Methodology

Data Sources

Percentile data are drawn from peer-reviewed scientific studies and large-scale population health surveys. Primary sources include:

All sources are listed on the references index.

Population Context

Several metrics, including BMI and body fat percentage, use data from the US population (NHANES). NHANES is one of the largest, most rigorously collected, and most frequently updated population health datasets available, with measured (not self-reported) values and nationally representative sampling.

However, the US has one of the highest obesity rates among major countries. US percentile distributions therefore skew higher than most other populations — the 50th percentile reflects the US median, not a health threshold.

Where this matters, context is added directly on the metric pages:

  • BMI: WHO category bands are overlaid on charts so bars visually land in Underweight/Normal/Overweight/Obese zones. An international comparator note references mean BMI by country from the NCD-RisC (Lancet, 2024).
  • Body fat percentage: A caveat notes that Americans carry 5 to 10 percentage points more body fat than European populations at the same age and BMI.
  • Blood pressure: Systolic and diastolic BP percentiles are from NHANES 2001-2008. Age-standardized mean BP and hypertension prevalence vary substantially across countries (NCD-RisC, Lancet 2021), so the upper percentiles may not generalize globally. Each overview page notes this caveat.
  • 30-Second Chair Stand: Normative data is from US community-dwelling older adults (Rikli & Jones, n=7,183). A German study (n=1,657, ages 65-75) found lower scores in comparable age groups, partly attributed to higher body weight.

Where international comparator data exists, it is linked on each metric's reference page.

Percentile Calculation

Five percentile points are reported for each metric: 5th, 25th, 50th (median), 75th, and 95th. Depending on the source study, these values are derived by one of the following methods:

  • Directly reported — the source study publishes the exact percentile values shown.
  • LMS curve fitting — the source provides L (skewness), M (median), and S (coefficient of variation) parameters by age; percentiles are computed using the standard LMS formula. Used for lean mass index and appendicular lean mass index (Kelly et al. 2009).
  • Normal distribution — percentiles calculated from reported means and standard deviations, assuming a normal distribution. Used where the source reports summary statistics but not full percentile tables.
  • Percentile proxy — when P5, P25, P75, or P95 is unavailable, the nearest reported value is substituted (e.g. P10 for P5, P20 for P25). Documented per metric.
  • Interpolation — when reported percentiles bracket the target (e.g. P20 and P30 available but not P25), linear interpolation is applied: P25 = (P20 + P30) / 2. Used for Tomkinson 2017/2018 youth norms.
  • Calculator-derived — where the study authors publish an interactive calculator exposing the full distribution, quartile values are read directly from it. Used for the powerlifting lifts (van den Hoek 2024, via thestrengthinitiative.com).
  • Category-boundary estimate — for tests scored in discrete performance categories (e.g. push-up norms by fitness category), category boundaries are mapped to approximate percentile positions.
  • Derived from microdata — where no published percentile table exists, percentiles are derived from publicly available raw survey data. Used for waist-to-height ratio (NHANES). See derivation details.
  • Equation-derived — where the source study publishes regression equations rather than pre-tabulated percentiles, those equations are evaluated at representative age, height, and sex values. Used for FEV1 (GLI-2012). See derivation details.

Each metric's reference page documents which method applies. When multiple studies report data for the same demographic, priority is given to larger sample sizes and more recent publication dates.

Derived Datasets

For some metrics, no published study provides pre-tabulated age- and sex-stratified percentile tables. Percentiles are instead computed from published regression equations or derived from public microdata. Full derivation details are documented on each metric's methodology page:

Age Brackets

Age brackets match the source study's native groupings to preserve data fidelity. Most metrics use decade brackets (20-29, 30-39, ... 80+), while some use 5-year brackets (20-24, 25-29, ...) when the source data supports finer granularity. The number of brackets varies by metric depending on the age range covered by the primary study.

For trend curves showing how metrics change with age, we use the midpoint of each bracket, treating it as a continuous interval (e.g., 25 for a 20-29 bracket, 22.5 for a 20-24 bracket, 85 for the open-ended 80+ bracket).

Percentile Approximation

Source studies rarely report all five target percentiles (P5, P25, P50, P75, P95) directly. The approximation method varies by source:

  • Tomkinson 2017/2018 (youth norms) reports P10, P20, P30, P70, P80, and P90. P25 is interpolated as (P20 + P30) / 2; P75 as (P70 + P80) / 2. P5 uses P10 as a proxy; P95 uses P90.
  • van den Hoek 2024 (powerlifting) reports deciles P10 to P90. P5 uses P10 as a proxy; P95 uses P90. P25 and P75 are read from the authors' online calculator, which exposes the full distribution.
  • Other sources — where P25 or P75 is absent and no interpolation bracket exists, the nearest available percentile is used as a proxy. Each metric page notes which percentiles are approximated.

Rating System

A five-tier rating system is used based on percentile rankings:

Rating Percentile Interpretation
Excellent 95th Top 5% of the reference population
Above Average 75th Higher than 75% of the reference population
Average 50th Median of the reference population
Below Average 25th Higher than 25% of the reference population
Poor 5th Bottom 5% of the reference population

For lower-is-better metrics (e.g. resting heart rate, reaction time), the scale is inverted: the 5th percentile is rated Excellent and the 95th percentile Poor.

Four body-composition metrics do not use the performance scale above, because no value on them is a good or bad score: waist-to-height ratio, waist-to-hip ratio, body fat percentage, and mid-upper arm circumference. These report where a value sits in the population distribution using a neutral five-tier label: Very low, Low, Average, High, Very high (mapped 5th to 95th percentile). Both very low and very high values can be atypical, so the labels describe position only, not performance or health.

Limitations

  • Population representation: Several metrics use US-only data (NHANES), and the US has unusually high obesity rates. Percentiles for body composition metrics will skew higher than in most other countries. See Population Context above for details and the specific caveats we add to affected pages.
  • Measurement methods: Different studies may use slightly different protocols. For example, VO2 max can be measured directly via metabolic cart or estimated from submaximal tests.
  • Selection bias: Clinical and fitness registry data may over-represent healthier, more active individuals who seek testing.
  • Temporal changes: Population fitness levels change over time. The most recent available data are used, but some studies may be several years old.
  • Special-population norms: Some metrics are norms for a specific subpopulation rather than the general public. The powerlifting lifts (squat, bench press, deadlift) are based on drug-tested competitive powerlifters, a highly trained group whose values are substantially higher than recreational gym-goers. Where this applies, we add a prominent caveat on the metric page and note it on the reference page.

Updates

Data are updated as new large-scale studies become available. Each metric page shows the publication year and sample size of its sources.