Assessing agreement among observers of fourth‐grade children's dietary intake at school meals

Abstract

For a dietary‐reporting validation study conducted over 3 school years, researchers unobtrusively observed the food items and per‐item amounts eaten by children during certain school meals. To investigate children's dietary recall accuracy, observed information was compared to reported information from a subsequent interview with the child. Amounts observed eaten were recorded in servings of standard school‐meal portions as none = 0, taste = 0.1, little bit = 0.25, half = 0.5, most = 0.75, all = 1.0, or the number of servings if >1. Two researchers observed the same child at the same meal and recorded intake (281 meals). We compared pairs of observations to assess interobserver reliability (IOR) among 6 observers. To summarize observer agreement, we investigated statistics for different levels of information. Pearson's correlation was 93% for servings per food item and 96% for servings per meal. The table shows percent agreement and Cohen's Kappa. Beverages served in opaque containers accounted for 94 of 257 items for which observers recorded amounts that differed by =0.25 servings. This led to using a tolerance of 0.25 servings for matches. All measurements indicated good levels of IOR. Funded by R01 HL074358.