Beyond personal factors: Multilevel determinants of childhood stunting in Indonesia.

BACKGROUND: Stunting is still a major public health problem in low- and middle-income countries, including Indonesia. Previous studies have reported the complexities associated with understanding the determinants of stunting. This study aimed to examine the household-, subdistrict- and province-level determinants of stunting in Indonesia using a multilevel hierarchical mixed effects model.
METHODS: We analyzed data for 8045 children taken from the 2007 and 2014 waves of the Indonesian Family and Life Surveys (IFLS). We included individual-, family-/household- and community-level variables in the analyses. A multilevel mixed effects model was employed to take into account the hierarchical structure of the data. Moreover, the model captured the effect of unobserved household-, subdistrict- and province-level characteristics on the probability of children being stunted.
RESULTS: Our findings showed that the odds of childhood stunting vary significantly not only by individual child- and household-level characteristics but also by province- and subdistrict-level characteristics. Among the child-level covariates included in our model, dietary habits, neonatal weight, a history of infection, and sex significantly affected the risk of stunting. Household wealth status and parental education are significant household-level covariates associated with a higher risk of stunting. Finally, the risk of stunting is higher for children living in communities without access to water, sanitation and hygiene.
CONCLUSIONS: Stunting is associated with not only child-level characteristics but also family- and community-level characteristics. Hence, interventions to reduce stunting should also take into account family and community characteristics to achieve effective outcomes.

1. Introduction

Stunting is an ongoing issue in many low- and middle-income countries. UNICEF/WHO and the World Bank [1] indicate that the number of stunted children is approximately 151 million, accounting for 22.2% of the children in the world. Moreover, the proportion of stunted children is concentrated in low-income (16%) and lower-middle-income (47%) countries compared to upper-middle-income (27%) and high-income (10%) countries [1]. Approximately 83.8 million stunted children live in Asia, mainly in Southern and southeastern Asia, 58.7 million in Africa and 5.1 million in Latin America and the Caribbean.

Indonesia is one of the countries with a high burden of malnutrition, including stunting [1]. Child health outcomes are poor, even though the Indonesian economy is the largest in Southeast Asia and the 17th largest in the world [2]. Data published by the Ministry of Health show that the incidence of stunting among children aged five years and below remains high at 30.8% [3]. The World Bank (2020) [4] noted that Indonesia has underperformed in terms of reducing the level of stunting compared to other upper-middle-income countries and other countries in the region. Given the high prevalence of stunting and its impact on children’s cognitive development, the productivity level of Indonesia’s next generation is predicted to be half of its potential [4]. Therefore, tackling child stunting remains a major government commitment, as asserted in the Indonesia Medium Development Goals 2015–2019 and 2020–2024 [5, 6].

The wider literature on stunting reveals that various child-, parental-, household- and community-level characteristics are associated with stunting [7-13]. At the parental and household levels, several dietary and socioeconomic factors have been shown to be correlated with the risk of stunting. With regard to risk factors, Beal et al. [7], for example, established that the risk of stunting in Indonesia is higher in households that have no access to safe drinking water. Household wealth status is another significant predictor, as children coming from poor households are more likely to be stunted [11, 12]. Meanwhile, at the community level, the prevalence of stunting has been shown to be higher in communities that lack access to health care [7]. In terms of protective factors for stunting, previous studies have shown that the likelihood of stunting is lower in communities where antenatal care services and integrated health and nutrition services are available [8, 9, 13]. In addition, consumption of diverse food within the household has also been found to lower the likelihood of stunting [14]. Furthermore, parental education has been shown to be significant, with children raised by educated parents having a lower risk of being stunted. Semba et al. [11] explored the channels through which parental education impacts stunting and argued that educated parents provide more care (in the form of having their children immunized and providing them with vitamin A and iodized salt), which would in turn lower the risk of stunting.

The prevalence of stunting in Indonesia varies by region, as illustrated in Fig 1 below [3]. It improved across all provinces between 2013 and 2018, except in East Kalimantan (Kalimantan Timur). The capital city of Jakarta Province had the lowest prevalence in 2018 at 17.7%, whereas East Nusa Tenggara recorded the highest at 42.6%. Provinces in the eastern part of Indonesia, where many development indicators lag behind other regions, have a higher prevalence of stunting. The World Bank (2020) [4] has also highlighted the regional variation in the incidence of stunting in Indonesia and further noted that the risk of stunting was higher in poor and populous districts where access to basic infrastructure of water, sanitation and hygiene (WASH) was lacking.

Fig 1

Stunting prevalence across provinces in Indonesia.

Source: Basic Health Survey (2018).

This paper examines the individual-, household-, subdistrict- and provincial-level determinants of stunting using a multilevel mixed effects model. Methodologically, this is an improvement over most of the previous literature in the field [see, for example, 13–15], as earlier studies examined the determinants of stunting in Indonesia using logistic and probit regressions. A multilevel model was appropriate for our analysis because of the hierarchical or clustered structure of our data. Young children living in the same household and community can be expected to have more similar stunting risks compared to those living in different households and communities [11, 12, 15]. Furthermore, children in the same province are more likely to have the same risk of being stunted because they have similar access to health care services and other infrastructure [8, 9, 13]. In addition, parental characteristics and household socioeconomic background have an influence on children’s eating habits and nutritional status [14, 15]. Multilevel modeling enables us to investigate individual heterogeneities and the heterogeneities between clusters and improves the estimation techniques used in previous stunting studies [see, for example, 13–15]. In addition, taking into account the clustering in the data generates more reliable standard errors of regression coefficients [16].

This paper addresses two main research questions: Do variations at the province, subdistrict, household and individual levels explain childhood stunting in Indonesia? What are the multilevel determinants of childhood stunting in Indonesia? The remaining parts of the paper are organized as follows. Section 2 discusses the data and methodology used in the paper, including the outcome variables and various control variables included in the estimation. Section 3 presents and discusses the results of our estimation. The last section presents the conclusions and provides policy recommendations based on the findings of the study.

2. Data and methods

2.1. Data source

An open access and rich dataset from the Indonesian Family Life Survey (IFLS) was used to estimate the multilevel determinants of stunting across households, subdistricts and provinces in Indonesia (https://www.rand.org/well-being/social-and-behavioral-policy/data/FLS/IFLS.html). The IFLS is a longitudinal survey representing 83% of the Indonesian population in 1993. The survey has five waves of data collected in 1993, 1997, 2000, 2007–2008 and 2014–2015. The survey was conducted by the RAND Institute with the cooperation of local universities and research centers in Indonesia. The first wave of the survey covered only 13 provinces, but the number was broadened to include more than 20 provinces in the last round to capture respondents’ mobility to other provinces. Ethical clearances for the surveys were provided by institutional review boards (IRBs) in the United States and Gadjah Mada University (UGM) for IFLS waves 3, 4 and 5 and by the University of Indonesia for IFLS waves 1 and 2.

This particular study focuses on the last two waves of IFLS—wave 4 (conducted between 2007 and 2008) and wave 5 (conducted between 2014 and 2015)–analyzed as repeated cross-sectional surveys. There were 13,500 households and 43,000 individuals interviewed in the fourth wave of the study. The number of respondents increased in the fifth wave to 15,900 households and more than 50,000 individuals. This study only covers young children aged five years old and below. The dataset includes a total of 8,290 young children, with 4,142 in wave 4 and 4,148 in wave 5. Using child anthropometry information, we excluded children identified as having biologically implausible values (n = 185, 2.2%) based on z-scores of height, following the World Health Organization (WHO) child growth standard, which reduced the number of children to 8,105. In addition, we removed young children for whom complete information on household and community characteristics was not available (n = 60, 0.7%). Hence, the number of young children included in our final analysis was 8,045. The proportion of respondents with missing values was insignificant (3%).

Four hierarchical levels were considered in our analysis: individual (child), household, subdistrict and province. Children (the lowest level in our mixed effects hierarchical model) were nested within households (level two). There were 6,437 households with 8,045 children aged 5 and below in the two waves of the IFLS. Approximately 20% of the total households had more than one child aged 5 and below. Households were nested within subdistricts (Kecamatan in Indonesian). There were 1,332 subdistricts recorded in the two waves of the survey. We considered the subdistrict as the third level in our model, as service provision and quality of care in primary health centers (PHCs) or puskesmas (which are expected to impact childhood stunting) are determined at the subdistrict level. The fourth level was the province, within which subdistricts were nested. We considered provinces as the highest level in our analysis; there were 21 provinces in the two waves of the IFLS used in our study.

The IFLS provides a wide range of information, including data on birth history, anthropometry of children (i.e., height, weight and body mass index), clinical and subclinical infection, dietary habits and demographic characteristics. Child health status data include both a self-reported measure of general health status and a biomarker measurement conducted by a nurse. Furthermore, the household data consist of information on parental education and parental anthropometry, antenatal care during pregnancy, consumption and wealth, and household access to WASH. Finally, the IFLS also includes other community-level data, consisting of access to nutrition-specific intervention programs in the PHCs in children’s neighborhoods.

2.2. Outcome variable

According to the 2005 World Health Organization child growth standard and Indonesian Ministry of Health guidelines [17], stunting is a measure of children’s nutritional status based on their height. Nutritional status is determined by comparing the height of children with their peers of the same age. Height was measured during the survey by trained interviewers. Then, the z-score of height was calculated using the height of children and the corresponding median and standard deviation for children of the same age, which were obtained from the anthropometry guidelines.

We then further grouped nutritional status into four categories using the calculated z-score values. The first group consists of severely stunted children whose height-for-age z-scores are more than three standard deviations below the World Health Organization child growth standards median. The second group consists of stunted children whose height-for-age z-scores are more than two standard deviations below the WHO child growth standards median. The third group is considered to have normal nutritional status, and their z-score is within two standard deviations of the WHO child growth standards median. The last group consists of children who are considered to have above normal nutritional status, with z-scores more than two standard deviations above the WHO child growth standards median.

