Your privacy, your choice

We use essential cookies to make sure the site can function. We also use optional cookies for advertising, personalisation of content, usage analysis, and social media.

By accepting optional cookies, you consent to the processing of your personal data - including transfers to third parties. Some third parties are outside of the European Economic Area, with varying standards of data protection.

See our privacy policy for more information on the use of your personal data.

for further information and to change your choices.

Skip to main content

Minimum Dietary Diversity for Women: precision of national surveys and accuracy of brief data collection instruments

Abstract

Background

Minimum Dietary Diversity for Women (MDD-W) has been identified as a promising indicator for monitoring diets globally. MDD-W questionnaires have been integrated into, amongst others, the Demographic and Health Surveys (DHS) and the Gallup World Poll (GWP). Disagreement between estimates of healthy diet metrics for the same country collected in the same year is not desirable.

Objectives

To determine the measurement agreement of MDD-W estimates collected through DHS and GWP (i.e., precision), to assess the impact of survey characteristics on potential discordance, and to examine the comparative validity of the brief data collection instruments used to estimate MDD-W (i.e., accuracy).

Methods

Using meta-data from DHS and GWP, we quantified the percentage points (pp) difference in food group consumption and MDD-W prevalence. Furthermore, we qualitatively examined the differences of four survey characteristics: food groups and sentinel foods used in the MDD-W questionnaire, sampling framework, survey timing, and data collection modality. In addition, using data from two non-inferiority studies in Ethiopia – which used either a weighed food record (WFR) or quantitative 24-hour recall (24-HR) as the reference method – we simulated the total magnitude of errors associated with non-quantitative open or list-based 24-HRs, and subsequently compared the pp differences in simulated food group consumption and MDD-W prevalence.

Results

MDD-W estimates from the GWP were significantly higher than those from the DHS in five of nine country-year sets, one difference was non-significant, and three estimates could not be statistically compared due to lack of reporting on margins of error. The absolute difference between MDD-W estimates from the DHS and GWP were >|5| pp for all country-year sets (range: -17 to + 21 pp). There was poor agreement between the DHS and GWP questionnaires regarding the choice and number of sentinel foods used for each food group in the same country (range: 21 to 65%). In general, GWP data collection covered fewer months and questionnaires were enumerated in fewer languages than the DHS, but the number of sentinel foods used per food group was more standardized across countries. Simulations indicated that the magnitude of errors associated with pilot tested non-quantitative open and extensive list-based 24-HRs were marginally lower than sentinel list-based 24-HRs in Ethiopia.

Conclusion

For global monitoring, standards must be defined for the preferred data collection instrument and survey platform for each healthy diet metric. This would facilitate cross-country comparability and help mitigate misinterpretations of change over time within countries and the selective reporting of national statistics. A harmonized methodology for developing, pilot testing, and continuously updating sentinel food lists is needed to further improve the accuracy of MDD-W questionnaires.

Peer Review reports

Introduction

Unhealthy diets are a leading cause of morbidity and mortality globally [1, 2]. While gradual progress has been made on collecting nationally representative quantitative dietary intake data, diets remain infrequently monitored on a large scale worldwide [3]. In response to the financial and human resource burden and lagged data dissemination associated with quantitative dietary assessment, low-burden brief data collection instruments have been developed – which are conventionally qualitative recalls of food groups consumption [4]. Nevertheless, there remains a lack of consensus on the measures and indicators that best capture the four priority sub-constructs of a healthy diet for monitoring purposes (i.e., diversity, nutrient adequacy, macronutrient balance, and moderation) [5]. Consequently, healthy diet indicators have been omitted in global monitoring frameworks, such as the Sustainable Development Goals and the World Health Assembly global nutrition targets. Due to the rapid pace of changes across food systems and the associated dietary transition, however, the importance of monitoring what people eat and drink across population groups and countries has never been more critical [6].

In 2023, Healthy Diets Monitoring Initiative (HDMI) – a partnership among FAO, UNICEF, and WHO – identified four healthy diet metrics for which routine data collection was deemed most promising at the global level [7]. Low-burden questionnaires for one priority metric, the MDD-W indicator [8], have already been included in two multi-country nationally representative data collection platforms, namely DHS (i.e., Phase 8 Woman’s Questionnaire) and GWP (i.e., Global Diet Quality Project), and many country-specific national and sub-national surveys. Nonetheless, the MDD-W questionnaires used in the DHS and GWP are known to differ in their disaggregation of MDD-W food groups, the total number of food groups included, and the number and examples of foods listed under each food group.

To track national level progress on healthy diets, which also informs global monitoring frameworks, timely, precise, and accurate dietary data are needed [9, 10]. Therefore, it is desirable that prevalence estimates of healthy diet indicators, such as MDD-W, for the same country collected in the same year, be consistent across various surveys to ensure robust monitoring at national level. Evidence for poor precision might hinder the combined use of data collection platforms and might indicate inaccurate data from one or multiple sources. In recognition of this concern, this study assessed the measurement agreement of food group consumption and MDD-W prevalence for available country-year sets, using meta-data from the DHS and GWP, and qualitatively examined the potential causes for disagreement.

However, without an objective reference method, the aforementioned data cannot determine which data collection platform yields the most accurate estimate of MDD-W. The non-quantitative list-based 24-HR method used for DHS and GWP has previously been shown to overestimate MDD-W prevalence among non-pregnant females, as compared to direct quantitative observations [11, 12] or quantitative 24-HRs across multiple countries [13]. Therefore, using data from two non-inferiority studies in Ethiopia, this study also aimed to compare the magnitudes of errors associated with each brief data collection instrument, by adjusting published estimates for the distinct reference measures used (i.e., accounting, where necessary, for respondent biases), and to compare the difference in simulated food group consumption and MDD-W prevalence.

Methods

MDD-W indicator

Achieving MDD-W is defined as a female consuming foods from at least five out of ten food groups over the previous 24 h. The ten predefined food groups are: (1) grains, white roots and tubers, and plantains; (2) pulses (beans, peas, and lentils); (3) nuts and seeds; (4) milk and milk products; (5) meat, poultry and fish; (6) eggs; (7) dark green, leafy vegetables; (8) vitamin A-rich fruits and vegetables; (9) other vegetables; and (10) other fruits. The proportion of females in the survey population who meet this cutoff is defined as the MDD-W prevalence [14,15,16].

Data sources: comparing food group consumption and MDD-W prevalence from DHS and GWP conducted in the same countries

Meta-data on food group consumption and MDD-W prevalence and survey details were manually extracted from reports and websites described in detail below.

In 2019, a questionnaire on food group consumption was included in the DHS Phase 8 Woman’s Questionnaire. The data collected through the non-quantitative sentinel, or irregularly extensive, food list-based 24-HR allows for computation of the MDD-W indicator among (non-pregnant) females aged 15–49 years. Recent DHS adaptations of the MDD-W questionnaire start from the public goods shared by the Global Diet Quality Project. The DHS Program reviews the sentinel foods and makes changes based on suggestions from DHS key informants, before sharing them with the implementing agencies. These agencies themselves frequently offer feedback, which is subsequently circled back to the Global Diet Quality Project to determine if changes can be accommodated. DHS enumerators are trained to read each sentinel food listed to the respondent under the relevant question [17]. On 24 January 2024, at the time of data analysis, the DHS Program website indicated that food group consumption data were collected, and MDD-W point estimates were published in final survey reports, in nine countries [i.e., Burkina Faso [18], Cambodia [19], Kenya [20], Nepal [21], Nigeria [22], the Philippines [23], Sierra Leone [24], Tajikistan [25], and Tanzania [26]], while data collection or final survey report finalization was ongoing for 14 more countries.

