Theoretical premise

The theoretical foundation of conceptualizing the impact of community-level women’s education on adolescent health is rooted in the ecosocial theory [9]. The theory cautions against a simplistic understanding of determinants of disease (and health) at the individual level. It views individuals as non-separable from society because social and biological determinants are intertwined at every level: Individual, family, community or neighborhoods, regional, national and global [9]. It, therefore, encourages epidemiologists to inquire into the social context of health using a multilevel perspective [9]. Drawing from this theory, we conceptualized adolescents as deeply embedded in their context and, by extension, the unequal distribution of low hemoglobin levels and undernutrition among them as an embodiment of their lived social environment. Extant literature has pointed out that socio-economic and political context, including cultural and societal values, social stratification based on gender, caste, class, and religion, are important social determinants of nutrition in adolescents [10]. These social determinants operate at global, regional, and local levels through complex intersecting pathways, determining the access to, and availability of, food, health, education, and security to adolescents [10]. An individual-focused perspective fails to take into account the causal influence of the context on adolescents’ undernutrition [9]. Exploring and addressing the contextual determinants of adolescent health is thus essential in addressing this stubborn problem. We argue that the value assigned to women in society is an integral aspect of the social context within which adolescents live and grow, acquire skills and education, and often work. We posit that this has relevance to adolescents’ health and nutritional status.

Patriarchal norms are deeply pervasive in Indian society with its manifestations and effects varying depending on socio-economic and regional contexts [11–13]. The lower value assigned to women as compared to men in Indian society is evident in a range of socio-economic and developmental indicators. For instance, India was ranked 112^th among 153 countries in the World Economic Forum’s Global Gender Gap Index, 2019–2020 [14]. Discrimination against women begins even before birth in India as evidenced by an inverse sex ratio—according to the latest estimates, 899 girls are born for every 1000 boys born [15]. This inverse sex ratio at birth has been argued to primarily result from gender-biased sex-selection practices rooted in son-preference [16, 17]. Evidence suggests that gender discrimination continues after birth with the neglect of the girl child—11% more girls than boys die under the age of five [18]. This is in stark contrast to the world average, where 7% more boys, than girls, dies under the age of five [18]. According to UNICEF, India is the only large country in the world that has an inverse sex ratio at birth accompanied by postnatal sex selection [18]. With 460 thousand girls missing every year between 2013 to 2017, India accounted for nearly one third of the world’s missing women as of 2020 [17]. India is also home to the world’s highest number of child brides [18]. Although child marriage in India is showing a declining trend, the rate remains high at 27% [19]. Further, nearly 27 percent married adolescents reported to have given birth to one child [20]. This has huge health and economic consequences on the adolescent herself, her off-spring, and the family [21]. Notably, female labor force participation in India is among the lowest in the world: 2020 data show that four out of five women were not working for wages [22]. This is lower than the average values for lower-middle income countries and the South Asian region. Indian women are also disadvantaged in ownership of assets and access to healthcare [19]. Moreover, women are commonly exposed to gender-based violence in India, with the latest nationally representative data suggesting that 33% of ever-married women between 15 to 49 years were exposed to at least one incidence of physical, sexual or emotional spousal violence [19].

While historically poor investment in girls’ education has resulted in an overall gender gap in literacy, India is rapidly making progress on closing the gender gap in literacy among children [23], and secondary school enrolment rates [24]. However, the gender gap in education among the socio-economically marginalized groups is still a problem [25, 26]. The gender gap in completion of secondary schooling (tenth-grade) also persists, likely because of the gender bias favoring boys regarding an investment in education [27] and the provision of free education only until the age of fourteen under the Right of Children to Free and Compulsory Education act, 2009, popularly known as Right to Education (RTE) Act, 2009 [28]. According to the Indian National Sample Survey of 2018, the literacy rate (percentage among persons of age 7 years and above) in men is 81.5% versus 65% in women [29]. The same source indicated that the percentage of persons not literate in those aged 15 years and above was 22.2% among men and 41.2% among women; while the proportion who had completed some primary schooling (grades 1 to 4 was 21.2% among men and 20.4% among women; -middle school (grades 5 to 7) was 19.8% among men and 14.4% among women;—secondary school (grade 8 to 10) was 17.3% among men and 12.5% among women;—higher secondary (grades 11 to 12) was 12% among men and 7.6% among women; and -graduate or above was 7.4% among men and 3.9% among women [29].

