Social media and communication
What social media does to people
The best correlational studies bound the effect near zero and the best experiments find a small negative one, and the disagreement between them is unresolved.
The evidence does not converge, and anyone claiming it does is picking a side. Two bodies of strong research point different ways. Analyses of large national datasets find associations between social media use and well-being so small they sit near the noise floor of survey research. Randomised deactivation experiments and quasi-experimental studies of platform rollout find effects that are also small, but negative and causally identified. The average effect on an average person lies somewhere between nothing and roughly a tenth of a standard deviation — a range covering both "irrelevant" and "worth a public health response", which is why the argument continues. What follows sets out both sides without reconciling them, because the literature has not reconciled them either. For a workplace-oriented example of technology measurement in practice, see this overview.
The correlational ceiling is very low
The most widely cited attempt to size the correlation pooled three national datasets and asked how much of the variation in adolescent well-being digital technology use could account for at all. The answer is a ceiling rather than an estimate, and it is low enough that the authors compared it to ordinary household variables.
Digital technology use explains at most 0.4% of the variation in adolescent well-being — a smaller association than regularly eating potatoes, and about a quarter the size of the association with being bullied.
Direction: Decrease. Strength of evidence: Strong.
Caveat All three datasets are cross-sectional and rely on self-reported screen time, so this bounds the size of a correlation rather than a causal effect and cannot rule out larger effects in small subgroups.
Meta-analysis of the narrower question — time on social media against internalising symptoms such as anxiety and low mood — lands in the same place, below the threshold the authors set for an effect worth interpreting.
The association between time spent on social media and adolescent internalising symptoms was r = .061, below the authors' r = .10 threshold for an interpretable effect.
Direction: No detectable effect. Strength of evidence: Mixed.
Caveat The search covered only PsycInfo and Medline, just 5% of included studies were preregistered, and the lead author is a long-standing public sceptic of media-harm claims.
The experiments find something, and it is also small
Correlations that size are compatible with no effect at all. Experiments are harder to dismiss because they create the variation rather than observing it. The largest deactivation study yet run paid people to give up their accounts for six weeks either side of a US election.
Six weeks off Facebook improved self-reported emotional state by 0.060 SD and six weeks off Instagram by 0.041 SD.
Direction: Decrease. Strength of evidence: Strong.
Caveat Fewer than 1% of invited users enrolled, the sample skews liberal and civically engaged, the outcome is three self-report items, and fieldwork ran during the unusually charged 2020 US election.
The other strong causal design exploits history: Facebook launched college by college, so some students got it a year or two before otherwise similar peers.
The staggered 2004–2006 rollout of Facebook worsened student mental health by 0.085 SD, which the authors estimate accounts for roughly 24% of the two-decade rise in severe depression among US college students.
Direction: Decrease. Strength of evidence: Strong.
Caveat The product studied had no feed algorithm, no video and no smartphone attached to it, so extrapolation to 2026 platforms is an assumption rather than a finding.
The two literatures are not measuring the same thing
The disagreement is not a matter of one side being sloppy. The correlational studies measure how self-reported use differs between people and what that difference predicts. The experiments measure what happens when a particular product is removed from a particular population for a set number of weeks. A near-zero between-person correlation and a real within-person effect of removal can both hold at once.
One finding cuts against everything built on the first approach: self-reported use is a poor proxy for what devices actually record.
Across 106 effect sizes, self-reported digital media use was rarely an accurate reflection of logged device use, with the weakest correspondence for measures of "problematic" use.
Direction: No detectable effect. Strength of evidence: Strong.
Caveat This is a measurement finding rather than an effects finding, so it lowers confidence in the surrounding literature without indicating which direction the bias runs.
It also constrains what can be said about whether technology use meets any clinical definition of addiction, which rests on the same self-report instruments.
Effects may be concentrated at particular ages
Averages hide structure. A British longitudinal study looked for periods when the year-on-year link between use and life satisfaction was strongest, rather than assuming it constant across adolescence.
Higher social media use predicted lower life satisfaction a year later specifically at ages 11–13 for girls, 14–15 for boys and around age 19 for both.
Direction: Decrease. Strength of evidence: Mixed.
Caveat Uses self-estimated rather than logged use with annual measurement intervals, and the authors do not claim causal identification.
Exposure is measured much better than harm
What children encounter is now counted directly, and the counts are high. Ofcom's 2026 research found that 73% of UK 11–17-year-olds reported meeting potentially harmful content in a four-week window, including 62% who saw bullying content and 23% who saw eating-disorder content (Children's Online Experiences Research Report, 2026). Seeing is not the same as being harmed by, and no study links those exposure rates to an outcome; the material on bullying that follows children home treats the same measurement problem.
Teenagers' own verdicts split along a familiar line. Pew Research Center found in 2025 that 48% of US teens said social media has a mostly negative effect on people their age, up 16 points since 2022, while 14% said the same about themselves. That gap is the classic third-person effect, a finding about perception rather than harm.
The short version
- The strongest correlational evidence caps the association between digital technology use and adolescent well-being at 0.4% of the variance (Orben & Przybylski, 2019).
- The largest randomised deactivation experiment finds a real but small improvement in mood, of 0.060 SD for Facebook over six weeks (Allcott et al., 2025).
- The best quasi-experimental estimate of a platform arriving, 0.085 SD, comes from a 2004-era product that shares little with current apps (Braghieri et al., 2022).
- Most of this literature rests on self-reported use, which corresponds only weakly to logged use, so its numbers should be treated as approximate (Parry et al., 2021).
- The average effect and the effect on a particular thirteen-year-old girl are different quantities, and only the first has been estimated with any precision.