In 2021–2023, a broader questionnaire on food group consumption, known as the Diet Quality Questionnaire (DQQ), was included in the GWP. The data collected through the non-quantitative sentinel food list-based 24-HR allows for the computation of multiple metrics, including the MDD-W indicator among females aged 15–49 years. The DQQ country-level adaptation protocol includes a desk review of relevant diet-related data and literature, iterative interviews with 5–10 key informant to identify the most commonly consumed (i.e., sentinel) foods and their vernacular names, identification of anomalous food items from regional-level comparisons of DQQs, and harmonization of terminology used [27]. Enumerators are instructed to read the DQQ exactly as written, without additional dialogue or probing. On 24 January 2024, the Global Diet Quality Project website indicated that MDD-W point estimates were available for 56 countries, including the nine aforementioned countries for which estimates are also available from DHS. Between 2023 and 2024, DQQ data collection was ongoing through the GWP for an additional 36 countries. On 19 June 2024, at the time of writing, disaggregated summary statistics for MDD-W were available for 85 countries on the DQQ Dataverse website [28].

Data sources: partitioning misclassifications by brief MDD-W data collection instruments as compared to reference methods in Ethiopia

Meta-data, and available micro-data, on food group consumption and MDD-W prevalence were extracted from two published non-inferiority studies described in detail below.

Uyar et al. (2023) analysed data from 488 females (aged 15–49 years) collected between November and December 2019 in five regions of Ethiopia, namely, Amhara, Oromia, SNNPR, Tigray, and Addis Ababa [13]. MDD-W point estimates from the DQQ (7.4%; n = 969) were compared to estimates from quantitative 24-HRs enumerated on the same day (1.3%; n = 969). Food group misclassifications could therefore only be attributed to DQQ-related misestimations (e.g., too few or incorrect sentinel foods on the food list). Indeed, without an objective reference measure, such as WFRs, respondent biases, such as intrusions (i.e., respondent recalls consuming a food group, when a WFR objectively indicates no foods belonging to that food group were consumed) and omissions (i.e., respondent fails to recall consuming a food group, when a WFR objectively indicates foods belonging to that food group were consumed) of foods and drinks, cannot be identified due to the correlated errors of both recall-based methods [29].

Hanley-Cook et al. (2024) analysed data from 431 non-pregnant females (aged 15–49 years) collected between June and August 2019 in the Amhara Region of Ethiopia [12]. MDD-W prevalence from non-quantitative open (12%) and (extensive) food list-based 24-HRs (16%) were compared to WFRs enumerated on the day prior (8%). This objective reference measure allowed food group misclassifications to be partitioned into errors related to respondent biases (e.g., memory lapses, social desirability) or brief data collection instrument misestimations (e.g., incomplete food list). For most food groups, the predominant sources of overreporting (i.e., > 80% of cases) by both brief data collection instruments were intrusions [12].

Statistical analysis

Data management, statistical analysis, and data visualizations were conducted in Stata version 16.1 [30]. Descriptive data are presented as frequencies and percentages, while absolute differences are presented in pp with 95% confidence intervals (CIs), whenever possible.

Degree of concordance between DHS and GWP

We used the point estimates of MDD-W prevalence and consumption prevalence of the 10 underlying food groups published in the final DHS reports and on the Global Diet Quality Project website to visualize (i.e., range plots) and subsequently assess the absolute difference (i.e., precision) between the DHS and GWP for nine overlapping countries. To assess if absolute differences were statistically significant, we used the 95% CIs from the GWPFootnote 1 and the available 95% CIs (i.e., ± 2 × standard error) published in final reports by DHS. If the absolute difference between the point estimates from the GWP and DHS was positive, we compared the magnitude of the difference between the lower limit of the GWP 95% CI against the upper limit of the DHS 95% CI.

The Global Diet Quality Project publishes consumption prevalence for “vitamin A-rich orange vegetables” and “vitamin A-rich fruits” separately, which does not allow for the calculation of national consumption estimates of the MDD-W indicator’s “vitamin A-rich fruits and vegetables” food group. Likewise, the Global Diet Quality Project publishes consumption prevalence for “citrus” and “other fruits” separately, which does not allow for the calculation of national consumption estimates of the MDD-W indicator’s “other fruits” food group as it strictly also includes citrus foods. Therefore, potential disagreement of consumption prevalence between the DHS and GWP could not be assessed for these two food groups. However, this has no effect on the discordance analyses for the other eight food groups or MDD-W prevalence.

In addition, in consultation with the HDMI, four broad survey characteristics were defined to qualitatively assess the potential disagreement between estimates from the DHS and GWP:

  1. i.

    Food group questionnaire (e.g., choice and number of sentinel foods).

  2. ii.

    Sampling framework and sample size (e.g., exclusion of regions).

  3. iii.

    Timing of data collection (e.g., in a single or across multiple seasons).

  4. iv.

    Data collection modality (e.g., face-to-face or phone-based interview, languages).

To quantitatively assess the agreement between the sentinel foods listed on the DHS questionnaire and DQQ, two nutrition experts on the study team standardized the food descriptors. For example, in Kenya, “jute mallow” or “mrenda” were both coded with a uniform descriptor “jute mallow,” with the aim to mitigate artificial disagreement based on terminology used. Subsequently, 2 × 2 contingency tables were used to compare the total number of sentinel foods on both questionnaires and the overlap of sentinel foods on the shorter questionnaire only (i.e., the share of foods on the shorter list that also appear on the longer list).

The rank correlation between the absolute difference in MDD-W prevalence and the percent agreement of sentinel foods was assessed using Spearman’s ρ.

Accuracy of DQQ as compared to simulated WFR in Ethiopia

We used the published meta-data in Uyar et al. (2023) [13] to estimate the absolute discordance (i.e., accuracy) of the 10 MDD-W food group consumption prevalence from the DQQ in Ethiopia, when accounting for the proportion of error attributable to respondent biases identified for the list-based 24-HR by Hanley-Cook et al. (2024) (i.e., assuming this best approximates using WFRs as a reference measure) [12]. The relative inflation factor (RIF) attributable to respondent biases was computed as follows:

\(\:\frac{(a-b)}{(\left|a-b\right|+c)}\) × 100

, where a is the number of individuals with intrusions for a specific food group, b is the number of individuals with omissions for the same food group, and c is the number of misestimations (i.e., food item on food list was consumed, but in < 15 g/day). Subsequently, the estimated pp difference in food group consumption prevalence between the DQQ and simulated WFR was computed as follows:

\(\:\left(\frac{100}{100-\text{R}\text{I}\text{F}}\:\right)\:\)× Δ quantitative 24-HR % – DQQ %

For food groups which, according to Hanley-Cook et al. (2024) [12], were not subject to misestimations but only respondent biases, the absolute inflation factor (AIF) was simply calculated as the difference between the proportion of individuals reporting intrusions minus and individuals reporting omissions during the (extensive) list-based 24-HR. Consequently, the estimated pp difference in the food group consumption prevalence between the DQQ and simulated WFR was computed as follows:

Δ quantitative 24-HR % – DQQ % + AIF.

