📄 medicine

Spatial multilevel and decomposition analysis of minimum dietary diversity among Indian children aged 6 to 23 months in NFHS-5 with an evaluation of the Aspirational Districts Programme

This study utilizes a comprehensive multi-method framework on NFHS-5 data to reveal that Minimum Dietary Diversity among Indian children remains critically low and spatially clustered with stunting, driven primarily by maternal education and wealth inequalities, while finding no statistically significant impact from the Aspirational Districts Programme.

Original authors: Hemeswari Bhuyan, Bhanita Barman, Manjunatha VK, Kangkana Talukdar, Mayuri Ojah, Nikhila Priya V, Srikant Behera, Jyoti Ranjan Mohanty, Paramjot Panda, Saumyasree Pradhan

Published 2026-07-02

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Hemeswari Bhuyan, Bhanita Barman, Manjunatha VK, Kangkana Talukdar, Mayuri Ojah, Nikhila Priya V, Srikant Behera, Jyoti Ranjan Mohanty, Paramjot Panda, Saumyasree Pradhan

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A Nutritional Report Card

Imagine India has 64,000 young children (aged 6 to 23 months) taking a "Nutrition Report Card." The test isn't about how much they eat, but how varied their diet is. To pass, a child needs to eat foods from at least 5 out of 8 different food groups (like grains, dairy, eggs, meat, and fruits) in a single day.

The study, using data from a massive national survey (NFHS-5), found that only about 1 in 4 children passed this test. Roughly 3 out of 4 children failed, meaning their diets are too repetitive and lack the variety needed for healthy growth.

The Main Findings: Where and Why?

1. The "Neighborhood Effect" (It's Not Just Your Home)

The researchers realized that a child's diet isn't just about what their parents buy; it's also about where they live.

The Analogy: Think of a child's diet like a plant. Even if you water the plant perfectly (good parenting), if the soil in the garden (the district) is poor, the plant won't thrive.
The Finding: About 24% of the difference in children's diets is due to the "soil" (the district and state they live in), not just the family. Some districts are like fertile gardens where diverse food is easy to find; others are like rocky deserts where it's hard to get a varied diet, regardless of how hard the parents try.

2. The "Teacher vs. Wallet" Debate

Many people assume that if a family is poor, the child will have a bad diet. While money matters, this study found something surprising: A mother's education is actually a bigger factor than the family's wallet.

The Analogy: Imagine two families. Family A has a bit more money but the mother didn't go to school. Family B has less money, but the mother is educated. The study suggests the child in Family B is more likely to have a diverse diet.
Why? An educated mother is better at knowing what to feed the child and how to stretch limited resources to get the most nutrition. She acts like a skilled chef who can make a gourmet meal out of simple ingredients, whereas a lack of knowledge can lead to a poor diet even if money is available.

3. The "Double Trouble" Zones

The researchers used a special map (like a heat map) to find areas where two bad things happen at the same time: children aren't eating well, and they are also failing to grow tall (stunting).

The Finding: They found 84 specific districts (mostly in the central and eastern parts of India, like Bihar and Uttar Pradesh) where these two problems overlap heavily. These are the "Double Trouble" zones where the need for help is most urgent.

4. The "Aspirational Districts" Experiment

The government launched a special program called the Aspirational Districts Programme (ADP) to fix the worst-performing districts. The researchers asked: Did this program help children eat better?

The Result: No. While the program helped children grow taller and gain weight (anthropometric outcomes), it did not improve the variety of food they ate.
The Analogy: Imagine a program designed to fix a car. It successfully fixed the engine (growth/weight), but it forgot to fix the fuel quality (dietary diversity). The car runs, but it's not getting the best fuel.

The "Secret Ingredients" for Success

Using advanced computer models (like a smart detective), the study ranked the most important factors that determine if a child passes the diet test:

Mother's Education: The #1 factor.
Family Wealth: The #2 factor.
Child's Age: Older babies (18–23 months) eat better than younger ones (6–11 months).
Prenatal Care: Mothers who visited the doctor 4+ times during pregnancy had children with better diets.

The Bottom Line

The paper concludes that to fix this problem, India needs to stop looking at just one thing.

Don't just give money: Focus on educating mothers.
Don't just treat the whole country: Target the specific "Double Trouble" districts where the problem is worst.
Fix the program: The government's special district program needs to add "dietary diversity" as a specific goal, not just weight gain.

In short: Three out of four Indian toddlers are missing out on a varied diet. The solution lies in empowering mothers with knowledge and targeting the specific neighborhoods where the problem is deepest.

Technical Summary: Spatial Multilevel and Decomposition Analysis of Minimum Dietary Diversity Among Indian Children (NFHS-5)

Problem Statement
Despite a decade of national nutrition initiatives (e.g., POSHAN Abhiyaan, Saksham Anganwadi), the prevalence of Minimum Dietary Diversity (MDD) among Indian children aged 6–23 months remains critically low. Existing evidence regarding MDD failure is fragmented, relying on disparate definitions, weighting schemes, and analytical methods. Key gaps in the literature include: (1) a lack of decomposition of rural–urban and tribal–non-tribal gaps into compositional and structural components; (2) an absence of formal partitioning of the contextual share of MDD variance (district and state levels) net of household characteristics; (3) no mapping of the spatial co-occurrence of MDD failure with anthropometric failure (stunting); and (4) a lack of evaluation regarding the impact of the Aspirational Districts Programme (ADP) on dietary diversity, despite its evaluation for stunting and underweight outcomes.