As the focus of our analysis in this research is stunted children, for our regression analysis, we categorized children into two groups—stunted and not stunted. The stunted group consisted of children who were severely stunted (as defined above) and was represented by a dummy variable equal to one. The remaining children were categorized as not stunted and were represented by a dummy variable equal to zero.

2.3. Individual-level variables

One of this study’s particular aims was to assess the impact of children’s dietary habits on stunting. The IFLS dataset provides data on dietary diversity and consumption frequency in the one week prior to the survey for each child in the family. Studies have suggested that adequate dietary diversity lowers the odds of stunting, whereas the consumption of unhealthy snacks raises it [14, 18–20]. We constructed a binary variable capturing the intensity of unhealthy snacking as a measure of children’s dietary habits. Following Wang et al. [21], the dummy variable for unhealthy snacking took the value of 1 for children consuming unhealthy snacks more than 7 times a week (or more than once per day) and took the value of 0 for those with lower consumption frequency. Snacks in the IFLS survey include instant noodles, fast food, carbonated beverages and sweet snacks that are usually high in salt, fat and sugar and low in micronutrient content. However, unhealthy snack data are only available in IFLS wave 5. Therefore, we conducted two estimations. Our first estimation used both waves of the IFLS (wave 4 and wave 5) and excluded the unhealthy snack variable, whereas our second estimation used only the IFLS wave 5 and included the unhealthy snack variable.

Demographic characteristics of children, such as gender, have also been considered predictors of stunting [15, 22, 23]. A dummy variable was constructed to test whether male children have a higher risk of being stunted, as earlier studies have found [15, 22, 23]. The study also included neonatal weight as one of the predictors of stunting, as suggested by Tiwari et al. [23] and de Silva and Sumarto [15]. According to the World Health Organization [24], a newborn baby weighing fewer than 2.5 kg is considered a small baby. Our dummy variable assumed a value of 1 if the neonatal weight was fewer than 2.5 kg and 0 otherwise.

Infections (both clinical and subclinical), such as enteric infections and diarrhea, contribute to stunting [25]. The IFLS dataset has information on children’s acute morbidity in the four weeks prior to the survey, including the occurrence of diarrhea (at least three times a day). We constructed a dummy variable that assumed a value of 1 for children who had experienced acute diarrhea (either with blood or mucous or who had a watery stool with a pale color) and 0 otherwise.

2.4. Household-level variables

Household-level variables consisted of both parental- and household-level factors that have been found to affect stunting in the wider literature. Previous studies have indicated that maternal education is associated with a lower risk of stunting [7, 15, 23]. Mothers are primary caregivers in most Indonesian households, and we controlled for maternal education using years of schooling. We also included the mother’s height in our model to control for maternal stature, in line with previous literature [11, 15, 26]. According to Beal et al. [7], a short mother (fewer than 145 cm tall) has a higher risk of having stunted children. Therefore, we created a dummy variable equal to 1 for mothers whose height was fewer than 145 cm and 0 otherwise.

Previous studies have shown a strong association between stunting and socioeconomic background. De Silva and Sumarto [15] and Mani [27] presented evidence that higher household consumption expenditure per capita and greater wealth lower the prevalence of stunting among children. The IFLS dataset provides detailed information on food and nonfood consumption. Food expenditure consists of spending on staples, vegetables, dried foods, meat and fish, beverages and prepared food. The expenditure on nonfood items includes spending on education, electricity, water, telephone, household items, recreation, entertainment, clothing and medical costs. We used the quartile of total consumption expenditure per capita as a measure of household wealth, categorizing the expenditure data as quartiles labeled Q1 to Q4, with Q1 being the lowest quartile of expenditure. We included three dummy variables that identified poor households in the bottom quartile of the consumption expenditure distribution as a base (coded 0) compared with the higher quartiles of Q2, Q3 and Q4.

To account for possible differences in health service delivery between rural and urban areas, we included a dummy variable for place of residence, which equaled 1 for children living in rural areas and 0 for those living in urban areas. Some earlier studies have suggested that childhood stunting is more prevalent in rural areas [15, 23, 28], and we examined whether this also held true for Indonesia.

Finally, this study also examined the contribution of household access to WASH. Previous studies have underlined the importance of sanitation and access to health services in lowering the risk of stunting [12, 15, 29, 30]. This study considered the association between access to clean WASH and the risk of stunting. Three dummy variables were introduced in the empirical model to capture WASH variables. The main source of drinking water in households was coded as 1 if the source of drinking water was either tap water (piped water and groundwater) or mineral water and 0 otherwise. The availability of a toilet was coded as 1 if there was a toilet with its own septic tank for sanitation and 0 otherwise. Additionally, the disposal of garbage was coded as 1 if garbage was disposed into a trash can that is collected by a sanitation service and 0 otherwise.

2.5. Community-level variables

We examined the effect of nutrition-specific intervention programs in PHCs in young children’s neighborhoods. The IFLS dataset provides information on access to three nutrition-specific services provided by PHCs: growth and development monitoring, additional treatment for malnutrition and additional nutrition for the poor. A dummy variable was created to differentiate between young children who have access to all three nutrition-specific services (coded as 1) and those who do not (coded as 0).

2.6. Statistical methodology

The study examined the individual-, household-, community- and province-level determinants of stunting using a multilevel mixed effects logistic model. Multilevel models allowed us to take into account the hierarchical structure in our data and calculate residual components at each level in the hierarchy. The individual-level variables we controlled for included children’s demographic characteristics, birthweight, history of diarrhea and dietary habits. The family-level variables included mother’s education, mother’s stature and family consumption. We also included contextual variables at the community and district levels in the form of access to clean water, hygiene and sanitation, availability of nutritional services and area of residence (rural/urban). The adjusted odds ratio (aOR) of the fixed effects reflects the likelihood of stunting in children. The intraclass correlation coefficient (ICC) measures the degree of homogeneity within clusters in the risk of stunting. The partitioning of the residual components at each level enables us to see the effect of “unobserved” province- and community-level characteristics on stunting. Another advantage of multilevel models is that they allow simultaneous estimation of group effects and the effect of group-level predictors.

We started our estimation by running mixed effects logit null models without covariates. The first null model introduced a random intercept term at level 4 (province level). The second null model included an additional random intercept term at level 3 (subdistrict level). Additionally, the third null model included an additional random intercept term at level 2 (household level). The null models provided an indication of how much variation in stunting each additional cluster accounts for. Predictors were introduced in a stepwise manner, starting with child-level covariates, and followed by family- or household- and community-level covariates.

3. Results

3.1. Characteristics of study participants

Table 1 depicts the children’s characteristics in the two waves of the IFLS data. It can be seen from the table that the proportion of stunted children was 26.29%, which consisted of 8.88% severely stunted and 17.41% stunted children.

Table 1

Characteristics of study participants.

	Summary statistics for 2007 & 2014 combined		2014		2007
Outcome variable
Stunted (%)	8045	26.29 (0.440)	4,044	28.73 (0.45)	4,001	23.82 (0.43)
Independent variables
Individual (child-)level data
Unhealthy snacking*
High frequency (%)	2282	56.43 (0.496)	2282	56.43 (0.496)
Low frequency (%)–base	1762		1762
Gender
Male (%)	4183	52 (0.499)	2116	52.32 (0.50)	2067	51.66 (0.50)
Female—base	3862		1928		1934
Baby size
Small baby (%)	339	4.21 (0.200)	176	4.35 (0.04)	163	4.07 (0.04)
Normal weight baby—base	7706		3868		3838
Diarrhea
Acute diarrhea (3 times/day in the past 4 weeks) (%)	1244	15.46 (0.362)	6959	17.19 (0.38)	549	13.72 (0.34)
Not experienced acute diarrhea—base	6801		3349		3452
Household level data
Mother’s education (years of schooling)**	8045	8.48 (4.272)	4044	8.94 (4.23)	4001	8.02 (4.27)
Mother’s stature
Mother short (<145 cm)	3634	45.17 (0.497)	1979	48.94 (0.50)	1655	41.36 (0.49)
Normal height (145 cm & above)–base	4411		2065		2346
Consumption quartile (Rupiah)
First quartile—poor (Q1)	2023	230,947 (107,719)	1010	317,224 (82,955)	1013	144,905 (38,446)
Second quartile (Q2)	2018	413,914 (166,956)	1010	570,282 (73,783)	1008	257,236 (35,790)
Third quartile (Q3)	2008	644,466 (243,638)	1022	867,705 (1,077,70	986	413,076 (60,048)
Fourth quartile (Q4)	1996	2,314,440 (4,812,522)	1002	2,270,691 (2,227,326)	994	2,358,541 (6,444,053)
Regional differences
Rural (%)	3553	44.16 (0.497)	1680	41.54 (0.49)	1873	46.81 (0.50)
Urban (%)–base	4492		2364		2128
Access to clean water (%)	8045	96.35 (0.188)	4044	96.98 (0.17)	4001	95.70 (0.20)
Access to sanitation (%)	8045	68.84 (0.463)	4044	74.18 (0.44)	4001	63.43 (0.48)
Access to hygiene (%)	8045	32.78 (0.47)	4044	36.35 (0.48)	4001	29.17 (0.45)
Community level variable
Nutrition-specific intervention (%)	7795	27 (0.444)	3930	26.41 (0.44)	3865	27.61 (0.45)

*) Data on unhealthy snacking is only available in wave 5 of the IFLS.

**) Continuous variable of mother’s years of schooling.

The table also shows that more than half of the children (56.43%) consumed unhealthy snacks frequently based on Wave 5 data. In terms of gender composition, 52% of the children were boys. Moreover, the proportion of children who suffered from acute diarrhea was 15.46%. However, only 4% had been small at birth, with a weight of less than 2.5 kg.

Mothers in our sample had, on average, 8 years of education, which is equivalent to reaching the second year of junior high school in Indonesia. In terms of stature, 45.17% were fewer than 145 cm tall and were, therefore, classified as short following Beal et al. [7]. For household-level variables, the data revealed that the monthly average household consumption expenditure per capita increases as we move from the bottom quartile to the top quartile. The average for the bottom quartile was 230,947 Rupiah (equivalent to $57.32 USD Purchasing Power Parity in 2014) compared to 2.3 million Rupiah (equivalent to $574.40 USD PPP in 2014) in the top quartile.

