Data sources & methodology

This page is the technical reference for HeightComparison3D. It describes the rules our build scripts apply to raw data, the published methods each calculator implements, the checks that stop a bad number reaching a page, and the known weaknesses of each approach. If you want to reproduce a number you see on the site, the equations and constants here are the ones the code uses. Every method is cited in the source list at the bottom of the page.

Datasets at a glance

All data is stored as static JSON and compiled into the site at build time. Nothing is fetched from a third party when you load a page, which keeps pages fast and means every figure you see was fixed, and reviewed, at a known date. The heights are stored in centimeters; feet and inches are always derived from the stored value, never stored separately.

People: how heights are chosen

Heights of real people start from Wikidata, the structured knowledge base maintained by the Wikimedia community, whose data is dedicated to the public domain under CC0. We use the height property (P2048). Wikidata is the right starting point because every statement can carry its own reference, its own rank and its own date, so the source of each number is traceable rather than anonymous. It is not the end point: Wikidata is edited by volunteers, and a single bad edit can make a 6′4″ president 6′8″. The rest of this section describes what the build does about that.

The selection rule for one person

A person on Wikidata can have several height statements: one copied from a sports database, one from a biography, one deprecated after an edit war. Our people build (scripts/build-people.mjs) resolves them with a fixed, ordered rule, so the same input always produces the same output:

1. Drop deprecated statements. Wikidata editors mark known-wrong values with the deprecated rank; we never use them.
2. Convert to centimeters. Values stated in meters, centimeters, millimeters, feet or inches are converted with exact factors (1 ft = 30.48 cm, 1 in = 2.54 cm). Any other unit uses Wikidata’s own SI-normalized amount.
3. Drop impossible values. Anything outside 50–275 cm is discarded and logged. The tallest reliably measured person, Robert Wadlow, was 272 cm, so the bound leaves room for real extremes while catching unit mistakes such as a height typed in inches but labeled centimeters.
4. Prefer preferred rank. If any remaining statement has Wikidata’s preferred rank, only preferred statements are kept.
5. Prefer referenced statements. If any statement carries a reference URL (P854) or a “stated in” source (P248), unreferenced statements are dropped.
6. Prefer the most recent. If statements carry a “point in time” qualifier (P585), only the latest date is kept. Athletes are often re-measured, and the newest roster figure is usually the most accurate.
7. Take the median of whatever is left, rounded to 0.1 cm. With one statement this is just that value; with two equally good statements it is their midpoint.

Every distinct value seen in step 3 is kept alongside the chosen one. Among the 300 most prominent people, any person whose surviving values spread by 2 cm or more is listed in the review queue (see below), even when the rule picked a sensible value, so a human looks at it. Weights follow a simpler rule (drop deprecated values and anything outside 30–300 kg, prefer preferred rank, take the median), and birth dates keep only the part that all statements agree on: if two sources disagree on the day, we store just the year and month.

A worked example: two good-looking values

When we built the data, Victor Wembanyama had two height statements on Wikidata: 88 inches (223.5 cm) at normal rank, referenced to his NBA draft prospect page, and 234.32 cm at preferred rank, referenced to ESPN. The rule keeps preferred statements first, so it chose 234.3 cm. That is exactly the kind of case the review step exists for: the NBA’s own profile lists him at 7′4″, and the NBA has published heights measured without shoes since 2019. We applied an override to 223.5 cm and cite NBA.com first on his page. The rule is deliberately mechanical; judgment is applied afterwards, in the open, with a logged source.

How precise a celebrity height is

Heights of public figures come from very different kinds of measurement. League rosters (NBA since 2019, many football clubs) are measured; actors’ and musicians’ heights are usually self-reported, often rounded to the nearest inch, and sometimes measured in shoes; historical figures come from military, medical or archival records. Adults also lose height with age (the NHANES table below shows US men in their 70s about 4 cm shorter on average than men in their 40s, partly because of shrinkage and partly because of birth cohort). Treat any single figure for a non-athlete as accurate to roughly ±2 cm unless the page cites a measured source.

