Should Gardner's multiple intelligences be used as a basis for instruction?

Verdict

Insufficient, deliberately — not debunked, and the distinction is the point of this file.

Learning styles was tested with the right design and failed. Multiple intelligences was never operationalised well enough for that to happen. Ferrero, Vadillo and León (2021) set a low bar — any study reporting a quantitative impact of an MI-based intervention on reading, maths or science using a pre-post design with a control group — collected 39 articles covering 3,009 students in 14 countries, and concluded that a valid evaluation is not yet possible. That is a verdict about the evidence base, and under this archive's vocabulary it is insufficient, not debunked.

Two claims have to be kept apart, because they have different evidential status:

The structural claim — that there are eight (or nine) largely independent intelligences. This has been tested and it does fail. debunked would be defensible for this claim alone.
The instructional claim — that teaching through the intelligences raises achievement. This is what a school actually decides, and it is unevaluable on the current literature.

The topic's verdict tracks the instructional claim, because that is the decision.

What the evidence shows

Source	Design	Grade	Key effect
Ferrero 2021	Systematic review + attempted meta of MI interventions; 39 studies, 3,009 students, 14 countries	C	Not validly estimable. Small samples, no active controls, unreported instruments and treatment content, publication/reporting bias
Visser 2006	Psychometric test: 2 tests per Gardner domain, 200 adults	D	Large g factor; independence not supported. Bodily-Kinaesthetic loads notably low
Waterhouse 2023	Critique / 40-year review	D	No researcher has ever directly looked for a brain basis for the individual intelligences
Waterhouse 2006	Critical review of MI, Mozart effect, EI together	D	All three lack adequate empirical support; each has a better-supported counterpart theory

Ferrero et al. is the load-bearing source and it says something specific. Not "MI instruction does not work" but "we cannot tell." Studies did not report which instruments measured the outcome. Studies did not report what was actually done during training. Control groups were passive where they existed. Sample sizes were small. Reporting bias was detectable. A literature in that condition cannot be aggregated into an effect size, and pretending otherwise would reproduce exactly the error this archive exists to correct.

The structural failure is cleaner. Visser, Ashton and Vernon built two tests for each of Gardner's eight domains from Gardner's own descriptions and factor-analysed them. A large general factor emerged, loading substantially on Linguistic, Logical/Mathematical, Spatial, Naturalistic and Interpersonal. Within-domain non-g associations were weak but present — so the domains behave like the group factors of a standard hierarchical model of intelligence, which is a very different thing from eight independent intelligences. The domains that escape g are the ones defined by sensory, motor or personality content, i.e. the ones that are not intelligences in the first place.

What has never been done is as telling as what has. Waterhouse (2023) makes the point that in 40 years no researcher has directly looked for a brain basis for the individual intelligences, that factor studies have not shown independence, and that MI teaching studies have not explored alternative causes of positive results. Gardner's own defence — that MI cannot be a neuromyth because he never advanced it as a neurological theory — sits uneasily with the modular-brain framing on which the theory was built and with which it is taught.

The literature citing Ferrero is its own evidence. Chasing forward from the 2021 review returns overwhelmingly small, non-randomised, single-site applied studies with researcher-administered outcomes — one representative specimen is recorded in db/sources/ as excluded (gebremeskel-2024-mi-reading-tasks-efl). Four years after a systematic review identified exactly these flaws, the field is still producing them.

Hereditarian-lens assessment

Risk: medium, which is why confidence is capped below high independent of source count.

The MI intervention literature is made of uncontrolled or passively controlled classroom comparisons. Classes selected into MI programmes differ from comparison classes in ways that track family background and child ability — the standard passive gene–environment correlation problem — and nothing in this literature is genetically informative. Any positive estimate it produces is therefore uninterpretable on its own terms before its reporting problems are even considered.

There is a deeper tension with the archive's premise. MI's central appeal is the promise that a child weak in the academically loaded domains is strong in another, independent one. The psychometric finding runs the other way: the cognitive domains share a strong general factor, which is also what the behaviour-genetic literature finds and what makes cognitive ability substantially heritable in the first place. The domains where MI's independence claim survives — bodily-kinaesthetic, interpersonal — survive because they are not measuring cognitive ability. That is not a consolation prize for a struggling reader; it is a category change presented as one.

Boundaries & what critics say

Gardner's own position is that MI is a claim about the architecture of mind, not a pedagogical prescription, and that he never endorsed the classroom applications sold under his name. If that is granted, then the instructional claim has no author defending it, and evaluating it against the intervention literature is fair.
insufficient is not refuted. Nothing here shows that MI-inspired teaching harms children. It shows that after four decades nobody has run a study capable of showing it helps. A founder is entitled to know that the shelf is empty, not that the shelf is full of negatives.
Visser 2006 is grade D and off-band. Cross-sectional, 200 adults, no children. It is retained because the instructional claim rests on the structural claim and no comparable child study exists — a real gap, not an oversight.
Do not run this argument into curriculum narrowing. "MI is not an evidence-based instructional framework" and "art, music and PE are worth teaching" are both true. This archive's motor-competence transfer verdict makes the same move: teach the thing for the thing, drop the spillover promise.
The one durable contribution may be rhetorical. MI gave teachers permission to value non-academic capability. That is a real cultural effect. It is not an evidence base.

Practical guidance

Do not profile children by intelligence type, do not timetable by it, and do not buy materials or professional development organised around the eight domains.
Do not use MI as an explanation for a struggling student. "He's a bodily-kinaesthetic learner" is a stopping point dressed as a diagnosis; it displaces the assessment that would actually help (decoding, fluency, prior knowledge, task working-memory load).
Teach the arts and physical skills because they are worth having. Justify them directly. A justification that rests on a transfer promise collapses when the promise is tested — and this one has not even been tested.
If someone presents MI evidence, ask two questions: was the control group active, and what instrument measured the outcome? Ferrero et al. found the field routinely cannot answer either.

Open questions

Whether a well-designed MI intervention trial would show anything is genuinely unknown; the honest answer is that one has not been run, not that one would fail.
No study has tested whether MI-framed instruction affects teacher expectations for individual children — the same unexamined risk that attaches to the essentialist form of the learning-styles belief.
No genetically informative design exists anywhere in the MI intervention literature.