Correlation and causation in screen-time headlines

Few research topics produce as many alarming headlines as children and screens. One week a study links phones to teenage anxiety; the next, another finds almost no effect. Parents reasonably ask how both can be true. Usually the answer lies in the difference between correlation and causation, and in a few habits of reporting that blur it.

This guide uses screen time as the running example because it's where the problem shows up most. The same reasoning applies to studies about diet, sport, homework, bilingualism or almost anything else.

What the words mean

Two things are correlated when they tend to go up and down together. Teenagers who spend more hours on social media, on average, report somewhat lower wellbeing in many surveys. That's a correlation.

Causation means one thing actually produces the other: if you changed the first, the second would change too. A correlation is consistent with causation, but it is also consistent with several other explanations. Telling them apart is most of the work.

Three explanations for any link

Say a survey finds that children who use tablets more sleep less. At least three stories fit the result.

1. Tablets reduce sleep. Late-night use delays bedtime or the content keeps children alert.
2. The arrow points the other way. Children who struggle to sleep pick up a tablet because they're awake anyway. This is called reverse causation.
3. Something else drives both. A chaotic household, a long commute, a parent working nights or an anxious temperament might lead to both more screen use and less sleep. That third factor is a confounder.

A single survey usually can't separate these. Often all three are partly true.

Why confounders are hard to remove

Researchers try to account for confounders statistically. They measure things like family income, parents' education and the child's age, and adjust for them. That helps, but there are limits. You can only adjust for what you measured, and many important things, such as family relationships, are hard to measure well.

Adjustment can also go wrong in the other direction. If researchers adjust for something that is actually part of the effect, they can hide a real link. Choosing what to adjust for is a judgment call, and reasonable researchers make different calls.

How big is the effect?

A study can find a "statistically significant" link that is tiny. With tens of thousands of participants, even very small differences pass statistical tests. So always ask how big the effect was, in terms you can picture.

Screen-time research offers a well-known example. In 2019, researchers at the University of Oxford analyzed three large datasets covering more than 300,000 adolescents and parents in the UK and US. They found that digital technology use was associated with lower wellbeing, but the association was very small, explaining at most about 0.4% of the variation in wellbeing. They compared it with other factors in the same data, and found that associations of similar size showed up for things like regularly eating potatoes, while factors such as being bullied or getting enough sleep showed much larger links.

That study doesn't prove screens are harmless. It shows that, averaged across many teenagers, total screen use was a weak predictor of wellbeing compared with other parts of their lives. Averages can hide children for whom the effect is larger.

"Screen time" isn't one thing

A video call with a grandparent, an hour of homework, a shooter game and scrolling at 2 a.m. all count as screen time in many surveys. Mixing them together blurs real effects. Newer research increasingly asks what children do on screens, when, and what the screen time replaces: sleep, exercise, homework or time with friends.

So when two studies disagree, check what each one measured. One may count hours on any device; another may look only at social media; a third may focus on use after bedtime. They can all be right about their own question.

Better designs, closer to cause

Some study designs get closer to causal answers:

• Longitudinal studies follow the same children over time and can test whether screen use comes before changes in mood or the other way round.
• Within-child comparisons check whether a child does worse in the months they use screens more than usual, which removes stable differences between families.
• Natural experiments use events that changed screen access for reasons unrelated to the children, such as when high-speed internet reached different areas.
• Randomized trials ask some participants to cut back and compare them with others who don't. These exist mostly for older teens and adults and tend to be short.

When these different approaches point the same way, the case for cause gets stronger. When they disagree, caution is the sensible response.

What official guidance says

Public health bodies have issued guidance despite the uncertainty, mainly for young children. The World Health Organization's 2019 guidelines recommend no sedentary screen time for babies under one, and no more than one hour a day for children aged two to four, with less being better. In early 2026 the American Academy of Pediatrics published a policy statement that moves away from fixed hour limits. It asks families to focus on content, context and a family media plan, and to keep screens out of bedrooms and off before bed. In 2023 the US Surgeon General issued an advisory saying there was not yet enough evidence to conclude that social media is sufficiently safe for young people, and calling for more research.

Guidance like this is a judgment made under uncertainty. It's reasonable to follow it while understanding that the underlying evidence is still mixed.

Reading the next headline

When a screen-time story appears, run through four quick checks:

1. Was it a trial, a study following children over time, or a one-off survey?
2. Does the headline say "causes" or "linked to", and which does the study support?
3. How big was the difference, in hours of sleep, points on a scale, or children out of 100?
4. What exactly was measured as "screen time"?

The study checklist asks the same questions and adds a few more. For the size question, the next guide on sample size explains why big studies find small effects and small studies find big ones.