Water insecurity potentially undermines dietary diversity of children aged 6-23 months: Evidence from India.

Abstract Dietary diversity is a crucial pathway to child nutrition; lack of diversity may deprive children of critical macro and micronutrients. Though water along with hygiene and sanitation is a known driver of child undernutrition, a more direct role of household water in shaping dietary diversity remains unexplored. Existing literature provides a sound theoretical basis to expect that water could affect dietary diversity among young children. Here, we test the proposition that suboptimal household access to water and low regional water availability associate with lower dietary diversity among young children. Using the nationally representative 2015–2016 India Demographic and Health Survey data, we conducted a probit analysis on the sample of 69,841 children aged 6–23 months to predict the probability that a child achieves minimum standards of dietary diversity (MDD). After controlling for relevant socioeconomic and gender‐related covariates, we found that children in household with suboptimal household water access were two percentage points less likely to achieve MDD, when compared with those from households with optimal water access. Children in high water availability regions had nine percentage points greater probability of achieving MDD compared with children from low water availability regions, accounting for household water access. As dietary diversity is central to nutrition, establishing the role of water access in shaping early childhood dietary diversity broadens the framework on how household material poverty shapes child malnutrition—independent of sanitation and hygiene pathways. This provides additional window for nutrition planning and intervention wherein water‐based strategies can be leveraged in multiple ways.

continue exclusive breastfeeding until 6 months of age (World Health Organization [WHO], 2003). Starting at 6 months, mother's milk alone cannot meet the growing infant's caloric and micronutrient needs (WHO, 2001). Therefore, gradual introduction of optimally diversified foods is recommended at 6 months, alongside continued breastfeeding up to 2 years of age (Na, Aguayo, Arimond, & Stewart, 2017;Black et al., 2008;Begin et al., 1999). Adequate and diverse complementary feeding among young children is known to be the most important window to mitigate early childhood malnutrition (Na, Aguayo, Arimond, & Stewart, 2017).
Dietary diversity among infants and young children has been associated with maternal and child characteristics, household resources, and other sociocultural factors (Rakotonirainy et al., 2018;Kumera, Tsedal & Ayana, 2018;Santoso et al. 2019). Maternal education, awareness, and knowledge about child feeding practices are recognized factors affecting child's dietary diversity (Ickes et al., 2015).
Among household resources, access to adequate and safe water is known to shape child nutrition and health outcomes, primarily through sanitation and hygiene pathways (Cumming & Cairncross, 2016;Subbaraman & Murthy 2015;Dewey & Mayers, 2011). However, a more direct role of water in child nutrition-specifically in infant and young child dietary diversity-remains largely unexplored. This is despite the fact that the literature linking water and food security does provide a sound basis for theorizing the possible role that household water might have in shaping child's access to dietary diversity.
Based on this literature, we can broadly identify three ways in which water might affect child's dietary diversity. First, regional water availability can affect child's dietary diversity, through its effect on household food security (Misra, 2014;United Nations Educational, Scientific, and Cultural Organization, 2012). This association has primarily been viewed from a community or productionist perspective, emphasizing production or availability of foods in the wake of growing demographic pressure and climate change (Sabo et al., 2017;Suweis, Rinaldo, Maritan, & D'Odorico, 2013;Qureshi, Hanjra, & Ward, 2013).
Water availability affects household food diversity through food choice because in water-scarce contexts, people may prefer to grow and eat less water-intensive food (Mabhaudhi et al., 2016;High Level Panel of Experts [HLPE], 2015;Field, 1987). Water-rich regions have opportunities for greater crop diversification and multicropping because they can more easily grow foods such as soybean, groundnut, and maize (Mabhaudhi et al., 2016). This increases the diversity of foods available in the food environment for purchase in the market and grown on-plot within the household and reduces child malnutrition (Bezner-Kerr, Berti, & Shumba 2011;HLPE, 2015). Greater local availability of water also increases the likelihood of households having water-based animal foods such as fish in their diet (Hoekstra, 2012), thus increasing consumption diversity. Second, at the household level, water access can affect dietary diversity through budget constraints that may force low-resource households to choose between buying water, food, or fuel. When they purchase water, this leaves insufficient funds for acquiring an adequate and diverse diet (Cumming & Cairncross, 2016;Collins et al., 2019;Mason, 2015). Third, water at the household level may affect child's dietary diversity because water is very important in preparation of household foods Wutich & Brewis, 2014). Complementary foods have high water requirements (Wichmann & Voyi, 2006). Mothers have reported increased demand for water during the first 1,000 days of child life, for preparation of complementary foods and breastfeeding (Collins et al., 2019). An assessment of recipes for complementary foods prepared in low-and middle-income countries identified water as a critical ingredient commonly comprising 70% to as high as 90% of ingredient proportions (Ferguson & Lehrfeld, 1998).
Not having enough water to prepare complementary foods in this appropriate manner was the main reason cited for feeding infants lesspreferred foods in substitution in a recent cross-cultural study of 19 sites in 16 low-and middle-income countries (Schuster, Butler, Wutich, Miller, Young, & HWISE Consortium, n.d.). All these waterfood and child feeding linkages are likely mediated by gender role arrangements given that women are usually managing both household water (Das & Safini, 2018) and child care (Rakotonirainy et al., 2018) duties in many parts of the world, including India.
Accordingly, we propose that inadequate availability and access to water is a key factor affecting child's consumption of minimum standard of dietary diversity-consuming foods from at least four of seven food groups as per WHO (2008). We test the three pathways outlined above that may link water insecurity to child's dietary diversity using large-scale, nationally representative data for India. First, we test the proposition that low household water access lowers a child's achievement of minimum dietary diversity (MDD) standards. We also test that the relative effect of water access on households varies based on the general diversity of diets. Here, we engage primarily in theory building because we cannot precisely predict about the direction of this effect, for example, the suppressive effect of inadequate water access may be more pronounced for diets already higher or lower in diversity. Second, we test the proposition that low regional water availability is associated with lower dietary diversity. Notably, these proposed pathways are distinct from the WASH (water grouped with sanitation and hygiene) mechanism that focuses on effects of safe drinking water, sanitation, and hygiene practices on child nutrition (e.g., Troeger et al., 2018;Gautam et al., 2017;Wolf et al., 2014;Fink et al., 2011;Fewtrell et al., 2007;Torlesse et al., 2016).

