We know from the USDA's Beltsville study that no matter how hard we try, almost all of us are unable to accurately report how much we eat.
Studies using "doubly labeled" water -- as close to a gold standard as exists -- find that the average non-overweight adult man needs about 3,050 calories a day to maintain a stable body weight, and the average woman about 2,400. The Food and Drug Administration's 2,000-calorie standard for food labels is 50 percent lower than the average for men and 20 percent lower than that for women. But many -- if not most -- Americans are gaining weight. Therefore, they must be eating more calories each day than they need to maintain a stable weight.
Research on diet and health in general, and on weight gain in particular, depends on knowing what people eat. But outside of a whole-body calorimeter or a controlled environment for metabolic studies, getting even reasonably accurate information about dietary intake is, to say the least, challenging. Indeed, we consider finding out what people eat the greatest intellectual challenge in the field of nutrition today. Why? We have no nice way of saying this. Whether consciously or unconsciously, most people cannot or do not give accurate information about what they eat. When it comes to dietary intake, pretty much everyone forgets or dissembles. This problem makes surveys of dietary intake exceedingly difficult to conduct and to interpret.
DIETARY INTAKE SURVEYS
Doubly labeled water methods make it possible -- although at considerable effort and expense -- to measure average calorie expenditures in individuals going about their daily lives for periods of up to about three weeks. The number of people whose calorie needs can be determined this way is limited. To date, scientists have not come up with any simple, inexpensive way to measure calorie intake accurately in large groups of individuals who are not incarcerated in a laboratory but are "free-living" and going about the normal business of daily life.
Short of duplicate meal analysis, a method in which chemists analyze the nutrient composition of an exact duplicate of everything someone eats, all other methods for obtaining information about dietary intake depend on self-reports. These are, again to understate the matter, inconsistent and unreliable. Merely asking people about what they eat influences what they tell investigators. Even duplicate meal analysis has its hazards, as knowing your meals will be checked can be enough to change your normal eating behavior.
The usual methods for obtaining information about dietary intake ask you to do one of three things:
- Report what you remember eating and drinking in the previous day (this is called a retrospective 24-hour diet recall).
- Keep a record of what you eat for a day or more (concurrent diet record).
- Mark off on a long list of foods the ones you ate within the last day, week, month, or year (retrospective food frequency questionnaire).
The lack of precision in these methods is legendary. People do not easily remember what they ate. You might forget or be uncomfortable about reporting late-night snacks, candy picked up on the run, something you ate in a car, alcoholic beverages, or the amounts of foods you are eating. Like most people, you probably tend to underestimate the sizes of your food portions and overlook intake of foods you perceive as unhealthy. And, of course, you do not eat the same foods every day.
Nutrition scientists have put enormous effort into trying to evaluate the magnitude of reporting errors. They find that people underestimate their true calorie intake by astonishing percentages, typically 30 percent, with a range of 10 to 45 percent depending on such factors as age, sex, body composition, and socioeconomic status. Underreporting of food -- and therefore calorie intake -- increases with age and is greater among women, people who are overweight, and those of low education and income status. People also tend to exaggerate intake of foods they think are supposed to be good for health .
Researchers are still debating whether one survey method is better than another, whether collecting information about portion sizes is either useful or necessary in dietary intake surveys of populations, and indeed whether any method can capture the complexities of diets that vary so much from person to person and from day to day. As an example of how hard it is to draw conclusions from surveys based on one day's reported food intake, a decades-old study done by the United States Department of Agriculture (USDA) deserves careful attention.
The USDA runs an agricultural research center in Beltsville, Maryland. In the early 1980s its scientists observed glaring inconsistencies in the results of the agency's national surveys of dietary intake. They noticed that the number of calories reported as consumed by men and women in 1977-78 was lower than the number reported in 1965. Not only that, but respondents to the 1977-78 survey reported intakes 300 to 400 calories below the amounts needed to maintain their weights. Even more suspicious, during the 13-year period from 1965 to 1978 the average heights and weights of survey participants had increased, meaning that they should have been eating more calories, not fewer. USDA scientists guessed that survey participants were underreporting calorie intake and decided to study whether 24-hour or even three-day dietary intake records could truly approximate habitual food intakes.
