Instructional Strategies

There is no best strategy, and that is the finding

Ask which teaching method works best and the honest reply is a question back: for whom, teaching what, and at what point in learning it? That is not evasion; it is one of the most replicated results in the field.

The sharpest demonstration is the expertise reversal effect. Kalyuga, Ayres, Chandler and Sweller assembled experiments in which the relative effectiveness of two instructional formats flipped as learners gained knowledge: support that novices needed became redundant for more knowledgeable learners, and processing it against knowledge they already held cost them. The authors' conclusion is blunt - design has to be tailored to the expertise of its intended learners, and without that tailoring the effectiveness of any design is close to arbitrary. A strategy is not a property of a lesson. It is a relation between a lesson and a learner.

Three dimensions do most of the choosing. Prior knowledge: the less a student knows, the more the instruction has to supply and the less it can ask the student to generate. Type of knowledge: a procedure with one correct execution is not learned the way an interpretive judgment is. Phase of learning: early learning wants dense guidance and a high rate of success, later learning wants support withdrawn, conditions varied and retrieval demanded. Run a strategy argument through those three questions and most of it dissolves.

Explicit instruction is not lecturing

This confusion is old, and it wrecks the argument before it starts. Hughes, Morris, Therrien and Benson wrote specifically to tighten the term, and their first point is that it is not a single intervention at all: it is a combination of more than a dozen teaching behaviors used to design and deliver instruction, which is why published definitions of it disagree. They also separate two things sharing a name - capital-letter Direct Instruction, meaning particular published programs, and direct or explicit instruction as a general approach.

Unpack the bundle and the difference from lecturing is not subtle. Explicit instruction breaks a skill into pieces small enough to learn, demonstrates each one, makes the reasoning audible rather than only the steps, requires responses from every student often enough to know who is lost, delivers feedback fast enough to catch an error before it is practiced, and arranges the work so most attempts succeed. A lecture supplies the demonstration and stops. It has no way of detecting whether the room followed, no way to correct anyone in time for the correction to matter, and no student production at all.

Rosenshine's synthesis supplies the number that makes the difference concrete: effective guided practice runs at a success rate of roughly eighty percent - high enough that students are rehearsing the correct thing, low enough that something is still being learned. A lecture has no success rate, because nobody is producing anything that could be right or wrong.

Worked examples, and knowing when to take them away

A worked example is a fully solved problem handed to a student to study rather than to solve. It helps novices because solving an unfamiliar problem and learning the method it teaches are two jobs competing for the same narrow attention.

The federal practice guides turn this into design advice. Organizing Instruction and Study to Improve Student Learning puts the examples inside the practice set itself, recommending at a moderate level of evidence that solved examples be alternated with problems to solve. The What Works Clearinghouse guide on secondary writing rates a model-practice-reflect cycle at strong evidence for grades six through twelve - the same sequence applied to a skill nobody would call procedural. Modeling is not just for arithmetic.

The decision people get wrong is the exit. Guidance that is essential at the start becomes redundant later, and redundancy is not neutral: the learner still has to process it. So plan the fade. Start with the full example, move to partially completed examples in which the student supplies the last step, then the second-to-last, then the whole thing. Two signals say it is time: the success rate climbing past the point where the remaining errors still teach something, and students saying the next step before you do. Leaving the scaffold up past that point is not generosity. It is the expertise reversal effect happening to your class.

Questioning: check the room, not the volunteer

Questioning is the cheapest instrument a teacher has and the one most often used wrongly. Rosenshine's contrast is the diagnostic: asking whether there are any questions is not a check for understanding, because the students who most need help are the least likely to answer. A real check makes every student produce something readable.

Two pieces of the technique have research behind them. First, wait time - the length of the pauses separating utterances in a classroom exchange. Tobin's review found that once average wait time passed a threshold of about three seconds, teacher and student discourse changed and higher cognitive level achievement was obtained in elementary, middle and high school science, with increases also reported in middle school mathematics. Stahl reframes the silence as think time: it is not the teacher waiting, it is the student processing - and the pauses worth managing include the one inside a student's answer and the one right after it.

Second, cold calling - choosing who answers rather than taking the raised hand. Dallimore, Hertenstein and Platt studied sixteen sections of one undergraduate accounting course, five with heavy cold calling and eleven with little. In the heavy sections significantly more men and women volunteered answers; women, who volunteered significantly fewer answers than men where cold calling was rare, answered about as many where it was common; and neither group reported discomfort. One course at one university, so hold it loosely - but it is direct evidence against the usual objection.

Keep the categories straight, though. 'No opt out', returning to a student who could not answer until they say the correct answer themselves, is Technique #1 in Doug Lemov's Teach Like a Champion - a descriptive book of routines drawn from watching classrooms. A fine place to get an idea; not the same kind of claim as a controlled comparison.

Discussion and Socratic method: real gains, narrower than advertised

Discussion is defended on principle more often than it is examined on results, so the meta-analytic picture is worth having exactly. Murphy and colleagues found that several discussion approaches produced strong increases in the amount of student talk with matching reductions in teacher talk, and substantial improvements in comprehension of text. They also found that few approaches were effective at increasing literal or inferential comprehension or critical thinking and reasoning, with effects moderated by study design, outcome measure and student academic ability. Most studies involved grades four through six.

