All articles
5 min readlearning / training

61% versus 38%: why mixing practice beats drilling one skill

I design AI trainings the way most trainers do: teach one technique, run ten exercises on it, move to the next technique. People leave the room fluent. Four weeks later someone emails me a problem from their own job and cannot tell which of the four techniques applies. Doug Rohrer's team put a number on that gap, and the fix costs nothing: shuffle the exercises.

Shuffling the worksheet moved scores 23 points.

In 2020, Doug Rohrer, Robert Dedrick, Marissa Hartwig and Chi-Ngai Cheung ran a preregistered cluster randomized trial across 54 seventh-grade mathematics classes 1. Both arms worked the same problems over 4 months. One arm got blocked assignments, a page of one skill at a time. In the other, no two consecutive problems called for the same strategy. Both groups then completed an identical interleaved review. A month later came an unannounced test: 61% for the interleaved group against 38% for the blocked group, d = 0.83. The teachers ran the intervention with no training and backed it afterwards in an anonymous survey.

The 2015 precursor, 126 seventh-graders over 3 months, shows where the benefit lives 2. Tested one day after practice, the interleaved advantage sat at d = 0.42. Tested 30 days after, it grew to d = 0.79. Blocking looks acceptable at the end of a session and decays from there.

A blocked worksheet hands you the strategy before you read the problem. You know the method from the header, so the hardest step never happens. An interleaved set forces the choice that a real exam, or a client's messy request, forces.

Weekly quizzes and batting cages answer the same way.

Faria Sana and Veronica Yan gave 155 students in grades 9 to 12 a weekly science quiz over 4 weeks, covering half the concepts taught that week 3. Some quizzes grouped questions by concept, others mixed concepts together. On a test one month after the final quiz, concepts from blocked quizzes scored 54% and concepts never quizzed scored 47%. Concepts from interleaved quizzes scored 63%, d = 0.35 over blocking. The whole intervention cost 10 to 12 minutes of class time a week.

Motor skills behave the same. Kellie Green Hall, Derek Domingues and Richard Cavazos gave a college baseball team 12 extra batting sessions over 6 weeks, 45 pitches each: 15 fastballs, 15 curveballs, 15 change-ups 4. One group of 10 saw them blocked by pitch type, one saw them in random order, one got no extra practice. On both transfer tests the random group scored above the blocked group, who scored above the control group. Same swings, same cage, a different order.

Interleaving fails on low-similarity material.

Christian Brunmair and Tobias Richter meta-analyzed 59 studies, 238 effect sizes across 158 samples 5. The overall interleaving effect came out at Hedges' g = 0.42, and the average hides the part you need. Paintings and other visual categories: g = 0.67. Mathematics tasks: g = 0.34. Expository texts and tastes: no reliable effect either way. Words: g = -0.39, where blocking wins.

Paulo Carvalho and Robert Goldstone explain the split 6. Interleaving sharpens the contrast between categories, so it pays when categories resemble each other and telling them apart is the hard part. Blocking exposes what the items inside one category share, so it pays when members of a category look nothing alike. In their experiments, interleaved study improved classification of new items for high-similarity categories, and blocked study improved it for low-similarity ones.

The working rule I take from that: mix what a learner could confuse, and keep together what a learner cannot yet see as one family. Four prompting techniques that overlap belong in the same shuffled pool. Vocabulary in a new language does not.

Learners rate the worse schedule higher.

Nate Kornell and Robert Bjork showed participants six paintings by each of 12 artists, blocked for half the artists and interleaved for the rest, then asked them to name the painter of canvases they had never seen 7. Interleaving won on the test. Participants still rated massing as more effective, after their own scores had contradicted them. Performance climbs inside a block, and that climb feels like learning.

Jamie Donenfeld, Zsuzsa Kaldy and Erik Blaser reran the experiment in 2026 with a forced-choice test, which strips out the response biases that the original identification task mixed in with memory 8. The interleaving effect survived.

That gap between the feeling and the result is the operational problem for anyone running a training. Interleave a session and your satisfaction scores can slip while retention rises. I say it out loud in the first ten minutes now.

The protocol.

  1. Count consecutive repeats in your next practice set. More than 3 items in a row calling for the same method, and the order needs a rewrite.
  2. Shuffle within a family, not across the syllabus. Mix the 3 or 4 techniques a learner is likely to confuse. Leave unrelated topics in their own sessions 6.
  3. Teach blocked, practice interleaved. Introduce a technique with 2 or 3 worked examples back to back, then drop it into the mixed pool and leave it there.
  4. Strip the labels. Remove section headers, exercise numbering, any tell about which method applies. The learner choosing the strategy is the mechanism.
  5. Judge the schedule on a test 30 days out. At one day the interleaved advantage was d = 0.42; at thirty days, d = 0.79 2. Scoring a session on how it ends picks the wrong schedule.
  6. Add one 10-minute mixed quiz per week. Sana and Yan bought 9 percentage points with that budget, measured a month after the last quiz 3.
  7. Warn the room. Tell participants the mixed set will feel harder and their sense of progress will drop. Then show them the 61 against 38 a month later.

Start with the worksheet you are already handing out this week. Reorder the problems, delete the headers, and schedule the retest for early October.

Sources.

  1. Rohrer, D., Dedrick, R. F., Hartwig, M. K., & Cheung, C.-N. (2020). A randomized controlled trial of interleaved mathematics practice. Journal of Educational Psychology, 112(1), 40-52. doi.org/10.1037/edu0000367
  2. Rohrer, D., Dedrick, R. F., & Stershic, S. (2015). Interleaved practice improves mathematics learning. Journal of Educational Psychology, 107(3), 900-908. doi.org/10.1037/edu0000001
  3. Sana, F., & Yan, V. X. (2022). Interleaving retrieval practice promotes science learning. Psychological Science, 33(5), 782-788. doi.org/10.1177/09567976211057507
  4. Hall, K. G., Domingues, D. A., & Cavazos, R. (1994). Contextual interference effects with skilled baseball players. Perceptual and Motor Skills, 78(3), 835-841. doi.org/10.2466/pms.1994.78.3.835
  5. Brunmair, M., & Richter, T. (2019). Similarity matters: A meta-analysis of interleaved learning and its moderators. Psychological Bulletin, 145(11), 1029-1051. doi.org/10.1037/bul0000209
  6. Carvalho, P. F., & Goldstone, R. L. (2014). Putting category learning in order: Category structure and temporal arrangement affect the benefit of interleaved over blocked study. Memory & Cognition, 42(3), 481-495. doi.org/10.3758/s13421-013-0371-0
  7. Kornell, N., & Bjork, R. A. (2008). Learning concepts and categories: Is spacing the "enemy of induction"? Psychological Science, 19(6), 585-592. doi.org/10.1111/j.1467-9280.2008.02127.x
  8. Donenfeld, J., Kaldy, Z., & Blaser, E. (2026). Testing the interleaving effect without response bias: A forced-choice reevaluation of Kornell and Bjork (2008). Psychonomic Bulletin & Review. doi.org/10.3758/s13423-026-02944-1

Keep reading