The key to unlocking a brighter future for students lies within the understanding and application of learning science. As a data scientist and edtech developer, I believe our job is not about inventing the next shiny digital device; it’s about improving education outcomes for students, and doing so demonstrably and empirically with research. And the starting point for that is looking at what we already know from the science of learning.
Of the litany of learning science research, there are three important pieces that guide my work, and I hope they will help support yours as well.
Bloom’s Two Sigma Problem
Benjamin Bloom, one of America’s leading educational psychologists, is most famous for “Bloom’s Taxonomy,” but in 1984 he wrote a seminal paper in learning science called the “Two Sigma Problem.” In the paper Bloom reports research conducted by his team that compared student learning under three instructional conditions:
Conventional. Students learn the material in a standard lecture-type setting. Periodically they are also tested to assess whether they have learned the material.
Mastery Learning. The instructional material in mastery learning is the same as in the conventional model. However, in mastery learning students are not allowed to move to the next stage of learning until they have demonstrated “mastery” of the prior unit of instruction. Also, testing is replaced by assessment in the form of continuous feedback and error correction. In the language of educators in mastery learning the emphasis shifts from summative assessments (testing) to formative assessments (feedback)..
Tutoring. Each student is assigned a personal learning coach. Students receive all the benefits of mastery learning but additionally have access to an expert human tutor.
Which mode worked the best? Using the conventional mode of instruction as a baseline, students under mastery learning saw a one-sigma (standard deviation) improvement in performance. Students who received one-on-one tutoring saw a two-sigma improvement.
A one sigma is roughly a one-letter grade in improvement. It can be the difference between a student failing a course and passing a course—and most educational interventions don’t come close. If one sigma of improvement is huge, two is monumental.
There is another subtle and often-missed aspect of Bloom’s findings. Personalized instruction, in the form of mastery learning and human tutoring, not only increases the mean of performance, but also decreases the standard deviation—meaning that students at the low end of the distribution over time can begin to catch up with those at the high end.
We can derive several important conclusions from Bloom’s research. First, given the right setting most students are capable of high levels of learning. Second, personalized instruction can lead to large learning gains. Third, personalized instruction can also help to close the achievement gap between high and low learners. Fourth, we have yet to devise scalable learning systems and solutions that can lead to one sigma and beyond of learning gains.
As an expert on experts, Swedish psychologist K. Anders Ericsson’s framework of “Deliberate Practice” provides a grand unifying principle for understanding the core elements of learning and skill development. One of the bedrocks of Ericsson's research is that you have to put in the time. There are no shortcuts, no matter your talent level—or even if you’re Michael Jordan. To attain the status of an elite performer on average it takes 10 years or roughly 10,000 hours of sustained practice.
We often think of athletes as “naturally talented.” At the elite level you have to be talented. There is no mistake about that. But Jordan didn’t emerge out of his mother's womb dunking basketballs and flying and twisting in mid-air towards the basket for a layup. Jordan honed his art by relentless practice. He surpassed other athletes by his work ethic, practicing harder and longer than most of his competitors.
Researchers disagree on exactly how much practice is needed for skill development. Drawing on Ericsson’s work, Angela Duckworth, another leading psychologist and MacArthur Fellow, has distilled the four primary characteristics of deliberate practice.
Intentional Practice. To achieve optimal skill development we don’t practice randomly. Practice has to have a very specific intention. It needs to match and be appropriate to the current level of skill development and the next targeted level.
Challenge Exceeds Skill. Learning occurs only when what we need to be able to do exceeds our skill level. Learners resist being in this zone. All skill development requires staying in a zone for prolonged periods, and we are naturally averse to that.
Immediate Feedback. In most learning situations the lag between performance and feedback is too long. This is a challenge to teachers given the constraints on their time and resources. Deliberate practice requires that the lag time between performance and feedback approach zero.
Repetition to Automaticity. You can’t become truly fluent in any new skill until you have repeated it to a level of automaticity and that you can practically do it without conscious effort.
A considerable body of research shows that we are notoriously fickle decision makers. Even when presented with the right information or “insight" we still end up making poor choices.
The poor choices are not just irrational, they are predictably so. Despite the wealth of information on proper nutrition and exercise, why are many of us still overweight? Why do insights about nutrition and exercise remain ineffective for most of us? Is the distinguishing fact will power?
We can learn something from behavioral economics. Research in this field indicates that while we act predictably irrationally in a variety of contexts, small and apparently insignificant details can lead us to change our behaviors.
Consider this example, described in “Nudge: Improving Decisions About Health, Wealth, and Happiness” by Richard H. Thaler and Cass R. Sunstein. Concerned that men don't pay much attention to where they aim in urinals, authorities at Schiphol Airport in Amsterdam sketched the image of a black housefly into each urinal. The predicted that if a man saw the fly, he’d aim at it. The fly-in-urinal trials at Schiphol Airport found that the fly etchings reduced spillage by 80 percent.
One of the aims of data science research then is to discover empirically the right insights along with the small nudges that can make the insights actionable. Think of a nudge as that small but important motivational push that can propel an insight to action. Data science research can help us both discover possible nudges and confirm that they are effective.
Personalized instruction, through a combination of machine intelligence and human intelligence informed by research, holds the greatest promise for significantly moving the needle on learning outcomes. But we also need to understand and design for the messy world of human motivation, perseverance, and irrationality.
Alfred Essa (@malpaso) is Vice President of R&D and Analytics at McGraw-Hill Education