The proportion of children living in rural areas was 44.16%. From the descriptive statistics for access to WASH facilities, we can see that more than 96% of households had access to clean water, only 68.8% had access to sanitation, and an even lower proportion (approximately 32.8%) had access to hygiene. Last, the data show that 27% of families had access to all three nutrition-related services from PHCs.

3.2. Multilevel analyses of stunting determinants

We present the results from the estimation of our three null models in Table 2. As noted previously, the first null model includes random effects at the province level, the second null model includes random effects at both the province and the subdistrict levels, and the third null model includes random effects at the province, subdistrict and household levels.

Table 2

Null models: With province effects (Model 1), province and subdistrict effects (Model 2) and province, subdistrict and household effects (Model 3).

Variable	Model 1 (95% CI)	Model 2 (95% CI)	Model 3 (95% CI)
Constant	-1.231 (-1.459, -1.004)	-1.371 (-1.572, -1.169)	-1.561 (-1.803, -1.318)
Between province variance	0.186 (0.069, 0.503)	0.116 (0.034, 0.378)	0.144 (0.043, 0.472)
Between subdistrict variance	-	0.378 (0.274, 0.522)	0.452 (0.326, 0.648)
Between household variance	-	-	0.795 (0.475, 1.329)
ICC (province)	0.053 (0.0205, 0.133)	0.0306 (0.009, 0.094)	0.031 (0.009, 0.09)
ICC (province and subdistrict)	-	0.131 (0.095, 0.177)	0.127 (0.009, 0.174)
ICC (province, subdistrict and household)	-	-	0.297 (0.226, 0.379)
Observations (young children)	8045	8045	8405
Group level	Province	Province; subdistrict	Province; subdistrict; household
Number of groups	21 provinces	21 provinces	21 provinces
		1332 subdistricts	1332 subdistricts
			6437 households
Likelihood ratio test (LR)	85.98	218.97	242.89
Prob > chi2	(0.000)	(0.000)	(0.000)

Model 1: Two-level model (individual and household).

Model 2: Three-level model (individual, household, subdistrict).

Model 3: Four- level model (individual, household, subdistrict and province).

ICC: Intraclass correlation coefficient.

We can see from the table that the likelihood ratio test statistic is highly significant for all three null models (Models 1–3). This shows that multilevel models are a better fit to the data than a single-level model. Moreover, we also conducted likelihood ratio tests comparing the four-level model (with random effects at the province, subdistrict and household levels) with three- or two-level models and found that the four-level model was a better fit for the data. Thus, stunting in Indonesia varies by province, subdistrict and household, and analysis of stunting needs to take into account variations at all these levels.

Fig 2 below shows a caterpillar plot of the residuals from all 1,332 subdistricts in the sample with 95 percent confidence intervals. The residual shows the departure of the risk of stunting for subdistricts from the average (overall mean). The graph reveals that for a substantial number of subdistricts, the 95% confidence interval does not overlap with the horizontal line at zero, indicating that stunting in these subdistricts is significantly above average (above the zero line) or below average (below the zero line).

Fig 2

Sub-districts caterpillar plot.

Table 2 also presents the intraclass correlation coefficient (ICC), which measures the degree of homogeneity in the risk of stunting within clusters (household/subdistrict/province). A higher ICC indicates a stronger correlation in the risk of stunting within a cluster. We can see from Model 3 that the level-4 intraclass correlation at the province level is 0.031, whereas the level-3 intraclass correlation at the subdistrict-within-province level is 0.127 and the level-2 intraclass correlation at the household-within-subdistrict level is 0.297. This indicates that within-subdistrict differences (i.e., differences between households and children) are more important in explaining the variation in the risk of stunting in Indonesia than differences between subdistricts and provinces.

We can also calculate the variance partition coefficients (VPCs) to see the contribution of unobserved cluster characteristics to the risk of stunting at each level in our model. The VPCs indicate that 3% of the variation in stunting is due to differences between provinces, whereas 10% is due to differences between subdistricts (within provinces). Differences between households (within-subdistricts) account for 17% of the variation, whereas individual differences between children account for 70% of the variation in stunting.

We present our estimation results in Table 3. Our most comprehensive model controls for household- and individual child-level covariates and community-level contextual characteristics (in particular, access to WASH). Two main estimates are generated from the stepwise regression in S1 Appendix, as discussed in the methodology section. The first estimation employs two complete waves of the dataset (waves 4 and 5); these results are presented in Columns 1 and 2 of Table 3. The second estimation relies on the latest wave of data (wave 5), as information about dietary habits, particularly unhealthy snacking behavior, is only available in wave 5; these results are presented in Table 3 Columns 3 and 4.

Table 3

Multilevel mixed effects model logistic regression results of stunting status.

	Two waves of data (2007&2014)		One wave of data (2014)
Variables	Adjusted odds ratio (aOR)	(95% CI)	Adjusted odds ratio (aOR)	(95% CI)
Individual level
Unhealthy snacking
High frequency	-	-	1.30***	(1.08–1.58)
Gender
Male	1.17***	(1.04–1.32)	1.26**	(1.04–1.51)
Baby size
Small baby	2.29***	(1.73–3.01)	2.51***	(1.63–3.88)
Diarrhea
Acute diarrhea (3 times/day in the past 4 weeks)	1.27***	(1.08–1.49)	1.30**	(1.02–1.65)
Family/household level
Mother’s education	0.96***	(0.94–0.97)	0.94***	(0.92–0.97)
Mother’s stature
Mother short (<145 cm)	1.19***	(1.05–1.34)	1.19*	(0.99–1.44)
Consumption quantile
2nd quantile	0.77*	(0.65–0.91)	0.76	(0.59–0.98)
3rd quantile	0.73***	(0.61–0.87)	0.71***	(0.54–0.93)
4th quantile	0.56***	(0.46–0.68)	0.50***	(0.37–0.68)
Community level
Clean water
No access	1.36*	(0.98–1.89)	1.22	(0.71–2.12)
Sanitation
No access	1.27***	(1.10–1.46)	1.23*	(0.98–1.54)
Hygiene
No access	1.52***	(1.28–1.80)	1.75***	(1.34–2.27)
Regional differences
Rural	1.19**	(1.02–1.40)	1.10	(0.86–1.41)
2014 time dummy	1.50***	(1.32–1.70)	-
Intercept	0.18***	(0.13–0.25)	0.22***	(0.13–0.37)
Province effect—coef (std.dev)	0.13 (0.07)		0.20 (0.13)
Subdistrict effect—coef (std.dev)	0.19 (0.05)		0.44 (0.13)
Household effect—coef (std.dev)	0.79 (0.210)		1.31 (0.56)
Observations	8045		4044

Note: Unhealthy snacking data are only available in Wave 5 of the IFLS.

An odds ratio is statistically significant at either 1 percent (***), 5 percent (**) or 10 percent (*) of the confidence intervals.

For the individual variables, our findings indicate that a high frequency of snack consumption is associated with a higher risk of children being stunted. In terms of magnitude, frequent snacking increases the risk of being stunted by 30% (95% CI 1.08–1.58), as shown in Table 3 (Column 3) [18, 28, 30–32, 34].

Additionally, our results indicate that boys have a higher risk of being stunted than girls, a finding that is consistent with previous studies. The likelihood of stunting was higher by 17% (95% CI 1.04–1.32) for boys than girls using two waves of data and by 26% (95% CI 1.04–1.51) using the Wave 5 dataset [32-35].

Another individual-level factor that is associated with a higher risk of stunting is neonatal weight. This study found that small babies weighing fewer than 2.5 kg at birth have a 2 times higher risk of being stunted than babies of normal weight. According to our model utilizing both waves of data (Table 3, Column 1), the odds of stunting for a small baby were 2.29 (95% CI 1.73–3.01).

The final individual-level predictor of stunting is infection with diarrhea. The estimations revealed that suffering from acute diarrhea (i.e., more than three times a day) in the four weeks prior to the survey is associated with a higher risk of being stunted, with an odds ratio of 1.27 (95% CI 1.08–1.49) in the model using Wave 4 data (Table 3, Column 3) and an odds ratio of 1.30 (95% CI 1.02–1.65) in the model using Waves 4 and 5 (Column 1).

Our results also indicate that maternal characteristics, such as maternal stature and years of schooling, and household characteristics, such as household wealth quartile, place of residence and access to basic infrastructure (WASH), are significantly associated with stunting. Children whose mothers are shorter than 145 cm have a 19% (95% CI [1.05–1.34] in the model using Waves 4 and 5) greater risk of being stunted. Moreover, the number of maternal years of schooling lowered the risk of being stunted (odds ratio of 0.96, 95% CI [0.94–0.97] in the model using data from Waves 4 and 5).

Our results also suggest that children from poor households have a higher risk of being stunted. The risk of stunting decreased by 23% (95% CI [0.65–0.91] in models using Waves 4 and 5 data [Table 3, Column 1]) for children in the second quartile of the wealth distribution. The risk of stunting was further lowered for children in the fourth quartile by 44% (95% CI [0.46–0.68] in the Waves 4 and 5 model).

It is apparent from our results that children living in rural areas have 19% (95% CI [1.02–1.40] in the Waves 4 and 5 model) greater odds of being stunted than children living in urban areas. This study also found that having no access to clean water increases the risk of being stunted by 36% (odds ratio of 1.36, [95% CI (0.98–1.89)]). Similarly, having no access to sanitation was associated with a 27% higher risk of being stunted (odds ratio of 1.27, [95% CI (1.10–1.46)]). A higher risk of stunting was also associated with a lack of access to hygiene at 52% (odds ratio of 1.52, [95% CI (1.28–1.80)]).

