Important Statistical Considerations Regarding Dietary Data Analyses
Before beginning an analysis of dietary data, it is important to have a basic understanding of some key statistical principles that may affect the results. This module gives an overview of measurement error, with a special emphasis on error related to day-to-day variation in intakes. The module reviews how to check for data symmetry and provides an outline of practical considerations for data analysis, including data sources that are appropriate for different types of analysis.
Twenty-four hour recalls are a primary source of dietary data in NHANES. However, one or two recalls do not accurately reflect an individual’s true usual (long-run average) intake. Because long-run average or “usual” intake is most often the measurement of interest, statistical adjustments are often necessary.
This section describes the basic concepts of both random and systematic measurement error and provides examples of each in dietary data.
Many statistical procedures are based on the assumption that data are normally distributed, and therefore, symmetrically distributed. However, the distribution of dietary intake data is often skewed because, for any given dietary component measured on a single day, many people may have zero intake and at least some people may have very large intakes. Therefore, it is important to check dietary data for symmetry.
Data from the dietary recalls, food frequency questionnaire, and supplement questionnaire each measure different things, cover different time periods, and are collected differently. Because of this, these various types of data lend themselves to different types of analyses and each type of analysis requires different statistical assumptions.