In peer instruction, students are given multiple choice questions to consider individually, followed by an individual vote on the question using a clicker. That’s followed up by a small group discussion which is followed by a re-vote. Typically the percentage of students getting the correct answer to the question jumps, often in my experience with nearly the entire class converging on the right answer following discussion. But does that jump happen because peer discussion helps students understand the material better, or because students with a weaker understanding are socially influenced by students with a stronger understanding?
This research paper has some data that suggest the former. The authors administered 16 different sets of PI questions to a large-lecture (n = 350) physics class. The questions were given in pairs of “isomorphic” questions, having different contexts and “cover stories” but assessing the same core concepts. First, a question was asked using the usual vote-discuss-revote model. Then the second, isomorphic question was asked with only individual voting taking place.
If the peer discussion was merely creating an environment of peer pressure to vote the same as the knowledgeable students, then what you’d expect is that the percentage of students answering correctly would be relatively low on the first vote on Question 1, then high on the revote, then back to low again on Question 2. But that’s not what the study found! On the first vote, an average of about 50% of students were voting correctly. On the revote of the first question, that average went up to just shy of 70%. On the first vote of the second question — testing the same concept but in a new context — the percent voting correctly was above 70%.
What’s more, the study found that even if students ended up in groups where nobody answered correctly on the first vote, they were still able to vote correctly on the revote and on the second question. That’s pretty profound. You have these groups of students, none of whom are answering a conceptual question correctly, but through putting their heads together and going through all the available options, they are able to come to the right conclusions and transfer that knowledge to a new (isomorphic) situation. In the study, the student response histograms were not shown after the first vote, either, so there’s no social pressure to suggest that one answer is right.
What would really drive this home for me is if students were given some sort of question — a more complex PI question or an open-ended problem — where they had to use the concept being tested in PI to solve the new problem, to see just how well the knowledge transfer was taking place. Since the questions are isomorphic after all, and because they happened in succession in the same class meeting, I’d want to know that student understanding could be applied to a more complex problem and in a different time/space environment. That might be a nice follow-on study. For now it’s nice to see some verification that PI can really help students learn, even when the groups they’re in don’t contain any experts.