Supplemental sources, overrides and the review queue

A hand-maintained file of supplemental heights sits beside the Wikidata cache. Every entry must name the person’s Wikidata ID, a height in centimeters, the figure as originally stated (for example “7′4″”), a source label and a source URL that shows the value; entries missing any of these are ignored by the build. An entry can do one of three things:

“Authoritative” has a specific meaning here: a league or club roster, an official biography or government record, a scholarly archive, or the person’s own reported statement in a named interview. Fan-measured celebrity height sites and IMDb are never used as the only source for a value. A source-only entry that turns out to disagree with Wikidata by 2 cm or more is not applied silently either; the build writes it to the review queue so an editor can decide whether it should become an override.

Examples of overrides applied

Some overrides come with a note that is shown on the person’s page, because the honest answer has two parts. Kevin Durant’s note explains that the NBA measured him at 6′10″ without shoes in 2019 while his current NBA.com profile lists 6′11″. Kevin Hart’s note records that he has described himself as 5′5″ in sneakers, while 5′2″ is the barefoot figure most often reported. An example of a source-only entry is Lionel Messi: Wikidata’s value agrees with Inter Miami’s player profile (5′7″, 170 cm) to within 2 cm, so the club listing is cited first and the Wikidata value stands.

Where possible, the fix is also made at the source: correcting Wikidata benefits everyone who reuses it, and once Wikidata is right, the override becomes unnecessary and is removed at the next review.

The review queue

Every run of the people build writes a review queue, a plain list that an editor works through by hand. It has these sections:

• Famous people whose Wikidata height differs from the widely reported figure by more than 5 cm. The build carries a short reference list of widely reported heights for the most-searched people (presidents, the tallest NBA players, headline celebrities). Any mismatch over 5 cm is flagged for an override or a Wikidata fix. At launch this section was empty because every flagged case had been resolved.
• Searched-for people with no usable height, who are left out rather than guessed.
• Every supplemental height, with its stated value, source and any weak-source warning.
• The verification log for the most prominent people, recording each decision (override, source added, or agreement without a citable listing).
• Conflicting values among the top 300 people (spread of 2 cm or more).
• Other issues: out-of-range values dropped, birth-date conflicts, probable duplicate Wikidata items, and English labels that look vandalized.

Which people get their own page

The comparison tool can load all 5,705 people, but only 100 have a dedicated height page. The selection is driven by search demand: people with measured monthly search volume for their height (from keyword research done at the start of the project) are ranked by that volume and take places first; any remaining places are filled by Wikipedia prominence, measured by the number of Wikimedia sitelinks (Wikipedia language editions and sister projects) that cover the person. The same logic chooses 75 character pages, 30 landmark pages and 99 country pages (the most populous countries plus those that are notably tall, short or frequently searched).

We limit pages this way for a quality reason, not a technical one. A dedicated page carries a researched article, a review date and a set of comparisons, and every one has to be checked. A smaller set of pages that people actually look for can be kept accurate; tens of thousands of thin pages could not. Everyone else remains searchable in the tool, with the same sources attached.

Fictional characters

Character heights are curated by hand. The preferred source is the publisher’s own material: an official databook, fanbook, guidebook, volume profile or creator Q&A. Because those books are rarely online, we usually cite a fan wiki page that names the book the number comes from, and we record which book it is. Each character carries a source kind label, and the page wording changes with it:

At launch 201 characters were databook-via-wiki and 22 were fan-wiki only. For example, Yuji Itadori’s 173 cm is cited to the Jujutsu Kaisen Wiki, which attributes it to the Volume 1 character profile.

Bounds such as “over 190 cm”

Some official profiles give a bound rather than a number. The Jujutsu Kaisen Official Fanbook and volume profiles describe Satoru Gojo, Aoi Todo and Panda as “over 190 cm”, Suguru Geto as “over 180 cm”, and Yuta Okkotsu as “almost 180 cm”. We store the stated figure (190, 180) so the character can be drawn and compared, and the page carries a note saying that the figure is a minimum (or, for “almost”, that the real height is slightly under it). We never invent a specific number such as 192 cm to fill the gap.

