The evidence on disconnecting

A considered reading of fifteen years of research on phones, attention, sleep, and mood — and what the studies actually support, as opposed to what the headlines imply.

---

There is a particular kind of essay about phones that I would like this one not to be. It is the essay that opens with a statistic about average screen time, pivots to moral alarm, gestures vaguely at "the science", and concludes that you should touch more grass. Such essays are everywhere. They are usually sincere. They are rarely useful, because they overclaim in one direction and then get corrected, reasonably, by researchers who think the evidence is more mixed than the alarm suggests.

The truth is that the research on phone use, attention, and wellbeing is genuinely complicated. Parts of it are strong, replicated, and boring in the way that good findings often are. Parts of it are weak, contested, or so methodologically compromised that the published effect sizes should not be taken literally. A careful reading needs to separate the two, and a careful reader deserves to be told which is which.

What follows is my attempt. It is long, which is deliberate. The topic has been discussed at enough superficial length for a generation, and shortcuts have done most of the damage.

The self-report problem

Before touching any of the findings on what phones do to people, there is a methodological issue that has to be named: most of the research on phone use is built on self-reported estimates of how much time people spend on their phones, and those estimates are wrong. Not in the sense of being slightly off. In the sense of being dramatically, almost randomly wrong.

In 2021, Douglas Parry and colleagues published a systematic review and meta-analysis in Nature Human Behaviour comparing self-reported digital media use to device-logged use across 106 studies (Parry et al., 2021). They found that only about 5% of self-reports matched actual logs with reasonable accuracy. Correlations between what people said they did and what their devices recorded were moderate at best, often weak, and systematically biased: heavy users tended to underestimate, light users to overestimate, and everyone to answer with whatever felt socially defensible in the moment.

This matters because a very large share of the published literature on screen time and wellbeing relies on self-report. When someone tells a researcher that they use their phone "about three hours a day" and the phone itself would have logged five and a half, the resulting correlation with, say, depression scores does not describe phone use and depression. It describes the overlap between people's beliefs about their phone use and depression, which is a different and considerably noisier thing.

This is not a reason to dismiss the field. It is a reason to weight studies carefully. Findings from passively logged data, RCTs that manipulate phone use directly, and large longitudinal cohorts deserve more credence than cross-sectional surveys that ask people to estimate their own hours. Wherever possible in what follows, I have tried to cite the stronger designs.

The cognitive cost of proximity

Perhaps the most striking finding in the recent literature concerns what happens when a phone is merely present, not in use. In a 2017 study published in the Journal of the Association for Consumer Research, Adrian Ward and colleagues at the University of Texas ran a series of experiments in which participants completed standard measures of working memory and fluid intelligence (the OSpan working-memory task and a set of items from the Raven's Progressive Matrices). The experimental manipulation was trivial: where the participant's own smartphone sat during the task. On the desk face down. In a pocket or bag. Or in another room entirely.

The phones were silenced throughout. Participants did not use them. They did not receive notifications. They did nothing at all with the phone except be near it.

The result was a gradient. Participants whose phones were in another room performed meaningfully better on the cognitive tests than those whose phones were on the desk, with "in a bag" landing in between. Ward et al. named the effect "brain drain" — the hypothesis being that the mere presence of the phone was consuming a small but measurable fraction of attentional resources, simply by existing as a salient stimulus the brain had to actively inhibit (Ward, Duke, Gneezy, & Bos, 2017).

It is a hard finding to dismiss. The manipulation is minimal, the design is clean, and the effect replicates across samples. It also has uncomfortable implications. Most people who are "not using their phone" at a given moment are still near it. The phone-on-the-desk-while-you-work scenario is the overwhelmingly typical one, and if the Ward results generalise — which further work suggests they do, with some boundary conditions — the cost of modern phone-proximity is not an occasional interruption but a continuous, low-grade drag on cognitive performance.

A related finding by Cary Stothart and colleagues at Florida State University found that receiving a phone notification, even without checking it, measurably degraded performance on a sustained attention task. The effect was comparable in size to actually reading a text message (Stothart, Mitchum, & Yehnert, 2015). Again, the cost was imposed by the signal, not by the action of responding to it.

