Two different failures wearing one name#
People underreport what they eat for two reasons that get filed under one word, and separating them is the whole answer to why it happens. The first is that being recorded changes what you eat: hand someone a food diary and their actual intake drops. The second is that what they write down is lower than what they ate. Both shrink the number at the bottom of the page. Only the second is a logging failure — the first is a real, if temporary, change in behavior that the log describes perfectly well.
The cleanest separation of the two came from a laboratory that measured both at once. Fifty-nine UK adults spent twelve days in a feeding suite where their food was weighed covertly, so the researchers knew the true intake without the participants knowing it was known. When participants were told they were being observed, women's actual energy intake fell by 8 percent (P < 0.001), with the drop concentrated in fat, down 12 percent; men's fell 3 percent, which was not statistically significant. Separately, and on top of that, what participants reported ran 5 to 21 percent below what they had actually eaten1. The authors named these the observation effect and the reporting effect. Almost every practical question about food diaries turns on which one you are dealing with.
The gap that is not a reporting error at all#
Start with the half nobody counts as an error. When a person knows their eating is being watched — by a researcher, by an app, by their own diary — they eat somewhat less, and they eat less of the food they would rather not be seen eating. That is not a flaw in the record. The record is accurate. What it is accurate about is an unrepresentative week.
This matters because the two effects call for opposite responses. If your log reads low because you skipped the second helping on the days you logged, no amount of better capture will fix it, and fixing it would not even be desirable — the behavior is the point. If your log reads low because you ate the helping and did not write it down, better capture fixes it entirely. Conflating the two produces the familiar and useless advice to simply try harder, which addresses neither.
The observation effect is the only measurement error in nutrition that improves the thing it is measuring. A diary that quietly talks you out of the second helping has not malfunctioned.
The asymmetry in Stubbs's data is worth pausing on: women changed their eating under observation and men essentially did not. Whatever drives the observation effect is not a universal human response to being watched — it tracks how much a person cares about how their eating looks.
What survives when the eating is held still#
Strip out the behavior change and you are left with the reporting effect proper: the distance between the fork and the page. In the same laboratory, that distance varied enormously depending on the instrument, and the ranking is the useful part.
| How intake was recorded | Under-report vs. actual (observation + reporting) |
|---|---|
| Weighed food record | ~10% |
| Food-frequency questionnaire | 7–15% |
| 24-hour recall | 15–21% |
| Diet history interview | ~21% |
Those totals combine both effects, which is why they run larger than the reporting effect alone1. The pattern is that methods relying on memory do worse than methods relying on writing things down as they happen — a weighed record, done at the moment of eating, lost about half what a diet-history interview lost. Memory is doing real damage here, and it is doing it in one direction, because what falls out of a recollection is not a random sample of the day.
Across the wider literature the numbers settle in the same neighborhood: a systematic review of 37 studies found roughly 30 percent of participants qualify as under-reporters and energy intake is underestimated by about 15 percent, with the magnitude similar across food records, recalls, and questionnaires6. The full audit of how far a personal calorie total can drift is the pillar's job — see how accurate calorie counting is and the realistic margin of error. What this article is after is the machinery underneath.
Shame is real, and it is probably not the engine#
The popular explanation is embarrassment: people shade their diaries toward the person they would like to be. There is genuine evidence for it, and there is genuine evidence against it being the main story, and the two bodies of work are worth setting side by side because they disagree about something specific.
For the case in favor: in a study of 484 adults aged 40 to 69 whose true energy expenditure was measured by doubly labeled water, social desirability — a questionnaire trait measuring the tendency to present oneself favorably — was among the leading predictors of underreporting on 24-hour recalls for both sexes, in models explaining 22 percent of the variance in women and 25 percent in men3. Fear of negative evaluation and weight-loss history came along with it.
For the case against: when researchers went looking for the same psychological signature under covert weighing — 59 participants over 14 residential days, then 182 participants across laboratory and home days — the associations largely evaporated. Personality traits, body image, and desirability bias produced scattered significant results that did not survive correction for false discovery rate, and in the combined stepwise models only sex and body fat were retained, together predicting 2.5 to 7.3 percent of the variance in misreporting. The authors' conclusion was blunt: personal and psychological characteristics have "little utility in predicting misreporting"2.
Two things separate these results, and neither is that one team was careless. Tooze's outcome was a residual computed against doubly labeled water in free-living adults, so it absorbs genuine variation in what people ate as well as what they reported, and the psychosocial scales were self-report questionnaires administered alongside the dietary ones — trait and outcome share a method. Hopkins's outcome was measured against food weighed covertly, which removes that overlap, and the analysis was corrected for testing many traits at once, which the earlier generation of studies generally was not. Read together, the honest reading is that the shame story is not wrong so much as small: it exists, it shows up most clearly in instruments that ask people to account for themselves out loud, and it is nowhere near large enough to be the mechanism.
