Jeremy Horpedahl Reviews The Anxious Generation: How the Great Rewiring of Childhood Is Causing an Epidemic of Mental Illness, by Jonathan Haidt

See What's a Parent to Do?

"Haidt argues very strongly that the relationship between social-media-equipped smartphones and teen mental health is causal, not a mere correlation. What is the evidence? For a book that asserts this so heavily, you might be surprised to learn that there is just one paragraph focusing on randomized-control trials that address this question. It can be found on page 148. In that paragraph he summarizes just two studies that attempt to measure the effect of social media on mental health. In a footnote, he tells us further that there are 14 RCTs showing harm, and another 6 that found no harm (but he regards these 6 as low quality studies), and then points us to an online Google Document that he put together with his collaborators.

There are multiple documents that he has put together on his website that relate to the research behind this book, but the one on Social Media and Mental Health runs 356 pages, longer than the text of the book itself! Haidt is to be applauded for putting this all online transparently, but as a social-science nerd, I would have liked to see this take up more than a single paragraph of the book. A few chapters perhaps? But that probably wouldn’t have landed the book on the New York Times bestseller list.

Anyway, back to the 356-page Google Document. The discussion of RCTs begins on page 168 (of the current version as I write—this is a living document) and runs for 20 pages. There are now 23 studies showing causal negative effects, and another 8 studies showing no effect, 11 more studies than when the book went to print just a few weeks ago. I won’t dive into all 31 of these studies, but given that Haidt is laser-focused for both the trends and policy recommendations on teenage girls, how many of the RCT studies would you guess are about teenage girls? The answer: just one. The other papers study college undergraduates, adults, or young adults.

It’s not that there is nothing we can learn about teenage mental health by studying people older than them. But what’s so striking about this fact—just one study of teenage girls!—is that Haidt is so confident in his overall hypothesis despite the evidence being so razor thin. And what of this one single study? Also interesting: it wasn’t even teens in the United States, but rather in the Netherlands. Again, there is nothing wrong about studies outside of the United States, but as I will argue below, the worst of the mental health problems seem confined to teenage girls in the United States, yet we have no studies of teenage girls in the United States.

The paper in question is well done. It randomly assigns girls to two groups, and one group shows manipulated Instagram photos that make the subjects more attractive. The girls that received the treatment reported lower body satisfaction, about 0.4 points on a 9-point scale. This result is statistically significant, but… is it enough to worry us? Is it good proof that social media is causing a mental health crisis, when you have a small change on the scale of body image from one single study of 144 teenage girls in the Netherlands, with no follow-up for long-term effects? This seems, to me, to be a very weak reed to build an entire apparatus of restricting phone use for teenagers around the world.

Is It Really Happening Everywhere?

While Haidt spends much of the book discussing evidence from the United States, he suggests that this is a global phenomenon. In Chapter 1, he spends three pages expanding his charts on teen mental health to other English-speaking countries (Canada, the United Kingdom, Australia). He has another two and half pages on “the rest of the world,” but the evidence here is pretty thin: a chart on psychological stress in Nordic nations, and a chart on alienation in school by broad regions (Asia, Europe, and English-speaking Latin America).

As with the summary of studies, there are other Google Docs to consult from Haidt and collaborators. For example, while the Nordic nations get just one paragraph and one chart in the book, the Nordic adolescent mood disorders online document runs over 100 pages. And much of that document is less certain and ambiguous than the text of the book. While there is plenty of evidence of rising mental health diagnoses and self-reports, evidence on self-harm and suicide doesn’t show increases. In some cases, it shows decreases. In Denmark, self-harm was reduced by almost 50 percent from 2007 to 2016 among teenage girls and boys—there had been a rise in the decade before 2007, but it came back down after that. Teenage suicides in Sweden exhibited a similar pattern, with a rise from 2000 to about 2008, then coming back down.

To go beyond the Nordic Nations, another Google Doc (Haidt is very thorough and transparent) on Adolescent mood disorders is useful. But the studies summarized in Section 3.6, looking outside of the Anglosphere (United States, Canada, United Kingdom, Australia, New Zealand) are not very convincing. The first study they summarize looks at suicides rates for ages 15-24 from 2006-2017 in several high-income countries. The only clear increases are in the Anglosphere, and even if we limit the analysis to girls, only Spain is added to the unfortunate group of rising youth suicide. France and Italy are declining, Germany and Poland are flat. There is no evidence that social media and smartphones have proliferated less in those countries than the Anglosphere. They also look at OECD data, but it is no more promising for their hypothesis: “Teenage suicides rates have, on average, declined slightly over the past two decades or so.”

There does seem to be something particularly bad happening in the United States and other large English-speaking countries, but our non-English-speaking peer nations aren’t seeing the same trends (though some are seeing the rise in reported mental health issues).

What Is to Be Done?

We can think of the question “What is to be done?” in two ways. Haidt is convinced that the evidence is overwhelming on the connection between social media, smart phones, and teen mental health. If you are also convinced, the thing to be done is find policy solutions or suggest changes in social behavior.

But the second way to think of “What is to be done?” is to think about what further research needs to be done to better understand the relationship between social media and teen mental health. Perhaps the two questions can be merged: targeted policy interventions could also produce good research results, to be used for future potential interventions.

On page 263-64, Haidt suggests just that merging of the questions, when he proposes that state educational authorities set up random-assignment of schools into one of four groups, such as phone-free, free-play, both, and a control group (status quo policy). What’s really important here is to note that no such studies exist or Haidt would have cited them. He is really making recommendations without much good evidence yet. But as Haidt notes earlier in the book (page 249), 77 percent of schools in the United States already say that they ban phones—they just aren’t enforcing the bans.

So clearly schools have this power (as they do have the power to limit all sorts of student behaviors and activities), they just aren’t using it. Haidt says: use that power but use it in a way that we can learn from it. If the governors of, say, Utah, Florida, and Arkansas (three states that have passed some restrictions on youth social media use) took this opportunity to conduct randomized experiments on schools, other states could learn from their experiments. It may seem cruel to treat school children as test subjects, but that’s actually we are already doing, we’re just doing it poorly and in a way that is hard to study the actual effects.

Unfortunately, it does not seem that any well-done studies have tried this randomized approach yet (you may have heard about a new paper supposedly on phone bans in Norway, but Haidt acknowledged on social media that the paper “does not really tell us much”)."