Collectively these studies make an uncomfortable point. The common intuition — I am not using my phone, therefore my phone is not affecting me — appears to be wrong. Presence is not neutral. Silence is not absence.

Attention fragmentation, measured longitudinally

The most important single researcher on what phones and screens have done to sustained attention is probably Gloria Mark, a computer scientist at the University of California, Irvine, who has been measuring attention behaviour in workplace settings for more than two decades.

Her early work, conducted with colleagues at UC Irvine and Humboldt University of Berlin, instrumented knowledge workers and measured how long they sustained attention on any one screen before switching. Her 2008 paper with Gudith and Klocke documented the cost side of this — that interrupted work was completed faster but at the expense of measurably higher stress, frustration, and effort (Mark, Gudith, & Klocke, 2008). The initial average attention span her team recorded, in around 2003, was approximately two and a half minutes.

Over the following two decades, Mark and her collaborators repeated these measurements with successive cohorts, and the averages did not hold steady. By 2012, sustained attention on a single screen had fallen to around seventy-five seconds. By the early 2020s, Mark's updated longitudinal data, summarised in her 2023 book Attention Span, reported an average of around forty-seven seconds. In roughly twenty years, the duration of uninterrupted single-screen attention had contracted to about a third of what it had been (Mark, 2023).

Two things about this finding deserve emphasis. First, the decline is not a matter of self-report; it is behavioural, measured on the device itself. Second, the decline does not appear to be a generational artefact. It holds across age cohorts and holds within individuals over time. The most parsimonious reading is that it reflects a property of the environment — specifically, an environment in which applications compete aggressively for each next glance — rather than a property of the people living in it.

Sophie Leroy, then at the Carlson School of Management at the University of Minnesota, identified a complementary mechanism in 2009. When we switch tasks, a portion of our attention does not come with us; it remains, briefly, oriented toward the previous task. Leroy called this attention residue, and showed in a series of experiments that the residue measurably impaired performance on the subsequent task until it dissipated (Leroy, 2009). If each of Mark's ~47-second attention episodes is followed by a brief residue before the next can proceed cleanly, the net effective time spent on any one cognitive thread shrinks still further.

Put together, these findings sketch a picture of modern attention as highly fragmented and serially contaminated. This does not mean people are unable to think. It means the conditions under which sustained thought occurs have become genuinely rare, and the effort of creating them — deliberately — has gone up.

Sleep: the strongest single finding

If I were asked to name the single most robust finding in the entire literature on screens and health, it would probably be the effect of evening phone use on sleep. The mechanisms are multiple and additive, and the effect sizes are meaningful.

The best-known mechanism is the physiological one. Evening light exposure, particularly in the short-wavelength blue range that smartphone screens emit, suppresses melatonin secretion, delays the circadian phase, and pushes sleep onset later. Anne-Marie Chang and colleagues at Harvard Medical School demonstrated this directly in a widely cited 2015 study published in PNAS. Healthy adults reading from an iPad for four hours before bed, compared with reading a printed book, showed significantly reduced evening melatonin, longer sleep latency, reduced REM sleep, and next-morning alertness deficits that persisted hours into the following day (Chang, Aeschbach, Duffy, & Czeisler, 2015).

The second mechanism is behavioural rather than physiological. Phones in the bedroom extend the window in which the mind is engaged with external input. A 2016 PLoS ONE study by Christensen and colleagues at UCSF, using objective phone-logging, found significant associations between higher evening smartphone screen time and reduced sleep quality, longer sleep latency, and worse sleep efficiency in a large adult sample (Christensen et al., 2016).

Lauren Hale and Stanford Guan, at Stony Brook, published a systematic review in Sleep Medicine Reviews covering 67 studies of screen time and sleep in school-aged children and adolescents. 90% of the reviewed studies reported adverse associations between screen use and at least one sleep outcome, most commonly later bedtimes, shorter sleep duration, and poorer sleep quality (Hale & Guan, 2015).

