Young people who spend more time on recreational screens often report more depression and anxiety symptoms. But does screen use cause those difficulties, or do other factors help explain why they occur together? A large Swedish twin study suggests that the answer is more complicated than a simple warning about hours spent online.
Researchers followed 21,797 twins from childhood into young adulthood and found associations between heavier recreational screen use and worse mental health in conventional analyses. When they compared twins within the same families, however, the associations became substantially smaller and did not remain statistically significant after correcting for multiple tests. The study, published in Nature Human Behaviour on 8 October 2026, provides unusually detailed evidence about why correlation should not automatically be treated as causation.
Why comparing twins changes the question
Imagine two teenagers who use their phones for different amounts of time. If the heavier user also has more anxiety symptoms, screen time may be one explanation. But the teenagers may differ in family circumstances, genetic predispositions, earlier emotional difficulties, bullying, sleep patterns and other experiences. Those factors can influence both screen habits and mental health.
Twins offer a way to narrow some of those differences. Identical twins share essentially all their inherited genetic variation and usually grow up in the same household. Fraternal twins share, on average, about half their segregating genetic variation and much of their family environment. Comparing twins who differ in screen use therefore helps researchers account for influences that ordinary population comparisons often cannot measure.
The approach is not equivalent to randomly assigning children to different screen-time limits. Twins can still have different friends, school experiences, stresses and individual characteristics. But if an association becomes much smaller when researchers compare twins within families, it suggests that shared familial factors contributed to the original relationship.
How the Swedish study was designed
Tong Gong and colleagues analysed the Child and Adolescent Twin Study in Sweden, known as CATSS. The initial cohort contained 21,797 twins, with 49.9% female participants and a reported response rate of 68.5% for the age-nine interview.
Parents provided information when the children were about nine years old. Follow-up questionnaires were completed by 16,027 twins at age 15, 13,050 at age 18 and 5,772 at age 24. The researchers also linked survey information with Swedish national health and medication registers to examine clinically recorded depression and anxiety outcomes.
Screen use was measured differently across ages. At nine, parents reported how frequently children watched television or videos, played games or used the internet. At 15 and 18, participants reported time spent on different recreational activities during weekdays and weekends. The authors distinguished passive activities, such as watching television and videos, from interactive activities, including gaming, chatting and browsing.
For adolescent analyses, the researchers classified more than six hours per day as heavy use and two hours or less per day as the comparison group. Those are categories used for analysis, not proof that any single number of hours is a universal safe or unsafe threshold.
What the conventional comparisons showed
Across the full cohort, heavier recreational screen use was associated with more internalising problems, the research term for symptoms such as depression and anxiety. The associations were generally more pronounced during adolescence than at age nine, and interactive screen activities were particularly relevant.
At ages 15 and 18, comparisons between heavy use and the lowest-use category produced standardised symptom differences of approximately 0.20 to 0.38 across the relevant analyses. In some cross-sectional comparisons, the estimates were larger among girls, with reported standardised differences of 0.37 to 0.51, compared with 0.19 to 0.22 among boys.
These figures describe statistical differences in symptom scores, not the percentage of teenagers who became depressed. A standardised coefficient expresses a difference relative to variation in the symptom measure. It should not be translated into a direct clinical risk percentage.
The researchers also found that heavy interactive use at age 15 was associated with higher levels of depression and anxiety symptoms at age 18. In the full sample, the reported standardised coefficients for these later symptoms ranged from 0.14 to 0.24. That prospective association is important because screen habits were measured before the later outcome. However, measuring one factor earlier does not, by itself, eliminate confounding.
What happened when twins were compared with each other
The key test was to repeat analyses within twin pairs whose screen habits differed. Among identical twins at age 15, the estimates were 32% to 95% smaller than the corresponding full-sample associations. The adjusted results no longer met the study’s statistical significance thresholds after accounting for multiple comparisons.
At age 18, the researchers also observed attenuation within identical twin pairs. For relevant comparisons, estimates were 16% to 45% smaller than in the full cohort. In cross-age analyses, associations between heavy interactive screen use at age 15 and later symptoms were reduced by 28% to 58% within discordant twin pairs and did not remain statistically significant after multiple-testing correction.
This pattern does not establish that screen use has no effects. A smaller within-family estimate may reflect confounding, limited statistical power or both. But it makes a straightforward causal interpretation of the larger population-level associations harder to defend.
The clinical outcomes still warrant attention
Questionnaire symptoms are one measure of mental health. The researchers also examined diagnoses and antidepressant dispensing in Swedish national registers. During follow-up from the age-nine survey, 2,523 of 21,797 participants (11.5%) had a recorded clinical outcome under the study’s definition. Among those who completed the age-15 survey, 2,843 of 16,027 (17.7%) had such an outcome during the corresponding follow-up period.
