HSW-0265 · How Studying Works
A learner solves the same kind of problem five times and becomes fast.
Then the surface changes.
The numbers look different, the wording changes, or the relevant rule is hidden inside a new context—and the method disappears.
Another learner studies several worked examples instead of repeatedly solving one problem. A third learner retrieves the rule across several different examples. Which route builds better transfer?
The answer depends partly on something study advice often treats as background: how variable the practice is and how much guidance the learner had before practice began.
This article owns the narrow interaction between practice variability, retrieval practice, worked examples and prior instruction when the goal is generalisation. The general retrieval mechanism remains with How Retrieval Practice Works. Worked examples remain with How Worked Examples Work for Performance. Variable retrieval remains with How Variable Retrieval Works. Here the job is to understand why those ingredients can change one another’s value.
Quick Answer
Practice is not defined only by the method label. Retrieval can mean repeatedly recalling the same answer or retrieving the same underlying rule across different problems. Worked examples can mean studying one repeated solution or comparing several examples that share a deep structure.
A 2026 intervention study by Meng Cao and Paulo Carvalho directly crossed retrieval practice versus worked examples with repeated versus varied problems. When learners first received instruction, retrieval practice produced better later generalisation than worked examples whether the practice problems repeated or varied. Without initial instruction, however, repeated worked examples outperformed repeated retrieval practice; varying the retrieval problems removed that disadvantage, producing generalisation comparable to varied worked examples. See Cao and Carvalho, 2026.
The important lesson is not that one method won. It is that the learning process changed when the learner had to induce the rule rather than merely retrieve and apply an already taught rule.
1. Repeating a Problem and Repeating a Principle Are Different
Suppose a student practises the distributive law only through problems that look like:
3(x + 4)
Five repetitions may make that surface familiar.
Now compare a varied set:
- 3(x + 4)
- 5(2y − 1)
- −2(a + b)
- 7p + 7q rewritten as a product
- an area model that requires the same distributive relation
The second set repeats the principle while changing its surface.
This forces the learner to decide what is invariant.
2. Variability Slows Practice for a Reason
Varied practice often feels worse in the moment.
The learner cannot simply reuse the last answer pattern. Each new item requires comparison, classification and rule selection.
In the 2026 study, learners improved more slowly during varied retrieval practice than during repeated retrieval. That slower practice did not necessarily signal worse learning. In the condition without initial instruction, variability improved the generalisation value of retrieval enough that varied retrieval became comparable with varied worked examples on the final transfer test.
Immediate fluency and later generalisation can therefore move differently.
3. Initial Instruction Changes the Cognitive Job
If a learner is explicitly taught the rule first, practice can focus on remembering, selecting and executing that rule.
Without initial instruction, the learner must do more:
- notice recurring features;
- compare cases;
- infer the hidden rule;
- distinguish relevant from irrelevant variation;
- test whether the inferred rule survives a new case.
That is induction, not mere retrieval.
A study strategy that is excellent for remembering an established rule can be weaker when the learner has not yet formed the rule worth remembering.
4. Why Worked Examples Can Win Before the Schema Exists
A worked example reduces search by showing a valid route.
For a novice who has not yet inferred the structure, repeated unguided problem solving can consume effort on local moves rather than schema formation. The learner may repeatedly retrieve a fragile or incomplete procedure.
A worked solution can expose:
- the relevant representation;
- the decision point;
- the order of operations;
- the condition that makes the method valid;
- the check that confirms the result.
Once the schema is available, retrieval practice can become more productive because there is now a coherent structure to retrieve.
5. Why Variability Can Rescue Retrieval During Rule Discovery
Repeated retrieval of one problem can strengthen one problem.
Varied retrieval forces the learner to carry the rule across changing instances. That creates evidence about which features matter.
In Cao and Carvalho’s second experiment, where no initial instruction was given, repeated retrieval practice produced poorer generalisation than repeated worked examples. But when retrieval used varied items, its final generalisation was comparable with varied worked examples.
The authors interpret this as evidence that variability can broaden what retrieval practice preserves, supporting induction and generalisation when learners must extract the rule themselves.
6. Do Not Turn One Two-Experiment Study Into a Universal Curriculum Law
The study used adults learning artificial-language and mathematical rules. It did not test every school subject, every age group or long-term classroom sequence.
Its strength is causal clarity about a specific interaction. Its limit is scope.
The educational application is therefore a design hypothesis to test, not a licence to declare:
- “Retrieval is always better after instruction.”
- “Worked examples are always better for novices.”
- “Variation is always good.”
- “If a learner struggles during varied practice, learning must be deeper.”
7. Mathematics: Vary the Structure Carefully
Suppose a learner has just been taught simultaneous equations.
Early practice may hold the structure stable while changing coefficients so the learner can establish the elimination procedure.
