How accurate is food tracking

"The app says I ate 1,840 kilocalories today. How accurate is that?" To perhaps ten or fifteen percent on a careful day, and much worse on a careless one — and knowing where the error comes from is what lets you decide how careful to be.

The error stack

Four layers, each with its own margin:

1. The database. A labelled product carries its label's value, which the EU guidance on the labelling regulation allows to deviate from analysis by a margin — on the order of a fifth for the main nutrients. A generic food carries an average over many samples of something that varies: two apples are not the same apple. A few percent either way, sometimes more.
2. The portion. The largest error for most people, and it runs in one direction. Estimated portions are systematically larger than people think; a "tablespoon" of oil, a "handful" of nuts and a "medium" potato are each routinely underestimated by a third or more. A scale removes this layer almost entirely.
3. Cooking. Oil absorbed, water lost or gained, the difference between raw and cooked weights of rice and meat. Log ingredients raw, or use a cooked entry for cooked weights, but not one for the other.
4. Memory. Anything logged in the evening from recollection is missing the tastes while cooking, the second helping, the drink. The classic measurement study on self-reported intake found people under-reporting by a large margin — in its small "diet-resistant" subgroup, by close to half — and the direction, if not the size, has been found repeatedly. Logging at the moment removes most of this layer.

What it adds up to

Weighed, logged at the time, with sensible entries: within roughly ten percent, which is about the same as the error on the expenditure estimate it is being compared with. Estimated portions from memory: often off by a third, and almost always low. Neither is a reason to give up tracking; the second is a reason to weigh for a week when the trend disagrees with the log.

What is good enough

It depends on the decision. To learn what your usual portions actually weigh: one careful week with a scale, once. To hold a moderate deficit over months: rough logging plus a trend on the scale, which corrects the log's bias for you. To answer "why am I not losing?": a careful week again — the eating-little page is that page. The log is the estimate; the trend is the measurement.

In the app

The diary accepts weights, "as planned" logging of the menu, and free text for everything else; the trend on the progress page is what the log is checked against, and the target moves with the trend, not with the log.