Physical activity is consistently associated with better mental health, particularly lower depression risk — but two related questions get asked here, and they require different study designs.

These shouldn’t be treated as the same question — the sections below focus on the prevention side.

Prospective Evidence: PA and Incident Depression

Schuch et al. conducted a large meta-analysis of 49 prospective studies involving 266,939 individuals. Higher PA was associated with lower odds of developing depression: adjusted OR = 0.83, meaning roughly 17% lower odds compared with lower PA. The association held across age groups — youth (OR ≈ 0.90), adults (OR ≈ 0.78), and older adults (OR ≈ 0.79).

The prospective design matters here: PA was measured before depression developed, which strengthens temporal inference (PA came first). It still doesn’t fully establish causality on its own, but it’s a meaningfully stronger design than a simple cross-sectional comparison.

Dose–Response: PA and Depression

Pearce et al. published a dose-response meta-analysis in JAMA Psychiatry, covering 15 prospective studies, 191,130 participants, 2.1 million person-years, and 28,806 incident depression events. Compared with adults reporting no activity:

Once again, the largest improvement showed up toward the lower end of the activity distribution — the same nonlinear pattern seen throughout PA epidemiology.

Population Impact of PA on Depression

Pearce et al. estimated that if insufficiently active adults reached the recommended activity level, roughly 11.5% of incident depression cases could potentially have been prevented, assuming the association is causal — about 1 in 9 cases. This is a good example of the gap between an individual-level relative risk and a population-level impact: a modest-sounding relative risk reduction can still translate into a meaningful number of prevented cases across a whole population.

An Important Limitation of the Depression Literature

The Pearce dose-response meta-analysis relied on self-reported physical activity, which the authors themselves flagged as a limitation — self-report is vulnerable to recall and social-desirability bias, and they specifically recommended more device-based research. This connects to a broader theme across PA epidemiology: how PA is measured genuinely affects what relationships get observed. (See the Accelerometry page for more on this.)

A Worked Example: MVPA and Depression Meta-Analysis

This next part is my own academic analysis (from a KIN 648 course project) — it’s a useful worked example, not an established pooled estimate from the broader literature.

StudyEffect Estimate
UK BiobankHR 0.67 (0.60–0.74)
TILDAHR 0.57 (0.49–0.66)
Nurses’ Health Study IIHR 0.80 (0.70–0.92)
Vallance / NHANESHR 0.37 (0.20–0.70)
HallgrenHR 0.71 (0.53–0.96)

A random-effects analysis of these studies showed substantial heterogeneity: I² ≈ 75.6%. Across the included studies, greater MVPA was generally associated with reduced depression risk, but the magnitude varied considerably. That high I² suggests the studies differ in ways that matter — population, age, PA measurement method, depression assessment, follow-up duration, exposure definition, and covariate adjustment could all be contributing. (See the Reading the Research page for more on interpreting I² and meta-analysis.)