What the research says

Peer Instruction: Why Talking to a Neighbour Improves Answers

Vote alone, argue it out with a neighbour, vote again. Ten years of Harvard physics data and a clever genetics study show why it works, and which questions it works best on.

By the Giga Quizzes team5 min read
A bright red and yellow toy telephone with a coiled cord
Photo: Matias Megapixel, Unsplash, Unsplash License, cropped.
In this post
  1. 1How one round works
  2. 2Ten years of results
  3. 3Learning, or just copying the clever kid?
  4. 4The sweet spot: 35 to 70 percent
  5. 5What the studies cannot tell you
  6. 6Trying it with any live quiz

In 1991 the Harvard physicist Eric Mazur changed how he taught his introductory course. Instead of lecturing straight through, he stopped every so often, put up a single conceptual question, and asked students to answer it on their own, then to convince the people sitting next to them, then to answer again. He called it Peer Instruction. Two studies, one looking back over ten years of his course and one designed to test whether the discussion is real learning or just copying, explain why such a simple routine works.

How one round works

Catherine Crouch and Eric Mazur describe the steps precisely, and the timings matter more than they look.

  1. A short question on one core idea goes up. Mazur calls these ConcepTests: they test understanding, not recall of a formula.
  2. Students get one or two minutes to think and commit to an answer alone.
  3. They then spend two to four minutes trying to convince their neighbours that their own answer is right, while the teacher walks round and listens.
  4. Everyone answers the same question again.
  5. The teacher explains the answer, pitched at what the discussion revealed.

The individual vote before the discussion is the part people tend to skip, and it is the part that makes the rest work. A student who has committed to an answer has something to defend, and a reason to listen when a neighbour disagrees.

Ten years of results

Mazur’s course gave the same conceptual test of mechanics, the Force Concept Inventory, at the start and end of every term from 1990 onwards. Crouch and Mazur report a figure called the normalised gain: the share of the possible improvement that students actually made. A class that starts at 70% and ends at 85% has gained half of the 30 points available, so its gain is 0.5.

YearTeachingNormalised gain
1990Traditional lectures0.25
1991First year of Peer Instruction0.49
1997Peer Instruction, refined0.74
Calculus-based physics at Harvard, from Crouch & Mazur (2001)

A common worry is that time spent on concepts comes out of time spent on calculation. The data do not show that. In one comparison, a final-exam problem from a year taught traditionally was set again for a class taught with Peer Instruction; the new class averaged 7.4 out of 10 against 5.5 the year before.

Looking at every ConcepTest across one semester, nearly half of all correct answers were reached only after discussion, and students switched from a right answer to a wrong one just 6% of the time. Even the strongest students were stretched: none of them answered correctly before discussion more than 80% of the time.

Learning, or just copying the clever kid?

There is an obvious objection to all of this. Maybe the second vote goes up because the student who knew the answer simply told everyone else. Michelle Smith and colleagues designed a study in an undergraduate genetics course to tell the two explanations apart.

Students answered a question, discussed it and voted again, as usual. Then came a twist: a second question on the same concept, built the same way but with different details, answered alone with no discussion. If students had only copied a neighbour, the second question would expose them. It did not. Their results showed that peer discussion improved understanding, even in groups where nobody had known the right answer to begin with.

That last finding is the remarkable one. Talking through a problem with people who are equally unsure still moves the whole group closer to the answer, because explaining a choice out loud forces a student to check the reasoning behind it.

The sweet spot: 35 to 70 percent

Not every question is worth discussing. Crouch and Mazur found that discussion raised the number of correct answers substantially when between 35% and 70% of the class got the question right on the first vote, and that the gain was largest at around half.

  • Under 35% right: the question may be ambiguous, or too few students understand the idea to have a useful argument. Teach a little more first.
  • 35% to 70% right: the class is split, and almost everyone has someone nearby who disagrees. Discuss and vote again.
  • Over 70% right: there is little to gain. Confirm the answer and move on.

This gives you a decision rule for the moment the first vote comes in, which is often the hardest moment in the lesson. It also says something about writing questions. Crouch and Mazur advise making every wrong option plausible, ideally based on a real misunderstanding, and a good source is last year’s exam scripts: the mistakes students actually made are the best wrong answers you can write.

What the studies cannot tell you

Both studies come from universities, and the Harvard figures come from one department that kept improving its course. Alongside Peer Instruction, Mazur replaced reading quizzes with short written responses before class and brought in a new textbook for part of the course, so not all of the rise from 0.49 to 0.74 can be credited to the discussions themselves. The genetics study was a single course too. What the two papers do show, together, is a large and lasting effect in real lecture halls, and a mechanism that is genuine understanding rather than copying.

Trying it with any live quiz

The routine needs nothing more than a way to collect answers from everyone at once. With a live quiz on students’ phones, the steps translate directly: show the question, let the answers come in, and pause before revealing which one is right. If the split is in the sweet spot, ask students to turn to the person next to them and convince them, then ask the same question again.

Two habits make it work. Keep the first vote private, so that nobody changes their answer to match a friend before they have thought. And give the discussion a real time limit, three minutes or so, announced at the start. When it ends, the second vote is your evidence of what the conversation achieved. The post on polls for quiet students has a worked example of the vote, talk and vote-again pattern.

Sources

  1. Crouch, C. H. & Mazur, E. (2001). Peer Instruction: Ten years of experience and results. American Journal of Physics, 69(9), 970–977.
  2. Smith, M. K., Wood, W. B., Adams, W. K., Wieman, C., Knight, J. K., Guild, N. & Su, T. T. (2009). Why peer discussion improves student performance on in-class concept questions. Science, 323(5910), 122–124.

More from the blog

All posts →