There’s been a torrent of recent articles about the massive increase in depression among young Americans, particularly girls, and the possibility that social media is to blame. Here’s a small sample of work from just the last few weeks.
Eric Levitz, “No, Teen Suicide Isn’t Rising Because Life Got Objectively Worse.” (New York Magazine)
Michelle Goldberg, “Don’t Let Politics Cloud Your View of What’s Going on With Teens and Depression.” (NYT)
Derek Thompson, “America’s Teenage Girls Are Not Okay.” (The Atlantic)
Jon Haidt, “Social Media is a Major Cause of the Mental Illness Epidemic in Teen Girls. Here’s the Evidence.”
For this essay, we can call this the “social media hypothesis.” My initial inclination was to treat this theory with skepticism. I tend to dislike moral panics that are used to justify government intervention in people’s personal choices or market forces, particularly of the “think of the children” variety. Moreover, I think there’s a tendency to scapegoat big tech, which I consider an unhealthy impulse that is rooted in both Luddism and anti-market bias, both of which I strongly oppose. In my interviews with Tyler and Marc Andreessen, each of them made the case against the anti-tech position, and as I find their politics congenial on most issues, they were able to convince me that perhaps the emerging conventional wisdom was wrong. A final reason I didn’t want to believe in the social media hypothesis was that I’d much rather blame wokeness, and the timing of the mental health crisis takeoff roughly matches the beginning of the Great Awokening. It’s annoying to me that intellectuals won’t even consider the possibility that radical new ideas about race and sex might be to blame, and the fact that the most liberal demographics, that is women and LGBT adolescents, are having the most mental problems indicates that there is something to the theory that wokeness causes misery.
In sum, nearly every ideological and political commitment I hold inclined me to be wary of jumping on this bandwagon. I therefore set out to do a deep dive into the most relevant research, convinced that I would probably find shoddy data and exaggerated claims about the negative effects of social media.
After looking at various kinds of evidence, however, I have changed my mind. This essay sets out to explain why I think that the rise of social media has had disastrous effects on the mental health of young people. First, randomized control trials show that quitting or cutting back on Facebook is good for your mental health. It’s true that some studies show a null or even opposite effect, but, as I explain, the studies supporting the hypothesis that social media causes misery tend to be larger and more convincing. Second, I looked to see whether the increase in teen depression since around 2010 can be found in other developed countries. The answer is mostly yes, and some of this data is extremely impressive in that much of it comes from sources that weren’t setting out to prove anything about the social media hypothesis, but found data that supported it anyway. Finding similar trends across the developed world makes it much less likely that something specific to the US like the rise of wokeness can be blamed for teen misery. Finally, I discuss the relevance of research on how covid affected mental health, how kids increasingly use their time, and what makes people happy.
Jonathan Haidt and Jean Twenge have put together a Google doc that contains all studies they can find relevant to whether social media use harms the mental health of young people. Most of them I don’t find very convincing. Correlational studies show that kids who use a lot of social media tend to be depressed. Everyone knows to say that “correlation is not causation,” but they tend to forget the adage when the correlational evidence supports a hypothesis they’re committed to.
Another kind of study engages in “quasi-experimental” analysis. For example, one paper looked at when Facebook was introduced at various colleges and finds that mental health got worse at places where it did. Another shows that areas in the UK that had faster broadband ended up with children with worse mental health. The problem here is that there isn’t true randomization; colleges that get Facebook earlier or neighborhoods that have faster broadband might differ from other colleges and neighborhoods in ten thousand large and subtle ways. Sure, social scientists try to “control for” various factors, and this can look very convincing to some people, but I see no reason to trust that researchers have good data on what to control for or even know what variables to include if they did.
I’m also not convinced by research on teen suicide. While people kill themselves due to unhappiness, only a tiny percentage of people ever take that step. It’s completely possible for teens to become unhappier in the aggregate while suicide goes down, or vice versa. So like the correlational and quasi-experimental studies, I’m going to set the suicide data aside.