As a comparative analysis, we assessed the impact of using only the AIF, rather than the RIF, on the difference in food group consumption prevalence between the DQQ and simulated WFR for food groups which were in fact subject to a degree of misestimation.

Accuracy of open and (extensive) list-based 24-HR as compared to simulated quantitative 24-HR in Ethiopia

We used the published meta-data in Hanley-Cook et al. (2024) [12] to estimate the discordance of the 10 MDD-W food group consumption estimates from the open and (extensive) food list-based 24-HRs (henceforth referred to as brief instruments) in Ethiopia, when completely discounting the proportion of error attributable to respondent biases (i.e., assuming that a perfect correlation of errors between recall-based methods approximates using quantitative 24-HRs as the reference measure). The relative deflation factor (RDF) attributable to respondent biases was computed identically to the RIF. Subsequently, the estimated pp difference in food group consumption prevalence between the brief instrument and simulated quantitative 24-HR was computed as follows:

Δ WFR % – brief instrument % × \(\:\left(\frac{100-\text{R}\text{D}\text{F}}{100}\:\right)\)

Moreover, we also re-analysed the individual-level micro-data used in Hanley-Cook et al. (2024) [12] to estimate the discordance of the MDD-W prevalence from the brief instruments in Ethiopia, when completely discounting the proportion of error attributable to respondent biases. To this end, individuals with intrusions for a food group, were recoded as non-consumers and individuals with omissions for a food group were recoded as consumers. Subsequently, we re-calculated the MDD-W prevalence from the adjusted brief instruments and assessed the pp difference as compared to the WFR prevalence.

As a comparative analysis, we estimated the discordance of MDD-W prevalence for brief instruments in Ethiopia (i.e., analyses repeated for open and list-based recall), when accounting for the proportion of errors attributable to respondent biases to the WFR. In this case, individuals with intrusions for a food group were recoded as consumers and individuals with omissions for a food group were recoded as non-consumers.

Results

Degree of concordance between DHS and GWP

MDD-W and food group consumption prevalence

The MDD-W point estimates from the DHS and GWP differed by >|5| pp for all nine country-year sets [mean:|11.6| pp; range: -17 pp in Sierra Leone and + 21 pp (95% CI: 13, 28) in Kenya]. Furthermore, the MDD-W estimates from the GWP were significantly higher than those from the DHS in five of nine country-year sets (mean: 12.8 pp) (Fig. 1; Table 1). While the GWP point estimate in Nepal was 8 pp higher than the DHS in Nepal, the difference was not statistically different (95% CI Δ: -1, 17 pp). No variability estimates were published in the DHS reports for three countries. Nevertheless, under the conservative assumption of a margin of error of ± 2 pp, the MDD-W estimates from the DHS in Nigeria and Sierra Leone would be significantly higher than those from the GWP. The opposite conclusion would be reached for Tajikistan, noting however that the DHS and GWP were collected four years apart. The food groups most prone to a higher consumption prevalence through the GWP were “beans, peas, and lentils” (i.e., seven countries), “dairy” (i.e., six countries), and “other vegetables” (i.e., seven countries) (Figs. 23).

Table 1 Description of gallup world poll (GWP) and demographic healthy surveys (DHS) for nine low- and middle-income countries for which minimum dietary diversity for women estimates are available
Fig. 1
figure 1

Difference between nationally representative Demographic and Healthy Survey (DHS) and Gallup World Poll (GWP) estimates of MDD-W prevalence, among females (15–49 years) in nine low- and middle-income countries. Y-axis ranges between a minimum of 20% and a maximum of 100%. Δ, absolute difference in percentage points between DHS and GWP; BFA, Burkina Faso; ISO, International Organization for Standardization; KEN, Republic of Kenya; KMH, Kingdom of Cambodia; NGA, Federal Republic of Nigeria; NPL, Nepal; PHL, Republic of the Philippines; SLE, Republic of Sierra Leone; TJK, Republic of Tajikistan; TZA, United Republic of Tanzania

Fig. 2
figure 2

Difference between nationally representative Demographic and Healthy Survey (DHS) and Gallup World Poll (GWP) consumption estimates of “starchy staples,” “pulses,” “nuts and seeds,” and “dairy” food groups, among females (15–49 years) in nine low- and middle-income countries. Y-axis range differs for each food group. Δ, absolute difference in percentage points between DHS and GWP; BFA, Burkina Faso; ISO, International Organization for Standardization; KEN, Republic of Kenya; KMH, Kingdom of Cambodia; NGA, Federal Republic of Nigeria; NPL, Nepal; PHL, Republic of the Philippines; SLE, Republic of Sierra Leone; TJK, Republic of Tajikistan; TZA, United Republic of Tanzania

Fig. 3
figure 3

Difference between nationally representative Demographic and Healthy Survey (DHS) and Gallup World Poll (GWP) consumption estimates of “flesh foods,” “eggs,” “dark green leafy vegetables,” and “other vegetables” food groups, among females (15-49 years) in nine low- and middle-income countries. Y-axis range differs for each food group. Δ, absolute difference in percentage points between DHS and GWP; BFA, Burkina Faso; ISO, International Organization for Standardization; KEN, Republic of Kenya; KMH, Kingdom of Cambodia; NGA, Federal Republic of Nigeria; NPL, Nepal; PHL, Republic of the Philippines; SLE, Republic of Sierra Leone; TJK, Republic of Tajikistan; TZA, United Republic of Tanzania

Food group questionnaires

Overall, the agreement between the (sentinel) foods on the DHS questionnaire and DQQ was relatively poor, ranging from 21% of total foods in Sierra Leone up to 65% in the Philippines. Nonetheless, the overlap of food items included only on the shorter of the two questionnaires, in other words considering the shorter list as the denominator, was > 75% in six countries, ranging between 48% in Tajikistan and 93% in Nepal (Table 2). The absolute difference in MDD-W prevalence was not significantly rank correlated with the percent agreement of sentinel foods (ρ=-0.12; P = 0.73).

Table 2 Agreement between food item examples listed under the 10 Minimum Dietary Diversity for Women (MDD-W) food groups from the Gallup World Poll (GWP) and Demographic Healthy Surveys (DHS)1

In seven countries, for the MDD-W relevant questions, the DQQ listed more sentinel food items than the DHS questionnaire. Furthermore, for all country-year sets, the DQQ was composed of more individual MDD-W relevant questions due to a greater disaggregation of the 10 food groups, leading to fewer sentinel foods per question. The DHS Sierra Leone questionnaire incorrectly included ice-cream and condensed milk as food item examples for the “milk and milk products” food group. The food groups with the greatest number of (sentinel foods) in the GWP and DHS were “starchy staples” [mean (SD): 17 [2] vs. 15 [4]], “flesh foods” [mean (SD): 20 [2] vs. 24 [17]], and “other fruits” [mean (SD): 11 [9]]. A maximum of 35 and 67 foods were listed under the “dark green leafy vegetables” in the DHS Nigeria and “flesh foods” in the DHS Sierra Leone.

