A meta-analysis of longitudinal peer influence effects in childhood and adolescence

Psychological Bulletin 147(7): 719–747, doi 10.1037/bul0000329. Read as the publisher's version deposited open access in the Utrecht University repository (hdl 1874/424961; "Open Access (free)"; Maes is Utrecht-affiliated). Read by researcher R5a, 5 October 2026. Provenance: papers/change/giletta2021_peer_influence_meta.provenance.json.

What was read

• Read in full: all 4,254 lines of pdftotext (reading-order) output of the 30-page file: the repository cover page, abstract, introduction, method, results, Tables 1–2, figure captions, discussion and references.
• Partly unreadable:
  • Table 1 (per-study descriptors) is scrambled by text extraction. It was read but not reliably parsed, and nothing below depends on it.
  • Minus signs. The text layer drops them in places (for example "ranged from .21 to .29"), so the signs below come from the prose.
• Not read: Figures 1–6 as images (PRISMA flow, behaviours, caterpillar plots, age and lag scatterplots, funnel plots); the OSF codebook and data.

What they did

• Search. PsycINFO (June 2018), named authors and reference lists; 5,273 reports screened.
• Inclusion:
  • longitudinal, with target and peer behaviour both measured at baseline;
  • peer affiliations identified by nominations or social-cognitive maps;
  • peers' behaviour reported by the peers themselves or by outside informants, never as the target perceived it;
  • the same measure for target and peer;
  • community youth with a baseline mean age of 18 or less.
• Excluded: stochastic actor-based network models, initiation-only studies, and estimates adjusted for covariates.
• Effect size: the standardised cross-lagged β of peers' behaviour at T1 on the youth's behaviour at T2, controlling for the youth's behaviour at T1, computed from three correlations.
• Model: four-level random effects with robust variance estimation.

Main results (verified)

• Data: 71 reports (62 articles, 9 dissertations; 1978–2017); 60 independent studies, 233 effect sizes, 47,423 participants.
  • Studies: US 34, Europe 15, Canada 6, Asia 5.
  • Baseline mean age 3.95–16.66 (mean 12.48).
  • Lags 3–96 months (mean 13.49, median 12); 80% of effect sizes span 6 or 12 months.
• Mean β = .08 (SE .01, 95% CI [.06, .09]). With consecutive waves only (k = 155), β = .08 [.07, .09].
• Variance shares: sampling 31.77%, within wave 27.17%, between waves 19.68%, between studies 21.38%.
• No moderation by type of behaviour (F(2, 192) = 1.38):
  • externalising .081, internalising .089, academic .053;
  • finer: substance use .09, problem behaviour .07, aggression .07, depression .09, other internalising .08, achievement .05, motivation .07.
• Relationship type: close friend .085 against several friends .066 (n.s.).
• Peer context (F(3, 210) = 2.81, p = .040): affiliations limited to the classroom gave .043, against .092 (grade), .101 (school) and .091 (beyond school).
• Age: no linear (p = .110) or quadratic (p = .808) effect, contrary to the authors' hypothesis of a peak in early adolescence.
• Time lag: a negative linear effect, smaller with longer lags (b ≈ −.026 per year, p = .030; in the multivariate model −.02, p = .034). The quadratic term was not significant.
• No publication bias, and no moderation by publication type, year, region (North America .079, Europe .077, Asia .046), gender, ethnicity or report type.

Limits

• Observational. Influence is inferred; the cross-lagged control may over- or under-correct for selection.
• Narrow coverage. Mostly 6–12-month lags. Only one effect size comes from youth older than 17 at baseline, and few from children under 10.
• Influence of peers' behaviour, not explicit pressure. Peer pressure as such was not measured.
• Inconsistencies found:
  • The abstract says four continents; the method codes three (North America, Europe, Asia).
  • "60 independent studies (reported in 71 studies)" should read 71 reports.

What it means for Kurisutina (inference)

• Lasting peer influence on real behaviour and attitudes is small: about 0.08 SD of change over roughly a year per SD of friends' behaviour. It is similar for drinking, aggression, depression and schoolwork, and weaker over longer lags.
• What lasts is far smaller than what shows in the moment. In-the-moment conformity is 25–37% of answers in Asch-type tasks (Franzen) and 7–33% of the gap in Knoll. That supports a small carry-over setting with decay.
• No age gradient in lasting influence, unlike the lab (Knoll). The in-the-moment age gradient may not carry into lasting change. Age moderation in B6 should not be assumed from lab conformity data alone.