The gold standard, of course, is randomized control trials. A list of them can be found under the heading “Question 3” in the Haidt and Twenge document. Conveniently, the authors break papers down into 14 studies that support the social media hypothesis and 6 that don’t. One always has to worry about the file drawer effect, in which studies that support a hypothesis tend to be the ones that get published. That being said, it is telling that the larger studies tend to be the ones to find an effect, while the smaller ones do not. The table below lists the randomized control studies highlighted by Haidt and Twenge under “Question 3,” excluding six of them because they did not involve asking people to reduce or give up on social media, and counting Sagioglou & Greitemeyer 2014 twice because it included two separate studies. I’m also adding an extra study that Haidt and Twenge put in the appendix (Tromholt 2016), as I think it’s relevant to the question.
.
As you can see, the studies supporting the social media hypothesis have an average of 572 participants. The studies not supporting it have an average of 197, just over a third as many. In fact, one study supporting the social media hypothesis alone (Allcott et al. 2020) has more participants than all the null or negative studies put together, and a second positive one (Tromholt 2016) only has about 100 fewer. This analysis seems to indicate that the negative studies are simply underpowered.
Allcott et al. is not only the largest study, but it was also pre-registered and easily the most impressive. Participants, about half of them under 30 years old, were paid to stay off Facebook for a month. Here are the main findings for the outcomes we care about.
The effect sizes for life satisfaction, anxiety, depression, and happiness all hover around a tenth of a standard deviation. Tyler argues that such effect sizes are small, and critiques Haidt for saying we should assume network effects. Haidt’s argument is that quitting Facebook shouldn’t be expected to do all that much if all your friends are still on it and miserable. Personally, I don’t find the effect size to be that small. As Allcott et al. point out, the result for subjective well-being is about equivalent to a $30,000 increase in income.
A tenth of a standard deviation seems small, but this is just one intervention that lasts a month. Other than getting rid of Facebook, you still have the same genes, had the same childhood, have the same job, grew up in the same culture, etc. If cutting out one app can alone take you from the 50th to 55th percentile of happiness, it’s definitely worth doing. And network effects aren’t some mysterious social force we should be skeptical of – the mental well-being of individuals rises and falls along with that of others in a society, which is why we can observe such striking trends over time. It seems to me very plausible that an intervention that only increases happiness by a small amount at the individual level might have massive effects if the same change was adopted by society as a whole. To expect much more than a tenth of standard deviation from one short intervention may be asking too much.
There are two more interesting things about the Allcott study, both of which make the results more impressive. First of all, those in the treatment condition reported reducing their Facebook use after the study was finished.
Deactivation clearly reduced post-experiment demand for Facebook. These effects are very stark, with by far the largest magnitude of any of our main findings. The effect on reported intentions to use Facebook as of the endline survey is a reduction of 0.78 standard deviations: while the average Control group participant planned to reduce future Facebook use by 22 percent, deactivation caused the Treatment group to plan to reduce Facebook use by an additional 21 percent relative to Control. In our post-endline survey a month after the experiment ended, we measured whether people actually followed through on these intentions, by asking people how much time they had spent on the Facebook mobile app on the average day in the past week. Deactivation reduces this post-endline Facebook mobile app use by 12 minutes per day, or 0.31 standard deviations. This is a 23 percent reduction relative to the Control group mean of 53 minutes per day, lining up almost exactly with the planned reductions reported at endline. However, online Appendix Table A13 shows that the reduction is less than half as large (8 percent of the Control group mean) and not statistically significant (with a t-statistic of −1.16) if we limit the sample to iPhone users who reported their usage as recorded by their Settings app, thereby excluding participants who were reporting personal estimates of their usage.
Moreover, 9 weeks after the end of the experiment, 5% of the treatment group still had their Facebook account deactivated, compared to 2.5% of the control group, indicating that people saw the benefits of scaling back use or completely giving up the site.