That says two things at once. Structured discussion reliably shifts who is talking and helps students understand a text; it does not automatically produce the higher-order reasoning it is usually justified by, and the approach matters more than the fact of discussing.

What the evidence supports specifically is the demand for explanation: the IES practice guide rates asking deep explanatory questions - why, how, what if, what if not - at strong evidence. A Socratic sequence is at its most defensible when it does that: take a student's claim, require the reasoning underneath, then test it against a case they did not consider. It is at its weakest as a guessing game in which the class hunts for the answer already in the teacher's head.

Cooperative learning versus students sitting in groups

Putting students in fours does not make the work cooperative. Johnson and Johnson, whose social interdependence theory sits under most of this literature and whose review counts more than 1,200 studies over eleven decades, specify five elements that have to be engineered into the task: positive interdependence, individual and group accountability, promotive interaction, deliberately taught social skills, and group processing in which the group reviews how it is working.

Two of those do the heavy lifting when you are designing a task. Positive interdependence means the task is built so no member can succeed unless the others do - one set of materials split between them, roles each holding a necessary piece, a product nobody can assemble alone. Individual accountability means each student's own learning is separately visible, so the group's product cannot hide a passenger. Unstructured group work usually has neither, which is why it reliably produces one student doing the assignment and three watching.

There is also a threshold question that gets ignored. Collaborative cognitive load theory predicts that group work pays off on tasks complex enough to overload a single learner, because the elements can be spread across several working memories; on simple tasks the cost of communicating and coordinating outweighs the benefit. If a competent student could do the task alone in five minutes, making it a group task adds overhead and subtracts practice.

Inquiry and problem-based learning: it is a guidance question

The public argument treats inquiry and explicit instruction as opposites. The meta-analytic evidence treats guidance as a dial, and that is the more useful frame.

Alfieri and colleagues ran two meta-analyses over 164 studies. Across 580 comparisons, outcomes favored explicit instruction over unassisted discovery under most conditions, d = -0.38. Across 360 comparisons, outcomes favored enhanced or assisted discovery over other forms of instruction, d = 0.30. Their reading of the gap is the practical one: unassisted discovery does not benefit learners, whereas feedback, worked examples, scaffolding and elicited explanations do. Discovery activities work when they carry the supports that unassisted discovery lacks.

Which is why the label on a strategy tells you almost nothing. Defenders of problem-based and inquiry learning argue that those approaches were wrongly conflated with unguided discovery and are in practice extensively scaffolded; that exchange belongs to the constructivism lesson. What matters for selection is that 'inquiry' names a range of guidance levels, and the guidance level is what the evidence attaches to. So the useful question about a proposed inquiry unit is not whether inquiry works. It is where a student gets told the thing they cannot reasonably discover, and where they get feedback before a wrong method sets.

Practice is a design decision, not homework

How practice is scheduled is as much an instructional strategy as how content is presented, and it is the part most often left to whatever order the textbook used. Spacing, interleaving and retrieval are covered elsewhere in this subject as things a student does while studying. The point here is that they are also things a teacher decides, weeks in advance, on a calendar.

Three of those decisions carry ratings in the IES practice guide. Spacing the same material across separated sessions rather than massing it is rated moderate. Using quizzes to re-expose students to key content is rated strong - which reclassifies the low-stakes quiz as an instructional event rather than a measurement one, and puts it in the lesson sequence rather than the gradebook. Interleaving problem types instead of blocking them is the conditional case: the meta-analytic average is a moderate benefit, but it is strongly moderated by material, clearly positive for visual categories and mathematics and negative for word learning. Interleaving helps when the difficulty is telling confusable things apart. It hurts when it fragments something that needed a run at it.

That is the whole lesson in miniature. Even inside one well-supported technique, the effect depends on what is being taught.

Reading strategy evidence without being sold a ranking

Teacher preparation is full of ranked lists of teaching influences with effect sizes beside them, most traceable to John Hattie's Visible Learning. Treat those numbers with real caution. Bergeron's published critique, written from a statistician's standpoint, identifies problems past quibbling: the book's common language effect figures can fall outside the range a probability can occupy, an error first flagged by Topphol in 2012; Cohen's d is treated as a unit-free universal measure permitting comparison across arbitrarily different studies, which it is not; disparate things are aggregated under single labels; and the baseline comparison, often arbitrary, decides whether an effect comes out positive or negative. Bergeron's verdict is that the methodology should be called pseudoscience. You need not adopt that verdict to accept the narrower one: a league table of teaching methods is not a decision procedure, and a number lifted from it is not a magnitude you can plan with.

The better model is already available. Federal practice guides rate each recommendation separately, name the population it was tested on and print the studies behind it - which is why one guide rates quizzing strong and pre-questions low, and why systematic instruction is rated strong specifically for elementary mathematics intervention, not for teaching in general. Ratings attach to specified practices for specified learners. That is a slower kind of claim than a ranking, and the kind you can act on.