↖︎ Vishal Singh
Youth Risk Behavior Survey · 2009–2023

Scars That Don’t Show

One in four American high schoolers is bullied — in the hallway, on the phone, or both. The bruises are rare. The despair is not: bullied teens report sadness, suicidal thinking, and suicide attempts at rates two to six times higher than their classmates, and the students hit from both directions fare worst of all.

23.5%
of high schoolers were bullied in the past year — at school, online, or both
2021–23 POOLED, WEIGHTED
76%
of teens bullied both at school and online felt sad or hopeless, vs 33% of the non-bullied
PAST-YEAR PERSISTENT SADNESS
30%
of doubly-bullied teens attempted suicide in the past year — vs 5.8% of the non-bullied
SELF-REPORTED ATTEMPT
6.8×
adjusted odds of a suicide attempt for teens bullied both ways, after demographic controls
AOR, 95% CI 6.0–7.7
01 · The exposure

Bullying never went away — it just found a second venue

Every two years since 1991, the CDC has asked a national cross-section of high school students about their lives. Since 2009 it has asked whether they were bullied on school property, and since 2011 whether they were electronically bullied — through texting, Instagram, or whatever platform arrived that year. The striking thing about both series is how little they move. School bullying has hovered near one in five students for fifteen years (dipping only in 2021, when many students simply weren’t in the building), and cyberbullying has sat near one in six since the question was first asked.

The stability hides a sharply unequal burden. In 2023, 21.9% of girls were bullied at school against 16.6% of boys, and the online gap is wider still: 20.7% of girls were electronically bullied against 12.0% of boys. Toggle the chart to see the split.

Bullied at school vs. bullied online, 2009–2023
Bullied on school property Electronically bullied 95% confidence band
FIG 1 Percent of U.S. high school students bullied in the 12 months before the survey. School bullying asked since 2009, electronic bullying since 2011. Shaded bands are 95% CIs from Taylor-linearized design-based standard errors. Source: YRBS national files, survey-weighted.
02 · The gradient

Two kinds of bullying, one compounding injury

Pooling the two most recent waves (2021 and 2023), students sort into four groups: the 76.5% who were not bullied, 7.5% bullied at school only, 6.2% bullied online only, and 9.8% bullied in both settings. Each mental-health outcome climbs the same staircase. Persistent sadness rises from 33% among the non-bullied to 56% (school only), 64% (online only), and 76% for both. Seriously considering suicide: 15% → 32% → 36% → 53%. Actually attempting it: 5.8% → 13.3% → 19.1% → 30.2%.

Note the ordering: online-only bullying is consistently at least as damaging as school-only bullying, despite involving no physical presence at all. A phone follows a teenager home; the school bus stops at the corner.

Mental-health outcomes by bullying exposure, 2021–23
Hover a bar for the estimate, CI, and sample size
Not bullied School only Online only Both
FIG 2 Past-12-month outcomes among U.S. high school students, by bullying exposure. Pooled 2021 & 2023 waves, n = 36,446 with both bullying items answered. Error bars are 95% CIs. Source: YRBS, survey-weighted.
The doubly-bullied are one student in ten. They account for a third of all suicide attempts reported by anyone in the survey’s cross-section.
03 · The adjustment

The association survives every control we can throw at it

Raw gaps can mislead: girls are bullied more and report more distress; younger students differ from older ones. So we fit survey-weighted logistic regressions on the pooled 2017–2023 waves (about 62,000 students), adjusting for sex, race/ethnicity, grade, and survey year, with standard errors clustered on the sampling unit. The gradient barely budges. Compared with otherwise-similar non-bullied classmates, students bullied both ways carry 6.2× the odds of persistent sadness, 5.9× the odds of seriously considering suicide, and 6.8× the odds of attempting it. Either form alone roughly triples each outcome.

A cross-sectional survey cannot prove that bullying causes the despair — distressed teens may also be targeted more. But the dose-response shape, the consistency across outcomes, and the sheer size of the adjusted association put the burden of proof on anyone claiming these are unrelated.

Adjusted odds ratios vs. non-bullied students, 2017–2023
Log scale. Whiskers span the 95% confidence interval.
FIG 3 Survey-weighted logistic regression, pooled 2017–2023 national YRBS. Each model adjusts for sex, race/ethnicity, grade, and year; SEs clustered by PSU. n = 61,779–61,922 (sadness/ideation), 53,585 (attempt).
04 · What it means

The policy target is exposure, not platform

Two facts should anchor any response. First, the overlap group dominates the harm: most cyberbullied teens are also bullied face-to-face (9.8% of all students are in both categories, versus 6.2% online-only), so “get them off the phone” addresses at most half the exposure. Second, the flat trend lines mean that a decade of anti-bullying programming has, at the national scale, mostly held the line rather than pushed it down — while the mental-health floor beneath bullied students has been sinking for everyone. A fifth of the student body carries measurably higher suicide risk that adults can actually observe: bullying, unlike private despair, has witnesses.