Going into this literature, I suspected that results could be biased by the fact that the experiments aren’t blind. In a drug trial, you don’t want participants knowing whether they got the placebo or the real medicine. Here that’s impossible, since people know whether they quit Facebook or not. So perhaps individuals only report better outcomes after quitting social media because that’s what they think they’re supposed to say, or for self-fulfilling prophecy reasons.
Thankfully, the authors thought of that. At the end of the study, they asked respondents what they thought the researchers were trying to prove. About 62% said they didn’t think they had an agenda or that they weren’t sure, while just over a third said that the point was to show Facebook was bad for people, with 3% thinking the researchers wanted to show that Facebook was good. Here are the treatment effects depending on which category people fell into. There’s no evidence of a statistically significant impact of participant expectations.
It’s impressive that even those who did not think that the researchers had an anti-Facebook agenda were half a standard deviation lower on the “post-experiment use index,” which included factors related to planned and actual reduction in use. Arguably, what people chose to do after the experiment was over is more impressive than what they said about how depressed or anxious they felt.
The second massive study was Tromholt 2016. It had 1,095 Danish participants who were asked to stay off Facebook for a week. From Table 1, I estimate quitting Facebook provides a 0.28 increase in life satisfaction and 0.19 standard deviation increase in emotional well-being, with the effects being higher for those who used Facebook more often. Haidt calls this a low-quality experiment because it didn’t monitor compliance, but the author argues that the fact that he included those who didn’t comply in the analysis should make the results even more impressive since it introduces noise.
Finally, it’s worth taking a look at the largest study that did not find support for the social media hypothesis, which was Przybylski et al. 2021. The experiment lasted two days. One group of undergraduates was allowed to use Facebook on day 1 and told to abstain on day 2, while the other group had the days switched around. There were 600 participants total, evenly divided across the US, the UK, and Hong Kong. Only about half complied with the conditions set and so were included in the analysis.
In all three locations, respondents felt less “socially related” on the day they abstained from Facebook. Those in the UK and HK also had more negative emotions and less satisfaction with their day.
Przybylski et al. 2021 can perhaps be reconciled with the other major studies by the length of the treatment. The Allcott experiment went for a month, and Tromholt for a week. If social media is addictive, cutting people off for a day might be like taking heroin users, asking them to stop cold turkey, and then coming back 24 hours later and determining that going sober would be a huge mistake.
In sum, there is a clear trend of larger randomized studies being more likely to support the social media hypothesis. There are a handful finding null results, but some of them are so underpowered that one can doubt whether they should have been expected to find an effect. And by far the largest study not supporting the social media hypothesis lasted only a single day, telling us nothing about the longer-term impacts of cutting back.
While studies show “mixed results,” with 10 experiments supporting the social media hypothesis and 6 contradicting it or finding null results, those numbers make the two sides of the debate appear much more evenly matched than they actually are.
If smart phones are making teens miserable, then we should see a pattern across developed countries. In contrast, if the trend starting around 2010 is due to some peculiar quirk of American culture, we would only find it in the US, and perhaps in other Anglophone countries subject to our influence. For that reason, I tried to find any data I could on trends in depression, anxiety, and other mental health problems for teens in other countries.
One thing that might be worth worrying about here is the possibility that papers and reports are only likely to be written when there’s a trend to report. For example, let’s say that in France there are five different depression surveys that were conducted in 2010 and 2019, and 4 of the 5 find no trend, while the fifth does. It’s completely possible that an academic writes a paper on the fifth study, while the other four get ignored and those of us investigating the social media hypothesis never hear of them.
