Skip to main content
Download PDF

The relative importance of households as a source of variation in child malnutrition: a multilevel analysis in India

International Journal for Equity in Health volume 20, Article number: 225 (2021) Cite this article

Abstract

Background

Child malnutrition remains a major public health issue in India. Along with myriad upstream and social determinants of these adverse outcomes, recent studies have highlighted regional differences in mean child malnutrition rates. This research helps policy makers look between urban and rural communities and states to take a population-level approach to addressing the root causes of child malnutrition. However, one gap in this between-population approach has been the omission of households as a unit of analysis. Households could represent important sources of variation in child malnutrition within communities, districts, and states.

Methods

Using the fourth round of India’s National Family Health Survey from 2015 to 2016, we analyzed four and five-level multilevel models to estimate the proportion of variation in child malnutrition attributable to states, districts, communities, households, and children.

Results

Overall, we found that of the four levels that children were nested in (households, communities, districts, and states), the greatest proportion of variation in child height-for-age Z score, weight-for-age Z score, weight-for-height Z score, hemoglobin, birthweight, stunting, underweight, wasting, anemia, and low birthweight was attributable to households. Furthermore, we found that when the household level is omitted from models, the variance estimates for communities and children are overestimated.

Conclusions

These findings highlight the importance of households as an important source of clustering and variation in child malnutrition outcomes. As such, policies and interventions should address household-level social determinants, such as asset and social deprivations, in order to prevent poor child growth outcomes among the most vulnerable households in India.

Background

Child malnutrition remains a major public health issue in India. Often indicated by measures of child anthropometry and low hemoglobin levels, over 30% of the world’s stunted children (i.e. children with < − 2 SD height-for-age Z score) lived in India in 2017 [1]. Furthermore, the prevalence of child wasting (< − 2 SD weight-for-height Z score), underweight (< − 2 SD weight-for-age Z score), anemia (< 11.0 g/deciliter), and low birthweight (< 2500 g) was 15.7, 32.7, 59.7, and 21.4%, respectively [1, 2]. In addition to being associated with an increased risk of infectious disease [3], child malnutrition is linked with impaired cognitive development that can lead to poor long-term educational and economic outcomes [4]. Furthermore, mild to moderate child malnutrition is associated with increased child mortality [2].

Social determinants such as household socioeconomic status, intergenerational poverty, and inadequate environmental conditions are all associated with child malnutrition [5,6,7]. Additionally, some studies highlight regional differences in mean child malnutrition rates. For example, districts with high rates of stunting are clustered in north and central India [8]. Other studies have shown that states such as Tamil Nadu are high performing in terms of child malnutrition outcomes, while other states, such as Bihar, Uttar Pradesh, Odisha, Madhya Pradesh, and Gujarat perform quite poorly [9]. Furthermore, multilevel analyses reveal that variations in malnutrition are most attributable to states and urban and rural communities as opposed to districts [10, 11]. Corruption, variations in health policy and service delivery, and differences in health care spending are some of the reasons associated with disparate child malnutrition outcomes across states and communities in India [11,12,13].

However, one gap in these multilevel analyses and population approaches has been the omission of households as a unit of analysis. This omission may stem in part from the assumption that within-population distributions remain constant over time and across populations [14, 15]. As a result, the environments, such as households, that people are embedded in within communities, districts, or states are often disregarded. However, a strictly population-level approach does not account for heterogeneity at lower levels. This is demonstrated by the fact that over 93% of the variation in child height-for-age Z score (HAZ) in India was attributable to between-individual variations [11]. Similarly, another study showed that 80–85% of the variation in child hemoglobin and anthropometry in India was due to within-population differences [14]. Households are important sites of etiologic action given previously established relationships between housing characteristics, conditions, and amenities, and health [15]. As such, households could represent an important unit of investigation in multilevel analyses within communities, districts, and states.

Given this background, the purpose of this paper is to examine the proportion of variation in child stunting, wasting, underweight, anemia, and low birthweight attributable to the household level relative to commonly analyzed levels such as communities, districts, and states. We conducted this study using data from India’s most recent National Family Health Survey (NFHS), and to our knowledge, no other papers using this dataset have included households as a unit of analysis. This research is significant given that India’s National Nutrition Strategy (NNS) seeks to dramatically reduce the burden of child malnutrition in the coming years [16]. Achieving this goal will require an approach that targets at-risk individuals and households given how much of the variation in child malnutrition is attributable to lower levels [17,18,19,20] in addition to a population-level strategy to reduce average rates of child malnutrition across states in India [11, 21]. Therefore, this research is important as it builds on prior studies by examining households, a lower level of inferential targeting, as a possible source of variation in child malnutrition. Understanding the role households play could help policy makers intervene both between and within populations in order to improve child malnutrition throughout India.

