Through the Looking Glass: A Systematic Review of Longitudinal Evidence, Providing New Insight for Motor Competence and Health

Children and adolescents first assessed between ages 2 and 18; 43 longitudinal/mediation/intervention studies extracted (32 longitudinal), sport-specific samples excluded

PROSPERO-registered (CRD42020155799) systematic review of longitudinal, mediation and experimental evidence testing every path of the Stodden (2008) developmental model; 585 titles screened, 152 full-text, 43 articles extracted, final search 8 November 2019. Evidence graded by proportion of supportive analyses (0-33% "0" no association; 34-59% "?" indeterminate; >=60% "+/-", upgraded to "++/--" when four or more studies agree; <=3 studies = insufficient). CRUCIAL to reading the numbers: for most pathways the review reports TWO proportions — one over all analyses, and one after deleting the two studies that ran very large numbers of comparisons (Antunes 2016, 72 analyses; Bryant 2014, 60 analyses) — and the two often fall on opposite sides of a rating boundary. The headline verdicts in the abstract follow the post-deletion figure. An earlier version of this record reported the post-deletion figure for the weight-status pathways and the pre-deletion figure for MC->PA, which flattered the paper's consistency in both directions; both proportions are now recorded on every row.

Deflates the popular 'build motor skills -> lifelong physical activity' cascade. The MC->PA path, the model's load-bearing claim, is rated NO ASSOCIATION on all 123 analyses (26%) and only reaches INDETERMINATE after the largest study is deleted (41%); the reverse path PA->MC is null on any reading (8%, still 27% after deletion). But the same pathway contains a strong positive the record must not hide: composite/total MC predicting later PA is 10/14 (71%), rated STRONG POSITIVE "++" — the null is carried by object control skills (1/29, 3%), not by MC as a whole. The other robust paths are MC->fitness (62%, "++"; locomotor/coordination 71%) and weight status->MC, i.e. relationships plausibly driven by shared physical capacity rather than a behavioural cascade; fitness->MC in reverse is indeterminate at best (27% overall, 59% post-deletion). Fitness also acts as a strong mediator between MC and PA in BOTH directions. Only 2 intervention studies exist to test causality and neither showed MC intervention effects on PA. Risk of bias in the 32 longitudinal studies is high on the items that matter: 81% biased on blinding of outcome assessors to exposure, 59% on participation rate of eligible people (<50% or undeterminable), 59% on loss to follow-up (>20% or undeterminable) and 50% giving no sample-size justification. No longitudinal study contained all variables and pathways of the model, so "there is no evidence base to support the model as a whole".

Not modelled — and the authors say so, listing genetics among the variables the review could not assess: "The broad scope of this review meant that we could not assess how other relevant variables (e.g., diet, genetics, cultural settings, growth and maturation, cognition, motivation) related to the core variables in the model." The corpus is longitudinal/observational by design, so selection and passive gene-environment correlation are unaddressed throughout. MC, fitness and PA are all substantially heritable and share genes, so the surviving MC->fitness and total-MC->PA paths are the ones most exposed to pleiotropic confounding.

Caution on independence: the "companion" review usually cited alongside this one (Hill 2023, cognitive and social-emotional outcomes) shares five authors with it — Barnett, Pesce, Valentini, Getchell and Robinson — so it is the same consortium extending to a different outcome domain, not an independent replication. The genuinely independent corroboration is Zi & de Geus 2025, a different author group reviewing 106 reviews/meta-analyses, which argues the residual MC-PA associations are familially confounded.