One should keep this possibility in mind throughout this section. A good way to test the social media theory in other countries would be to track down large datasets that haven’t been publicly used yet and see what they say. For now, I’ve been looking for the best data I can find from across the developed world. Some of the results come from government sources, where a file drawer effect might be less likely, although not impossible. Other papers and reports below are based on studies conducted during covid-19, which incidentally show a decline in mental well-being in the years leading up to the pandemic. I’ve generally excluded numbers from the first year or so of covid, and looked for data that shows trends from circa 2005-2012 up to say 2015-2019, or to 2021 or later, although I sometimes include studies motivated by research into the mental health effects of covid if they cover the correct time periods. These are particularly credible, since the researcher goes into his project interested in what happened in 2020, and only incidentally finds an earlier rise in a negative outcome. Probably the least credible sources are put out by activist groups, since they have a clear agenda and thrive on publicity surrounding their pet issue, but I’ve included the few that I found regardless.
I’ve looked for survey data that asks the same question over time, related to depression, emotional well-being, or anxiety. I’ve ignored diagnoses of mental health disorders or trends on things like the use of SSRIs, since it’s possible for conditions to be diagnosed more often without an underlying change in the degree to which a population suffers from a disorder. Keep in mind that if I included this kind of data, the case for the social media hypothesis would appear even stronger. For example, one paper from Germany shows a large increase in depression from 2009 to 2017 based on insurance data. This has the benefit of being comprehensive, including everyone with statutory health insurance, which is 87% of the population. The biggest increase is for 15-19 year-olds, and then 20-24 year-olds. Since I wasn’t able to find survey data from Germany, however, I’m not including that paper in this review, and this analysis therefore remain agnostic about whether teen depression is actually on the rise in that country.
With those caveats and qualifications in mind, here’s what I’ve been able to find.
Mission Australia: “Psychological distress in young people in Australia”: Reports that the number of 15-19 year-old females in Australia experiencing psychological distress between 2012 and 2020 went from 22.4% to 33.5%. For boys, the numbers were 12.6% and 16.8%.
Australian Institute of Health and Welfare, “Australia’s youth.”: We see a rise in mental health problems among young women from 2011 to 2015, followed by a stabilization over the next few years.
Among young people aged 18–24, the rate of high or very high psychological distress increased from 12% to 15% between 2011–12 and 2014–15. However, there was little change between 2014–15 and 2017–18.
For females aged 18–24, the rate of high or very high psychological distress also rose between 2011–12 and 2014–15, from 13% to 20%. However, there was little change between 2014–15 and 2017–18.
Didier Garriguet, “Portrait of youth in Canada: Data report.”: Finds a massive decline in metal health for both sexes starting around 2010, being particularly pronounced among girls.
Direction de la recherche, des études, de l’évaluation et des statistiques, “Crise sanitaire: hausse des syndromes dépressifs et des consultations pour ce motif”: The French Ministry of Health reports about a doubling of those having depressive symptoms for 15-29 year-olds from 2014 to 2019, from 4.2% to 10.1%, and then another doubling during covid, up to 22%. Interestingly, as of 2019, there wasn’t much of a gap between men and women, which is different from most other countries.
East Hunter, “No social media driven depression epidemic in Hungary.”: This is a researcher on Substack who looked at Hungarostudy, which involves population-representative surveys and has had four waves, in 2002, 2006, 2013, and 2021. Each survey talks to 5,000-10,000 people. He finds a slight decrease in depression for those under 24 years-old between 2006 and 2021 and also an increase in well-being among the same age group according to the World Health Organization.
“My World Survey 2”: Based on a large dataset funded by the Irish government, between 2012 and 2019, the number of adolescents with severe or very severe depression went from 8% to 15% (p. 117-18, Figures 8.1 and 8.2). The numbers for severe or very severe anxiety increased from 11% to 22%.
Simone Amendola, “Burden of mental health and substance use disorders among Italian young people aged 10–24 years: results from the Global Burden of Disease 2019 Study”: An analysis of the Global Burden of Disease (GBF) finds not much going on in Italy.
If you squint you can see an increase in depression among girls starting around 2010, but it’s slight. That being said, there are some massive confidence intervals here. And so I went to the GBD website, and couldn’t really make sense of how to use it. I couldn’t figure out how to search for depression in the US for girls 10-24 years old, instead it gives me a number called “Years Lived with Disabilities” for depression, and it tells me there’s been no change over that time period, which seems hardly believable. So this seems like a very strange and perhaps worthless data source.
