Size Recommender

Method

Each measurement has its own least-squares regression fitted offline on the ANSUR II male dataset (US Army, 2012 — 4,081 men, 93 measured dimensions each). Predictors were chosen per measurement by forward selection against 5-fold cross-validated error, from a pool of height, weight, age, BMI, √(weight/height) and log and square terms. Girths overwhelmingly selected √(weight/height), which is the expected physical scaling for a circumference.

The ± figure after each value is that measurement's cross-validated RMSE: the typical error against a real measured man. Small bands are dependable, large bands are rough.

Build and waist

ANSUR II has no body-type column, so build was recovered from the data: it is whatever the fourteen residuals still share once height, weight and age have been accounted for, and it accounts for 24% of that leftover. The axis contrasts limb girth against midsection girth, which is simply mass conservation — at a fixed height and weight, the weight sits in one place or the other. Picking a build applies that category's average deviation, up to 4.3 cm on the waist and 1.7 cm on the calf. It barely touches the lengths, because those are skeletal, not build.

Leaving the field on Regular / not sure applies nothing and keeps the wider error bands, so not answering never overstates confidence.

The two optional measurements do different jobs. Waist pins the midsection, and mainly helps bicep and seat. Trouser length pins how your height divides between torso and legs, and is the only input that improves jacket length (2.58 to 2.41 cm) and sleeve (1.84 to 1.71 cm) — build does not move them at all, because they are skeletal. Anything you enter is shown as given and used as measured.

A separate regression is fitted for each combination of supplied measurements, and the page keeps whichever cross-validated lowest rather than assuming more input is always better: knee, calf and leg opening really are predicted better by build alone than by build plus a waist.

One caveat on Stomach. ANSUR measures girth at the navel and has no trouser-line girth at all, so the relationship between the two is an assumption of mine rather than something fitted: stomach is the waist plus an allowance of 1.5 to 5 cm that grows with build. When you supply a waist, Stomach is marked derived and keeps its full error band, because it is not a measurement you gave us. If you wear your trousers well below your navel the gap can be larger than that allowance, and Stomach will read low.

Where the measurement differs from the survey landmark

• Jacket length and sleeve length are anchored to skeletal landmarks (collar-to-seat span, and shoulder-to-wrist outseam), then shifted by a constant fitted against Hockerty's published centimetre size chart. Those two residuals were flat across the whole chart, confirming a pure difference of definition.
• Shoulder width blends acromion-to-acromion breadth with deltoid breadth, since a tailor's tape runs over the shoulder caps rather than bone to bone. This lands within 0.3 cm of Hockerty's chart across its three most common sizes without being fitted to it.
• Chest, stomach and seat are taken straight from ANSUR. Hockerty's chart was rejected as a calibration source for these: its residuals trend steeply rather than sitting flat, because the height and weight columns appended to that chart are not realistic pairings (it lists a 122 cm chest at 99 kg, where real measured men of that height and weight average 109 cm).
• Bicep is ANSUR's flexed measurement less 2 cm, as tailors measure relaxed.
• Knee is not in ANSUR. It is derived from the girths that bracket it — just above the knee, and the calf — and averages 37.2 cm, inside the 34.9–38.8 cm range reported by published men's size surveys.
• Waist deducts a small build-dependent allowance from the navel measurement, since trousers sit off the navel. It is an assumption, not a measured quantity.
• Leg opening is a style choice, not a body dimension. Shown here is the classic 38 cm finished opening scaled gently by ankle and calf. Slim is roughly 35 cm, wide roughly 42–44 cm.

Limitations worth knowing

• ANSUR II measures serving US Army personnel, who are leaner and more muscular than the general public. Expect chest and bicep to read a little generous, and accuracy to fall off at high BMI.
• Height, weight and age cannot express build, which is what the Build field is for. Even with it set, the skeletal proportions — jacket length and front rise — keep wide bands, because build does not move them.
• The build figures assume you classify yourself accurately. Real self-assessment is noisier than the survey data behind it, so expect a little less in practice.
• These are body measurements, not finished garment measurements. No ease is added.
• Treat this as a starting point for a conversation with a tailor, not a replacement for a tape measure before you place an order.