Post (7): Effect of Peer-Driven Knowledge Sharing on Skill Production in K-12 Classrooms

Research Study

Skill development in K-12 classrooms differs across students due to their varying learning motivations, abilities, and needs. Educational Technology (EdTech) has proven to improve learning productivity -- the rate at which learning efforts are converted into academic success -- for students with varying learning needs. A prominent example is personalized computer-aided learning, which identifies students’ knowledge gaps and offers customized educational content to improve their skill development. However, it can only help those who are motivated to use and have adequate support to learn from these applications. Since students from lower-income families are generally less engaged with education and have lower educational resources than high-income families, they may not benefit from such EdTech without adequate inducement and support. For this reason, most successful EdTech interventions for underprivileged students were provided in structured classes (inside or outside school hours) with teaching support. However, it is hard to scale such interventions given the constraint on the availability of teaching and other school resources. Surprisingly, few EdTech applications have utilized classroom peers -- an abundantly available yet underutilized resource in the K-12 educational ecosystem -- to improve learning. In the current research, we design an EdTech platform that leverages peers to improve skill production in K-12 classrooms.

Our educational platform (called EPInc) has three key features. First, it identifies students’ knowledge gaps based on their homework performance. Second, it manipulates students’ peer groups in classrooms by organizing students of different abilities in classrooms into balanced teams. Thus, each team has high-ability students who can help other teammates with their knowledge gaps. Finally, it creates team contests in classrooms, so the motivation to win contests induces peers to help students with their knowledge gaps.

The EPInc platform works in the existing school setup as follows. Teachers give class instructions per the existing curriculum and then offer students base homework on the platform. Once students complete the homework, the platform identifies students’ knowledge gaps (topics in which they did poorly) based on their homework performance. The platform displays the topic-wise performance of team members and nudges students to seek help from their teammates who have done well in their knowledge gaps so that they can win class tournaments as a team. Then the platform generates personalized homework, which contains more questions in areas of students’ knowledge gaps. After a few homework assignments, the platform organizes class tournaments in which students answer questions individually, but they obtain team-based rankings based on the aggregate team score. Thus, teams obtain high ranks if their members individually perform well.

Team-based incentives, such as team contests, can induce students to share knowledge with teammates and, thus, improve overall skill production in K-12 classrooms. In contrast, the existing individual-based incentives in classrooms encourage high-ability students to invest in education with a higher likelihood of success while discouraging the low-ability ones from investing in education. Thus, we may expect a lower dispersion of skill acquisition in classrooms with team-based than individual incentives. However, the overall classroom skill production and its dispersion under the two incentive schemes depends on how the team members’ interactions evolve in practice. It is unclear whether high-ability students share knowledge with their low-ability teammates or find it frustrating and time-consuming, leaving them with less time for self-study. Similarly, low-ability students could feel demoralized from their interactions with high-ability peers instead of feeling motivated if adverse team dynamics emerge. Overall, how the skill production may vary from the two technologies -- team and individual based -- is an empirical question to be tested.??

To estimate the effect of peer-driven knowledge diffusion, we conduct a field experiment on 3294 students in 92 Grade-3 to Grade-6 math classes in 13 Indian schools. We performed within school random assignment of classes into three groups: team, individual, and control classes. Students in team classes use the EPInc platform in balanced teams comprising students of different abilities. Students in individual classes use the EPInc platform individually, and those in control classes don’t use the EPInc platform.

At the start of the academic session, we conducted baseline assessments of all students based on the math syllabus of their previous grades. Teachers in all classes give students class instructions as per the existing curriculum and pedagogy in the schools. After lectures, teachers offer homework to students on a few topics. While students in control classes get traditional paper-based homework, those in team and individual classes get digital homework on the EPInc platform. After a few homework, teachers organize class tournaments on the EPInc platform in team and individual classes. While students in the individual classes are encouraged to self-study to do well individually in tournaments, those in the team classes are encouraged to help their teammates to help their team win tournaments.?

In the current study, we utilize data on the homework and tournaments completed by 733 students in 21 team classes and 628 in 17 individual classes till April 2024. The baseline scores of all students and students in each ability terciles in the team and individual classes were statistically indistinguishable, indicating they had a similar stock of math skills at the start of the experiment. During the experiment, students acquire skills by listening to class lectures, doing homework, and studying. Students in the two types of classes had similar educational resources, except that While students in team classes help their team members with their knowledge gaps, those in individual classes self-study. We measure students’ math skills with their tournament scores.\footnote{similar to assessments for measuring skills, students individually answer questions on different math topics in limited time in tournaments} A simple comparison of class-level aggregate tournament scores reveals a significantly higher average value and less dispersion of tournament scores in team classes than in individual classes. This result indicates model-free evidence of higher and less dispersed skill production under team-based incentives in classrooms. Examining tournament scores across students of the two groups, We find students in team classes obtain 0.31σ higher tournament scores than those in individual classes. We found similar effect magnitudes across students in the top, middle, and bottom ability terciles. These are intent-to-treat estimates indicating that team-based incentives result in higher skill production but do not shed light on the underlying cause(s).

The EPInc platform is designed to improve skill production by informing team members about each others’ knowledge gaps (feedback) and organizing team contests to motivate students to invest in learning (improved learning motivation) and share relevant knowledge with their teammates (induced knowledge sharing). We provide empirical evidence for these underlying mechanisms. Since team members know about the student’s knowledge gaps when they attend homework, the student would obtain higher support when more team members do homework with her. We find that a student’s tournament scores improve by 0.15σ ?with an additional teammate attending homework with her than a student in the individual class. In general, students of all abilities benefit from attending homework with both high and low-ability teammates.??

Since teams are balanced in team members’ abilities, all teams should have similar chances of winning the tournaments. We find evidence of this in our data -- class ranks of teams vary significantly across tournaments. In contrast, high- (low-) ability students consistently get high (low) ranks in individual classes. With a high chance of being a part of the winning team, low-ability students may invest more in learning in the team classes, especially if their team had obtained high ranks in past tournaments. We find that a team student scored 0.33σ higher than the individual student if her team ranked in the top half of the class in the previous tournament. The effect of high rank in past tournaments is more pronounced in students in the bottom and middle ability terciles. This provides evidence that team-based incentives promote higher investment in learning among team members.

Finally, we examine the effect of interaction with teammates during homework on student’s tournament scores. The EPInc platform asks students to declare the name(s) of teammates with whom they discussed math topics before personalized homework. We use this information to compute the average number of teammates a student interacts with per homework before a tournament. Using team assignment as an instrumental variable for teammate interactions, we find that interacting with an additional teammate would improve a student’s tournament score by 1.43σ. [1] This provides direct evidence that peer interactions - probably knowledge sharing among team members -- drive students’ skill production.

Overall, we find strong evidence that inducing peer-led knowledge sharing with team-based incentives inside classrooms can result in significantly higher and more equitable skill production in K-12 classrooms. While the effect of technology -- especially personalization technologies -- in upgrading educational production is well-studied, we show how inducing peers to help and support students while using these technologies can further improve skill production. Thus, our findings contribute to the literature on peer effects and the role of technology in education. Our proposed intervention requires very little investment, as it works with the existing educational resources, curriculum, and pedagogy in K-12 schools. It can be implemented at scale to improve skill production in K-12 classrooms, especially in under-resourced educational systems. Our platform can also easily integrate other entities in the K-12 educational ecosystem (such as parents, teachers, and educators) with peers to further improve skill production.????

[1] On average, a student interacts with 0.2 teammates per homework before a tournament in our data