Evidence Desk · Cause or correlation

Teen social media checking is linked to diverging brain responses. Which came first?

A brain-imaging study in the top 1% of its field for citations found habitual checkers' responses rose with age while other teens' fell. The groups were already different at the start.

vizmaya desk · 2 November 2026

The study

One cohort, three scans, a lot of attention

Paper
Maza, Fox, Kwon and colleagues, JAMA Pediatrics 177(2), 160–167
Published
3 January 2023; figure labels corrected 13 February 2023
Design
Longitudinal cohort with functional MRI
Sample
169 adolescents, mean age 12.8 at the first scan
Read the paper The correction

Cited 115 times, with 46 citations in 2026 so far, more than any full year before.

Source: OpenAlex (CC0), read 2 November 2026; 2026 is a partial year. Citations measure attention, not strength of evidence.

How the finding was produced

Checking was measured once. Brains were scanned three times.

At the first wave, each teen reported how often they checked social media. In the scanner, they did a task in which they waited for social feedback, either reward or punishment.

Scans followed roughly a year apart. Checking was never measured again.

Source: Maza et al. 2023, p.163, Results; n = 169 overall, not every teen at every wave.

The exposure

169 teens, one self-report score, three groups

Of the 169, 79 scored 0, 34 scored 1 to 15 and 56 scored above 15.

Answers were recoded and summed into a score from 0 to 54. It is not a phone log, and "low" does not mean a teen never used social media: the lowest answers were recoded to zero.

One figure per teen. Source: Maza et al. 2023, pp.161–163, Self-reported Social Media Use and Results.

The finding in a number

Same brain region, opposite directions

+0.11High checkers' age slope in a left amygdala, posterior insula and ventral striatum cluster. 95% CI 0.04 to 0.18.
−0.12Low checkers' age slope in the same cluster. 95% CI −0.19 to −0.06.

FactFrom about age 12.8 to 14.7, habitual checkers' response to anticipated social feedback in this cluster rose, and non-checkers' fell, by similar amounts.

BenchmarkThe units are the model's activation units, so 0.11 has no everyday or clinical scale. The benchmark is zero, meaning no change with age: both intervals exclude it.

Source: Maza et al. 2023, p.163 and Figure 1 (corrected labels). n = 169.

The effect in context

In all four regions, the groups moved in opposite directions

High checkers' slopes ran from +0.09 to +0.19; low checkers' from −0.10 to −0.13. Every interval excludes zero.

The regions were picked because they showed significant clusters, so these are not four independent tests. No other study measured the same exposure and outcome, so this chart compares the study with itself.

Age-related β with 95% CI (thin line), in model activation units. Brain is a schematic of textbook locations, not the study's cluster coordinates. Source: Maza et al. 2023, p.163, Figures 1–4, with the 13 Feb 2023 label correction. n = 169.

The catch

At age 12, before any follow-up, the groups were already different.

The model estimated a gap between habitual and non-habitual checkers at the first scan in all four regions. So their opposite trajectories cannot tell us whether checking shaped sensitivity, or sensitivity shaped checking.

Source: Maza et al. 2023, p.163 (age-12 group coefficients) and p.165 (Limitations: the authors say direction cannot be determined).

What the design can show

Order and change over time, yes. Cause, no.

It can show

  • What came firstChecking was reported before the second and third scans.
  • Change with ageRepeated scans trace how responses shifted in the sampled teens.
  • Different paths by groupThe trajectories differed with baseline checking.

It cannot show

  • CauseChecking was not assigned, so other differences can drive both.
  • Within-person effectsWith one checking measure, we cannot see whether a teen who checks more later responds differently.
  • HarmA change in activation is not shown to be damage, addiction or poorer attention.

Source: Maza et al. 2023, pp.161–166; vizmaya reading of the design.

The alternatives

Three stories fit the same data

The study cannot tell these apart. Both directions could also run at once; the data do not say how much each contributes.

Social environment and maturation are candidate common causes worth testing, not causes shown in this dataset.

Source: Maza et al. 2023, p.165, Limitations; diagram by vizmaya.

The lens chart

Up the evidence ladder, no trial tests years of checking

Studies on neighbouring questions, from correlational to randomized. Each uses a different measure, so their values cannot share one axis. The top rung, a randomized trial of sustained checking with a brain-imaging outcome, was not found.