In this backdrop, we conceptualized community-level women’s education as a proxy for the value that a community places on women’s well-being. This is based on the assumption that societies where women are valued, ‘as women’, tend to have a greater access to education for women [30]. We recognize that women’s education is a result of the amalgamation of a range of structural, socio-cultural, and household-level factors [31]. We posit that societies that value their women, see the value in educating their women, and in increasing the social desirability and social acceptability of women getting an education. We also surmise that these societies tend to invest in, and generate resources to facilitate, women’s access to education. Thus, we argue, one feature of societies that value their women could be a greater proportion of women with higher levels of education.

We have not empirically examined the potential pathways through which the contextual effect of the value assigned to women’s wellbeing in a community manifest. However, contextual effects and diffusion of behaviors, knowledge and values have been conceptualized in several disciplines, for instance in political science [32], in demography and fertility transition [33] as well as in public health such as neighborhoods and health [34] and health behavior change [35]. Drawing from these scholars as well as the ecosocial theory [9], we conceptualized the pathways linking the contextual influences of the value assigned to women’s wellbeing in a community on adolescent nutrition.

Communities where women are valued are likely to assign a higher value to women’s knowledge, skills, and abilities [30, 36]. Women in these communities might have a higher self-worth and confidence (intangible resources) to assert their preferences and be able to better leverage or even generate material resources to enact their choices [30]. These changes could alter the social norms in a community and influence the decision-making of adolescents in those communities. Women in such communities might have a greater say in decision making within the household as well as in the public sphere, such that they are able to negotiate the use of resources and the benefits accruing from their use, to achieve the goals that they value, which likely includes the betterment of the nutritional status of adolescents in their communities [30, 37]. Women in influential positions in communities have the potential to transform social institutions and promote the growth of public services, including health and nutrition services in a community [37]. All of these have the potential to impact adolescent nutrition. Extant literature has suggested that often women are endowed with primary care-giving responsibility within the household [38]. Women also have a central role in maintaining household food security and household health environment [38]. In such a context, the influence of women’s empowerment and/or education on child health has suggested that women who have access to resources and a greater say in decision-making tend to prioritize pro-child resource allocation such as towards food, health care, and education [37–39]. Extending this to adolescent children, we argue that women’s health-conducive and -progressive behaviors related to adolescent health, nutrition, education, sanitation and hygiene, diffuse within the community, through social interactions. This potentially alters the normative expectations within the community and reinforces health-conducive norms and practices among family members of the adolescents, regardless of their own individual or family characteristics. This could be illustrated by a thought exercise where we compare two families with similar socio-economic backgrounds, having two adolescent children (a boy and a girl). Say family A lives in an area where generally women are valued as opposed to a family B living in a community where women are less valued. Family A would have a higher chance of exposure to health-enhancing choices made by a larger proportion of empowered women in the neighborhood. Social interaction among them likely provides an opportunity for social learning and social influence, potentially reinforcing health-promoting practices and norms. This might alter family-A’s beliefs and practices related to adolescent care, benefitting adolescents in family-A, irrespective of the status of women in family-A. Such an effect might operate through women, men and/or adolescents themselves within family-A. Adolescents in family-A might also benefit from being less exposed to gender-unequal social norms in their social environment, which might promote their self-worth and confidence that may subsequently help them negotiate gender-equal situations within a familial setting. On the other hand, family-B has a smaller chance of social interaction with women who make health-promoting choices, as the proportion of such women in their community is lower. Therefore, the social pressure to reinforce health-promoting behaviors is likely much lower. Adolescents in family-B therefore might not benefit from a health-promoting contextual effect of living in a community that values women unlike adolescents in family-A. Furthermore, adolescents growing up in a context that is more gender unequal, will more likely be imbued with those values, resulting in their own lower self-worth and lower confidence to assert their choices in a familial setting, a potentially more direct impact for adolescent girls. This might manifest in their lower ability to bargain for nutrition- or health-related resources and social support, for instance: consuming nutritionally-dense food, getting access to health-care or access to hygienic environments, products and practices especially related to menstruation.

To summarize, despite being theorized, our extensive literature search could not locate any theory-driven research that empirically demonstrated the influence of context on the hemoglobin level, anemia, nutritional status, or health of adolescents, in low- or lower-middle-income countries, including India. To our knowledge, this is the first study that investigates the relationship of community-level women’s education with adolescents’ nutritional status. Notably, our interest in community-level women’s education is not limited to its immediate exposure. Rather, we conceptualize it as capturing the strength of the socio-cultural norms in the community that value women’s well-being. Thus, our inquiry has relevance not only for Indian adolescents, but might have implications for the millions of adolescents growing up in societies that are embedded in patriarchy.