How does GBD work? According to this paper.
GBD estimates incidence, prevalence, mortality, years of life lost (YLLs), years lived with disability (YLDs), and disability-adjusted life-years (DALYs) due to 369 diseases and injuries, for two sexes, and for 204 countries and territories. Input data were extracted from censuses, household surveys, civil registration and vital statistics, disease registries, health service use, air pollution monitors, satellite imaging, disease notifications, and other sources.
So they look at “satellite imaging” and “air pollution monitors” and from that try to estimate things like depression, but not even depression, but as far as I can tell, “years lived with disability”? Anyway, this seems like a completely useless source to me, but I’m describing my experiences with it here just so others can look into the GBD (data here) and see if they can benefit from it.
M.E. De Looze, “Trends over time in adolescent emotional wellbeing in the Netherlands, 2005-2017.”: A large study of almost 22,000 adolescents sees a slight decrease in emotional well-being between 2009 and 2013, but no change between 2013 and 2017. Psychosomatic symptoms, which we are told include things like headaches and “feeling low,” similarly increased between 2005 and 2013, but stabilized by 2017.
Peter G. van der Velden et al., “Mental health problems among Dutch adolescents of the general population before and 9 months after the COVID-19 outbreak.”: Table 3 shows no significant change among 16-20 year-olds in anxiety-depression symptoms or use of mental health services between 2012 and 2016. The 2012 cohort had 175 members, and 2016 had 134, so these are very small sample sizes.
Aftenporten, “Ungdom opplever at hovedårsaken til de psykiske helseplagene er skole.”: We see a major spike in depression among 14-17 year-olds in Norway between 2006 and 2015, being particularly pronounced in girls. This appears to be very high-quality data, with over 24,000 respondents. Statista has an English-language figure showing the results.
Thomas Potrebny et al., “Health complaints among adolescents in Norway: A twenty-year perspective on trends.”: Finds an increase in psychological and psychosomatic health complaints between 1994 and 2010, but small changes between 2010 and 2014. This was based on a World Health Organization survey, and not specifically about depression or anxiety, so I don’t know how relevant it is here, but I’m including the paper in case someone wants to dive more into it.
Statista, “Share of students who experienced depression within the last 12 months in South Korea from 2011 to 2021”: The link to the original source is broken, so I’m directly using Statisa here. In what looks like a large government database, we see a decline in depression among South Korean middle and high school students between 2011 and 2021.
Depression actually decreases during covid here, and I don’t know what to make of that, since it contradicts most of what we see in the rest of the world. Is it possible that East Asians have been so enthusiastic about masking and social distancing because they find it less stressful to cut off contact with other people? We don’t have enough data to say that, but it’s an intriguing possibility. This would also be an explanation for why the iPhone hasn’t increased depression among young people in South Korea.
Folkhälsomyndigheten, “Ängslan, oro eller ångest.”: Data from the Public Health Agency in Sweden finds that the number of 16-29 year-olds with anxiety went from 34% in 2010 to 52% in 2018.
For the population as a whole, male anxiety went from 25% to 32%, and for women it was 38% to 46%, although the report does not give us the data for young people broken up by sex.
Office for National Statistics, “Young people’s well-being in the UK: 2020”: This reports increasing depression or anxiety among women 16-24, up to 31% in 2017-2018, from 26% five years earlier. I infer from the report that there was no statistically significant change among boys, otherwise it would have been reported.
Office for National Statistics, “Mental Health of Children and Young People in England 2017” (link to source here): For 11-15 year-olds in England between 2004 and 2017, we see an increase in emotional disorders among girls, from 6.6% to 9.4%, and depressive disorders from 1.8% to 3.6%. Boys see a similar percentage increase in emotional disorders, but are mostly stable otherwise.
