Youth is not what it used to be. On both sides of the Atlantic, whether it is the share of teenagers who consider themselves failures or the percentage who are hospitalised for self-harm, young people’s mental health is in freefall. The numbers here are not disputed; since 2012, levels of anxiety and depression have skyrocketed. What had been disputed is why this was happening.
Earlier this year, the partial release of the Centers for Disease Control and Prevention’s (CDC) biannual Youth Risk Behavior Survey in the US showed that most teen girls (57 per cent) now say they experience persistent sadness or hopelessness (up from 36 per cent in 2011), and 30 per cent of teen girls say that they have seriously considered suicide (up from 19 per cent in 2011). Boys are doing badly too, but their reported rates of depression and anxiety are not as high, and their increases since 2011 are smaller. The big surprise in the CDC data is that the Covid-19 pandemic didn’t really affect the overall trends, which marched on as they have since around 2012. Teens were already socially distanced by 2019, which might explain why Covid restrictions added little to their rates of mental illness, on average. (Of course, many individuals suffered greatly.)
[See also: Smartphone use is having a radical effect on young people’s mental health]
The media responded to this new data by lamenting the rise and searching for causes. Many noted that the rise pre-dated the pandemic and then pointed to their preferred explanation – everything from rising academic pressure to rising global temperature and misogyny. Some considered the role of smartphones and social media but then repeated a variation of what has become a standard saying among journalists: “Gosh, we just don’t know if it’s social media, because the evidence is all correlational and the correlations are really small.” The data-orientated writer Derek Thompson wrote a widely read essay in the Atlantic on the multiplicity of possible causes, noting that “the academic literature on social media’s harms is complicated”.
Thompson then quoted an academic studying the issue, Jeff Hancock, of Stanford University: “There’s been absolutely hundreds of [social media and mental-health] studies, almost all showing pretty small effects.”
Thompson’s scepticism – and wider scepticism in the media and society about the link between smartphones and declining adolescent mental health – was justified in 2019, but it is not justified in 2023. A lot of new work has been published since 2019, and there has been a surprising convergence among the opponents in the debate about the size of the correlation coefficient. There is now a great deal of evidence that social media is a substantial cause – not just a tiny correlate – of depression and anxiety, and therefore of related behaviours, including self-harm and suicide.
In our book, The Coddling of the American Mind, Greg Lukianoff and I tried to explain what’s happened to Gen Z. We focused on overprotection (“coddling”), but in our chapter on anxiety, we included six pages discussing the possible role of social media, drawing heavily on Jean Twenge, professor of psychology at San Diego State University and author of dozens of pioneering studies on the topic, particularly her book iGen. The evidence back in 2017, when we were writing, was mixed, so we were appropriately careful, ending the section with this: “We don’t want to create a moral panic and frighten parents into banning all devices until their kids turn 21. These are complicated issues, and much more research is needed.”
Our book was released in September 2018. Four months later, two Oxford University researchers – Amy Orben and Andrew Przybylski – published work that was hailed as the most authoritative study on the matter. “The association between adolescent well-being and digital technology use” deployed an advanced statistical technique called “specification curve analysis” (SCA) on three large data sets in which teens in the US and UK reported their “digital media use” and answered questions related to mental health.
Orben and Przybylski wrote that the average regression coefficient (measuring screen-time use to predict positive mental health) was negative but tiny, indicating a level of harmfulness so close to zero that it was roughly the same size as they found (in the same datasets) for the association of mental health with “eating potatoes” or “wearing eyeglasses”. The authors concluded that “these effects are too small to warrant policy change”.
Orben and Przybylski’s study had a profound influence on journalists and researchers. The vivid, memorable comparison to potatoes was written up in Forbes, Vox, the New York Times and Wired magazine. After the study’s publication, whenever a journalist suggested there was little or no relationship between social media and mental illness, Orben and Przybylski were usually cited.
When I read the study, I began to have doubts myself. After all, it was the largest and most impressive one ever done on the matter, and it was published by researchers who had been studying social media far longer than I had. Might Greg and I have got it wrong? Might we have been contributing to yet one more unjustified moral panic over technology?
More studies were published in 2019, yielding conclusions on both sides of the question. It was a confusing time. So I decided to compile in one document all the relevant studies I could find. I invited Jean Twenge to join me on the project since she was far more knowledgeable about the various datasets. We posted our Google document online in February 2019 and invited comments from critics, as well as the broader research community. Each section ended with a request to tell us what we had missed.
