Why Food Diaries Under-Report Intake, and What That Means for Reading Any Logged Total

What studies that checked self-reported intake against doubly labeled water and urinary biomarkers found about the size, causes and limits of the gap

Quick answer: Self-reported energy intake runs low. In large studies that checked diaries and recalls against doubly labeled water, an objective measure of energy expenditure, food records and 24-hour recalls understated intake by roughly 12% to 21% on average, and food frequency questionnaires by roughly 28% to 38%. The gap tends to widen as body mass index rises and varies widely between people, so a logged energy total is better read as an estimate that probably runs low than as a measurement. The trend in body weight over several weeks is a useful cross-check.

This note sets out how the gap was measured, who it affects most, why it arises, and what follows for anyone reading a food log, whatever tool produced it. It describes research on measurement and is not advice for an individual. Anyone making health decisions from a food record should speak with a doctor or registered dietitian.

How do researchers know that food diaries under-report?

A diary cannot be checked against itself, so the key studies compare it with measures that do not depend on memory or willingness to disclose. For energy, the reference is doubly labeled water, which measures total energy expenditure over one to three weeks of ordinary life (Westerterp 2017). In people whose weight is stable, expenditure equals intake, so it serves as a marker of intake. For protein, sodium and potassium, 24-hour urine collections play the same role. The National Cancer Institute’s dietary assessment primer calls these recovery biomarkers because they show a direct relationship with absolute intake.

The first comprehensive review of studies that paired reported intake with doubly labeled water expenditure (Livingstone and Black, 2003) concluded that they conclusively demonstrate a widespread bias toward underestimating energy intake. A 2019 systematic review of 59 studies in 6,298 free-living adults (Burrows and colleagues) agreed: most studies reported significant under-reporting, few found over-reporting, and the gap was highly variable even within the same method. Its authors note two limits: doubly labeled water tracks intake only when weight is stable, and its measurement window does not always match the period a diary covers.

How large is the gap?

Two U.S. studies and a pooled analysis that includes one of them give consistent averages, expressed relative to the measured value.

• Food records and 24-hour recalls. In the OPEN study of 484 adults aged 40 to 69 (Subar and colleagues, 2003), men under-reported energy on 24-hour recalls by 12% to 14% and women by 16% to 20%. In a 2018 study of 1,075 adults aged 50 to 74 (Park and colleagues), the shortfall was 15% to 17% on automated self-administered recalls and 18% to 21% on four-day food records. A pooled analysis of five validation studies (Freedman and colleagues, 2014; 263 to 544 participants per study) averaged 15% for a single 24-hour recall.
• Food frequency questionnaires. The gap was larger: 31% to 36% in men and 34% to 38% in women in OPEN, 29% to 34% in the Park study, and 28% in the pooled analysis.

Across the 59 studies in the 2019 review, significant group-level under-reporting ranged from 11% to 41% for food records (19 studies; one other found over-reporting of 8% and three found no significant difference), from 4.6% to 42% for food frequency questionnaires, and from 8% to 30% for 24-hour recalls, the least variable method. A 2009 systematic search of 37 studies in adults (Poslusna and colleagues) put the typical picture at about 30% of participants classed as under-reporters and energy intake underestimated by about 15%.

The gap is smaller for some nutrients: the Park study found under-reporting greater for energy than for protein, potassium or sodium. The U.S. figures above are averages from research volunteers. The pooled authors caution that reporting in the wider population may be somewhat poorer, because their participants were nearly all non-smokers; the Park authors note that their mostly white, older sample limits generalizability.

Who under-reports more?

Body size is among the most consistent predictors. In the pooled analysis, body mass index strongly predicted under-reporting of energy and protein, with education and age also related, and the Park study found under-reporting more prevalent among people with obesity. Macdiarmid and Blundell’s 1998 review found it more common in women than in men and in overweight and obese groups, with a prevalence of 18% to 54% in large nutritional surveys and up to 70% in some subgroups. Hill and Davies’s 2001 review adds dietary restraint and socioeconomic status.

The clearest clinical illustration is a 1992 study in the New England Journal of Medicine by Lichtman and colleagues. In a study of 224 consecutive people with obesity presenting for treatment, the researchers evaluated in detail ten (nine women and one man) with a history of diet resistance, meaning repeated failure to lose weight despite reporting an intake under 1,200 kcal a day. Their measured energy expenditure was within 5% of predicted values, but they under-reported actual intake by an average of 47% (SD 16) and over-reported physical activity by 51% (SD 75). The authors concluded that the lack of weight loss reflected an intake higher than reported, not abnormal thermogenesis. Ten people selected for reported diet resistance is a small group, so the study shows what can be happening in that situation, not how often it happens.

