Log each meal soon after it, and expect the daily total to look identical either way#
The best time to log a meal is while it is still in front of you or shortly after — within an hour or two, not at bedtime. The reason is not willpower or habit; it is that the accuracy of a food record falls measurably with the number of hours between the fork and the entry. The reason almost nobody notices is stranger and more useful: the daily calorie total is the one number that does not move much when you log late.
That pairing is the whole article. In the best-controlled test of logging delay anywhere in the literature, researchers watched 480 fourth-graders eat actual school meals — so the true intake was observed, not estimated — then interviewed them either about an hour after lunch or about twenty-one hours after it. Measured on the total, the two records were the same: reported energy came to 88.4% of observed energy at the short delay and 86.2% at the long one, a difference with a P value of 0.83. Measured on the contents, they were not the same at all. The share of the lunch that was reported correctly — right food, right amount — fell from 70.6% to 53.2%, and the share of the report that described food the child had not eaten rose from 17.6% to 32.7%1.
This is about when you record, not when you eat. Whether the clock on the meal itself moves anything is a different question with a different literature behind it.
Twenty hours of delay, priced#
Baxter's design compared two conditions with the observation as the referee. The short interval was a recall covering the prior 24 hours taken in the afternoon — roughly one hour after lunch and five after breakfast. The long interval was a recall covering the previous calendar day taken the next morning: about twenty-one hours after lunch, twenty-five after breakfast.
| Short delay | Long delay | |
|---|---|---|
| Lunch reported correctly | 70.6% | 53.2% |
| Lunch report that was not eaten | 17.6% | 32.7% |
| Breakfast reported correctly | 67.2% | 42.4% |
| Breakfast report that was not eaten | 31.3% | 55.0% |
| Reported energy ÷ observed energy, lunch | 88.4% | 86.2% (P=0.83) |
Data: Baxter et al., 2016; 480 fourth-graders, school meals directly observed.
Read the breakfast rows. After twenty-five hours, more than half of what the children reported for breakfast had not been on their tray, and fewer than half the calories they actually ate were described correctly. The errors run in both directions — food falls out, and food that never happened wanders in — which is why they are separate rows rather than one.
An earlier study by the same group, with 60 children, put numbers on both directions at once: shortening the recall window cut omissions by about a third, cut intrusions by about half, and cut total inaccuracy by about a third, all at P<0.012.
One caveat has to be stated plainly and kept in view for the rest of the article: these are nine- and ten-year-olds recalling institutional meals. Adults recalling their own food have better vocabulary for portions and worse structure to their day. The direction is almost certainly the same; the magnitudes probably are not.
The total is the number that hides it#
The P=0.83 row deserves more attention than the headline result, because it explains why almost nobody has noticed this effect in their own logging.
A daily total is a sum. Omissions subtract from it and intrusions add to it, and there is no reason for the two to be equal in any single day — but across a sample they substantially offset, leaving a total that looks stable while the record underneath it has quietly degraded. The authors are blunt about the implication for research: dividing reported energy by observed energy "without accounting for which items or amounts were reported correctly distorts the picture of reporting accuracy."
A late log can arrive at the right total by being wrong twice in opposite directions. The number is unharmed. The record is ruined.
Which matters for anyone using a log to learn something rather than just to hit a number. If your reason for tracking is a single daily figure to steer by, delay costs you less than you would think. If your reason is to find out where your calories actually come from — the point of reviewing the log at all — then a record assembled at midnight is the wrong instrument, and its reassuringly normal total will not tell you so.
The meal that suffers is breakfast, not dinner#
The common intuition about late logging is that the evening meal is at risk because it is closest to the moment you sit down to write. The observation data says the opposite, and the reason is arithmetic: retention interval is a property of each meal, not of the day.
Across 374 children interviewed in six combinations of recall window and time of day, accuracy for school breakfast was best in morning recalls covering the prior 24 hours, and accuracy for school lunch was best in afternoon recalls3. Each meal was reported best when it was the most recent thing being asked about.
A single catch-up session, whenever you schedule it, therefore hands your earliest meal the longest delay. Log everything at 10 p.m. and breakfast is fifteen hours old while dinner is ninety minutes old. Log everything at lunchtime and dinner has not happened yet. There is no time of day at which one session serves the whole day well — which is the actual argument for logging as you go, and it is a different argument from the habit one, where the cue you reliably meet beats the theoretically correct cue.
The adult evidence is thinner, and it splits for a nameable reason#
One trial has tested this in adults, and at first reading it undercuts everything above. Thirty-three UK adults aged 18 to 64 completed dietary recalls two ways in a crossover: the standard single next-day recall, and a "progressive" version split across the day — breakfast before noon, lunch in the afternoon, dinner in the evening, stragglers the next morning, each prompted by text. The progressive version shortened the average retention interval by 15.2 hours4.
