The Problem School Leaders Face

One in five children aged 8-16 in England had a probable mental disorder in 2023, up from around one in nine in 2017. Among 17-19 year olds, the rate has more than doubled over the same period, from 10.1% to 23.3%. Young women in that age group are twice as likely to be affected as young men. CAMHS waiting lists have reached record lengths, with referrals to children’s mental health services more than doubling between 2017 and 2022.

Against this backdrop, the debate about social media’s role is loud, polarised, and largely unresolved. Jonathan Haidt’s The Anxious Generation (2024) argues smartphone adoption directly caused declining adolescent wellbeing. Researchers at Oxford’s Internet Institute, using equally large datasets and more rigorous methods, reach a starkly different conclusion. Meanwhile, school leaders are being asked to respond, often under media or parental pressure, without clear guidance on what the evidence actually supports.

What the Research Shows

Screen time, taken alone, predicts almost nothing.

One of the largest and most methodologically rigorous studies to date, Orben and Przybylski (2019, Nature Human Behaviour), analysed over 355,000 individuals and found digital technology use explains at most 0.4% of variation in adolescent wellbeing. That is roughly comparable to the association between eating potatoes and wellbeing. Przybylski and Weinstein’s earlier preregistered study of 120,115 English adolescents similarly found only a weak, nonlinear relationship between screen time and mental wellbeing, with effects varying by platform, context, and time of day, not volume of use alone.

Knowing how many hours a student spends online tells you almost nothing about their mental health.

Problematic use is a different matter entirely.

When researchers ask not how long but how, whether use is compulsive, displaces sleep, involves upward social comparison, or triggers anxiety-driven checking, the picture changes. Shannon et al. (2022, JMIR Mental Health), drawing on 18 studies and over 9,000 participants, found moderate but statistically significant correlations between problematic social media use and depression (r=0.273), anxiety (r=0.348), and stress (r=0.313). A 2024 review in Nature Reviews Psychology by Orben, Przybylski and colleagues identifies specific behavioural, cognitive, and neurobiological pathways through which social media amplifies developmental risk, reinforcing that how platforms are used matters far more than time spent on them.

Girls on visual platforms are a higher-risk group.

NHS Digital’s 2023 survey found rates of probable mental disorder among 17-19 year old young women at 31.6%, compared to 15.4% in young men. Twenge et al. (2020, Journal of Adolescence), analysing three large datasets, found heavy users among girls were often twice as likely to experience mental health difficulties as boys with equivalent usage. Instagram and TikTok are structurally built around social comparison, appearance-based evaluation, and follower metrics, mechanisms that consistently show stronger associations with poor mental health in adolescent girls than boys.

Offline factors remain the dominant predictors.

Family support, sleep quality, socioeconomic circumstances, and prior mental health status explain far more variance in adolescent wellbeing than social media use across the literature. NHS Digital data shows children aged 11-16 with a probable mental disorder were five times more likely to have been bullied in person, and those from financially strained households were substantially overrepresented. Bohnert et al. (2023, Journal of Adolescence) further demonstrates that low-SES adolescents are more negatively affected by heavy screen time than their peers. Social media risk is not evenly distributed. For many young people, social media also provides genuine peer support and community that carries real protective value.

What This Pilot Found

This ten-week pilot study tracked 12 Year 10 students (6 female, 6 male; aged 14-16) at Oakwood Academy, a pseudonymised South London comprehensive. Mood was assessed weekly using the validated WEMWBS scale (Tennant et al., 2007). Screen time was captured through a combination of self-report and objective app data (80% compliance). Semi-structured interviews and sleep diaries validated against actigraphy provided qualitative context.

This is an explicitly small pilot. Statistical findings are indicative rather than definitive; the sample size does not support generalisable conclusions. What it offers is pattern recognition, tested against the broader published evidence base.

Three students showed clear social media-mood links. All three were female.

Ava (15; 4.2 hrs/day, Instagram and TikTok) experienced a 15-point WEMWBS decline following a cyberbullying incident on Instagram Stories. Night-time scrolling reduced her sleep to around five hours per night. Sofia (15; 3.5 hrs/day, TikTok) linked algorithm-driven content to a nine-point mood decline and persistent anxiety. Jamie (14; 3 hrs/day, Snapchat) experienced 11-point mood dips linked to compulsive checking and validation-seeking through likes. Across this subgroup, problematic use correlated with mood at r=0.47 (p=0.04), accounting for 22% of mood variance. Mean sleep: 5.3 hours per night.