The Prince’s Trust Tesco Youth Index 2021 (data visualization here): Between 2009 and 2019, the percentage of young people feeling always or often anxious went from 37% to 56%, and the percentage feeling often or always depressed went from 26% to 38%.
Overall, we find strong evidence of declines in teenage mental health over the last decade or so based on government data with large sample sizes not only in the US, but also Australia, Canada, France, Ireland, Sweden, Norway, and the UK. Only in Hungary, the Netherlands, and South Korea do we see the absence of such trends. Overall, this makes an extremely strong case that the problem goes way beyond the United States, or even English-speaking countries. That being said, there does seem to be a pattern in which countries that are culturally closer to the US tend to see trends that are worse than those that are more distant.
Across the world, we saw massive increases in mental distress throughout covid.
A simple theory explains why. Humans need contact with others, and when deprived of that contact they tend to start malfunctioning. The theory that smart phones caused increasing depression is based on the exact same idea.
In her book iGen, Twenge shows that kids are spending more time on social media, and the opportunity cost of doing so is less socialization. Here’s a figure showing that getting together with friends fell off a cliff beginning around 2000. (see this thread for other interesting charts from the book)
The trend starts as soon as the internet came along, but really took off with the rise of smart phones and social media. It may be that social media isn’t even that bad, the problem could simply be that it reduces socialization and human contact. This would make it like drug and alcohol addiction, in that much of the harm comes from what it takes away from other areas of life.
If that’s not enough for you, we have good data showing that people’s highest levels of subjective happiness tend to be when they’re either outside or spending time with romantic partners or friends.
The social media hypothesis is convincing because we have a simple, intuitive story backed up by various lines of evidence that all point to the same conclusion. Human beings evolved to gain pleasure from spending time with others. When young people in their developmental years were given addictive devices that stopped them from doing that, they started getting depressed and anxious. In 2020, governments decided to take away much of the remaining social interaction young people had left, and their mental well-being plummeted even further. This exact same pattern — iPhones preceeding increases in mental distress, and covid-19 preceeding an even larger deterioration — is found across most developed countries. We have a straightforward theory that fits the vast majority of the data. It is supported not only by correlational data linking higher levels of social media use to poor mental outcomes, but also a preponderance of the evidence in randomized control trials that ask people to reduce or eliminate social media use.
One can debate country-specific differences that might explain secondary questions like why the gender gap is different across nations, or why young people in the Netherlands seem to be unique in having a relatively mild decrease in mental well-being between 2005 and 2017, compared to larger changes in places like the US and France. But I think the big picture here is clear. Smart phones and social media have had a negative impact on young people, particularly girls.
I still don’t think government intervention is necessarily the answer to the problem. Often, when the public health community speaks with one voice, it can make parents change their behavior. Since the 1980s, for example, there has been a massive increase in breastfeeding, as parents have been told that it has health benefits for the child and even increases IQ. This is based on junk science, since more educated parents tend to listen to what doctors and health experts tell them to do, and it’s far from certain whether breastfeeding actually causes babies to end up better off in the long run. But that doesn’t matter — the point is that breastfeeding is inconvenient and often hurts pretty badly, so if expert advice alone can get women to switch to the practice, then they can probably be convinced to wait a little longer to buy their kids an iPhone. Parents are still best positioned to know the circumstances of each child and understand what they are able to handle once families are given the proper information.
Although I find the case for the social media hypothesis quite convincing, it would be nice to know more, particularly about countries I didn’t find any data for. If someone wants to take the international data from this essay and put it into a Google Doc, or if Haidt and Twenge want to add the studies I review here to their own document, be my guest. Hopefully others would be able to add to what I’ve found.
We could of course also use more randomized control trials on cutting back or eliminating social media use. In an ideal world, entire communities, schools, or even states could consider trying more coercive measures to keep kids away from social media and seeing what happens. In China, the national government sometimes will roll out a policy in some regions but not others so it can learn from the results of its experiment. The US federal government doesn’t have similar power, but perhaps local entities up to and including states can find ways to experiment on their own.
Leave A Comment