Methods

The National Family Health Survey 2015–2016

We used the fourth round of the NFHS from 2015 to 2016 to conduct this study. Households were defined as a group of individuals who normally live together and take their meals from a common kitchen. Overall, this survey used a stratified two-stage sampling frame (states, and urban/rural areas within states) to select households and participants [22]. More specifically, households were selected from primary sampling units, defined as groups of adjacent households, which were villages in rural areas and census enumeration blocks in urban areas. We collectively refer to these village and census enumeration blocks as communities in this paper. As such, this dataset contains data from each of India’s 36 states/union territories, all 640 districts, 28,522 out of over 650,000 communities, 601,509 households, and 699,686 women between the ages of 15–49. For the purposes of this paper, the survey included data on a total of 259,627 children from 180,227 households.

Variables and sample sizes

The outcomes included in our analysis were child height-for-age Z score (HAZ), weight-for-height Z score (WHZ), weight-for-age Z score (WAZ), standardized hemoglobin measures (HB), and birthweight (measured in grams). We also included stunting (<− 2 SD HAZ), wasting (<− 2 SD WHZ), underweight (<− 2 SD WAZ), anemia (< 11.0 g/dL), and low birthweight (< 2500 g). The dataset had complete HAZ, WAZ, and WHZ data for 225,002 children from 164,664 households. The dataset had complete HB data for 209,496 children from 157,746 households, and birthweight data for 184,852 children from 140,572 households. Sample sizes are fully described in Fig. 1 below. The number of states/union territories [23] and districts (640) stayed the same in all the analyses hence they are not reported in the figure.

Fig. 1
figure 1

Flow chart of sample sizes for primary and secondary analyses for all outcomes

Full size image

Covariates included 19 risk factors known to be associated with child malnutrition [7, 24, 25]. We included three different nutrition variables, early breastfeeding initiation, vitamin A supplementation (only asked in reference to children between 6 and 59 months), and the use of iodized salt. Each of these three variables was dichotomized yes/no. We included four environmental variables, household access to improved sanitation, household use of clean cooking fuel, safe child stool disposal, and household access to improved drinking water. These variables were also dichotomized yes/no. A total of six health coverage variables were included. These were whether the child experienced an infectious disease in the past two weeks, complete child vaccinations, presence of skilled birth attendant at birth, provision of oral rehydration therapy after diarrhea, care seeking for pneumonia, and whether family planning needs were met. All of these variables were dichotomized yes/no. Household wealth, mother’s level of education, and child’s birth order were included as socioeconomic risk factors. Households were dichotomized by either being in the poorest wealth quintile, or in all other quintiles. Mother’s education was dichotomized based on whether the mother had received no education or at least primary education. We dichotomized birth order as before or after the 6th birth. Finally, we included three maternal characteristic risk factors which were mother’s age at marriage, maternal height, and maternal body mass index (BMI). Age at marriage was dichotomized as above or below 18. Mother’s height was dichotomized as above and below 145 cm, and mother’s BMI was dichotomized as above and below 18.5 kg/m2.

After performing listwise deletion for all missing observations for the above covariates, our fully adjusted sample sizes were 204,979 children with HAZ, WAZ, and WHZ outcome data, 190,227 children with HB outcome data, and 167,838 children with birthweight outcome data.

Statistical analysis

Our statistical analysis consisted of a series of four- and five- level models in order to compare the addition of households as a level of analysis. Additionally, each set of models included an ‘unadjusted’ and ‘adjusted’ model. The unadjusted models contained only child age (in months) and sex, while the fully adjusted models contained all 19 risk factors outlined above. Further, our analysis was composed of primary and secondary analyses. The primary analysis included the full sample of households, with varying numbers of children per household, whereas the secondary analysis focused on a restricted subsample composed of only those households with more than one child. This approach was taken in order to examine whether household variance estimates were attenuated by between-child estimates in households with more than one child.