Nine students showed stable or positive outcomes, regardless of how much time they spent online.

Noah (15; 4 hrs/day) maintained a stable WEMWBS score throughout. Gaming communities provided genuine relaxation alongside strong offline friendships. Priya (14; 2.5 hrs/day, Instagram) used social platforms primarily for cultural connection with extended family; her mood showed no correlation with use (r=0.09). Liam (16; 3.8 hrs/day, TikTok) curated his feed deliberately; family stress, not screen time, was the sole significant mood predictor (b=0.38).

Across the full cohort: total screen time b=0.14 (p=0.52); family support b=0.42 (p<0.01); sleep b=0.31 (p=0.02). Overall R²=0.68.

The gender pattern is not coincidental. It maps directly onto what the literature predicts and has direct implications for how schools design their pastoral response.

What This Means for School Leaders

Blanket phone restrictions address neither the problem nor the evidence. Total screen time is a poor proxy for risk. Policies focused on reducing minutes online are unlikely to identify or support the students who need help, and risk restricting access to platforms that serve genuinely positive social functions for the majority.

Problematic use patterns are identifiable. Compulsive checking, sleep displacement, validation-seeking, and heavy exposure to social comparison content are screenable at pastoral level without specialist input. A brief annual check, far less resource-intensive than most schools assume, can identify the minority genuinely at risk.

Girls using visual platforms need a differentiated response. Any pastoral or PSHE approach that treats all students identically will systematically under-serve the group most at risk. Differentiated content addressing social comparison, appearance-based evaluation, and feed curation specifically is not an optional refinement. It is where the evidence points.

Offline protective factors are the highest-leverage intervention available. Sleep routines, family connection, and financial stability consistently predict wellbeing outcomes more powerfully than anything happening on a screen. Parent engagement that builds these factors is among the most cost-effective actions a school can take.

Recommended Actions

1. Annual pastoral screening for problematic use A brief check using validated indicators: compulsive checking, sleep displacement, FOMO-driven anxiety, mood dependence on social feedback. Identifies the minority at genuine risk without treating all students as vulnerable. Owner: Head of Pastoral | Term 1, 2026/27 | approx. £500 staff training | Anchor: DfE Teaching Online Safety in Schools (2023); KCSIE 2025
2. Differentiated PSHE provision Four sessions per year covering social comparison, sleep hygiene, algorithmic feed curation, and managing compulsive checking. Content differentiated for girls, particularly around visual platform use. Owner: PSHE Lead | From Easter 2026 | Staff time only | Anchor: PSHE Association Digital Wellbeing Guidance (2023)
3. Parent and carer workshops Focused on building offline protective factors: sleep routines, family connection, physical activity. One of the highest-leverage, lowest-cost interventions available to schools. Owner: Family Support Lead | Summer 2026 | approx. £200 | Anchor: Online Safety Act 2023
4. Trust-wide digital wellbeing policy A coordinated policy distinguishing screen time restrictions (weak evidence) from targeted support for problematic use (stronger evidence). Provides governance consistency and demonstrates evidence-informed leadership. Owner: CEO/DSL | Academic year 2026/27

A Note on the Evidence Base

This is a genuinely contested field. Haidt’s work is prominent and the mechanisms he identifies, platform architecture driving compulsive behaviour and visual platforms harming a subset of girls, are plausible and partially supported. His causal claims, however, are disputed by researchers using equally large datasets and more rigorous methods. The Anxious Generation synthesises correlational evidence, not causal longitudinal data.

The honest position is that social media is one contributing factor for a minority of vulnerable young people. The drivers of the current mental health crisis are multifactorial: academic pressure, socioeconomic stress, post-pandemic disruption, and severely constrained CAMHS capacity all play a role. Leaders approving actions in this area should look for policies grounded in problematic use identification, offline buffer-building, and differentiated support for higher-risk groups, not blanket measures unsupported by the evidence.

An Emerging Consideration: AI and the Developing Risk Landscape

The evidence base in this case study addresses social media specifically. A separate but related question is beginning to surface in schools: what role does artificial intelligence play in adolescent mental health, and should school leaders be paying attention to it now?

