Opinion | Learning Strategies

If This Is the End of Average, What Comes Next?

By Daniel T. Willingham     Jun 11, 2018

If This Is the End of Average, What Comes Next?

Personalized learning—the notion that children’s educational experiences should be tailored to their interests and abilities—has a long history, but the digital age has brought new promise to the idea. Computers are cheaper and more powerful than ever, and learning algorithms are more sophisticated, promising better tuning of lessons to students’ passion and performance. Both the Gates Foundation and Chan Zuckerberg Initiative have committed significant resources to developing personalized learning.

Todd Rose, the director of the Mind, Brain, and Education program at the Harvard Graduate School of Education, has emerged as a central intellectual figure behind the movement. In particular, his 2016 book, “The End of Average,” is seen as an important justification for and guide to the personalization of learning.

In Forbes, Tom Vander Ark said the book “poured gasoline on the personalized learning wildfire sweeping American education.” Education Week called it “a kind of bible for the personalized learning movement,” and EdSurge said the book “vaulted [Rose] in the public spotlight.”

The momentum of the personalized learning movement—and the centrality of “The End of Average” to it—prompted me to read the book, and I was left with serious concerns about how Rose puts together conclusions about psychological and neuroscientific studies to support his thesis.

Let’s start by clarifying what Rose argues against. He holds that our culture is obsessed with measuring and finding averages—averages of human ability and averages of the human body. Sometimes the average is held to be the ideal. Rose points to a World-War-II era statue of a woman based on the average measurements of thousands of women, and held up as an ideal of beauty.

At other times, the average is not the ideal, but the comparator. We don’t want to be of average intelligence, but are very concerned with how we compare to that bar. Are we above or below average? By how much?

Rose aims to upend our fascination with the average, and offers three principles as evidence that it is flawed.

The Jaggedness Principle

The jaggedness principle means that many of the attributes we care about are multi-faceted, not of a whole. For example, human ability is not one thing, so it doesn’t make sense to talk about someone as “smart” or “dumb.” That’s unidimensional. Someone might be very good with numbers, very bad with words, about average in using space, and gifted in using of visual imagery. Indeed, any teacher will have noticed that different students have different strengths and weaknesses of this sort.

Since the 1930s, psychologists have debated whether intelligence is best characterized as one thing or many. If “smart” were one thing, your ability to work with space and with numbers ought to be correlated because each is a component of smart. Much research has been done in the last 80 years testing that prediction. Most studies show that the measures of these abilities are correlated, but the correlation is not very big. Historically, some psychologists pointed to these results and said “see, ability is all of a piece—scores on all ability tests are related.” Others said “Look how small those correlations are. There are multiple abilities.” Rose falls in the latter camp, describing the correlations as “not particularly strong.” (p. 90)

But most psychologists stopped playing this game in the 1990s. The resolution came through the work of John Carroll, who developed a third model in which abilities form a hierarchy. We can think of abilities as separate, but nested in higher-order abilities. Hence, there is a general, all-purpose intelligence, and it influences other abilities, so they are correlated. But the abilities nested within general intelligence are independent, so the correlations are modest. Thus, Rose’s jaggedness principle is certainly not new to psychology, and it’s incomplete.

The Context Principle

The second of Rose’s principles holds that personality traits don’t exist, and there’s a similar problem with this claim: Rose describes a concept with limited predictive power as having none at all. The most commonly accepted theory holds that personality can be described by variation on five dimensions: Extraversion, Openness, Conscientiousness, Neuroticism, and Agreeableness. If I know where you fall on each of these dimensions, how much can I predict about your behavior?

Some, but not that much. For one thing, context matters. If you’re at a party where everyone knows one another but you know no one, that situation will have a sizable impact on your propensity to act extroverted, whatever your personality. For another thing (and as Rose emphasizes) these dimensions can vary within an individual, according to context—I may act extroverted in business situations, but introverted in social situations. So sure, if you add more information (context) to personality traits, you will be better at predicting what I do. But that doesn’t mean that traits alone are meaningless.

