Why mixing problem types in one session can beat practicing one type at a time, what the research shows, and why it feels worse while you do it.
By Bleart Emini, founder · Published · Last reviewed · 7 min read
Most homework is blocked: ten problems on the same rule, in a row, right after you learned it. That feels productive because by problem three you know exactly which method to use. The catch is that you were never asked to decide. Interleaving mixes problem types in one session so that choosing the method becomes part of the practice. The research mostly favors it, with real exceptions, and it nearly always feels worse in the moment. That feeling is part of the story.
Blocked practice groups by type: AAA BBB CCC. Interleaved practice mixes them: ABC BCA CAB. Same problems, different order.
Under blocked practice, the lesson title tells you the answer to "which method?" before you read the problem. On an exam, nothing does. Interleaving trains the step that textbooks usually skip: looking at a problem and working out what kind it is.
It is easy to confuse this with spacing, which means spreading practice across days. They overlap, because mixing types also spaces out each type, but they are different manipulations, and some studies below separate them.
Here is a correct example. Suppose you just finished the product rule, the quotient rule, the chain rule and u-substitution. A blocked set repeats one rule: five product-rule derivatives, then five chain-rule derivatives. A mixed set looks like this:
Problems 1 and 3 both contain "sin" and look alike, yet need different rules. Problems 3 and 4 are nearly the same expression, one to differentiate and one to integrate. A blocked set would let you run on autopilot. The mixed set forces you to read each problem before reaching for a tool.
I am currently in Calculus. I do a few free-response problems, then a quick concept check, and repeat. How many problems varies. I will not oversell this: it is my routine, not an experiment, and I am not claiming it improved anything. A free-response problem makes me carry out a procedure. A concept check, like the f·g question above, asks me to explain why the procedure works.
Math practice. Rohrer and Taylor (2007, Instructional Science) ran two experiments with college students. In the second, students who practiced mixed problem types did far better a week later than students who practiced the same types in blocks. Taylor and Rohrer (2010, Applied Cognitive Psychology) then tested children practicing four kinds of math problems and held spacing constant, so that only the order differed. Interleaving hurt performance during the practice session but roughly doubled scores on a test a day later, and their error analysis suggested the gain came from better matching of each problem to the right procedure.
A larger study ran in real classrooms. Rohrer, Dedrick, Hartwig and Cheung (2020, Journal of Educational Psychology) randomly assigned classes of seventh graders in Florida schools to mixed or blocked worksheets over eight assignments. The What Works Clearinghouse reviewed it and rated it as meeting its standards without reservations. On a test given 43 days later, the mixed group averaged about 62 percent and the blocked group about 38 percent. That is a big gap, but it comes from one study, one grade level and one set of four problem types, tested with an author-developed test.
Category learning. Kornell and Bjork (2008, Psychological Science) had UCLA undergraduates study paintings by different artists, either grouped by artist or mixed, then identify the artist of new paintings. In Experiment 1a, accuracy was 61 percent after mixed study and 35 percent after grouped study. And here is the part to remember: 78 percent of participants did better with mixing, yet 78 percent said grouping was as good or better. People felt that blocked study was working when it was not.
The broader picture, and the caveats. Brunmair and Richter (2019, Psychological Bulletin) combined 59 studies and, in the accepted manuscript of the paper, found a moderate overall benefit of interleaving (Hedges' g = 0.42). It was largest for paintings and other visual materials (0.67) and smaller for mathematical tasks (0.34). For expository texts and tastes, results were ambiguous. For studies using words, blocking did better (g = -0.39). The authors conclude that interleaving can help inductive learning, meaning learning categories from examples, but that the material matters and that it should be used with caution for texts and words.
That last result matters for flashcard users. Memorizing a list of vocabulary definitions is not the same job as telling apart problem types. Interleaving is best supported where the hard part is distinguishing similar things, such as which integration technique, which kind of statistics test, or which artist painted this. For plain fact recall, the evidence does not back mixing as a clear win, and the main driver there is spacing and retrieval, which other guides here cover.
Other limits are worth knowing. Many studies use undergraduates or school-age children, not adults studying alone. Most delays are days to a few weeks, not a semester. In the Kornell and Bjork paper, a contrived fill-in-the-blank word task showed the opposite pattern, with massing better (34 vs 22 percent), so the benefit is not universal. And practice-session performance is worse, which is a real cost if it discourages you.
In both the Taylor and Rohrer and the Kornell and Bjork results, the worse method felt better. Blocked practice gives fast, fluent success. Interleaved practice gives more errors early and a feeling of being slower. If you judge your study method by how it feels during the session, you will pick the weaker one. The right check is a delayed, mixed test, not your mood on problem six.
All study articles · Features · Sign up
Features · Pricing · Explore public decks · Student discount · Study guides · Study articles · Help · Sign up · Log in · Privacy Policy · Terms of Service
Cardvoo runs in your browser and needs JavaScript turned on.