A few entries are species or group averages rather than individuals, such as an average adult male Na’vi or an average male hobbit. They are labeled as averages, cite their source (for hobbits, the “Heights” essay in The Nature of Middle-earth), and are excluded from character rankings. Where manga, anime and film adaptations disagree, the article on the character’s page discusses the difference, but the headline value follows the cited primary source.

Landmarks, objects and planets

“How tall is a building?” has three defensible answers, and published figures often mix them up. We follow the Council on Tall Buildings and Urban Habitat (CTBUH) height criteria, which define:

Our headline figure for buildings is the architectural height, which is the figure CTBUH uses to rank the world’s tallest buildings. It is also the most stable: antennas are added and removed over a building’s life, while the architectural top is part of the design. When the tip is notably higher, the page says so. The Shard, for example, is recorded at its CTBUH architectural height of 306 m, with a note that the spire tip reaches about 309.6 m. Burj Khalifa’s 828 m is likewise cited to its CTBUH Skyscraper Center entry.

Non-buildings follow the convention of their owners. Statues use the operator’s official statistics (the Statue of Liberty figures come from the National Park Service), and towers that are not buildings are measured to their tip where that is how they are officially described. Tilted structures get a note: the Leaning Tower of Pisa is recorded at 56.67 m on its high side, 55.86 m on its low side, and 58.36 m including the base according to the Opera della Primaziale Pisana. Mountains are given as elevation above sea level, the standard way summits are measured, rather than height above the surrounding land.

Everyday objects (doors, cars, furniture, animals) are labeled as typical sizes, because no single door is “the” door; each cites the reference it comes from. Planet and Sun sizes come from NASA’s National Space Science Data Center fact sheets, for example the Sun fact sheet.

Average height by country

Country averages come from the NCD Risk Factor Collaboration (NCD-RisC), a network of health scientists that pooled 1,472 population-based studies with measured height from 18.6 million people and published mean adult height for 200 countries in eLife in 2016. We use the figures as republished by Our World in Data under CC BY 4.0, which mirrors the NCD-RisC estimates in a machine-readable form. Our build script downloads that file once, caches it, and keeps the most recent birth cohort for each country plus the trend for the 1896, 1926, 1956, 1976 and 1996 cohorts.

What “birth cohort” means

NCD-RisC reports height by year of birth, not year of measurement. The latest cohort on our country pages is 1996: the estimated average height those people reached as adults (age 18). Height is essentially fixed by the end of adolescence, so a birth-cohort figure describes a generation rather than everyone alive today. Because data for recent cohorts is thinner in some countries, NCD-RisC estimates it with a Bayesian hierarchical model that borrows strength across time, age and neighboring countries, and publishes uncertainty intervals that are wider where measurements are scarce.

Why national surveys give different numbers

Readers often find a national statistic that does not match our country page. Usually both are right, because they measure different things:

• Cohort versus cross-section. A health survey averages every adult alive at the time, including people born in the 1930s who are shorter and have lost height with age. For the United States, NCD-RisC gives 177.1 cm for men born in 1996, while NHANES 2021–2023 measured a mean of 175.1 cm across all men aged 20 and over.
• Measured versus self-reported. NCD-RisC only uses measured heights. Surveys that ask people their height tend to come out higher, because people round up.
• Modelled versus raw. Where a country has few recent surveys, the NCD-RisC figure is a model estimate informed by its neighbors; a single national survey may be more current but has its own sampling error.
• Who is counted. Military conscript data, for example, covers young men only.

Where a country’s own statistics office publishes measured data, the country article discusses it and explains the difference. Percentile estimates on country pages assume heights are normally distributed around the NCD-RisC mean with the spread measured in the US (standard deviation 7.6 cm for men and 7.0 cm for women, derived from the NHANES 5th–95th percentile range). That is an approximation; real spreads differ slightly by country.

US adult averages (NHANES)

Every “average American” figure on the site comes from one table: Table 7 of the CDC/NCHS report Anthropometric Reference Data for Children and Adults: United States, August 2021–August 2023 (Fryar et al., Vital and Health Statistics series 3, number 50, June 2025). NHANES staff measure participants’ standing height with a stadiometer in a mobile examination center, so these are measured heights from a nationally representative sample, not self-reports. Table 7 gives the mean and the 5th, 10th, 15th, 25th, 50th, 75th, 85th, 90th and 95th percentiles for adults aged 20 and over and for each decade of age.

