Jeff Bezos recently announced that he would commit two billion dollars to two initiatives, one of which was to create a network of full-scholarship preschools in underserved communities.

Reaction has been “wary,” focusing mostly on the lack of detail in the announcement. (There was also one of those periodic meditations on tech moguls’ love affair with Montessori education.) The two professional organizations of Montessori educators—no doubt hoping for an unprecedented spotlight on and promotion of work they hold dear—issued statements that fizzed with enthusiasm (see here for AMI & AMS).

​For my part, I was focused on one word in the announcement

Despite this headline in Chalkbeat, Bezos is not proposing to launch and operate Montessori preschools, but rather Montessori-INSPIRED preschools. That’s a huge difference, because although there are some studies showing an advantage to the Montessori method (see here, here, and here) research also shows that fidelity matters—children in Montessori classrooms “supplemented” with non-Montessori materials learned less than children in high-fidelity classrooms (see here and here).

I hope that Mr. Bezos and whoever he listens to on education matters are keeping in mind that the method has a lot of components, and, excepting the consequences of adding materials to the classroom, we don’t have data on outcomes when the method is tampered with.

What happens if you

• employ teachers who lack Montessori training? (There are thousands of teacher training programs in the US. Fewer than 25 offer Montessori training.)
• eliminate or shorten the 3 hour work cycle typical of Montessori preschool classrooms?
• eliminate or change the multi-year age groupings?
• eliminate or change Montessori scripted lessons? (Did you know that Montessori uses scripted lessons?)
• eliminate or change the curriculum?
• eliminate or change the Montessori conception of a prepared environment?
• change what is usually a high student-teacher ratio?

I don’t know the answer to any of these questions. I don’t think anyone does.

The Executive Director of the Yale Education Studies program wondered, in a New York Times op-ed, why Bezos didn’t make use of existing institutions to promote his preschool vision, rather than creating a network out of whole cloth? One obvious answer is that he wants tighter control to shape the organization as he sees fit, and to populate it with people he trusts.

Another likely reason is that he’s (rightly) suspicious that Montessori educators will be sticklers about the method, and he wants the flexibility to adapt the method as he sees fit. This may even be what he meant by another phrase in the brief announcement that drew a lot of attention: “the child will be the customer.”

Fair enough, it’s his money. But if that’s true, you may as well drop the “Montessori-inspired” bit.

Indeed, I’m predicting that picking and choosing elements of the Montessori toddler program (and not adopting it wholesale) will yield student outcomes (academic and social) that are indistinguishable from other preschools. I think the components interlock and all are integral to its success.

​Montessori is, indeed, “inspiring” but using the education program as a jumping off point for your own homebrew will, I predict, disappoint.

On September 15 I tweeted about a new meta-analysis that examines the impact of auditory distraction on reading. It’s an issue of broad concern, as many of us read at work in noisy office environments, and when we read for pleasure we may be on a subway, at a playground, and so on. Students and educators are keenly interested in this issue, because some students like to read with music on in the background and some educators wonder whether that affects comprehension.

The article concluded that that background noise, speech, and music all have small but reliable negative impacts on reading comprehension.
In response, several folks on twitter commented as much to say “ok, so we should tell kids not read with music on.”

I am not so sure.

This is a point of interpretation around which Todd Rose framed his book, The End of Average. Now I didn’t care much for this book, because I thought Rose took a valid concern and ran much too far with it, but here it’s applicable.

An average is meant as a summary that gives you a sense of the central tendency of a distribution. That doesn’t mean it is a good representative of every data point. To use Rose’s example, if you measure a large group of airplane pilots and find their average height is 69 inches, and then design airplane cockpits assuming “pilots are 69 inches tall,” well, you’ll be disappointed. The cockpit will be a good fit for a few, but will be too big or too small for most.

I criticized Rose’s book because I argued that (1) many principles of the mind do apply pretty well across the board—everyone’s attention is limited, for example and; (2) psychologists are generally aware of the problem Rose identifies. The entire subfield called individual differences is devoted to identifying ways in which we all differ.

The influence of background music on reading may be a case where Rose’s warning is pertinent. The meta-analysis reports a small, consistent cost to reading comprehension when listening to music. Looking at the breakdown of individual studies it’s easy to see that the studies trend towards the stated conclusion. 

But there is also a lot of variability—I’m not referring to the dot representing the mean of each study, but to the dotted lines around each of those dots, which shows the variability associated with that mean.

