What Longitudinal Data Would Actually Show About Screens and Loneliness

Jonathan Haidt’s The Anxious Generation, published in March 2024, makes a causal claim: the shift from play-based to phone-based childhood in the early 2010s produced a measurable rise in adolescent loneliness, anxiety, and depression. The claim has drawn sustained pushback, and the collaborative evidence review that Haidt and Jean Twenge maintain alongside the book includes dissenting studies rather than suppressing them. That is unusual for an advocacy document, and it is worth taking seriously precisely because it makes the disagreement legible.

Most of that disagreement traces back to a single design problem. The bulk of the evidence linking social media use to adolescent loneliness is cross-sectional: a survey asks teenagers how much time they spend on a platform and how lonely they feel, at one point in time, and reports the correlation. That design cannot distinguish three very different stories that would produce an identical result.

Three stories, one correlation

The first story is that heavy platform use causes loneliness — displacing in-person contact, disrupting sleep, or triggering social comparison that erodes a sense of belonging. The second is the reverse: adolescents who are already lonely, for reasons unrelated to their phones, turn to social media as compensation, and the loneliness was there first. The third is that some third factor — family instability, a diagnosed mental health condition, a shift in neighborhood or school — drives both variables independently, so that phone use and loneliness rise together without either causing the other.

A cross-sectional survey, no matter how large, cannot adjudicate among these. It can establish that the association exists and estimate its size. It cannot say which way the arrow points, or whether there is a common cause upstream of both. The Harvard Making Caring Common report from February 2021 found that 61% of young adults aged 18 to 25 reported serious loneliness, and that 43% said their loneliness had increased since the pandemic began — figures drawn from a single wave of measurement. They describe a moment, not a trajectory, and the report does not claim otherwise.

A longitudinal design solves this differently, though not perfectly. By measuring the same individuals repeatedly over time, it can establish temporal precedence — whether a change in platform use at time one predicts a change in loneliness at time two, after accounting for the loneliness level already present at time one. That is a meaningfully stronger claim than a snapshot correlation, though it still falls short of a randomized experiment, since unmeasured confounders that also change over time can produce the same pattern.

What the small number of longitudinal studies actually show

The evidence base here is thinner than the volume of commentary on the topic would suggest. The BMC Public Health review of the state of loneliness research, published in June 2023, singles out inconsistent measurement as a structural barrier to the field — different studies use different loneliness instruments, different social media exposure measures, and different follow-up windows, which makes even the longitudinal studies that exist difficult to compare or pool.

One useful exception is the JACSIS study out of Japan, which tracked isolation and loneliness prevalence across 2020 and 2021 using a repeated national sample rather than a one-off survey. Because it followed the same population structure over two points during a period of dramatic behavioral change, it can speak to how isolation and loneliness moved together — and moved apart — during the pandemic in a way a single cross-sectional wave cannot. It does not focus on adolescent social media use specifically, but it demonstrates the kind of design that the phone-and-loneliness debate needs more of: repeated measurement of a defined population through a period of known exposure change.

A second relevant thread comes from research that treats isolation and loneliness as genuinely separate constructs rather than interchangeable terms. The Scientific Reports study on isolation, age, and loneliness during the pandemic found that the relationship between the two varies by age group — meaning a teenager who is structurally isolated (few peer contacts, little in-person activity) is not necessarily the same teenager who reports subjective loneliness, and the gap between the two categories itself changes across the life course. This matters for the phone debate because most of the discussion collapses “time on a screen” into a single isolation proxy and “reported unhappiness” into a single loneliness proxy, when the mechanism Haidt proposes actually implies a specific sequence: displacement of in-person contact (a change in isolation) preceding a change in subjective loneliness. A longitudinal design built to test that sequence would need to measure both constructs separately, at multiple time points, in the same adolescents — a design that, as far as the current literature shows, has not yet been executed at scale for this question.

The Frontiers in Psychology study on college students coming out of the pandemic offers a partial, indirect data point. Rather than measuring platform time, it asked which kinds of social connection predicted wellbeing among students who had lived through both extremes — pandemic isolation and a return to in-person contact. The finding relevant here is not about social media at all: it is that the quality and regularity of contact, not the sheer count of interactions, predicted wellbeing. If that generalizes to the mechanism Haidt proposes, it suggests any longitudinal test of phone use and loneliness should track not just hours on a platform but what kind of contact those hours are displacing.

What a genuinely adequate study would look like

Following the same set of adolescents from before smartphone acquisition through several years afterward, measuring platform use with something more precise than self-reported hours (screen-time logs, ideally), and measuring loneliness and isolation as separate constructs at each wave, would let researchers test whether a change in use precedes a change in loneliness within individuals — the pattern the causal story predicts — rather than merely coexisting with it across a population. It would also let researchers separate adolescents who substitute screen time for in-person contact from those who add it on top of an unchanged social schedule, since the displacement story only holds for the former group.

No such study currently exists in the literature the Center has reviewed. What exists is a set of cross-sectional snapshots, a handful of shorter panel studies not designed around this specific question, and a public debate that has outrun the design of the evidence supporting it. Haidt’s collaborative evidence review is candid about including dissenting studies, which is more than most single-author positions in this space offer. But candor about the disagreement is not the same as resolution of it, and the resolution requires a study design that, at the moment, has not been built.