Later variability can change:
- whether substitution or elimination is efficient;
- whether equations are already aligned;
- whether fractions appear;
- whether a word problem must first be represented;
- whether one equation is redundant or inconsistent.
The learning job shifts from “execute elimination” to “recognise the family of problems and select a valid route”.
8. English: Vary the Passage, Preserve the Reasoning Job
A learner practising inference should not answer the same passage repeatedly.
Instead, preserve the reasoning relation while changing the surface:
- narrative motive;
- speaker attitude;
- cause implied across two sentences;
- meaning inferred from contrast;
- tone inferred from word choice.
The varied cases teach the learner what counts as evidence for inference rather than what one passage’s answer looks like.
9. Science: Vary the Context, Keep the Mechanism Honest
For heat transfer, students can retrieve the same mechanism across a saucepan, house insulation, animal adaptation and laboratory apparatus.
But variation should not erase boundary conditions. A learner who says “heat always rises” has generalised a slogan, not a mechanism.
Good varied practice changes context while requiring the learner to identify which physical process is actually present.
10. The Variability Ladder
Variation can be staged rather than dumped onto the learner at once.
- Exact repetition: establish the basic route.
- Parameter variation: change numbers, names or superficial details.
- Representation variation: move among equations, diagrams, prose or tables.
- Method-choice variation: mix neighbouring problem types.
- Context variation: place the same principle in a new setting.
- Boundary variation: include cases where the familiar rule should not be used.
- Unsignalled transfer: remove the label and let the learner decide what applies.
11. Retrieval Practice Needs Something Worth Retrieving
Retrieval practice is often presented as content-neutral: close the book and recall.
But the quality of retrieval depends on the representation already built.
If the learner stored an answer pattern without its conditions, retrieval can make the wrong abstraction more fluent.
Before increasing retrieval volume, ask:
- What is the invariant?
- What conditions make the method valid?
- What nearby method could be confused with it?
- What variation would reveal whether the learner actually sees the structure?
12. Worked Examples Need Active Comparison Too
Varied worked examples are not automatically educational simply because they are varied.
If the learner reads each solution as an isolated answer, the shared structure may remain invisible.
Comparison prompts can help:
- What stayed the same?
- What changed?
- Which step changed because the problem changed?
- Which step was invariant?
- What would make this method invalid?
13. Practice Variability vs Interleaving
Variability and interleaving overlap but are not identical.
Variation can occur within one problem family. Interleaving typically mixes distinct categories or methods so learners must discriminate among them.
A varied retrieval set can therefore be blocked by principle while still changing surface instances. Later, interleaving can add method-selection demands.
14. Practice Variability vs Variable Retrieval
Variable Retrieval owns the broader idea of changing cues so knowledge can be reached through more than one route.
This article is narrower: it asks how variation changes the relative value of retrieval practice and worked examples, especially when learners have or have not received initial instruction.
15. A Four-Question Design Check
- Does the learner already know the rule?
- Is the goal retention, fluency, discrimination or generalisation?
- Does practice vary the right features while preserving the invariant?
- Will the final test require independent rule selection under changed surface conditions?
These questions are often more useful than asking whether “retrieval practice” or “worked examples” is the superior method in the abstract.
16. Parent and Tutor Guide
If a student can solve ten near-identical questions but fails an unfamiliar one, do not immediately add twenty more near-identical questions.
Ask the student to explain:
- what makes all the successful questions members of the same family;
- which feature actually triggers the method;
- what could change without changing the method;
- what change would require a different method.
If those answers are weak, add comparison and variation before adding volume.
17. Delayed Transfer Check
After practice, wait until the examples are no longer fresh.
- Present a new surface form.
- Do not label the method.
- Include at least one plausible competing method.
- Require the learner to justify the choice before executing it.
- Score method selection and execution separately.
If the learner can execute only after being told the method, procedural fluency may have grown without generalisation.
18. Evidence Boundary
The 2026 Cao and Carvalho experiments provide strong evidence for a specific interaction among initial instruction, practice method and item variability in adult rule-learning tasks. They do not establish a universal sequence for every curriculum.
The broader variability literature supports the idea that varied instances can help abstraction and transfer, but benefits depend on what varies, how much prior knowledge exists, how complex the task is and whether learners can identify the invariant structure.
Educationally, variability should therefore be treated as a design variable rather than a virtue in itself.
19. Return: Change the Surface Until the Learner Finds the Rule
Repetition can build reliability.
Variation can reveal structure.
Worked examples can make a route visible. Retrieval can make the learner carry it.
The art is not choosing one forever. It is matching the method to the learner’s current job.
Teach enough structure that retrieval has something coherent to recover. Vary enough that the learner must find the invariant. Reduce guidance. Delay. Then test whether the rule survives a problem that no longer looks like practice.
Continue through How Retrieval Practice Works, How Worked Examples Work for Performance, How Variable Retrieval Works, the How Studying Works Numbered Series Reading Index and the How X Works Hub.