Key messages
• Suboptimal household water access is associated with failure to meet minimum dietary diversity standards among children aged 6-23 months in India.
• Children living in regions of higher water availability are relatively buffered.
• This role of water in shaping access to dietary diversity among children can be leveraged for nutritional improvements, given that dietary diversity is central to child nutrition.

| Water and dietary diversity in India
India is home to the largest number of malnourished children in absolute terms. Nearly 38% of children under 5 years of age in India are stunted, or too short for their age, a commonly accepted indicator of childhood nutrition (Global Nutrition Report, 2018). Access of children to a minimally diverse diet is also very low in India; only 22% of children aged Child's access to minimally diverse diet and nutrition is also variable across regions of India. In terms of child nutrition, southern states of India perform better than northern and central states (IIPS & ICF, 2017). Western states (except Goa) have mediocre achievement in this regard. Eastern states have very high level of child undernutrition (Pathak & Singh, 2011). Similar variation is reflected in child's access to dietary diversity (Raykar, Majumder, Laxminarayan, & Menon, 2015). Southern and western states have generally higher levels of dietary diversity compared with eastern and northern states (Borkotoky, Unisa, & Gupta, 2018;Das, 2014;IIPS & ICF, 2017).
Access to safe water is recognized as one of the key factors underlying high malnutrition in India through hygiene and sanitation mechanisms (Reddy et al., 2017;Flederjohann et al., 2015;Ravillion & Jalan, 2003;Million Death Study Collaborators, 2010;Rah et al., 2015;IIPS & ICF, 2017;Ngure et al., 2014). However, the role of water conceptualized more broadly in shaping children's dietary diversity remains unexplored. India is considered as a water stressed country (UN-IDfA;Croninet al., 2016). A total of 54% of India faces high to extremely high stress in terms of surface water (Shiao et al., 2015).
Most regions of India suffer from problems in terms of water quality, quantity, or both. The majority of northwest and Central India are water stressed, whereas South India-except the water abundant state of Kerala-has low to medium water availability (Piesse, 2017;Chakrapani, 2014;ADB, 2007). This regional variation in water availability in India provides a suitable context for discerning the role that both household water access and regional availability might play in shaping children's dietary diversity. Here, we examine the effect of water availability and access on an important infant and young child feeding indicator-minimum dietary diversity (MDD)-in context of India using the nationally representative Demographic and Health Survey (DHS) data.

