The goal of ManyClasses2 is to systematically assess the replicability of the benefits of retrieval practice in routine college settings, using a randomized controlled replication carried out over two academic semesters and embedded in 200+ college classes distributed over 10 institutions. Retrieval practice is one of cognitive science’s principal contributions to our modern understanding of student learning. Decades of research have demonstrated that retrieval practice, the effortful act of trying to retrieve information from memory, helps to improve memory for that information, even better than restudying the same information. However, as an education intervention, the implementation of retrieval practice in classroom settings, and its benefits, are not so consistent. Most of these classroom studies have used non-random experimental designs and/or have compared retrieval practice with depleted no-treatment controls. The low quality of this evidence makes it difficult to identify how different instructional designs moderate the benefits of retrieval practice, which is problematic when considering retrieval practice’s distinctive efficacy in laboratory settings. While retrieval practice surely works, the education community is presently at a loss for programmatic details of where it works, and how to implement it most effectively. The goal of the current proposal is to fill this gap, by systematically replicating a rigorous, paradigmatic retrieval practice experiment across many diverse classroom implementations, and by using these findings to produce structured recommendations for educators to realize the benefits of retrieval practice. We are currently actively seeking funding to support this effort; more information will be posted at a later date.
ManyClasses2 will be implemented using Terracotta, and we will provide support for participating institutions to integrate Terracotta into their Canvas LMS instance.
The target sample for ManyClasses2 (200+ classes) was established from a power analysis demonstrating that the number of classes are the primary limiting factor in inferring the effects of class-level moderators on the benefits of a student-level manipulation.