Chapter 1 offers another example of Rose dismissing a concept with limited predictive power as having none at all. He cites a brain imaging experiment in which the location of brain activity associated with a particular task showed a lot of variation across individuals—so much variation that the average was deceptive; it’s like the man who put his feet in a freezer and head in an oven and was, on average, quite comfortable. The average brain described a state that no brain was in during the task.

Rose is, of course, right that this can be a hazard of averaging. But he goes much, much too far when he claims that “The guiding assumption of decades of neuroscience is unfounded.” (p. 22) Many cognitive functions are localized in the same place—roughly—in different brains. Damage to the occipital cortex results in predictable consequences to vision, for example. The brain damage caused by Parkinson’s Disease leads to predictable motor and cognitive deficits. Yes, there’s variation, especially as you try to make the predictions (either in the brain or in behavior) more fine-grained, but that doesn’t mean that averaging is meaningless.

The Pathways Principle

Rose’s third principle suggests that there are multiple ways to reach a goal like walking or reading, and that there is not a fixed set of stages through which each of us passes. I think he’s absolutely right about this, and I have written about the lack of support for stage theories of development.

What does Rose see as the implications of these three principles? If this is the end of average, what comes next?

There’s a chapter on recommendations for business, but I’ll focus on the higher education chapter. The recommendations are surprising in that they don’t connect to the principals from Part II. That is, you might expect Rose to write “the current system uses personality information, but I’ve just shown you that personality traits don’t exist, so that means we should do things as follows.” Direct ties to the three principles are absent.

Also surprising is that Rose begins the chapter by declaring a single purpose for higher education (in his view). That purpose is preparation for a career, and the declaration will bring some readers to a full stop, because it’s not the purpose many students envision: some people go to college because they are intellectually curious, for example, or to fulfill their vision of being an educated person, or to learn about themselves. What’s surprising is that a book devoted to individuality limits higher education to a single goal.

Rose’s recommendations follow from his vision of education’s purpose. He thinks students should earn credentials, not diplomas. In other words, a school would not certify that you’re “educated in computer science” but that you have specific knowledge and skills—that you can program games on handheld devices, for example. He think grades should be replaced by testaments of competency; the school affirms that you’ve mastered the skills and knowledge, period. Finally, Rose argues that students should have more flexibility in choosing their educational pathways.


These are interesting, bold proposals. There’s plenty of room for questions—for example, whether it’s really plausible that schools could coordinate to define a single criterion for a credential as Rose wants. But after all, Rose’s purpose here is not a detailed blueprint, but a vision that will spark conversation.

My concern lies not with the details, but with the fact that the recommendations are only loosely related to personalization. Other than the call for greater choice, the other ideas seek to change what college is for, not how it works, with the goal of making it more personal.

What’s perhaps most striking to me about “The End of Average” is any perception that educators need to be convinced regarding the book’s main thesis. Tuning educational experiences to individual children has been a goal for education researchers since the 1950s, at the least—“one size fits all” remains a common pejorative applied to educational practices, not an accolade. Educators don’t fail to personalize because it never occurred to them to do so. The obstacles are (1) cost in personnel, materials et al., (2) reluctance to greatly increase learning from screens, and most formidably; (3) lack of knowledge. To personalize effectively, you need to know which differences among children are meaningful and how to cater to them.

“The End of Average” doesn’t offer new evidence as to what those differences might be, or how to honor them. The difficult problem is not persuading people that abilities are jagged, but figuring out how to reliably identify those abilities, and how to help students capitalize on them to meet educational goals. Likewise, the challenge is not persuading people that personality interacts with context, but identifying which contexts matter for which children. And if there is more than one way to reach an educational goal (proficiency in algebra, for example) we need to know how to identify those different pathways, and how select a pathway for a given child. Rose answers these questions by his personal story—he figured out what works for him. But most children don’t, and adults can’t expect them to.

If This Is the End of Average, What Comes Next?

