Learning theories compared and explained: what each family can predict

What follows is learning theories explained for someone with a decision in front of them: four families, set against the same questions, with each one's blind spots stated rather than implied. It is the version we needed ourselves before writing the five briefs that lean on it.

Why the vocabulary decides the argument

Almost every unresolvable staffroom argument about teaching is a vocabulary problem wearing an evidence problem's clothes. One person is using learning to mean a durable change in what someone can do unaided; another is using it to mean participation in the practice of a group. Both then reach for studies that measure their own definition, and neither can be moved by the other's evidence, because the two are not disagreeing about the world. They are disagreeing about what would count as the world changing.

This brief exists because every later brief on this site leans on this vocabulary. The adult-education brief argues that prior structure matters more than age; that claim is only meaningful if you know which theory treats prior structure as decisive and why. The curriculum brief states sequencing as a cost; that cost is measured in the currency cognitivist accounts provide. So this page is not a survey. It is the shortest set of distinctions that make the other five briefs legible.

Four families do most of the work in education research: behaviourist, cognitivist, constructivist, and socio-cultural or situated. They are not four answers to one question. They are four questions, and the reason the field looks quarrelsome from outside is that its participants rarely say which one they are asking.

The four families of learning theories, compared on the same three questions

With the learning theories compared on identical questions, the families stop being rivals and start being instruments. The three questions worth asking of any account of learning are: what does it count as learning, what does it accept as evidence that learning happened, and what teaching decision does it license?

Behaviourist accounts

Behaviourism is treated as a historical curiosity in teacher education and as settled infrastructure everywhere else. Every spaced-repetition app, every driving-test question bank, every clinical skills checklist runs on it. Its claim is narrow and durable: behaviour that is followed by reinforcement becomes more frequent, and schedules of reinforcement change how resistant that behaviour is to extinction.

What it explains well is the acquisition of components: recall of facts, fluency in a procedure, the automaticity that frees attention for something harder. What it cannot explain is anything that requires talking about what the learner understood, because it declines the question. That refusal was a methodological virtue in 1930 and is a practical limit now: a learner who has been reinforced into producing the right answer for the wrong reason is invisible to it.

The honest version of the behaviourist contribution is therefore not "drill works" but something more specific: components benefit from schedules, and understanding does not decompose into components. Confusing the two produces a curriculum that can teach a procedure to fluency and still leave a learner unable to say when to use it.

A concrete case: a nursing student who can recite a drug-calculation formula but cannot do it under time pressure has a fluency problem, and spacing the practice will fix it. A student who applies the formula to a case where it does not apply has an understanding problem, and more practice will make the error faster rather than rarer. The two look identical on a wrong answer and need opposite interventions, which is the whole practical value of keeping the families distinct.

Cognitivist accounts

The cognitivist turn asked what happens between stimulus and response, and answered it in terms of memory. The version with the most practical purchase is cognitive load theory, which starts from two facts that are as close to settled as this field gets: working memory holds very little novel information at once, and long-term memory is effectively unbounded. Instruction is then the problem of getting material into the second without overrunning the first.

What makes this family the most useful for design is that it produces predictions specific enough to be wrong. Worked examples should beat unguided problem solving for novices. Splitting attention between a diagram and text that explains it should cost more than integrating them. Redundant explanation of something a learner already knows should hurt rather than help. Each of those has a shape you can see in a lesson, and each has been tested repeatedly.

The finding that matters most for anyone designing for mixed audiences is the expertise reversal effect: the support that helps a novice most is the support that hinders an expert. A worked example is a scaffold for a learner who cannot yet generate the steps and an interruption for one who can. This single result kills the idea of a universally optimal lesson more thoroughly than any argument about styles.

The design consequence is that support has to be planned as something that will be withdrawn, on evidence, rather than as a permanent feature of the material. In practice that means a course needs a way to notice that a learner has moved past the scaffold — which is a measurement problem the theory does not solve for you.

The limit is scope. Cognitive load theory is a theory of processing, and it is close to silent on why anyone bothers. It cannot tell you why a learner who solved the worked examples stops coming, or why the same material lands in one workplace and not the next.

Constructivist accounts

Constructivism claims that a learner builds new understanding out of what they already have, which means that prior conceptions are not empty space to be filled but structure that filters everything arriving. Its strongest empirical support is the misconceptions literature: stable, teachable, intelligent people hold ideas about force, evolution, probability and heat that survive years of instruction because the instruction never engaged the idea already in place.

