Distributed practice

The basic finding

Hermann Ebbinghaus established the core pattern in the 1880s, memorising nonsense syllables on himself, work he published in 1885 while based in Berlin. He found that spreading repetitions across days required fewer total repetitions to reach the same criterion than massing them in a single session.

The effect has been replicated so many times, across so many materials — vocabulary, prose, mathematics, motor skills, surgical procedures — that it is among the most robust findings in memory research. When researchers at Washington University in St. Louis and elsewhere run meta-analyses, the spacing advantage holds up. The effect size varies, but it is rarely trivial and almost never reverses.

The mechanism behind it remains genuinely contested. One influential account is the encoding variability hypothesis: when you return to material after a gap, your retrieval cues are slightly different from those active during the original encoding, giving you more varied hooks into the memory.

Massed practice can produce performance that looks equivalent, or even superior, immediately after the session.

Another account, associated with work by Robert Bjork at the University of California, Los Angeles, treats the gap itself as a desirable difficulty — forgetting partially before returning to something forces more effortful retrieval, and that effort does consolidation work that massed practice skips. Both accounts predict similar results in most circumstances, which is one reason they have been hard to disentangle empirically.

What the evidence actually shows

Performance during practice is not the same as long-term retention, and the difference is easy to miss. The gap only becomes visible when you return days or weeks later. This is why the spacing effect is so counterintuitive: the experience of massed practice feels effective, because fluency during the session signals mastery that doesn’t last.

The size of the optimal gap is less settled than the existence of the effect. Research suggests the ideal inter-session interval scales with the intended retention interval — a test tomorrow favours shorter gaps, a test in six months favours longer ones. But this relationship is probabilistic and context-dependent; precise prescriptions should be treated with scepticism. What the evidence supports confidently is that almost any spacing is better than none, and that the gains are largest when the retention interval is long.

Endel Tulving’s work on encoding specificity also bears on the limits of the effect: distributed practice helps most when the conditions at practice and test are reasonably matched. If the material itself changes substantially between sessions — as it does in some procedural skills — or if the test format diverges sharply from the practice format, the advantage can shrink.

Strength and limits

The spacing effect replicates unusually well by the standards of cognitive psychology. A 2006 paper by Henry Roediger and Jeffrey Karpicke at Washington University in St. Louis helped sharpen interest in retrieval-based accounts, and subsequent work has generally confirmed that spacing and retrieval practice are complementary rather than redundant. Spacing a series of retrieval attempts is more powerful than spacing a series of re-reads.

The main caveat is transfer. Most evidence for spacing comes from retention of the same material practiced, which is near transfer. Whether distributed practice improves the ability to use knowledge in genuinely novel contexts — far transfer — is much less well-established. The data from laboratory settings, where the material is often word lists or paired associates, also generalise imperfectly to complex, interconnected subject matter where understanding the structure matters as much as retaining individual facts.

A second limit is practical: massed practice is often all that is available. A deadline compresses time; access to a course or instructor is bounded. The spacing effect tells you something important about how memory consolidates, but it cannot conjure extra days. What it does suggest is that, wherever the schedule permits any distribution at all, front-loading revision into one long session is a worse use of that time than most people assume.