The stepwise estimation using the multilevel mixed effects model shows that the risk of stunting is higher if the children have no access to nutrition services programs in PHCs, but this association is not statistically significant (odds ratio of 1.06, [95% CI (0.88–1.26)]). Finally, the estimation shows that the odds of being stunted for young children in 2014 was higher than that for young children in 2007 (odds ratio of 1.50, [95% CI (1.32–1.70]). The data show that the proportions of stunted children in 2007 and 2014 were 24% and 29%, respectively.

4. Discussion

Our analysis found that stunting is associated with several individual-, family-/household- and community-level variables. Frequent unhealthy snacking, male sex, low birthweight and diarrhea increase the risk of being stunted. Family characteristics that contribute to stunting risk include short maternal stature and having a family with a low socioeconomic status. In terms of community characteristics, this study found that living in rural areas increases the risk of stunting by 19%. The risk of stunting is also higher for children living in a community with a lack of access to clean WASH.

The results show that there is a positive association between a high frequency of snack consumption and the risk of children being stunted. Studies have indicated that snacking is becoming more prevalent in Indonesia, both in rural and in urban areas. A study by Sekiyama et al. [19] showed that one-third of food consumed by young children in West Java, Indonesia can be categorized as snack food. High snack consumption has a detrimental effect on children’s development because snacks contain mostly fat (59.6%) and energy (40%) but have a lower density of protein and micronutrients. Black et al. [31] and Tarwotjo et al. [32] suggested that a lack of micronutrient intake, such as calcium and vitamin A, adversely affects children’s linear growth. The World Health Organization has reported that chronic deficiencies in micronutrients are experienced by more than 2 billion people worldwide [29]. Micronutrients are vital for children’s development because they have a significant role in bone formation (calcium), long bone growth (zinc) and intrauterine femur length increase (supplements) [33-35].

The importance of micronutrients for children has also been established by other studies measuring the effect of the School Feeding Program (SFP) on children’s development [36-38]. A quasi-experimental study by Metwally et al. [36] revealed that feeding children pie made of flour fortified with vitamins and minerals has a positive effect on cognitive development. Nevertheless, the effect of SFP on nutritional status takes a longer time to manifest, so the effect is not statistically significant. Another study of SFP in rural Kenya showed that the intake of minerals, such as zinc and iron, and vitamins may increase children’s appetite, muscle growth and physical activity [37]. The importance of micronutritients on children’s diet has also been found in the impact evaluation of SFP programs in rural Uganda [38]. It was found that children fed one or two eggs per day gained more height and weight because eggs are a source of 13 essential micronutritients and protein, which are essential for children’s development.

Our finding that boys have a higher risk of being stunted than girls is consistent with previous studies. Adair and Guilkey [35], Moestue [34] and Wamani et al. [33], for example, found that the height-for-age z-score for girls is higher than that for boys; thus, girls have a lower prevalence of stunting in Sub-Saharan Africa and China. The literature suggests that higher stunting prevalence among boys may be explained by complementary feeding practices. Boys receive premature complementary foods, as parents perceive that breastfeeding is not sufficient to fulfil the greater energy intake they believe is required for baby boys [18]. A study by Tumilowicz et al. [39] showed that among Guatemalan children, more boys than girls aged 2–3 months are fed complementary foods. The IFLS dataset for young Indonesian children also shows that complementary feeding is initiated for boys earlier than for girls. On average, complementary feeding for Indonesian young children in the IFLS data of Waves 4 and 5 started at 19.74 weeks for baby girls and 18.91 weeks for baby boys; the difference was statistically significant in a t-test.

In addition, these complementary feeding practices do not benefit boys because they are more likely to be fed more meals than girls. Tumilowicz et al. [39] presented evidence that baby boys are fed two more meals than girls in a 24-hour period. Premature complementary feeding practices have a detrimental effect on young children because they pose a greater risk of catching infectious diseases [40, 41]. In addition, early introduction of food before babies reach 6 months of age has no significant effect on children’s length or weight development [23].

As presented in the Results section, we found that babies of lower birth weight (less than 2.5 kg at birth) have a two times higher risk of being stunted than babies of normal birthweight. This finding is consistent with a study by Tiwari et al. [23], which found that average and above average weight newborn babies have lower odds of being stunted than smaller babies. Furthermore, A Saleemi [42] and Varela-Silva et al. [43] also showed that the risk of smaller babies being stunted is three times higher than that for other babies.

According to Schmidt et al. [26], neonatal weight and particularly length are good indicators of the status of children’s nutrition in the future. Low neonatal weight and short length may be an indication of intrauterine growth restriction (IUGR), meaning that babies are not growing at a normal rate inside the womb during pregnancy. Lower neonatal weight and height may also be related to maternal malnutrition during pregnancy, which in turn influences the development of the baby [23]. Moreover, small babies may also have been born prematurely, which means they may not have fully developed during pregnancy [23]. In the long run, IUGR may lead to numerous developmental issues, such as growth retardation, lower cognitive ability development and poor neurodevelopmental outcomes.

In terms of infection with diarrhea, our findings are consistent with previous studies by Bardosono et al. [44], Beal et al. [7] and Tiwari et al. [23]. These authors also found that diarrhea is associated with stunting. According to Richard et al. [45], diarrhea is a particularly common health issue in developing countries, as households lack access to clean water and sanitation. Drinking from unimproved water sources and drinking untreated water increase the risk of diarrhea due to intestinal infections from a variety of bacteria, parasites and viruses. Richard et al. [45] suggest that the impact of diarrhea on linear growth is particularly strong when young children suffer from multiple episodes of diarrhea in the first 24 months of their lives. Furthermore, diarrhea may lead to growth retardation if the occurrence of diarrhea coincides with a lack of good-quality food and poor access to health care. Diarrhea has a detrimental effect on linear and ponderal growth, as it lowers dietary intake, escalates metabolic demands and lessens nutrient absorption in the gut [45].

Our analysis found a strong association between maternal stature and childhood stunting. Semba et al. [11] similarly established a strong association between maternal stature and stunting, and Schmidt et al. [26] contended that maternal stature and neonatal weight are the strongest predictors of child stunting. Maternal stature is perceived to be a good indicator of intragenerational undernutrition. Prendergast and Humphrey [46] argued that having a stunted mother is relevant in explaining stunting prevalence in children because of the importance of the nutritional status of the mother on children’s stunting. Mothers suffering from undernutrition may have a higher risk of having stunted children because of their significant influence, especially in the first 500 days of a child’s life.

The association between maternal years of schooling and risk of stunting is also supported by previous studies [see, for instance, 8, 19, 41]. A better educated caregiver is perceived as having appropriate maternal nutritional knowledge [44]. Furthermore, Semba et al. [11] found that better educated parents tend to engage in more protective caregiving; for example, they may do this by ensuring that their children receive vitamin A capsules, are fully immunized, have access to better sanitation and consume iodized salt.

The results also provide evidence that children from poor households are at higher risk of being stunted. The strong association we found between family wealth and stunting has been established in the literature [7, 11, 13, 27, 47–50]. Bardosono, Sastroamidjoo and Lukito [44] argued that poor families lack the resources to consume high-quality nutritional foods and access health care.

Our result shows the positive association between living in rural areas and the risk of being stunted. Previous studies support the finding that children living in rural areas have a higher risk of stunting than their peers in urban areas [see, for instance, 15, 23, 28]. According to Mahendradata et al. [51] and Mulyanto, Kurst and Kringos [52], both the demand and the supply of health care vary between urban and rural areas. People living in urban areas have more access to health care and other related infrastructure, such as roads that reduce the travel time to health care facilities. Meanwhile, access to health services in rural areas is more limited. According to Sparrow and Vothknecht [53], 6.3% of subdistricts in Indonesia have no access to PHCs. These districts are mostly in rural areas outside Java Island. Furthermore, their study reported that 4.2% of PHCs in rural areas have no physician serving in health facilities. Physical health infrastructure, such as working incubators, lab facilities and outpatient polyclinics, is also more limited in rural areas.

Another study by Schmidt et al. [26] suggested that a higher prevalence of stunting in rural areas compared with urban areas is related to a greater sensitivity to changes in food prices. Families in rural areas are more sensitive to food price increases because they allocate two-fifths of their budget for staple needs. As the price of foods increases, the purchasing power of rural families declines, making it harder to fulfil the essential nutritional requirements of their children.

Our results also show that a lack of access to WASH is associated with stunting among young children in Indonesia. Secure access to WASH infrastructure is critical. Young children are more prone to diarrhea, intestinal worm infection and environmental enteropathy when households have poor WASH facilities [9]. These infections may lead to nutritional issues. For example, children may lose their appetites, so they might consume less food than they need. In addition, these types of infections can lead to the malabsorption of nutrition and chronic immune activation. Finally, infections may induce fever, which requires the body to burn more food and exert energy to fight the infection instead of using it for physical development.

The insignificant association we found between stunting and access to nutrition services in PHCs may be due to the limited capacity of the Indonesian nutrition program, as shown in a report by UNICEF [54]. In terms of coverage of the nutrition-specific interventions recommended by The Lancet [55], the Indonesian government has only adopted 4 out of 10. Of the remaining six programs, four are partially covered, and two are not included. In terms of health infrastructure, the proportion of nutritionists per head in the population is low in Indonesia and varies across regions. In the larger provinces on Java Island, such as in Central Java, the number of nutritionists to every 1,000 people was as high as 43.14 in 2017. However, the number was much lower in the provinces outside Java Island. For example, in East Nusa Tenggara, one of the provinces with a high prevalence of stunting, the number was only 12.04.

Our results also show that the risk of stunting for young children was higher in 2014 than in 2007. Previous studies have revealed that after a significant decline in stunting prevalence in Indonesia in the 1990s and early 2000s [28], the prevalence remained unchanged in the 2000s [56]. Our finding aligns with national data from the Indonesia Socio-Economic Survey and Basic Health Survey, which shows that stunting prevalence was higher in 2013 than in 2007. A World Bank publication acknowledges the important role that nutrition programming and surveillance at the village level in the 1980s played in reducing the prevalence of malnutrition in Indonesia. However, this massive program has experienced setbacks and lost the close attention of the government. Furthermore, Indonesia has undergone decentralization, which has reduced the effectiveness of nutrition programs in improving children’s nutritional status due to weak management and poor governance [57].