Is it forgetting, or does recording change what people eat?

Both, in the studies that tried to separate the two. Goris and Westerterp (1999) followed 24 lean, highly motivated female dietitians through a week of weighed food recording. Recorded water intake plus metabolic water matched measured water loss closely (r = 0.93), which the authors took as evidence of precise recording. Even so, body mass fell by 0.6 kg during the recording week, against a change of 0.1 kg the week before, and the 16% shortfall in reported intake was attributed to eating less while recording.

In 30 men with obesity (Goris and colleagues, 2000), the average shortfall was 37%: about 26 points were attributed to eating less during the recording week and about 12 points to incomplete recording. The reported share of energy from fat fell as under-reporting rose, which the authors read as selective under-reporting of fat. Macdiarmid and Blundell’s review suggested something compatible: foods with a negative health image, such as cakes and sweets, are more likely to be under-reported, and genuine under-eating has to be distinguished from under-reporting. The 2019 review calls the behavior change reactivity and notes it may come from a wish to simplify recording or to appear to eat healthily.

A record therefore describes the period in which it was kept, and that period may not match usual eating.

Does under-reporting make self-reported diet data useless?

Researchers disagree, and the two positions are worth stating exactly. Archer and colleagues (2013) compared reported energy intake in U.S. national survey data from 1971 to 2010 (28,993 men and 34,369 women aged 20 to 74) with estimated basal metabolic rate and estimated expenditure. They judged intake physiologically implausible for 58.7% of men and 67.3% of women, with the largest average shortfalls, up to 716 kcal a day in men and 856 in women, in people with obesity. Because expenditure was estimated rather than measured, this is weaker evidence than the biomarker studies, though it points the same way.

Dhurandhar and colleagues (2015) argued that self-reported energy intake and physical activity are “so poor that they are wholly unacceptable for scientific research” and that reliance on them should be discontinued. A 2015 commentary by Subar and colleagues answered claims of that kind. It recommended continuing to collect self-report data, which carry valuable information about the foods and beverages consumed, but not using self-reported energy intake as a measure of true energy intake, and using it instead to energy-adjust other nutrients, which improves estimates of diet-health associations. The camps agree on one point: a reported energy total is not a measurement of true intake.

How should a logged total be read?

What follows is this note’s reading of the evidence above, not a finding of any single study.

Treat it as an estimate that probably runs low, by an unknown amount. The averages do not transfer to one person: the 2019 review found the gap highly variable within a single method, and the pooled analysis found weak person-level agreement between reported and true energy (correlations of 0.21 to 0.31), so adding a flat percentage to a log is not defensible. As arithmetic only, because these gaps are expressed against the measured value, a 15% shortfall means true intake is about 18% above the logged figure, and a 21% shortfall about 27% above.

Prefer ratios to absolute amounts. In the Park study, protein and sodium densities were close to biomarker values on all three instruments, although potassium density on food frequency questionnaires ran 26% to 40% high. The 2009 search found iron, calcium and vitamin C intakes about 30% lower in low-energy reporters, while nutrient density tended to be higher, though not consistently.

Check the log against body weight. When weight is stable, intake equals expenditure (Livingstone and Black). A log showing a large deficit alongside weeks of unchanged weight is a prompt to question the log, as the Lichtman study illustrates, and to consider medical causes; a persistent mismatch is a reason to see a clinician rather than to cut intake further.

Do not chase precision. If recording food has become a source of distress or rigid rules, speak with a clinician; see our framework for recognizing when food tracking becomes disordered. Related reading: interpreting a single nutrient result covers intake data alongside laboratory results, and why self-monitoring works for some people and fails for others covers what trials report about weight.

Food diaries and recalls understate energy intake in most studies that have checked them against an objective measure, by roughly 12% to 21% on average for records and recalls and more for food frequency questionnaires, with larger gaps at higher body weight and wide variation between individuals. The shortfall reflects both incomplete recording and eating differently while recording. A logged energy total is best read as a rough estimate that probably runs low, ratios hold up better than absolute amounts, and the trend in body weight is the cross-check. Decisions about medication, pregnancy, diabetes or medically supervised weight loss belong with a doctor or registered dietitian.