The 20+ rows are based on 2,690 men and 3,307 women. By decade, mean height falls from 176.3 cm for men aged 20–29 to 170.5 cm for men aged 80+, and from 163.5 cm to 156.8 cm for women.

Percentiles between the published points

The percentile calculator interpolates between tabulated percentiles in z-score space rather than linearly in percent, because height is close to normally distributed: each published percentile p is converted to its standard normal z = Φ⁻¹(p), the height is located between the two nearest points, and the result is converted back with Φ. Beyond the 5th and 95th percentiles it extrapolates with the standard deviation implied by the published spread:

SD = (P95 − P5) / (2 × 1.6449)
z  = z_P95 + (height − P95) / SD      (above the 95th percentile)
percentile = Φ(z) × 100

Simpler readouts on conversion and comparison pages interpolate linearly between published percentiles and say only “below the 5th” or “above the 95th” outside the table, so they never claim more precision than the table has.

Children’s growth charts (CDC LMS method)

The growth chart and the child mode of the percentile calculator use the 2000 CDC growth charts for stature-for-age and weight-for-age, ages 2 to 20. CDC publishes these as LMS parameters for every half month of age: L (the Box-Cox power that removes skewness), M (the median) and S (the coefficient of variation). Cole’s LMS method turns a measurement X into a z-score:

Z = ((X / M)^L − 1) / (L × S)        when L ≠ 0
Z = ln(X / M) / S                    when L = 0

X = M × (1 + L × S × Z)^(1/L)        (value at a given z, L ≠ 0)
X = M × exp(S × Z)                   (L = 0)

percentile = Φ(Z) × 100

L, M and S are interpolated linearly between the tabulated ages, as CDC recommends. Φ, the standard normal cumulative distribution, uses the Abramowitz and Stegun 7.1.26 approximation (absolute error under 1.5 × 10⁻⁷), and its inverse uses Acklam’s rational approximation (relative error under 1.2 × 10⁻⁹). Both are far more precise than the data. For example, the median stature for boys at 24.5 months is M = 86.86 cm, and for girls at 240 months (age 20) it is M = 163.34 cm.

How the tables are verified

The CDC data files are converted by scripts/build-growth.mjs, which refuses to write output unless every check passes:

1. Known points match: boys’ stature median at 24.5 months is 86.8616 cm and girls’ at 240 months is 163.3383 cm.
2. For every row of both files, the 3rd, 5th, 10th, 25th, 50th, 75th, 90th, 95th and 97th percentiles that CDC publishes are recomputed from that row’s L, M and S and must agree within 0.01 (cm or kg).
3. Ages are strictly increasing and span exactly 24 to 240 months for both sexes.
4. The Khamis-Roche coefficients (below) predict, for a CDC-median child with CDC-median parents, an adult height within 3 cm of the CDC age-20 median at every age from 4 to 17.5.

Because cdc.gov blocks scripted downloads, the committed copies of the CDC files were taken from Internet Archive snapshots of the exact CDC URLs; check 2 confirms that nothing was altered in transit. CDC recommends the WHO growth standards for children under 2, which this site does not chart; for infants, ask your health provider for a WHO chart.

Adult height prediction

The height predictor shows up to three methods side by side, because they use different information and none is exact.

1. Mid-parental (target) height, Tanner 1970

Boys:  (father + mother + 13 cm) / 2
Girls: (father + mother − 13 cm) / 2
Target range: ± 8.5 cm (about ± 2 SD)

The 13 cm term is the average difference between adult men and women, so each parent’s height is put on the child’s sex scale before averaging. Tanner and colleagues introduced this adjustment for growth standards that allow for parents’ heights. It ignores the child’s own growth so far, so it describes a genetic target rather than a forecast. For parents of 178 cm and 165 cm, a son’s target is (178 + 165 + 13) / 2 = 178 cm and a daughter’s is 165 cm.

