Statistical estimation of dietary parameters: implications of patterns in within-subject variation--a case study of sampling strategies.

Patterns within intraindividual variation in energy intake were described previously. Using case studies based on the same Beltsville One-Year Dietary Intake Study data set, we examined the interaction between random and nonrandom variation and the choice of sampling strategy in estimation of individuals' usual intakes over 1 y. Mean intake estimates derived from adjacent-day samples were less reliable and more likely to be biased than were those based on randomly selected days. A finite adjacent-day sample fails to encompass longer-term trends. Because adjacent-day samples underestimate true within-subject variation, by customary tests they appear more reliable. This may present an interpretational problem. Comparisons of random weekend and week-day samples confirm that failure to proportionately sample both will bias the estimation of the usual (1-y mean) intake and the within-subject variance.