To calculate MDD-W, data from either 22 (e.g., Nepal) or 23 (e.g., Burkina Faso) questions on the DQQ were required. In contrast, the available DHS included between 14 (e.g., Sierra Leone) and 16 relevant questions (e.g., Cambodia) for MDD-W computation.

Sampling framework and sample size

The DHS samples females and, frequently, males aged 15–49 years and their children aged 0–59 months from the entire non-institutionalized civilian population. At present, MDD-W data are collected among females only. Data are collected through face-to-face, computer-assisted personal interviews. DHS usually follows a stratified two-stage sampling design. The first stage involves selection of clusters consisting of enumeration areas, generally from the most recent census frame. The enumeration areas are selected with a probability proportional to their size within each sampling stratum (e.g., region, urban and rural areas within regions). In the second stage, ~ 25–30 households are selected systematically from each cluster. A complete household listing is conducted in each of the selected clusters and acts as the sampling frame.

Due to the non-proportional allocation of households to the different regions and sampling stratums, and the differences in response rates, sampling weights are required. Sampling weights are based on sampling probabilities separately for each sampling stage and for each cluster. For households, the sampling weight is the inverse of its household selection probability multiplied by the inverse of the household response rate in the stratum. Subsequently, for females, the individual survey weight is the household weight multiplied by the inverse of the individual response rate for females in the stratum.

The GWP samples individuals aged 15 years and over from the entire non-institutionalized civilian population. Most data in low- and middle-income countries are collected through ~ 1 h face-to-face, computer-assisted personal interviews. The GWP’s clustered sampling method is as follows. Primary sampling units (i.e., ~ 125 clusters of ~ 8 households) are stratified by population size (e.g., cities, towns, and villages) or geography (e.g., region) and clustering is achieved through one or more stages of sampling. Where population information is available (e.g., census listings), household sample selection is based on probabilities proportional to population size, otherwise simple random sampling is used. Random route procedures are used to select sampled households. Random respondent selection is achieved by using the Kish grid method.

The GWP constructs base sampling weights to account for oversampling and household size (i.e., to adjust for the probability of selection). Thereafter, post-stratification weights are constructed. Population statistics are used to weight the data by gender, age, and, where reliable data are available, education, or socioeconomic status.

In countries where telephone coverage represents at least 80% of the population (e.g., the Philippines) or is the customary survey methodology, ~ 30-minute telephone surveys are used which are based on either random-digit-dial or a nationally representative list of phone numbers. Random respondent selection is achieved by using either the latest birthday or Kish grid method. Typically, the design is not stratified but otherwise the other processes and procedures follow those used in the face-to-face design.

For the countries included in this study, the sample sizes from the GWP ranged between 416 females in Nepal and 595 in Sierra Leone, whereas the sample sizes from the DHS ranged between 10,718 females in Tajikistan and 41,821 in Nigeria. The much smaller sample sizes from the GWP are reflected in the larger variability around the MDD-W point estimates (Table 1).

Exclusion areas

The DHS Kenya excluded one cluster in Mandera County due to insecurity, from a total of 1,692 clusters across the country. Mandera County as a whole is home to ~ 2% of the Kenyan population, and therefore one cluster within it represents an even smaller proportion of the total population.

The GWP in Burkina Faso excluded some communities in the East and Sahel regions for security reasons, representing 4% of the population. The DHS Burkina Faso report indicated that the MDD-W prevalence was 12.5% (n = 948; 95% CI: 7.9–17.1) and 24.6% (n = 618; 95% CI: 20.4–28.8) in the East and Sahel regions, respectively [national estimate: 25.1% (23.8–26.5)].

Likewise, the GWP in Nigeria excluded the states of Adamawa, Borno, and Yobe for safety and security reasons, representing 7% of the population. The DHS Nigeria report indicated that the MDD-W prevalence in these states were 41.9% (n = 903), 38.4% (n = 1,469), and 66.7% (n = 1,327), respectively [national estimate: 56%].

Furthermore, the GWP in Cambodia excluded the Koh Kong, Stueng Treng, Otdar Meanchey, and Kep provinces, representing approximately 3% of the population. The DHS Cambodia report indicated that the MDD-W prevalence was 48.8% (n = 681; 95% CI: 42.1–57.4) in the Koh Kong province, 62.0% (n = 809; 95% CI: 57.5–66.5) in the Stueng Treng province, 51.8% (n = 712; 95% CI: 44.5–59.1) in the Otdar Meanchey province, and 67.8% (n = 746; 95% CI: 63.7–72.0) in the Kep province (national MDD-W estimate: 57.3%; 95% CI: 55.9–58.7).

Timing of data collection and modality

For most countries, the DHS data collection period covered ~ 5 months, while the GWP covered ~ 2 months, with the notable exception of the DHS Philippines which covered less than two months. The most noticeable differences in the timing of DHS and GWP data collection efforts were in Cambodia (i.e., dry vs. wet season), Kenya (summer dry and longer wet vs. winter dry season), Nepal (winter and spring vs. early autumn), and Tanzania (long rainy vs. dry season) (Table 1).

Furthermore, with the exception of the GWP in the Philippines (i.e., mobile telephone-based interview), all MDD-W questionnaires were enumerated face-to-face. In eight countries, the DHS questionnaire was translated and could be enumerated in more languages (range: 2–17 languages) than the DQQ (range: 1–5 languages) (Table 1).

Accuracy of DQQ as compared to simulated WFR in EthiopiaFootnote 2

For the “nuts and seeds,” “flesh foods,” and “other fruits” food groups, the largest proportion of total error, using WFR as the reference, was attributable to respondent biases. Nonetheless, due to a larger number of misestimations (i.e., using quantitative 24-HR as the reference), our simulations – which accounted for respondent biases – indicated that the DQQ’s consumption prevalence for the “dairy,” “dark green leafy vegetables,” “vitamin A-rich fruits and vegetables,” and “other fruits” food groups were the least accurate (i.e., > 5 pp difference). Our comparative analyses confirmed the main findings (Table 3).

Table 3 Estimated discordance of the 10 Minimum Dietary Diversity for Women (MDD-W) food group consumption prevalence from the Diet Quality Questionnaire (DQQ) in Ethiopia (n = 969), without and with accounting for the estimated proportion of errors attributable to respondent biases

Accuracy of open and (extensive) list-based 24-HR as compared to simulated quantitative 24-HR in Ethiopia

When discounting the proportion of total error attributable to respondent biases, using WFR as the reference, our simulations indicated that both the open and list-based 24-HRs were highly accurate for eight food groups (<|1.5| pp difference), while the consumption prevalence for the “beans, peas, and lentils” food group were the least accurate (i.e., > 9 pp difference) (Tables 4 and 5, Supplemental Table 1).

