What the Learner Is Building: Rules, Schemas, and Retrieval
The choice between retrieval practice and worked examples is a choice about what the learner must build next. Retrieval practice asks someone to bring an answer or procedure back from memory; a worked example lets them study how a solution is organized. The useful judgment is not which activity looks more active or feels more difficult, but whether the learner needs to construct a representation, strengthen access to one, or apply it beyond the example that introduced it.
This distinction gives the solve-first question a more precise mechanism. Problem solving before instruction can expose an incomplete representation that later explanation reorganizes. But if a learner already has the rule, asking them to generate it can serve a different purpose: retrieval can make access more reliable. If the learner has no workable representation at all, repeated requests to retrieve an answer may not be a demanding route to understanding. They may simply make the learner repeat a guess or remember an item without seeing its structure.
Retrieval and Examples Put Work in Different Places
Retrieval practice is an active recall task. The learner sees a prompt, closes the book or moves beyond the worked solution, and tries to produce the relevant fact, relationship, method, or answer. The effort matters because successful recall exercises access to the representation rather than mere recognition of it. A learner who rereads a rule may experience it as familiar. A learner who must state the rule or use it without the example has to find it.
That effort is not magic, and it is not the same as guessing without information. Retrieval works best as part of an environment in which the learner has something meaningful to retrieve and can compare the attempt with corrective information. In the controlled study examined in this series, retrieval trials included immediate feedback that indicated whether the response was correct and showed the correct answer. The resulting evidence is therefore about retrieval practice with feedback. It should not be generalized automatically to every quiz, every delay in feedback, or every situation in which a learner is asked to guess.
A worked example allocates more of the solution generation to the instructional material. The learner studies a problem together with a step-by-step solution and an account of the rule or procedure. That arrangement can reduce the burden of holding a problem in mind while simultaneously deciding what operation to perform, recalling a possible method, carrying out each step, and checking whether the result makes sense. When the learner is still forming a procedure, this reduction can leave more capacity for noticing how the steps fit together.
The two activities therefore strengthen different links in a learning process. Retrieval practice emphasizes access: can the learner bring the relevant knowledge to mind when the supporting display is gone? A worked example emphasizes organization: can the learner see how a solution is assembled and which features of the problem control the steps? The contrast is not absolute. Studying an example can support later recall, and retrieval can contribute to learning a procedure. But the activities put the learner’s limited attention in different places.
A simple scenario makes the difference visible. Suppose learners are trying to understand the distributive property. A worked example might show why an expression such as 2(x + 3) becomes 2x + 6, with each step tied to the multiplication of both terms. The learner can inspect the relationship between the original expression and the transformation. A retrieval prompt might instead ask the learner to transform 4(x + 5) without seeing the solution. That prompt tests whether the learner can access and apply the rule. If the learner already understands the distributive property, the retrieval attempt can strengthen independent access. If they have never recognized why both terms must be multiplied, an incorrect answer may need explanation before another retrieval attempt can teach much.
This is why a correct answer during retrieval is not the only relevant outcome. The learner may remember the output for one item without representing the operation that produced it. Conversely, an error can be useful if feedback identifies the relation that was missed and the learner then applies that relation to a changed case. The learning process depends on the connection among prompt, attempted response, feedback, and subsequent application.
Retrieval practice should also be kept separate from fluency practice. Fluency practice aims to increase speed, accuracy, and automaticity, often through repeated or timed responses. Retrieval practice can eventually contribute to faster access, but its defining feature is active recall, not speed. The controlled experiments used self-paced tasks and did not include a speeded response component. Their findings therefore support conclusions about retention and generalization under those conditions, not about automatic performance or timed mastery.
A worked example can look passive, but “looking at the answer” is not a complete description of how example-based learning works. A learner can passively copy visible steps and miss the reason for them. The value of an example depends on whether the learner attends to the relation between the problem’s features and the solution’s moves. A process-oriented explanation can help, but the evidence considered here does not imply that every worked example is equally clear or that simply increasing the number of examples will produce a schema.
The useful comparison is therefore not active versus passive. Both methods can involve demanding cognitive work, and both can be implemented badly. The useful comparison is which part of learning the activity makes possible. Retrieval gives the learner practice accessing a representation. A worked example can make the organization of a representation inspectable. An open-ended first attempt can reveal a representation that neither activity can safely assume is already present.
Prior Knowledge Changes the Comparison
Cognitive load helps explain why the same problem-solving task can be productive for one learner and overwhelming for another. Working memory can hold only a limited amount of new information at once. In a complex problem, a novice may need to search for a method, remember several conditions, generate a next step, execute it, and monitor the result. If the learner has not yet formed the relevant schema, each part may remain a separate demand. The learner can spend the available attention on managing the task without discovering the structure that the task was meant to reveal.
A worked example can lower some of that unnecessary generation. The learner does not have to invent every step before knowing what kind of operation the problem requires. That guidance can make it easier to inspect the relation among the steps and the underlying principle. This is the logic behind the worked-example advantage often associated with novice learning: guidance can protect cognitive resources while a coherent procedure is being built.