The honest answer is that the peer-reviewed evidence on AI and teen mental health is not yet at the level of the social media literature. We do not yet have the equivalent of Orben and Przybylski for AI-generated content. What we do have is a set of plausible and documented mechanisms that warrant serious attention.

Algorithmic recommendation systems, which sit underneath TikTok, Instagram, and YouTube, are AI systems. The compulsive use patterns identified in this case study are not accidental features of those platforms. They are the intended output of optimisation engines trained to maximise engagement. Understanding that the harm mechanism is algorithmic rather than simply social reframes the problem for school leaders: the question is not just what young people are consuming but what systems are deciding what they see next.

Beyond recommendation engines, three further developments are worth monitoring. AI-generated imagery is making social comparison on visual platforms more acute, not less. The idealised bodies and lifestyles that drive upward comparison among adolescent girls are increasingly synthetic, removing even the limited anchor of knowing that a real person looked that way. Deepfake technology has introduced a new category of image-based abuse that is beginning to appear in secondary schools. And AI companionship tools, while largely unresearched in adolescent contexts, raise questions about displacement of peer relationships that the wider social media literature suggests could be significant.

KCSIE 2025 has begun to reflect this shift, explicitly referencing AI and generative technology as areas where schools should exercise informed caution. The DfE’s guidance on generative AI in education, updated in 2024, provides a starting framework, though it focuses primarily on academic integrity rather than wellbeing.

The practical implication for school leaders is straightforward: the pastoral screening and PSHE provision recommended in this case study should be reviewed annually to incorporate emerging evidence on AI-specific risks. The mechanisms that make social media harmful to a subset of young people, compulsive engagement, social comparison, sleep displacement, identity validation through feedback, are the same mechanisms that AI systems are increasingly optimised to exploit. The evidence base will catch up. Schools that have already built the infrastructure to identify and respond to problematic use patterns will be well positioned when it does.

References

Bohnert, M. et al. (2023). Digital use and socioeconomic inequalities in adolescent well-being. Journal of Adolescence, 95(4). https://doi.org/10.1002/jad.12183

Department for Education (2023). Teaching Online Safety in Schools. https://www.gov.uk/government/publications/teaching-online-safety-in-schools

Department for Education (2025). Keeping Children Safe in Education 2025. https://www.gov.uk/government/publications/keeping-children-safe-in-education–2

Haidt, J. (2024). The Anxious Generation. Penguin Press.

Newlove-Delgado, T. et al. (2023). Mental Health of Children and Young People in England, 2023 Wave 4 follow-up. NHS England. https://digital.nhs.uk/data-and-information/publications/statistical/mental-health-of-children-and-young-people-in-england/2023-wave-4-follow-up

Odgers, C.L. and Jensen, M.R. (2020). Annual research review: adolescent mental health in the digital age. Journal of Child Psychology and Psychiatry, 61(3). https://doi.org/10.1111/jcpp.13190

Orben, A. and Przybylski, A.K. (2019). The association between adolescent well-being and digital technology use. Nature Human Behaviour, 3. https://doi.org/10.1038/s41562-018-0506-1

Orben, A. et al. (2024). Mechanisms linking social media use to adolescent mental health vulnerability. Nature Reviews Psychology, 3. https://doi.org/10.1038/s44159-024-00282-4

PSHE Association (2023). Digital Wellbeing Guidance. https://pshe-association.org.uk/guidance/ks1-5/digital-wellbeing

Przybylski, A.K. and Weinstein, N. (2017). A large-scale test of the Goldilocks hypothesis. Psychological Science, 28(2). https://doi.org/10.1177/0956797616678438

Shannon, H. et al. (2022). Problematic social media use in adolescents and young adults: systematic review and meta-analysis. JMIR Mental Health, 9(4). https://doi.org/10.2196/33450

Tennant, R. et al. (2007). The Warwick-Edinburgh Mental Well-Being Scale: development and UK validation. Health and Quality of Life Outcomes, 5(63). https://doi.org/10.1186/1477-7525-5-63

Twenge, J.M. et al. (2020). Gender differences in associations between digital media use and psychological well-being. Journal of Adolescence, 79. https://doi.org/10.1016/j.adolescence.2019.12.018

Ward, J. et al. (2025). Admission to acute medical wards for mental health concerns among children and young people in England from 2012 to 2022. Lancet Child & Adolescent Health, 9(1). https://doi.org/10.1016/S2352-4642(24)00259-3