Minimum Dietary Diversity for Women: Partitioning Misclassifications by Proxy Data Collection Methods using Weighed Food Records as the Reference in Ethiopia.

Background: Nonquantitative list-based or open 24-h recalls (24-HRs) have been shown to overestimate the prevalence of Minimum Dietary Diversity for Women (MDD-W), as compared with direct quantitative observations. However, the main sources of error are unknown. Objectives: To assess the measurement agreement of proxy data collection methods for MDD-W, as compared with weighed food records (WFRs). Methods: Applying a noninferiority design, data were collected from 431 nonpregnant females in Ethiopia. MDD-W estimates from both proxy data collection methods were compared with the WFR prevalence by McNemar's chi-square tests, Cohen's Kappa, and receiver operator characteristic analyses. Ten-point food group diversity scores (FGDS) were compared by Bland-Altman plots, Wilcoxon matched-pairs tests, and weighted Kappa. Food group misclassifications were partitioned into errors related to respondent biases or the questionnaire development. Results: List-based and open 24-HRs overreported MDD-W by 8 and 4 percentage points, respectively, as compared with WFR (objective MDD-W prevalence: 8%). Furthermore, list-based 24-HRs overestimated FGDS by 0.4 points (limits of agreement [LOA]: -1.1, 2.0), whereas open 24-HRs led to a 0.3 point (LOA: -1.2, 1.7) overestimate. Food groups most likely to be misreported using proxy data collection methods were "pulses," "nuts and seeds," "dairy products," and "other fruits." Underreporting of consumption occurred among <4% of females for all food groups. Furthermore, respondent biases were the predominant cause of food group overreporting, except for the "pulses" and "other vegetables" food groups, where food items incorrectly included on the food list were the main source of errors. Conclusions: Food group consumption misclassifications by proxy data collection methods were mainly attributable to females overreporting consumption because of respondent biases or the criterion for foods to be counted, rather than the suboptimal development of the food list in Ethiopia. To obtain precise and accurate MDD-W estimates at the (sub)national level, rigorous context-specific food list development, questionnaire pilot testing, and enumerator training are recommended to mitigate identified biases.