One of the first comments we received was that some researchers doubted that the mental illness epidemic was real. That led Jean and me to create a second document, “Adolescent mood disorders since 2010: A collaborative review”. Immediately, we found that there was a simple and obvious structure for the social media literature review: nearly all of the published studies fell into one of three categories: correlational, longitudinal, or experimental. We therefore structured the document around the three questions addressed by studies of those types. I’ll summarise what we’ve found about causality from those three kinds of studies.
Is there an association between social media and poor mental health outcomes?
The typical study here asked hundreds or thousands of adolescents to report how much time they spend on social media, or digital media more generally, and then report something about their mental health. Of course, correlational studies can’t prove causation, but they are a first step; they tell us what goes with what, then we can figure out which way the causal arrows go later.
The majority of studies found a positive correlation between time spent on social media and mental health problems, especially mood disorders (depression and anxiety). There were 55 studies listed in our review that found a significant correlation; 11 found no relationship, or nearly no relationship. The “winning side” is not determined by a simple count, as we explain in the “cautions and caveats” section. But the correlations are widely found and they aren’t randomly distributed.
In fact, there is a revealing pattern found across many studies and literature reviews: those that look at all screen-based activities (including television) for all children (including boys) generally find only small correlations, but as you zoom in on social media for girls the correlations rise, sometimes substantially. The general finding in these correlational studies is a dose-response relationship such as the one pictured below, from Kelly, Zilanawala, Booker and Sacker (2019), who analysed data from the large Millennium Cohort Study in the UK, which followed roughly 19,000 British children born around the year 2000 as they matured through adolescence.
Three features of this chart are common across many studies. First, the rates of mood disorders are higher for girls than boys. Second, the lines are curved: moderate users are often no worse off than non-users, but as we move into heavy use, the lines rise more quickly. Third, the dose-response effect is larger for girls. For boys, moving from two to five hours of daily smartphone use is associated with a doubling of depression rates. For girls, it’s associated with a tripling.
How can this large effect of social media use on girls be reconciled with the Orben and Przybylski study, which also examined the same UK dataset? What about the famous potatoes finding? Twenge and I argued in a published response paper in the same journal that Orben and Przybylski made six analytical choices – each one defensible – that collectively ended up reducing the statistical relationship and obscuring a more substantial association.
The first issue to note is that the potatoes comparison was what Orben and Przybylski reported for all “digital media use”, not for social media use specifically. Digital media covers all screen-based activities, including watching TV or Netflix, which routinely turn out (in correlational studies) to be less harmful than social media. In their own published report, when you zoom in on “social media”, the relationship is between two and six times larger than for “digital media”. Crucially, Orben and Przybylski combined both boys and girls, while many other studies have found that the correlations with mood disorders are larger for girls. So even if the association is weak for all children using all screens, the association is much larger if you focus just on girls using social media.
[See also: Are you mentally ill, or very unhappy? Psychiatrists can’t agree]
Twenge and I later reran Orben and Przybylski’s analysis on the same datasets (teaming up with researchers Kevin Cummins and Jimmy Lozano.) When we used Orben and Przybylski’s assumptions, we replicated their results exactly, finding a very small correlation. When we limited the analysis to social media for girls, we found relationships that were many times larger.
Despite years of heated debate, a consensus has now emerged about just how large the correlation is between social media use and mood disorders. In the SCA paper Twenge, Lozano, Cummins and I wrote, we compared the association of social media time with mental illness to other variables found in the same datasets. In that same UK dataset, mood disorders were more closely associated with social media use than with marijuana use and binge drinking, though less closely associated with sleep deprivation. I’m not saying that a day of social media use is worse for girls than a day of binge drinking. I’m simply saying that if we’re going to play the game of looking through lists of correlations, the proper comparison is not potatoes and eyeglasses; it is marijuana use and binge drinking. Amy Orben published a paper in 2020, reviewing many studies, which obtained a correlation coefficient (for boys and girls together) that was just a little smaller than ours.
Few parents would knowingly let their daughters become heavy users of anything shown to correlate with mental illness at this level. The effects might be even larger for younger teen girls, who are just beginning puberty, according to a more recent study by Orben and Przybylski. Granted, these correlations don’t prove causation, but the frequent finding that the correlations are consistently higher for social media, and higher for girls, shows that we’re not looking at random noise here. What would it take to show that social media use was causing teen girls to become depressed and anxious? At this point, social scientists generally move on from correlational studies to longitudinal studies and then to true experiments.
Does social media use at one time predict anything about mental health at another?
