When examining data on an individual's requirement for any nutrient, it is essential to consider the quality of the intake data. The most valid intake data are those collected from the metabolic study protocols in which all food is provided by the researchers, amounts consumed are accurately measured, and the nutrient composition of the food is determined by laboratory analyses. Such protocols can be used for balance studies with a small number of subjects, but they are seldom possible for larger studies. Thus, intake data are often self-reported (for example, 24-hour recalls of food intake, diet records, or food frequency questionnaires), which have inherent limitations. Potential sources of error in self-reported intake data include over- or under-reporting of portion sizes, omission of foods, and inaccuracies in tables of food composition. These and other sources of dietary intake errors have been discussed in several reviews (Kohlmeier et al., 1997; LSRO/FASEB, 1986; Thompson and Byers, 1994; Willett, 1990) and at two recent conferences on dietary assessment methods (Buzzard and Willett, 1994; Willett and Sampson, 1997). The general conclusion is that self-reported dietary data are subject to a number of inaccuracies and biases. Therefore, the values reported by nationwide surveys or studies that rely on self reporting may be somewhat inaccurate and possibly biased.
Because of day-to-day variation in dietary intakes, the distribution of 1-day (or 2-day) intakes for a group is wider than the distribution of usual intakes, even though the mean of the intakes may be the same. Statistical adjustments have been developed (NRC, 1986; Nusser et al., 1996) that require at least 2 days of dietary data from a representative subsample of the population of interest. These adjustments have been made to the U.S. population intake data from the 1994 U.S. Department of Agriculture (USDA) Continuing Survey of Food Intake of Individuals (CSFII) (Cleveland et al., 1996), which are used in this report to more accurately estimate intakes of specific life stage and gender groups. However, this method does not adjust for the underreporting of intake, which may be as much as 20 percent (Mertz et al., 1991).
Finally, food composition databases that are used to calculate nutrient intake from self-reported and observed intake data introduce errors due to random variability, genetic variation in content, and use of poor analytical methods. In general, when estimating nutrient intakes for groups, the effect of errors in the composition data