The third mechanism, which has received comparatively less attention but which clinicians consistently flag, is that phones make it too easy to resume engagement in the middle of the night. A waking at 3am, which twenty years ago would have been resolved by rolling over, is now often resolved by picking up the device and scrolling for forty minutes.

The practical implication is one of the strongest in the whole literature: the phone should not be in the bedroom. This is not a matter of willpower or digital-wellness virtue. It is the direct consequence of three independent, replicated lines of evidence. Buy a cheap alarm clock. Charge the phone in the hallway. The next morning that you wake up without reaching for a screen is, on the evidence, measurably better than the one before.

Mood and anxiety: the contested middle

Here the evidence becomes more complicated, and this is the part of the literature where most popular writing goes wrong in both directions.

On one side, Jean Twenge at San Diego State University has argued since 2017 that the rapid rise of adolescent smartphone ownership — which reached saturation between roughly 2012 and 2014 — coincided with a sharp rise in depression, anxiety, and suicidality among American adolescents, particularly girls, and that the coincidence is substantive rather than coincidental (Twenge, Martin, & Spitzberg, 2019). Jonathan Haidt has extended and popularised this argument in The Anxious Generation (2024), framing the phone-based childhood as the single most plausible cause of a generational mental health decline.

On the other side, Amy Orben and Andrew Przybylski at the Oxford Internet Institute have argued, from the same datasets, that the effect sizes being cited are extremely small — often no larger, in comparable terms, than the association between teenage wellbeing and eating potatoes — and that the causal inferences being drawn from them are unwarranted. Their 2019 Nature Human Behaviour paper argued that the correlation between adolescent digital technology use and wellbeing accounts for approximately 0.4% of the variance in wellbeing, a magnitude that should not support the confident claims being made about it (Orben & Przybylski, 2019). Their subsequent work, including a 2019 paper in PNAS using large longitudinal samples, reinforced that conclusion (Orben, Dienlin, & Przybylski, 2019).

Both sides have a point. The Twenge–Haidt position is strongest on the inflection. The timing of the adolescent mental health deterioration is genuinely striking, and it aligns closely with the transition from feature phones to smartphones plus the launch of image-based social platforms. The Orben–Przybylski position is strongest on the individual associations. Within any given cross-section, the statistical relationship between daily screen-time hours and wellbeing is small, and many heavy users are fine.

There is a reasonable synthesis. The small individual-level effects and the large population-level inflection are compatible if a particular subset of users — disproportionately adolescent girls with particular vulnerabilities, using particular platforms at particular developmental windows — is experiencing large effects while the average across all users is diluted toward zero. Several recent analyses, including work by Kelly and colleagues on the UK Millennium Cohort Study, support something like this heterogeneous pattern (Kelly, Zilanawala, Booker, & Sacker, 2019).

The strongest direct evidence for a causal effect of social media on mood comes from randomised experiments in which people are paid to reduce or stop using a platform. Melissa Hunt and colleagues at the University of Pennsylvania ran a three-week trial in which undergraduates were randomly assigned to limit their Facebook, Instagram, and Snapchat use to ten minutes per platform per day, or to continue as usual. The limited-use group reported significant reductions in loneliness and depression over the three weeks, with larger effects among participants who had started with the higher baseline symptoms (Hunt, Marx, Lipson, & Young, 2018).

A larger and economically more sophisticated RCT by Hunt Allcott and colleagues, published in the American Economic Review, recruited 2,743 adult Facebook users and paid a subset of them to deactivate their accounts for four weeks ending just after the 2018 US midterm elections. The deactivated group reported significantly increased subjective wellbeing, reduced political polarisation, reduced factual news knowledge, and — on returning to the platform after the study — spent on average about 23% less time on it than they had before deactivation (Allcott, Braghieri, Eichmeyer, & Gentzkow, 2020). The wellbeing effects were, in the authors' own conservative framing, modest but real.

Reading all of this together, the most defensible position is this. For most adults, on average, moderate smartphone use is probably not doing substantial harm to mood. For some users, in some conditions — heavy users, adolescent users, users of particular platforms at particular developmental stages — the effects are larger, and in the aggregate they appear to be real. The intervention studies that actually cut use consistently find wellbeing improvements, which is the most direct kind of evidence available.