The second group of studies Twenge and I looked at were longitudinal ones. In these, hundreds or thousands of people are tracked over a set time period and measured repeatedly along the way. Typically, participants fill out the same survey annually, allowing researchers to measure change consistently over time. These studies have an interesting property that allows researchers to infer causality: you can look to see if an increase or decrease in some behaviour at one point in time predicts a change in other variables at the next measurement time.
So, let’s say a teen reports that she spends two hours a day on social media, on average, across a ten-week study. If in week three she suddenly reduces her time to zero, what do we expect to happen to her mood in week four? Will she be happier or sadder? If there is, on average, a change in happiness the week after people quit or reduce their social media time, then we can infer the change was caused by the change in behaviour the prior week.
What do we find in these studies? As of February 2023, we have 40 longitudinal studies in section two of our Collaborative Review Google document. Twenty-five of them (62.5 per cent) found evidence indicating causation, and 15 of them largely failed to find such evidence. Once again, you can’t just count up the studies and let the majority side win; scientific studies that fail to find an effect are sometimes harder to get published in an academic journal, because they lack an exciting outcome. But our collaborative review made it easy for us to acquaint ourselves with the range of studies and see what differentiates those that found evidence of harmful effects from those that did not.
As we read through the reports we noticed something: the studies that used a short time interval (a week or less between measurements) mostly failed to find an effect, which makes sense if social media is addictive, as much evidence suggests. Going cold turkey doesn’t make you happy, it makes you anxious and dysphoric for a few weeks. So we should not expect to find benefits to mental health in the short-interval studies.
My research assistant Zach Rausch created a table to categorise all the studies in section two as either a short interval (a week or less) or long interval (a month or more), and as finding an effect versus no effect. He found that seven studies used a week or less (five of them were daily), and only one of the seven found an effect. But 33 studies used a month or more (20 were annual) and of these, 24 found a significant effect. So a simple dose-response model in which social media is like poison (where cutting consumption on Monday makes you feel better on Tuesday) does not seem to be supported. But 73 per cent of the studies that looked for causal effects a month or more in the future found them.
Do experiments using random assignment show a causal effect of social media use on mental health?
Now we come to the gold standard in the social sciences for testing causality: the experiment. We looked at true experiments, in which participants were randomly assigned to either a treatment condition or a control condition, and then some dependent variable related to mental health was measured. Most experiments are done on university students or young adults – it’s hard to get parental consent to do experiments on minors – so we did not limit this section to studies on adolescents. There are few experiments out there (compared to correlational and longitudinal studies), so we also included some that used adults if it was clear that many or most were relatively young.
By February 2023, Twenge and I had 18 true experiments in Section 3 of our document, of which 12 (67 per cent) found evidence of a causal effect and six did not. Some of the studies randomly assigned university students or young adults to reduce their social media use for a while and then measured self-reported mental health outcomes, compared to the control group which was instructed to make no changes. One 2018 study randomly assigned students to either greatly reduce their use of social media platforms or not at all, and then measured their depressive symptoms four weeks later. The group that had limited use showed “significant reductions in loneliness and depression over three weeks” compared to those who continued their use as normal.
Some studies exposed girls and young women to time on Instagram, or to experiences designed to mimic Instagram, and then looked at the psychological after-effects. Another study randomly assigned teenage girls to be exposed either to original selfies taken from Instagram, or to selfies that were manipulated to be extra-attractive. Perhaps unsurprisingly, exposure to manipulated photos led directly to lower body image. A 2020 study randomly assigned female students to use Facebook or Instagram, or perform an emotionally neutral task (the control condition) on an iPad. The finding: “Those who used Instagram, but not Facebook, showed decreased body satisfaction, decreased positive affect, and increased negative affect.”
[See also: How to save the NHS]
Turning to the six experiments that failed to find significant effects, it is noteworthy that four involved asking participants to reduce or eliminate social media for one week or less. As we saw in the examination of longitudinal studies, going cold turkey brings immediate discomfort to addicts; the benefits only kick in after a few weeks when the brain has adapted to the loss of chronic stimulation. So if we remove all the studies that used a week or less, the final tally becomes ten experiments that found evidence of social media being harmful (80 per cent) and two that did not.
In sum, there are now many true experiments using a variety of methods to test questions such as whether reducing or eliminating exposure to social media confers benefits (it does, when continued for at least a month), or if exposing girls and women to Instagram damages their mood or body image (it does). These experiments provide direct evidence that social media – and particularly Instagram – is a cause, not just a correlate, of bad mental health, especially in teenage girls and young women.
Do “quasi-experiments”, using the arrival of Facebook or high-speed internet, show a causal effect of social media access on mental health?
