Two people spend five hours a day with screens. One uses a phone to talk with family across a border, then watches a film after work. The other scrolls through a feed late into a night already made difficult by isolation and unpaid bills. Their time totals match. Their circumstances, reasons for use and possible routes to better well-being do not.
The OECD’s 2025 brief Screen Time and Subjective Well-being: Insights from a Few Countries Worldwide examines reported personal screen time and well-being in a poll across 14 countries. It finds an association between long personal use and poorer reported outcomes, while identifying sleep, activity, financial difficulty and loneliness as important parts of the same picture. This guide explains what the poll actually measured, what its statistical comparisons mean and why no universal time limit follows from them.
The question and the sample
The OECD’s Centre on Well-being, Inclusion, Sustainability and Equal Opportunity conducted the Digital Well-being Poll in collaboration with Cisco. The brief analyses 14,611 responses collected in February and March 2025 across Australia, Brazil, Canada, France, Germany, India, Italy, Japan, Korea, Mexico, the Netherlands, South Africa, the United Kingdom and the United States. Each country contributed about a thousand respondents, with a larger sample in India. This is a substantial cross-country snapshot, but it is not a continuous record of how the same people changed over time.
The analysis distinguishes screen use for personal purposes from work-related use. Its personal-use question asks about average daily time during the past week across computers, tablets, phones, televisions and gaming consoles. That breadth matters: a single number combines activities with different social and cognitive meanings. The work measure is especially hard to interpret because some occupations require screens and others do not; the brief therefore focuses its main model on personal use.
The title’s phrase “a few countries worldwide” is appropriately modest. Fourteen countries provide wider scope than one national sample, yet they cannot stand for every population. Country averages can hide variation within countries. The poll’s stratification and quality controls strengthen comparisons within its design, but self-reports and unmeasured context still matter. The accompanying OECD technology brief describes the specialist fieldwork and clarifies that its analysis uses collected survey responses rather than open crowd-sourced submissions to the public-facing hub.
Three meanings of subjective well-being
The brief does not use one all-purpose happiness score. It examines mental well-being with the WHO-5 index, life satisfaction with a 0–10 evaluation, and eudaimonia with an item about whether life feels worthwhile. Each captures a different aspect of how people assess their lives. Someone may find daily life meaningful while also reporting stress or low mood; another may feel comfortable yet question their direction.
The OECD’s updated guidelines on measuring subjective well-being emphasize that evaluation, affect and meaning should be distinguished. That distinction helps prevent a common reading error: a lower life-satisfaction score is not, by itself, a clinical diagnosis. The WHO-5 is a brief self-report measure used to flag possible concerns; the poll does not establish that an individual has a mental disorder. Its outcomes should be named precisely, especially when a chart is summarized in a headline.
The three measures also have different time frames and response habits. A person’s answers may reflect a recent event, long-standing circumstances or cultural norms about rating one’s life. Cross-country survey design can reduce but not erase those differences. A useful interpretation asks whether a pattern appears across multiple measures and whether the direction remains after the authors account for other observed factors.
What the screen-time pattern looks like
In the descriptive figures, reported well-being tends to be highest around lower or moderate personal screen-use categories and lower at the high end. The shape resembles an inverted U rather than a straight line in which every added minute predicts the same decline. The group reporting no screen use is small, about one percent of the sample, and its estimates vary widely. It would be a mistake to infer that zero use is itself harmful from that group alone.
The brief uses moderate use, one to three hours daily, as a comparison category in its regression models. In the base model, people reporting more than five hours of personal use had higher odds of low WHO-5 well-being, low life satisfaction and low eudaimonia than the comparison group. The reported increases were 44%, 47% and 62%, respectively. In a model that includes loneliness and related interactions, the corresponding estimates were 28%, 32% and 47%. These are model-based associations, not an experiment assigning people to five hours of use.