Table 4 Estimated discordance of the 10 Minimum Dietary Diversity for Women (MDD-W) food group consumption prevalence from the non-quantitative list-based 24-HR in Ethiopia (n = 431), with and without discounting the proportion of errors attributable to respondent biases
Table 5 Estimated discordance of the 10 Minimum Dietary Diversity for Women (MDD-W) food group consumption prevalence from the non-quantitative open 24-HR in Ethiopia (n = 431), with and without discounting the estimated proportion of errors attributable to respondent biases

Adjusting the individual-level food group consumption data from the brief instruments (i.e., discounting errors due to respondent biases in the micro-data) showed that the open and list-based 24-HRs estimated MDD-W prevalence to within ~ 1 pp of the reference method (Table 6). Our comparative analyses, adjusting the individual-level food group consumption data from the WFR (i.e., accounting errors due to respondent biases in the micro-data) indicated that the brief instruments estimated MDD-W prevalence to within < 3 pp of the reference method (Supplemental Table 2).

Table 6 Estimated discordance of Minimum Dietary Diversity for Women (MDD-W) prevalence from non-quantitative open and list-based 24-HRs in Ethiopia, when discounting the proportion of errors attributable to respondent biases from brief data collection instruments

Discussion

MDD-W estimates from the GWP were significantly higher than those from the DHS in five of nine country-year sets, while point estimates were likely to be statistically higher for an additional country but lower for two other countries. The variability around the MDD-W point estimates from the GWP was ~ 3.5 larger than those from the DHS in each country. The lack of precision between the DHS and GWP poses problems for reconciling national estimates of MDD-W across and within countries, when collected using different data collection platforms. Moreover, a calibration of prevalence estimates does not seem viable as the direction and the absolute or relative magnitude of the observed measurement disagreement was not systematic across countries.

Overall, sparse MDD-W estimates from other national surveys are more closely aligned with DHS than GWP. To illustrate, the 2023 Burkina Faso National Nutrition Survey reported an MDD-W prevalence of 18% (95% CI: 15–20) [31], as compared to 25% (95% CI: 24, 27) from the 2021 DHS [18]. Furthermore, the 2016 Nepal National Micronutrient Status Survey indicated an MDD-W prevalence of 49% (95% CI: 46–52) [32], in comparison with 50% in the 2016 DHS [33]. Moreover, the 2021 Sierra Leone National Nutrition Survey published an MDD-W prevalence of 74% (95% CI: 72–78) [34], which is 18 and 35 pp higher than the 2019 DHS [24] and 2021 GWP, respectively. Lastly, the 2016 National Nutrition Survey in Tajikistan estimated MDD-W prevalence at 81% [35], which is within 1 pp of the 2017 DHS [25].

We identified that there was poor agreement between the DHS and GWP questionnaires regarding the choice and number of sentinel foods used for each MDD-W food group. However, DHS has recently partnered with the Global Diet Quality Project, which has developed a global database of over 120 DQQs with relevant questions, to fine-tune their own MDD-W questionnaires moving forward [36]. Therefore, a harmonized methodology for developing and updating sentinel food lists, both within and across countries, and a standardized questionnaire used across national data collection platforms is likely to enhance the comparability and accuracy of MDD-W estimates moving forward. For most country-year sets, GWP data collection covered fewer months and questionnaires were enumerated in fewer languages than the DHS. The number of foods listed per question was more standardized in the GWP (i.e., no more than ~ 7 sentinel foods) but for seven country-year sets, the total number of foods enumerated was greater than the DHS due to the larger number of questions and greater disaggregation of MDD-W food groups.

The observed variability in the length of data collection and the seasons covered are likely to have contributed to the differences between the MDD-W estimates from the DHS and GWP. A longitudinal study among 503 mothers in Uganda showed significant fluctuations in the MDD-W prevalence across six-month intervals between November 2014 and May 2016 (i.e., females were more likely to achieve MDD-W shortly after the rainy season) [37]. Similarly, a longitudinal study among 167 mothers in Timor-Leste found significant variability in MDD-W prevalence across three to five month intervals between 2017 and 2018 [38]. Moreover, two longitudinal studies among 200 females of reproductive age in Malawi and Zambia showed that across eleven 24-HRs between September 2017 and May 2018 there were significant seasonal fluctuations in the percentage of females achieving MDD-W, peaking in the harvest season and lowest in the lean season [39]. Likewise, a longitudinal study among over 350 females of reproductive age in Ethiopia indicated that between February 2017 and August 2018 the MDD-W prevalence significantly differed across seasons (i.e., lowest during the Orthodox fasting and irrigation season) [40]. In addition, high-frequency longitudinal data from over 1,700 pregnant females in Burkina Faso showed that the MDD-W prevalence was responsive to seasonal variations, with peaks at the end of the dry season (i.e., April or May) and troughs in the rainy season (i.e., August) [41]. Lastly, a four-year longitudinal analysis of 2,701 mothers enrolled in a cluster-randomized controlled trial in Bangladesh indicated that MDD-W prevalence was prone to considerable seasonal fluctuations and was significantly higher in the pre- and early-monsoon months (May-July) [42]. These studies highlight that the consistent timing of data collection is critical for repeated surveys, such as DHS every ~ 5 years, even when using the same brief instrument, to ensure nonbiased inferences of change and trends in dietary diversity over time. Hence, various national surveys should also aim to align the months across which they collect dietary intake data – most conservative for MDD-W being the lean season – to improve the compatibility and comparability of estimates across platforms and countries. To illustrate, the GWP was implemented between August and September in Burkina Faso, which coincides with the lowest point of MDD-W prevalence, while the DHS was conducted between July and November, a period when MDD-W prevalence is likely to be substantially higher [41]. Therefore, the observed disagreement between the two data collection platforms might in fact be an underestimate for some countries. Neither the DHS or GWP publish data which allow an assessment of the distribution of dietary data collection across weekdays.

In addition, the narrower range of local languages in which data were collected through the GWP might have increased respondent burden and differential response bias among certain subgroups of the female population, as compared to DHS [43]. In DHS Jordan, enumerators have still reported the need to translate food or drinks items into local or slang names on the spot [17]. This finding reinforces the added-value of developing and adapting food lists with local experts, selecting local enumerators, and high-quality training and continuous feedback on the MDD-W questionnaire. Furthermore, the variable placement of the MDD-W questionnaire (e.g., near the beginning or the end of the survey) might be another cause for the differences between the DHS and GWP. To illustrate, a study that randomized the placement of the MDD-W questionnaire in a multi-topic phone-based survey in Ethiopia reported that delaying the questioning of food group consumption by 15 min led to a decrease in the 10-point food group diversity score [mean (SE) difference: 0.3 food groups (0.1)], although no difference in the MDD-W prevalence was observed [44]. In the DHS, the MDD-W questionnaire has been integrated into the “Child Health and Nutrition” section, which is the 6th of 12–16 sections of the Woman’s Questionnaire for the available country-year sets. To our knowledge, national GWP questionnaires, including the placement of the tailored DQQs, are not available as a global public good. Therefore, no comparisons could be made between the DHS and GWP for this survey characteristic.

