Hi. I’m Chelsea Troy. I’m a computer science educator at the University of Chicago.
You can see all my posts about teaching by visiting the teaching tag right here. You’ll find pieces about designing a syllabus, choosing topics to cover in a session, and selecting session activities. As you can see right here, my students do a lot of work in groups.
This post covers my approach to group work this quarter.
Critical Background Context
My entire pedagogical approach optimizes for a student that wants to learn. I don’t focus on stopping students from getting A’s or squeaking through without learning.
I am confident that the skills I teach are valuable. I am also confident that my teaching methods make it easier for students to obtain those skills than figuring it out on their own. Therefore, I believe that students who avoid learning in my class are cheating themselves and will regret it.
I do not focus on disciplining or “gotcha”-ing people, because that places the student that wants to learn under unproductive stress.
I also teach an elective course in a master’s program. Students in my class have already spent years making choices that brought them to me. Teachers who teach required courses, or who teach courses at an earlier point in students’ education, face challenges that I don’t face. I acknowledge this. My approach to teaching might not work everywhere. I’m still going to write about it.
Group Work: Replacing Fear with Appeal for Hard Problems
I teach future software engineers. Software engineers spend their careers on unsolved problems in a kaleidoscope of languages and frameworks. I have to prepare my students for that by asking them to try to do things that I have not taught them to do yet.
I do not find attaching a grade to new, challenging problems to be productive, particularly before I’ve taught investigation and induction skills. This, again, adds unproductive stress. I want productive stress; I want students to have opportunities to fail that do not feel like permanent, consequential reflections on their skills. I grade students on homework assignments that they should be able to do, based on what I taught them. I save the stretch challenges for in-class, ungraded exercises.*
*I could also theoretically use extra credit. Sometimes I do put an extra credit item on homework for students to attempt individually. This occasional stretch is not enough to fulfill my students’ challenge needs, though, and if I offered extra credit for every stretch quest I gave them, students could theoretically cover up unacceptable gaps in one topic with their stretch points from another topic.
It can be hard to challenge oneself without a short-term consequence: ask anyone who has struggled for years to get to the gym. So I provide for my students what trainers advise for beginning gym-goers: accountability partners. I divide them into groups and asking them to approach the challenges together.
Circumventing Student Hatred for Group Work
Many students hate group-work. I’ve identified two reasons for this and attempted to circumvent them in my class.
The first source of group work hatred has to do with being forced to coordinate four busy schedules outside of class. This is an especially relevant issue in my class because I teach adults, some of whom have jobs and dependents. I avoid asking students to do this by setting aside time for group work in class, when they’ve already agreed to be present by registering.
The second source of group work hatred has to do with grades. Students who work hard don’t want to get stiffed the bill by groupmates who force them to do more than their fair share for the same grade. I avoid this by not grading students on their outcomes during group challenges.
Contrary to popular concern, this does not seem to result in students shirking effort on the challenges. First of all, again, I’m optimizing for a student who wants to learn. Such a student isn’t going to eek by for a grade and deliberately ignore an activity that I:
- designed specifically for maximum learning stimulus with minimum stress.
- carved out time for in class so they don’t have to do it on their own time.
Second of all, even if an individual student might want to do that, they’re subjected to peer pressure. Who wants to risk rejection with “Hey, group, let’s just sit on our tuchuses for twenty minutes and not do our challenge?”
I divide my students into groups of four, if I can. These groups of four will tackle problem-solving challenges together, like modifying complex code, annotating a risk diagram, or discussing what they would do if they found out their employer was behaving unethically.
Why four? Two reasons:
1. A group of four is small enough that students can respond to prompts by going round-robin. Once the number of people gets much higher than four, this works less well, and unstructured conversations degenerate into the participants with the highest caucus scores doing the vast majority of the talking. (We’ll talk more about caucus scores in a minute).
2. A group of four is large enough to produce a high probability that there will be multiple students in the group who genuinely want to learn. So if there’s a slacker somewhere in the class, their group still has three other people who can help each other tackle the challenge, and the one slacker doesn’t totally tank the mood.
3. (Bonus side effect) If I have groups of four, I can divide students into pairs for some activities and keep their pairs within their groupmates.
I do not assign the groups randomly. Instead, I use students’ caucus scores.
The caucus score is a scale that I developed as a proxy metric for an individual’s usual experience in meetings characterized by unstructured discussion. Without explicit rules for who gets to talk, how much, and when, implicit rules take over—and the implicit rules can get pretty gross.
I have described caucuses and their impacts on team members of different caucus scores in more detail right over here. The three point summary:
- People with high caucus scores tend to have an easy time talking in unstructured discussions, so they think those discussions work great. This tends to include people with decision-making power, which is why caucuses are so common even though they’re not nearly as productive as these people think they are.
- People with low caucus scores tend to have a hard time contributing to unstructured discussions. They find those discussions stressful and unproductive.
- Unstructured discussions are stellar ways to lose the contributions of participants with low caucus scores through caucal collaborative degeneracy: when people never receive the opportunity to share critical information or insights at their disposal because other people talked over them.
Sound familiar? One fantastic way to fix that loss at your organization is to replace the caucus with some explicit meeting structure, especially when the size of the meeting surpasses about four people (recall, past this group size, round robin fails).
