This page isn’t specific to physical activity — it’s a general reference for reading any health research, including the breakdowns on the Research Explained page. Understanding these few concepts makes it much easier to tell what a study’s numbers actually mean.
Understanding Meta-Analysis
A meta-analysis statistically combines results across multiple studies into one pooled estimate. A forest plot — the typical way results get displayed — shows individual study effect estimates, their confidence intervals, each study’s weight in the pooled result, and the overall pooled estimate.
- Fixed-effect model — assumes all studies are estimating one single underlying true effect.
- Random-effects model — assumes the true effect can genuinely vary across studies.
For heterogeneous PA epidemiology research — where populations, exposures, measurement methods, and outcomes all differ — random-effects models are usually the more appropriate choice.
Heterogeneity — I²
I² estimates the proportion of variability among effect estimates that’s attributable to real between-study heterogeneity, rather than just sampling variation (random noise). A rough traditional interpretation:
- ~25% — low heterogeneity
- ~50% — moderate heterogeneity
- ~75% — high heterogeneity
But I² shouldn’t be read mechanically as just a number to check against a threshold — a high I² is a prompt to ask why studies differ (population, age, measurement method, follow-up duration, and so on), not just a red flag to note and move past.
Relative Risk, Odds Ratio, and Hazard Ratio
These three show up constantly in PA and health research, and they’re easy to blur together — but they’re not interchangeable.
Relative Risk (RR)
RR = risk in the exposed group ÷ risk in the comparison group. An RR of 0.75 means 25% lower risk in the exposed group.
Odds Ratio (OR)
Compares odds rather than probability. An OR of 0.80 indicates 20% lower odds — not technically the same as 20% lower absolute risk, since odds and probability diverge more as an event becomes more common.
Hazard Ratio (HR)
Compares event rates over follow-up time rather than a fixed endpoint. An HR of 0.67 means the exposed group experiences roughly a 33% lower hazard during follow-up, under the model’s assumptions.
Relative vs. Absolute Risk
This distinction is extremely important, and headlines get it wrong constantly. Suppose disease risk is 10% for an inactive group and 7% for an active group.
- Relative risk: 7 ÷ 10 = 0.70 → a 30% relative risk reduction.
- Absolute risk reduction: 10% − 7% = 3 percentage points.
So “30% lower risk” does not mean 30 fewer people out of every 100 will develop the disease — it means 3 fewer people out of 100. Both numbers are true and both are useful, but they answer different questions: relative risk tells you about the strength of an association, while absolute risk tells you how much that association actually matters for an individual or a population. Always check which one a headline or abstract is actually reporting.