| DATA AND METHODS
The data has been obtained from 2015-2016 India DHS, a nationally representative survey that provides comprehensive information on child care, child feeding practices, and relevant socio-demographic information on the household. The sample covers children aged 6-23 months (N = 69,841), nested in relatively smaller number of households; each household may contain more than one child in the sampled age range.

| Outcome variable: Minimum dietary diversity
Dietary diversity is in the pathway of child feeding and child nutritionwhich is not limited to stunting and is inclusive of micronutrient deficiencies. Failure to consume MDD is associated with lower micronutrient intake (Kennedy, 2009). Therefore, we have chosen not to use child linear growth or stunting as an outcome in our analysis as we are focused on explicating the degree to which water affects what foods are available to and prepared in the household for infants and young children. This approach aligns with recent scholarship recognizing that a focus on linear growth and stunting as ultimate nutrition outcomes "discounts the importance of other outcomes" and risks overlooking the "positive, meaningful, and observable effects before linear growth improves" of other interventions (see Leroy & Frongillo, 2019).
Accordingly, our outcome variable is whether or not a child achieves minimum standards of dietary diversity as estimated from data collected through daily recall. This is a dichotomous variable taking the value "1" if child achieves MDD and "0" if child does not achieve MDD. According to WHO (2008), a child is classified as achieving MDD if during the previous day s/he was fed with at least four of seven food groups: grains, roots, and tubers; legumes and nuts; dairy products (milk, yogurt, and cheese); flesh foods (meat, fish, poultry, and liver/organ meats); eggs; vitamin A-rich fruits and vegetable; and other fruits and vegetables.

| Household water access
Household water access is a key dimension of water insecurity (Jepson et al., 2017;Wutich et al., 2017) and includes dimensions of sufficient quantity, time to fetch water, and meeting household needs.
We characterize water access levels based on an assessment of the household's water sources. The DHS solicits information on household's water sources through the question, "What is the main source of water used by your household for other purposes such as cooking and handwashing?" The response options include piped water, piped into dwelling, piped into yard/plot, bottled water, piped to neighbour, public tap, tube well water, tube well or borehole, protected well, protected spring, tanker truck, dug well, unprotected well, unprotected spring, river/dam, rainwater, and others. Following WHO standards described in Howard and Bartram (2003), water access based on levels of needs met is defined as (a) optimal-when all consumption and hygiene needs are met, (b) intermediate-when consumption needs are assured and basic personal and food hygiene are met, and (c) basic or no access-when consumption needs are difficult to fulfil and hygiene needs may or may not be assured.
Building on the WHO/UNICEF Joint Monitoring Program ladder for household water services (WHO/UNICEF 2017), we recoded the source of water variable-our main independent variable-in the DHS data in following manner: 1. Optimal access-where sources of water include piped water, piped into dwelling, piped into yard, and bottled water. We expect that among the available sources listed, these four sources are most likely to ensure water access in desirable quantity and quality, meeting all consumption and hygiene needs in comfortable manner.
2. Intermediate access-where sources of water include piped to neighbour, public tap, tube well water, tube well or borehole, protected well, protected spring, and tanker truck. These sources are more likely to assure that consumption needs are met and basic hygiene needs are likely to be met because even public sources are also protected.
3. Basic or no access-where sources of water include dug well, unprotected well, unprotected spring, river/dam, rainwater, and others. In these cases, we expect that consumption needs may be met with difficulty, whereas minimum hygiene norms can hardly be assured.

| Regional water availability
Our second water-related predictor variable is relative regional water availability. For this, we classified India spatially on the basis of surface F I G U R E 1 Regional classification of India based on the water availability index Source: Adapted from WBCSD (2019)