Continuous outcomes

We used multilevel modeling to decompose the proportion of variation in our continuous outcomes – HAZ, WHZ, WAZ, HB, and birthweight – attributable to children at level one, nested in urban/rural communities at level two, districts at level three, and states at level four. Multilevel modeling is a statistical methodology commonly used in the field of public health to elucidate the effects of both compositional and contextual factors on health [26,27,28]. For this four-level model, we estimated equation (1), which took the basic form Yijkl = β0 + β1Xijkl + (e0ijkl + u0jkl + v0kl + f0l) where Yijkl is one of the outcomes for child i nested in community j, district k, and state l. In this model, Xijkl is a vector of covariates, and the random effects e0ijkl, u0jkl, v0kl, and f0l are the residual differentials for children, communities, districts, and states, respectively. We then included households such that we decomposed the proportion of variation in the same continuous outcomes attributable to children at level one, nested in households at level two, communities at level three, districts at level four, and states at level five. For this five-level model, we estimated equation (2), which took the basic form Yijklm = β0 + β1Xijkl. + (e0iyjkl + h0yjkl + u0jkl + v0kl + f0l) where Yiyjkl is one of the outcomes for child i nested in household y, community j, district k, and state l. In this model, Xiyjkl is a vector of covariates, and the random effects e0iyjkl, h0yjkl, u0jkl, v0kl, and f0l are the residual differentials for children, households, communities, districts, and states, respectively. In both the four-level and five-level models, the residual differentials for children, households, communities, districts, and states are assumed to be normally distributed with a mean of 0 and a variance of \({\sigma}_{e0}^2\), \({\sigma}_{h0}^2\), \({\sigma}_{u0}^2\), \({\sigma}_{v0}^2\), and \({\sigma}_{f0}^2\), respectively. The variances in the four and five-level models are the parameters of interest and signify the between-child (\({\sigma}_{e0}^2\Big)\), between-household (\({\sigma}_{h0}^2\)), between-community (\({\sigma}_{u0}^2\)), between-district (\({\sigma}_{v0}^2\Big)\), and between-state (\({\sigma}_{f0}^2\)) variations in child i experiencing the outcome. Finally, we repeated the five-level analysis for a subsample of households with more than one child.

Binary outcomes

For the binary outcomes stunting, wasting, underweight, anemia, and low birthweight, we estimated four level models for the probability of a child i, in community j, in district k, in state l experiencing the outcome Yijkl = 1 as equation (3) logit(πijkl) = β0 + β1Xijkl + (u0jkl + v0kl + f0l), where πijkl is the log odds of the outcome in child i, Xijkl is a vector of covariates, and the random effects are the residual differentials for communities (u0jkl), districts (v0kl), and states (f0l). We then added households in order to decompose the proportion of variation in the same binary outcomes attributable to children at level one, nested in households at level two, communities at level three, districts at level four, and states at level five. For this five-level model, we estimated the probability of child i, in household y, in community j, in district k, in state l experiencing the outcome Yiyjkl = 1 as equation (4) logit(πiyjkl) = β0 + β1Xiyjkl + (hyjkl + u0jkl + v0kl + f0l). In this case, the random effects hyjkl, u0jkl, v0kl, and f0l are the residual differentials for households, communities, districts, and states, respectively. The same assumptions and parameter definitions used for equations (1) and (2) were applied to equations (3) and (4). However, the variance of the lowest levels cannot be estimated when considering binary outcomes. As such, the child-level random effect and variance is not shown. Again, we repeated the five-level analysis for households with more than one child.

We used the Monte Carlo Markov Chain (MCMC) method with a burn-in of 5000 cycles monitored over 50,000 chain iterations in MLwiN 3.05 software to conduct the analysis and produce the estimates for the continuous and binary outcomes in the four and five level models [29, 30].

Results

Sample characteristics

In the full sample of the 225,002 children with anthropometric data, the mean HAZ value was − 1.48 (std. dev. 1.68). The average WAZ and WHZ values were − 1.52 (std. dev. 1.22) and − 0.97 (std. dev. 1.39), respectively. Approximately 38.3% of the children were stunted, 20.4% experienced wasting, and 34.5% were underweight. The average HAZ, WAZ, and WHZ values among the 128,197 children in households with more than one child were − 1.59 (std. dev. 1.69), − 1.60 (std. dev. 1.21), and − 0.98 (std. dev. 1.37). In the restricted sample, 41.2% of children were stunted, 36.7% were underweight, and 20.4% experienced wasting.

Of the full sample of 209,496 children with hemoglobin data, the average HB value was 10.6 (std. dev. 1.54). Approximately 57.5% of the children in our sample were anemic. In the restricted sample, the mean HB value was 10.5 (std. dev. 1.55) among the 119,668 children with complete hemoglobin data, and approximately 58.1% of the children were anemic. Finally, the average birthweight was 2.82 kg among the 184,852 children with birthweight data. Approximately 17.7% of the children were classified as having low birthweight. In the restricted sample, the mean birthweight was 2.81 kg among the 99,667 children with birthweight data, and 17.9% were classified as having low birthweight.

Household level variance for continuous outcomes

We estimated the household level variance for HAZ, WHZ, WAZ, HB, and birthweight. These results are presented in Table 1. Overall, we found that the unadjusted and adjusted household variance estimates for HAZ were 0.47 (0.01) and 0.41 (0.01), respectively. For WHZ, the unadjusted and adjusted household variance estimates were 0.32 (0.007) and 0.30 (0.008). For WAZ, the unadjusted and adjusted household variance estimates were 0.36 (0.005) and 0.31 (0.005), respectively. For hemoglobin, the unadjusted household variance estimate was 0.33 (0.008), while the adjusted estimate was 0.32 (0.008). Finally, the unadjusted and adjusted household variance estimates for birthweight were 0.086 (0.002) and 0.082 (0.002), respectively. Overall, we found that except for the child level, the largest proportion of variation was attributable to households for HAZ, WHZ, WAZ, HB, and birthweight.