​Contrast that with the studies on the effect that background speech has on reading comprehension. 

What this indicates is that, while mean of the grand distribution may show a small hit to comprehension when background music plays, it's NOT the case that every child reads a little worse with background music on. Part, but not all, of that variability is noisy measurement.

As the article notes, researchers have sought variables that differentiate why music hurts, fails to influence, or even helps comprehension. For a while they thought introversion/extraversion might be the answer, but that didn’t pan out. Still, I think this is a case where individual difference play an important role. 

As far as practice goes, I think this finding could be offered as support for a decision not to play music to every child in a classroom. In a big sample, you’d say it will reduce mean comprehension. But I don’t think it supports telling individual children not to listen to music while they read. (Note too, I see this as one factor among many a teacher would consider in a decision of this sort.)

Here’s another reason I personally wouldn’t be too quick to interpret this meta-analysis as showing people should never listen to music while reading. Some of my students say they like music playing in the background because it makes them less anxious. It could be that a laboratory situation (with no stakes) means these students aren’t anxious (and hence show little cost when the music is off) but would have a harder time reading without music when they are studying. In other words, the laboratory situation may underestimate the frequency that music provides a benefit for a subset of students. 

Adam Grant published an op-ed in the New York Times yesterday with the subhead “Advice for college students: The best experts sometimes make the worst educators.” I know he didn’t write that description, but in this case, it’s apt. The article offers advice to students about how to pick professors. Unfortunately, I think Grant gets it wrong.

He emphasizes that people who know a lot aren’t necessarily great teachers; to teach well you need a different type of knowledge, knowledge not just of complex ideas, but also of how to communicate those ideas. That’s commonly called pedagogical content knowledge and it’s old stuff to K-12 educators. Teachers spend a great deal of time thinking about how to communicate what they know to someone who knows much less.

Grant points out (rightly, I think) that some college instructors spend little time thinking about it. He cites data showing that those who know more about their content area are likely to be even worse in communicating what they know. This is commonly called the curse of knowledge; as your knowledge gets more and more advanced, it becomes harder and harder to remember what it’s like to be a novice; you can't take their perspective as well when teaching. For example, some experiments show that children more successfully learn from other children who are just kind of coping with a skill than from a child who has mastered it.

Grant proposes three principles by which college students might choose instructors:

1. Following the curse of knowledge, he suggests taking courses from people who have more recently learned themselves, such as graduate students, rather than more experienced faculty.
2. In addition, he suggests that people who had to work really hard to get to where they are will better understand the struggle of learning. (As opposed to someone with a natural talent for the field who picked it up easily.)
3. Students should focus on how well someone communicates, not just on how well they know. (As measured, I guess, by their prominence in their field.)

 
I see real problems with this advice.
 
First, it predicts an inverse correlation between instructional quality and research productivity. Measures of both constructs are controversial, but when researchers have tried, the correlation is no different than zero (see here and here). Being a great (or indifferent) researcher predicts nothing about the quality of your teaching.
 
Second, picking instructors based on their inexperience ignores the fact that factors other than the curse of knowledge contribute to teaching effectiveness. Most instructors improve with experience as they gain feedback on their teaching. Grant seems to predict that instructors, on average, become less and less effective as they grow more distant from their initial training. We know that’s not true of K-12 educators, and for the same reason, I doubt it’s true for college instructors.
 
Third, two out of the three principles Grant suggests aren’t really practical for students. #1 rests on information that a student can’t access: how will a student know if an instructor worked hard to get where they are? And #3 sounds a lot like “pick a good teacher.”
 
I’ll try to do better. Here’s my advice.

1. Attend a college where undergraduate education is the primary concern. If you get into MIT or Harvard, congratulations! But you should know that the criterion for being allowed to teach there is not excellence in teaching. Grant’s absolutely right about that.
   
   What he doesn’t mention is that good teaching is the key criterion for employment at most colleges. You just haven’t heard of most of them because prestige comes from faculty scholarship and research dollars. Attend a small college and you’ll get more attention from faculty who care about and focus their energy on teaching.
   