The previous three questions all asked about individual-level effects: what happens to individuals who are exposed to more or less social media? But this fourth category of studies is different – and critical – for this reason: it is the only one that allows us to look at emergent network effects. These studies look at how whole communities changed when social media suddenly became much more available to that community. These studies are sometimes called “quasi-experiments” because the researchers take advantage of natural variation in the world as though it was random assignment.
One study conducted in 2022 took advantage of the fact that Facebook was originally offered only to students at a small number of colleges. As the company expanded to new colleges, did mental health change in the following year or two at those institutions, compared to colleges where students did not yet have access to Facebook? Yes, it got worse. The authors say:
“We find that the roll-out of Facebook at a college increased symptoms of poor mental health, especially depression, and led to increased utilisation of mental healthcare services. We also find that, according to the students’ reports, the decline in mental health translated into worse academic performance. Additional evidence on mechanisms suggests the results are due to Facebook fostering unfavourable social comparisons.”
We also found five studies that used a similar design applied to the roll-out of high-speed internet (HSI). It’s hard to have a phone-based childhood when data speeds are very low. So what happened in Spain, for example, as fibre-optic cables were laid and high-speed internet came to different regions at different times? The same thing, except with clearer evidence of a gendered effect. The authors analysed “the effect of access to high-speed internet on hospital discharge diagnoses of behavioural and mental health cases among adolescents”. Their conclusion:
“We find a positive and significant impact on girls but not on boys. Exploring the mechanism behind these effects, we show that HSI increases addictive internet use and significantly decreases time spent sleeping, doing homework, and socialising with family and friends. Girls again power all these effects.”
They also discovered that the arrival of high-speed internet had a particularly damaging effect on the quality of father-daughter relationships.
In sum, we found six quasi-experiments that looked at real-world outcomes in real-world settings when the arrival of Facebook or high-speed internet created large, sudden emergent network effects. All six found that when social life moves rapidly online, mental health declines, especially for girls. Not one study failed to find a harmful effect.
Social media is a major cause of mental illness in girls, not just a tiny correlate
We are now 11 years into the largest epidemic of teen mental illness on record. As the CDC’s report showed, most girls are suffering, and nearly a third have seriously considered suicide. Why is this happening, and why did it start so suddenly around 2012?
It’s not because of the global financial crisis. Why would that hit younger teenage girls hardest? Why would youthful mental illness rise steadily throughout the 2010s as the American economy steadily recovered? Why did a measure of loneliness at school go up around the world only after 2012, as the global economy got better and better? And why would the epidemic hit Canadian girls just as hard when Canada didn’t have much of a financial crisis?
It’s not because of the 9/11 attacks, wars in the Middle East, or school shootings. As Émile Durkheim showed long ago, people in Western societies don’t kill themselves because of wars or collective threats; they kill themselves when they feel isolated and alone. Also, why would American tragedies cause the epidemic to start at the same time among Canadian and British girls?
There is one giant, obvious, international and gendered cause: social media. Instagram was founded in 2010. The iPhone 4 was released then too – the first smartphone with a front-facing camera. In 2012 Facebook bought Instagram, and that’s the year that its user base exploded. By 2015, it was becoming normal for 12-year-old girls to spend hours each day taking selfies, editing selfies and posting them for friends, enemies and strangers to comment on, while also spending hours each day scrolling through photos of other girls and wealthy female celebrities with (seemingly) superior bodies and lives. The hours girls spent each day on Instagram were taken from sleep, exercise, and time with friends and family. The arrival of smartphones rewired social life for an entire generation. What did we think would happen to them?
Facebook knows what is happening: in one slide from an internal presentation on Instagram’s mental health effects, the presenter notes that “parents can’t understand and don’t know how to help”. The slide explains: “Today’s parents came of age in a time before smartphones and social media, but social media has fundamentally changed the landscape of adolescence.”
The collaborative review document that Jean Twenge, Zach Rausch and I have put together collects more than 100 correlational, longitudinal and experimental studies, on both sides of the question. Taken as a whole, it shows strong and clear evidence of causation, not just correlation. There are surely other contributing causes, but the collaborative review document points strongly to this conclusion: social media is a major cause of the mental illness epidemic in teen girls. It is now time for parents, heads of schools and governments to act.
This essay is adapted from a post at Jonathan Haidt’s “After Babel” Substack
[See also: The great attention deficit: what’s fuelling the rise in adult ADHD?]
Content from our partners
Related
This article appears in the 29 Mar 2023 issue of the New Statesman, Easter Special