populations, in which feeding is controlled and about the same for all individuals in the group (e.g., nursing homes). In these special instances it may be possible that the variance of intakes in the group could become small enough to create a problem. In this case, it might be better to assess adequacy using the probability approach rather than its short cut.
The assumption of symmetry of the requirement distribution is inappropriate for at least one important nutrient: iron requirements in menstruating women. As will be evident by inspection of the simulation results, when this assumption does not hold the performance of the Estimated Average Requirement (EAR) cut-point method for estimating the prevalence of nutrient inadequacy leaves much to be desired. In cases where it is known that the distribution of requirements is skewed, use of the probability approach is recommended to assess adequacy of nutrient intake for the group. In the case of iron, for example, the estimate of prevalence that would result from applying the probability approach and using a log-normal model for the requirement distribution will be less biased than that resulting from application of the EAR cut-point method. This is likely to be true even if the log-normal model is not the correct model for requirements.
The model used for simulating intakes and requirements in this section differs from the ones described in previous sections. Here, the simulation model was based on one proposed by the Food and Agriculture Organization/World Health Organization (FAO/WHO, 1988) to describe iron requirements. It has been established that daily losses of iron are 0.77 mg, and menstrual losses of iron are modeled as log-normal random variables with a mean (in natural log units) of −0.734 and standard deviation of 0.777. The specification of the model also assumes high iron availability in the diet (a bioavailability of 15 percent). For the simulation, the skewness of the requirement distribution was varied, and five values considered: 0.6, 1.3, 2.5, 3.2, and 5.7. Recall that for a symmetrical distribution, the value of the skewness coefficient is equal to zero; thus, increasing skewness reflects increasing departures from symmetry. Intakes were simulated independently as normal random variables with a mean intake of 12 mg, and standard deviation of 3 mg resulting in a CV of intake of 25 percent.
Rather than repeatedly sampling groups of 2,000 from the population