   There are trade-offs, however, notably in the opportunities for proactive, self-starting students. When I taught at Williams College, my Introduction to Cognitive Psychology class had about 25 students (at UVa I teach 350) and in my lab I mentored only undergraduates--I didn’t have to divide my attention with graduate students. Sounds great, but if you were a Williams student interested in cognitive psychology I was it. At UVa you can study cognition and education with me…or the epigenetics of cognition with Jess Connelly, or social aspects of cognition with Jamie Morris, or memory with Chad Dodson, or perception with Dennis Proffit, or cognition in the elderly with Tim Salthouse, or computational modelling, cognition and the brain with Per Sederberg or Nicole Long. (I have more good things to say about the undergraduate experience at UVa, but I'll save them for another time.)
   
   
2. Students don’t take every college course for the same reason, or at least they shouldn’t. Grant points to “learning” as the reason to take courses, but students take some courses for the experience, and the possibility of kindling interest or even passion, like the senior engineer taking an art history course, or the psychology major taking an environmental science course. They don't care if they master this introductory content, or even if they remember details a year later, and I get that. 
   
   Student course evaluations don’t measure teaching quality very well, but they probably do measure how much people like the way the professor teaches. So when “fun and interest” are the right criteria, it makes sense to take a course that your peers have said is fun and interesting.
   
   
3. For other courses, mastering content is essential, as when the second year psychology student takes their first course in statistics and methods. They will call on this content in future courses, so it must be mastered and the entertaining professor may or may not get them there. I think about the best one can do is to ask more senior students who have taken follow up courses whether they felt well prepared by the instructor.
   
   

A final point to consider. Expectations for independent work are very high in college. By the time you get there you’ve had 12 years of experience learning, and so professors figure you are pretty resourceful in teaching yourself—students are usually expected to put in 3 to 6 hours on their own for every hour they are in class. That’s not a free pass for poor college instructors, but it does highlight the possibility for success even in courses taught by an indifferent professor.

It’s been 13 years since I wrote about the lack of evidence supporting learning styles theories in American Educator. My editor thought it would be good to review the evidence published since 2005 and write an update. To be honest, I wasn’t that keen on the idea, as I’m pretty sick of writing on this topic. (I've written pieces for several other outlets since 2005.) But she was right—there are some new findings of interest.  The whole article is open access, and you can find it here. What follows is a brief summary.

1. The best known review (that of Pashler et al in 2008) prompted a small rush of new studies. Like previous studies, many of these were poorly designed. I identified 16 studies that were interpretable. Three offered support for two different learning styles theories, which I suggested indicated these theories merit further investigation. Thirteen studies did not support styles theories. In short, this review leads to the same conclusion that previous reviews have. There are many theories, but none come close to the standard of evidence we'd want before advising teachers to revise their practice to align with the theory.
2. What’s new: Previous reviews have consistently shown that people believe they have a learning style, but only recently has evidence emerged showing that people will act on that belief, recoding information to be consistent with their style, if given the chance. This conclusion is tentative (there are only a handful of studies showing it, but the studies were very tightly crafted) and it only applies to a couple of learning styles theories. Acting on your learning style does not bring any benefit—you don’t do the task any more efficiently.
3. There are, however, substantial task effects. That is, you’re much more successful doing certain tasks verbally, and treating other tasks visually. The cost or benefit of using one or the other type of processing is the same, whether you’re a “verbalizer” or “visualizer.”

 
I close the article with some considerations for the classroom, focusing on these task effects.

The recent kerfuffle concerning Hart & Risley (1995) and the 30 million word gap offers an object lesson in science, the interpretation of science, and the relation of science and policy.

Let’s start with the new science. Douglas Sperry and colleagues sought to replicate Hart & Risley, who reported the 30 million word gap—that’s the projected difference in total number of words directed to a child by caregivers when comparing children of parents on public assistance and children of parents in professional positions.  Sperry and his team claim not to find a statistically reliable difference among parents of different social classes.
​
Twitter was quick to pounce:

​Coverage from NPR made it sound like Hart & Risley had been debunked, with the headline “Let’s stop talking about the 30 million word gap.”

But the Sperry report doesn’t really upend Hart & Risley.

First, Sperry et al. claim that the Hart & Risley finding has never been replicated. I am not sure what Sperry et al. mean by “replicate,” because the conceptual idea that socioeconomic status and volume of caregiver→child speech has been replicated. (The following list is not offered as complete—I stopped looking after I found five.)

Gilkerson et al (2017)
Hoff (2003)
Hoff-Ginsberg (1998)
Huttenlocher et al (2010)
Rowe (2008)

None of these is an exact replication---they have variations in methods, population, and analyses. The same is true of Sperry et al, and funnily enough that study has a fairly significant difference—they didn't include a group of professional parents which is key if your main concern is the size of the gap between professional and public assistance parents.

