What youth online harm data exists, and where are the gaps?

Within this index, Ofcom and Australia’s eSafety programs provide broader online-harm coverage than the CDC National YRBS, which covers Bullying. Ofcom’s COST covers six mapped categories while eSafety’s Keeping Kids Safe Online covers seven.

At the US state level, several states conduct notably large-scale in-school surveys that include some data on online harms. For example, the 2025 wave of the Pennsylvania Youth Survey (PAYS) reports 267,379 valid responses, compared with 10,710 in the CDC National YRBS 2025 wave (both reports were released in 2026). Some State surveys have broader online-harm coverage than the CDC YRBS, but none of the State entries in this index measures as many mapped categories as Ofcom’s COST or eSafety’s Keeping Kids Safe Online.

The National index entries cover a wide range of age ranges. Some cover only children under 13; others include younger children and teens, older respondents, or adults reporting childhood experiences retrospectively.

19 of the 41 National entries have indexed age ranges reaching below 13. Ofcom’s Children’s Online Behaviour and Attitudes survey reaches age 3 through parent reports, with children answering from age 8. OfCom’s Children and Parents Media Literacy Tracker covers ages 8-17. Ofcom’s Children’s Online User Ages entries cover ages 8–12 within the full 8–17 surveys. Australia’s eSafety Behind the Screen entry covers ages 8–12 within the full 8–15 survey, while Keeping Kids Safe Online covers a 10–15 subgroup within its full 10–17 survey. eSafety’s follow-up evaluation of social media age restrictions covers ages 10–15. In contrast, in the US, several of the main probability-based surveys are only 13+, including Pew and USC / Stanford.

US state surveys partly fill younger-population gaps: all nine entries include middle-school students, and their lowest grades are 5 or 6 (School-based surveys often cover grade levels vs. age ranges). More regular data on younger users in the US would meaningfully improve understanding of their experiences; this is also the population where underage-use and early-exposure questions matter most.

30 of the 41 National entries document app-level measurement of a harm or risk factor, including underage app use or account ownership. Not all of these surveys report true per-app prevalence data. Many report platform attribution (the share of victims who name a given platform) rather than a prevalence rate (the share of an application’s users who report experiencing the harm on that application). Attribution reflects both an app’s reach and its users’ reported harm rate, so rankings can differ. Our harm rates by app analysis summarizes reported or derived per-user rates from USC Neely / Stanford CDH, Pew, Thorn, Common Sense among US girls, the Youth Endowment Fund, Ofcom’s Children’s Online Safety Tracker, and Australia’s eSafety Commissioner.

10 of the 41 National entries and all 9 State entries measure at least one offline or broader all-context analogue of an online harm. Differences in wording, populations and recall periods mean the National and State surveys do not always provide offline-only rates or directly comparable online and offline estimates.

The most common comparison of online and offline harm is bullying. Most state comparisons involve bullying. The USC Neely / Stanford CDH survey asks about five indexed harm categories online and in person. Its published article reports online and offline rates for bullying, unwanted contact, violent content and self-harm content; its photo-misuse estimates are broader than NCII. Common Sense Girls & Mental Health compares online and "real life" negative peer experiences, including bullying, as well as self-harm and eating-disorder content. Finkelhor’s retrospective study compares online child sexual abuse with broader child sexual abuse before age 18. Oregon compares technology-mediated dating control with broader physical dating abuse/control, while Georgia measures actual self-harm and suicide attempts across all contexts alongside its social-media-attributed reasons questions.

There is no standard instrument for measuring online harms — surveys differ in question wording, recall window, and the denominator used to compute rates, so headline numbers for the “same” harm are frequently not comparable. Specifically:

1. Construct wording. For example, cyberbullying is defined and measured differently across surveys. The CDC YRBS asks a single yes/no item prefaced by a formal definition that limits bullying to repeated peer-to-peer aggression with a power imbalance (“when 1 or more students … over and over again,” explicitly excluding “2 students of about the same strength or power”). Cyberbullying Research Center uses a 9-item victimization scale prefaced by a broader definition with no peer or power-imbalance restriction (“when someone repeatedly and intentionally harasses, mistreats, or makes fun of another person…”); Pew’s 2022 Cyberbullying survey lists six specific behaviors, while its 2026 Cross-Platforms survey asks a three-item battery per platform; USC Neely / Stanford CDH asks one yes/no question covering five harassment behaviors; and eSafety’s Keeping Kids Safe Online (KKSO) uses a twelve-item battery framed around an intent to humiliate.

   State instruments differ too. Colorado and Oregon ask about electronic bullying; Rhode Island asks separately about cyberbullying by schoolmates and by others; Utah asks about threats or harassment through the internet, email or cell phones. Pennsylvania asks about bullying and then where it happened, with online as one option. Oklahoma asks whether cyberbullying is a problem at school, rather than whether the student was personally cyberbullied.

2. Harm recall windows. These differ by survey: windows used by surveys in our index include 30 days, 4 weeks, 3 months, 12 months, the current school year, and lifetime/ever. Australia’s eSafety Keeping Kids Safe Online measures both lifetime and 12-month prevalence in the same instrument; 12-month rates are about 60–85% of lifetime rates for most aggregate harms in its Digital Use and Risk report.
3. Baseline / denominators for prevalence calculations. Reported percentages use different denominators, including all respondents; users of a given platform (USC Neely / Stanford CDH); respondents with at least one social-media account (ABCD); or victims who then name a platform (eSafety’s KKSO attribution).

Cross-survey comparisons thus require care, and there may be opportunities to standardize instruments to allow for more direct comparisons going forward.