Stronger for a causal claim as you go down

  1. Cross-sectionalBrain structure study, NeuroImage, 2026
    7,614 adolescents at one ABCD wave    −4.1 ± 1.2100 × standardized β ± SEM: use time vs cortical thickness, one region
  2. Longitudinal cohortMaza et al. 2023, this study
    169 adolescents, three scans    +0.11 [0.04, 0.18]Age slope, high checkers, left amygdala cluster
  3. Longitudinal, registered reportMiller et al. 2023, Cortex
    Exploratory network analysis    β 0.0067 to 0.025Screen time and change in 4 network measures; none significant
  4. Longitudinal cohort, structureNivins et al. 2024, Scientific Reports
    Four-year follow-up    β −0.03, p = 0.16Baseline use and total cerebellar change
  5. Controlled experimentSherman et al. 2016, Psychological Science
    Likes assigned to photos in one session    No comparable effect sizeImmediate brain response, not years of checking
  6. Randomized trial, sustained checkingNot located
    Targeted search, not a systematic review    NoneThe rung that could show cause

Sources: NeuroImage Table 2; Maza p.163; Miller §3.6.2; Nivins p.6; Sherman Results. Links on the sources slide.

Lens read: cause or correlation Interpretation

Linked, in a known order. Not shown to cause.

How it was produced. Each teen got one checking score, which was then related to three yearly scans. That gives a longitudinal association, not a comparison of randomly assigned habits.

What cause would need. Teens at different checking levels would have to be alike in everything else that shapes social sensitivity: their brain responses before they began checking, their social world, and who missed scans. Self-report error would need handling too.

The strongest alternative. Reverse direction or selection. Teens already more or less tuned to social feedback may have formed different checking habits, and the first scan already shows the groups apart.

The verdict. The repeated scans beat a one-time snapshot, but they cannot turn a baseline difference into a cause.

The other side

A registered report finds little

Miller and colleagues (Cortex, 2023) found screen time did not significantly predict change in four brain-network measures. Adding screen and brain factors barely changed how well models explained eight cognitive tests.

It does not refute Maza: it measured screen time and resting networks, not checking and social-feedback responses, and its network analysis was exploratory.

Accounted covariance (AC). Corrected threshold p < .004. n = 7,559 to 7,635 per test. Source: Miller et al. 2023, Table 3, §3.3.

Since then

Later studies widen the picture, not the causal case

  • Structure over four years

    Nivins et al., Scientific Reports, 2024

    Baseline social media use did not significantly predict total cerebellar change (β = −0.03, p = 0.16). Repeated-exposure models found a small link to the shape of the trajectory.

  • Same team, daily diaries

    Maza and colleagues, Developmental Cognitive Neuroscience

    In 91 teens scanned two years before keeping diaries, the link between daily use and daily affect differed with neural sensitivity. Not an independent replication.

  • A big snapshot

    NeuroImage, 2026

    7,614 adolescents at one wave of the ABCD study: use time was linked to cortical thickness in adjusted models. One time point, so direction stays open.

Where it stands

Mostly correlational, few trials

The annotated research collection this study comes from rates the evidence on attention, cognition and behaviour as "mostly correlational; few RCTs", with causal evidence thin outside smartphone-blocking trials.

InterpretationWhat would move this question, in our reading:

  • Checking measured at every wave, next to every scan, so within-person change can be tested.
  • A randomized trial that reduces checking and measures the same social-feedback brain response.
  • Evidence tying these brain changes to how teens actually feel and function, the link the authors ask for.
  • An independent replication with the same exposure and task. We did not find one.

Go further

Read the evidence yourself

Study: Maza, M. T., Fox, K. A., Kwon, S.-J., et al. (2023). Association of habitual checking behaviors on social media with longitudinal functional brain development. JAMA Pediatrics, 177(2), 160–167. https://doi.org/10.1001/jamapediatrics.2022.4924. Selected from: Social Media and Its Effects on People and Society: An Annotated Research Collection (2026 edition). Citation counts: OpenAlex (CC0).

Sources and claims log

What we checked, and what we could not

25claims logged
23verified
2unverified, kept off the page
0contested
  1. Every number comes from the paper itself (printed pages and Figures 1–4), not the collection's summary. Group labels follow the 13 Feb 2023 correction.
  2. Kept off the page: any claim that checking causes harmful brain development (the design cannot show it), and an exact independent replication (not located).
  3. Not found or not audited: a preregistration or public data for the original study, its supplement, and any adjusted versus unadjusted pair of checking estimates.
  4. No clinical meaning is assigned to the neural coefficients; the paper gives no clinical threshold.
  5. Citation counts read from OpenAlex on 2 November 2026: 115 total, field-weighted impact 49.4, top 1% of its field.
  6. The direction question is open by the authors' own account, not contested between studies.

vizmaya desk · 2 November 2026 · No photographs: the open image libraries we query were unreachable while this story was built.