sample sizes. For nutrients with an Adequate Intake (AI) for age groups older than infants, new research and data that allow replacement of the AIs with EARs and RDAs will greatly aid the assessment of nutrient adequacy. In addition, information on the distribution of requirements is needed so that the appropriate method for assessing the prevalence of inadequacy for groups can be determined (EAR cut-point method vs. full probability approach).

Research should be undertaken to allow Tolerable Upper Intake Levels (ULs) to be set for all nutrients and to generate information on ways to identify and conceptualize the risk of exceeding the UL.

The estimation and amelioration of bias (such as under- or over-reporting of food intake) is a relatively unexplored field. Efforts in the management of bias during data analysis are very preliminary and far from satisfactory at present. This is seen as a high priority area waiting for new initiatives and innovative approaches.

Advances in behavioral research to determine why people underreport food intake would allow development of improved dietary data collection tools that would not trigger this behavior. Such information would also help in the derivation of statistical tools to correct the bias associated with this phenomenon.

Better ways to quantify the intake of supplements are needed. A large proportion of the population in the United States and Canada consumes dietary supplements. Using intakes only from food sources in dietary assessment is certain to result in a faulty estimate of nutrient inadequacy, as well as inaccurate estimates of the percentage of the population with intakes above the UL.

Food composition databases will need to be updated to include the forms and units that are specified by the DRIs. Chemical methodology to facilitate analysis of various forms of certain nutrients (e.g., α-tocopherol vs. γ-tocopherol) may be required for comparison to the DRIs.

Methods for developing standard errors for prevalence estimates should be investigated. Some sources of variance (primarily associated with intake data) can currently be quantified but many (such as those associated with requirement estimates) cannot. Without a standard error estimate, it is not possible to determine if an esti-