Methodology
This study utilizes a unified framework integrating spatial, multilevel, decomposition, machine learning, and quasi-experimental methods using the National Family Health Survey-5 (NFHS-5, 2019–21) dataset. The analytic sample comprises 64,123 children aged 6–23 months, weighted using NFHS-5 sampling weights ( $v005$ ), strata ( $v022$ ), and primary sampling units ( $v021$ ).

Outcome Definition: The primary outcome is the WHO/UNICEF (2021) MDD indicator, defined as the consumption of $\ge$ 5 of 8 food groups in the previous 24 hours (including breast milk). A sensitivity analysis employed the WHO 2008 definition ( $\ge$ 4 of 7 groups, excluding breast milk).
Multilevel Modeling: A three-level random-intercept logistic regression (child nested within district, nested within state) was estimated to calculate Variance Partition Coefficients (VPC) and Median Odds Ratios (MOR), quantifying contextual effects.
Inequality Decomposition: Wealth-related inequality was assessed using the Erreygers-corrected Wagstaff-normalized concentration index. The rural–urban and tribal–non-tribal gaps were decomposed using Blinder–Oaxaca and Fairlie methods, respectively, to separate explained (compositional) and unexplained (structural) components.
Spatial Analysis: Global Moran's $I$ and Local Indicators of Spatial Association (LISA) were computed to identify spatial clustering. A bivariate LISA analysis mapped the co-clustering of MDD failure and district-level stunting.
Machine Learning: An XGBoost classifier with SHapley Additive exPlanations (SHAP) values was trained to rank non-linear determinants and visualize feature interactions.
Policy Evaluation: A difference-in-differences (DiD) design with state fixed effects evaluated the ADP's impact on MDD between NFHS-4 and NFHS-5, testing for pre-trend balance using NFHS-3 and NFHS-4 data.

Key Results

Prevalence: The national MDD prevalence under the 2021 definition is 22.9% (95% CI: 22.4–23.5), meaning approximately three in four children fail to meet the criterion. Only 11.3% meet the Minimum Acceptable Diet. State-level prevalence ranges from 14.9% (Rajasthan) to 53.5% (Meghalaya).
Contextual Variance: Multilevel modeling reveals substantial contextual effects: 17.8% of residual variance is attributable to the district level and 6.1% to the state level (combined VPC $\approx$ 23.9%). The district-level MOR is 1.86, indicating that a child moving from a low- to a high-MDD district faces 86% higher odds of meeting MDD, independent of household characteristics.
Inequality Drivers: The Erreygers-corrected concentration index is +0.144 (pro-rich). Decomposition attributes the largest shares of inequality to maternal education (41.2%) and household wealth (22.7%), followed by antenatal care (12.4%).
Spatial Clustering: Global Moran's $I$ (0.41, $p < 0.001$ ) confirms strong spatial autocorrelation. Bivariate LISA identified 84 "High–High" co-cluster districts where both MDD failure and stunting are high, concentrated in Bihar, Uttar Pradesh, Jharkhand, and Madhya Pradesh. These districts contain ~20% of the under-two population but account for ~33% of the national MDD and stunting burden.
Determinants: XGBoost-SHAP analysis ranks maternal education, household wealth, child age, and antenatal care as the top predictors (AUC = 0.83). Interaction effects show maternal education has a larger marginal impact in rural and ADP districts.
ADP Impact: The DiD analysis yields a coefficient of +0.7 percentage points (95% CI: -0.6 to +2.1; $p = 0.31$ ) for the ADP $\times$ NFHS-5 interaction. This indicates no statistically significant effect of the Aspirational Districts Programme on MDD between the two survey rounds, contrasting with its documented impact on anthropometric outcomes like underweight.

Significance and Contributions
The paper claims four primary contributions to the literature:

Integrated Framework: It provides the first integrated multilevel decomposition of NFHS-5 MDD, demonstrating that approximately one-quarter of the variance is contextual (district/state level), a figure higher than previously documented for complementary feeding.
Reframing Inequality: It establishes that maternal education and antenatal care, rather than household wealth alone, are the dominant decomposable drivers of dietary diversity inequality, suggesting a policy shift toward education and frontline counseling.
Operationalizing "Double Burden": The bivariate LISA framework operationalizes the concept of "double-burden" districts, providing a direct tool for inter-sectoral targeting of the 84 identified co-cluster districts.
Policy Evaluation: It offers the first quasi-experimental estimate of the ADP's impact on a complementary feeding indicator, finding that while the program has improved anthropometric outcomes, it has not yet detectably moved the dietary diversity indicator.

The authors conclude that current deficits are spatially clustered and shaped by factors beyond household composition. They recommend embedding MDD as a tracked indicator in the ADP, POSHAN 2.0, and Saksham Anganwadi frameworks, prioritizing the identified co-cluster districts, and strengthening complementary feeding counseling within antenatal care.