Microlearning statistics, with the studies behind them
Spacing and recall tests have supporting research. That does not give every short lesson a fixed retention gain. Check what a study measured, who took part and what the comparison actually was.
Learning Cadence editorial, , updated , 6 min readHow we work
A useful learning statistic names its population, activity, comparison, measure and follow-up period. A percentage without those details cannot tell you whether a particular lesson or app will help.
This is a selected reading of primary studies, not a systematic review or a trial of Cadence. Below, each finding separates the result from its practical limits. The worked example then shows how to check a headline before using it.
What the research says
Findings, with the study behind each one.
Where a widely repeated number has no traceable source, this article says so instead of repeating it.
01
One replication measured relearning savings, not all everyday memory
Plain-language finding: In the 2015 replication, one participant learned and relearned lists of nonsense syllables; the outcome was savings in relearning effort.
Savings means less work needed to relearn previously studied material. It is not the percentage of a workplace lesson remembered.
Study: Murre, J. M. J., and Dros, J. (2015). Replication and Analysis of Ebbinghaus' Forgetting Curve. PLOS ONE, 10(7).
What it means in practice: Do not use a schematic forgetting curve to predict exactly what a particular learner will lose tomorrow.
A large review investigated distributed verbal practice
Plain-language finding: The 2006 review included 839 assessments from 317 experiments across 184 articles.
Spacing and the delay before testing jointly affected recall. The spacing associated with best retention changed with the retention interval.
Study: Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., and Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132, 354 to 380.
What it means in practice: This supports investigating review timing. It does not rank every short lesson above every long one.
Plain-language finding: The 2008 study taught facts to more than 1,350 people and varied review and test delays.
Optimal gaps were roughly 20 to 40% of a one-week test delay and 5 to 10% of a one-year delay. Those are findings from this task, not a personal scheduling formula.
Study: Cepeda, N. J., Vul, E., Rohrer, D., Wixted, J. T., and Pashler, H. (2008). Spacing effects in learning: A temporal ridgeline of optimal retention. Psychological Science, 19, 1095 to 1102.
What it means in practice: Record your actual recall and adjust review. A calendar recommendation needs the task, initial knowledge and desired retention period.
Recall tests helped delayed retention in a particular comparison
Plain-language finding: In two 2006 experiments, students studied prose and then restudied or took recall tests.
Restudy performed better after five minutes. Prior testing performed better after two days and one week.
Study: Roediger, H. L., and Karpicke, J. D. (2006). Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention. Psychological Science, 17(3).
What it means in practice: A check can both assess and support learning. It does not guarantee mastery or replace explanation and corrective feedback.
Plain-language finding: A 2009 paper reports up to 150% better final recall with an optimal gap in its laboratory studies.
The two studies used three sessions with vocabulary, facts and names of visual objects. This is a specific spacing result, not a universal microlearning improvement.
Study: Cepeda and colleagues (2009). Optimizing distributed practice: theoretical analysis and practical implications. Experimental Psychology, 56(4), 236 to 246.
What it means in practice: Do not reject or repeat a large percentage on appearance alone. Check the measure, comparison, population and delay.
Original hypothetical example: a comparison group recalls 40 of 100 items and another recalls 60 of 100. The recall rates are 40% and 60%. The difference is 20 percentage points. Relative to the first group, the increase is (60 - 40) / 40 = 50%.
Both calculations can describe the same result. Neither means everyone remembered half more of every real-world lesson. You still need to know how groups were assigned, what they studied, when they were tested and how uncertain the estimates are.
Self-check: if recall rises from 20% to 30%, the difference is 10 percentage points and the relative increase is 50%. If the starting value is zero, that relative calculation is undefined. These are invented arithmetic examples, not study results.
In practice
Build a claim card before citing a number.
Copy the original study link and write five short fields. If the citation leads only to another summary, continue to the original or leave the number out. Do not convert an unanswered search into proof that a study cannot exist.
Population: who or what was studied, and who was not?
Task and comparison: what actually differed between conditions?
Outcome: recall, completion, return visits and workplace performance are different measures.
Timing: when was the outcome measured, and how long did the activity last?
Uncertainty and fit: what limits the result, and how closely does it match your decision?
In practice
Compare like with like.
A five-minute module and a six-week course offer different commitments. Their completion percentages alone cannot show which teaches better. Likewise, returning to an app is not the same measure as remembering or applying an idea.
The 150% result above illustrates why context matters. A real laboratory finding can still be misused as a claim about every microlearning product. Read the study that supports the exact claim you intend to make.
In practice
What this means for Cadence.
We make Cadence. Its short lessons, saved course progress and checks provide ways to organize self-directed learning. These studies did not test Cadence and do not establish that a particular lesson length is optimal.
Use the public exercises below to inspect an explanation and try a check. For a workplace skill, later use in an appropriate real situation matters too. Accredited or safety-critical work may require formal teaching, qualified supervision and a different kind of assessment.
Put it into practice
Start it today, one lesson at a time.
If you want to keep going, Cadence turns what brought you here into a course with practice and saved progress. Learn at your own pace. Cadence is our app.
Does research prove a fixed microlearning retention gain?
No fixed gain follows from the selected studies here. Their tasks, comparisons and outcome windows differ. Use the original study for a claim about its own result, not a generic app percentage.
Can a 150% improvement have a real source?
Yes. The linked 2009 spacing paper reports up to 150% better final recall in its laboratory comparisons. That does not establish a 150% gain for every short course or for Cadence.
What is the difference between percent and percentage points?
In the invented example above, 40% to 60% is a 20-percentage-point difference and a 50% relative increase. Always state the starting value and measure.
Does a forgetting curve predict my memory?
The cited replication measured relearning savings for a particular task and participant. It is not a personal prediction of the percentage you will forget.
What should I cite in a business case?
Cite the original paper, with the population, task, comparison, measure and follow-up period. If those do not match your proposal, state the extrapolation and plan to evaluate your own outcome.
See it in concrete terms
Explore example courses for your goal.
If the underlying question interests you, these Cadence courses cover habit design and reading evidence critically.
Selected primary-study summaries organized by task, comparison and limits, plus original arithmetic and a claim-card template. This is not a systematic review.
What we checked directly: On 29 September 2026 we read the author or publisher abstracts for the spacing and retrieval papers and the full 2015 replication methods. The 2009 result corrects our earlier blanket treatment of 150% claims.
Limits of this review: We did not run an original learner experiment, reanalyze data or test Cadence against another product. These examples are planning and interpretation aids, not validated learning interventions.