This is where the field's most expensive confusion lives. Constructivism is an account of how knowledge is built. It is not an instruction to minimise guidance. The step from "learners construct their own understanding" to "therefore let them discover it" is a non-sequitur, and the 2006 argument by Kirschner, Sweller and Clark made the cost explicit: a discovery task loads working memory with the search itself, which is precisely the resource the learner needs for the concept. Their opponents replied, correctly, that guided inquiry is not the same thing as unguided discovery. The exchange is worth reading in full because it is the clearest available case of two camps agreeing on the evidence and disagreeing about what the label licenses.

What constructivist accounts license is therefore narrower and more demanding than the slogan: find out what conception the learner is running, give them a situation their conception predicts wrongly, and then supply the better account. That sequence needs more instructional control, not less.

Socio-cultural and situated accounts

The socio-cultural family, following Vygotsky, treats learning as something that happens between people before it happens inside one. Its most-borrowed and most-abused idea is the zone of proximal development: the band between what a learner can do alone and what they can do with help. The abuse is that it gets used as a synonym for "slightly difficult". Its actual content is a claim about mediation — the help is not encouragement, it is a more capable partner doing part of the thinking out loud until the learner can do it themselves.

Lave and Wenger pushed this further in 1991 by making participation, not acquisition, the unit of analysis. In their account a newcomer learns a practice by being allowed to do progressively less peripheral parts of it, and what they learn includes an identity within the community. This explains something the other three families handle badly: why an intervention that demonstrably taught the skill still fails to change what people do at work. The skill was acquired; the practice did not admit it.

The limit of this family is measurement. Once learning is defined as changed participation, the evidence becomes what a community lets someone do, and that is genuinely harder to establish than a test score. Studies in this tradition are often rich and rarely comparable, which is a real cost and not a slander.

What scaffolding means, explained properly

Scaffolding is the most widely adopted and least faithfully implemented idea in this literature, so it is worth stating what the original claim requires. A scaffold is temporary support that lets a learner complete a task they could not complete alone, and it has three properties that the classroom version usually drops.

It is contingent: the support responds to what this learner just did, which means the teacher has to find out where the difficulty actually is rather than delivering a prepared hint. It fades: support that stays is not a scaffold, it is a permanent prosthesis, and a learner who can only perform with it has not learned the task. And it transfers responsibility: at each step the learner takes over a part of the thinking the helper was doing, which is why a worked example read passively scaffolds nothing.

Written out, the demands are obvious and expensive. A worksheet with sentence starters is a scaffold only if somebody removes the starters later and checks what happens. Most published scaffolding is the first property alone, made general enough to print, which is why the effect in practice is so much smaller than the theory implies.

Claims from this literature that did not survive

A brief that only reports what holds is not much use for reading the field, because the discredited claims are the ones still being sold. Three are worth naming.

Learning styles

The proposal is that instruction matched to a learner's preferred modality produces better outcomes. The observation underneath it is real: people do report stable preferences, and they report them consistently enough to be measured. What has repeatedly failed to appear is the interaction the theory needs: the matched group does not reliably outperform the mismatched one. The distinction to hold on to is that a preference can be genuine and a prescription built on it can still be unsupported.

The learning pyramid

The diagram assigning retention percentages to activities — ten per cent of what we read, ninety per cent of what we teach — has no traceable study behind the numbers, and the figures conveniently arrive in multiples of ten. It survives because it flatters an intuition that is partly correct: explaining something to somebody else is a demanding retrieval task and does help. That intuition needs no invented percentages, and citing them costs credibility with anyone who checks.

Brain-based prescriptions

Left-brain/right-brain teaching, critical windows after which a subject can no longer be learned, and the claim that we use a fixed fraction of the brain all pattern the same way: a real neuroscientific finding, generalised past what it can support, and then converted into a classroom instruction it never licensed. The test that catches all three is to ask what specific teaching decision the finding changes, and to notice when the answer is a slogan.

Transfer: the problem none of them solved

Transfer — using what was learned in one setting somewhere else — is the point at which all four families become modest. The empirical pattern has been stable for decades and is uncomfortable: near transfer, to problems that resemble the taught ones, is reliable. Far transfer, to genuinely different situations, is rare, small, and hard to reproduce. The general thinking skills that curricula promise transfer least well of all.

Each family has an explanation. The behaviourist one is that the new stimulus does not resemble the trained one. The cognitivist one is that transfer requires abstracting a schema, and abstraction requires more examples than instruction usually supplies. The constructivist one is that the learner's conception was tied to the surface features of the teaching case. The situated one is that the second setting is a different practice, so the question was malformed. All four are partly right, which is why the practical advice converges: if you want transfer, teach with varied cases, make the abstraction explicit rather than hoping it emerges, and expect less than the brochure promises.

