It Takes a Class: Peer Learning and Children Social Networks

We evaluate the effect of the introduction of a new pedagogical online platform to teach students digital sciences (“sciences numériques” in French). Note that the introduction of digital science to the elementary school curriculum is a part of national education reform. Thus, control classes will also have a similar curriculum of the same duration (2 hours per week), with the difference that they will not use the peer-learning platform. The platform contains two core features:
1. The “peer help” component encourages students to ask for help whenever they do not understand one of the activities and to help their peers when they need help. When a student needs help, he clicks on a button labeled “J’ai besoin d’aide” (“I need help” in English). This sends a help request to all the peers in the class who can volunteer to help. Note that the help requests are always anonymous. When a student volunteers to help, he physically joins the student who needs help at his desk.
2. The “peer evaluation” component implies that students evaluate each other’s work, without any intervention from teachers. In practice, when a student has finished a task on his computer, he indicates that he needs to be evaluated and the platform randomly assigns a corrector who will physically join the student at his desk to review the task with him/her.
3. Finally, the “cooperative learning” component means that students learn through a combination of individual and collaborative projects. Collaborative projects involve in-person interactions between students who seat next to each other in front of a computer to perform a task.

Students use the new platform for 2 hours each week, which corresponds to about 50 hours of exposure to the peer-learning pedagogy over the academic year. Peer evaluations and cooperative learning not only increase the number of interactions that students have with their peers, but they also affect which students they are interacting with. The random assignment of helpers, graders, and group projects partners encourages interactions students tend to naturally shy away from, such as interactions between boys and girls and between students from different social backgrounds.

2022-01-31

2025-06-30

This project investigates how the adoption of a peer-learning pedagogy affects five main outcomes:
(i) the number and diversity of interactions between students of a class
(ii) the composition of students friendship network over time
(iii) students communication with their class peers
(iv) students cooperation with their peers

Over a period of six months, students in the treated and control groups will take monthly tests to measure the evolution of four main outcomes:

1. The number and diversity of interactions between students of a class. Each month, we ask students to name students they play during class breaks and students they worked with today. We are particularly interested in checking whether peer learning increases interactions between boys and girls, as well as between students from different social and ethnic backgrounds.

2. Network of friends. Each month, we ask students to name their closest friends. Combined with information on gender, social background, and ethnic origin, this information will allow us to document monthly changes in the number of friends for students in the treated and control groups, but also monthly changes in the peers composition over time. We are particularly interested in checking whether peer learning increases friendship between boys and girls, as well as between students from different social and ethnic backgrounds.

3. Communication between students. To measure communication, we have designed a game which we carry-out over a three-month period. We will measure communication using the probability that the students know a fact previously shared with other students from the same class. If that probability is higher in the treated group than in the control group, this shows that peer learning increases communication between students.

4. Cooperation between students. We measure cooperation using a standard cooperation game in experimental economics (the prisoner’s dilemma game). We will investigate whether cooperation increases more over time in the treated group than in the control group.

We evaluate the effect of peer learning using a Randomized Controlled Trial. The first year of the evaluation (academic year 2021/2022), 65 volunteer classes participate in the evaluation. 33 classes are randomly assigned to the treatment group— these classes benefit from a peer-learning pedagogy from January 2021 onward—while 32 classes are assigned to the control group, who do not have access to the peer-learning technology in 2021/2022 academic year. With 25 students per class and an expected consent rate of about 85%, our final sample will contain about 1,380 students.

The experiment will continue during the academic years 2022/2023 with a larger sample of 300 schools with at least one class per school (at least 6,375 students). We will randomly assign schools to the treatment and control to avoid spillover effects between treated and control classes of a given school.

During the two years of the experiment, classes or schools will be assigned to the treatment or control condition using a stratified random lottery with strata defined on the basis of school location and school status.

Not available

Computerized randomization using Stata.

Pilot phase (2021/2022): Randomization is done at the class level.
Scale-up phase (2022/2023): Randomization is done at the school level to avoid within-school between-class spillover effects.

Yes

Pilot phase (2021/2022): 65 clusters, as we will use a stratified randomization at the class level with 65 classes.

Scale-up phase (2022/2023): 300 clusters, as we will use a stratified randomization at the school level with 300 schools.

Pilot phase (2021/2022): With 25 students per class, an expected consent rate of 85% and 65 classes, our final sample will contain about 1380 students. Scale-up phase (2022/2023): With 25 students per class, an expected consent rate of 85%, 300 schools and at least one class per school, our final sample will contain at least 6,375 students.

There is only one treatment arm, so answer is the same as above.

For the pilot phase, the minimum detectable effect on the number of classmates that students report as close friends is +0.717 with the following assumptions: - The randomization is conducted at the class level, with 65 classes and 25 students per and an expected consent rate of 85%. - Mean number of close friends in control classes: 4.730; standard deviation: 3.184. - Intra-class correlation: 0.059. - Power: 0.8; alpha: 0.05. The minimum detectable effect on the number of coins that students contribute to the common moneybox is +0.051 with the following assumptions: - The randomization is conducted at the class level, with 65 classes and 25 students per and an expected consent rate of 85%. - Mean number of coins in control classes: 0.720; standard deviation: 0.269. - Intra-class correlation: 0.027. - Power: 0.8; alpha: 0.05. The minimum detectable effect on the number of rotating facts that students guessed correctly is +0.098 with the following assumptions: - The randomization is conducted at the class level, with 65 classes and 25 students per and an expected consent rate of 85%. - Mean number of facts guessed in control classes: 1.794; standard deviation: 0.408. - Intra-class correlation: 0.073. - Power: 0.8; alpha: 0.05. We will compute minimum detectable effects for the scale-up phase of the experiment based on the data collected during the pilot phase.