Learn More about Statistical Modeling
In science, we often wish to measure phenomena that cannot be observed directly. One approach to overcoming this problem is statistical modeling, which estimates how variables are related to each other.
Directly capturing [glossary term:] usual dietary intake (Learn More about Usual Dietary Intakes) by individuals with an acceptable level of [glossary term:] accuracy and [glossary term:] precision is virtually impossible due to [glossary term:] measurement error (see Key Concepts about Measurement Error) inherent in self-reported intake data. This measurement error includes [glossary term:] day-to-day variation, which reflects that intakes change from day to day. To advance nutrition research, investigators have focused on how to analyze dietary data in a way that accounts for and minimizes the impact of error. Statistical modeling has been shown to be extremely helpful in this context.
- For example, for estimating distributions of usual dietary intake among groups and given repeat [glossary term:] 24-hour dietary recalls (24HR) on at least a subsample of the group of interest, statistical models can be used to separate [glossary term:] within-person variation from between-person variation and remove the excess within-person variation (which results primarily from day-to-day variation in intake).
- In terms of examining relationships between diet and health [glossary term:] outcomes using self-reported intake data, statistical models such as [glossary term:] regression calibration models (Learn More about Regression Calibration) can be used to adjust for measurement error in estimated associations.
Assessing an individual's usual intake remains an unsolved problem in the clinical setting. The challenges inherent in estimating usual intake at the individual level, together with recommendations, are outlined elsewhere.