A third study complicates it further. Among 65 women assessed against doubly labeled water, the frequent under-reporters did have higher social desirability, higher BMI, and greater body dissatisfaction — but the single variable contributing most to distinguishing them was lower income7. Whatever that reflects — irregular eating, less familiarity with portion language, less practice with forms — it is not embarrassment.
Not a discount, a distortion#
If underreporting were a flat scaling-down, it would be almost harmless: a consistent 20 percent low is still a usable thermometer, which is the argument for reading a log as a trend rather than a truth. It is not flat. In 323 Danish adults whose protein intake was checked against 24-hour urinary nitrogen output, degree of obesity predicted underreporting of both energy and protein — but relative to the total energy they reported, protein was over-represented5. The reported diet was not a smaller version of the real one. It was a cleaner one, with the composition tilted toward the foods people are most confident describing and away from snack-type items eaten between meals.
That directional lean — the categories that quietly go missing, and the drinks, oils, and bites they consist of — is where the practical damage lands. A composition error cannot be corrected by adding a flat percentage, because the missing calories are not spread evenly across the plate.
The diary decays while you are keeping it#
One last mechanism gets missed because it only shows up if you look at the days in order. In the UK National Diet and Nutrition Survey, 6,932 people aged 13 and over kept 4-day food diaries. Reported daily energy fell by 164 kJ — about 39 kcal, roughly 2 percent of the 1,698 kcal daily mean — between day 1 and day 4 (P < 0.001), and the decline was significantly steeper in adults with a BMI above 304.
Two percent is small, and the size is not the point. The point is that day 4 of a diary is not the same instrument as day 1 — it is either recording a person who has quietly started eating less, or a person who has quietly started writing less. The study cannot tell those apart, which is precisely the ambiguity Stubbs's covert weighing was built to resolve. Either way, a week of logging is not seven independent samples of a stable habit, and the tail of a recording period drifts.
Which points each fix at a specific mechanism rather than at willpower:
- Memory decay responds to shortening the delay. Recalls lost roughly twice what weighed records did, so capture at the table beats reconstruction at bedtime — one of the more durable arguments for logging by voice or photo in the moment.
- Composition bias responds to targeting, not effort. The lean is in fats, snacks, and drinks; a standing correction on those beats a general resolution to be thorough.
- The observation effect responds to nothing, and should not. It is behavior change, not error, and it fades on its own as logging becomes routine — which is also why the first week of any diary flatters you.
- The reporting effect under social pressure responds to lowering the stakes of the entry. It is largest in the methods that make a person account for themselves, and smallest in the ones that just record.
FAQ#
Do people underreport their food on purpose?#
Mostly not. Deliberate shading exists — social desirability was among the leading predictors of underreporting on 24-hour recalls in one 484-person study3 — but when misreporting was measured against covertly weighed food and corrected for multiple testing, psychological traits explained only 2.5 to 7.3 percent of it2. Memory and the composition of what gets left out do far more work than intent.
Does keeping a food diary change what you eat?#
Yes, measurably. Under covert weighing, women's actual intake fell 8 percent while they knew they were being observed, with fat down 12 percent; men's fell 3 percent, which was not significant1. That is a genuine behavior change rather than a recording error, and it means your first logged week probably understates your normal eating even if every entry was correct.
Does underreporting get smaller as you get more practice?#
Within a single recording period it goes the other way. Across 6,932 four-day diaries, reported intake fell about 2 percent from day 1 to day 4, and more steeply in people with a BMI above 304. What improves with practice is portion vocabulary and consistency — not the underlying lean, which is why the trend on the scale, not the log, is the thing that catches it.
Sources#
- Stubbs RJ, O'Reilly LM, Whybrow S, et al. Measuring the difference between actual and reported food intakes in the context of energy balance under laboratory conditions. Br J Nutr. 2014;111(11):2032-43.
- Hopkins M, Michalowska J, Whybrow S, Horgan GW, Stubbs RJ. Identification of psychological correlates of dietary misreporting under laboratory and free-living environments. Br J Nutr. 2021;126(2):264-275.
- Tooze JA, Subar AF, Thompson FE, Troiano R, Schatzkin A, Kipnis V. Psychosocial predictors of energy underreporting in a large doubly labeled water study. Am J Clin Nutr. 2004;79(5):795-804.
- Whybrow S, Horgan GW, Macdiarmid JI. Self-reported food intake decreases over recording period in the National Diet and Nutrition Survey. Br J Nutr. 2020;124(6):586-590.
- Heitmann BL, Lissner L. Dietary underreporting by obese individuals — is it specific or non-specific? BMJ. 1995;311(7011):986-989.
- Poslusna K, Ruprich J, de Vries JHM, Jakubikova M, van't Veer P. Misreporting of energy and micronutrient intake estimated by food records and 24 hour recalls, control and adjustment methods in practice. Br J Nutr. 2009;101 Suppl 2:S73-85.
- Scagliusi FB, Ferriolli E, Pfrimer K, et al. Characteristics of women who frequently under report their energy intake: a doubly labelled water study. Eur J Clin Nutr. 2009;63(10):1192-9.