4.1. Study limitations and strengths

This study highlights the importance of family- and community-level variables in stunting, in addition to individual characteristics. The assessment of the different levels of clusters in this study (i.e., province, districts and households) facilitates understanding of how contextual factors contribute to stunting in children. Hence, our study demonstrates the urgency of addressing not only personal- or individual-level factors but also household- and community-level factors to reduce stunting prevalence.

The limitations of this study are listed here. First, regarding the source of the data, the two waves of the IFLS data used in our paper are representative of approximately 83% of the Indonesian population, covering 21 provinces; thus, some areas of eastern Indonesia were excluded from our analysis. Additionally, IFLS Wave 4 data do not capture unhealthy snacking behavior among young children. This variable is considered an important proxy for measuring children’s dietary habits; therefore, we used only Wave 5 data to examine this issue. Second, from a methodological perspective, the cross-sectional nature of our analyses limited our ability to infer causation. Moreover, some data were based on self-reported information, for example, the birth weight of the children, and thus may be susceptible to measurement error. In addition, our study did not control for the random slope component in the model. The assessment of nutritional services was conducted from the supply side (i.e., the availability of nutritional services in the community). However, we adjusted for several important confounders and took into account some unobserved characteristics through multilevel modeling.

5. Conclusion

Stunting remains a development issue in Indonesia, with approximately 30% of young children being stunted. This study examined the multilevel determinants of stunting among young children in Indonesia. At the contextual level, the ICC showed that there is a correlation in the risk of stunting for children living in the same province. The correlation becomes stronger for children living in the same subdistricts. Finally, the strongest correlation of the risk of stunting was found among children living in the same household. Moreover, the likelihood ratio tests revealed that stunting in Indonesia varies by province, subdistrict and household level, and analysis of stunting needs to consider variations at all these levels.

At the child and family levels, our results identified several statistically significant determinants of childhood stunting. In terms of individual characteristics, being a boy, having a low neonatal weight and experiencing acute diarrhea were associated with stunting. In terms of family characteristics, we found that mothers’ characteristics, specifically maternal stature and maternal education, were associated with stunting. In contrast, living in a family with a higher socioeconomic status lowered the risk of stunting, and the risk of stunting was much lower for children in the highest quartile of the wealth distribution. Finally, in terms of community characteristics, we found that living in rural areas increased the risk of stunting by 20%. The risk of stunting was also higher for children living in a community with lack of access to clean WASH.

From a policy perspective, our findings suggest that tackling stunting in Indonesia requires substantial effort to create spaces that assist policy implementation in establishing supportive multilevel conditions. These include addressing both individual- and household-level factors that support good child nutrition and development. Healthy eating habits, mothers’ education and awareness, socioeconomic characteristics and the availability of WASH matter.

(DOCX)

Click here for additional data file.

5 Jul 2021

PONE-D-21-06348

Beyond Personal Factor: Multilevel Determinants of Childhood Stunting in Indonesia

PLOS ONE

Dear Dr. Mulyaningsih,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

The study was focusing on assessing the Multilevel Determinants of Childhood Stunting in Indonesia. However, fundamental issues are indicated to be provided by aggressive editing for most of the study sections. Consider revising the spelling and grammar throughout the manuscript for increased clarity.

Please note that your manuscript was reviewed by 2 experts in the field. There is consensus agreement that the idea of the article is interesting but also the majority detected sections that required additional work. The reviewers identified many important problems and provided copious comments (enclosed).

Please submit your revised manuscript by Aug 19 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Ammal Mokhtar Metwally, Ph.D (MD)

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: https://www.youtube.com/watch?v=_xcclfuvtxQ

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: No

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The statistical tests applied and the technicality of the manuscript are sound. However, the manuscript requires some revision with regard to its structure and language. The manuscript is very lengthy and difficult to read it.

Reviewer #2: MAJOR COMMENTS

(1)

How much missing data was there on each variable? How was missing data handled in the models?

(2)

The conclusion to this paper needs re-working. Too much of the conclusion is a repetition of material from earlier in the paper.

A suggested structure for the conclusion would be as follows:

• Brief review of main results

• Strengths of the study

• Limitations of the study, including any unobserved confounders which may affect the results

• Policy implications of the study

• Further work which should be undertaken using this data or in future studies

MINOR COMMENTS

There are a number of places in which the grammar could be improved.

TITLE OF PAPER

“Beyond Personal Factor: Multilevel Determinants of Childhood Stunting in Indonesia”

suggest

“Beyond Personal Factors: Multilevel Determinants of Childhood Stunting in Indonesia”

P 2

“Children health outcomes are poor in Indonesia despite the..”

suggest

“Child health outcomes are poor in Indonesia despite the..”

“Regarding to risk factor, Beal et al., (2018) (4), for example, establish…”

suggest

“With regards to risk factors, Beal et al., (2018) (4), for example, establish…”

P 3

“In term of protective factors of stunting, previous studies show…”

suggest

“In terms of protective factors for stunting, previous studies show…”

P 4

“…taking into account the clustering in the data generates a more reliable standard errors of regression coefficients…”

suggest

“…taking into account the clustering in the data generates more reliable standard errors of regression coefficients…”

“The survey has five waves of data: 1993, 1997, 2000, 2007-2008 and 2014-2015 rounds.”

suggest

“The survey has five waves of data, collected in 1993, 1997, 2000, 2007-2008 and 2014-2015.”

P 5

“The first wave of the survey had covered only 13 provinces but the number had been broadened to include…”

suggest

“The first wave of the survey covered only 13 provinces but the number has been broadened to include…”

“Using a-repeated cross-sectional survey…”

suggest

“Using a repeated cross-sectional survey…”

P 6

“The children health status data is a self-reported measure of general health status…”

suggest

“The child health status data is a self-reported measure of general health status…”

P 7

“Following Wang et al., (2018) (17), the dummy variable for unhealthy snacking will take on the value of 1 for children consuming unhealthy snacks for more than 7 times a week…”

suggest

“Following Wang et al., (2018) (17), the dummy variable for unhealthy snacking will take the value 1 for children consuming unhealthy snacks more than 7 times a week…”

“Demographic characteristics of children such as gender is also considered as a predictor…”

suggest

“Demographic characteristics of children such as gender are also considered as a predictor…”

P 8

“This study further constructs the expenditure into quartile data of the bottom 25 percentile (Q1), between 25-50 percentile (Q2), between 50-75 percentile (Q3) and those in the top 25 percentile (Q4).”

suggest

“This study encodes the expenditure data as quartiles, labelled Q1 to Q4, with Q1 being the lowest quartile of expenditure.”

P 12

“…incorporating mother stature, mother education and socio-economic status of the family…”

suggest

“…incorporating mother’s stature, mother’s education and socio-economic status of the family…”

“After cleaning the data, there are 8105 number of children used in the analysis.”

suggest

“After cleaning the data, there are 8105 children used in the analysis.”

P 18

“On average, complementary feeding is started at 19.74 weeks for baby girls and 18.91 weeks for baby boys and the difference is statistically significant using t-test.”

suggest

“On average, complementary feeding is started at 19.74 weeks for baby girls and 18.91 weeks for baby boys; this difference is statistically significant using a t-test.”

TITLE FOR TABLE 3

“Stepwise of Multilevel Mixed Effect Model Logistic Regression”

suggest

“Results of stepwise Multilevel Mixed Effect Logistic Regression Models”

P 22

“Children whose mothers are shorter than 145cm have 19 percent [95% CI (1.05-1.34) in model using wave 4 and 5] more risk to be stunted.”

suggest

“Children whose mothers are shorter than 145cm have a 19 percent [95% CI (1.05-1.34) in model using Waves 4 and 5] greater risk of being stunted.”

“Prendergast & Humphrey (2014) (39) argue that stunted mother is relevant in explaining stunting prevalence considering the importance of the nutritional status of the mother on children stunting.”

suggest

“Prendergast & Humphrey (2014) (39) argue that having a stunted mother is relevant in explaining stunting prevalence in children because of the importance of the nutritional status of the mother on child stunting.”

P 23

“Moreover, maternal years of schooling lowers the risk of being stunted [odds ratio of 0.96 95% CI (0.94-0.97) in model using two waves of 4 and 5].”

suggest

“Moreover, the number of maternal years of schooling lowers the risk of being stunted [odds ratio of 0.96 95% CI (0.94-0.97) in the model using data from Waves 4 and 5].”

“Our estimation results also suggest that children from poor households have a higher risk to be stunted.”

suggest

“Our results also suggest that children from poor households have a higher risk of being stunted.”

“According to Mahendradata et al (2017) (44) and Mulyanto, Kurst and Kringos (2019) (45), both demand and supply of health care are varied across urban and rural areas.”

suggest

“According to Mahendradata et al (2017) (44) and Mulyanto, Kurst and Kringos (2019) (45), both demand and supply of health care vary between urban and rural areas.”

P 24

“People living in the urban areas have more access to health care and other related infrastructure…”

suggest

“People living in urban areas have more access to health care and other related infrastructure…”

“The physical health infrastructure such as working incubators, lab facilities and outpatient polyclinics are more limited in the rural areas.”

suggest

“The physical health infrastructure such as working incubators, lab facilities and outpatient polyclinics are more limited in rural areas.”

“Families in rural areas are more sensitive to a food price increase because they allocate 2/5 of their budget for a staple.”

suggest

“Families in rural areas are more sensitive to food price increases because they allocate two fifths of their budget for staple needs.”

P 25

“In terms of health infrastructure, the proportion of nutritionists per population is low and varied across regions in Indonesia.”

suggest

“In terms of health infrastructure, the proportion of nutritionists per head of population is low and it varied across regions in Indonesia.”