Regarding data collection modality, the difference in MDD-W estimates from the Philippines is unlikely to be attributable to the mobile phone-based interview used in the GWP. A test-retest study comparing phone-based against face-to-face interviews among females of reproductive age in Kenya, with an interval of approximately one-week, indicated that the data collection modality had no effect on the MDD-W prevalence [45]. Moreover, it is improbable that the exclusion of specific regions in the GWP contributed substantially to the higher MDD-W estimates. With the exceptions of the East region in Burkina Faso and the Kep province in Cambodia, the DHS reports indicated that the MDD-W estimates from the excluded regions were comparable to the national prevalence.

These findings comparing DHS and GWP, however, fail to elicit which data collection platform provided the most accurate MDD-W estimate in each country-year set. For that purpose, (nationally-representative) dietary intake data would have to be collected among the same individuals using the DHS questionnaire, the DQQ, and a reference method of dietary intake, such as WFR on the day prior or quantitative 24-HR recall on the same day. These data would allow an assessment of the discordance of MDD-W estimates obtained through the DHS and GWP, as compared to a more objective MDD-W estimate or an examination of the predictive performance of these MDD-W estimates against reference metrics of nutrient adequacy (e.g., mean adequacy ratio of 11 micronutrients).

Our simulations indicated that the magnitude of errors associated with pilot tested non-quantitative open and (extensive) list-based 24-HRs were lower than a sentinel list-based 24-HRs in Ethiopia. However, when using the lenient non-inferiority criterion of ± 10% point from the reference method set by Uyar et al. (2023) [13], all brief instruments accurately estimated MDD-W prevalence as compared to (simulated) quantitative 24-HRs. These exploratory findings should not be extrapolated to other settings or population groups, as dietary patterns in Ethiopia are known to be more homogeneous and less diverse than in other countries [11, 13] and food group diversity and MDD-W prevalence has been shown to vary substantially across sexes in the same country [16]. Furthermore, for robust global monitoring of healthy diets, brief instruments must achieve stricter non-inferiority criteria. To this end, sentinel food lists might be optimized by relying more on national quantitative dietary intake and market data, rather than on expert opinion, pilot tested, and continuously updated for each country, given the rapid transition of dietary patterns globally. Nevertheless, as was evident from Hanley-Cook et al. (2024), a large share of error from brief instruments is attributable to respondent biases [12]. Hence, future studies might also explore whether the magnitude of respondent biases from MDD-W questionnaire are systematic. These findings might allow an informed judgement of whether MDD-W estimates across countries are equally inaccurate, and therefore comparable for global monitoring.

To increase the precision of MDD-W estimates from the GWP, future research should further assess whether MDD-W is also a valid indicator for micronutrient adequacy among males aged 15–49 years across contexts [16, 46]. This might allow for the entire GWP sample of ~ 1,000 adults to be exploited for global monitoring and enable valid comparisons of MDD-W prevalence across sexes. Moreover, given the limited data collected from females aged 15–49 years in certain GWP rounds (e.g., 182 for Switzerland 2023), reporting a moving average of MDD-W from multiple GWP rounds might be more appropriate, as is done for Sustainable Development Goal indicator 2.1.2: Prevalence of moderate and severe food insecurity in the population, based on the Food Insecurity Experience Scale. In any case, national MDD-W prevalence estimates from a single 24-hour recall, which fail to account for day-to-day variability, are likely to be an under- or overestimate of usual MDD-W prevalence depending on whether the population mean dietary diversity is below or above five food groups (i.e., due to extreme values regressing to the mean) [47].

This study has several limitations that should be considered. First, the discordance between sentinel foods on the DHS questionnaire and DQQ is possibly an overestimate. To clarify, when one questionnaire listed a generic food item descriptor such as “milk” and the other listed “curdled milk,” “fermented milk,” and “powdered milk” these were deemed as food items that disagreed, while one might argue that at least one or all three agree. Second, our simulations made the untestable assumption that all respondent biases were carried over from the non-quantitative DQQ to the quantitative 24-HR enumerated in series on the same day, whereas inherent difference in these data collection instrument may influence respondent biases in distinct ways. Third, the two non-inferiority studies were conducted at different time points and in different locations in Ethiopia, so the observed magnitude of errors might not truly be comparable. Lastly, the research team did not have access to the micro-data used in Uyar et al. (2023) [13]. Therefore, simulations comparing sentinel list-based 24-HRs to WFRs could not be run.

In conclusion, defining both the preferred brief instrument and standard data collection platform to be used for a healthy diet indicator might facilitate its uptake in current and forthcoming global monitoring frameworks. In addition, setting international standards for healthy diet metrics would simplify cross-country comparisons, help mitigate biased inferences of change over time within countries, help prevent selective reporting of national statistics, and ultimately improve assessment of global progress on healthy diets. Therefore, for reporting purposes, a hierarchy of nationally-representative country-year data sources was advanced in a recent proposal to include MDD-W as an additional indicator for SDG 2 – starting from quantitative 24-HRs through to MDD-W questionnaires collected through nutrition surveys, DHS, and GWP. Lastly, pilot testing and continuous amendment and improvement of context-specific food lists is crucial to ensure the accuracy and precision of food list-based questionnaires, such as those used to collect the data required for MDD-W.

Data availability

Nationally representative data and brief data collection instruments supporting the reported results are publicly available on The Demographic and Healthy Surveys Program website (https://dhsprogram.com/, both micro- and meta-data) and Diet Quality Questionnaire Dataverse website (https://dataverse.harvard.edu/dataverse/dqqdata, meta-data only), while meta-data and brief data collection instruments from the non-inferiority studies are publicly available in Hanley-Cook et al. (2020, 2024) [11, 12] and Uyar et al. (2023) [13].

Notes

  1. Personal communication with Dr. Anna Herforth for the Sierra Leone 2021 round. At the time of analysis, only the aggregated statistics (i.e., mean and CI) for the three GWP rounds conducted between 2021 and 2023 were publicly available (n = 1,412). Individual-level microdata for each round has since been published on the DQQ Dataverse website (28).

  2. At the time of writing, data collection for the 2024–2025 DHS Ethiopia was ongoing. Therefore, the agreement between the DHS and DQQ food lists could not be assessed, as the former is published as part of the final report. Consequently, no speculations as to whether these findings are also relevant to the DHS Ethiopia have been made.

Abbreviations

24-HR:

24-Hour recall

AIF:

Absolute inflation factor

CI:

Confidence interval

DHS:

Demographic and Health Survey

DQQ:

Diet Quality Questionnaire

FAO:

Food and Agriculture Organization of the United Nations

HDMI:

Healthy Diets Monitoring Initiative

GWP:

Gallup World Poll

MDD-W:

Minimum Dietary Diversity for Women

pp:

Percentage points

RDF:

Relative deflation factor

RIF:

Relative inflation factor

SD:

Standard deviation

WFR:

Weighed food record

References

  1. Afshin A, Sur PJ, Fay KA, Cornaby L, Ferrara G, Salama JS, et al. Health effects of dietary risks in 195 countries, 1990–2017: a systematic analysis for the global burden of disease study 2017. Lancet. 2019;393(10184):1958–72.

    Article  Google Scholar 

  2. Willett WC, Rockström J, Loken B, Springmann M, Lang T, Vermeulen S, et al. Food in the anthropocene: the EAT–Lancet commission on healthy diets from sustainable food systems. Lancet. 2019;6736(18):3–49.