An odds increase is not a percentage-point increase in the chance that a person feels unwell. The baseline rate matters. The categories also do not identify a biological threshold at exactly five hours. They are bins used to compare people in a survey. A person at four hours and fifty-nine minutes does not suddenly change state when the clock passes five. The measured pattern can motivate a closer question, but it cannot supply an individual prescription by itself.
The brief’s abstract notes an association above two hours, while the fuller report presents categories and highlights the clearer high-use contrast with one to three hours. Reading both together prevents the abstract from being mistaken for a simple rule that all personal screen use beyond two hours damages well-being. The full report repeatedly warns that correlation does not establish causation.
Why the causal direction is unsettled
Long screen use might displace sleep, physical activity or face-to-face contact for some people. A distressing online experience might worsen well-being. Those are plausible pathways, but this cross-sectional poll cannot tell which one produced the observed association. People who are lonely, unemployed, unwell or financially strained may also spend more time with screens because of those conditions. Both directions can operate at once.
The OECD includes several factors in its models, but statistical adjustment cannot remove every difference between people. The poll measures self-reported time over a recent week and current subjective outcomes. It does not observe what a given respondent would have felt if they had spent one hour less online while everything else stayed the same. That unobserved comparison is what a causal claim would require.
The activity inside the screen total is another major omission for interpretation. A long video call, job application, online support group and endless autoplay session may have distinct effects. Combining them in one duration can obscure benefit and harm. The companion OECD brief on experiences with technology distinguishes passive exposure from active engagement and invites closer study of how people use tools. The screen-time brief is a starting point for such work, not a substitute for it.
Self-report introduces measurement error. People may estimate time inaccurately, especially when several devices are involved or professional and personal use overlap. Device logs could sharpen duration measurement but would still miss the meaning of a conversation, the quality of a relationship or why someone stayed online. Better measurement needs both behavioral and contextual information.
Loneliness is a distinct signal
The brief finds strong associations between loneliness and low well-being. In its models, respondents who reported loneliness had substantially higher odds of poor outcomes than those who did not. The analysis also examines whether loneliness changes the relationship between high screen time and well-being. It reports an additional concern for poor mental well-being among lonely high-use respondents.
These results should not be read as “screens cause loneliness” or “lonely people should go offline.” The report itself says the direction of the screen-time and loneliness relationship is ambiguous. A person may turn to digital spaces because in-person connection is unavailable; a platform may provide real support; or a pattern of use may leave someone feeling more excluded. The same duration can contain different social experiences.
For practice, this means a useful question is “What contact or activity would you like more of?” rather than “How many hours can we remove?” A reduction that cuts off a valued friendship could worsen the situation. A change that protects sleep while keeping a supportive call might help. The OECD data do not test either intervention, so these are design questions for further evaluation, not recommendations proved by the poll.
Sleep, movement and material conditions
Sleep duration, physical activity and difficulty making ends meet were important correlates of well-being in the OECD models. Very short sleep and financial hardship had stronger associations with low well-being than the personal screen-time categories in several comparisons. This is a crucial corrective to a narrow technology story. A person who cannot afford housing or has no control over a night shift faces constraints that cannot be solved by a phone timer.
At the same time, these factors may interact. A device can delay sleep, but it can also provide work, care coordination or a route to help. Unemployment may increase screen time while independently lowering well-being. A policy that treats time as the only lever risks overlooking the conditions that shape both the time and the outcome.
The evidence invites attention to practical pathways. If a person says late-night use interrupts sleep they value, the relevant change may be a different bedtime routine or a less intrusive interface. If a student uses a screen to reach a community unavailable locally, preserving that connection may be important. If an employer expects constant after-hours responses, the solution may be a work norm rather than individual willpower.
Adults, adolescents and the Goldilocks idea
The OECD compares its curved pattern with the “Goldilocks” hypothesis tested by Przybylski and Weinstein in adolescent data. That earlier study explored whether moderate digital engagement could coexist with good well-being while very high use was associated with poorer outcomes. It is a useful historical comparison, but the two studies differ in participants, measures, devices and time. Their curves should not be treated as a single universal dose-response function.
