Objective: To estimate the habitual dietary intake distribution in a population on the basis of repeated short-term measurements, especially of multiple 24 h diet recalls.
Methods: Six different statistical methods were evaluated and compared. The comparison referred to theoretical assumptions, admitted data transformations, statistical foundations, available software packages, and applications to real data of dietary intake.
Results: The Nusser method and a simplified version of it proposed in the paper have proved to be universally applicable methods for estimating the usual intake distribution for food groups and nutrients. Also, the Buck method seemed to be a robust estimation procedure suitable for the description of food consumption data, whereas the other considered methods were only applicable for log-normally distributed intake data or required a comprehensive data simulation. Characteristics of the estimated usual intake distribution were a decreased standard deviation, increased lower percentiles, and decreased upper percentiles compared to the observed sample distribution of individual means. Empirical results concerning total fat and vegetable intake in three different European consumption surveys showed that the estimated percentiles of the usual intake distribution did not depend markedly on the number of sampling days.
Conclusions: Repeated short-term measurements like 24 h diet recalls can be used to describe the habitual dietary intake distribution in food consumption surveys. Recommended is a sampling design of two non-consecutive sampling days. The sampling days of all participants should be selected in such a way that they cover all seasons and days of the week.