Joeri Dros spent about 70 hours learning strings of nonsense syllables and relearning them after varying gaps: 20 minutes, one hour, nine hours, one day, two days, six days and 31 days. Ten lists at each interval, except nine hours, which had nine. Sixty-nine lists in the experimental phase.
He did it to retest the work of Hermann Ebbinghaus, who in 1885 had made himself the subject of exactly this procedure. Jaap Murre and Dros published the result in PLOS ONE in 20151.
The short conclusion: the forgetting curve replicates.
Why rerun an experiment everybody already believes
The answer is technical. There had been earlier replications, but they were mostly very old studies presenting items at 2 seconds each, where Ebbinghaus learned at 0.4 seconds per item. That gap is wide enough that nobody could be sure the two were measuring the same thing.
Murre and Dros followed the original method closely, using Ebbinghaus's own measure: savings, the share of time saved when relearning compared with learning the first time. Four curves — Ebbinghaus's, two from another replication, and Dros's — share many characteristics, despite the century between them and despite the new study running in Dutch. The authors call the resemblance remarkable1.
One discrepancy, and what it teaches
The one mismatch sits at 31 days. Dros's savings came in at 0.0410, where the other three curves are around 0.20.
The authors traced the cause: Dros's learning time rose steadily over the course of the study. Corrected for that trend, the figure becomes 0.137 — closer, still lower.
That detail matters more to training designers than it first appears. Someone grinding through 70 hours of the same task gets worse at the task itself. Anyone who has ever designed a four-day consecutive programme should pause there.
The bump at 24 hours
The interesting finding is not that the curve goes down. Everyone knows memory fades.
The finding is that the curve is not smooth. The authors conclude it most probably shows a jump upwards starting at the 24-hour data point. The fitted boost was 0.030 for Ebbinghaus's data and 0.131 for another participant.
The natural explanation is sleep consolidating memory, which would lift the 24-hour point above what a smooth curve predicts. The authors raise that possibility and say directly that it remains to be established1.
But the forgetting curve is not the useful part
This is where most training decks go wrong. Someone shows the curve, says learners forget a given percentage within a day, and concludes that you need reinforcement. True, but it leaves the real question unanswered: reinforcement when?
That question has an answer. In 2006 Nicholas Cepeda and colleagues published a meta-analysis of the spacing effect in Psychological Bulletin, covering 839 assessments across 317 experiments drawn from 184 articles2.
Their headline finding is not that spaced beats massed practice, which was already well established. It is that the gap between study sessions and the length of time you need to remember work together: the gap producing the best retention grows as the retention interval grows.
Translating that for whoever books the room
More precisely, in their data:
- Where the material is needed for less than a day, the gap between sessions makes little or no difference.
- Where it is needed for one day, performance rose significantly as the gap widened from 1–15 minutes to one day.
- Where it is needed for 2 to 28 days, a one-day gap clearly beat gaps of 1 to 15 days, with a slight drop beyond one day.
- Where it is needed for over a month, the qualitative evidence suggests the best gap is longer than one day.
And here is the sentence that matters most for companies: for most practical purposes the retention interval is months or years, so the optimal gap will likely be well in excess of one day2.
A selling-skills programme the company expects people to use for the next two years should not have its only refresher 48 hours later.
Where to be honest about the limits
Cepeda's own tables say something that rarely gets quoted: the evidence base is heavily tilted towards very short intervals. At retention intervals of 2–59 seconds there are 174 performance comparisons; at 1 minute to 2 hours, 259. At one day it drops to 27. At 30 days or more there are 23 comparisons, 34 data points and 19 effect sizes2.
Memory science has studied remembering for minutes in great depth, and remembering for months only thinly — which is exactly the range companies care about. The authors say plainly that much more detailed study is needed here.
When somebody hands you a reinforcement schedule precise to the day, ask where the numbers came from.
Redesigning a programme you already run
Say you run a two-day supervisory skills course each quarter and expect people to use it for at least a year.
The evidence-based rewrite is not to chop two days into ten short sessions. It is to keep the live practice, then place the follow-up touchpoints at widening intervals: one after a few days, one after a few weeks, one after a few months. Each touchpoint has to make people retrieve rather than reread, because relearning effort is the measure in both studies above.
And because the long-interval data is thin, measure your own results rather than trusting a borrowed number.
