NHANES (CDC) — derived WHR percentiles
About this reference
No published study provides sex- and age-stratified WHR percentile tables for US adults. NHANES collects both waist circumference (BMXWAIST) and hip circumference (BMXHIP) in the body measures examination, but hip circumference was not measured in the 2011–2016 NHANES cycles, so the WHR pool starts in 2017. FitnessNorms derived WHR percentiles by downloading public XPT microdata files for two NHANES blocks, merging body measures with demographics, and computing weighted empirical quantiles (P5, P25, P50, P75, P95) for each sex × 5-year age bracket from 20–24 through 80+. Two cycles were pooled to improve tail stability: 2017–March 2020 (P_BMX / P_DEMO, weight WTMECPRP, the CDC-prescribed pre-pandemic weight) and 2021–2023 (BMX_L / DEMO_L, weight WTMEC2YR). Pooling divided each cycle's MEC weight by 2; this gives approximately equal cycle contribution because post-filter cycle weight totals are within 0.5% of each other (229.5M vs 230.5M), but the divide-by-N rule is not a normalisation that would force equality if cycle totals diverged. Pooling is a FitnessNorms design choice, not official CDC guidance for this combination of releases. The 2021–2023-alone table was compared against the pooled table at every published quantile (P5, P25, P50, P75, P95), not only at the median. One cell crossed the 0.02 red-flag threshold: female 20–24 P75 differs by 0.022 (0.866 single-cycle vs 0.888 pooled). The other 129 (sex × bracket × quantile) comparisons stayed below the threshold (max 0.018). Pooled is retained as primary because halving every cell's sample size to fix one tail in one cell would degrade the other 25 brackets; the 0.022 deviation is documented in the limitations and FAQ. The 2021–2023-alone output is retained as an internal sensitivity check. Weighted quantiles were computed using linear interpolation on the empirical cumulative distribution (analogue of Type 7). Sampling weights were applied to produce weighted percentile estimates; survey-design standard errors were not computed (this is an internal derivation, not a peer-reviewed publication).
Known limitations
- US population only. Percentiles reflect the US adult population, which has one of the highest obesity rates globally; US WHR distributions sit higher than older reference values from leaner populations. The percentiles are not directly comparable to the WHO 2008 cut-offs (>0.90 men, >0.85 women) because the WHO cut-offs are calibrated for a different waist-measurement protocol (midpoint between the lowest rib and iliac crest) than the NHANES iliac-crest protocol used here.
- This is an internal derivation by FitnessNorms from CDC public microdata, not a peer-reviewed publication. Weighted empirical quantiles are used; survey-design-correct standard errors are not reported.
- Equal-cycle pooling (dividing each cycle's weight by 2) is a FitnessNorms design choice. CDC does not prescribe a pooling method for this specific combination of releases, particularly across the pandemic break. The divide-by-2 rule yields approximately equal cycle contribution here only because post-filter cycle weight totals already happen to be within 0.5% of each other.
- Female 20–24 P75 shows a 0.022 difference between the pooled table (0.888) and the 2021–2023-alone table (0.866), narrowly exceeding the internal 0.02 red-flag threshold. All other quantile cells stayed below the threshold (max 0.018). Pooled is retained as primary; users at the upper-quartile boundary of this cell should treat the value as 0.866–0.888. The discontinuity check in scripts/nhanes_whr_percentiles.py exits non-zero on any red flag that is not in its ACCEPTED_DEVIATIONS allow-list, so future regenerations cannot silently ship pooled values that breach the threshold.
- Hip circumference was not measured in NHANES 2011–2016 cycles, so the pooled WHR sample (n≈13,727) is smaller than the comparable WHtR pool (n≈18,700) and limited to two recent cycles rather than three.
- RIDAGEYR is top-coded at 80 in NHANES public data; the 80+ bracket includes all participants aged 80 and over.
- Cross-sectional design: values reflect population averages at the time of survey, not individual trajectories.