    Google Scholar 

  3. Food and Agriculture Organization of the United Nations. Intake. Global report on the state of dietary data. Rome, Italy. 2022.

  4. Hanley-Cook GT, Gie SM, Hoogerwerf S, Padula de Quadros V, Balcerzak A, Neufeld LM et al. Cross-context equivalence and agreement of healthy diet metrics for national and global monitoring: A multi-country analysis of cross-sectional quantitative 24-hour dietary intake studies. American Journal of Clinical Nutrition. 2024;120(5):1093–104. Available from: https://doi.org/10.1016/j.ajcnut.2024.07.010

  5. Neufeld LM, Hendriks S, Hugas M. Healthy Diet: A Definition for the United Nations Food Systems Summit 2021. In: Science and Innovations for Food Systems Transformation. 2023;21–30.

  6. Schneider KR, Fanzo J, Haddad L, Herrero M, Moncayo JR, Herforth A, et al. The state of food systems worldwide in the countdown to 2030. Nat Food. 2023;4:1090–110.

    Article  PubMed  PubMed Central  Google Scholar 

  7. Verger EO, Savy M, Martin-prével Y, Coates J, Frongillo E, Neufeld L. Healthy diet metrics: a suitability assessment of indicators for global and National monitoring purposes. Switzerland: Geneva. 2023.

    Google Scholar 

  8. FAO. Minimum dietary diversity for women. An updated guide for measurement: from collection to action. Rome, Italy: FAO 2021;158.

    Google Scholar 

  9. Frongillo EA, Baranowski T, Subar AF, Tooze JA, Kirkpatrick SI. Establishing validity and cross-context equivalence of measures and indicators. J Acad Nutr Diet. 2019;119(11):1817–30.

    Article  PubMed  Google Scholar 

  10. Kirkpatrick SI, Baranowski T, Subar AF, Tooze JA, Frongillo EA. Best practices for conducting and interpreting studies to validate self-report dietary assessment methods. J Acad Nutr Diet. 2019;119(11):1801–16.

    Article  PubMed  Google Scholar 

  11. Hanley-Cook GT, Tung JYA, Sattamini IF, Marinda PA, Thong K, Zerfu D, et al. Minimum dietary diversity for women of reproductive age (MDD‐W) data collection: validity of the list‐Based and open recall methods as compared to weighed food record. Nutrients. 2020;12(7):1–13.

    Article  Google Scholar 

  12. Hanley-Cook GT, Hoogerwerf S, Parraguez JP, Gie SM, Holmes BA. Minimum dietary diversity for women: partitioning misclassifications by proxy data collection methods using weighed food records as the reference in Ethiopia. Curr Dev Nutr. 2024;8(7):103792.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  13. Uyar BTM, Talsma EF, Herforth AW, Trijsburg LE, Vogliano C, Pastori G, et al. The DQQ is a valid tool to collect population-level food group consumption data: A study among women in Ethiopia, Vietnam, and Solomon Islands. J Nutr. 2023;153(1):340–51.

    Article  PubMed  Google Scholar 

  14. Martin-Prevél Y, Arimond M, Allemand P, Wiesmann D, Ballard TJ, Deitchler M et al. Development of a Dichotomous Indicator for Population-Level Assessment of Dietary Diversity in Women of Reproductive Age. Curr Dev Nutr. 2017;1(12):cdn.117.001701.

  15. Verger EO, Eymard-Duvernay S, Bahya-Batinda D, Hanley-Cook GT, Argaw A, Becquey E et al. Defining a dichotomous indicator for population-level assessment of dietary diversity among pregnant adolescent girls and women: A secondary analysis of quantitative 24-h recalls from rural settings in Bangladesh, Burkina Faso, India, and Nepal. Curr Dev Nutr. 2024;8(1).

  16. Hanley-Cook GT, Hoogerwerf S, Parraguez JP, Gie SM, Holmes BA. Minimum dietary diversity for adolescents: Multi-country analysis to define food group thresholds predicting micronutrient adequacy among girls and boys aged 10–19 years. Curr Dev Nutr. 2024;8(3):102097.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  17. Benedict RK. Nutrition Data Collection Procedures: Enhancements and Innovations: DHS Qualitative Research Studies 26. Rockville, Maryland, USA. 2024 [cited 2025 Feb 12]. Available from: https://www.dhsprogram.com/pubs/pdf/QRS26/QRS26.pdf

  18. INSD ICF. Enquête démographique et de Santé du Burkina Faso 2021. Maryland, USA: Burkina Faso and Rockville. 2023.

    Google Scholar 

  19. NIS, MoH ICF. Cambodia Demographic and Health Survey 2021–22 final report. Maryland, USA: Phnom Penh, Cambodia & Rockville. 2023.

    Google Scholar 

  20. KNBS ICF. Kenya Demographic and Health Survey 2022. Volume 1. Maryland, USA: Nairobi, Kenya, and Rockville. 2023.

    Google Scholar 

  21. Ministry of Health and Population, New ERA. ICF. Nepal demographic and health survey 2022. Nepal: Kathmandu. 2023.

    Google Scholar 

  22. National Population Commission, ICF. Nigeria demographic and health survey 2018. Nigeria, and Rockville, Maryland, USA: Abuja. 2019.

    Google Scholar 

  23. PSA ICF. 2022 Philippine National Demographic and Health Survey (NDHS): Final Report. Quezon City, Philippines, and Rockville, Maryland, USA. 2023.

  24. Statistics Sierra Leone, ICF. Sierra Leone demographic and health survey 2019. Sierra Leone, and Rockville, Maryland, USA: Freetown. 2020.

    Google Scholar 

  25. Statistical, Agency. Ministry of Health and Social Protection, ICF. Tajikistan Demographic and Health Survey 2017. Republic of Tajikistan, and Rockville, Maryland, USA: Dushanbe. 2018.

  26. MoH NBS, OCGS ICF. Tanzania Demographic and Health Survey and malaria indicator survey 2022 final report. Maryland, USA: Dodoma, Tanzania, and Rockville. 2022.

  27. Herforth A, Uyar B, Vogliano C, Spray Bulungu A, Gonzalez C, Sokourenko K, Diet Quality Questionnaire (DQQ) Adaptation Methods. Boston USA et al. 2022 [cited 2025 Feb 12]. Available from: https://drive.google.com/file/d/1zxf1eVrQ4VX7TOnVE1OprSsE2EigmUxe/view

  28. Global Diet Quality Project, Dataverse DQQ. 2024. DQQ Results Dataset 2021–2023. Available from: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/KY3W8A

  29. Whitton C, Ramos-García C, Kirkpatrick SI, Healy JD, Dhaliwal SS, Boushey CJ, et al. A systematic review examining contributors to misestimation of food and beverage intake based on short- term self- report dietary assessment instruments administered to adults. Adv Nutr. 2022;13(6):2620–65.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  30. StataCorp LLC. Stata Statistical Software: Release 16.1. College Station, Texas, USA. 2022.