Dataset and sample characteristics

This analysis is based on publicly available, de-identified data from the fourth round of the National Family Health Survey (NFHS-4), India, undertaken in 2015–16 [40]. It was a nationally representative survey yielding estimates at the district and state levels [19]. The sampling design treated villages (occasionally village-clusters) and census-enumeration wards (in urban areas) as the Primary Sampling Units (PSUs) [19].

In the present study we have operationalized the PSU as the “community”. The sample households in the clusters (PSUs) are a mix of different caste-groups residing in the same neighborhood in both urban as well as rural areas. However, the proportion of households of different caste groups across clusters might vary depending on the caste composition of the cluster. Our operationalization of the community therefore captures geographic proximity of the households. The sample population might be conceptualized as living in the same neighborhoods. Physical proximity of residences and shared interests related to availability of public infrastructure typically facilitate social interaction among the residents. In the rural areas, the villages or cluster of villages are frequently the part of the same local government (Gram Panchayat). Social interactions might be further strengthened by the involvement of women in local self-help groups (women’s microfinance groups) or children sharing the same pre-school (Anganwadi). Similar opportunities for greater social interaction and the development of a sense of community are observed even in urban areas. While the CEBs are spaces with relatively higher population density, they often have shared sources of water, shared public toilets and other facilities such as Anganwadi centers. They might also share geographic access to markets, infrastructure, and services. All this might promote social interaction among the local residents. Thus, PSUs might not only serve as a delimited geographical and/or administrative region. It is likely viewed by the residents as a social space with a sense of belonging in the form of ‘our village’ or ‘our place’ or ‘our neighbors.’ However, other factors influencing social interaction such as caste and religious dynamics do not get adequately captured in this operationalization of the community, which remains a limitation, even though a sample from a PSU includes households having different caste and religious characteristics. Conceptually, neighborhoods have emerged as potentially relevant in public health as they capture the social and physical aspects of contexts that have theoretically plausible causal association with health [34]. Scholars have previously operationalized PSUs as communities while estimating the impact of social characteristics of neighborhoods on health outcomes in the Indian context [39]. A total of 601,509 households were included in NFHS-4, which were selected by employing a two-stage stratified random sampling design [19], (for additional details on sampling design refer [19]) from 640 districts of all the 36 states/union territories (UT) in India. Hemoglobin was measured in a total of 117,711 girls and 17,912 boys aged 15 through 19 years. We restricted our analysis to unmarried adolescents between 15 through 17 years of age, as information on parents’ education was available only for children below 18 years of age [19]. We linked the hemoglobin levels of 75,696 adolescents to their mother’s education. Information on father’s education was collected only if the father permanently lived in the same household as that of the adolescents [19] (information not available for instance if father had migrated for work). Therefore, we excluded 11,106 adolescents who were missing the information on their father’s education. After excluding observations that were missing data (including the response “don’t know”) on any of the other covariates, our analytical sample size was 62,648 adolescents for the hemoglobin analysis. Following a similar process, of 18,984 boys and 1,25,333 girls aged 15 through 19 whose body mass index (BMI) was measured, we linked the BMI of 75944 adolescents to their mother’s education. After deleting observations with missing data on covariates we obtained an analytical sample of 62,846 adolescents aged 15 through 17 years for the BMI analysis.

Measures

The outcome variables were blood hemoglobin level in adolescents measured in grams per deciliter (g/dl), and BMI measured as weight in kilograms divided by the square of height in meters (kg/m²) on continuous ratio scales, excluding biologically implausible values [41, 42]. Hemoglobin estimation was conducted on-site with a battery-operated portable HemoCue Hb 201+ analyser [19]. Hemoglobin levels were adjusted for smoking and altitude in areas that were above 1,000 meters [19]. Weight and height were measured on-site using a Seca 874 digital scale and a Seca 213 stadiometer respectively [19].

