Trends in US youth suicide by gender, 1999–2020
Do suicide rates fit the hypothesis that smartphones and social media are making teenagers (and especially female teenagers) miserable?
Via the contrarian-yet-not-stupid-about-it AliceFromQueens (once fleetingly on Substack), I bumped, again, into the long-running debate about whether smartphones and social media are making American teens sadder and more mentally ill.
Ted Gioia posted a line chart, reportedly based on data summarized by Jean Twenge, of “Depressive symptoms in US 8th, 10th and 12th graders”. Lest lurkers missed the point, Gioia asked “What happened in 2012?” next to the chart, which showed 3 types of depressive self-reports trending upward in US teenagers from 2008 onward. AliceFromQueens reacted by linking “Matt” linking a David Wallace-Wells piece warning that the upward trends were likely affected by ascertainment/sampling bias:
In 2011, as part of the rollout of the Affordable Care Act, the Department of Health and Human Services issued a new set of guidelines that recommended that teenage girls should be screened annually for depression by their primary care physicians and that same year required that insurance providers cover such screenings in full. In 2015, H.H.S. finally mandated a coding change, proposed by the World Health Organization almost two decades before, that required hospitals to record whether an injury was self-inflicted or accidental — and which seemingly overnight nearly doubled rates for self-harm across all demographic groups. Soon thereafter, the coding of suicidal ideation was also updated.
In short, hospital-derived and diagnosis-based measures of teenage unhappiness have likely been inflated since 2011 by better screening and record-making.
I thought of a data source which, with a dollop of luck, might not be affected by those changes: rates of suicide, as tallied from death certificates. As far as I know the US hasn’t significantly changed how death certificates record suicide in recent years, and one dissertation reckons that death certificates “from 2003 through 2017” were exceptionally accurate (“99.57%”) in identifying suicide, leaving little room for coding changes to seriously distort the numbers.
Online CDC system WONDER (“Wide-ranging ONline Data for Epidemiologic Research”) allows the public to count up deaths according to “Underlying Cause” and other variables, including gender, age, state or county, and month for the period 1999 through 2020. (The catch: if one slices the data too finely, WONDER hides the exact numbers for the smallest slices to avoid breaching privacy.) I downloaded counts and crude rates of suicide1 by gender, 5-year age bracket, and year for the entire US.2
Suicide rates in US teenagers by gender, 1999–2020
To get started, I plot suicide rates among 10–14-year-old boys (represented in blue) and girls (represented in red), the youngest age bracket in which, tragically, there are enough suicides per year that WONDER discloses their numbers without hiding any: about 2 per 100,000 of the girls per year and about 4 per 100,000 of the boys per year.
There is indeed an upward trend in suicide from 2012. To the extent that suicide is a reliable index of misery, 10–14-year-olds in the US became more miserable since 2012. As for the hypothesis that smartphones and social media are to blame for that, the hypothesis is dubious if it insists on pinning the timing to the year 2012:
in both boys and girls the suicide rate began picking up in 2008, not 2012, and
strangely abrupt single-year leaps in suicide came after 2012, namely in 2013 for girls and in 2017 for boys.
Indeed, on that last point, one team of researchers posited that a single TV show, 13 Reasons Why, might explain “a significant increase in monthly suicide rates among US youth aged 10 to 17 years”. I am rather skeptical — for one thing, the researchers found the show’s release to be associated with increased suicide among boys, not girls, despite the show’s viewers skewing female — but it illustrates the difficulty of shoring up a causal hypothesis just by pointing at time series. Why was there a 66% year-on-year leap in suicides among the girls, and a 30% year-on-year leap (from a higher baseline) among the boys, but 4 years apart? It’s not because the underlying numbers of suicides are minuscule and overwhelmed by statistical noise; the jump among girls represented an absolute increase from 85 suicides to 141, and the jump among boys an absolute increase from 265 to 348.
10–14-year-olds are only half of the story, so here are the rates for ages 15–19.
The male spike in 2017 persists in this age group, but the female spike in 2013 is gone. The anomaly of both male and female trends starting in 2008, not 2012, persists.
Cross-gender comparisons
Gioia presented the increase in depression as a phenomenon afflicting youth regardless of gender, not just in his X/Twitter xeet, but in his Substack postletter “Crisis in the Culture: An Update”.3 However, the originators of the blame-smartphones-and-social-media hypothesis, Jean Twenge and Jonathan Haidt, put a gendered spin on it: the trends are worse for girls and teenage women than for boys and teenage men (see also Haidt asking “What could have changed right around 2012 that hit tween and young teen girls hardest?”).
The suicide data don’t necessarily sit well with the gendered hypothesis. The year one takes as the starting point makes a big difference. Return to the 10–14-year-olds’ data. From 2012 to 2020, the girls’ suicide rate increased by 139% and the boys’ by only(!) 70%. That’s a pretty clear gender difference. But take 2009 as the start year instead of 2012: then the girls’ suicide increase is 130% and the boys’ 120%, a much more trifling gender difference; is it really big enough to warrant a fiddly, complicated hypothesis about “particularly Instagram” “and TikTok” hitting girls especially hard?