It’s also worth noting that Sperry et al speculated that their results may be more representative of how parents actually talk, because the researchers used an unobtrusive recording system. Hart and Risley (and most other researchers) had a researcher observing parents and children, so perhaps parents in different SES groups reacted differently in the presence of researcher, the guess being that poor people might clam up, or the wealthier might show off by talking more. I'll leave alone the assumptions underlying that speculation, but I will point out that, first, I doubt observer effects would count for much because the observations occurred over the course of years; people get used to being observed. Second, Gilkerson et al (2017) used the same unobtrusive system that Sperry did and observed the association of SES and caregiver speech.

Another odd thing about the Sperry et al paper is their emphasis on bystander speech (i.e., speech that is not directed to the child but happens in the child’s presence.) This is odd because multiple studies indicate that child *can* learn from such speech, but more often learn little or nothing (e.g., here and here).

Sperry points out that in some cultures children are seldom addressed directly, yet learn to talk. But maybe children in those cultures learn “if someone’s talking, I should listen, even if it’s not addressed to me because they may say something that’s important to me.” In most households in the US, if you’re not being addressed it’s less likely that the speech is important to you, so the child likely does not redirect attention from whatever he or she was doing to the speech.

So all in all, I don’t think this failure to replicate overturns Hart & Risley, coming as it does in the face of several successful replications. As to whether the gap is 30 million or some other figure…I don’t know, maybe somebody thought the absolute value mattered. I doubt any psychologists did. We would care about the predictive power of the caregiver speech. On the whole, there’s still pretty good reason to think there’s an association between SES and child-directed speech from parents. (For more on this issue, see the recent blog by Roberta Golinkoff and her colleagues.)

BUT thinking that there’s pretty good evidence for the association is not AT ALL the same as thinking it ought to influence policy. There are two issues here.

First, do we understand this phenomenon well enough to intervene? Second, should we?

In answer to “do we understand enough?” I’d say “no.” The volume of words is the variable you hear about most, but it may not be the most important. It may be the conversational back and forth that matters. Or the diversity of speech. Or the gestures that go with speech. And oral language is only one contributor to vocabulary size and syntactic complexity. Maybe we should intervene to get more parents reading to their children, or better, using dialogic reading strategies. At the very least, I’d like to see a small-scale intervention study (not just a correlational study) showing positive results of asking caregivers to talk their children more, before I would be ready to draw a strong conclusion that volume of caregiver talk is causal to children’s language capabilities.

The second question—if we were pretty sure we knew that a factor is causal, should we intervene?—is much more fraught. As I have considered  at length elsewhere, questions like this are outside of the realm of science. You’re contemplating using science, but whether or not to intervene is not a scientific question. It’s a question of values. You are seeking to change the world. That brings costs and the promise of benefits. Will it be worth it? It depends on what you value.

That’s what people on Twitter were responding to on this issue (some explicitly, some implicitly)—the assumption that parents in poverty ought to parent more like middle-class parents. Then their kids would be successful…according to middle class values.  That conclusion entails the obvious corollary that parents can eliminate any disadvantage their child has, so if they don’t, well, it’s no one’s fault but their own.

I agree with this argument to a point. The prospect of using science to tell people how to parent makes me very uneasy.

On the other hand, should we fiercely defend parenting practices in the name of cultural equality or because we don’t want to let powerful institutions off the hook if we know those practices put children at a disadvantage in school, and later, in the job market? (Reminder, I don’t think that such evidence currently exists on parental speech volume.) Wealthy parents keep pace with what researchers suggest will help children flourish, and then defend the right of parents living in poverty to use parenting practices that put poor children at a disadvantage? In a long series of tweets (in which she raises many of the criticisms of the Sperry et al study that I raised above) Twitter user @kimmaytube closed with this pointed comment

What is the role of a scientist in these difficult application issues? For better or worse, I have come to what seems the obvious resolution: give people the fullest information you can and let them decide. 

People who are concerned about the impact of proposed applications of science will have on low-income communities and individuals raise valid points, which they sometimes undercut with rash claims about the invalidity of the scientific studies.

