Children's school-breakfast reports and school-lunch reports (in 24-h dietary recalls): conventional and reporting-error-sensitive measures show inconsistent accuracy results for retention interval and breakfast location.

Validation-study data were analysed to investigate retention interval (RI) and prompt effects on the accuracy of fourth-grade children's reports of school-breakfast and school-lunch (in 24-h recalls), and the accuracy of school-breakfast reports by breakfast location (classroom; cafeteria). Randomly selected fourth-grade children at ten schools in four districts were observed eating school-provided breakfast and lunch, and were interviewed under one of eight conditions created by crossing two RIs ('short'--prior-24-hour recall obtained in the afternoon and 'long'--previous-day recall obtained in the morning) with four prompts ('forward'--distant to recent, 'meal name'--breakfast, etc., 'open'--no instructions, and 'reverse'--recent to distant). Each condition had sixty children (half were girls). Of 480 children, 355 and 409 reported meals satisfying criteria for reports of school-breakfast and school-lunch, respectively. For breakfast and lunch separately, a conventional measure--report rate--and reporting-error-sensitive measures--correspondence rate and inflation ratio--were calculated for energy per meal-reporting child. Correspondence rate and inflation ratio--but not report rate--showed better accuracy for school-breakfast and school-lunch reports with the short RI than with the long RI; this pattern was not found for some prompts for each sex. Correspondence rate and inflation ratio showed better school-breakfast report accuracy for the classroom than for cafeteria location for each prompt, but report rate showed the opposite. For each RI, correspondence rate and inflation ratio showed better accuracy for lunch than for breakfast, but report rate showed the opposite. When choosing RI and prompts for recalls, researchers and practitioners should select a short RI to maximise accuracy. Recommendations for prompt selections are less clear. As report rates distort validation-study accuracy conclusions, reporting-error-sensitive measures are recommended.