Table 1 Four and five-level variance estimates (standard error) for continuous anthropometric, hemoglobin, and birthweight outcomes
Full size table

Household level variance for categorical outcomes

We estimated the household level variance for stunting, wasting, underweight, anemia, and low birthweight. These results are presented in Table 2. The unadjusted and adjusted household variance estimates for stunting were 0.27 (0.013) and 0.74 (0.09), respectively. For wasting, the unadjusted household variance estimate was 0.89 (0.04), and 0.87 (0.05) in the adjusted model. The unadjusted and adjusted household variance estimates for underweight were 1.29 (0.04) and 0.93 (0.35), respectively. The unadjusted and adjusted household variance estimates were both 0 (0) for anemia. Finally, for low birthweight, the unadjusted and adjusted household variance estimates were 1.49 (0.24) and 0.45 (0.48), respectively.

Table 2 Four and five-level variance estimates (standard error) for stunting, wasting, underweight, anemia, and low birthweight
Full size table

Comparing four level and five level variance estimates

We found that state and district variance estimates remained stable between the four and five-level models for each of the continuous outcomes. However, we found differences in the community variance estimates between the fully adjusted four-level and fully adjusted five-level models for all five continuous outcomes. These differences are presented in Table 3. For HAZ, the four-level adjusted community-level variance estimate was 17% higher than the five-level adjusted community variance estimate. For WHZ, the four-level adjusted community-level variance estimate was 15% higher than the five-level adjusted community variance estimate. For WAZ, the four-level adjusted community-level variance estimate was 25% higher than the five-level adjusted community variance estimate. For HB, the four-level adjusted community-level variance estimate was 11% higher than the five-level adjusted community variance estimate. For birthweight, the four-level adjusted community-level variance estimate was 50% higher than the five-level adjusted community variance estimate.

Table 3 Differences between four and five level variance estimates in adjusted models for continuous outcomes
Full size table

Similarly, we found that the adjusted four-level models overestimated the child-level variance for all five outcomes when compared to the adjusted five-level models. For HAZ, the four-level adjusted child-level variance estimate was 16% higher than the five-level adjusted community variance estimate. For WHZ, the four-level adjusted child-level variance estimate was17% higher than the five-level adjusted community variance estimate. For WAZ, the four-level adjusted child-level variance estimate was 25% higher than the five-level adjusted community variance estimate. For HB, the four-level adjusted child-level variance estimate was 17% higher than the five-level adjusted community variance estimate. Finally, for birthweight, the four-level adjusted child-level variance estimate was 26% higher than the five-level adjusted community variance estimate.

There were also differences in the variance estimates between the fully adjusted four-level and fully adjusted five-level models for the binary outcomes. For stunting, the state, district, and community variance estimates in the fully adjusted four-level model were 0.04 (0.012), 0.05 (0.004), and 0.19 (0.008), respectively, compared to 0.05 (0.015), 0.06 (0.006), and 0.19 (0.01) in the fully adjusted five-level model. For wasting, the state, district, and community variance estimates were 0.14 (0.04), 0.12 (0.009), and 0.29 (0.01), respectively, compared to 0.18 (0.05), 0.16 (0.01), and 0.31 (0.01) in the fully adjusted five-level model. For underweight, the state, district, and community variances in the fully adjusted four-level model were 0.15 (0.04), 0.07 (0.005), and 0.17 (0.008), respectively, compared to 0.19 (0.06), 0.09 (0.01), and 0.15 (0.01) in the fully adjusted five-level model. For low birthweight, the state, district, and community variances in the fully adjusted four-level model were 0.19 (0.06), 0.07 (0.006), and 0.28 (0.014), respectively, compared to 0.21 (0.06), 0.08 (0.012), and 0.27 (0.019) in the fully adjusted five-level model.

Households with more than one child

We estimated the household level variance for all five continuous and binary outcomes in a subsample of the population containing homes with more than one child. These results are presented in Table 4. The unadjusted and adjusted household variance estimates for HAZ were 0.49 (0.01) and 0.42 (0.012), respectively. For WHZ, the unadjusted and adjusted household variance estimates were both 0.29 (0.008). For WAZ, the unadjusted and adjusted household variance estimates were 0.36 (0.006) and 0.31 (0.006), respectively. For HB, the unadjusted household variance estimate was 0.36 (0.006), while the adjusted estimate was 0.35 (0.01). The unadjusted and adjusted household variance estimates for birthweight were both 0.09 (0.002).

