Peer-to-peer tutoring in France

This study evaluates the impact of a peer-to-peer tutoring programme in French and Mathematics, “Duos d’apprentissage”, for French primary students in CM1 (equivalent to 4th grade).

The program was developed by researchers at Aix-Marseille University, specialising in French language teaching, and the University of Geneva, specialising in mathematics education, with the support of the Ampiric pilot centre and the Innovation, Data and Experiments in Education (IDEE) project . It was piloted in 2024-2025 with partner schools from the Ampiric pilot center (Aix-Marseille Academy). The programme will be implemented at scale during the 2025–2026 school year as part of a nationwide randomized controlled trial, allowing for a rigorous evaluation of its effects prior to any broader dissemination within the educational community.

Tutoring pairs will be formed at the beginning of the intervention based on baseline (pre-test) performance. Higher-performing students will be paired with lower-performing peers. The intervention will take place over two six-week periods between February and June 2026, and will include three 25-to-30-minute tutoring sessions per subject per week.

During these sessions, the pairs will have to complete a series of activities in a shared notebook distributed to schools by the research team. In written language, for example, the exercises will cover four areas from the curriculum: reading and text comprehension, vocabulary and lexicon, grammar and syntax, and spelling and written production. In mathematics, particular emphasis will be placed on the numbers and calculations and quantities and measurements components of the curriculum, based on problem solving and metacognitive activities. This programme provides a very structured approach, based on materials designed and structured to be easy for teachers to use and easily reproducible with a high degree of accuracy. The activities are aligned with the CM1 curriculum and fit naturally into the teachers' work.

2026-02-23

2026-06-30

Post-intervention test performance:
Scores on study-specific post-tests administered in class on a platform, under the control of teachers, after completion of the intervention.

Standardized achievement measures:
Scores from national evaluations conducted at the beginning of CM2 (5th grade)

Longer-term academic outcomes:
Medium- and longer-term academic indicators, including subsequent achievement in later grades (e.g., middle school), subject to data availability.

Class climate
Diversity of friendship networks
Self-efficacy
Motivation
Willingness to learn

Recruitment
Schools and classes interested in participating in the study were identified during the first half of the 2025–2026 school year. Participation was voluntary, and schools registered through the Ministry of Education.

All interested schools/classes received instructions to complete the pre-test and student questionnaire. The sample used for randomization consists of all classes that successfully completed the baseline assessments and surveys by the randomization deadline, excluding schools that no longer wished to participate in the programme (see details below).

Data collection

The impact of the intervention will be analysed by collecting data in both groups on: pupils' performance in mathematics and French; their motivation and self-efficacy, measured using validated scales; the classroom atmosphere and friendship networks, measured using validated scales; treatment fidelity (actions carried out by the pairs in the treatment group based on the workbooks distributed); two questionnaires for teachers (at midline and endline).

This data will be collected through pre- and post-tests and surveys using a platform developed for this purpose; it will ultimately be supplemented by the Ministry’s DEPP's national assessments at the beginning of CM1 and CM2. Longer-term monitoring of DEPP data may be considered if the results are conclusive.

Additionally, qualitative interviews with a number of teachers will be conducted to discuss the advantages and challenges of implementing tutoring in their classrooms.

All students will participate in the experiment as part of their Math and French classes. However, data will not be collected for students whose parents have refused the participation in the study.

Randomization
The randomization sample consisted of 78 schools and 92 classes that completed the baseline pre-tests and surveys.

Participating units were randomly assigned to one of two experimental conditions. In the Resources group (control), schools/classes receive French and mathematics workbooks for individual student use but do not implement peer-to-peer tutoring. In the Duos/Tutoring group (treatment), schools/classes receive workbooks containing the same exercises adapted for peer-to-peer tutoring and implement the Duos d’apprentissage tutoring programme.

The unit of randomization was the school whenever multiple eligible classes participated within the same school, in order to avoid within-school contamination. Otherwise, the participating class/school was randomized as a single unit.

Randomization was conducted in six batches to accommodate variation in the timing of baseline data collection. Within each batch, assignment was stratified by académie (a geographical administrative unit in the French education system). When a batch-académie cell contained more than seven schools, the stratum was further split by baseline academic performance, measured as the school-level average of students’ mathematics and French pre-test scores. We aimed to form strata of four schools where possible.

The final randomization procedure produced 20 strata. Six strata contained fewer than four schools. One school was randomized alone using a Bernoulli trial.

After randomization, one class dropped out of the evaluation before the teacher was informed of its assignment. This class belonged to a school with multiple participating classes, so this exclusion did not change the number of schools. One additional school dropped out after randomization but before learning its treatment status. Because both these dropouts occurred with no knowledge of the randomization status, it cannot bias the treatment allocation and we can take those two classes out of the analysis.

The final baseline sample therefore contains 77 schools, of which 38 are assigned to treatment and 39 to control, and 90 classes, of which 46 are assigned to treatment and 44 to control.

Not available

Randomization is conducted in-office using Stata, following pre-defined strata.

School

Yes

77

~1500

Realized: 38 school (46 classes) treatment, 39 schools (44 classes) control

We conducted power calculations with a significance level of 5% and a power level set at 80%. Randomization is performed at the school level, with 50% of the schools being randomly assigned to the treatment group. We expect an attrition rate of 10% and a take-up rate of 90% (with no crossovers). We initially use an R² of 20%, which we consider conservative depending on the outcomes of interest. We set the ICC to 10%. For a sample of 77 classes with an average of 20 students per class, these parameters lead to a minimum detectable effect size (MDES) of 25.5% of a standard deviation (SD). If treatment-take up was 80% instead, the MDES would increase to 28.5%. Conversely, if the proportion of outcome variance explained by baseline covariates (R²) were 40% (keeping take up at 90%), the MEDS would decrease to 22% SD.