I’ve seen this repeatedly on the subject of grit and self-control. Again, there are very legitimate questions to ask about the values that underlie the assumption that we should make kids more self-controlled and/or more gritty, and questions about the costs to children and institutions should we try to intervene that way. These are separate issues than questions about the scientific standing of grit and self-control as explanatory constructs.

​Twitter notwithstanding, I encourage you to bear the distinction in mind. 

E. D. Hirsch celebrated his 90th birthday a few days ago.

What better time to remind ourselves of his contributions to American education? I hope Hirsch will forgive me if I do not dwell here on his practical and arguably greatest contribution—the establishment of the Core Knowledge Foundation, which has both produced outstanding curricular materials (many distributed without cost) and advocated for equitable, outstanding education for all. (I sat on the board of the foundation for some years.)

Instead, I'll focus on three profound ideas that Hirsch developed and explicated, and that have had a substantial influence on my thinking. 

1. The role of knowledge in reading. Background knowledge is the main driver of language comprehension, whether written or spoken.   Disadvantaged students are disproportionately dependent on schools to provide the background information that will make them effective readers because wealthy students have greater opportunity to gain this knowledge at home. These were the key ideas in Cultural Literacy. That 1987 volume became a best seller mainly because of the list at the back of the book, “What Literate Americans Know.” The list also gave Hirsch the undeserved reputation of an ultra-conservative because he was apparently advocating that school children spend most of their time memorizing the names of dead white males. You couldn’t hold that opinion if you actually read the book, but most people didn’t.
2. The importance of shared knowledge in citizenship. The American Founders recognized that this country, as a multi-ethnic society, faced a peculiar dilemma among nations; how to encourage a feeling of commonality and mutual responsibility among diverse citizenry? They saw a common body of knowledge as crucial to the cohesiveness of American citizenry where individuals held allegiance to other tribes—English, Scottish, German, etc. In The Making of Americans Hirsch argues for a “civic core,” and for the idea that each of us as individuals can and should have commonality in the public sphere, even as we have individuality and different group allegiances in the private sphere. The former does not diminish the latter.
3. The seeds of Americans' denigration of knowledge. Why would it be controversial to argue that children should share some common knowledge? The seeds of that idea lay in the Romantic response to the Enlightenment. Whereas Enlightenment thinkers esteemed knowledge of the world, the Romantics emphasized feeling, emotion, and especially esteemed the impulse of the individual. Whereas Enlightenment thinkers would emphasize social institutions as beneficial to human well-being and flourishing, Romantics depicted social institutions as problems, and portrayed humankind in its natural state as sanctified. In The Schools We Need and Why We Don't Have Them and in Why Knowledge Matters, Hirsch has argued that early educational theorists were influenced by Romantics to a degree few appreciate, and that we today are inheritors of their mostly flawed assumptions about human nature. These assumptions lead to a reverence for individuality and for nature, and a corresponding denigration of knowledge deemed important enough for all to know.

 
Needless to say, a paragraph doesn't begin to do justice to each of these ideas. If they are not familiar, I encourage you to explore them further--I've already made it easy by including the links to buy the books!

Yesterday, the day of the student walkouts to protest gun violence, Robert Pondiscio published a blog  suggesting that participating students ought to face the usual consequences (whatever they might be at their school) for missing class.

I posted a link to this blog on social media with the provocative (but not inaccurate) title "Why students participating in the walkout today ought to be punished." A few people expressed puzzlement and a few expressed outrage, so I thought I'd explain.

Pondiscio's point is easily summarized with this quote: "By its very nature, an act of civil disobedience means the protester refuses to comply  with rule, norms, and expectations.”

Pondiscio goes farther than I would, saying that “compliance rob[s] the protest of any meaning,” and Andy Rotherham seems to agree in his own blog, saying that if students know in advance a walkout is consequence-free “it’s theater.” I think it's still meaningful to show support, but I don’t see how you can argue that it’s an act of civil disobedience or a walkout.

Rotherham touches on another theme that I agree with: adults coordinating the walkout, seeking ways to make it easy and "safe" for kids, fits a more general pattern of adults today exercising too much control over kids' lives, and keeping them safe in ways that ultimately backfire. (Watch for Greg Lukianoff and Jon Haidt's book on this subject, The Coddling of the American Mind.)

Both Pondiscio and Rotherham made another point that I found much more telling, and is the reason I thought the blog worth sharing: teachers and administrators allowing students to attend walkouts sets a terrible precedent.