Table 4 Five level variance estimates (standard error) for households with more than one child
Full size table

For stunting, the unadjusted and adjusted household variance estimates were 0.88 (0.03) and 0.76 (0.04), respectively. The unadjusted and adjusted household variance estimates for wasting were 0.33 (0.02) and 0.31 (0.02). For underweight, the unadjusted household variance estimate was 1.31 (0.04) compared to 1.13 (0.05) in the adjusted model. For anemia, the unadjusted household variance estimate was 0.017 (0.001) while the adjusted variance was 0 (0). Finally, for birthweight, the unadjusted and adjusted household variance estimates were 1.57 (0.07) and 1.52 (0.08), respectively. Table 5 below shows a comparison between the household-level variance estimates between the full sample and households with more than one child.

Table 5 Comparison of household-level variance estimates between full sample and households with more than one child
Full size table

Discussion

This study had three salient findings. First, in the four-level models, the majority of the variance was attributable to the community and child levels. However, in the five-level models, of the four levels that children were nested in (households, communities, districts, and states), the greatest proportion of variation in child HAZ, WAZ, WHZ, HB, and birthweight was attributable to households. This was also the case for stunting, wasting, underweight, and low birthweight. It is possible this was not the case for anemia given the high prevalence of anemia in the overall sample, and because over 50% of the households with more than one child either had all children anemic or none anemic (described in supplementary Table 1). Second, we found that community and child-level variance estimates are overestimated in four-level models that omit the household level for HAZ, WAZ, WHZ, HB, and birthweight. Third, while the greatest proportion of the variation was still attributable to households in the secondary analysis, there was no substantial attenuation in the variance estimates when compared to those from the primary analysis in any of the adjusted models. However, the five-level adjusted variance estimates for underweight and low birthweight (< 2500 g) were higher in the secondary analysis than in the primary analysis, possibly due to the reduced number of households in the sample.

There are four data limitations with this study. First, while NFHS data are considered extremely high quality and representative [31], it should be noted that questions about the covariates included in this study were self-reported and not verified by enumerators, a potential source of measurement error. Child anthropometric data, however, were directly measured and collected by trained field staff, and birthweight was collected using birth records. Second, there were missing observations for the outcome measures and the covariates that were included in the study. These missing observations were excluded from the study, potentially biasing the variance estimates given that the missingness might not have occurred randomly. Third, these are secondary data that were not explicitly collected for the purposes of this study. Fourth, only 5% of the households in our sample had more than one mother. As such, we were unable to disentangle the proportion of variation attributable to households versus mothers within a household. However, supplementary Table 2 shows that for HAZ, the variance estimates are very similar between 5-level and 6-level models, and 6-level models with only households with more than one mother.

Our findings are policy-relevant for several reasons. Decentralization initiated in the early 1990s has expanded the role of states and districts in designing and implementing policies. As an example, the National Nutrition Strategy, implemented by the National Institution for Transforming India, implements district-level interventions as a means to improving child nutrition outcomes [16]. Yet, our findings show that with the exception of HAZ and stunting, a greater proportion of variation in child malnutrition is attributable to states and communities than districts. This is confirmed by our fully adjusted four and five-level models. These findings are consistent with findings from prior studies that also show the relative importance of states and communities compared to districts when examining between-population differences in child malnutrition [11, 13, 32].

Moreover, findings from this paper underscore the importance of households as an important source of clustering and variation in child malnutrition outcomes. This has also been found previously [33], albeit in a different context, and is exemplified by the fact that of the four levels children are nested in, households accounted for the largest source of variation for all outcomes in both the unadjusted and adjusted models. This is further reflected by the fact that when the household level is omitted in the four-level models, the variance estimates at the community and child levels are overestimated by 11 to 50% and 16 to 26%, respectively. The implications of this finding are that between-population differences, such as community, district, or state level contextual factors, are still important determinants of child health. Yet the environments within those geographic levels that people are embedded in, such as households, also function as sources of variation in child malnutrition. This is a particularly important implication given the heterogeneity of child malnutrition outcomes within geographic levels [32]. Therefore, our findings signal the importance of designing policies and interventions that account for the most vulnerable individuals and households within a population in addition to addressing contextual factors at the state and community levels aimed at lowering mean rates of child malnutrition between populations [17,18,19,20]. Doing so could help alleviate between and within population disparities in child malnutrition throughout India.

