What do people really eat? Consolidating datasets on food consumption as a reference for analyzing dietary recommendations.

Transforming diets is vital for environmental sustainability and human health. Dietary recommendations play a crucial role in this effort. To understand the changes they imply, a clear understanding of what people eat is essential, but often difficult given misreporting in intake surveys and the complexity added by composite foods. Using Bayesian statistics, we combine quantities available for human consumption at the market level, average food intake quantities from a survey, recipes, and four other datasets, creating a consolidated quantity flow along the food value chain in Germany. It allows us to calculate the quantities consumed as ingredients in composite foods while accounting for biases and errors in the original datasets. Quantifying intake this way, instead of solely using the intake of unprocessed raw primary commodities reported in an intake survey, is more accurate and implies substantially different dietary changes when used as a reference for dietary recommendations.