My concern is that educators who suspended the usual consequences for students missing class did so because they agreed with the cause of the protesters. I do too, but it seems pretty clear that I can’t suspend a policy only for causes I agree with. So what happens when people want to cut class not to protest gun violence, but to support gun owners rights, or to lower the drinking age, or to show support for Nancy Pelosi or Donald Trump? “Slippery slope” arguments often make me roll my eyes, but in this case, I think it’s apt. Are students to be allowed to walk out of classes for any protest? (I imagine middle-school me, at the pizza place during math class, telling the vice principal between bites "This is the way I protest congressional inaction on term limits.")

Maybe there's a good argument for educators taking the role of sanctioning or punishing protests based on their content. I haven't heard it. 

In a 2014 commentary, Gail Lovette and I argued that many educators have a misconception about the teaching of reading comprehension. We suggested that they often think of comprehension as a transferable skill—as reading comprehension improves, it improves for all texts. We suggested, in contrast, the comprehension is highly text-specific and dependent on background knowledge.

Further, we suggested that all-purpose comprehension processes (e.g., monitoring whether you’re understanding, remembering to coordinate meaning across sentences and paragraphs) makes a contribution, but is not much susceptible to practice. As evidence, we cited eight meta-analyses that examined data from studies of comprehension strategy instruction. All of these analyses showed a sizable benefit for strategy instruction, but the amount of instruction or practice had no impact on the benefit. Our interpretation was the strategy instruction told students (who didn’t already know it) that things like coordinating meaning was a good thing to do, but such instruction can’t tell you how to do it, because the how depends on the particular meaning. The instruction can’t be all-purpose.

A new meta-analysis shows the same pattern of data.

Amy Elleman summarized data in a meta-analyis of 25 studies that used various methods to teach children to make inferences, and to apply them to texts. She examined three separate measures of comprehension: general comprehension, inferences in particular, and understanding of things literally stated in the text. She also separated the benefit of instruction to skilled and less-skilled readers.

The data showed “moderate to large” effects of instruction to general comprehension and to making inferences for both skilled and less skilled readers. The pattern differed for the “literal” measure, however, with skilled readers showing almost no gain but unskilled readers again showing a sizable gain. 

It’s somewhat surprising that these students showed such a large gain on an outcome for which they received no instruction…but it must be remembered that less skilled readers are often characterized as somewhat passive in their reading. Hence, instruction may have improved literal comprehension by prompting them tackle the task with more cognitive resources.
​
Especially noteworthy to me was that Elleman observed no effect of what she called “Instruction intensity” i.e., number of hours devoted to inference instruction, as Lovette and I noted for the other eight meta-analyses.

This finding was not discussed in the article, but supports Willingham & Lovette’s interpretation of the effect of comprehension instruction: it alerts students to the importance of making inferences, and perhaps more broadly (for less skilled readers) that it is important to THINK while you read. But practicing inferences does not lead to a general inferencing skill for two reasons. One, as noted, inferencing depends on the particular text, and two, whatever cognitive processes contribute to inferencing are already well practiced from use in oral language---we continually draw inferences in conversation.
​
As Willingham & Lovette suggested, comprehension instruction is a great idea, because research consistently shows a large benefit of such instruction. But just as consistently, it shows that brief instruction leads to the same outcome as longer instruction. 

Not long ago, a friend told me he was going across country to visit his friend who had lost his wife six months previously. He mentioned that he had not gone to the funeral. “I don’t get that much time off so I can only go once. Everyone’s at the funeral. Somebody needs to be there six months later.”

It’s important to keep this perspective in mind as we continue to process the horrific school shooting in Parkland, Florida. Just as my friend knew that losing your spouse is not resolved in six months, we might guess that the trauma associated with attending a high school where murder took place would have long term consequences.

In fact, Louis-Philippe Beland and Dongwoo Kim have examined the educational consequences for survivors. Using the Report on School Associated Violent Deaths from the National School Safety Center, they identified 104 shootings categorized as homicidal and 53 as suicidal. (Shootings took place on the property of a public or private US school, or while a person was attending or on their way to or from a school-sponsored event.)

School performance data were obtained from each state’s Department of Education website. The researchers used other schools in the same district for comparisons, on the reasoning they would be roughly matched for demographics. (I wonder about the soundness of this assumption.) The researchers examined three main outcomes.