That our findings highlight households as a significant source of variation in child malnutrition outcomes should not undermine efforts aimed at addressing the upstream and social determinants of child malnutrition. This is an important point given that the utility of multilevel modeling is to draw attention away from individuals and households in order to examine place-based effects on health [26,27,28]. However, underscoring households as important sites of clustering is not at odds with a social epidemiological approach to improving child malnutrition outcomes. This is highlighted by a growing body of literature which demonstrates the relative importance of upstream factors within households when compared to individual-level factors in determining child malnutrition. For example, recent studies show the relative importance of household socioeconomic status and maternal characteristics over nutrition-specific risk factors, such as child feeding practices [5, 24, 25, 34, 35]. Previous work also suggests that investing in early nutrition for girls at the household level is important given that maternal height is strongly associated with child height in the subsequent generation [7, 23, 36]. To this end, a recent study found that children born to mothers who received consistent meals in school had significantly higher HAZ scores than children born to mothers who did not receive this benefit [37]. Another recent study from Odisha shows that providing conditional cash transfers to mothers in poor households can also lead to significant reductions in stunting and anemia among children under five [38]. Evidence suggests that cash transfers can empower women who then invest in the health and wellbeing of their children [8, 24, 39]. Future research should examine the associations between other housing environmental conditions, such as sanitation, clean cooking fuel, and refrigeration, all of which are associated with faltering child growth outcomes [40,41,42].

Conclusion

The purpose of this paper was to examine the proportion of variation in child stunting, wasting, underweight, anemia, and low birthweight attributable to the household level relative to commonly analyzed levels such as communities, districts, and states. Overall, we found that of the four levels that children were nested in (households, communities, districts, and states), the greatest proportion of variation in child HAZ, WAZ, WHZ, hemoglobin, birthweight, stunting, underweight, wasting, anemia, and low birthweight was attributable to households. Furthermore, we found that when the household level is omitted from models, the variance estimates for communities and children are overestimated. This implies that households are an important lower-level responsible for a considerable proportion of the variation in child malnutrition.

Availability of data and materials

The data that support the findings of this study are openly available in India: Standard DHS, 2015–2016 Dataset at https://dhsprogram.com/data/dataset/India_Standard-DHS_2015.cfm?flag=0.

References

  1. 2018 Global Nutrition Report [Internet]. [cited 2021 Jan 31]. Available from: https://globalnutritionreport.org/reports/global-nutrition-report-2018/.

  2. Swaminathan S, Hemalatha R, Pandey A, Kassebaum NJ, Laxmaiah A, Longvah T, et al. The burden of child and maternal malnutrition and trends in its indicators in the states of India: the global burden of disease study 1990–2017. Lancet Child Adolesc Health. 2019;3(12):855–70.

    Article  Google Scholar 

  3. Black RE, Morris SS, Bryce J. Where and why are 10 million children dying every year? Lancet. 2003;361(9376):2226–34.

    Article  Google Scholar 

  4. Grantham-McGregor S, Cheung YB, Cueto S, Glewwe P, Richter L, Strupp B. Developmental potential in the first 5 years for children in developing countries. Lancet. 2007;369(9555):60–70.

    Article  Google Scholar 

  5. Corsi DJ, Mejía-Guevara I, Subramanian SV. Risk factors for chronic undernutrition among children in India: estimating relative importance, population attributable risk and fractions. Soc Sci Med. 2016;157:165–85.

    Article  Google Scholar 

  6. Martorell R, Young MF. Patterns of stunting and wasting: potential explanatory factors. Adv Nutr. 2012;3(2):227–33.

    Article  Google Scholar 

  7. Kim R, Rajpal S, Joe W, Corsi DJ, Sankar R, Kumar A, et al. Assessing associational strength of 23 correlates of child anthropometric failure: an econometric analysis of the 2015-2016 National Family Health Survey, India. Soc Sci Med. 2019;238:112374.

    Article  Google Scholar 

  8. Menon P, Headey D, Avula R, Nguyen PH. Understanding the geographical burden of stunting in India: a regression-decomposition analysis of district-level data from 2015–16. Matern Child Nutr. 2018;14(4):e12620.

    Article  Google Scholar 

  9. Cavatorta E, Shankar B, Flores-Martinez A. Explaining cross-state disparities in child nutrition in rural India. World Dev. 2015;76:216–37.

    Article  Google Scholar 

  10. Ackerson LK, Kawachi I, Barbeau EM, Subramanian SV. Geography of underweight and overweight among women in India: a multilevel analysis of 3204 neighborhoods in 26 states. Econ Hum Biol. 2008;6(2):264–80.

    Article  Google Scholar 

  11. Mejía-Guevara I, Krishna A, Corsi DJ, Subramanian SV. Individual and ecological variation in child undernutrition in India: a multilevel analysis. J South Asian Dev. 2015;10(2):168–98.