The payoff was small. Evening meals gained a food: 5.2 items reported against 4.2 (P=.005). Across the whole day the difference was 13.9 items against 12.7 (P=.12), and mean energy came out at 1,530 against 1,669 kcal (P=.18). Both statistically indistinguishable.
These two bodies of evidence are not in disagreement about what delay does to memory, and it is worth naming exactly what separates them, because the difference is instructive rather than awkward. Baxter had observers recording what each child actually ate, so she could count omissions and intrusions separately. Osadchiy had no gold standard, so the available outcomes were item counts and energy totals — and Baxter's data shows precisely that item counts and energy totals are the measures that stay flat while the underlying record decays. The adult study measured the insensitive quantity. That is not a criticism of it; without observation there was nothing else to measure. It does mean the null result is weak evidence of no effect.
The adult trial's other finding is the one to actually act on. Sixty-five percent of interviewed participants said their memory was better under the progressive method — and 65% still preferred the traditional single sitting, because the prompts landed at inconvenient moments and evening reporting cut into the part of the day they wanted back. Only 30% preferred the progressive version. A method that is more accurate and less tolerable is a method people abandon, which is the constraint every timing rule has to survive.
What to actually do about the clock#
Four rules, ordered by how much error each removes per second spent:
- Capture at the meal; resolve whenever. The decaying asset is your memory of what and how much, not your ability to look up a database entry. Two seconds of a spoken note or a photograph freezes the perishable part; the calorie work can wait for the train.
- If you batch, batch forward, not backward. Logging at lunch covers a fresh breakfast. Logging at bedtime covers a fifteen-hour-old one. Two short sessions beat one long one for the same total effort.
- Prioritise the unstructured occasions. Meals have plates, seats and start times to hang a memory on; the afternoon handful does not, which is why snacks and drinks dominate the foods that go missing. Those are the entries where the first hour matters most.
- Do not read a normal total as a clean record. The comforting sum is exactly the statistic that survives a bad recall. If you want to know where your calories came from, the entries have to be near-contemporaneous; the wider mechanics of misreporting explain the rest of the gap that timing cannot close.
And one thing not to do: do not treat a late entry as worthless. A 53% correspondence rate is still an enormously better record than nothing, and the alternative to logging dinner at midnight is usually not logging dinner. Timing is a way to improve a log you are already keeping — not another standard to fail. The rest of the workflow sits in the counting method.
FAQ#
How long after eating should I log a meal?#
Within an hour or two, and sooner for anything eaten without a plate. In directly observed school meals, lunch reported about an hour later was 70.6% correct against 53.2% at roughly twenty-one hours, with the share of invented food nearly doubling1. Capture immediately if you can; the calorie lookup can be done later without loss.
Is it fine to log the whole day in one sitting at night?#
It is much better than not logging, and much worse than logging as you go — and the meal it damages is breakfast, not dinner. Recall accuracy tracks each meal's own delay: breakfast reports were most accurate in morning recalls and lunch reports in afternoon ones3. A single evening session gives your first meal a fifteen-hour head start on decay.
Why does my daily total look the same whether I log right away or at bedtime?#
Because forgotten food and invented food push the sum in opposite directions. In the observed-meal data, reported energy divided by observed energy was 88.4% at a short delay and 86.2% at a long one (P=0.83), while correctly reported energy collapsed from 70.6% to 53.2%1. A stable total is not evidence of an accurate record.
Sources#
- Baxter SD, Guinn CH, Smith AF, et al. Children's school-breakfast reports and school-lunch reports (in 24-hour dietary recalls): conventional and reporting-error-sensitive measures show inconsistent accuracy results for retention interval and for breakfast location. Br J Nutr. 2016;115(7):1301-1315.
- Baxter SD, Smith AF, Litaker MS, et al. Recency affects reporting accuracy of children's dietary recalls. Ann Epidemiol. 2004;14(6):385-390.
- Baxter SD, Guinn CH, Royer JA, Hardin JW, Mackelprang AJ, Smith AF. Accuracy of children's school-breakfast reports and school-lunch reports (in 24-h dietary recalls) differs by retention interval. Eur J Clin Nutr. 2009;63(12):1394-1403.
- Osadchiy T, Poliakov I, Olivier P, Rowland M, Foster E. Progressive 24-hour recall: usability study of short retention intervals in web-based dietary assessment surveys. J Med Internet Res. 2020;22(2):e13266.
