The Anxious Generation Argument: What Would Actually Test It

Jonathan Haidt’s The Anxious Generation, published in March, makes a specific causal claim, not a vague one: the replacement of play-based childhood with phone-based childhood in the early 2010s drove a measurable rise in adolescent anxiety, depression, and loneliness. That claim is testable in principle. Whether the evidence currently available tests it is a separate question, and the answer is mostly no.

The mechanism, stated plainly

Haidt’s argument, built substantially on Jean Twenge’s generational survey work, has a timing component and a dose component. Timing: the shift should track the diffusion of smartphones and front-facing social media around 2010-2012, not gradual technology adoption stretching back decades. Dose: within a cohort, heavier users should show worse outcomes than lighter users, and the relationship should hold after accounting for the fact that already-anxious or already-lonely adolescents may simply use their phones differently.

Both components matter, because a correlation between phone ownership and loneliness satisfies neither on its own. A generation could get phones and get lonelier for reasons that have nothing to do with the phones — a labor market, a housing market, a pandemic, a decline in unsupervised free play that predates the smartphone. Haidt argues the timing lines up too precisely to be coincidence. Critics, catalogued in the collaborative evidence review that Haidt and Twenge maintain alongside the book, argue the opposite: that effect sizes in the individual studies are small, that reverse causation (lonely teens retreat to phones) is rarely ruled out, and that global cross-national data do not show the pattern the timing argument predicts.

What the cross-sectional numbers can and cannot show

The Harvard Making Caring Common survey found 61% of Americans aged 18-25 reporting serious loneliness in 2021, against 36% of Americans overall — a striking age gradient. Gallup’s 2023 global survey of 142 countries found the same shape: adults 19-29 report loneliness at 27%, versus 17% among those 65 and older. These figures are consistent with Haidt’s account. They are also consistent with several others. Both surveys are cross-sectional snapshots; neither followed the same people through childhood to see whether smartphone acquisition preceded the loneliness or followed it. A single wave of data cannot distinguish “phones caused this” from “something else caused this, and this generation happens to have phones.”

The American Enterprise Institute’s 2021 friendship survey adds a wrinkle that a phone-only explanation struggles with: the decline in close friendship it documents — the share of men with no close friends rising from 3% in 1990 to 12% by 2021, and to 15% among men specifically — spans adults who were already out of adolescence before smartphones existed. If the mechanism were confined to the phone-based childhood of people born after roughly 1995, it would not obviously predict a friendship decline among men in their forties and fifties. Something broader may be operating, of which the phone-based childhood effect could be one strand among several. The WBUR reporting on American Time Use Survey data similarly finds that face-to-face socializing was declining before the pandemic and continued declining after it, a trajectory too long and too broad to be explained by any single cohort-specific mechanism, including Haidt’s.

None of this refutes the phone-based childhood argument. It shows that the population-level trend data on offer — age gradients in one-time surveys, multi-decade friendship decline across all ages — is the wrong shape of evidence to confirm or disconfirm a claim about a specific cohort and a specific causal pathway. The BMC Public Health review of the loneliness research field in 2023 makes a related point in general terms: the field’s measurement is inconsistent enough that comparing findings across studies, let alone establishing timing and dose within one, remains a basic obstacle.

What would actually test it

A study built to adjudicate Haidt’s mechanism would need three things the current evidence base largely lacks in combination. First, longitudinal tracking of the same adolescents from before smartphone acquisition through several years after, with loneliness and mental health measured repeatedly rather than once — allowing timing to be checked directly rather than inferred from cohort comparisons. Second, a way to separate dose from selection: ideally natural experiments, such as staggered smartphone or social media access by school policy or region, rather than self-reported usage that already-lonely adolescents might inflate or deflate. Third, cross-national replication using countries where smartphone diffusion happened on a different timeline than the United States, to see whether adolescent loneliness rose in lockstep with phone adoption specifically rather than with global trends common to the period regardless of device penetration.

Some pieces of this exist in fragments across the literature Haidt and Twenge’s own review compiles, including dissenting studies that find weak or null effects once confounds are addressed. What does not yet exist is a single design that runs all three tests at once. Until it does, the honest description of where this stands is not that the phone-based childhood hypothesis has been confirmed or debunked, but that it has been argued from evidence that was never built to settle it.