    Article  Google Scholar 

  12. Balarajan Y, Selvaraj S, Subramanian S. Health care and equity in India. Lancet. 2011;377(9764):505–15.

    CAS  Article  Google Scholar 

  13. Mohanty SK, Kim R, Khan PK, Subramanian SV. Geographic variation in household and catastrophic health spending in India: assessing the relative importance of villages, districts, and states, 2011-2012: geographies of household and catastrophic health spending in India. Milbank Q. 2018;96(1):167–206.

    Article  Google Scholar 

  14. Rodgers J, Kim R, Subramanian SV. Explaining Within- vs Between-Population Variation in Child Anthropometry and Hemoglobin Measures in India: A Multilevel Analysis of the National Family Health Survey 2015–2016. J Epidemiol [Internet]. 2019 [cited 2020 May 15]; Available from: https://www.jstage.jst.go.jp/article/jea/advpub/0/advpub_JE20190064/_article.

  15. Galobardes B. Indicators of socioeconomic position (part 1). J Epidemiol Community Health. 2006;60(1):7–12.

    Article  Google Scholar 

  16. Kumar A, Kaur S, Singh A, Menon P. Accelerating progress on nutrition in India: What Will It Take? 2020. https://www.niti.gov.in/sites/default/files/2020-10/AbhiyaanMonitoringReport.pdf.

  17. Rose G. Sick individuals and sick populations. Int J Epidemiol. 2001;30(3):427–32.

    CAS  Article  Google Scholar 

  18. Mejía-Guevara I, Corsi DJ, Perkins JM, Kim R, Subramanian SV. Variation in anthropometric status and growth failure in low- and middle-income countries. Pediatrics. 2018;141(3):e20172183.

    Article  Google Scholar 

  19. Subramanian SV, Kim R, Christakis NA. The “average” treatment effect: a construct ripe for retirement. A commentary on Deaton and cartwright. Soc Sci Med. 2018;210:77–82.

    CAS  Article  Google Scholar 

  20. Subramanian SV, Mejía-Guevara I, Krishna A. Rethinking policy perspectives on childhood stunting: time to formulate a structural and multifactorial strategy: rethinking policy perspectives on stunting. Matern Child Nutr. 2016;12:219–36.

    Article  Google Scholar 

  21. Subramanian SV, Kim R. Sick countries and sick individuals: reassessing the inferential targets in global health research. Lancet Glob Health. 2018;6(7):e726–7.

    CAS  Article  Google Scholar 

  22. International Institute for Population Sciences (IIPS) and ICF. National Family Health Survey (NFHS-4), 2015-16. Mumbai: IIPS; 2017.

  23. Subramanian SV. Association of maternal height with child mortality, anthropometric failure, and Anemia in India. JAMA. 2009;301(16):1691.

    CAS  Article  Google Scholar 

  24. Bhutta ZA, Ahmed T, Black RE, Cousens S, Dewey K, Giugliani E, et al. What works? Interventions for maternal and child undernutrition and survival. Lancet. 2008;371(9610):417–40.

    Article  Google Scholar 

  25. Ruel MT, Alderman H. Nutrition-sensitive interventions and programmes: how can they help to accelerate progress in improving maternal and child nutrition? Lancet. 2013;382(9891):536–51.

    Article  Google Scholar 

  26. Krieger N. A glossary for social epidemiology. J Epidemiol Community Health. 2001;55(10):693–700.

    CAS  Article  Google Scholar 

  27. Diez-Roux AV. Bringing context back into epidemiology: variables and fallacies in multilevel analysis. Am J Public Health. 1998;88(2):216–22.

    CAS  Article  Google Scholar 

  28. Berkman LF, Kawachi I, editors. Social epidemiology. New York: Oxford University Press; 2000. 391 p.

    Google Scholar 

  29. Browne WJ. MCMC estimation in MLwiN v3.03: University of Bristol: Centre for Multilevel Modelling; 2019.

    Google Scholar 

  30. Charlton C, Rabash J, Browne WJ, Healy M, Cameron B. MLwiN Version 3.05: University of Bristol: Centre for Multilevel Modelling; 2020.

    Google Scholar 

  31. Corsi DJ, Neuman M, Finlay JE, Subramanian SV. Demographic and health surveys: a profile. Int J Epidemiol. 2012;41(6):1602–13.

    Article  Google Scholar 

  32. Rajpal S, Kim J, Joe W, Kim R, Subramanian SV. Small area variation in child undernutrition across 640 districts and 543 parliamentary constituencies in India. Sci Rep. 2021;11(1):4558.

    CAS  Article  Google Scholar 

  33. Vaezghasemi M, Ng N, Eriksson M, Subramanian SV. Households, the omitted level in contextual analysis: disentangling the relative influence of households and districts on the variation of BMI about two decades in Indonesia. Int J Equity Health. 2016;15(1):102.

