Checkpoint: Name the Learning Job Before Choosing the Method
The central subject is still the order and form of learning: solving before instruction, studying a worked example, retrieving a rule, and applying it to a new case. The useful judgment now is diagnostic. Before choosing a method, identify what the learner can already represent, what the next task demands, and what kind of transfer matters. That single pause connects the sequence question to the mechanisms of memory, induction, guidance, variation, and feedback.
The point is not to memorize a ranking. Problem solving first can prepare a learner for later instruction when the attempt is informative and the design supports reorganization. Worked examples can be better matched to a learner who needs a procedure made visible. Retrieval can strengthen access once the rule exists. Variation can widen the representation when the learner must generalize. These are different answers to different learning jobs.
The Same Activity Can Solve Different Problems
A method label tells us what an activity looks like, but not what it is doing for a particular learner. “Problem solving” could mean an open-ended attempt before any explanation, a structured question after instruction, or repeated selection from answer choices with feedback. “Instruction” could mean a concise definition, a worked solution, or a comparison of several approaches. “Practice” could strengthen a remembered item or require a learner to infer a relation across changed cases. The surface label is not enough to predict the learning outcome.
The sequence matters because it changes the learner’s starting point. In instruction followed by problem solving, an explanation supplies an initial structure that practice can exercise. In problem solving followed by instruction, the learner’s first work supplies material that the explanation can organize or challenge. The meta-analytic evidence favors PS-I on average, especially when it is implemented with high fidelity to Productive Failure, but that average belongs to a particular comparison. It should not be converted into the claim that all independent attempts are better than all explanations.
The process model adds another distinction. Retrieval practice is effortful access to information that is already represented well enough to retrieve. A worked example makes a solution process available for inspection and can reduce the burden of generating every step while a procedure is still being formed. A first attempt can reveal what the learner currently believes before either access or organization is secure. The three activities can appear in one learning episode, but they do not perform the same job.
This gives us a more useful contrast than easy versus hard. Suppose a learner has been shown a rule but cannot use it without looking at the example. The immediate need is not another explanation in the abstract; it is an opportunity to retrieve and apply the rule, receive feedback, and discover whether access holds under changed conditions. Now suppose another learner can imitate the steps in an example but cannot explain why the steps work. Repeated retrieval of the final answer may strengthen the wrong target. The learner needs organization and comparison before retrieval can do its best work.
A third learner may have no stable account of the relationship at all. Asking for a correct answer can then confuse two failures: failure to retrieve a known rule and failure to construct the rule in the first place. A worked example may provide the necessary guidance.