The gendered hypothesis arguably does even worse with the suicide data for 15–19-year-olds. In that age bracket, the male suicide rate went up by 26% from 2012 to 2020, and the female suicide rate by 30% — equal to within statistical error. In fact, if I fit a rigorous model4 to all of the post-2011 data instead of just taking ratios of the rates for 2020 and 2012, I get a marginally faster increase in the male suicide rate (4.1% per year) than the female (3.7% per year).
It may not be a coincidence that recently Haidt seems to start the clock not in 2012 but 2010, or flickers between using both years, which restores the finding of faster increases in suicide among female teens (though that gender difference, from what I can see, remains statistically insignificant).
Comparisons to adults aged 20–34
With suicide rates by age readily available, I find it natural to compare growth in teenage suicide to growth in adult suicide. After all, that could be informative about the blame-smartphones hypothesis, which posits that smartphones and social media particularly harm teenagers. If suicide turns out to have increased about as much among teenagers as among older adults, that would hurt the credibility of teenager-oriented hypotheses, including the blame-smartphones hypothesis.
So, here are male suicide rates for teenagers and now adults aged 20–24, 25–29, and 30–34. I switch to using a log scale so that the lowest line (for 10–14-year-olds) isn’t crushed next to zero.
As it happens, the increase in suicide among 10–14-year-old boys is plainly faster than for the older age groups (the youngest boys’ line is visibly steeper than the other lines after 2007). On the other hand, the increase among 15–19-year-olds looks similar to the increases among older adults. Being more rigorous,5 and taking 2010 as the start year (à la Haidt), the annual growth rates of the suicide rate are 3.8% for ages 15–19, 3.2% for ages 20–24, 2.4% for ages 25–29, and 3.2% for ages 30–34. Suicide is up among older teenagers, but most of that increase is accounted for by a broader-based increase that also occurred among adults aged 20–34. Changing the start year to 2012 doesn’t change that basic result: that shifts the annual growth rates to 4.1% for ages 15–19, 3.9% for 20–24, 2.6% for 25–29, and 3.7% for 30–34. The growth in the suicide rate of male older teenagers is only modestly higher than that of men aged 20–24 or 30–34.
Here’s another log-scale plot of suicide rates, for the same age brackets, this time for girls and women.
As for boys, the increase in suicide among 10–14-year-old girls is much steeper than for older girls and women. And again, as for boys, older age brackets have more-similar growth rates. Taking 2010 as the start year, the annual growth rates are 4.9% for ages 15–19, 3.7% for 20–24, 2.9% for 25–29, and 2.4% for 30–34. Alternatively, starting at 2012, the annual growth rates are 3.7%, 3.6%, 2.9%, and 1.9% respectively. As for boys, most of the growth in suicides at ages 15–19 is of a piece with the growth in suicides among adults aged 20–34.
Youth suicides are up, but why?
US teenagers have been dying by suicide at increasing rates since about 2007, regardless of gender, and the increase has itself been faster among younger teenagers (10–14-year-olds). Some commentators hypothesize that those increases started in 2012 or 2010 due to teenagers gaining wider access to smartphones and social media, especially female teenagers.
When I do the ground-clearing exercise of simply looking at suicide rates among teenagers and young adults over time, I notice features of the data that fit the smartphones-and-social-media hypothesis less than perfectly:
there are strangely abrupt spikes in suicide rates that happened in different years for different genders;
most of the increases in teenage suicide began before the 2010s;
suicide increased comparably quickly in male and in female 10–19-year-olds, leaving less basis for a gendered hypothesis; and
suicides have increased in 20–34-year-old adults too, albeit less than in teenagers.
The suicide data convince me that more US teenagers are experiencing the worst nadirs of mental distress, but the causes of that increase remain unconfirmed. Teenagers having greater access to smartphones and social media is likely a contributing factor, but my suspicion is that it’s a minor contributor rather than a major one.
Update, 10 November: corrected “The suicide rate of male older teenagers is only modestly higher” to “The growth in the suicide rate of male older teenagers is only modestly higher”. Older male teenagers did not have higher suicide rates than men aged 20–34, that was an error where I meant to write that they showed higher increases in suicide (but that higher increase was from a lower base).
WONDER represents underlying causes of death as ICD-10 codes, so to count suicides I counted deaths with ICD-10 codes of “Y87.0” (“Sequelae of intentional self-harm”), “X60-X84” (“Intentional self-harm”), and “U03” (“Terrorism Intentional (Suicide)”), just the same as the CDC does it.
I wanted to slice further to month and single years of age, but that would’ve made the slices too thin.
The postletter includes Financial Times charts showing rising self-reports of depressive symptoms in both US and UK teens. Notably, while the post-2010 trends are upward in UK teens too, the UK trends, which are presumably unaffected by the recent US sampling biases, are visibly shallower than for US teens. I also note that from June 2010 until last month the UK elected only Conservative governments, governments infamous for imposing austerity.
A quasi-Poisson generalized linear model, with a log link function, of suicide rate regressed against gender and gender interacted with year, if you wanna get technical. This amounts to fitting one exponential-growth curve to the male time series and another exponential-growth curve to the female time series.
That is, fitting the same basic kind of model as in the previous endnote, but substituting age bracket for gender.