Opinion | Learning Strategies

If This Is the End of Average, What Comes Next?

By Daniel T. Willingham     Jun 11, 2018

If This Is the End of Average, What Comes Next?

Personalized learning—the notion that children’s educational experiences should be tailored to their interests and abilities—has a long history, but the digital age has brought new promise to the idea. Computers are cheaper and more powerful than ever, and learning algorithms are more sophisticated, promising better tuning of lessons to students’ passion and performance. Both the Gates Foundation and Chan Zuckerberg Initiative have committed significant resources to developing personalized learning.

Todd Rose, the director of the Mind, Brain, and Education program at the Harvard Graduate School of Education, has emerged as a central intellectual figure behind the movement. In particular, his 2016 book, “The End of Average,” is seen as an important justification for and guide to the personalization of learning.

In Forbes, Tom Vander Ark said the book “poured gasoline on the personalized learning wildfire sweeping American education.” Education Week called it “a kind of bible for the personalized learning movement,” and EdSurge said the book “vaulted [Rose] in the public spotlight.”

The momentum of the personalized learning movement—and the centrality of “The End of Average” to it—prompted me to read the book, and I was left with serious concerns about how Rose puts together conclusions about psychological and neuroscientific studies to support his thesis.

Let’s start by clarifying what Rose argues against. He holds that our culture is obsessed with measuring and finding averages—averages of human ability and averages of the human body. Sometimes the average is held to be the ideal. Rose points to a World-War-II era statue of a woman based on the average measurements of thousands of women, and held up as an ideal of beauty.

At other times, the average is not the ideal, but the comparator. We don’t want to be of average intelligence, but are very concerned with how we compare to that bar. Are we above or below average? By how much?

Rose aims to upend our fascination with the average, and offers three principles as evidence that it is flawed.

The Jaggedness Principle

The jaggedness principle means that many of the attributes we care about are multi-faceted, not of a whole. For example, human ability is not one thing, so it doesn’t make sense to talk about someone as “smart” or “dumb.” That’s unidimensional. Someone might be very good with numbers, very bad with words, about average in using space, and gifted in using of visual imagery. Indeed, any teacher will have noticed that different students have different strengths and weaknesses of this sort.

Since the 1930s, psychologists have debated whether intelligence is best characterized as one thing or many. If “smart” were one thing, your ability to work with space and with numbers ought to be correlated because each is a component of smart. Much research has been done in the last 80 years testing that prediction. Most studies show that the measures of these abilities are correlated, but the correlation is not very big. Historically, some psychologists pointed to these results and said “see, ability is all of a piece—scores on all ability tests are related.” Others said “Look how small those correlations are. There are multiple abilities.” Rose falls in the latter camp, describing the correlations as “not particularly strong.” (p. 90)

But most psychologists stopped playing this game in the 1990s. The resolution came through the work of John Carroll, who developed a third model in which abilities form a hierarchy. We can think of abilities as separate, but nested in higher-order abilities. Hence, there is a general, all-purpose intelligence, and it influences other abilities, so they are correlated. But the abilities nested within general intelligence are independent, so the correlations are modest. Thus, Rose’s jaggedness principle is certainly not new to psychology, and it’s incomplete.

The Context Principle

The second of Rose’s principles holds that personality traits don’t exist, and there’s a similar problem with this claim: Rose describes a concept with limited predictive power as having none at all. The most commonly accepted theory holds that personality can be described by variation on five dimensions: Extraversion, Openness, Conscientiousness, Neuroticism, and Agreeableness. If I know where you fall on each of these dimensions, how much can I predict about your behavior?

Some, but not that much. For one thing, context matters. If you’re at a party where everyone knows one another but you know no one, that situation will have a sizable impact on your propensity to act extroverted, whatever your personality. For another thing (and as Rose emphasizes) these dimensions can vary within an individual, according to context—I may act extroverted in business situations, but introverted in social situations. So sure, if you add more information (context) to personality traits, you will be better at predicting what I do. But that doesn’t mean that traits alone are meaningless.

