1. Introduction
Childhood is a critical period for proper development of the body and brain [
1]. However, more than 200 million children in developing countries experience developmental deficits [
2]. Inadequate child development is a critical problem, as these children will be likely to subsequently have poorer levels of educational achievement, poorer health in subsequent life stages, a lower probability of employment, and lower earnings [
3,
4,
5]. In addition, impaired child development, in turn, negatively affects developmental outcomes among future generations through lasting effects on educational attainment and livelihoods. Thus, improving children’s developmental potential plays an important role in cutting the chain of intergenerational poverty transmission [
3,
6].
It is well known that inadequate physical growth and cognitive development share common risk factors such as poverty, malnutrition, infectious diseases, lower parental SES, and family-environmental adversity [
7,
8,
9,
10,
11]. Although the community context has been highlighted as a critical level for interventions [
12,
13], much research on child growth and development has focused on individual-level factors especially within developing countries. Further, existing research assessing links between poor community conditions and adverse child development has dealt with either physical or cognitive outcomes, but not both [
14,
15,
16,
17,
18,
19]. Few studies have examined clustering of child physical growth and cognitive outcomes co-occurrence at the community level nor considered the shared environmental factors influencing both dimensions of child development.
In this paper, we bring together two bodies of literature on child physical growth and cognitive development using a multivariate multilevel analytical approach that enables us to jointly regress multiple outcomes on multilevel explanatory variables [
20]. Multivariate analysis is the analysis that assesses more than two outcomes simultaneously, which is frequently confused with multiple or multivariable regression analyses in the literature [
21]. Taking a multivariate multilevel approach uniquely allows us to examine community-level predictors as well as individual level predictors that simultaneously affect both physical growth and cognitive development, which could not be done if regression models were separate for each outcome. As a result, we can explore the extent to which communities simultaneously affect these two child development dimensions, adjusting for community-level compositional effects. Given that community dynamics are complex social processes, we pose an ecological and systemic perspective embracing environmental, psychological, and material characteristics of communities [
22,
23], rather than using the urban–rural divides or a single community dimension as community-level variables. If our study finds that inadequate child physical growth and cognitive development are clustered within communities, our findings would support investments in community-level interventions to prevent the co-occurrence of these two problems [
21].
We chose data from India for our study due to the persistently high rates of stunting and underweight children in India [
24]. A recent report showed that about 48% of children and more than 44% of children in India were stunted and underweight, respectively, during 2009–2013 [
24]. The aim of this study is (1) to investigate the variation, covariation, and correlation of these two outcomes at the individual and community level, and (2) to identify community-level characteristics associated with clustering of child physical growth and cognitive development.
3. Results
Table 1 presents descriptive information of the family backgrounds of the sample children. More than 53% of caregivers had no formal education. The mean of mother’s height was 151.4 cm (SD 6.5). On average, communities had 2.7 pollution problems (SD 1.7), 1.8 social problems (SD 1.1), 11.5 accessible local services (SD 7.1), 27.9 programs run by governments and NGOs/charities (SD 7.1), and 2.1 kinds of health resources (SD 2.6).
Supplementary Table S2 shows results from the sensitivity analysis to examine whether there were differences in estimates between the original model and a model only for children that lived in the same community after the previous 2002 survey. Results from the sensitivity analysis suggest that there was little difference between the models.
We found individual- and community-level factors associated with the outcomes, in terms of magnitude and significance (
Table 2). There were no common community factors that were associated with both physical growth and cognitive development. Local healthcare resources were associated with increasing physical growth. Local programs run by the government and NGOs/charities were only associated with better cognitive development. Local social problems were associated with lower math scores. Increasing child age was negatively associated with WAZ, but positively associated with PPVTZ and MATHZ. Higher caregivers’ education was more likely to have better physical growth and cognitive development than those whose caregivers had no education. Birth order of third or greater was inversely associated with better physical growth and cognitive development than the first child. Mother’s older age and greater height were positively associated with better physical growth. Wealth index was positively associated with both physical growth and cognitive development.