Our main predictor was the education of women at the community-level, represented as the proportion of all women in NFHS-4 above 19 years of age in a PSU who had completed formal education at least till the tenth-grade. The education data of a total of 880,431 women across 28524 PSUs were used to compute this variable. NFHS-4 collected information about the education level of each family member of the household above the age of four years. We therefore had information about women’s education for 1,041,971 women in total. After removing those aged below 19 years, we had information from a total of 880,850 women above 19 years of age. Further removing those who were missing data on education yielded data on the education of women above 19 years of age from a total of 880, 431 women. This information was utilized to compute the community-level women’s education variable.

Our selection of a tenth-grade education for operationalizing community-level women’s education is based on the idea that the completion of a tenth-grade education among women involves valuing women’s education beyond the elementary level (Grades 1 to 8), even when it is not compulsory and free for the children under the RTE Act, 2009 [28]. Further, historically, completion of the tenth grade is an important milestone in the Indian education system [26]. One of the reasons for this is that it involves clearing an examination that is conducted by either a central or a state secondary school examination board [26]. This examination requires students to appear for the exam at a place that is centrally decided, often different from their own school, sometimes in a different village or town than their own. In this standard examination, a student’s academic performance is assessed by external evaluators. There is no such requirement for any lower grades. Furthermore, after the introduction of the Right to Education Act 2009, a no-detention policy until grade 8 was implemented in India [28]. In many schools, this likely resulted, unintentionally, in an improper assessment of children in elementary school [43]. This is described as one of the reasons why children who were poorly assessed in the elementary grades tended to lose interest in studies when they performed poorly at higher grades [44]. Despite the several critiques of the board examination [45], we believe that for the above-mentioned reasons, this examination might serve as a tool to assess with greater integrity the academic potential of a student. This implies that we expect a student who has passed the tenth-grade to manifest a greater level of the desirable impacts of schooling such as critical thinking and a progressive attitude towards change in social norms than students who have basic numeracy and literacy skills. Further, the tenth-grade examination plays a decisive role in making further education opportunities available to the child in the Indian education system [26]. Passing a tenth-grade exam carries a value in the labor market as well, opening up several job opportunities for the candidate (for instance in one of the greatly sought-after employment sectors, the government [46]). In India, secondary schooling is reported to be giving an 11.4% rate of return to education for every additional year of schooling at the secondary schooling level, as compared to the same for primary and middle school at 5.5% and 6.2% respectively [47]. On the other hand, the rate of return to education for every additional year of schooling at higher-secondary and graduate levels are 12.2% and 15.9% respectively [47]. Thus, the rate of return to schooling is reported to be much higher for completion of secondary schooling in comparison to primary and middle school. Additionally, one of the major points at which substantial dropout from school is observed in India is between grades 9 and 11 [48]. Considering these points, we believe that for the purpose of the present study, the tenth-grade serves as an appropriate cut-off to measure the importance given to women’s education and, by extension, we argue, the importance given to women themselves, in a community.

We used the following variables as covariates in our analysis. We operationalized the age of adolescents (in years), education of adolescents, adolescent’s mother, and adolescent’s father (all three as years of formal schooling completed), as well as household size (as the number of family members) as continuous variables. We operationalized gender (girls or boys, reference = boys), religion (Hindu, Muslim, Christian, and others; reference = Hindu), social group (Scheduled Tribe-ST, Scheduled Caste-SC, other backward classes- OBC, and general; reference = SC), household wealth (as quintiles of the distribution of the population on a household asset score included in the original dataset; the five quintiles were poorest, poorer, middle, richer or richest; reference = poorest), place of residence (urban or rural; reference = urban), and sub-national regions (groups of states in the North, East, West, Central, South and Northeast; reference = North) as categorical variables. Social groups denoted as SC, ST or OBC are historically socially and economically marginalized caste groups in India as recognized by the Government of India. These labels are employed by the Government of India in official documents and widely used in daily life in the Indian context. The wealth quintile data that we have used were provided in the original dataset [19, 40] (for further details on how the wealth index was created and the computation of wealth quintile, please refer to [19]). Variables representing the socio-economic condition of the community were created as the proportions, respectively, of total households in the PSUs following the Muslim religion, belonging to ST/SC groups, and the poorest wealth category.