The headline ought not to be "phones destroy your mental health". It also ought not to be "phones are fine". It is: cutting back consistently helps, the benefit is larger for heavier users, and the strongest effects are in populations nobody should want to experiment with lightly.

Memory and offloading

A smaller but genuinely interesting strand of research concerns what phones do to memory. The seminal paper here is Betsy Sparrow and colleagues' 2011 Science paper, widely referred to as the study of "Google effects" (Sparrow, Liu, & Wegner, 2011). In a series of experiments, the authors showed that when participants expected information to be accessible later — because they believed it was saved in a computer file — they remembered the information itself less well, but remembered the location where it could be found more accurately. The memory had not disappeared; it had been reorganised. The brain offloaded the content to the external store and held on only to the pointer.

This is not new behaviour. Humans have been offloading memory onto artefacts for millennia — shopping lists, phone books, diaries. The phone accelerates and extends the tendency. What is worth noticing is that the offloading is not free. When the information being offloaded is trivial, the trade-off is fine. When it is information that would have been useful for other cognitive work — connecting ideas, noticing patterns, making arguments — the offloading has a hidden cost that is difficult to measure directly but is very hard to argue does not exist.

This strand of the literature is less developed than the attention or sleep work, and the philosophical implications — for what it means to "know" something in an environment of ubiquitous retrieval — have been discussed more in philosophy than in empirical psychology. It deserves more of the latter.

The social cost of the phone on the table

A quietly devastating paper by Andrew Przybylski and Netta Weinstein at the University of Essex, published in 2013, demonstrated that the mere presence of a phone on the table during a two-person conversation measurably reduced both the quality of the conversation and the reported closeness of the relationship afterwards (Przybylski & Weinstein, 2013). The phones were not used. They were simply present. Participants were not told the study was about phones.

This finding has been replicated, with some methodological debate, by several subsequent groups, and the effect size is not enormous. But the direction is robust and the mechanism is plausible: a device that both parties know can interrupt the interaction at any moment imposes a small but continuous tax on how fully present either participant can be.

The colloquial term phubbing — phone-snubbing — has been used for the broader behaviour, and it has been associated in survey studies with reduced relationship satisfaction among couples and with small but measurable decreases in the perceived warmth of parent–child interactions. These findings should be held with the caveats above about self-report, but they are consistent with the Przybylski and Weinstein result and with common experience.

The practical conclusion is undramatic and direct. When a conversation matters, the phone should not be on the table. This is one of the cheapest interventions in the entire literature, and it appears to have meaningful effects on the quality of the social life it is applied to.

Creativity and the brain's default network

I have written a longer essay on this elsewhere on this site — Your brain on boredom — but a brief summary belongs here, because it completes the picture.

The brain contains a large-scale network, the Default Mode Network, that is most active when attention is not being directed outward toward a specific task. It is the neural substrate of autobiographical memory consolidation, future simulation, social cognition, and creative incubation (Raichle et al., 2001; Buckner, Andrews-Hanna, & Schacter, 2008). Baird and colleagues at UC Santa Barbara showed that participants performed significantly better on creative problem-solving tasks after a period of undirected mind-wandering than after a period of rest or focused effort (Baird et al., 2012). Sandi Mann and colleagues at the University of Central Lancashire demonstrated experimentally that exposure to boredom prior to a creative task increases creative output — the brain, starved of external input, begins to generate its own (Mann & Cadman, 2014).

The implication for phones is straightforward. Every moment of incidental idleness that is filled by scrolling is a moment in which the DMN had been preparing to do some of its most useful work, and in which that work is instead displaced by an external input of marginal value. Over thousands of such moments a week, the cumulative cost is not obviously zero.

A synthesis

Taking all of the above together, the shape of the evidence is clearer than the noisy middle of the literature might suggest.

