Spaced Repetition

Spaced repetition (the distributed practice effect) is the robust finding that study time spread across separated sessions beats the same study time massed into one session — and the machine built on that finding: an algorithm schedules each item’s next review at an interval calibrated to the learner’s demonstrated recall, so items are revisited just before they would be forgotten.

The core evidence

• The effect is large and old. Demonstrated since Ebbinghaus (1885) and Jost (1897). Cepeda, Pashler, Vul, Wixted & Rohrer’s 2006 meta-analysis pooled 839 assessments of distributed practice across 317 experiments in 184 articles; across the 254 studies (14,000+ participants) reviewed in Dunlosky et al. 2013, students recalled 47% after spaced study vs. 37% after massed study. ¹
• Top-rated technique. Dunlosky et al. (2013, Psychological Science in the Public Interest) evaluated 10 popular learning techniques across four generalizability dimensions (learning conditions, student characteristics, materials, criterion tasks). Distributed practice and practice testing were the only two rated “high utility” — they work across ages, materials, and long delays, and are easy to implement. Rereading and highlighting — students’ two most-used techniques — rated low utility. ²

Key findings for practice

1. Optimal spacing scales with how long you want to remember. The inter-study interval (ISI) producing maximal retention increases as the retention interval increases: cramming works for tomorrow’s test; spaced weeks-apart reviews win for retention months out. ISI and retention interval operate jointly — a core Cepeda et al. result. ³
2. Expanding intervals look promising but the evidence is thin. Intuition (and Landauer–Bjork) says re-learn at progressively longer gaps. Cepeda et al. found expanding ISIs generally beat fixed ones, but with large between-study variability and possible confounds (feedback provision), so conclusions are tentative. Modern SRS algorithms (SM-2 and descendants) bake this in anyway.
3. Spacing alone is weaker than spacing + retrieval. The “reminding” account holds that a later presentation works partly by forcing retrieval of the first episode — which is why spaced repetition systems pair scheduling with active recall (flashcards, quizzes), not passive re-exposure.
4. Learners underrate it. Students rate their learning as higher after massed study even after experiencing spacing’s benefits (Kornell & Bjork, 2008) — the massed session feels fluent, a metacognitive illusion. Dunlosky et al. note students may need demonstration plus instruction to be convinced.

Why it works (candidate mechanisms)

Multiple mechanisms likely contribute (Dunlosky et al. survey the live theories):

• Deficient processing — back-to-back re-study invites shallow processing; the learner works too easily and is misled by fluency.
• Reminding / retrieval — spaced re-encounters force retrieval of the first episode, itself a powerful memory enhancer (the testing effect — the other “high utility” technique in Dunlosky et al.).
• Consolidation — later study benefits from offline consolidation of the first trace.
• Encoding variability — spaced encounters occur in varied contexts, yielding richer retrieval cues.

From effect to algorithm

An SRS implementation is the effect operationalized: each item carries state (last review, interval, ease/lapse history); a correct recall multiplies the interval (≈1–2.5×), a failure resets or shortens it. The result approximates the “review at the point of forgetting” ideal the ISI–retention-interval curve implies. Anki (SM-2), SuperMemo, and Duolingo’s variant all implement this family. The anti-pileup rule matters operationally: missed days compound the due queue, and unmanageable queues are the main reason learners abandon SRS — matching the metacognitive-discipline finding above.

Open questions

• Non-verbal materials: the meta-analytic base skews to verbal recall; less is settled for procedural/motor skills and deep conceptual material.
• Expanding vs. fixed schedules at long retention intervals — genuinely unresolved in the 2006 synthesis and not much clearer since.
• Optimal intervals per-domain: the ISI × retention-interval interaction means no universal schedule; per-learner adaptation (modern ML-flavored SRS) is an active area.

Related pages

• gonzo-futurism — the OODA loop is a tempo argument about learning under uncertainty; SRS is the same bet applied to memory: short, frequent, feedback-coupled cycles beat heroic cramming
• filter-bubble — both pages are about exposure scheduling: algorithms decide what you re-encounter and when, for retention or for engagement
• post-normal-times — personal knowledge infrastructure (SRS, this wiki) as an individual-scale answer to institutional knowledge decay

Sources

• Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). “Improving Students’ Learning With Effective Learning Techniques.” Psychological Science in the Public Interest, 14(1), 4–58. 2013
• Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). “Distributed practice in verbal recall tasks: A review and quantitative synthesis.” Psychological Bulletin, 132(3), 354–380. 2006

Footnotes

1. 2006 ↩

2. 2013 ↩

3. 2006 ↩