Chapter 1 offers another example of Rose dismissing a concept with limited predictive power as having none at all. He cites a brain imaging experiment in which the location of brain activity associated with a particular task showed a lot of variation across individuals—so much variation that the average was deceptive; it’s like the man who put his feet in a freezer and head in an oven and was, on average, quite comfortable. The average brain described a state that no brain was in during the task.

Rose is, of course, right that this can be a hazard of averaging. But he goes much, much too far when he claims that “The guiding assumption of decades of neuroscience is unfounded.” (p. 22) Many cognitive functions are localized in the same place—roughly—in different brains. Damage to the occipital cortex results in predictable consequences to vision, for example. The brain damage caused by Parkinson’s Disease leads to predictable motor and cognitive deficits. Yes, there’s variation, especially as you try to make the predictions (either in the brain or in behavior) more fine-grained, but that doesn’t mean that averaging is meaningless.

The Pathways Principle

Rose’s third principle suggests that there are multiple ways to reach a goal like walking or reading, and that there is not a fixed set of stages through which each of us passes. I think he’s absolutely right about this, and I have written about the lack of support for stage theories of development.

What does Rose see as the implications of these three principles? If this is the end of average, what comes next?

There’s a chapter on recommendations for business, but I’ll focus on the higher education chapter. The recommendations are surprising in that they don’t connect to the principals from Part II. That is, you might expect Rose to write “the current system uses personality information, but I’ve just shown you that personality traits don’t exist, so that means we should do things as follows.” Direct ties to the three principles are absent.

Also surprising is that Rose begins the chapter by declaring a single purpose for higher education (in his view). That purpose is preparation for a career, and the declaration will bring some readers to a full stop, because it’s not the purpose many students envision: some people go to college because they are intellectually curious, for example, or to fulfill their vision of being an educated person, or to learn about themselves. What’s surprising is that a book devoted to individuality limits higher education to a single goal.

Rose’s recommendations follow from his vision of education’s purpose. He thinks students should earn credentials, not diplomas. In other words, a school would not certify that you’re “educated in computer science” but that you have specific knowledge and skills—that you can program games on handheld devices, for example. He think grades should be replaced by testaments of competency; the school affirms that you’ve mastered the skills and knowledge, period. Finally, Rose argues that students should have more flexibility in choosing their educational pathways.


These are interesting, bold proposals. There’s plenty of room for questions—for example, whether it’s really plausible that schools could coordinate to define a single criterion for a credential as Rose wants. But after all, Rose’s purpose here is not a detailed blueprint, but a vision that will spark conversation.

My concern lies not with the details, but with the fact that the recommendations are only loosely related to personalization. Other than the call for greater choice, the other ideas seek to change what college is for, not how it works, with the goal of making it more personal.

What’s perhaps most striking to me about “The End of Average” is any perception that educators need to be convinced regarding the book’s main thesis. Tuning educational experiences to individual children has been a goal for education researchers since the 1950s, at the least—“one size fits all” remains a common pejorative applied to educational practices, not an accolade. Educators don’t fail to personalize because it never occurred to them to do so. The obstacles are (1) cost in personnel, materials et al., (2) reluctance to greatly increase learning from screens, and most formidably; (3) lack of knowledge. To personalize effectively, you need to know which differences among children are meaningful and how to cater to them.

“The End of Average” doesn’t offer new evidence as to what those differences might be, or how to honor them. The difficult problem is not persuading people that abilities are jagged, but figuring out how to reliably identify those abilities, and how to help students capitalize on them to meet educational goals. Likewise, the challenge is not persuading people that personality interacts with context, but identifying which contexts matter for which children. And if there is more than one way to reach an educational goal (proficiency in algebra, for example) we need to know how to identify those different pathways, and how select a pathway for a given child. Rose answers these questions by his personal story—he figured out what works for him. But most children don’t, and adults can’t expect them to.

Get our email newsletterSign me up
Keep up to date with our email newsletterSign me up