Experimental evidence on data-driven remedial instruction in the Dominican Republic
How to cite
Lopez, Carolina; Pineda, Astrid. 2023. "Experimental evidence on data-driven remedial instruction in the Dominican Republic." AEA RCT Registry. AEARCTR-0012383. https://doi.org/10.1257/rct.12383-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
Learning losses due to COVID-19 have been substantial, especially for students coming from low socioeconomic backgrounds, further worsening existing learning deficits in many developing countries. To address these losses, the use of tutoring and computer-assisted instruction holds promise for accelerating learning recovery. However, there is limited knowledge on how education systems can effectively implement these approaches at scale. Computer adaptive learning (CAL) softwares are particularly noteworthy for their ability to cater to students' individual learning levels. However, most evidence on CAL is based on after-school settings and primary-school-aged children, making it challenging to extrapolate to older students or in-school settings. Tutoring, while effective, faces scalability challenges due to cost and availability of qualified tutors. In this proof-of-concept, we aim to evaluate an in-school intervention combining CAL with group tutoring as a potentially more scalable alternative to accelerate learning recovery among teenagers. Moreover, we aim to generate evidence on whether using CAL-generated data for targeted tutoring and teacher support can lead to better student outcomes. We plan to use in-depth data, including on teaching practices and teachers’ time use, to investigate the underlying mechanisms behind the observed changes.
Intervention
In this proof-of-concept, we aim to evaluate an in-school intervention combining the use of a Computer Adaptive Learning (CAL) software with group tutoring as a potentially scalable alternative to accelerate learning recovery among teenagers. Moreover, we aim to generate evidence on whether using CAL-generated data for targeted tutoring and teacher support can lead to better student outcomes. The intervention has two components:
Component #1 introduces the use of CAL as part of schools’ regular instruction, substituting two out of seven weekly hours scheduled for math instruction. The software selected has been adapted to the Dominican Republic’s math curriculum. This arm includes “CAL aides” to support teachers in leveraging the rich source of data the platform provides to assess student learning and adapt their instruction. CAL aides will meet for one hour every two weeks with teachers to support them in accessing and monitoring students’ assessments, tracking their progress, identifying topics that students do not yet master for lesson planning, and using resources offered by the software to ease their workload (e.g., assigning and grading homework and tests).
Component #2 introduces targeted, data-driven (based on CAL software data) small-group tutoring. This component aims to explore complementarities between CAL and more
traditional forms of instruction, such as tutoring, by providing up to three weekly hours of tutoring to high-need students. Students participating in tutoring sessions will have access to the platform (component #1), and tutors will be instructed to use the data to monitor student progress, identify learning gaps, and plan their lessons. These lessons will be implemented during school hours devoted to extracurricular activities.
Experimental design
Classroom-level assignment, participating classrooms will be randomly assigned to two groups:
1. Treatment. Students and teachers in selected classrooms will receive access to the CAL software and general training on its use (component #1).
2. Control group. The control group will not receive access to the CAL software..
Our design would follow a matched pair randomization design, where classrooms would be matched based on their size. One of each classroom in a pair would be randomly assigned to treatment. In addition, among classes in the treatment arm, we will cross-randomize component #2 (covering 20% of the students in each selected classroom with the lowest math performance).
Each classroom will receive one of three treatment combinations: no intervention (pure control), CAL software without tutoring, and CAL software with tutoring for students with the lowest math performance (bottom 20%).
Design details
Cited by
No citation data available for this registration yet.