The spacing effect, one of the most robust phenomena in learning research, refers to the finding that memory is enhanced when study sessions are distributed over time rather than massed together. First reported by Ebbinghaus (1885) and confirmed in hundreds of subsequent studies, the spacing effect has been formalized in several mathematical frameworks that attempt to explain both why spacing helps and how much spacing is optimal.
Encoding Variability Theory
The encoding variability account (Estes, 1955; Bower, 1972) proposes that each study episode encodes a sample of contextual features along with the target item. Spaced presentations occur in more diverse contexts, creating a richer set of retrieval cues. The probability of recall is related to the overlap between the test context and at least one stored context:
With massed practice, the study contexts are nearly identical, providing redundant cues. With spaced practice, the diverse contexts increase the probability that the test context will match at least one study context.
Study-Phase Retrieval Theory
An alternative account emphasizes study-phase retrieval (Thios & D'Agostino, 1976; Benjamin & Tullis, 2010): when an item is presented for the second time after a delay, the learner must retrieve the first presentation, and this retrieval act itself strengthens the memory. The optimal spacing interval is one where retrieval is still possible but effortful:
where S(t) is the current memory strength, r(t) is the retrieval success probability at the time of the second study, and α is a learning rate. This "desirable difficulty" framework predicts that spacing should be challenging enough to require effort but not so long that retrieval fails entirely.
ACT-R and the Optimal Spacing Function
Pavlik and Anderson (2005) developed a model within the ACT-R framework that provides a quantitative account of spacing effects. Each practice trial updates the chunk's base-level activation, but the activation gain depends on the current activation at the time of practice. When activation is high (massed practice), the benefit is small; when activation has decayed (spaced practice), the benefit is large:
This model predicts an optimal spacing function: the ideal inter-study interval increases with the desired retention interval, a prediction confirmed by Cepeda et al. (2008), who found that the optimal gap is roughly 10-30% of the test delay.
Practical Implications
Mathematical models of spacing have direct implications for educational technology. Spaced repetition systems such as SuperMemo and Anki use algorithms derived from these models to schedule review of flashcards at expanding intervals. The Leitner system uses a simplified version of this principle, while more sophisticated algorithms estimate individual item difficulty and learner parameters to optimize the scheduling of thousands of items.
A crucial finding is that the optimal spacing interval depends on when you will be tested. Short spacings are optimal for short retention intervals, while long spacings are optimal for long retention intervals. This interaction rules out any single-mechanism explanation and motivates multi-process models that combine encoding variability, study-phase retrieval, and forgetting dynamics.