LabNorms Population Percentiles

Methodology

LabNorms curates population-level percentile data for lab tests and biomarkers. We do not generate original data. Every value is derived from a published, peer-reviewed source. Our role is to select sound sources, compute percentiles from the underlying data, and present them clearly.

What "normative values" means here

Some users search for normative values or normative data. On LabNorms, that term refers to descriptive population distributions and percentiles drawn from published datasets, not healthy, ideal, or prescriptive ranges.

Unit of Output

Each page represents one analyte, one demographic group, and one data source. Percentiles are not pooled across sources unless explicitly stated. Where multiple sources exist for the same analyte, each is presented and attributed separately.

What We Do

For each analyte and demographic group, we identify a suitable published dataset, apply defined and documented inclusion and exclusion criteria, and compute percentile values (P5, P10, P25, P50, P75, P90, P95) from the participant-level microdata. The same computation and filtering approach is applied consistently across pages unless explicitly stated otherwise. The source, criteria, and sample size are stated on every data page. All values can be independently reproduced from the stated source, survey cycle, and exclusion criteria.

Source Selection

We only use sources that are:

  • Peer-reviewed and publicly available
  • Based on a defined population with documented sampling methodology
  • Measured using standardized laboratory methods
  • Of sufficient sample size for the demographic group
  • Representative of a defined measurement context (for example, fasting vs. non-fasting, clinical vs. general population)
  • Free from obvious conflicts of interest

Sources vary by analyte. Each is documented in full on its reference page. No assumption is made that values from different sources are directly comparable.

How Percentiles Are Computed

Percentiles are computed from participant-level microdata using the method stated for the source. For NHANES-based pages, we use survey-weighted quantile estimation and apply the appropriate MEC or fasting-subsample weights so the resulting distributions reflect the reference population defined by the survey design. If a future source uses a different sampling structure or a published methodology that does not call for survey weighting, that will be stated explicitly on the relevant pages.

We report P5, P10, P25, P50 (median), P75, P90, and P95 for each demographic group and survey cycle. We stop at P5 and P95 because percentiles further out, such as P1 or P99, rest on the few participants beyond them, and in the smallest groups that is one or two people.

Download all LabNorms percentiles (CSV) Values at full stored precision; tables on this site may round for display. Licensed under CC BY 4.0.

Confidence Intervals

NHANES selects counties (or groups of small counties) first, then neighborhoods, households and people within them, so participants from the same area resemble each other more than a simple random sample would. The precision of a percentile therefore depends on that design as well as on the number of participants. Each percentile on a slice page carries a 95% confidence interval, except where the assay floor rules one out (see below). Each interval is computed by the Woodruff method (1952), which the CDC uses for percentiles in its Second National Report on Biochemical Indicators of Diet and Nutrition in the U.S. Population (2012), shortened here to the CDC nutrition report. The interval uses the full NHANES design, its strata (SDMVSTRA) and primary sampling units (SDMVPSU), and treats each demographic group as a domain of the whole survey, not as a separate sample. The method is described in the report's appendix on confidence intervals for percentiles, posted on the CDC site as Appendix D: Confidence Interval Estimation for Percentiles.

For each percentile, we estimate the weighted share of the group at or below it and that share's standard error under the survey design. The interval on that share is the percentile's level (5% for P5) plus or minus the standard error times a multiplier from Student's t distribution. Both ends are converted back into values with the same quantile rule as the percentile itself, so a percentile always lies inside its interval. The t multiplier depends on the group's design degrees of freedom: the number of primary sampling units that contain members of the group, minus the number of strata that do. The CSV download gives this number for each row in the design_df column.

When the interval on the share reaches 0% or 100%, the bound is the smallest or largest value in the group. This happens mostly for P5 and P95, in the smallest groups and where many results share the same reported value. Shared values can also make a bound, or both bounds, equal the percentile itself.

Some results fall below the assay's detection limit. NHANES stores these as an imputed value (the detection limit divided by the square root of 2), not a measurement. On the sex hormone pages (androstenedione, estradiol, FSH, LH, progesterone and SHBG) and for CRP, ALT, GGT, fasting insulin and vitamin C, the imputed values lie below every measured result. On their slice pages, a lower bound at or below the imputed value is shown as "assay floor" instead of a number. The CSV columns p5_ci_low to p95_ci_high hold the interval bounds at full stored precision. For an "assay floor" bound, the CSV keeps the computed bound and lists the percentile in the ci_low_at_floor column. When the upper bound is also at the imputed value, the whole interval sits there: the page shows "not estimated (assay floor)" and the CSV leaves both bound cells empty.

Small groups give less stable percentiles, and tail percentiles rest on the few participants beyond them. A percentile from a group with fewer participants than the threshold in the table below is marked † on the page. It is still published, with its interval.

Sample sizes below which a percentile is marked †
Percentile Marked when n is below Threshold source
P5, P95224CDC nutrition report
P10, P90112CDC nutrition report
P25, P7545Derived by LabNorms from the NHANES III rule
P5042Derived by LabNorms from the NHANES III rule

The thresholds come from the NHANES III reporting rule (1995). A percentile at P25 or below, or P75 or above, needs the expected number of participants beyond it (n × 5% for P5) to be at least 8 times the design effect. A percentile between P25 and P75 needs n of at least 30 times the design effect. The CDC nutrition report (2012, page 19) applies the rule with an assumed design effect of 1.4, which gives 224 for P5 and P95 and 112 for P10 and P90. That report publishes no threshold for P25, P75 or P50, so LabNorms applies the same rule and design effect to get 45 for P25 and P75, and 42 for P50. P50's threshold is lower than P25's because the rule tests the middle range separately. The design effect of 1.4 is the CDC nutrition report's assumption, not an estimate from LabNorms data. On four analytes we checked (hemoglobin, fasting glucose, TSH and zinc), the observed design effect at P5 and P95 had a median of 1.5, close to the assumed 1.4. It was above 2 for 27 of the 96 estimates (48 groups, each at P5 and P95), mostly in ages 50-69, and for those estimates the threshold of 224 understates the sample needed.