Children and adolescents also have different developmental contexts from adults in an international poll. UNICEF Innocenti’s 2025 research on childhood in a digital world examined multi-country evidence and emphasized that content and negative online experiences matter, while total time alone did not show a consistent strong link to the mental-health outcomes it assessed. This does not cancel the OECD adult association; it shows why age group and outcome must be kept explicit.
The contrast is a safeguard against misusing the OECD brief to justify a blanket rule for every child. A school deciding how to structure device use should ask what the device is for, what harm it may expose students to, what benefits would be lost, and how learning and well-being will be measured. The OECD poll cannot answer those questions on behalf of a school.
How to read the models without overreading them
Begin with the denominator. The poll covers people in selected countries who answered its survey, not everyone everywhere. Then identify the exposure: self-reported personal screen time across several device types in the prior week. Next identify the outcome: low score on one of three self-report well-being measures. This sequence keeps the study from becoming a vague claim about “technology and mental health.”
Look at the reference group. The comparisons are generally against moderate users, not against a hypothetical person who never uses a screen. Read odds ratios as associations after specified adjustments. An adjusted model can be informative without proving that changing the exposure would change the outcome by the same amount.
Ask whether the policy suggestion follows from the measured variable. A finding about duration does not reveal which platform design is responsible. A finding about loneliness does not tell us which kind of online connection is supportive. A finding about work screen time is confounded by occupation and autonomy in ways the brief acknowledges. The most credible conclusion may be a research priority rather than a rigid rule.
Finally, compare the magnitude and uncertainty of different factors without turning the model into a contest. Sleep, hardship and social support are not merely covariates to be “controlled away”; they are parts of people’s lives. A well-being policy should be able to address them directly.
What the evidence does not show
The OECD brief does not show that five hours of personal screen use causes low well-being, that two hours is an optimal dose for each person, or that removing screens would improve every high user’s life. It does not distinguish every beneficial from every harmful online activity. Its cross-sectional survey and broad duration categories cannot resolve the direction of influence between use and well-being.
Nor should the poll be used to dismiss people’s reported experiences. Someone may accurately notice that a particular app disrupts sleep or that an online group helps them endure isolation. A population association does not override that observation; it supplies context and suggests what to measure more carefully.
The report is most useful when it prevents two opposite simplifications: that all screen time is equally damaging, and that duration can never carry information. High personal use is a signal worth investigating. The investigation should ask what the person does, what they gain, what they lose, what constraints they face and what outcome a change is meant to improve.
Return to the two people
Their five-hour totals do not tell us which one needs a change. A careful response would ask each what that time contains and whether it supports the life they want. One may need a reliable channel to family; the other may need rest, affordable support or relief from an interface designed to keep them scrolling. The clock matters, but it is not the whole story.
The OECD’s contribution is a cross-country warning that long personal screen use and low subjective well-being appear together more often in its survey, especially amid loneliness and other vulnerabilities. Its equally important contribution is to keep the rest of the picture visible. Better digital well-being requires a richer account of circumstances than a single daily number can provide.
Research references
- OECD. (2025). Screen time and subjective well-being: Insights from a few countries worldwide. OECD Policy Insights on Well-being, Inclusion and Equal Opportunity, No. 24. View source
- OECD. (2025). How do people experience new technologies and generative AI? Insights from a few countries worldwide. OECD Policy Insights on Well-being, Inclusion and Equal Opportunity. View source
- OECD. (2025). OECD guidelines on measuring subjective well-being: 2025 update. OECD Publishing. View source
- Przybylski, A. K., & Weinstein, N. (2017). A large-scale test of the Goldilocks hypothesis: Quantifying the relations between digital-screen use and the mental well-being of adolescents. Psychological Science, 28(2), 204–215. View source
- UNICEF Innocenti. (2025). Childhood in a digital world. View source