“This finding is also correspondents with national data of Indonesia Socio-Economic Survey and Basic Health Survey that the stunting prevalence was remained high in 2013 compared to the 2007 survey and the figure is even slightly higher for 2013 survey.”

suggest

“This finding is in agreement with national data from the Indonesia Socio-Economic Survey and Basic Health Survey that stunting prevalence was higher in 2013 compared to the 2007 survey.”

P 26

“Nevertheless, the massive program had been set backs that government had loss attention and in the same time Indonesia undergo decentralization that reduced the effectiveness of nutrition program due to weak management and poor governance

(49).”

This sentence is unclear and needs re-writing.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Atnafu Mekonnen Tekleab

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

Submitted filename: Reviewed May 2021 Stunting in Indonesia.docx

Click here for additional data file.

18 Aug 2021

Responses to Reviewers

Article titled “Beyond Personal Factors: Multilevel Determinants of Childhood Stunting in Indonesia”

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Yes

Response: Thanks for your comment.

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

Response: Thanks for your comment.

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

Response: Thanks for your comment.

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: No

Response: Thanks for your comment.

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The statistical tests applied and the technicality of the manuscript are sound. However, the manuscript requires some revision with regard to its structure and language. The manuscript is very lengthy and difficult to read it.

Response: Thanks for your comment. We have streamlined the paper following your suggestion, for example we have shortened the method section. Regarding language, the paper has now been proofread by a professional editor.

Reviewer #2: MAJOR COMMENTS

(1) How much missing data was there on each variable? How was missing data handled in the models?

Response: Thanks for pointing this out. We have revised the manuscript and provide a clearer description of the data in section 2, sub section 2.1. paragraph 2.

“This study only covers young children aged five years old and below. The dataset includes a total of 8,290 young children, with 4,142 in wave 4 and 4,148 in wave 5. By using the child anthropometry information, we excluded children identified as having biologically implausible values (n=185, 2.2%) based on z-scores of height, following the World Health Organization (WHO) child growth standard reducing the number of children to 8105. In addition, we removed young children for whom complete information on household and community characteristics was not available (n=60, 0.7%). Hence, the number of young children used in our final analysis is 8,045. The proportion of respondents with missing values is insignificant, at 3 percent”.

(2) The conclusion to this paper needs re-working. Too much of the conclusion is a repetition of material from earlier in the paper.

A suggested structure for the conclusion would be as follows:

• Brief review of main results

• Strengths of the study

• Limitations of the study, including any unobserved confounders which may affect the results

• Policy implications of the study

• Further work which should be undertaken using this data or in future studies

Response: Thanks for your constructive feedback. We have restructured the conclusion accordingly to include the main results, the strengths of the study, the limitations of the study, policy implications and potential future research.

Brief review of main results:

- The intraclass correlation coefficient (ICC) showed that there is a correlation of the risk of stunting for children living in the same province. The correlation becomes stronger for children living in the same sub-districts. Finally, the strongest correlation of the risk of stunting was found among children living in the same household.

- Moreover, the likelihood ratio tests revealed that stunting in Indonesia varies by the province, sub-district and household level, and analysis of stunting needs to consider variations at all these levels.

- This study conducted a stepwise estimation and consisted of two main estimations. The first estimation employed a complete set of data utilizing two waves of the survey and the second estimation relied on the wave 5 dataset. IFLS wave 4 data does not capture unhealthy snacking behavior among young children. This variable is considered an important proxy for measuring children’s dietary habits, and thus we used wave 5 data to examine this issue.

- Our results show that frequent snacking increases the risk of being stunted. As this analysis shows that the associations between other covariates and stunting are robust, it is possible that if data had been available for the snacking variable in wave 4, the pattern of the association between child dietary habits and stunting might have also been found for that wave.

Strengths of the study

- Thus, this study adds value by taking into account the hierarchical structure of the data and the role of unobserved characteristics at the household, sub-district and provincial level.

- This study also uses a repeated cross-sectional survey providing recent data about young children in Indonesia, which is an improvement on previous studies.

- This study focuses on the last two waves of the IFLS survey: wave 4 collected in 2007 and wave 5 collected in 2014. The data covers 8,045 young children aged five years old living in 21 provinces and 1,332 sub-districts in Indonesia, with 4,001 children in wave 4 and 4,044 children in wave 5.

- Utilizing a multilevel mixed effects modelling takes the nesting in our data into account. Children are nested into households, which in turn are nested into sub-districts and with sub-districts further nested into provinces. Further, there are children living in the same sub-districts are nested into provinces.

- In addition, the multilevel mixed effects model was appropriate for examining the effects of unobserved household, sub-district and provincial characteristics, as well as child level characteristics, on the probability of children being stunted.

- This study highlights the importance of family and community level variables in stunting.

- Our study also demonstrates the urgency of addressing not only personal or individual level factors, but also household and community level factors to reduce stunting prevalence.

- The findings of this study are consistent with the results from previous studies. In terms of individual characteristics, being a boy, having a low neonatal weight and experiencing acute diarrhea are associated with stunting.

- In terms of family characteristics, we found that mothers’ characteristics, specifically maternal stature and maternal education, are associated with stunting. In contrast, living in a family with better socio-economic status lowers the risk of stunting, and the risk of stunting is much lower for children in the highest quartile of the wealth distribution.

- Finally, in terms of community characteristics, we found that living in rural areas increases the risk of stunting by 20 percent. The risk of stunting is also higher for children living in a community with lack of access to clean water, sanitation and hygiene.

Limitations

- Another data limitation associated with the IFLS sample is that the two waves of data used are representative of only about 83% of the Indonesian population, covering 21 provinces; thus, the analysis excludes some areas in eastern Indonesia.

- There are other limitations of the study from a methodological perspective. First, the cross-sectional nature of our analyses limits our ability to infer causation.

- Moreover, there were some potential biases or confounders in this study. Some data is based on self-reported information, for example the birth weight of the children, and thus may be susceptible to information bias.

- Additionally, heterogeneity in the tastes and preferences of parents for health investment for their children was not able to controlled for. The assessment of nutritional services was conducted from the supply side (i.e., the availability of nutritional services in the community). However, we adjusted for several important confounders and have taken into account some unobserved characteristics through multilevel modelling.

Policy implications

- From a policy perspective, our findings suggest that tackling stunting in Indonesia requires substantial effort to create spaces that assist policy implementation in establishing supportive multilevel conditions. These include addressing both individual and household-level factors that support good child nutrition and development. Healthy eating habits, mothers’ education and awareness, socio-economic characteristics and the availability of WASH matter.

- In 2017, the Indonesian government established the National Strategy to Accelerate Stunting Prevention (StraNas Stunting), a four-year strategy to integrate critical services related to stunting across national, regional and community programs (54). The strategy involves 22 ministries covering the areas of health, early childhood education and development, water, sanitation and hygiene (WASH), food security, and social protection. It adopts a multi-sectoral approach. It aims to prevent 2 million children from becoming stunted between 2018 and 2022. However, our research relates to a period before the implementation of this national policy.

Future study:

- Thus, future research should incorporate this policy measure and aim to evaluate the impact of this strategy on decreasing stunting in Indonesia.

- Future research may also expand the investigation of the determinants of stunting to eastern Indonesia, which is a less developed region than the western part of the country. The availability of the IFLS East survey conducted in 2012 will enable a similar analysis to be conducted in this region in the future.

MINOR COMMENTS

There are a number of places in which the grammar could be improved.

TITLE OF PAPER

“Beyond Personal Factor: Multilevel Determinants of Childhood Stunting in Indonesia”

suggest

“Beyond Personal Factors: Multilevel Determinants of Childhood Stunting in Indonesia”

P 2

“Children health outcomes are poor in Indonesia despite the..”

suggest

“Child health outcomes are poor in Indonesia despite the..”

“Regarding to risk factor, Beal et al., (2018) (4), for example, establish…”

suggest

“With regards to risk factors, Beal et al., (2018) (4), for example, establish…”

P 3

“In term of protective factors of stunting, previous studies show…”

suggest

“In terms of protective factors for stunting, previous studies show…”

P 4

“…taking into account the clustering in the data generates a more reliable standard errors of regression coefficients…”

suggest

“…taking into account the clustering in the data generates more reliable standard errors of regression coefficients…”

“The survey has five waves of data: 1993, 1997, 2000, 2007-2008 and 2014-2015 rounds.”

suggest

“The survey has five waves of data, collected in 1993, 1997, 2000, 2007-2008 and 2014-2015.”

P 5

“The first wave of the survey had covered only 13 provinces but the number had been broadened to include…”

suggest

“The first wave of the survey covered only 13 provinces but the number has been broadened to include…”

“Using a-repeated cross-sectional survey…”

suggest

“Using a repeated cross-sectional survey…”

P 6

“The children health status data is a self-reported measure of general health status…”

suggest

“The child health status data is a self-reported measure of general health status…”

P 7

“Following Wang et al., (2018) (17), the dummy variable for unhealthy snacking will take on the value of 1 for children consuming unhealthy snacks for more than 7 times a week…”

suggest

“Following Wang et al., (2018) (17), the dummy variable for unhealthy snacking will take the value 1 for children consuming unhealthy snacks more than 7 times a week…”

“Demographic characteristics of children such as gender is also considered as a predictor…”

suggest

“Demographic characteristics of children such as gender are also considered as a predictor…”

P 8

“This study further constructs the expenditure into quartile data of the bottom 25 percentile (Q1), between 25-50 percentile (Q2), between 50-75 percentile (Q3) and those in the top 25 percentile (Q4).”

suggest

“This study encodes the expenditure data as quartiles, labelled Q1 to Q4, with Q1 being the lowest quartile of expenditure.”

P 12

“…incorporating mother stature, mother education and socio-economic status of the family…”

suggest

“…incorporating mother’s stature, mother’s education and socio-economic status of the family…”

Response: Yes, we have revised accordingly.

p. 12

“After cleaning the data, there are 8105 number of children used in the analysis.”

suggest

“After cleaning the data, there are 8105 children used in the analysis.”

Response: Yes, thanks for your input. We have revised the sentence and provides more detail information regarding the data cleaning process in the note under table 3 as below:

• For model 1-4, we analyzed data from 8105 children, who met the inclusion criteria, and with anthropometric measures within the biologically plausible value.