The investigators asked their fellow employees at Beltsville to volunteer for a demanding experiment: reporting everything they ate or drank every single day for one entire year. Amazingly, they were able to induce 29 USDA employees working in staff, scientific, and administrative positions to sign up for this undertaking. The investigators knew that the act of reporting might influence what the volunteers ate or said they ate. As a reliability check, they also collected duplicate meals from each volunteer for one full week on four occasions during that year. They reported the results of this herculean effort in a supplement to the American Journal of Clinical Nutrition in 1984 .
The study population included 13 men and 16 women, ages 20 to 53. They were instructed to eat as they always did but to measure the sizes and weights of everything they consumed with rulers, measuring cups, measuring spoons, and a food scale. They were also to write down brand names, food preparation methods, and recipes. From this mountain of data, the USDA scientists converted information about food intake into nutrient and calorie intake through the use of food composition tables.
The results? The men in the study reported consuming an average of 2,760 calories and the women 1,850 calories a day. Their weights did not change during the year. These figures were 200 to 300 calories higher than averages reported in dietary intake surveys of that era. Even so, the investigators knew that the calories had to be underreported. They had the proof. During the four weeks in which they analyzed duplicate meals, they found actual calories to exceed reported calories by an average of 13 percent. On this basis, they urged readers to be cautious when interpreting balance studies done with subjects consuming self-selected diets .
In subsequent studies, USDA statisticians had some fun with these data. They calculated the number of days on which dietary intake data would have to be collected to arrive at an estimate of calorie intake that fell within plus or minus 10 percent of the average for the 365 days of the study. The minimum number of days? Fourteen. And this was for the volunteers whose diets varied least from day to day. For the group as a whole, the number of reporting days required to get within 10 percent of the average number of calories consumed was 27. For the man whose diet varied the most, it took 84 days of food records to get within 10 percent of his average daily calorie intake for the year .
The moral: one 24-hour dietary analysis is unlikely to represent average dietary intake. Given this problem, we can understand why even the most carefully conducted dietary surveys find average calorie intakes to be well below those observed in doubly labeled water studies. But researchers hope that the results of taking 24-hour recalls or collecting food frequency records from large numbers of people -- the more the better -- will average out and indicate intake levels or dietary patterns that apply to the group as a whole, if not to specific individuals within the group .
Excerpted from Marion Nestle and Malden Nesheim's Why Calories Count: From Science to Politics (University of California Press)
 IOM. Dietary Reference Intakes for Energy. Washington, D.C.: National Academies Press, 2005:107-264. See also: Basiotis PP, Lino M, Dinkins JM. Consumption of food group servings: People's perceptions vs. reality. Nutrition Insights 20, October 2000, at www.cnpp.usda.gov.
 Mertz W, ed. Beltsville one-year dietary intake study. AJCN 1984;6(suppl):1323s-403s.
 Kim WW, et al. Evaluation of long-term dietary intakes of adults consuming self-selected diets. AJCN 1984;40:1327-32.Kim WW, et al. Effect of making duplicate food collections on nutrient intakes calculated from diet records. AJCN 1984;40:1333-37.
 Basiotis PP, et al. Number of days of food intake records required to estimate individual and group nutrient intakes with defined confidence. J Nutrition 1987;117:1638-41.
 Investigators often use food frequency questionnaires to obtain dietary information from 100,000 individuals or more. These studies typically report calorie intakes much lower than those obtained in dietary surveys or measurement studies. For a review and critique of this method, see Lee R, Nieman D. Nutritional Assessment, 5th ed. New York: McGraw-Hill, 2009.