Statistical analysis

Descriptive statistics were used to depict the distribution of the sample, hemoglobin levels and BMI levels, across the covariates. We applied individual-level sampling weights to compute descriptive statistics. We fitted a series of five-level random intercept multiple linear regression models to estimate the association of community-level women’s education with adolescent hemoglobin levels and, separately, with BMI. This was an appropriate choice given our interest in the influence of the context on an individual-level outcome [49]. The five hierarchical population levels were adolescent i, nested in household j, PSU (community) k, district l and state m. The random components at each level were assumed to follow a normal distribution, with a mean of zero, and a constant variance; and assumed to be mutually independent. Alpha was set at 0·05. The unadjusted Model 1 included community-level women’s education as the predictor. We incrementally added individual-level covariates: adolescent’s gender, age and, adolescent’s education in Model 2, and mother’s education in Model 3. The fully adjusted Model 4 additionally included father’s education, household-level covariates: family size, social group, religion, household wealth; the community-level covariates: wealth, and social and religious composition; and the state-level covariate: region. Firstly, we fit unweighted multilevel models. Additionally, in order to account for the complex sampling design, rescaled weights were computed at the individual-level, using two methods [50]. Method A scaled the weights so that the new weights sum to the cluster sample size and method B scaled the weight so that the new weights sum to the effective cluster size [50]. Method A and method B are the nomenclature used by Carle (2009), following Asparouhov’s labels [50]. We ran our fully adjusted five-level models for hemoglobin and body mass index (BMI) applying the scaled weights obtained using both the methods at the individual-level. We have included these findings in S3 File Weighted multilevel models. Statistical analysis was carried out using Stata, version 12·1 [51].

Ethical concerns

We carried out a secondary data analysis of publicly available deidentified data collected as a part of NFHS-4 India, accessed via the Demographic and Health Surveys (DHS) Program website. Necessary approvals were sought from the DHS Program to use NFHS-4 data for this project. As we were undertaking secondary data analysis of publicly available deidentified data, and were not collecting any new data, our study was exempt from review at the Institutional Ethics Committee of the host institute of the authors. International Institute for Population Sciences (IIPS), Mumbai, was the nodal agency for the NFHS-4 project [19]. The IIPS Institutional Review Board and the ICF Institutional Review Board granted ethical approval to the NFHS-4 project [19]. Several field agencies selected by IIPS, Mumbai, collected the data for NFHS-4 [52]. Specially trained health investigators hired by the field agencies carried out anthropometric measurements and biomarker testing [52]. The health investigators obtained the written informed consent from the parent or guardian or any other responsible adult in the household for unmarried adolescents aged 15 through 17 years, separately for each measurement and biomarker testing [52]. After receiving the informed consent, the health investigators sought the adolescent’s assent for performing the measurements and tests [52].

Descriptive statistics

Our analytical sample for the hemoglobin analysis consisted of a total of 62,648 adolescents aged 15 to 17 years (average: 15.95; SD: 0·81), from 57,638 households, 23,448 PSUs, 640 districts, and 36 states/UT (Table 1). Average hemoglobin value for girls was 11·72 (SD 1·57) g/dl which is lower than the normal range (12 through 14 g/dl) for girls and would be considered as anemia. The interquartile range [25th percentile- 75th percentile] of hemoglobin value among girls was [10.8 g/dl, 12.8 g/dl]. On the other hand, for boys, average hemoglobin value was 13·52 (SD 1·69) g/dl which is within the normal range (13 through 15 g/dl) for boys. The interquartile range of hemoglobin value among boys was [12.5 g/dl, 14.7 g/dl].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. The distribution of the adolescents and average hemoglobin levels in gram/deciliter (g/dl) (N = 62,648) by gender and sociodemographic characteristics.

https://doi.org/10.1371/journal.pone.0251427.t001

The analytical sample for the BMI analysis consisted of a total of 62,846 adolescents aged 15 to 17 years (average: 15.95; SD: 0·81), from 57,813 households, 23,499 PSUs, 640 districts, and 36 states/UT (Table 2). Average BMI values for girls was 19.21 (SD 2.88) kg/m2 and for boys was 19 (SD 3.12) kg/m2 which were within the normal range (18.5 through 23 kg/m2) for both girls and boys, with girls having higher average values than boys. The interquartile range of BMI value among girls was [17.34 kg/m2, 20.64 kg/m2] and the same among the boys was [16.94 kg/m2, 20.44 kg/m2]. Further, an economic gradient was observed in average BMI values among both boys and girls. The average BMI values were lowest among the poorest quintiles. The average BMI values were higher at higher levels of the wealth index, with the richest having the highest average BMI values among both boys and girls. While average BMI values for the girls were within the normal range across all the wealth quintiles, the trend in the interquartile ranges of BMI for girls across the wealth quintiles suggested an inverse association of BMI with wealth (S1 File, Wealth index wise interquartile range of BMI). Fitting this pattern, the average BMI values were lower than normal among the boys in the poorest and poorer categories. The samples comprised mainly of girls (about 87%). The distribution of adolescents by background characteristics in our samples were comparable with that of the Indian adolescent population (Census 2011) [53].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. The distribution of the adolescents and average body mass index in kilogram/squared meter (kg/m²) (N = 62,846) by gender and sociodemographic characteristics.