Table 3,
Table 4 and
Table 5 provide variance, covariance, and correlations between HAZ, WAZ, PPVTZ, and MATHZ at the individual and community level to demonstrate how both outcomes varied, covaried, and correlated across the levels. In
Table 3, when individual-level covariables were included (Model 2), the community-level variations in HAZ and WAZ decreased by 78.5% (0.14 vs. 0.03) and 92.8% (0.14 vs. 0.01), respectively. PPVTZ and MATHZ decreased by 35.3% (0.17 vs. 0.11) and 36.4% (0.22 vs. 0.14), respectively. Further, when community characteristics were included in Model 3, variations in outcomes remained significant. In
Table 4, the community-level covariance between physical growth (HAZ and WAZ) and cognitive development (PPVT and MATHZ) decreased by including individual-level covariables, resulting in a covariance range of 0.012–0.098 (between physical growth and language skills) and 0.001–0.129 (between physical growth and mathematics scores). In
Table 5, correlations between physical growth and cognitive development after controlling for individual-level covariables varied according to combinations between outcome indicators (correlation coefficients: 0.18–0.71 (at community level model 2), 0.11–0.17 (at individual level model 2)). Notably, a more pronounced correlation between physical growth and cognitive development was shown at the community level (vs. individual level correlation).
4. Discussion
We used a multivariate multilevel approach to (1) investigate the variation, covariation, and correlation in children’s physical growth and cognitive development using the Young Lives study India dataset, and (2) identify community-level characteristics associated with the two outcomes. We found a stronger correlation between physical growth and cognitive development at the community level than at the individual level. This may suggest that children’s physical growth and cognitive development tend to cluster within communities even after accounting for natural lottery cluster within a child (i.e.,: gene inheritance influence). Further, physically delayed children were also more likely to be cognitively delayed. In addition, we found significant associations between several community-level characteristics and outcomes. Local pollution was associated with worse cognitive development. Children living in communities with more local healthcare resources were likely to have better physical growth. Local programs run by the government and NGOs/charities were associated with better receptive language skills in children.
A novel aspect of our study was to quantify the co-occurrence of the physical growth and cognitive development of the child at the individual level and the community level simultaneously as well as separately. We found correlations in patterning of between-community variations of the outcomes. A more pronounced correlation was observed at the community level, suggesting that a larger extent of the variation in co-occurrence was between communities. This finding implies that communities have the potential to simultaneously promote physical growth and cognitive development in children. This is the first study to quantitatively demonstrate the simultaneous importance of the community-level for connected outcomes.
Second, the number of local healthcare resources was associated with physical growth, a result supported by previous studies suggesting that regional healthcare resources may impact child nutritional status and child height by providing access to treatment for common infectious diseases especially in poor settings [
39,
40]. Children may benefit from community-level healthcare resources enough to gain more weight. However, our findings of community-level variation in child development outcomes may reflect inequity in India due to an imbalance in resource allocation, inadequate physical access to healthcare facilities and human resources, and access to antenatal care and infant and young child immunizations [
41,
42]. Further, many rural practitioners are not formally trained or licensed [
43].
Third, we found that children’s cognitive development was positively associated with local programs run by the government and NGOs/charities. Existing studies showed that local social protection and universal education programs supporting impoverished families were an effective approach to reducing poverty in low-income countries, consequently also supporting child development [
44]. A few studies have provided evidence of the importance of community-level programs and child development using the Young Lives data. For example, a Peru study reported that an early child development program promoted child physical growth and cognitive development [
45]. In addition, the Ethiopian social assistance program known as the Productive Safety Net Program decreased child work for pay, reduced child labor time, and consequently increased the highest children’s grade [
46]. Finally, an Indian study reported that the Midday Meal Scheme implemented as a security net for children boosted cognitive scores as well as buffered adversity from malnutrition [
47].
Fourth, our findings of associations between local social problems and worse child cognitive development are consistent with previous studies showing that danger and crime in the community may adversely influence child development [
48,
49]. Previous studies explained that children living in communities that lack informal control or collective efficacy may have difficulty accessing resources such as after school programs or extracurricular activities that might foster child cognitive developments [
50,
51]. Parents may be also less likely to allow their child to play outside in such communities [
52,
53], and consequently, reduce opportunities to enhance their children’s cognitive skills.
Our study has several limitations. First, we adopted a longitudinal approach to ensure qualified inferences regarding the cause-effect relations; however, causality cannot be inferred due to unmeasured confounding. Second, we cannot eliminate the possibility of other unmeasured mediators, even though we included several covariates in order to reduce confounding. Third, we cannot avoid the bias caused by non-random attrition. Similar to other longitudinal surveys, disadvantaged households were more likely to drop out in the Young Lives Survey even though attrition rates were relatively small when compared with other longitudinal studies in developing countries [
26]. Fourth, our result has the limitation of generalizability, since the survey only included children living in Andhra Pradesh and Telangan. Last, we were not able to measure the quality dimension as well as the quantity aspect of community characteristics due to the limitation of secondary survey data. For example, the survey asked only whether a public hospital was currently available in the locality, rather than the number or quality of available hospitals in the locality. Future longitudinal studies should further investigate whether community characteristics also have long-term effects on child physical growth and cognitive development.