For teachers, learners, and designers of learning technology, a compact model matters because it promises clear, transferable guidance. If one function can describe learning for short drills and for extended projects, it becomes easier to plan practice that builds lasting skill without relying on fragile rules of thumb. The paper’s focus on an invariant unit of measurement points toward a shared language for comparing studies and for making practical decisions about when to revisit material.
This research raises practical questions about how we schedule review, measure progress, and design tasks so practice pays off across contexts. Readers interested in how a single mathematical principle might change everyday learning will want to explore the full article to see the model, the datasets it was tested on, and what this could mean for more equitable, scalable approaches to teaching and training.
Abstract
Despite their potential to significantly improve the durability of learning and their proven predictive power on many occasions, theories of distributed practice have not yet been widely adopted by educators. The reluctance may be attributed to the enormous strain they impose on learners or the fragmented nature of the evidence, as well as the generality of available prescriptions. The aim of this study is twofold: (1) to propose a unified theoretical model for the effects of practice, forgetting and spacing, which to this day unfortunately still does not exist; and (2) to put this new model to the test in multiple realistic situations. To achieve the first objective, scale invariance was used as a unifying principle and applied to three distinct scales. This approach resulted in a parsimonious equation for the general case of scale-invariant learning (SIL). For the second objective, the model was tested on a large dataset obtained during a locally conducted experiment. In addition, it was applied to five other datasets extracted from reputable and well-known distributed practice experiments. Overall, the goodness of fit of the SIL was acceptable in all cases. It is argued that the SIL model, to the best of our knowledge, appears to be a very robust model. The model allows the same general function to be used for any sequence of tasks—be they simple, complex, short, long, unique, repeated or elaborate. Therefore, it enables strict and real-world application in authentic situations, while proposing an invariant unit of measurement for learning.