| Other covariates
To take relative household material wealth into account in the models, we used categories based on the standard DHSconstructed household wealth index. The wealth index is derived from housing characteristics and the number and kinds of consumer goods and amenities owned (Rutstein & Johnson, 2004).
Household wealth categories are constructed by dividing the nranked distribution into quintiles based on the absolute scores for the entire country sample.
Educated mothers are expected to be better aware on optimal child feeding practices. Our variable on maternal education is whether mother has completed primary, secondary, or high education, as given in DHS data. We also tested for the gender of head of household (female vs. male), given the relative social and economic vulnerability of the former compared with the latter.
Other variables included in the model are place of residence (rural or urban), child's age and gender, number of children under 5 years of age in the household, religion, and social group. In India, religious affiliation significantly shapes cultural preference for foods; for example, Hindus are likely to eat less animal product or eat it less frequently (Yadav, 2006;Bhalotra et al., 2009). We operationalized social group as household-level caste or tribe (instead of the constitutional classification 2 ) because dietary preferences are likely more variable between castes and tribes as compared with within castes (Ferro-Luzzi, 1975).
Notably, we have not included variables on access to safe drinking water, hygiene, and sanitation because our outcome of interest is dietary diversity, upstream of measured nutrition status. This aligns with our objective to identify the role of household water access-not access to safe drinking water or associated hygiene practices (which we would expect to affect nutrition outcomes through the WASH pathways)-on child dietary diversity. Research on dietary diversity typically does not include sanitation or hygiene variables (see Rakotonirainy et al., 2018).

| Estimation method
Because our outcome variable is a dichotomous/binary variable, we use the probit regression to estimate the model. We also ran ordinary least square regression using child's dietary score as continuous dependent variable. However, the results-not reported here-did not differ from probit regression. As per the probit model, the probability that a child achieves minimum dietary diversity can be expressed as where Y = 1, if the child was fed with minimum dietary diversity; Y = 0, otherwise; X is the vector of variables that could determine the probability that a child achieves MDD; α is the constant terms in the equations; γ is the vector of coefficients of explanatory variables in the equations; and ε is the error term.
We first estimate this model for all of India including various water availability regions as dummy variable. Then we also estimate separate models for each water availability region. The rationale is that in varying context of water availability, the relationship between household water access and child's access to MDD may vary.
Because our analysis covers the entire India wherein a case represents each child aged 6-23 months, we do not weight our analysis. Sample weights are used to account for survey design used for sample selection. However, as instructed in the technical guide to DHS, we do not need to use sample weights while estimating relationships such as performing regression models (Rutstein and Rojas, 2006). The data analysis has been conducted using Stata v16 software.  access. In fact, proportion of households with optimal water access in LWAR and HWAR is nearly same at around 25%.

| Sample characteristics
2 India is traditionally a caste stratified society. In support of affirmative action, the Indian constitution defines certain population groups as Scheduled Caste and Schedule Tribes, based on their social and economic status. Table 3 gives results from probit regression on child's access to minimum dietary diversity. To account for heteroskedasticity in data, robust standard errors have been reported for all probit estimates.

| Both household water access and regional water availability context are associated with child's access to dietary diversity
We tested overall model fit using Hosmer-Lemeshow test. With a significance value of p < .05, the test described our model as adequately describing the data.
Household access to water appears as a significant predictor at p < .01 level of significance, though the size of effect is not big. As compared with children in households with optimal water access, a child in a household with intermediate water access and with basic or no water access has lower probability of achieving MDD by 1.9 and 2.1 percentage points, respectively. Regional water availability in India is statistically significant as well. As compared with a child in MWAR, a child in LWAR has 9.2 percentage point lower probability, and a child in HWAR has 3 percentage point higher probability of achieving MDD standards. We also introduced an interaction variable representing rural and urban areas of various water availability regions.
In terms of demographic factors related to gender, child in male headed households had a 2.6 percentage point higher probability of achieving MDD compared with child in female headed households. As compared with a child whose mother did not attend formal schooling, those with a secondary educated mother had 3.4 percentage point higher probability and those with high educated mother has 5.1 percentage point higher probability of achieving MDD standards. Other household factors predicting a child's probability of achieving minimum dietary diversity included household wealth; a child from the middle, higher, and highest wealth class has 2.1, 3, and 3.6 percentage point higher probability, respectively, of achieving minimum standards compared with the lowest wealth class. Children in rural households has 1.2 percentage point lower probability compared with those in urban households. In terms of religion, as compared with a Hindu child, a child from any other religion has higher probability of achieving MDD. Similarly, child belonging to a caste group had lower probability of achieving MDD as compared with a child from a tribe or other group by 2.3 and 8 percentage points, respectively. Table 4 shows results from separate probit model by regional water availability. In each water region, there is a significant association between household water access and child's probability of achieving MDD. Table 4 also gives test results on overall significance of household water access, and it appears to be significant at 0.01 level in each region. In LWAR and MWAR, the probability of child achieving MDD was lower in households with intermediate water access by 1.8 percentage point and 6.2 percentage point, respectively, compared with children in households with optimum water access. However, in HWAR, a child in household with intermediate water access had a 7.7 percentage point higher probability of achieving MDD as compared with those with optimal water access. In all water availability regions, there is however no significant difference in the probability of a child achieving MDD between households with optimal water access and household with basic water access. Other covariates appear to follow a pattern similar to the overall probit model including all regions (Table 3).

