“I barely eat” — under-reporting is nutrition’s best-replicated finding

The best-replicated result in nutrition

Doubly labelled water is the reference method for energy expenditure. You drink water carrying two harmless isotope tracers, and the rate they leave your body over one to two weeks gives your CO2 production — and from that, the calories you actually burned. No diary, no honesty required. When researchers compare that measurement with what the same people report eating, the reports come up short. Not occasionally — in cohort after cohort, across four decades of studies.

The average gap in adults is around 20% (Trabulsi & Schoeller, 2001). The famous case is Lichtman et al. (1992): people who insisted they could not lose weight on under 1,200 kcal a day reported eating 1,028 kcal. Doubly labelled water measured 2,081 — a 47% shortfall. The same group over-reported their exercise by 51%. Their metabolisms were normal. Their reports were not.

Measurement error, not character

It reads like an accusation. It is not one. Under-reporting shows up in paid volunteers who know they are being measured, and in trained dietitians — who miss less, but still miss. The mechanism is mundane: memory. Cooking oil, the dressing, the flat white, the tastes while stirring, the second helping. Small items that never register as “a meal” and so never register at all. Portion sizes get estimated by eye, and eyes run low.

Be clear about what a food log can tell you. It records what you noticed and chose to write down. It cannot tell you what you ate. The difference between those two numbers is not a lie; it is the measurement error of a human memory doing an accountant’s job.

The gap is not evenly distributed

Under-reporting rises with body weight: roughly 14% in normal-weight adults, and 30–40% in adults with obesity in most doubly-labelled-water comparisons. It rises with dieting — restrained eaters under-report more than unrestrained ones. And it is selective: studies consistently find that snacks and foods people label “bad” vanish from records faster than foods they label “good”. The log does not shrink uniformly. It shrinks exactly where the calories are.

Log lighter, audit harder

You cannot recall your way to accuracy, so stop trying to. Shrink the distance between eating and recording instead: log at the table, photograph the plate, scan the barcode while the packet is still in your hand. Then treat every daily total as an estimate with a margin of about 20% — because that is what it is.

Audit with something memory cannot edit: your weight trend. A genuine 500 kcal daily deficit moves a 7-day rolling average by roughly 0.4–0.5 kg a week. Water weight — salt, carbs, a menstrual cycle — can hide fat loss for a week or two, which is why the audit window is three weeks, not one. If your log claims the deficit and the trend stays flat for three weeks, the trend is telling the truth and the log is short by about that much. One weigh-in is noise. The rolling line is the audit.