Why comparing twins changes the answer
Ordinary observational studies compare adolescents who use screens heavily with adolescents who use them little. Those two groups differ in many ways that are hard to measure: genes, parenting, household routines, socioeconomic background. The paper's central move is to compare a twin who uses screens heavily with the co-twin who does not. Monozygotic twins share all inherited genetic factors and their rearing environment; dizygotic twins share their rearing environment and, on average, half of their segregating genes. A within-pair comparison therefore removes much of the confounding that population-level comparisons cannot address.
The authors applied this co-twin control design to three questions at once: cross-sectional associations between screen use and internalizing symptoms, cross-age associations in both directions, and time-to-event analyses for clinically diagnosed depression and anxiety. The sample was 21,797 Swedish twins from the Child and Adolescent Twin Study in Sweden, followed at ages 9, 15, 18 and 24, with questionnaire data linked to national registers for diagnoses and dispensed antidepressants.
The logic of the design also sets the interpretation. If confounding were purely genetic, associations would vanish within monozygotic pairs and be partly reduced within dizygotic pairs. If confounding were purely shared environment, associations would appear at the population level and disappear within both types of pairs. The pattern the authors report — attenuation that is partial rather than complete — is what they read as a mix of shared liability and, at most, a modest causal component.
Population-level associations rise with heavier use
Before any within-pair adjustment, the cohort showed the pattern that motivates public concern. At ages 15 and 18, longer daily screen time corresponded to higher symptom scores in a dose–response fashion, with standardized coefficients of 0.20 to 0.38 when heavy use (more than 6 hours per day) was compared with restrictive use (2 hours per day or less). The estimates were larger in girls than in boys for interactive activities such as gaming, chatting and browsing: heavy use was associated with standardized scores of 0.37 to 0.51 in girls and 0.19 to 0.22 in boys.
The clinical outcomes pointed the same way. Among the age-15 participants, 17.7% (2,843 of 16,027) had a clinical outcome recorded during follow-up to age 28. Heavy interactive screen use at age 15 was associated with higher risk than restricted use, with adjusted hazard ratios of 3.37 for weekday use and 2.30 for weekend use. At age 9, daily gaming reported by parents was associated with an adjusted hazard ratio of 1.38 for later depressive or anxiety disorders, and 1.56 in boys. Passive activities such as watching TV, video and DVD showed mostly non-significant associations with later symptoms or diagnoses.
These numbers describe associations in the full sample. They are the baseline against which the twin comparisons have to be read, not evidence of causation on their own.
Within twin pairs, most estimates lose significance
The within-pair results are the paper's decisive evidence. For cross-age associations, heavy interactive use at age 15 was associated with elevated depressive or anxiety symptoms at age 18 in the full sample (standardized coefficients 0.14 to 0.24), but within discordant twin pairs the estimates fell by 28–58% to 0.06 to 0.13 and were no longer significant after correction for multiple testing. Heavy interactive use at age 18 and depressive symptoms at age 24 followed the same pattern: full-sample coefficients of 0.28 to 0.33, within-pair estimates of about 0.19 attenuated by 32–42%, and non-significance after correction.
The cross-sectional comparisons at ages 15 and 18 showed a graded pattern that the authors treat as informative. Associations remained significant within dizygotic pairs but were attenuated further within monozygotic pairs, where estimates dropped by 32–95% at age 15 and were non-significant after correction. That ordering — population, then dizygotic, then monozygotic — is consistent with partial genetic and shared environmental confounding rather than a clean causal effect.
For clinical outcomes, heavy interactive use at age 15 showed a 57–59% reduction in the within-pair estimates. The weekday estimate remained elevated (adjusted hazard ratio 1.65, 95% confidence interval 1.07–2.56) but did not survive correction for multiple testing. At age 9, the association between daily gaming and later diagnoses was not observed within discordant twin pairs at all.
The authors are explicit that attenuation was often incomplete: estimates frequently did not fall fully to the null. Their reading is that shared familial factors explain a substantial portion of the association, while the residual within-pair estimates provide only limited evidence of a clear causal effect.
Symptoms also predict later screen use
The cross-age analyses tested the reverse direction as well. Depressive and anxiety symptoms at age 15 were associated with heavier interactive screen use at age 18 on weekends, with an odds ratio of 1.04 (95% confidence interval 1.02–1.06) per standard deviation increase in standardized symptom scores. That is a small estimate, but it indicates that the temporal ordering is not one-way.
This matters for interpretation because a causal story in which screens drive symptoms is not the only ordering the data support. The authors describe the cross-age findings as bidirectional around mid-adolescence to young adulthood. Combined with the within-pair attenuation, the bidirectional pattern weakens any simple account in which restricting screen time alone would reliably reduce internalizing problems.
The paper also reports that the consistent female excess seen in the cross-sectional analyses did not reappear as a consistent pattern in the cross-age associations, and no clear sex difference was observed in the reverse direction.
What the twin design still leaves open
The co-twin design controls for factors shared within a pair, but it does not control for factors that differ between the twins themselves. The authors note that the within-pair contrast can be read as a potential causal effect within families only if the effect is approximately constant across families. They also note that screen use variables at earlier timepoints could not be adjusted for in the mid- and late-adolescent models because of multicollinearity, and that physical activity and bullying victimization were handled only in sensitivity analyses, where estimates attenuated only slightly.
Measurement differs across ages in ways that affect how much the within-pair comparison can detect. Parent-reported screen use at age 9 was highly correlated within twin pairs (correlations of 0.94 to 0.97 for monozygotic and 0.46 to 0.92 for dizygotic pairs), which the authors suggest may explain the limited within-pair evidence at that age. Self-reported use at ages 15 and 18 was less correlated within pairs, leaving more variation for the comparison to use.
The authors also state plainly that although associations were consistent across cross-sectional, cross-age and longitudinal approaches, the results remain correlational. The co-twin estimates complement randomized evidence with a larger sample and longer follow-up, but they do not replace it.
The residual signal sits with interactive use
The most useful distinction in this paper is not between screen time and no screen time, but between interactive and passive use, and between population-level and within-pair estimates. Passive activities such as watching TV, video and DVD showed mostly non-significant associations with later symptoms and clinical outcomes in both the full sample and within pairs. The associations that survived longest involved interactive activities — gaming, chatting and browsing — during adolescence.
Even there, the surviving evidence is thin. The within-pair clinical estimate for weekday heavy use at age 15 remained elevated at an adjusted hazard ratio of 1.65 but did not pass the multiple-testing threshold. The authors treat the heavy interactive use and clinical risk association as a reason to investigate mechanisms linking digital behaviours and youth mental health, not as a demonstrated causal pathway.
For readers weighing this evidence, the practical implication follows from the comparison structure: an association observed only at the population level, which shrinks when twins are compared, is more plausibly a marker of shared familial liability than a lever that screen restriction alone would move. The paper's own priority is to understand what distinguishes heavy interactive use within families, since that is where the residual signal, however modest, remains.
Sources used in this answer
Recreational screen use and internalizing problems from preadolescence to young adulthood in a longitudinal twin cohort with co-twin comparison
Recreational screen use was associated with internalizing symptoms and diagnosed depression and anxiety in the full cohort, but associations were substantially attenuated within discordant twin pairs and did not remain significant after correction for multiple testing, suggesting shared genetic and environmental factors partly explain the link.