| DISCUSSION
Our analysis contributes to existing literature by clearly establishing the basic association between water and child's likelihood of achieving MDD, which may shape child nutrition outcomes. Our finding that suboptimal household water access is associated with lower probability of a child achieving MDD advances the emergent body of research that offers insights into a few ways through which the linkage between water access and dietary diversity unfold (e.g., Collins et al., 2019;Gibson, Ferguson, & Lehrfeld, 1998;Schuster, Butler, Wutich, Miller, Young, & HWISE Consortium, n.d). Household water access can affect child's dietary diversity also through gendered and competing demands on caregiver time. A 19-site cross-cultural study identified household water insecurity as qualitatively associated with delays in infant feeding, as the caregiver spent more time obtaining and managing water and less time preparing food for feeding the infant (Schuster, Butler, Wutich, Miller, Young, & HWISE Consortium, n.d.).
F I G U R E 3 Predicted probability of a child aged 6-23 months achieving World Health Organization's minimum dietary diversity standards by household water access and regional water availability T A B L E 4 Estimates from probit regression predicting achievement of minimum dietary diversity for a child aged 6-23 months across India across each of three regional water availability contexts Our second finding that at regional level, water availability can affect child's dietary diversity by altering the food availability dimension through its contribution to agricultural and allied activities, is also in conformity with the water-food literature (Mabhaudhi et al., 2016;HLPE, 2015). Our analysis further adds to the literature by suggesting how the relationship between household water access and a child achieving MDD varies regionally. Particularly, living in a higher water region suppresses or buffers the ill-effect of low household water access on children's dietary diversity. This could be a possible explanation for our finding that the child in a household located in HWAR but having intermediate water access has higher probability of achieving minimum dietary diversity standards compared with the child in a household with optimal water access, located in LWAR. More than half of HWAR is covered by the northeastern states of India, which is a water-rich region with heavy rainfall. However, this region is primarily hilly and thus relies more on tube wells and natural water sources (e.g., water bodies and springs warranted to explore how household water access and community/ regional water context interact and are mediated by the gender dimension to shape child's dietary outcomes and in turn child nutrition.

| Limitations
The cross-sectional nature of the data constraints intertemporal prediction of proposed relationships. We recognize that even though our results show statistically significant association between household water access and the probability of child achieving MDD, the magnitude of this difference by water access is low. However, our findings unfold possible pathways linking water to MDD, which can be further examined through mixed method research.
quality and time to access water that may vary within and between water sources. For example, a study from Uganda (Lauer et al., 2018) showed that water from improved sources were contaminated (Lauer et al. 2018). However, the measure of water access used in this study is considered by water experts as an acceptable proxy in water insecurity studies .
The data set did not provide a measure of caregiver's time allocation across household water management or child care activities, which constrains our insight into the underlying relationships. Furthermore, we have used child's achievement of MDD as a binary variable that might have compromised some precision in data. Finally, we do not have data to show how water access affected cooking patterns and irrigation activities.

| CONCLUSIONS
Our analysis provides evidence that inadequate water accessassociated as it may be with low household wealth-provides additional pathway to child nutrition as defined in terms of access to minimum dietary diversity. Suboptimal water access may intersect with limited household resources to reinforce reduced dietary diversity, especially in low overall water availability regions, and for household where resource shortages are more concentrated. To the extent child's dietary diversity is associated with sufficient macronutrient and micronutrient consumption, this may have implications for child nutritional status. This latter part may be an area for further research. Our findings also indicate the need for more elaborate data collection on gendered roles and water management.

Support for the analysis was provided by Fulbright Visiting Scholar
Program award to NC.