When the standard deviation of the prevalence estimate is not known, formal inferences cannot be made about the prevalence of nutrient inadequacy in a group; for example, one cannot determine whether a prevalence estimate differs from zero, or whether prevalence estimates in two groups are different. The statistical approaches included in this report can be used to partially estimate the standard deviation of a prevalence estimate, but these approaches account only for the uncertainty in the estimates of usual intakes in the group.

Uncertainty also exists in requirement estimates. Although the Estimated Average Requirement (EAR) is a fixed and known quantity, based on data reported in the scientific literature, it is also an estimate of an unobservable median requirement for a group. Statistical methods for estimating the standard deviation of the EAR and the standard deviation of the usual intake distribution are, in principle, available. More difficult from a statistical point of view is combining the two sources of uncertainty into an estimate of the standard deviation for the prevalence of nutrient inadequacy.

Research is needed on ways to better match the biomarkers used to set requirements with the effect of dietary intake on those same biomarkers. Research is also needed on the appropriate biochemical data to collect so that these data can be combined with dietary intake data in assessment. Biomarker and other biochemical data are usually too expensive, time-consuming, or both, to collect on large numbers of individuals. However, when this information is available, it can be used in combination with intake data to give a more accurate estimate of the probability of inadequacy. Because biomarker and intake data are very different proxies for the same unobservable variable (nutrient status), combining the information they provide into an estimate of nutritional status for each individual in a group is a challenging statistical task.

Additional research is also needed for applications that assess the nutrient intakes of different subgroups of the population. In particular, evaluations of nutrition assistance programs typically compare nutrient intakes for program participants and a similar group of nonparticipants. A difficult and not fully explored research question is how to estimate differences in the prevalence of inadequacy between subgroups, after controlling for other factors that also affect nutrient intake. Chapter 7 describes a possible approach to addressing this question based on multiple regression analysis, but research is needed to apply this approach to existing survey data sets such as the Continuing Survey of Food Intakes by Individuals and the National Health and Nutrition Examination Surveys.