In the full cohort, teenagers reporting more than six hours of interactive screen use on weekdays at age 15 had an adjusted hazard ratio of 3.37 for later clinical outcomes compared with the lowest-use group. The 95% confidence interval was 1.98 to 5.73. Heavy weekend use was associated with a hazard ratio of 2.30, with a 95% confidence interval of 1.47 to 3.60.
A hazard ratio compares rates of events over follow-up, not the proportion of individuals who will necessarily develop a condition. These findings should therefore not be described as proof that heavy screen use triples a teenager’s chance of depression.
After comparing twins within families, the weekday heavy-use estimate fell to a hazard ratio of 1.65, with a 95% confidence interval of 1.07 to 2.56. Although that interval excludes one, the association did not survive the authors’ correction for multiple statistical tests. The result leaves uncertainty rather than confirming either a harmful causal effect or complete absence of risk.
Screen use and mental health may influence each other
The researchers also investigated the reverse direction: whether earlier emotional symptoms predicted later screen habits. Higher depression or anxiety symptoms at age 15 were associated with slightly heavier interactive screen use at age 18. The reported odds ratio was 1.04 per standard-deviation increase in symptoms, with a 95% confidence interval of 1.02 to 1.06.
This is a small association, but it illustrates an important possibility. Young people experiencing emotional difficulties may turn to games, online communities or browsing for distraction, social contact or relief. In other circumstances, digital experiences may affect mood. Both directions can operate alongside family and individual factors.
The study was not designed to determine whether specific online activities are supportive or harmful for particular individuals. Its broad categories combine very different experiences. Playing a cooperative game with friends, encountering online harassment and scrolling alone late at night are not interchangeable exposures.
Why the study matters for parents and policymakers
Many public discussions treat total screen time as a simple measure of risk. The Swedish findings suggest that this can be misleading. A correlation between hours online and mental health may partly reflect pre-existing vulnerabilities and shared family influences. Limiting time alone may therefore be an incomplete response to the problems identified in observational research.
Parents and educators can still take concerns about excessive screen engagement seriously, particularly when it displaces sleep, schoolwork, exercise or face-to-face relationships. But a more useful conversation may examine what a young person is doing online, why they are doing it, whether the experience is distressing and what else is happening in their life.
The findings also have implications for research-based guidance. Policies should distinguish between evidence that identifies a risk marker and evidence that shows an intervention changes mental health outcomes. A study linking heavy use with later diagnoses can help identify groups who may need support, even when it cannot prove that reducing screen time would prevent those diagnoses.
Limitations that shape the interpretation
The study has several important limitations. Screen habits were reported by parents or participants rather than continuously measured from devices, and the questionnaires changed with age. Broad activity categories did not capture platform content, online experiences or the circumstances in which screens were used.
Participation also declined across follow-ups: 73% of the original cohort completed the age-15 survey, 60% the age-18 survey and 27% the age-24 survey. Differences between people who continued and those who did not could affect estimates. The Swedish cohort may not represent adolescents in countries with different family structures, digital access, school systems or healthcare arrangements.
Within-pair comparisons control shared family influences but cannot automatically control factors unique to one twin, including bullying, relationships or individual stress. They also require twins to differ enough in screen exposure to make informative comparisons. When the number of discordant pairs is small, estimates become less precise.
Finally, the authors performed many analyses across ages, screen activities, symptoms and clinical outcomes. Correcting for multiple testing reduces the chance of declaring chance findings significant, but it also means some suggestive estimates remain statistically inconclusive. Non-significance is not proof of zero effect.
What the findings add to the debate
The strongest conclusion is not that screens are harmless or that they inevitably damage mental health. It is that family background and inherited characteristics appear to explain a substantial share of the association between recreational screen use and internalising problems in this cohort.
The study reinforces the need for more precise research on what people do online, which young people are most vulnerable and which changes in digital behaviour actually improve wellbeing. It also offers a practical reminder: when a striking risk estimate becomes much smaller under a stronger research design, that change is itself a major finding.
Source Information
Original research: Recreational screen use and internalizing problems from preadolescence to young adulthood in a longitudinal twin cohort with co-twin comparison.
Authors: Tong Gong, Emma Frans, Anna Ohlis, Shuyang Yao, Ruyue Zhang, Anders Nilsson, Yasmina Molero, Miriam A. Mosing, Isabell Brikell, Henrik Larsson, Paul Lichtenstein, Lisa B. Thorell, Patrik K. E. Magnusson, Ralf Kuja-Halkola and Yi Lu.
Journal: Nature Human Behaviour.
Published: 8 October 2026.
Study type: Peer-reviewed longitudinal observational twin cohort study with co-twin comparisons and national register linkage.
Sample: 21,797 Swedish twins at baseline, with smaller follow-up samples.
DOI: 10.1038/s41562-026-02607-0.
