We analysed data from India’s National Family and Health Survey (NFHS-3), collected by 18 research organizations under the direction of the Indian Ministry of Health and Family Welfare between November 2005 and August 2006 [30]. The data are publicly available and free of charge from the Demographic and Health Surveys website [31]. A probability proportional to size sampling method was selected, with a two-stage design (villages, households) in rural areas and a three-stage design (wards, Census enumeration blocks, households) in urban areas. A stratified sampling method was used in the first stage to ensure representativeness of the sample on the basis of village size, primary labour modality, caste, female literacy, HIV prevalence, and a variety of other indicators [32].
The sample, which is representative at both the national and state level, includes 124,385 women aged 15–49 and 74,369 men aged 15–54 across each of India’s 29 states. Our analytic sample was limited to regular female residents of the household (interviewed visitors were excluded). Population weights, which adjust for the sampling design and nonresponse (response rate = 93.5 %), were applied to all analyses using the weighting variable created by the NFHS survey team [33]. Less than 5 % of eligible women were not sampled due to not being home (2.9 %), postponing the interview (.1 %), refusing to participate (1.5 %), being incapacitated (.3 %), or some other reason (.3 %). Data on food and water consumption were collected only for living children born within the five year period preceding the interview. Since introduction of water into the diet is not recommended for children under 6 months of age, we restricted the sample to living children aged 6–59 months (n = 30,656) Additional file 1.
In the first stage of the analysis, we evaluated the prevalence of children who received no water using a synthetic cohort—that is, examining water consumption patterns within each 1 month age group. Water consumption was measured based on the mother’s report of children’s consumption from a questionnaire covering the preceding 24 h for 24 food items. For children under the age of 5 years, mothers were asked: “Now I would like to ask you about liquids (NAME) drank yesterday during the day or at night. Did (NAME) drink:” The list included plain water, fruit juice, tea or coffee, tinned, powdered, or fresh milk, commercially produced formula, and other liquids, in addition to a variety of solid foods. Children whose mothers reported “No” to plain water were coded as not consuming water; children whose mothers reported “No” to all beverage questions were reported as not consuming any beverages. An additional question asked how many times the child was breastfed in the preceding 24 h; children who were reported as having been breastfed 0 times during that period were coded as not having breast milk in the last 24 h. Missing data did not exceed 4 % for most food intake variables in the analysis, but was higher for water consumption (23.1 %, n = 10,516). To estimate the population-level prevalence rates, we used UN population figures for 2005 which indicated a total population of 6–59 month old Indian children of 148,698,600 [34].
While water consumption is preferred, for children who do not receive water, consuming alternative beverages may help to reduce the risk of dehydration. In the second stage of the analysis, we examined whether alternative beverages were consumed by those children who did not drink any water in the past 24 h. In the final stage of the analysis, we evaluated which children were at greatest risk of not drinking water using logistic regression models. This included a vector of potential household and environmental risk factors of receiving no water. One factor was type of water source, including dummy variables for four categories, as follows: piped, tanker, or bottled water (coded 1 if piped into dwelling, into yard, tanker truck, cart with small tank, and bottled water); well water (coded 1 if tube well or borehole, protected well, unprotected well); public tap; and river, spring, or rainwater (coded 1 if protected, unprotected spring, river, dam, lake, ponds, stream, canal, irrigation channel, rainwater, and other). We also evaluated WHO/UNICEF measure [35], of whether or not the household uses an improved water source (coded 1 if piped into dwelling, into yard, public tap, tube well or borehole, protected well, protected spring, rainwater; 0 if tanker truck, car with small tank, bottled water, unprotected well, unprotected spring, river, dam, lake, ponds, stream, canal, irrigation channel, rainwater, and other sources; WHO & UNICEF, 2013). We also expected children living in deprived households to be at greater risk. This was assessed using a dichotomous indicator of whether the household was deprived, based on a collapsed version of the standard DHS household wealth index available in the survey data (poorest and poorer were coded as deprived; middle wealth, richer, and richest were coded as not deprived). We also included place of residence (urban non-slum, urban slum, and rural).
Models evaluated potential disparities for age, gender, and social position, based on previous literature suggesting such inequalities in nutritional outcomes [36–38]. As a validity check, we included an indicator for whether the child had diarrhoea in the 2 weeks preceding the interview, as such a child should have been more likely to receive water. We also adjusted for maternal characteristics including maternal age, a categorical educational attainment measure (no schooling, primary school, secondary school, higher than secondary), religious affiliation of the household head (Hindu, Muslim, Christian, and other); caste (scheduled caste, scheduled tribe, other backwards class, and other/no caste). All models were estimated using STATAv12.1