al. (2023), students in threes and in sixes alike said they felt accountable and included.
4. **Enough time.** One student in Meaders and Vega (2023) admitted giving way to a dominant voice when rushed.

## A worked example: one 50-minute period

A class of 28 and a ten-question quiz fit one period, with ten minutes left to teach.

| Minutes | What happens |
| --- | --- |
| 0 to 3 | Students open the first quiz with its code |
| 3 to 23 | Individual stage: ten questions, no answers shown |
| 23 to 26 | You close the first quiz; seven groups of four each open the copy on one device |
| 26 to 38 | Group stage: the same ten questions, each answer agreed, then checked |
| 38 to 48 | The questions most groups missed, with the whole class |
| 48 to 50 | You release the individual scores |

## How does EdProcto run the two stages?

EdProcto runs a two-stage exam as two quizzes, and it is the best tool for the job when the questions come from your own course: the group copy takes one press, and the group stage gets the scratch card's instant verdict with your explanation attached. Every answer is marked on the server.

1. **Build the individual quiz.** Paste your notes into the quiz maker on the [home page](https://edprocto.com/) and edit what Procto AI drafts. Leave the kind at "Exam" ("handed in once, marked at the end") and leave "An account is needed to take it" on, so each student gives a name and an email address. Set "Time limit" to "Set" with your "Minutes overall", and "Show the score" to "When released": nothing of the score or the answers reaches a student until you release it.
2. **Make the group copy.** Press "Duplicate this quiz" at the foot of the builder. The copy is a draft titled "… (copy)" with the same questions, keys and settings, its own code and no submissions.
3. **Fit the copy to a group.** Change its kind to "Study quiz". Switch off "An account is needed to take it", set "Show the score" to "After submitting", shorten "Minutes overall", and delete any question you do not want discussed.
4. **Run stage one.** Press "Publish & share"; students open the quiz with its code. When time is up, press "Close the quiz".
5. **Run stage two.** Press "Share with a code" on the copy. One device a group opens it and types the group's name, each one different, under "Your name". After each answer the group presses "Check" and reads "Right." or "Not this time.", the answer and your explanation. In a study quiz a checked answer stands, so the group argues first.
6. **Combine the marks.** On each results page press "Download the marks": a spreadsheet file with a row for each attempt, students in one and groups in the other, each with a "Score %". Add each student's group to their row, look up that group's "Score %", and take the larger of the individual score and 0.85 of it plus 0.15 of the group's. Then press "Release the scores" on the first quiz. The combined figure is yours to publish.

"Each question" on both results pages gives the share who "Got it", alone and together. For rival groups and a scoreboard, set the copy's kind to "Quiz night": see [hosting a quiz night](https://edprocto.com/guides/how-to-host-a-quiz-night-with-a-live-scoreboard).

Free takes thirty students a quiz, Tutor 100 and Educator 500, and Educator adds the full analysis. On Free the copy counts as one of the month's ten quizzes, so a two-stage exam uses two. See [pricing](https://edprocto.com/pricing).

## Quick questions

### What is a two-stage exam?

A test sat twice in one sitting: alone for most of the mark, then in a group of three or four that must agree each answer.

### How much should the group stage count?

Between 15 and 25 per cent in the studies we read. Add a rule that it never lowers a mark. EdProcto gives each stage its own marks file.

### Do two-stage exams improve learning?

Group scores rise every time. Individual gains weeks later appear in some studies and not in others, so check with a later quiz of your own.

### Can a two-stage exam run online?

Yes. In EdProcto both stages open by code on any device, and one member of each group enters the answers: see the page [for universities](https://edprocto.com/solutions/universities).