This convergence matters more than the disagreements. Where the four families give the same advice, that advice is about as well supported as anything in education research.

Choosing the vocabulary for a decision

The test of a framework here is not truth but fit to the decision in front of you. Four common decisions, and the vocabulary that actually helps:

That table is the practical payoff of having the learning theories compared rather than ranked: it is a routing instruction, not a hierarchy. Someone who only owns one of these vocabularies will diagnose every problem as the kind their vocabulary can see, which is the most common failure in professional development and the reason the teacher-development brief treats theoretical range as a practical skill.

Where the field still disagrees

Three disagreements are live, and it is worth knowing which is which.

How much guidance, and when

Nobody serious now defends unguided discovery for novices, and nobody serious claims that expert learners need worked examples. The live question is the shape of the fade between those poles, and it is live because it depends on the learner's current structure, the thing that is hardest to measure in a room of thirty.

Whether general skills exist as skills

Critical thinking, problem solving and metacognition are either domain-general capacities that can be taught directly, or names for what expertise looks like from outside. The evidence leans towards the second, which is unpopular because whole programmes are sold on the first. The honest statement is that the transferable part is thinner than the promise and thicker than zero.

What counts as evidence

The deepest disagreement is not about learning at all but about method: whether an effect that only appears under controlled conditions tells you what to do in a classroom, and whether a study that describes one classroom richly tells you anything about the next. This one will not resolve, and pretending it has is how a field ends up with two literatures that never cite each other.

Which is why learning theories explained as rivals mislead: the families are instruments with different reaches. What follows for practice is deliberately unglamorous. Use the vocabulary that fits the decision, expect transfer to be the weak link, and treat any claim that one framework explains everything as a claim about the speaker rather than about learning.

Works referred to

1. Mind in Society: The Development of Higher Psychological Processes — L. S. Vygotsky, 1978.
2. Situated Learning: Legitimate Peripheral Participation — Jean Lave and Etienne Wenger, 1991.
3. Why Minimal Guidance During Instruction Does Not Work — Paul Kirschner, John Sweller and Richard Clark, Educational Psychologist, 2006.
4. Cognitive Load Theory — John Sweller, Paul Ayres and Slava Kalyuga, 2011.
5. How People Learn: Brain, Mind, Experience, and School — John Bransford and others, National Research Council, 2000.
6. Model-Based Approaches to Learning: Using Systems Models and Simulations to Improve Understanding and Problem Solving in Complex Domains — Sense Publishers, 2010.
7. Threshold Concepts within the Disciplines — Sense Publishers, 2008.

Questions readers send

Is there one learning theory that is simply correct?

No, and the question mistakes what these frameworks are. Each family answers a different question well: behaviourist accounts describe how contingencies shape observable behaviour, cognitivist accounts describe the processing limits that make instruction hard, constructivist accounts describe how prior structure filters new material, and situated accounts describe why the same lesson transfers in one setting and dies in another. A course design that needs all four answers needs all four vocabularies.

Does constructivism mean students should discover things for themselves?

That inference is the single most consequential category error in the field. Constructivism is a claim about how knowledge is built in a learner's head, not an instruction to minimise guidance. Kirschner, Sweller and Clark argued in 2006 that treating it as a method contradicts what is known about working memory, and the exchange that followed is still the clearest statement of the difference between an epistemology and a lesson plan.

What is the practical use of cognitive load theory?

It is the most operational of the four: it makes specific, testable predictions about worked examples, split attention, redundancy and the expertise reversal effect, and those predictions tell you what to change in a slide, a worksheet or a demonstration. Its limit is scope: it says little about motivation, identity or why a learner cares.

Why do learning-styles claims keep appearing if the evidence is against them?

Because the underlying observation is real and the inference from it is not. Learners genuinely differ, and they report stable preferences. What repeatedly fails to appear is the interaction effect the theory needs: matching instruction to a reported style does not reliably improve outcomes. The preference is real; the prescription is not supported.

How much of this matters for adult learners specifically?

The processing limits are the same. What differs is the amount and correctness of prior structure an adult brings, which is exactly the variable most of these theories treat as decisive. That is why the adult-education brief is built on this one rather than beside it.

Where to go next

1. Adult education What adult education research supports and what it only asserts: the six andragogical assumptions checked one at a time, plus the participation evidence.
2. Curriculum design Backward design, spiral sequencing and constructive alignment explained as a coverage-versus-depth trade, with the cost of each choice stated in teaching hours.
3. Teacher development Which forms of teacher professional development change classroom practice, by how much, and how long the effect lasts before it decays.