The implication is not that novices should never solve problems. It is that a solve-first attempt must be matched to what the learner can meaningfully represent. A short, structured challenge may expose a useful idea. A complex task that requires several unknown concepts may only produce overload. The distinction is the same one that matters for Productive Failure: difficulty is not productive because it is difficult; it is productive when the learner can generate material that later instruction can organize.
Prior knowledge changes the balance in the other direction as well. Once a learner has a stable representation of a rule, a worked example may spend time showing a route they can already reconstruct. Retrieval practice can then perform a more diagnostic and strengthening function. It asks the learner to access the rule without relying on the visible solution and to use it in a problem. The effort of retrieval may reveal that the learner understands the rule in principle but cannot yet bring it to mind under changed conditions.
This is a reason to resist a single method ranking. The word “novice” itself can be too broad. Someone may be a novice in a subject but already know the mathematical or logical operations needed for the particular task. Another learner may recognize the vocabulary but lack the structure that makes the procedure coherent. What matters is the relevant prior representation for the current problem, not a global label attached to the person.
The Knowledge-Learning-Instruction framework discussed in the evidence distinguishes among learning processes that include memory, induction and refinement, and understanding or sense-making. The categories are useful because they connect the activity to the kind of change expected. Memory-oriented learning asks the learner to encode and retrieve discrete information. Induction requires finding a pattern or rule across cases. Refinement improves a procedure through application and correction. Understanding and sense-making connect parts into an account that can guide reasoning.
Retrieval practice is naturally suited to memory strengthening, but it is not confined to isolated facts. When prompts vary while preserving an underlying structure, retrieval can also give the learner repeated opportunities to apply and compare a rule. The comparison across cases may support induction, especially when feedback makes the relevant relation visible. That does not mean retrieval has stopped being a memory-based activity. It means the scope of what is retrieved and the structure of the prompts can change what the learner is able to generalize.
Worked examples are naturally suited to guided organization, but they are not confined to the first exposure to a concept. A learner can study examples to refine a procedure, compare two solution paths, or notice when a familiar method does not apply. Yet examples may offer less retrieval effort than an independent attempt. If the learning goal is stable access under conditions where the example will not be present, example study alone may leave a gap between recognition and use.
The decision can be made clearer by examining four questions. First, is the underlying rule or procedure available enough for the learner to retrieve, or must it still be constructed? Second, does the task require memory for an item, induction across cases, or application of a known procedure? Third, would independent generation consume so much attention that the learner cannot notice the structure? Fourth, will the learner receive feedback and a new case that reveal whether the representation transfers?
These questions also expose why short-term ease is an unreliable guide. Worked examples often require less time than problem-solving practice because the learner is not generating a complete solution. In the controlled study, retrieval practice required more learning time than studying worked examples, and the difference was larger when the items varied. That extra time did not make retrieval inferior in the condition where initial instruction had already presented the rules. Nor did the greater effort automatically make it better when learners had to infer the rules without initial instruction.
The same study found a different pattern under those two conditions. When learners received explicit rule instruction before practice, retrieval practice led to better final generalization than worked examples, regardless of whether the items were repeated or varied. When initial instruction was removed, worked examples produced better generalization than repeated retrieval, while varied retrieval performed comparably to varied worked examples. These results fit the process distinction: once a rule is available, retrieval can strengthen access; when the rule must be induced, guidance and comparison become more important.
The result does not show that every worked example is better than every retrieval activity for a learner who lacks prior knowledge. The experiments used artificial language and mathematical rules, adult participants, immediate feedback, and immediate generalization tests. The retrieval condition was multiple-choice rather than unrestricted solution generation, and the varied and repeated groups in the no-instruction experiment were not equal in size after exclusions. Those details matter because they define what the comparison can support.
They also show why initial instruction is not a minor preface. It changes the learner’s starting state. With instruction, practice begins after a rule has been made available, so the central demand can be retaining and applying it. Without instruction, practice must help create the rule. A worked example supplies more guidance during that construction. Varied retrieval can partly compensate by giving learners multiple related instances to compare, but repeating one item mainly offers repeated contact with that item.
The broader lesson is that retrieval, examples, and solve-first attempts are not interchangeable containers for “practice.” They distribute cognitive work differently. A first attempt can expose a hypothesis. A worked example can organize a procedure. Retrieval can make an available representation accessible. Variation can widen the set of instances from which the learner must retrieve and compare. Feedback can tell the learner whether the relation they used is the one that transfers.
A learner’s state can also move during one lesson. The learner may begin without a stable rule, study an example or receive an explanation, and then become ready for retrieval. After a successful retrieval, a novel case may reveal that the rule is still too narrow. The appropriate response is not loyalty to the original method. It is a return to the learning job: provide more organization, a better contrast, a corrected attempt, or another retrieval opportunity.
The central comparison is therefore between kinds of support, not between identities. Worked examples are not “easy mode,” and retrieval is not automatically “real learning.” Each can be the right or wrong fit depending on what the learner can currently represent and what the next task requires. Once the rule is available, retrieval can make it usable without the example. Before the rule is available, guidance and meaningful comparison may be necessary to make retrieval possible at all.
The model in the diagram is a set of possible transitions, not a fixed timetable. It explains why a method can change roles as learning develops. The next practical question is how to diagnose these states before choosing the amount of guidance, repetition, variation, and independent problem solving. That diagnosis will be more useful than asking whether one method wins in the abstract.