Weighed Food Record

A weighed food record is a prospective record in which the participant weighs each food and beverage immediately before consuming it, and weighs any leftovers afterwards. The resulting portion figures do not rely on memory, on estimation against household measures, or on portion-size photographs.

Within dietary assessment it occupies a specific and narrow role: it is the per-meal reference method. When a study needs to establish what was actually on a plate — as opposed to what a person expended over a fortnight — the weighed record is the instrument that establishes it.

Where its remaining error comes from

A scale removes portion error. It does not remove composition error.

Once 150 g of a food is established, converting that mass to energy and nutrients requires a value from a composition database such as FoodData Central. That value is itself an average across samples, cultivars, cuts, and preparation methods. For standardised commodities the residual uncertainty is small; for a composite home-cooked dish with an unknown quantity of cooking oil absorbed, it is not.

The practical consequence is that a weighed record is close to exact for mass and merely good for energy. Validation studies that describe their reference as "weighed portions against a reference composition table" are stating this two-stage structure precisely, and the phrasing is worth reading carefully when it appears.

Reactivity: the characteristic failure

The burden of weighing every item changes what people eat. The documented direction of the effect is toward simplification: fewer mixed dishes, fewer snacks, fewer items eaten away from home, because each of those is harder to weigh and record.

This produces a record that is internally accurate and externally unrepresentative — a precise description of an atypical week. It is the main reason weighed records are normally limited to three to seven days and are not used to characterise habitual intake, a role for which the food frequency questionnaire is better suited despite being less precise per item.

Its role in evaluating estimation technology

Photograph-based and sensor-based estimation systems are validated by comparing their output against weighed reference portions for the same meals. The design is straightforward: prepare and weigh a set of meals, present them to the system, compare.

Two properties of that design deserve emphasis, because both are frequently elided when results are quoted:

1. The composition of the meal set drives the result. A system that performs well on flat plated food and poorly on deep composite bowls will score very differently depending on what the investigators cooked. This is why a single independent measurement, however carefully executed, does not establish that a figure generalises — and why replication by an unrelated group with its own meal set does substantially more work than a narrower confidence interval from the same group.
2. It measures the instrument, not the user. A weighed-portion study evaluates whether a system can estimate a plate placed in front of it. It does not evaluate whether a person using that system remembers to log the cooking oil. The larger real-world error lives in the second question, and no amount of instrument accuracy addresses it.

For the free-living counterpart of this question, see doubly labelled water, which measures expenditure across one to two weeks and detects exactly the under-reporting that a weighed-portion design cannot see.

Protocol notes

• Weigh raw or cooked consistently, and record which. 100 g of dry rice becomes roughly 250 g cooked; mixing the two conventions is the single most common arithmetic error in food records.
• Weigh leftovers. Served mass is not consumed mass.
• Record cooking fat separately, and apportion it across servings rather than assigning the whole pan to one plate.
• Include a weekend day. Weekday-only records systematically understate intake.