The Age of Anxiety

Reviewing Haidt's book "The Anxious Generation"

In reading/listening to Jonathan Haidt’s new book (The Anxious Generation: How the Great Rewiring of Childhood is Causing an Epidemic of Mental Illness), I was reminded of how musicians Win Butler and Regine Chassagne of Arcade Fire kicked-off their 2022 album, We, with a song entitled “The Age of Anxiety,” which does much to capture the confusing feelings and emotions plaguing many residents of affluent countries in recent years:

“It's the age of doubt; And I doubt we'll figure it out. Is it you or is it me? The age of anxiety. Fight the fever with TV; In the age where nobody sleeps; And the pills do nothing for me; In the age of anxiety; When I look at you, I see what you want me to; When you look at me, you see what I want you to see;
It's a maze (It's a maze) of mirrors (of mirrors); It's a hologram of a ghost; And you can't quite touch it; Which is how it hurts us the most; It's all about you; It's not about you; It's all about you.”

These lyrics capture the temptation to become an online supplicant. The excessive concern for how we are perceived by the desirables--those outside of our social circle but inside the outer circle we strive to inhabit—and the desire to distort ourselves to conform to the imagined desires of these desirables is at the heart of why social media is dangerous and why it saps authenticity and propels us away from our nature, the higher and most excellent version of ourselves. These desires may take over our self-control and re-direct it toward unattainable ends of renown, based on shallow contributions of mere influence. Anxiety and depression accompany the focus on things we cannot control: such as the desire of the desirables.

The other side—less tempting to agreeable people—is to become a troll. Even those capable of fostering indifference to the views of strangers may nonetheless find themselves harmed by excessive social media use. The virtues of courage and fairness are activated when we see morally repulsive behavior, and our desire to stop or punish such behavior is strong, especially among the most virtuous among us. Yet, social media bathes us in a cesspool of morally repulsive behavior, as companies have learned that exposing us to this content drives our engagement. The result is that we foster fear, anger, and distrust, as we wallow in sins of strangers.

I regard Jonathan Haidt’s new book as excellent but incomplete. The rise in time spent in the virtual world is documented with alarm, using compelling logic and data; his practical recommendations to deter screen time and social media are entirely wise; I fully agree with each of them. My only serious complaint about the book is its vague description of the real world. Surely, there’s more to be said about the conditions of an optimal childhood than restoring technology, risk, and play to pre 21^st century levels. We get hints of this in his chapter on spiritual elevation, which is among the book’s strongest, but it is so focused on how social media descends from this plane to fully engage the subject. I also think it is fair to say that his evidence concerning over-protection in the real world and its consequences is weak.

Before I elaborate on the book’s strengths and weaknesses, let me summarize the key points. Haidt makes one major claim: We’ve given children too much protection in the real world and not enough from the virtual world. In establishing why protection is needed in virtual world, he makes a persuasive distinction between the two worlds, showing the fundamental inferiority of the virtual world in its capacity to foster and maintain healthy relationships. Real-world relationships are embodied, involve synchronous communication in small groups that are bound together by fate or a shared identity, such that they are costly to dissolve. In establishing why over-protection has been a problem in the real world, he distinguishes two mindsets for young adults: discover mode—characterized by a desire for learning opportunities, novelty, and demonstrations of self-efficacy, and defend mode: characterized by fear, a sense of scarcity in recognition or resources. Haidt, sensibly, wants adolescents and young adults to primarily live in discover mode, and believes that adults have stifled risks and opportunities to such an extent, that children are raised to be dependent and incompetent. It’s as if generations of intrepid boy and girl scouts—or farm hands and apprentices—have been replaced by a generation of entitled lords and ladies, who can scarcely function without their servants—usually parents. Their incompetence leads to anxiety, because they correctly assume they lack basic but important skills, like the ability to cook or drive a car. Their parents’ anxiety about exaggerated dangers, meanwhile, is perceived and incorporated.

As implied above, Haidt does an outstanding job of laying out a theoretical foundation for valuing real-world relationships for adolescents with a discovery-oriented mindset. Social media relationships are uncanny from an evolutionary perspective in their features and characteristics, and it requires a great deal of effort to remind ourselves of this and avoid the temptation to be a troll or a supplicant, or simply losing ourselves in the entertaining but empty watching and judging of others vying toward those ends. Barring or greatly restricting access to social media use for children and adolescents is wise, as it is to restrict their access to illegal drugs.

I agree with and value Haidt’s call for a play-based childhood. As an athletic and energetic boy, I looked forward to recess most of all and the opportunity to compete against the best boys in any sport available to us to play. To this day, I savor competition, physical, grinding competition in sports like basketball, soccer, and football. Such play affords many benefits: a healthy outlet for aggressive human instincts—which I certainly have, a chance to forge strong friendships, to learn and demonstrate courage in facing pain, enduring discomfort, and fairness in adhering to rules, valuing your teammates’ contributions, distributing opportunities.

But are kids playing less? When it comes to sports, the evidence say no. There has not really been any change since 1999 in sports participation, except what is likely a temporary and small dip in 2021, as a result of COVID restrictions (which I opposed). Haidt may be aware of this. At one point in the book he makes disparaging remarks about sports—too many rules and not kid-led—in favor of “play,” and that partly resonates: I found practice rather boring. My favorite way to play sports was in games I organized with my friends without adults present, but I also loved playing on teams and got many benefits from those teams, including friendships outside of my neighbors. In any case, the same data source from the CDC shows no decline in broader physical activity among U.S. High School students.  If there are data showing that children are playing less, Haidt does not present it. We know children are spending huge amounts of time on social media, but we know less about what they are giving up in exchange.