  31. Ministere de la Santé et de l’Hygiene Publique. Burkina Faso Enquête Nutritionelle Nationale 2023. Ouagadougou, Burkina Faso. 2023 [cited 2025 Feb 12]. Available from: https://drive.google.com/drive/folders/1W42IwTfVgZJ49MitVzm0_kpkWpyjKY9g

  32. Ministry of Health and Population, UNICEF NERA, EU, USAID CDC. Nepal National Micronutrient Status Survey, 2016. Kathmandu, Nepal. 2018 [cited 2025 Feb 12]. Available from: https://www.unicef.org/nepal/media/1206/file/Nepal National Micronutrient Status Survey Report 2016.pdf

  33. FAO, IFAD, UNICEF, WFP, WHO., The State of Food Security and Nutrition in the World 2020. Transforming food systems for affordable healthy diets. Rome, Italy. 2020.

  34. Ministry of Health and Sanitation, UNICEF. Sierra Leonne National Nutrition Survey 2021, Freetown S. Leone. 2021 [cited 2025 Feb 12]. Available from: https://portal.mohs.gov.sl/wpfd_file/sierra-leone-national-nutrition-survey-2021-final-print-no-crop-marks_compressed/

  35. Ministry of Health and Social Protection, World Bank, UNICEF. National Nutrition Survey in Tajikistan 2016. Dushanbe, Tajikistan. 2016 [cited 2025 Feb 12]. Available from: https://www.unicef.org/tajikistan/media/491/file/National Nutrition Survey in Tajikistan 2016.pdf#30

  36. Herforth A, Beal T, Rzepa A, Gallup World Poll N. 2020 [cited 2024 May 23]. Global Diet Quality Project Aims to Bridge Data Gap. Available from: https://news.gallup.com/opinion/gallup/321968/global-diet-quality-project-aims-bridge-data-gap.aspx

  37. de Bruyn J, Thomson PC, Darnton-Hill I, Bagnol B, Maulaga W, Alders RG. Does village chicken-keeping contribute to young children’s diets and growth? A longitudinal observational study in rural Tanzania. Nutrients. 2018;10(11).

  38. Bonis-Profumo G, Stacey N, Brimblecombe J. Maternal diets matter for children’s dietary quality: seasonal dietary diversity and animal-source foods consumption in rural Timor-Leste. Matern Child Nutr. 2021;17(1):1–15.

    Article  Google Scholar 

  39. Ahern MB, Kennedy G, Nico G, Diabre O, Chimaliro F, Grace, Khonj et al. Women’s Diet Divers Changes Seasonally Malawi Zambia. 2021. https://alliancebioversityciat.org/publications-data/womens-dietary-diversity-changes-seasonally-malawi-and-zambia

  40. Baye K, Mekonnen D, Choufani J, Yimam S, Bryan E, Grifith JK, et al. Seasonal variation in maternal dietary diversity is reduced by small-scale irrigation practices: A longitudinal study. Matern Child Nutr. 2021;18(2):e13297.

    Article  PubMed  PubMed Central  Google Scholar 

  41. Hanley-Cook G, Argaw A, de Kok B, Toe LC, Dailey-Chwalibóg T, Ouédraogo M, et al. Seasonality and day-to-day variability of dietary diversity: longitudinal study of pregnant women enrolled in a randomized controlled efficacy trial in rural Burkina Faso. J Nutr. 2022;152(9):2145–54.

    Article  PubMed  PubMed Central  Google Scholar 

  42. Waid JL, Wendt AS, Kader A, Sobha S, Gabrysch S. Impact of a homestead food production program on dietary diversity: seasonal and annual results from the Cluster-Randomized FAARM trial in Sylhet, Bangladesh. J Nutr. 2023;154(1):191–201.

  43. Gibson RS, Charrondiere UR, Bell W. Measurement errors in dietary assessment using self- reported 24- hour recalls in low- income countries and strategies for their prevention. Adv Nutr. 2017;8(6):980–91.

    Article  PubMed  PubMed Central  Google Scholar 

  44. Abay KA, Berhane G, Hoddinott JF, Tafere K. Assessing Response Fatigue in Phone Surveys: Experimental Evidence on Dietary Diversity in Ethiopia. World Bank Policy Research Working Paper. 2021;(No.9636).

  45. Lamanna C, Hachhethu K, Chesterman S, Singhal G, Mwongela B, Ngendo M, et al. Strengths and limitations of computer assisted telephone interviews (CATI) for nutrition data collection in rural Kenya. PLoS ONE. 2019;14(1):1–20.

    Article  Google Scholar 

  46. Gómez G, Monge-Rojas R, Vargas-Quesada R, Previdelli AN, Quesada D, Kovalskys I et al. Exploring the FAO minimum dietary diversity indicator as a suitable proxy of micronutrient adequacy in men and women across reproductive and non-reproductive ages in 8 Latin American countries. Food Nutr Bull. 2024;45(Suppl 2):S55–S65.

  47. Barnett AG, van der Pols JC, Dobson AJ. Regression to the mean: what it is and how to deal with it. Int J Epidemiol. 2005;34(1):215–20.

    Article  PubMed  Google Scholar 

Download references

Acknowledgements

The authors express gratitude to Nancy Aburto (FAO) and members of the HDMI who provided feedback on results presented as part of a technical report, namely, Jennifer Coates (Tufts University), Edward Frongillo (University of South Carolina), Lynnette Neufeld (FAO) and Isabela Fleury Sattamini (WHO). We also thank Anna Herforth (Harvard T.H. Chan School of Public Health) and Ty Beal (Global Alliance for Improved Nutrition) for kindly providing methodological details and relevant MDD-W summary statistics from the GWP.

Funding

This work was supported by the Bill and Melinda Gates Foundation (grant numbers: INV-053345 and INV-063321). The funder had no role in the study design, data analysis, or preparation of the manuscript.

Author information

Authors and Affiliations

Authors

Contributions

The authors’ responsibilities were as follow: GH-C: designed research, GH-C and JP: analysed and performed statistical analysis; GH-C: curated and managed data; GH-C and SG: wrote the paper; GH-C and BH: have primary responsibility for final content; and all authors: read, edited, and approved the final manuscript.

Corresponding author

Correspondence to Giles T. Hanley-Cook.

Ethics declarations

Ethics approval

The study did not include any interaction or intervention with human subjects or include any access to non-anonymized data; hence, no approval was required from an Institutional Review Board.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Disclaimer

The views expressed in this publication are those of the authors and do not necessarily reflect the views or policies of the Food and Agriculture Organization of the United Nations.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Rights and permissions

Open Access The Article is licensed under the terms of the Creative Commons Attribution 3.0 IGO Licence (https://creativecommons.org/licenses/by/3.0/igo), which permits use, sharing, adaptation, distribution, and reproduction in any medium or format, as long as appropriate credit is given to the IGO, a link is provided to the Creative Commons licence, and any changes made are indicated.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Hanley-Cook, G.T., Gie, S.M., Parraguez, J.P. et al. Minimum Dietary Diversity for Women: precision of national surveys and accuracy of brief data collection instruments. BMC Nutr 11, 104 (2025). https://doi.org/10.1186/s40795-025-01065-7

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s40795-025-01065-7

Keywords