The mark depends only on a group's sample size, so the CSV has no flag column. Compare its n column with the thresholds above.

Reference Populations

The reference population varies by analyte. Some pages use the full survey population for a demographic group; others apply additional filters (for example, excluding pregnant participants or restricting to fasting samples). Criteria are chosen to match the measurement protocol of the source dataset. There is no universal reference definition applied across all analytes. Exclusion criteria are listed on each data page.

What a Percentile Means

A value at P50 is the median: half the reference population measured above it, half below. A value at P5 means 95% of the reference population measured higher.

Percentiles reflect measured values, not biological absolutes. A result is influenced by pre-analytic conditions (fasting status, sample handling, time of day), assay method, and calibration. The distributions here reflect those real-world conditions as documented in the source study.

Percentiles describe distributions. They do not define what is healthy, normal, or optimal. The same value can carry different clinical significance depending on context, symptoms, and other factors only a clinician can assess.

Thyroid Panel Data Source

Thyroid hormone analytes (TSH, total T4, free T4, total T3) use a different source than other analytes on this site. NHANES collected thyroid data only in 2007-2008, 2009-2010, and 2011-2012 as part of the Thyroid Profile component, measured on a one-third random subsample of participants (Subsample A). This component was discontinued after 2012 and is not available in more recent NHANES cycles including the 2017-March 2020 files used for most other analytes.

LabNorms pools the 2009-2010 (THYROD_F) and 2011-2012 (THYROD_G) cycles after confirming distributional equivalence between cycles (medians within 10%, no meaningful directional shift). The Subsample A two-year weight WTSA2YR is divided by 2 when pooling to maintain correct population-level estimates. Because this is a one-third subsample, sample sizes per age and sex group are lower than for full MEC analytes. See the NHANES 2009-2012 Thyroid reference page for full details.

Nutrition Data Sources

Nutrition analytes use several NHANES sources. 25-hydroxyvitamin D, vitamin A (retinol), vitamin E (alpha-tocopherol), and vitamin C come from the NHANES 2017-2018 single cycle (J-cycle files VID_J, VITAEC_J, VIC_J) using the cycle's published two-year MEC weight WTMEC2YR applied directly. Serum and red blood cell folate come from the NHANES 2017-March 2020 pre-pandemic combined cycles (P_FOLFMS and P_FOLATE) using the pre-pandemic folate subsample weight WTFOLPRP applied directly.

Vitamin B12 and methylmalonic acid use NHANES 2013-2014 (VITB12_H and MMA_H) with WTMEC2YR, because those analytes were not published in the 2017-2018 or 2017-March 2020 public lab files. Serum zinc, serum copper, and serum selenium use the NHANES 2015-2016 serum trace-elements file (CUSEZN_I) with the published WTSA2YR one-third subsample weight.

Single-cycle and older-cycle estimates are not pooled across sources. Tail values (P5 and P95) carry wider effective intervals than analytes pooled across cycles, especially for one-third subsample trace-element analytes. Each page states its source cycle, weight, and unweighted sample size.

No participants are excluded for supplement use on any page, because NHANES does not capture supplementation reliably enough to apply such exclusions consistently.

25-Hydroxyvitamin D values are stored in ng/mL (US-clinical convention) and converted from NHANES nmol/L by dividing by 2.496. The unit toggle exposes the SI alternate.

Race and Ethnicity

Creatine kinase pages are also broken down by race and Hispanic origin. These use the NHANES RIDRETH3 variable, a recode of self-reported race and Hispanic origin, and keep NHANES's category names: Mexican American, Other Hispanic, Non-Hispanic White, Non-Hispanic Black, and Non-Hispanic Asian.

NHANES also codes people who reported another race or more than one race in a single combined category (RIDRETH3 code 7). Because it combines many different groups, one distribution for it would not describe any one group, so it is not published as a separate page. Those participants are still included in the age-based pages.

NHANES oversamples some groups by design. The survey weights account for this, so each distribution reflects the US population in that group.

Race and ethnicity pages pool all adults aged 20 and older and are not age-standardized. The groups differ in age composition, so part of any difference between them reflects age. The site reports the observed distributions and makes no claim about the causes of differences between groups.

Limitations

  • Survey design matters. Percentiles depend on the source design and weighting approach. For NHANES pages we apply survey weights, but results still describe the surveyed reference population and should not be overgeneralized beyond it.
  • Population scope. All percentile data on LabNorms comes from NHANES, a US survey, and may not generalize to other countries or to groups the survey underrepresents.
  • Cross-sectional. Survey data is a snapshot. Values from older cycles may not reflect current distributions.
  • Measured, not biological. Distributions reflect what was measured under source study conditions. Pre-analytic variation, assay differences, and calibration shifts are not corrected for.
  • Assay variability. Methods and calibrations differ between studies. Values from different sources are not always directly comparable.
  • Small subgroups. Narrow demographic cuts may rest on small samples, reducing the stability of P5 and P95. Percentiles from groups below the sample sizes in the confidence interval section are marked with †.
  • No clinical validation. These are descriptive statistics, not validated diagnostic thresholds.
  • Source heterogeneity. Sources vary in population, design, and measurement standards. Each is evaluated individually; no claim is made that values across sources are directly comparable.

What This Site Does Not Do

LabNorms does not diagnose, interpret results clinically, define healthy or optimal ranges, personalize results, or give advice of any kind. Individual lab results should always be interpreted by a qualified clinician.