2. Percentile projection

For ages 2–20, the child’s CDC stature-for-age z-score is carried forward to the age-20 distribution. The range reflects that children’s and adults’ heights are only partly correlated (the Aberdeen growth study found correlations of roughly 0.7–0.8 from about age 3). We use r = 0.75, so the expected adult z-score is r × z with a residual standard deviation of √(1 − r²) ≈ 0.66, and the displayed range is the approximate 80% interval around that regression estimate.

3. Khamis-Roche, 1994 (with the 1995 erratum)

The Khamis-Roche method predicts adult stature from the child’s current height and weight and the parents’ mid-height, without a bone-age X-ray. It was derived from the Fels Longitudinal Study and is valid from age 4 to 17.5. The equation works in inches and pounds:

adult height (in) = β0 + β1 × height (in) + β2 × weight (lb) + β3 × ((mother + father) / 2, in)

The coefficients change with age and sex and are tabulated every half year (56 rows in all); between rows we interpolate linearly. The original 1994 paper printed weight coefficients that were corrected in a 1995 erratum (Pediatrics 95:457), and some software still carries the uncorrected values. Our coefficient file was cross-checked as follows: the intercept, stature and mid-parent columns are identical in two independent machine-readable copies (the matuR R package and the University of Washington Adult Stature Predictor by M. L. Richardson). The weight column follows the erratum-corrected UW copy; the matuR copy carries the pre-erratum values, a factor of ten smaller, with sign and scale anomalies at boys’ ages 5.0, 7.5 and 9.0. The corrected weights also match a third independent table at all 56 rows. Finally, the median-child check in the growth build (above) confirms that the coefficients behave sensibly at every age.

These are the authors’ published bounds for the derivation sample. Accuracy is lower for children who are entering puberty unusually early or late, and for populations unlike the largely white, Ohio-based Fels cohort.

BMI and ideal body weight

Body mass index is weight in kilograms divided by the square of height in meters, classified with the World Health Organization’s adult cut-offs:

BMI = weight (kg) / height (m)²
healthy-weight range at a height = 18.5 × h² to 24.9 × h²   (h in meters)

The body visualizer also reports waist-to-height ratio with NICE’s central-adiposity bands (under 0.5 healthy, 0.5–0.59 increased, 0.6 and above high), which picks up abdominal fat that BMI can miss.

Ideal body weight formulas

The ideal weight calculator reports four classic equations side by side. Each adds a fixed amount per inch of height over 5 feet (152.4 cm):

IBW = base + slope × (height in inches − 60)

These equations were created for drug dosing, not as health targets: Devine’s 1974 formula appeared in a paper on gentamicin dosing, and Pai and Paloucek’s 2000 review traces how it spread into general use without validation against health outcomes. The four disagree by several kilograms at most heights (about 5.0 kg for a 5′10″ man), which is why we show them all next to the BMI range instead of naming one “ideal”. Below 5 feet the per-inch slope is applied in reverse, as most calculators do, and the tool warns that the formulas were not designed for that range. Where adjusted body weight is shown, it is IBW + 0.4 × (actual − IBW), after Traynor and colleagues.

Body shape rules

The body shape calculator classifies women’s measurements with the Female Figure Identification Technique (FFIT) for Apparel (Simmons, Istook and Devarajan, 2004), an apparel-industry rule set that compares bust, waist, high-hip and hip circumferences in inches. The rules are evaluated in the published order and the first match wins:

FFIT’s oval and diamond types need stomach and abdomen girths, which we do not ask for. Following the plus-size modification work of Sokolowski and colleagues, a waist at least as large as both bust and hips is reported as “apple (oval)” before the FFIT rules run. For men there is no comparable validated standard, so the calculator uses a transparent heuristic labeled as such: oval if the waist is at least the chest and hips, triangle if hips exceed chest by 2″ or more, otherwise by the tailoring “drop” (chest minus waist): 10″ or more inverted triangle, 5–10″ trapezoid, under 5″ rectangle.

FFIT has real limits. Parker and colleagues (2021) found that its categories are not always reliable: small measuring differences can move a person across a boundary, because the thresholds are sharp cut-offs. Shape labels describe proportions for clothing fit; they say nothing about health.