Strong findings. The presence of a phone imposes a small but reliable cognitive cost, even when it is not being used (Ward et al., 2017; Stothart et al., 2015). Sustained attention, measured behaviourally and longitudinally, has contracted markedly over the past twenty years in environments saturated with interruptions (Mark, 2023). Evening phone use measurably impairs sleep through multiple independent mechanisms (Chang et al., 2015; Hale & Guan, 2015; Christensen et al., 2016). Reducing social media use, in randomised trials that actually manipulate use rather than observe it, produces measurable improvements in wellbeing (Hunt et al., 2018; Allcott et al., 2020).

Medium-strength findings. Correlational studies of phone use and adolescent mental health reveal a mixed picture that is nevertheless consistent with heterogeneous effects: small on average, larger among certain subgroups, plausibly real (Twenge et al., 2019; Orben & Przybylski, 2019; Kelly et al., 2019). The presence of a phone during in-person conversation reduces both conversational quality and reported closeness (Przybylski & Weinstein, 2013).

Weaker or still-emerging findings. The effects of information offloading on long-term memory and self-assessment are real but incompletely characterised (Sparrow, Liu, & Wegner, 2011). The relationship between notification frequency, dopaminergic reward learning, and habit formation in phone use is discussed widely but rests on a smaller primary literature than the popular framing implies.

What the evidence does not support. A confident, monocausal claim that phones are the sole or primary driver of contemporary unhappiness. The actual evidence is more specific and in some respects more useful: phones are a large factor in some contexts, a small factor in others, and there are concrete, narrowly targeted things that help.

What to actually do, given the evidence

Here, finally, is the practical part, derived directly from the strongest findings above. None of it is original; most of it is the boring, unromantic implication of the research.

1. Remove the phone from the bedroom. This is the single highest-confidence intervention in the entire literature. Evening screen use impairs sleep through multiple independent mechanisms, and sleep is upstream of almost everything else the brain does.

2. Put distance between yourself and the device during focused work. The Ward et al. (2017) results suggest that the physical location of the phone during a cognitive task has a real, measurable effect. Another room is better than a drawer; a drawer is better than a pocket; a pocket is better than the desk.

3. Take the phone off the dinner table, and out of real conversations generally. The Przybylski and Weinstein (2013) finding is one of the cheapest, most consequential changes you can make in social life. It costs nothing.

4. Reduce, don't eliminate, use of the highest-friction social platforms. The RCTs that find wellbeing improvements (Hunt et al., 2018; Allcott et al., 2020) are consistent in suggesting that even moderate reductions — capping use, deactivating for a period, or removing the app from the phone — have measurable benefits. Total abstinence is not required.

5. Allow occasional, deliberate boredom. The Default Mode Network research suggests that the incidental empty moments you used to have — the walk, the queue, the waiting room — were doing cognitive work you benefited from without noticing. Reclaiming even a fraction of those moments is free and has a meaningful upside.

6. Trust objective over subjective measures. Phones underestimate their own impact when you are the one reporting it. If you want to know how much you use your phone, look at the screen-time tracker. The number will probably be larger than the one in your head.

A note on certainty

I have tried throughout to be specific about what the evidence does and does not support, because nothing about this topic is served by overclaiming. Future research will likely sharpen some of the findings I have reported, and probably soften others. The best current reading of the literature is the one above, but it is a current reading, not a final one.

What I am reasonably confident about is this. The ambient sense that something is being lost in the constant availability of screens is not a moral panic. It maps, in places imprecisely and in places with unsettling precision, onto a body of research that has been accumulating for fifteen years and whose broad findings are converging rather than diverging. The best response to the research is not dramatic — it rarely is — but it is more than nothing, and the interventions that are best supported are also the simplest.

Put the phone in another room for a while. The rest mostly follows from that.

