into a distribution that can be modeled and used in selecting random members to contribute to the final requirement distribution. When the final distribution representing the convolution of components has been derived, then the median and ninety-seventh and one-half percentile of the distribution can be directly estimated. It is recognized that, in its simplest form, the Monte Carlo approach ignores possible correlation among components. In the case of iron, however, expected correlation is built into the modeling of the requirement where components are linked to a common variable, such as growth rate, so that not all sources of correlation are neglected. These methods and their statistical basis are described in detail in the DRI report Dietary Reference Intakes: Applications in Dietary Assessment (IOM, 2000a).

The EAR may also be used in the assessment of the intake of groups (IOM, 2000a) or, together with an estimate of the variance of intake, be used in planning for the intake of groups (IOM, 2003) (see Chapter 8).

If sufficient scientific evidence is not available to calculate an EAR, a reference intake called an Adequate Intake (AI) is provided instead of an RDA. The AI is a value based on experimentally derived intake levels or approximations of observed mean nutrient intakes by a group (or groups) of healthy people. The AIs for children and adults are expected to meet or exceed the amount needed to maintain a defined nutritional state or criterion of adequacy in essentially all members of a specific apparently healthy population. Examples of defined nutritional states include normal growth, maintenance of normal circulating nutrient values, or other aspects of nutritional well-being or general health.

As is the case in this report for determining the recommended intakes of water, potassium, sodium, and chloride, the AI is set when data are considered to be insufficient or inadequate to establish an EAR on which to base an RDA. For example, for young infants for whom human milk is the recommended sole source of food for almost all nutrients for the first 4 to 6 months, the AI is based on the daily mean nutrient intake supplied by human milk for healthy full-term infants who are exclusively fed human milk. For adults, the AI may be based on data from a single experiment, on esti-