What “learning styles” are and whether they exist is a longstanding debate in the education world. Some educational researchers argue there are no distinct learning styles. Recently, a chorus has been growing who insist this is a proven fact.

To me, it’s pretty obvious that different learning styles exist — that is, that different ways of learning certain concepts are more or less productive for certain students. But until now, we’ve simply lacked the tools to measure them properly.


Thirty years ago, developmental psychologist Howard Gardner laid out his theory of multiple intelligences. The theory posited that rather than measure intelligence as one general ability, we should measure types of intelligence: linguistic, logical-mathematical, musical, bodily-kinesthetic, spatial, interpersonal, and intrapersonal.

People immediately gravitated toward Gardner’s theory. Teachers, especially, saw in it something that could explain what they perceived as different learning preferences in the classroom. (Every teacher occasionally sees a hands-on demo, or a great educational video, engage students in a way a lecture wouldn’t have.)

But as time went on, some educators began casually to conflate “multiple intelligences” with “learning styles” — to sub out “spatial intelligence” for a “spatial learning style” or “bodily-kinesthetic intelligence” for a “bodily-kinesthetic learning style.” As in: “I learn best by dancing!” — which I think we can all agree is silly.


But in some circles the debate has moved beyond multiple intelligences to strongly imply that learning styles in any form don’t exist. “If learning styles were obviously right it would be easy to observe evidence for them in experiments. Yet there is no supporting evidence,” says UVA professor Daniel Willingham, one member of the opposition.

Willingham and others acknowledge that lack of evidence doesn’t constitute definitive proof of nonexistence. But their rhetoric’s strong overall suggestion is that learning styles don’t exist.

In an article on the same topic, Willingham and co-author Cedar Riener state that their arguments against learning styles are based primarily on their “most popular current conception” — their conflation with multiple intelligences. Willingham says elsewhere that no other learning styles theory has shown merit either.

This article is just one example of a wider trend. In taking an extremely narrow understanding of learning styles and strongly denying it, an optical illusion emerges. Learning-styles-deniers seem to be making a strong claim — that learning styles aren’t real. But they are actually making a weak claim — that very broad theories of learning styles (I learn best through song!) haven’t been measured to have any effect by the limited means available until now. Yet the general rhetoric (Willingham’s article is titled “The Myth of Learning Styles”) forcefully implies that learning styles don’t exist. It is these strong claims that are echoed throughout the blogosphere, parroted by education groupies eager to demonstrate that they know the “latest science.”

This isn’t to say this is the intention, or even choice, of any of the serious educational researchers involved. There’s an understandable appetite for decisiveness in this debate. And inadvertent selection bias inherent in scientific journal publishing tends to promote claims that at least appear to be decisive. Strong claims like “learning styles don’t exist” are publishable; weak claims like “we don’t yet have evidence either way” are not.


Students learn hundreds of thousands of discrete concepts in their academic careers. There are over one billion active students in the world at any one time. As I see it, for learning styles not to exist, it would have to be the case that for every single concept that a person ever learns, varying the exact way in which she learns it has zero statistically significant effect on how quickly she learns it, how deeply she learns it, and/or how long she retains it.

Education researchers are free to point out that learning styles haven’t been proven to work. But it is these kinds of standards they should be compelled to gainsay, not some “multiple intelligences” straw man.


At Knewton, we talk about learning strategies instead of learning styles, but the idea is the same. We think of strategies in the game theory sense, as something a person does whether she intends or not.

A learning strategy is a student’s learning path, whether deliberate or not, through particular content, and the resulting learning outcomes. And Knewton can measure the effects of these strategies, to the percentile, for every student, at the concept level.

There are myriad learning strategies that Knewton measures and adjusts for. A sample of the ones we’re focused on now: the amount of content covered per session, the format of the learning experience (text vs. video vs. game vs. physical simulation vs. group discussion, etc.), the difficulty level of prose explaining a given concept, the difficulty of accompanying practice questions, the time of day, whether content contains mnemonic devices, whether it confuses cause and effect, whether it makes use of lists, student attention span, student engagement with particular learning content, strategic modalities (e.g., does the content define a procedure vs. address a common misconception vs. use a concrete example?), the presence or absence of learning aids (e.g., hints), user-specific features (e.g., difficulty relative to the student’s current proficiency), and many, many more.

While many learning strategies are best personalized via materials, rather than in class, the resulting data sets can give teachers insights they’ve never had before. How can materials help a particular student make that “A-ha!” connection during study time away from class? At what point in the learning process would it be best to introduce offline, hands-on exploration, and how can teachers quickly evaluate its effectiveness for each student? Which students would benefit most from group work, and with whom should they be matched up to maximize learning (e.g., because each student happens to be strong at different concepts)? Which students are able to guide themselves through a specific online lesson, and which are more likely to require extra help? And so on.

Knewton has built a $100 million platform to measure learning outcomes at the atomic concept level, down to the percentile of proficiency for each student. We’ve got millions of students on our platform every day, producing unimaginable amounts of anonymized data. And we’ve got some of the world’s best data scientists and statisticians examining those data. And we’re certain that learning strategies do exist. They uniquely characterize each individual. And they are about to become very important.


Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>