https://doi.org/10.1371/journal.pone.0251427.t002

The community-level descriptive statistics were identical for the hemoglobin and BMI analyses, while the sample sizes were 23,448 for the hemoglobin analysis and 23499 for the BMI analysis (Table 3). The proportion of women with formal education was 73.73% in the urban areas as compared to 49.93% in the rural areas. The average (SD) proportion of women having a tenth-grade education in urban PSUs was 0.40 (0.21) whereas the same for rural PSUs was 0.17 (0.13). We have plotted the average value of women’s education (as the number of completed years of schooling) at community (PSU)-level using a histogram (Fig 1). The average number of completed years of schooling among women at the community-level was between four to five years in the maximum number of communities. The average education level among women (in the number of completed years of schooling) at the state- and regional-levels are provided in the S2 File (Women’s education at State and Region levels).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Frequency distribution of primary sampling units by average level of women (above 19 years of age)’s education in completed years of schooling.

(n = 23499) (Graph prepared using Stata 12 [54]).

https://doi.org/10.1371/journal.pone.0251427.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Summary measures of community-level factors (N = 23,448 for hemoglobin analysis/23499 for BMI analysis)^*.

https://doi.org/10.1371/journal.pone.0251427.t003

Hemoglobin analysis

In the bivariate Model 1 (Table 4), a one standard deviation (0·19) higher community-level women’s education was associated with (0·569(β)*0·19) 0·11 g/dl higher hemoglobin level, on average (p<0·001). Adding individual level covariates reduced the value of the coefficient (β = 0·423 in Model 2), but retained the positive association of the primary predictor observed in Model 1 (p<0·001). Adding mother’s education (Model 3) further reduced the β (0·286), but retained the positive association between the primary predictor and the outcome (p<0·001). The average hemoglobin level in adolescents was higher by 0·045 g/dl if the proportion of women who were educated up to the tenth-grade or above in a community was 19 percentage points higher (p<0·001) in the fully adjusted model (Model 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. The association (regression coefficients and 95% CI) of community-level women’s education with adolescent hemoglobin level in unadjusted and adjusted models (N in all models = 62,648 adolescents).

https://doi.org/10.1371/journal.pone.0251427.t004

BMI analysis

Similar to the relationship of contextual education with hemoglobin, a higher proportion of women educated at least until the tenth-grade in the community was associated with higher adolescent BMI. The unadjusted Model 1 for BMI analysis estimated that one standard deviation higher community-level women’s education was associated with ((β) 1.911*0.19) 0.36 kg/m² higher average BMI (p<0.001), which was reduced in the fully adjusted Model 4 to (0.776(β)*0.19) 0.15 kg/m² (p<0.001). The positive relationship of the main predictor with the outcome was sustained across all the models (p<0.001). Additionally, Model 4 for BMI analysis suggested that higher proportions of poorest or Muslim households in a community were significantly positively associated with BMI. Table 5.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. The association (regression coefficients and 95% CI) of community-level women’s education with adolescent BMI in unadjusted and adjusted models (N in all models = 62,846 adolescents).

https://doi.org/10.1371/journal.pone.0251427.t005

Notably, nearly 28% of the PSUs in our sample had less than 10% of the women with a tenth-grade education. If 100% of the women in these communities acquire at least a tenth-grade education, our findings suggest that it might lead to (0·9*0·234(β)) 0·21 g/dl higher hemoglobin level, and (0·9*0·776(β)) 0·7 kg/m² higher BMI, on average, among all adolescents in those communities. Similarly, in the nearly 52% of the PSUs having less than 20% of the women with a tenth-grade education, the estimated gain in hemoglobin level would be 0·19 g/dl and in BMI would be 0·62 kg/m², on average, for all adolescents in such communities.

In all the models, the unexplained variance at all the five levels remained statistically significant, indicating the importance of contextual factors at the household-, community-, district-, and state-levels on adolescent hemoglobin level and BMI.