• For model 5-10, we analyzed data from 8045 children, who met the inclusion criteria, and with anthropometric measures within the biologically plausible value, and with complete individual and household characteristics.

• For Model 11, we analyzed data from 7795 children who met the inclusion criteria, and with anthropometric measures within the biologically plausible value, and with complete individual and household characteristics, and also have information on nutritional services access. However, this model is not used in the following estimation, due to insignificant findings.

P 18

“On average, complementary feeding is started at 19.74 weeks for baby girls and 18.91 weeks for baby boys and the difference is statistically significant using t-test.”

suggest

“On average, complementary feeding is started at 19.74 weeks for baby girls and 18.91 weeks for baby boys; this difference is statistically significant using a t-test.”

TITLE FOR TABLE 3

“Stepwise of Multilevel Mixed Effect Model Logistic Regression”

suggest

“Results of stepwise Multilevel Mixed Effect Logistic Regression Models”

P 22

“Children whose mothers are shorter than 145cm have 19 percent [95% CI (1.05-1.34) in model using wave 4 and 5] more risk to be stunted.”

suggest

“Children whose mothers are shorter than 145cm have a 19 percent [95% CI (1.05-1.34) in model using Waves 4 and 5] greater risk of being stunted.”

“Prendergast & Humphrey (2014) (39) argue that stunted mother is relevant in explaining stunting prevalence considering the importance of the nutritional status of the mother on children stunting.”

suggest

“Prendergast & Humphrey (2014) (39) argue that having a stunted mother is relevant in explaining stunting prevalence in children because of the importance of the nutritional status of the mother on child stunting.”

P 23

“Moreover, maternal years of schooling lowers the risk of being stunted [odds ratio of 0.96 95% CI (0.94-0.97) in model using two waves of 4 and 5].”

suggest

“Moreover, the number of maternal years of schooling lowers the risk of being stunted [odds ratio of 0.96 95% CI (0.94-0.97) in the model using data from Waves 4 and 5].”

“Our estimation results also suggest that children from poor households have a higher risk to be stunted.”

suggest

“Our results also suggest that children from poor households have a higher risk of being stunted.”

“According to Mahendradata et al (2017) (44) and Mulyanto, Kurst and Kringos (2019) (45), both demand and supply of health care are varied across urban and rural areas.”

suggest

“According to Mahendradata et al (2017) (44) and Mulyanto, Kurst and Kringos (2019) (45), both demand and supply of health care vary between urban and rural areas.”

P 24

“People living in the urban areas have more access to health care and other related infrastructure…”

suggest

“People living in urban areas have more access to health care and other related infrastructure…”

“The physical health infrastructure such as working incubators, lab facilities and outpatient polyclinics are more limited in the rural areas.”

suggest

“The physical health infrastructure such as working incubators, lab facilities and outpatient polyclinics are more limited in rural areas.”

“Families in rural areas are more sensitive to a food price increase because they allocate 2/5 of their budget for a staple.”

suggest

“Families in rural areas are more sensitive to food price increases because they allocate two fifths of their budget for staple needs.”

P 25

“In terms of health infrastructure, the proportion of nutritionists per population is low and varied across regions in Indonesia.”

suggest

“In terms of health infrastructure, the proportion of nutritionists per head of population is low and it varied across regions in Indonesia.”

“This finding is also correspondents with national data of Indonesia Socio-Economic Survey and Basic Health Survey that the stunting prevalence was remained high in 2013 compared to the 2007 survey and the figure is even slightly higher for 2013 survey.”

suggest

“This finding is in agreement with national data from the Indonesia Socio-Economic Survey and Basic Health Survey that stunting prevalence was higher in 2013 compared to the 2007 survey.”

Response: Yes we have revised accordingly.

P 26

“Nevertheless, the massive program had been set backs that government had loss attention and in the same time Indonesia undergo decentralization that reduced the effectiveness of nutrition program due to weak management and poor governance (49).”

This sentence is unclear and needs re-writing.

Response: Yes, we have revised it accordingly.

“However, this massive program has experienced setbacks and lost the close attention of government. Furthermore, at the same time Indonesia has undergone decentralization, which has reduced the effectiveness of nutrition programs in improving children’s nutritional status due to weak management and poor governance (53)”.

6. Abstract section: Not structured

Response: Thanks for your comment and we have revised the abstract accordingly, as follows:

Background: Stunting is still a major public health problem, including in Indonesia. Studies have reported the complexities of understanding the determinants associated with stunting. This study aims to examine the household, sub-district and province level determinants of stunting in Indonesia using a multilevel hierarchical mixed effects model.

Methods: We used data from the Indonesian Family and Life Surveys (IFLS) waves 4 and 5 (for the years 2007 and 2014). Data from 8,045 children were analyzed. We included individual, family/household and community level variables in the analyses. A multilevel mixed effects model was employed to take into account the hierarchical structure of the data. Moreover, the model captures the effect of unobserved household, sub-district and province level characteristics on the probability of children being stunted.

Results: Our findings showed that the odds of childhood stunting vary significantly by province, sub-district and household levels. Among the child-level covariates included in our model, dietary habits, neonatal weight, a history of infection, and gender significantly affect the risk of stunting. Household wealth status and parental education are significant household-level covariates associated with a higher risk of stunting. Finally, the risk of stunting is higher for children living in a community without access to water, sanitation and hygiene.

Conclusions: Stunting is associated with not only children’s individual factors, but also family and community level characteristics. Hence, interventions to reduce stunting should also take into account family and community characteristics to achieve effective outcomes.

7. Introduction section: Authors mixed up reference citation styles (Vancouver vs Harvard styles) - see paragraph 1, lines 9-10.

“Therefore, fighting against stunting still remains the main agenda for the government (Indonesia Medium Development Goals 2014-2019 and 2020-2024)”.

Response: Thanks for pointing this out and we have revised it accordingly.

“Therefore, fighting against stunting still remains the main agenda for the government as asserted in the Indonesia Medium Development Goals 2014-2019 and 2020-2024”.

8. Methods section:

• The authors have mentioned that there were 4,309 children in wave 4 and 4,235 children in wave 5 who were less than five years old. However, they also mentioned that there were 5,975 households who had children aged 5 and below. How can the number of households who had children under five be more than the number of children age less than five? Does that mean some of the children did belong to more than one households?

Response: Thanks for pointing this out. We have revised the draft in section 2, sub section 2.1. paragraph three and hopefully now it is clearer. The revision is below.

“Four hierarchical levels are considered in our analysis – individual (child), household, sub-district and province. Children (the lowest level in our mixed effects hierarchical model) are nested within households (level two). There are 6,437 households with 8,045 children aged 5 and below in the two waves of IFLS. Approximately 20 percent of the total households had more than one child aged 5 and below. Households are nested within sub-districts (Kecamatan in Indonesian). There are 1,332 sub-districts recorded in the two waves of the survey.

• The authors need to elaborate the following issues:

o Which WHO curve was used to determine the nutritional status (stunting) of the children? The 2006 or….?

Response: Thanks for your question. This paper refers to the World Health Organization (2005) publication that is used by the Indonesian government in defining children’s nutritional status as stipulated by the Keputusan Menteri Kesehatan Republik Indonesia Nomor: 1995/Menkes/SK/XII/2010 .

o Who determined the nutritional status (stunting), the data collectors at the time of the data collection or the authors of this manuscript (by analyzing the raw data i.e. height-to-age of the child)?

Response: Thanks for your question. We (authors) determined the nutritional status by analyzing the raw data (i.e. height-to-age of the child).

• The community level variable that is “access to nutrition specific services” lacks clarity. The authors categorized the variable based on the accessibility of a child to three nutrition specific services provided by PHC in the area. What does “access” means? Does it necessarily imply the utilization of the mentioned services? How can a child be categorized if those mentioned services are available but child is not utilizing any of the services?

Response: Thanks for pointing this out. The access to nutritional specific services refers to the supply of the services in the neighborhood. Thus, we added this as a limitation of the study in the conclusion, as follows:

“Some data is based on self-reported information, for example the birth weight of the children, and thus may be susceptible to information bias. Additionally, heterogeneity in the tastes and preferences of parents for health investment for their children was not able to controlled for. The assessment of nutritional services was conducted from the supply side (i.e., the availability of nutritional services in the community). However, we adjusted for several important confounders and have taken into account some unobserved characteristics through multilevel modelling”.

• I am not sure if it is necessary to describe the details of the mathematical formulas that were used by the authors to run the different prediction models as it is done by the authors of this manuscript. Because of such lengthy description, it is no attractive for reading. Better only to state the models and why those models are preferred.

Response: Thanks for your comment. We have streamlined the description of the methodology and kept the final empirical model of equation 6.

“The final model in which we estimate determinants of stunting, adjusting for provincial, district, community and family unobserved characteristics, is presented in the equation below:

logit (y_iljk=1| x_ijk,ω_ljk,u_jk,v_k )= β_0+ β_1 X_(1iljk )+ β_2 X_2ljk+ β_3 X_3jk+ β_4 X_4k+ v_0k+u_0jk+ω_0ljk

Where β_1 is the vector of coefficients for predictor variables at level 1 (X_(1iljk )); β_2 is the vector of coefficients for predictor variables at level 2 (X_2ljk); β_3 is the vector of coefficients for predictor variables at level 3 (X_3jk); and β_4 is the vector of coefficients for predictor variables at level 4 (X_4k).

• Section 2.7 describes the authors’ findings. But it is kept under the “Data and Methods” section. Better to move it to the “Result section” as section 2.7 contains the research findings not research method.

Response: Thanks for pointing this out. We have relocated section 2.7 to the results section.

9. Section 2.7. characteristics of study participants:

• Table 1: Standard deviation the variable “stunted” is mentioned as 0.440. How can a categorical variable have a standard deviation?