## References

- Burgess, A., van Diggele, C., Roberts, C., & Mellis, C. (2020). Team-based learning: Design, facilitation and participation. *BMC Medical Education, 20*(Suppl. 2), Article 461. [https://doi.org/10.1186/s12909-020-02287-y](https://doi.org/10.1186/s12909-020-02287-y)
- Connell, G. L., Donovan, D. A., & Theobald, E. J. (2023). Forming groups in a large-enrollment biology class: Group permanence matters more than group size. *CBE Life Sciences Education, 22*(4), Article ar37. [https://doi.org/10.1187/cbe.22-08-0172](https://doi.org/10.1187/cbe.22-08-0172)
- Cooke, J. E., Weir, L., & Clarkston, B. (2019). Retention following two-stage collaborative exams depends on timing and student performance. *CBE Life Sciences Education, 18*(2), Article ar12. [https://doi.org/10.1187/cbe.17-07-0137](https://doi.org/10.1187/cbe.17-07-0137)
- Cortright, R. N., Collins, H. L., Rodenbaugh, D. W., & DiCarlo, S. E. (2003). Student retention of course content is improved by collaborative-group testing. *Advances in Physiology Education, 27*(1–4), 102–108. [https://doi.org/10.1152/advan.00041.2002](https://doi.org/10.1152/advan.00041.2002)
- DiBattista, D., Mitterer, J. O., & Gosse, L. (2004). Acceptance by undergraduates of the immediate feedback assessment technique for multiple-choice testing. *Teaching in Higher Education, 9*(1), 17–28. [https://doi.org/10.1080/1356251032000155803](https://doi.org/10.1080/1356251032000155803)
- Dimeo, S. P., Sudario, G., Sharma, S., Bellman, L., Gallion, A., Andreeff, K., & Boysen-Osborn, M. (2020). Pediatric seizure team-based learning. *Journal of Education & Teaching in Emergency Medicine, 5*(3), T1–T41. [https://doi.org/10.21980/J8MD22](https://doi.org/10.21980/J8MD22)
- Esmonde, I. (2009). Mathematics learning in groups: Analyzing equity in two cooperative activity structures. *Journal of the Learning Sciences, 18*(2), 247–284. [https://doi.org/10.1080/10508400902797958](https://doi.org/10.1080/10508400902797958)
- Giuliodori, M. J., Lujan, H. L., & DiCarlo, S. E. (2009). Student interaction characteristics during collaborative group testing. *Advances in Physiology Education, 33*(1), 24–29. [https://doi.org/10.1152/advan.90161.2008](https://doi.org/10.1152/advan.90161.2008)
- Ives, J. (2014). *Measuring the learning from two-stage collaborative group exams*. arXiv. [https://doi.org/10.48550/arXiv.1407.6442](https://doi.org/10.48550/arXiv.1407.6442)
- Ives, J., de Jong Van Lier, M., Sumah, N. K., & Stang, J. B. (2016). *Examining student participation in two-phase collaborative exams through video analysis*. arXiv. [https://doi.org/10.48550/arXiv.1607.03960](https://doi.org/10.48550/arXiv.1607.03960)
- Kinnear, G. (2021). Two-stage collaborative exams have little impact on subsequent exam performance in undergraduate mathematics. *International Journal of Research in Undergraduate Mathematics Education, 7*(1), 33–60. [https://doi.org/10.1007/s40753-020-00121-w](https://doi.org/10.1007/s40753-020-00121-w)
- Leight, H., Saunders, C., Calkins, R., & Withers, M. (2012). Collaborative testing improves performance but not content retention in a large-enrollment introductory biology class. *CBE Life Sciences Education, 11*(4), 392–401. [https://doi.org/10.1187/cbe.12-04-0048](https://doi.org/10.1187/cbe.12-04-0048)
- Meaders, C. L., & Vega, Y. (2023). Collaborative two-stage exams benefit students in a biology laboratory course. *Journal of Microbiology & Biology Education, 24*(1), Article e00138-22. [https://doi.org/10.1128/jmbe.00138-22](https://doi.org/10.1128/jmbe.00138-22)
- Miller, K., Kestin, G., & Miller, O. (2022). How gender composition and group formation impact the effectiveness of group work in two-stage collaborative exams. *Physical Review Physics Education Research, 18*(2), Article 020137. [https://doi.org/10.1103/PhysRevPhysEducRes.18.020137](https://doi.org/10.1103/PhysRevPhysEducRes.18.020137)
- Rempel, B. P., Dirks, M. B., & McGinitie, E. G. (2021). Two-stage testing reduces student-perceived exam anxiety in introductory chemistry. *Journal of Chemical Education, 98*(8), 2527–2535. [https://doi.org/10.1021/acs.jchemed.1c00219](https://doi.org/10.1021/acs.jchemed.1c00219)