Unit conversion and rounding

Since the international yard and pound agreement of 1959, the inch has been defined as exactly 25.4 mm, as NIST documents. Every conversion on the site follows from that definition, with no approximate factors such as “2.5 cm per inch” or “30 cm per foot”:

cm     = (feet × 12 + inches) × 2.54
inches = cm / 2.54
feet   = floor(inches / 12), remainder = inches − 12 × feet

5′11″ = 71 in × 2.54 = 180.34 cm
180 cm = 70.866 in = 5 ft 10.87 in → shown as 5′11″

Rounding rules

Rounding happens only at display time and only once. Feet and inches are carried correctly, so a value that rounds to 12″ becomes the next foot (5′11.6″ shows as 6′0″, never 5′12″). Differences between two characters are computed from the displayed one-decimal values, so a 188 cm versus 190.5 cm comparison always reads 2.5 cm. Sums and percentages use the stored values, so the converter and the calculators never compound rounding errors.

Comparisons, body landmarks and the 3D model

The height difference calculator subtracts the shorter height from the taller one and expresses the gap as a percentage of the shorter height. “Where the head reaches” uses average body-segment proportions from the biomechanics literature, after Drillis and Contini (1966) as reproduced by Winter:

The comparison tool and the body visualizer draw figures that are procedural, not scanned. Each 3D mannequin is generated in your browser from a small set of smooth cross-sections whose positions are fractions of stature from the same segment model (head about 0.13 H, shoulder 0.818 H, leg length about 0.47 H, knee 0.285 H). The crown sits exactly at the stated height, so the figures are to scale with each other. Chest, waist and hip girths are estimated from BMI when you do not enter them, calibrated so a BMI-22 adult lands near typical survey means; when you do enter them, your values are used. The models are deliberately stylized: they show height and overall proportion, not anyone’s real likeness, and no 3D scans or downloaded body models are used.

The height-from-photo tool uses simple proportion: the person’s length in pixels divided by a reference object’s length in pixels, times the reference’s real height. It is only as good as the reference, which must stand at the same distance from the camera as the person.

Known limitations

• Celebrity heights other than measured rosters may be self-reported, rounded or measured in shoes.
• Wikidata can be wrong or vandalized between our reviews; the review queue reduces, but cannot eliminate, that risk.
• Character heights depend on what creators published; adaptations sometimes contradict them.
• Country averages describe a birth cohort and are model-based where measured data is sparse.
• Country percentiles assume a normal distribution with the US spread.
• Growth charts cover ages 2–20 with US reference data; other countries use other references.
• Height predictions carry errors of several centimeters and do not account for early or late puberty.
• IBW formulas and body-shape labels are conventions, not health measures.

The calculators on this site provide general estimates for information only and are not medical advice. Talk to a doctor about any concern about growth, weight or health.

Corrections policy

If you believe a height or statistic is wrong, email hello@heightcomparison3d.com with the page address, the value you believe is correct, and a link to a reliable source. We aim to review every request within 7 days, update the page if the evidence supports it, record the source and refresh the page’s “Last reviewed” date. See the contact page for details. Accepted corrections to people’s heights go through the same override process described above, and significant changes are listed in the changelog below.

Changelog

2026-10-17: launch

• Published with 5,705 people, 223 characters, 75 landmarks and 200 countries; all data accessed and reviewed on this date.
• US adult figures moved to NHANES August 2021–August 2023 (Table 7, published June 2025) as the single source for “average American” statements site-wide.
• Khamis-Roche weight coefficients verified against the 1995 erratum before launch.

Corrections applied before and at launch

• Abraham Lincoln: 204 cm on Wikidata corrected to 193 cm (6′4″), per the House Divided Project at Dickinson College.
• Victor Wembanyama: 234.3 cm (preferred-rank Wikidata value) corrected to 223.5 cm (7′4″), per his NBA.com profile.
• Tacko Fall: 236.2 cm corrected to 228.6 cm (7′6″), per NBA.com; his 7′7″ in shoes from the 2019 combine is noted.
• Kevin Durant, Charles Barkley, Luka Dončić, Klay Thompson and Anthony Edwards aligned with current NBA.com profiles.
• Kevin Hart, Ariana Grande and Jason Momoa aligned with their own reported statements, with notes where figures conflict.