• Table 1 is confusing and difficult to understand because of the following reasons:

o The purpose of calculating “standard deviation” for categorical variables is not clear.

o Look at the following two observations which are directly taken from the table ( similar problem exists for other variables too):

Gender

Male= 8105

Female= 0

Baby size

Small baby= 8105

Normal baby=0

From the above observations, what readers can understand is that there were no female babies and normal size babies.

Response: Table 1 has been revised accordingly and we hope now this is clearer.

• Table 4: The table cells contain “starred” values. What do those “stars” imply? Needs description as foot note under the table.

Response: Thanks for your question. We have added information below table 4 as below:

odds ratio is statistically significant at either 1 percent (***), 5 percent (**) or 10 percent (*) of the confidence intervals.

• In the methods section the authors mentioned that there were 4,309 children in wave 4 and 4,235 in wave 5 of the survey. However, in the result section they mentioned that they have included 4015 children of wave 4 and 4090 children of wave 5 (Total of 8105 children). The authors need to explain how and why some of the children were excluded from the analysis.

Response: Thanks for pointing this out. We have revised the manuscript and provide a clearer description of the data in section 2, sub section 2.1. paragraph 2.

“This study only covers young children aged five years old and below. The dataset includes a total of 8,290 young children, with 4,142 in wave 4 and 4,148 in wave 5. By using the child anthropometry information, we excluded children identified as having biologically implausible values (n=185, 2.2%) based on z-scores of height, following the World Health Organization (WHO) child growth standard reducing the number of children to 8105. In addition, we removed young children for whom complete information on household and community characteristics was not available (n=60, 0.7%). Hence, the number of young children used in our final analysis is 8,045. The proportion of respondents with missing values is insignificant, at 3 percent”.

• The total number of children included in the analysis is 8105 (the sum of children in wave 4 and wave 5). It is not clear why the authors summed up the number of children from the two waves and treated the two group of populations as single population. The population of children in wave 4 and wave 5 are separate since data collected from the two group of population was collected at different times (2007 and 2014) and should not be summed up and should not be treated as homogenous population.

Response: Thanks for the feedback. Yes, we have added characteristics of study participants in table 1 for the two waves separately (2014 and 2007) and the combined two waves.

• Result and discussion are mixed. I advise the authors to follow the journal’s template.

Response:

Thanks for your comment. Yes, we have divided results and discussion into separate sections.

• Authors need to mention the limitation of their study

Response: Thanks for your comment. Yes, we have added the study’s limitations in the conclusion, as follows.

- Another data limitation associated with the IFLS sample is that the two waves of data used are representative of only about 83% of the Indonesian population, covering 21 provinces; thus, the analysis excludes some areas in eastern Indonesia.

- There are other limitations of the study from a methodological perspective. First, the cross-sectional nature of our analyses limits our ability to infer causation.

- Moreover, there were some potential biases or confounders in this study. Some data is based on self-reported information, for example the birth weight of the children, and thus may be susceptible to information bias.

- Additionally, heterogeneity in the tastes and preferences of parents for health investment for their children was not able to controlled for. The assessment of nutritional services was conducted from the supply side (i.e., the availability of nutritional services in the community). However, we adjusted for several important confounders and have taken into account some unobserved characteristics through multilevel modelling.

10. Conclusion:

• Under this section, the authors tried to summarize their manuscript rather than putting the main finding of their research. Better to revise this section in such a way.

Response: Thanks for your comment. The conclusion has been revised accordingly based on the below structure for the conclusion as follows:

• Brief review of main results

• Strengths of the study

• Limitations of the study, including any unobserved confounders which may affect the results

• Policy implications of the study

• Further work which should be undertaken using this data or in future studies.

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

20 Sep 2021

PONE-D-21-06348R1Beyond Personal Factor: Multilevel Determinants of Childhood Stunting in IndonesiaPLOS ONE

Dear Dr. Mulyaningsih,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Great effort was made by the authors to utilize the feedback that was provided for them to correct. I find it interesting and improved with respect to the original submission. However, there are still major things to adjust in addition to the enclosed reviewers’ comments.  Despite the known influence of the dietary behaviours and habits on growth and the effect of the national school feeding programs on improving stunting among school children, yet the introduction and discussion lack a lot of references concerning national figures for stunting and its determinants among children in low- and middle-income countries with similar context. Commenting on the impact of the national school Feeding programs on growth and subsequently on development and school achievement as a solution and remedy way for improving stunting in the discussion section (which is vital) is still missing. Please elaborate on this in structural relationships within the discussion, and recommendations.

Please consider reviewers’ comments for more details

Please submit your revised manuscript by Nov 04 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Ammal Mokhtar Metwally, Ph.D (MD)

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #3: (No Response)

Reviewer #4: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #3: Yes

Reviewer #4: Partly

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #3: Yes

Reviewer #4: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #3: Yes

Reviewer #4: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #3: Yes

Reviewer #4: No

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #3: All comments are addressed properly. The manuscript is technically sound, statistical analysis has been performed appropriately and rigorously and manuscript is presented in an intelligible fashion but need some grammatical corrections.

Reviewer #4: 1. The manuscript is unnecessarily long such that the core issues are masked

2. The methodology section need not include detailed formulae used in deriving the results. This section needs revising to make it shorter and succinct

3. The results tables are too crowded. Summarised tables can be given in the results and more elaborate tables provided as addendum, if necessary

4. The conclusions read like another discussion of results. This section needs to be focussed on answering the objectives of the study

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #3: No

Reviewer #4: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

19 Oct 2021

Response to specific reviewer and editor comments.

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #3: (No Response)

Reviewer #4: (No Response)

Response: We have revised the manuscript according to editor and reviewers’ inputs as described below.

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #3: Yes

Reviewer #4: Partly

Response: Thanks for your comment. We have revised the conclusion as per reviewer #4 suggestion in point 6 below.

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #3: Yes

Reviewer #4: Yes

Response: Thanks for your comment.

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #3: Yes

Reviewer #4: Yes

Response: Thanks for your comment.

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #3: Yes

Reviewer #4: No

Response: The manuscript has been proof read by American Journal Experts to assure that the manuscript is presented in an intelligible fashion, clear, correct, unambiguous and have no grammatical errors.

6. Review Comments to the Author

Reviewer #3: All comments are addressed properly. The manuscript is technically sound, statistical analysis has been performed appropriately and rigorously and manuscript is presented in an intelligible fashion but need some grammatical corrections.

Response: The manuscript has been proof read by American Journal Experts to assure that the manuscript is presented in an intelligible fashion, clear, correct, unambiguous and have no grammatical errors.

Reviewer #4:

1. The manuscript is unnecessarily long such that the core issues are masked

2. The methodology section need not include detailed formulae used in deriving the results. This section needs revising to make it shorter and succinct

Response: Regarding reviewer#4 input number 1 and 2, we have removed the empirical model in the methodology section and make the section shorter. Hopefully, the revised manuscript is more concise.

3. The results tables are too crowded. Summarised tables can be given in the results and more elaborate tables provided as addendum, if necessary

Response: Thanks for the suggestion. We relocated table 3 of stepwise results into appendix.

4. The conclusions read like another discussion of results. This section needs to be focussed on answering the objectives of the study.

Response: Thanks for the suggestion. We have revised the conclusion so it is more focus on answering the objectives of the study. Moreover, we relocated the study limitation from conclusion into discussion. We added a subsection at the end of the discussion section to explain the study limitation.

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #3: No

Reviewer #4: No

Response: Yes, noted.

8. Despite the known influence of the dietary behaviours and habits on growth and the effect of the national school feeding programs on improving stunting among school children, yet the introduction and discussion lack a lot of references concerning national figures for stunting and its determinants among children in low- and middle-income countries with similar context. Commenting on the impact of the national school Feeding programs on growth and subsequently on development and school achievement as a solution and remedy way for improving stunting in the discussion section (which is vital) is still missing. Please elaborate on this in structural relationships within the discussion, and recommendations.

Response:

Thanks for the suggestion.

- We have added a paragraph in the introduction presenting data concerning figures for stunting among children in low- and middle-income countries with similar context.

- We have added citation of articles discussing the effect of the national school feeding programs on improving nutritional status among school children in the discussion section to explain the role of dietary habits on children nutritional status.

- There are three additional literatures that we have added to explain the importance of dietary habits in lowering risk of nutritional issues such as stunting as below.

1. Metwally AM, El-Sonbaty MM, el Etreby LA, Salah El-Din EM, Abdel Hamid N, Hussien HA, et al. Impact of National Egyptian school feeding program on growth, development, and school achievement of school children. World Journal of Pediatrics. 2020 Aug 13;16(4).

2. Murphy SP, Gewa C, Liang L-J, Grillenberger M, Bwibo NO, Neumann CG. School Snacks Containing Animal Source Foods Improve Dietary Quality for Children in Rural Kenya. The Journal of Nutrition. 2003 Nov 1;133(11).

3. Baum JI, Miller JD, Gaines BL. The effect of egg supplementation on growth parameters in children participating in a school feeding program in rural Uganda: a pilot study. Food & Nutrition Research. 2017 Jan 6;61(1).

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

8 Nov 2021

Beyond Personal Factor: Multilevel Determinants of Childhood Stunting in Indonesia

PONE-D-21-06348R2

Dear Dr. Mulyaningsih,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Ammal Mokhtar Metwally, Ph.D (MD)

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

A great effort was made by the authors to utilize the feedback that was provided for them to correct. I find it interesting and improved with respect to the original submission.

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #3: All comments have been addressed

Reviewer #4: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #3: Yes

Reviewer #4: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #3: Yes

Reviewer #4: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #3: Yes

Reviewer #4: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #3: Yes

Reviewer #4: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #3: The manuscript is written in an intelligible fashion and statistical analysis are performed appropriately and rigorously.

Reviewer #4: (No Response)

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #3: No

Reviewer #4: No

11 Nov 2021

PONE-D-21-06348R2

Beyond Personal Factors: Multilevel Determinants of Childhood Stunting in Indonesia

Dear Dr. Mulyaningsih:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Professor Ammal Mokhtar Metwally

Academic Editor

PLOS ONE