This video explains how you might add structure to your discussions to prevent caucal collaborative degeneracy:
As valuable as the skill of running an inclusive meeting is, it’s a whole, separate, advanced skill. I need to get my students working together quicker than the time it would take for them to develop that skill. But I still need to avoid caucal collaborative degeneracy during group work; as an educator I cannot afford to lose the contributions of my students with lower caucus scores.
So here is what I do:
- I calculate students’ caucus scores with this diagnostic survey. That’s a template link, so you can copy that survey into your google drive and send it to your students. It is a simplified diagnostic from the one I put in the original caucus score writing, so don’t be surprised that the questions and scoring are different from what you’ve read in my other writing. The max caucus score is 16. The minimum score is 0. That form should calculate the student’s score for you.
- I put the students in a list in order from highest caucus score to lowest.
- Top 4 scores in a group. Next 4 scores in a group. So on, to the bottom.
After the first group activity, I send students a survey with two questions on it plus an “additional comments” section. Here are the two questions:
- In your breakout room, were you able to voice your perspective?
- In your breakout room ,did you get a chance to hear from everyone else in the group?
Here is a sample of some survey results I have seen:
If an individual student lets me know that they were unable to contribute to their group, I switch them with the person with the highest caucus score in the next caucus score group down (modulo adjustments as I learn about individual students). I then ask these follow-up questions again in the next class, after the switches, to make sure everyone is able to contribute to their group.
Once everyone is responding “yes” to both of the above questions, I leave the groups the same for the rest of the quarter. This allows the students to have some continuity from session to session, and their familiarity with their groupmates puts them in a comfortable environment to take risks as the discussions and challenges get harder over the course of the quarter.
I had some ideas about how this procedure might go. Some were right; some were wrong.
1. I thought folks from dominant demographics would coalesce toward the top groups, with folks of more marginalized identities towards the bottom. The evidence doesn’t support that for me so far (that could change as I do this with more classes). So far I see a small trend, but it’s not anywhere near distinct enough to attribute to something besides random chance, especially at the small statistical power of my meager data.
2. I thought students with lower caucus scores would have a better time with these groups than they do in normal group work, and students with higher caucus scores would have a worse time. The evidence doesn’t support that, either.
First, everybody seems to like their group okay, and the majority love their group:
Also, three quarters of students are having a better experience with these groups than they have with group work, historically. This is true despite the fact that we have to be remote this quarter.
Is there a trend to who is having which experience? Not that I can find. The students who said they are having a much better experience with group work in this class than in group work historically are scattered across every caucus score group.
The students who are having a little worse experience with group work in this class than in group work historically each pegged something besides the group split as the reason. One said the remote format makes things harder than in-person group work. The other said that the regroup discussions are the challenging part (which is fair—I’ve tried like five different things for making the regroups productive, and I haven’t been satisfied with any of them so far. It’s a work in progress).
3. I thought the groups of students with lower caucus scores might work differently than the groups of students with higher caucus scores. For this, I do see a little bit of evidence.
For each group assignment lately, I have provided template documents where each group can take notes, annotate diagrams, or answer questions. As they are working, I can scan through these documents to see which groups are approaching completion.
Groups with higher caucus scores tend to go broad. They finish the activities the fastest of all the groups. On one diagram annotation assignment, the top caucus score group finished in around 8 minutes, followed by the next one around 9 and a half minutes, and so on. Two lower caucus score groups did not quite finish.
But groups with lower caucus scores tend to go deep. They don’t always finish the exercise, but the part that they do finish includes ideas and considerations that other groups don’t make. On the same diagram annotation assignment I mention above, the diagrams from groups of students in the top half of caucus scores possessed, in total, 12 ideas that did not appear in the diagrams of groups with lower caucus scores. A lot of these were in the part of the diagram that the groups with lower caucus scores did not get to. But in just the part that those groups did get to, their diagrams had 12 ideas that did not appear in the diagrams of the groups with the higher caucus scores.
I have seen similar results to these across two other activities I’ve done with the students.
There might be a commentary to be made here, when the going gets tough or all the obvious answers are exhausted, to look to the quiet person in the corner of the room. But I don’t have nearly enough data to formally express that commentary, so I’m not doing it…not officially, anyway ;).
I do not have enough data to draw ironclad conclusions, and even if I did, those conclusions could still be wrong.
My situation depends on my specific context. I have a very small sample size of nineteen students (so far). I do not have a control group, unless you count the fact that I asked the students to compare their experience in this class to other group experiences and you assume that “other group experiences” involved randomly assigned groups.
No one is saying that this is statistically rigorous so far, especially not me, The Person Who Debunks Computer Science Claims on Their Statistical Rigor for Fun.
That said, I’d love to hear results from more instructors.
I’d be happy to advise how I can to help you try this in your classroom, and I’d be thrilled to hear your observations and results. Maybe if we try this in enough places, we can draw conclusions that show my conclusions were total outliers. If this appeals to you, feel free to reach out to me: chelsea at this blog’s address.
In the next post on teaching, I’ll more directly address adaptations to remote learning.
If you liked this piece, you might also like:
The debugging posts (a toolkit to help you respond to problems in software)
The Listening Series (Prepare to question much of what you know about how to be good at your job.)
Skills for working on distributed teams (including communication skills that will make your job easier)