Spacing Effects in Learning: A Temporal Ridgeline of Optimal Retention

The spacing effect, where study split across separated sessions beats the same study crammed together, was already well established before this paper. What Cepeda and colleagues added was a map of the relationship between two things people usually treat separately: the gap between study sessions and how long you need to remember the material afterward.

The study design

More than 1,350 participants learned a set of obscure trivia facts. They then returned for a single review session after a gap ranging from a few minutes to about 3.5 months. A final test followed at a further delay of up to one year. Because the researchers crossed many gap lengths against many test delays, they could chart performance across the full surface rather than at one convenient point. They called the line of best gaps across that surface a temporal ridgeline.

The core finding

At any fixed test delay, lengthening the gap between study and review first raised final-test performance and then gradually lowered it. There is an optimum, and crucially that optimum moves. Longer retention intervals call for longer absolute gaps.

The proportion is the more useful number. Measured as a fraction of the test delay, the optimal gap declined from roughly 20 to 40 percent at a one-week delay down to roughly 5 to 10 percent at a one-year delay. So the often-quoted “review at about 10 to 20 percent of your target retention interval” is a reasonable rule of thumb for moderate horizons, but it is a simplification. The ratio is not constant. It shrinks as the horizon stretches, and for a full year the optimal gap is closer to 5 to 10 percent.

What this is not

The study did not test an expanding schedule like 1, 3, 8, 20, 50 days, and it does not endorse that specific sequence. Those numbers come from spaced-repetition heuristics in the Leitner and Pimsleur tradition, not from this paper. Cepeda and colleagues used a single review, so their work speaks to the gap-to-retention ratio, not to the shape of a multi-session expanding ladder. The paper also found that within their range, a wide band of gaps near the peak worked nearly as well, so precision matters less than avoiding the extremes of cramming or waiting far too long.

Why it matters for scheduling review

For tutoring and ed-tech, the lesson is that review timing should be indexed to the goal. Scheduling a review for an exam next month is a different problem from building knowledge meant to last a year, and a single fixed interval cannot serve both. Systems that ask when retention is needed, then place review at a modest and decreasing fraction of that horizon, will tend to land near the ridgeline. Getting the gap roughly right is far better than massing study, and slightly too long beats slightly too short.