The Order of Learning Changes the Learning
When someone is learning a new idea, should they try to solve a problem before seeing an explanation, or study a worked example first? The subject here is not a contest between two classroom personalities—“let students discover” versus “tell students clearly.” It is a design choice about the order in which learners attempt, receive guidance, organize an idea, and use it again. This series examines that choice because the order can change what learners remember and, more importantly, what they can do with the idea when the next problem looks different.
The practical judgment is conditional. A learner who has never formed a usable representation of a rule may need help noticing structure. A learner who already understands the rule may need something different: effortful retrieval that makes the rule accessible without the example in front of them. The same activity can therefore be useful at one point and poorly matched at another. To make the distinction clear, we need to keep three outcomes apart: performance during practice, retention of what was studied, and transfer to a new situation.
The Question Is Not Simply Which Method Feels Easier
Imagine two students meeting a new kind of problem. One receives a worked solution, studies its steps, and then solves a similar problem. The other tries first, produces an incomplete or incorrect approach, and only afterward sees an explanation. The second student may look less successful during the first encounter. That observation alone does not tell us which student learned more. The first student may have copied a visible pattern; the second may have exposed a misconception or a useful partial idea that makes the later explanation more meaningful. The decisive test is what happens when the surface details change.
This is why transfer matters. Transfer means applying what was learned to a new problem, example, or situation rather than recognizing the exact item from practice. A learner can answer familiar questions fluently and still fail to identify the underlying relationship in an unfamiliar case. Conversely, a difficult practice period can coexist with better later application if the difficulty makes the relevant structure something the learner must work out and then connect to an explanation.
The broadest evidence supplied for this question is a meta-analysis of 53 studies containing 166 comparisons. It compared problem solving followed by instruction, abbreviated PS-I, with instruction followed by problem solving, abbreviated I-PS. The average result favored PS-I: Hedges’ g was 0.36, with a 95% confidence interval from 0.20 to 0.51. The result is meaningful, but it is an average across different studies, learners, tasks, and durations—not a promise that every learner should always struggle first. The same abstract reports stronger effects when PS-I was implemented with high fidelity to Productive Failure, while different patterns appeared for younger learners in roughly grades two through five and for domain-general skills.
The phrase “solve first” therefore needs more precision. It does not mean withholding all support, accepting random guessing, or assuming that frustration is educational. It names an initial position in a sequence. What learners do during that position, what instruction does afterward, and whether they get a chance to use the organized idea again are part of the design.
The diagram’s point is not that every lesson must pass through four rigid stages. It is that the initial order only makes sense in relation to the learning job. The rest of the series will ask what a first attempt contributes, how examples and retrieval work differently, and when variation helps a learner move beyond a practiced item.
Two Different Comparisons Hide Inside the Question
The PS-I versus I-PS comparison concerns sequence. It asks whether putting problem solving before instruction tends to prepare learners for later learning better than giving instruction first. Productive Failure is a particularly important version of this question because it treats the initial difficulty as something that must be designed to support later explanation and learning. The meta-analysis reported that high-fidelity implementations produced effects ranging from Hedges’ g 0.37 to 0.58, stronger than the overall PS-I estimate in the abstract. That finding makes implementation part of the question, not a footnote.
A second source examines a related but different choice: retrieval practice versus worked examples. Retrieval practice requires learners to bring information back from memory. A worked example presents a problem together with a solution process for study. These activities can both improve learning, but they put different demands on the learner. Retrieval practice is closely associated with strengthening access to an available representation. Worked examples can guide attention through a procedure and reduce the burden of generating every step while a learner is still trying to form a coherent schema.
Those comparisons should not be collapsed. Recalling an answer to a problem is not identical to open-ended problem solving before instruction. Studying a worked solution is not identical to every form of teacher explanation. A study can find that PS-I is helpful on average while another finds that worked examples outperform repeated retrieval when learners have not yet been taught the rule. The results are not automatically contradictory; the interventions and the learning demands differ.
The distinction also prevents a common mistake in interpreting difficulty. Retrieval practice is not the same as fluency practice. Retrieval asks whether a learner can recall and use information; fluency practice emphasizes speed, accuracy, and automaticity, often through timed repetition. The controlled study considered here did not use speeded responses, so its retrieval findings should not be presented as evidence that learners became faster or automatic. Precision about the activity protects the conclusion from becoming a slogan.
Four Learning Jobs Behind One Classroom Decision
A useful way to organize the subject is to ask what the learner needs to accomplish now.
Induce means infer a rule or relationship from cases. The learner does not merely need another answer; they need to notice what stays constant while the examples change. A first attempt can make their current idea visible. Varied cases can make comparison possible. A later explanation can name the structure that was difficult to see alone.
Organize means connect the idea into a usable procedure or schema. Even after an attempt, learners may have several partial strategies, errors, or observations that do not yet form a coherent account. Instruction and worked examples can help select the important features and show how the parts relate. This is where “instruction after” matters: the explanation must do more than reveal the correct answer. It must help reorganize what the learner tried.
Retrieve means bring an understood rule or procedure to mind when the supporting example is gone. Here, repeated effortful recall can be valuable. It tests whether knowledge is accessible rather than merely familiar. Feedback can correct an error and reinforce the intended relation. If the learner has no stable representation to retrieve, however, repeated requests may strengthen an isolated answer more readily than the underlying rule.
Transfer means apply the organized and retrievable knowledge to a new case. Transfer is not produced by variation simply because variation is present. The cases must differ in a way that makes the shared structure worth finding, and the learner must receive enough information to correct a mistaken interpretation. A practice set can become harder without becoming more transferable if its changes are arbitrary or if no feedback clarifies what the cases have in common.
These jobs can overlap. A learner may induce a rule while organizing it, retrieve it while refining it, and discover during transfer that the representation is incomplete. That is why the central diagnostic question is not “Which method is best?” It is “What is missing from the learner’s current knowledge, and what must the next activity make possible?”
The two supplied sources give complementary reasons to ask that question. The meta-analysis suggests that a well-designed PS-I sequence can prepare learners for instruction and later learning. The controlled experiments suggest that the presence of initial instruction changes the relative value of retrieval practice and worked examples. When rules were explicitly presented before practice, retrieval practice produced better immediate generalization than worked examples, and adding item variability did not produce a significant final-test advantage. When initial instruction was removed, repeated worked examples outperformed repeated retrieval, while varied retrieval reached performance comparable to varied worked examples. These findings point toward a changing learning state rather than a permanent winner.
That changing state is the key to the question this series will pursue. A first attempt can be valuable when it gives later instruction something to explain and compare, but only if the task and support make that attempt informative. Worked examples can be valuable when a learner needs guidance to construct a procedure. Retrieval can be valuable when the rule is available but access is fragile. Variation can be valuable when generalization requires comparison across instances, even if it slows practice. Feedback can determine whether an error becomes information or simply becomes another repeated error.
So the next question is narrower and more demanding than “Is failure good?” What turns a first attempt into Productive Failure rather than wasted effort? Answering it requires following the sequence itself: what the learner brings to the problem, what the attempt reveals, and what the instruction that follows must accomplish.