---

References

• Allcott, H., Braghieri, L., Eichmeyer, S., & Gentzkow, M. (2020). The welfare effects of social media. American Economic Review, 110(3), 629–676.
• Baird, B., Smallwood, J., Mrazek, M. D., Kam, J. W. Y., Franklin, M. S., & Schooler, J. W. (2012). Inspired by distraction: mind wandering facilitates creative incubation. Psychological Science, 23(10), 1117–1122.
• Buckner, R. L., Andrews-Hanna, J. R., & Schacter, D. L. (2008). The brain's default network: anatomy, function, and relevance to disease. Annals of the New York Academy of Sciences, 1124(1), 1–38.
• Chang, A. M., Aeschbach, D., Duffy, J. F., & Czeisler, C. A. (2015). Evening use of light-emitting eReaders negatively affects sleep, circadian timing, and next-morning alertness. Proceedings of the National Academy of Sciences, 112(4), 1232–1237.
• Christensen, M. A., Bettencourt, L., Kaye, L., Moturu, S. T., Nguyen, K. T., Olgin, J. E., Pletcher, M. J., & Marcus, G. M. (2016). Direct measurements of smartphone screen-time: Relationships with demographics and sleep. PLoS ONE, 11(11), e0165331.
• Haidt, J. (2024). The Anxious Generation. Penguin Press.
• Hale, L., & Guan, S. (2015). Screen time and sleep among school-aged children and adolescents: A systematic literature review. Sleep Medicine Reviews, 21, 50–58.
• Hunt, M. G., Marx, R., Lipson, C., & Young, J. (2018). No more FOMO: Limiting social media decreases loneliness and depression. Journal of Social and Clinical Psychology, 37(10), 751–768.
• Kelly, Y., Zilanawala, A., Booker, C., & Sacker, A. (2019). Social media use and adolescent mental health: Findings from the UK Millennium Cohort Study. EClinicalMedicine, 6, 59–68.
• Leroy, S. (2009). Why is it so hard to do my work? The challenge of attention residue when switching between work tasks. Organizational Behavior and Human Decision Processes, 109(2), 168–181.
• Mann, S., & Cadman, R. (2014). Does being bored make us more creative? Creativity Research Journal, 26(2), 165–173.
• Mark, G., Gudith, D., & Klocke, U. (2008). The cost of interrupted work: More speed and stress. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '08), 107–110.
• Mark, G. (2023). Attention Span: A Groundbreaking Way to Restore Balance, Happiness and Productivity. Hanover Square Press.
• Orben, A., Dienlin, T., & Przybylski, A. K. (2019). Social media's enduring effect on adolescent life satisfaction. Proceedings of the National Academy of Sciences, 116(21), 10226–10228.
• Orben, A., & Przybylski, A. K. (2019). The association between adolescent well-being and digital technology use. Nature Human Behaviour, 3, 173–182.
• Parry, D. A., Davidson, B. I., Sewall, C. J. R., Fisher, J. T., Mieczkowski, H., & Quintana, D. S. (2021). A systematic review and meta-analysis of discrepancies between logged and self-reported digital media use. Nature Human Behaviour, 5(11), 1535–1547.
• Przybylski, A. K., & Weinstein, N. (2013). Can you connect with me now? How the presence of mobile communication technology influences face-to-face conversation quality. Journal of Social and Personal Relationships, 30(3), 237–246.
• Raichle, M. E., MacLeod, A. M., Snyder, A. Z., Powers, W. J., Gusnard, D. A., & Shulman, G. L. (2001). A default mode of brain function. Proceedings of the National Academy of Sciences, 98(2), 676–682.
• Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google effects on memory: Cognitive consequences of having information at our fingertips. Science, 333(6043), 776–778.
• Stothart, C., Mitchum, A., & Yehnert, C. (2015). The attentional cost of receiving a cell phone notification. Journal of Experimental Psychology: Human Perception and Performance, 41(4), 893–897.
• Twenge, J. M., Martin, G. N., & Spitzberg, B. H. (2019). Trends in U.S. adolescents' media use, 1976–2016: The rise of digital media, the decline of TV, and the (near) demise of print. Psychology of Popular Media Culture, 8(4), 329–345.
• Ward, A. F., Duke, K., Gneezy, A., & Bos, M. W. (2017). Brain drain: The mere presence of one's own smartphone reduces available cognitive capacity. Journal of the Association for Consumer Research, 2(2), 140–154.