    Article  Google Scholar 

  34. Black RE, Allen LH, Bhutta ZA, Caulfield LE, de Onis M, Ezzati M, et al. Maternal and child undernutrition: global and regional exposures and health consequences. Lancet. 2008;371(9608):243–60.

    Article  Google Scholar 

  35. Kim R, Mejía-Guevara I, Corsi DJ, Aguayo VM, Subramanian SV. Relative importance of 13 correlates of child stunting in South Asia: insights from nationally representative data from Afghanistan, Bangladesh, India, Nepal, and Pakistan. Soc Sci Med. 2017;187:144–54.

    Article  Google Scholar 

  36. Perkins JM, Subramanian SV, Davey Smith G, Özaltin E. Adult height, nutrition, and population health. Nutr Rev. 2016;74(3):149–65.

    Article  Google Scholar 

  37. Chakrabarti S, Scott SP, Alderman H, Menon P, Gilligan DO. Intergenerational nutrition benefits of India’s national school feeding program. Nat Commun. 2021;12(1):4248.

    CAS  Article  Google Scholar 

  38. Chakrabarti S, Pan A, Singh P. Maternal and child health benefits of the Mamata conditional cash transfer program in Odisha, India. J Nutri. 2021;151(8):2271-81. https://doi.org/10.1093/jn/nxab129.

  39. Branca F, Piwoz E, Schultink W, Sullivan LM. Nutrition and health in women, children, and adolescent girls. BMJ. 2015;14:h4173.

    Article  Google Scholar 

  40. Prüss-Üstün A, Bos R, Gore F, Bartram J, World Health Organization. Safe water, better health: costs, benefits and sustainability of interventions to protect and promote health [Internet]. 2008 [cited 2021 Jan 25]. Available from: http://whqlibdoc.who.int/publications/2008/9789241596435_eng.pdf

  41. Fink G, Günther I, Hill K. The effect of water and sanitation on child health: evidence from the demographic and health surveys 1986–2007. Int J Epidemiol. 2011;40(5):1196–204.

    Article  Google Scholar 

  42. Karlsson O, Kim R, Joe W, Subramanian SV. The relationship of household assets and amenities with child health outcomes: an exploratory cross-sectional study in India 2015–2016. SSM - Popul Health. 2020;10:100513.

    Article  Google Scholar 

Download references

Acknowledgments

The authors would like to thank the Demographic and Health Surveys program for making the fourth National Family Health Survey data freely accessible.

Funding

This research was funded by Bill & Melinda Gates Foundation, INV-002992.

Author information

Authors and Affiliations

  1. Global Health & Social Medicine, Harvard Medical School, Boston, MA, 02115, USA

    Anoop Jain

  2. Harvard Center for Population and Development Studies, Cambridge, MA, 02138, USA

    Justin Rodgers

  3. Division of Health Policy & Management, College of Health Science, Korea University, Seoul, 02841, South Korea

    Rockli Kim

  4. Interdisciplinary Program in Precision Public Health, Department of Public Health Sciences, Graduate School of Korea University, Seoul, 02841, South Korea

    Rockli Kim

  5. Harvard Center for Population and Development Studies, MA, 02138, Cambridge, USA

    S. V. Subramanian

  6. Department of Social and Behavioral Sciences, Harvard T.H. Chan School of Public Health, Boston, MA, 02115, USA

    S. V. Subramanian

Authors
  1. Anoop Jain
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Justin Rodgers
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Rockli Kim
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. S. V. Subramanian
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

Conceptualization and Design: SVS, AJ, JR, RK. Data Acquisition and Analysis: AJ. Data Interpretation: AJ, SVS, JR, RK. Drafting of the Manuscript: AJ. Critical revisions to Manuscript: SVS, JR, RK. Overall Supervision: SVS. All authors read and approved the final manuscript.

Corresponding author

Correspondence to S. V. Subramanian.

Ethics declarations

Ethics approval

NA

Consent for publication

NA

Competing interests

The authors have no competing interests to disclose.

Additional information

Publisher’s Note

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

Supplementary Information

Additional file 1: Supplementary Table 1

: Prevalence of anemia among children in households with more than one child.

Additional file 2: Supplementary Table 2

: Comparison of variance estimates for HAZ between 5-level and 6-level models, and 6-level models with only households with more than one mother.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Jain, A., Rodgers, J., Kim, R. et al. The relative importance of households as a source of variation in child malnutrition: a multilevel analysis in India. Int J Equity Health 20, 225 (2021). https://doi.org/10.1186/s12939-021-01563-7

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12939-021-01563-7

Keywords

  • Social determinants
  • Child malnutrition
  • India
  • Multilevel modeling
  • Population health
  • Omitted level
  • Households

International Journal for Equity in Health

ISSN: 1475-9276

Contact us