I am less concerned than Haidt is about declining hospital admissions, sexual intercourse, work, and alcohol consumption among U.S High School students. Some of these are obviously undesirable, including sex, from my perspective, and the case for and against adolescents taking menial low-paying unskilled jobs could be made. I had several such jobs, and it is unclear to me how it contributed to my development. It did afford me more opportunities to take girls out on dates, go to movies with friends, and buy video games, which were probably my main desires for working. I can’t say it contributed to my present career in any way whatsoever, though perhaps it was helpful—to my work as a labor economist—to have firsthand experience with a labor union, as I did while working at a grocery store (I was happy to be paid an extra dollar relative to my other job, but not happy about the strict rules governing my hours, break, and lunch schedule; I was often hungry during slow periods but had to take a break during the rush hour periods). Otherwise, the work itself was mostly tedious and dull.

What about effect sizes?

There is much to like about the book, as I’ve said, but to me the most frustrating thing is Haidt’s reluctance to offer any insight into the effect size of social media (or lack of play) on youth mental health. Yes, yes, we know the effects are “statistically significant,” but many are without being particularly large or important, and statistical significance does not tell us how important the effects are relative to other things. This is what I mean when I say the book shortchanges the real world, even while it strongly criticizes the virtual world.

What causes youth mental health? The absence of technology certainly is not the answer. On the plus side, technology makes humans a heck of a lot richer and growing up without material deprivation is very good for mental health. There may also be various benefits in terms of learning, communicating, and accessing information. Going to college now strikes me as an even more exciting prospect than it was for me given the proliferation of data-analysis software, databases, and online scholarly resources.

Haidt is admirably transparent about his data and the scholarly literature. His publicly available Google docs are outstanding gifts to the public. Yet, even though these resources provide effect sizes, Haidt never says what they are in the book. I suspect Haidt does not tell the reader about the details because they are unimpressive.

The cross-section effect of screen time or social media use on adolescent mental health is statistically significant but small by most standards. In my own work, I estimate a Cohen’s d of 0.09 for social media use effect on mental health and a larger effect (d=0.21) for social media use on negative body image. To translate these into “percent of variance explained,” you can square both terms because Cohen’s d is approximately the same value as a correlation coefficient of two standardized variables, and r^2 is the amount of variation explained. That gets you a value of 0.8% and 4.4%. Moreover, the mental health effect is fully mediated by parent-child relationships and the regulation of screen time. The body image issues are only partly mediated by these things. With respect to the mental health findings, my work is consistent with a meta-analysis by Jeff Hancock and colleagues, which found effect sizes ranging from 1% to 2% for depression and anxiety (r values of 0.10 to 0.14). Experimental evidence using RCTs puts the effect in the same ballpark.

So, Haidt’s book—which is mostly about broad mental health—is devoted (mostly) to a problem that explains <1% of the variation in youth mental health. Parenting, on the other hand, measures by practices and relationship quality—explains 36% of the variation in youth mental health. So parenting, which Haidt barely mentions, is about 45 times more important than social media.

It is important to point out—as Jean Twenge has argued on X/Twitter (including with me in a friendly exchange)—that social media could still have a large effect on the change in adolescent mental health (even if not the level), because social media use has changed rapidly from zero to 5 hours a day (on average) over such a short period (about 15 years). I would have loved to see Haidt think through estimates of what the effect size might look like on the change and the level, given available information, which is certainly limited. I do realize that he wrote the book for a popular audience and not for scholars like me, so I’d have been happy if this was merely in an appendix.

Here's my back-of-the envelop effort. Say the cross-sectional effect of social media use on suicide for 15-19 year-olds is 2% of the variation; that is 2% of a suicide rate of 15 per 100,000 (or 0.3 per 100K) in 2021. This suicide rate increased by 5 per 100,000 since 2010. If social media explains 0.3 of the additional 5 suicides, that would be 6% of the increase (0.3/5). The estimate would be 3% if we assumed a cross-section effect of 1% (which is closer to my estimate). In other words, if I were launching a new study attempting to estimate the effect of social media use on the change in youth suicides, my prior estimate (using a Bayesian approach) would be that it would explain 3% to 6% of the increase, and that is what I would expect to find.

Social media is harmful, yes, but there is much more to life, like your relationship with your parents, friends, relatives, and your financial security, cognitive ability, social skills and character, and many other things beside. The harm is more than enough to justify keeping smartphones away from students during school hours and deterring (by law or social norms) children from having social media accounts on video and photo-sharing platforms like YouTube, TikTok, Instagram, Facebook, etc, unless they are closely monitored or controlled by their parents and the accounts do not allow access to the public adult-oriented information dumps that adults are currently exposed to. Haidt has other sensible recommendations, and I like them all, but I also want parents to realize how much power they have over their child’s behavior and mental health.

One thing good parents can do is control their child’s access to technology. If their children lie to them or try to sneak smartphone use behind their back, those should be considered punishable offenses. One appropriate punishment is the removal of the smartphone. If the child’s friends are so obsessed with a social media platform that they won’t be friends with someone who is not on the platform, they are not the sort of friends you want your children to have. Good parenting is stronger than bad technology.