​First, they examined whether enrollment in a school would go down after a shooting. (Note: all of the effects described apply to homicidal shootings. There were no effects of suicidal shootings on any of the outcomes.) They found that it did decrease, presumably as parents who could selected other schools. This effect was only observed in 9th grade enrollments, however. Perhaps families with children already attending a school felt more committed to that school. 

Second, they tested whether deadly shootings lowered test scores in later years. They found that it did. (Click for larger image.)

Based on the first, result, it could be that lower scores are a consequence of the opt-out; maybe it’s the most capable 9th graders students who choose not to attend the school where the shooting took place. To test the possibility the researchers examined a subset of the data from California schools, where they could access student-level data. The effect replicated. In other words, it’s not due to changes in the population. When researchers examine test scores of individual students year to year, those scores dropped after the shooting.

Third, the researchers examined behavioral outcomes including graduation rates, attendances, and suspensions. They observed no effects.

On the one hand, it may seem unsurprising that school shootings affect academic outcomes three years later. On the other hand, there is a rich research literature showing that we often overestimate how long we’ll feel distressed in the face of a negative event. But in this case predictions of negative consequences are accurate. Attending a high school where a homicide takes place prompts trauma, and that impacts students school experience and achievement.

​The needs of the students who remain at Marjory Stoneman Douglas High School may be deemed less urgent that those of the immediate families of the slain. That’s a fair assessment. But the needs of the survivors are real, and we must ask how we can address them. And we must not forget the students who attend these schools where murder took place within the last three years:

• Marshall County High School
• Aztec High School
• Rancho Tehama Elementary School
• Freeman High School
• North Park Elementary School
• Townville Elementary School
• Alpine High School
• Jeremiah Burke High School
• Antigo High School
• Independence High School
• Mojave High School.

It's one of the most familiar (and frustrating) problems teachers encounter. Students learn something new (say, a standard solution technique for a standard mathematical problem) but then fail to recognize the problem type when they encounter it again. For example, students may learn the idea of a "common factor" in equation form in an algebra class and fail to see that the same idea can be applied in a word problem. 

This is usually called the problem of transfer, and a classic laboratory problem was devised by Mary Gick & Keith Holyoak. 

It's a difficult problem, and most people are unable to solve it. If I tell you that this story may provide some inspiration, you'll probably get it. 

The interesting finding re: transfer is that if you simply ask people to read the military story and then to try to solve the radiation problem, most people don't see the analogy. They have no problem seeing it when told to look. But they don't spontaneously see it. And of course there's usually not someone around to give you a gentle elbow in the ribs, and drop a hint. You have to think of it on your own. 

This problem has proven very difficult to solve. One promising solution is to give subjects something to do that forces them to focus on the underlying structure of the problem. The underlying structure in the example above is "to avoid collateral damage, disperse your forces and converge at the point of attack." The underlying structure is expressed with tumors and rays in the problem and via a fortress and armies in the story. 

Dedre Gentner and her collaborators have tried (with some success) to improve transfer by asking people to compare problems. When you compare problems with the same deep structure, that obviously focuses attention on that deep structure and so you'll remember it better later.

A disadvantage of comparison is that the instructor must provide parallel versions for all problems. Ricardo Minervino and his colleagues sought a technique that would provide similar benefits, but might be more applicable to classroom situations.

In their experiment, subjects first read the fortress-army story (along with two other stories). At transfer, everyone was asked to solve the tumor-rays problem, but one group was also first asked to invent an analogous problem. Examples of the problems subjects invented appear below:

Subjects who invented an analogous problem were more likely to successfully solve the radiation problem compared to subjects not asked to invent a problem, 25% vs 10%. Independent raters judged the quality of the invented problems as analogous to the target problem, and further analysis showed that the better the analogy they created, the more likely they were to solve the target.

The experimenters also showed that it wasn't just the deeper thought required by the problem creation that made those subjects more likely to get the problem right. Another control group was given the tumor-rays problem and were asked to create an analogous problem before solving it, but they did not read the fortress-army story beforehand. Just 10% of these subjects solved the radiation problem.

In terms of the psychological mechanism behind this effect, it's not a huge surprise. Again, it's a technique that prompts people to focus attention on the deep structure, just as comparison does. What's nice about this technique is that, as the authors note, it removes the burden from the instructor to devise parallel problems. But that also gives students the freedom to create an "analogous" problem that isn't really analogous. So the instructor needs to check up on the problems students create.

Still, it's useful to know about the consequences when students create a new problem, and teachers may find it useful in some contexts