Ethnoracial child health inequalities in Latin America: Multilevel evidence from Bolivia, Colombia, Guatemala, and Peru

Abstract

Using Demographic and Health Survey (DHS) data for Bolivia, Colombia, Guatemala, and Peru, between 1986 and 2015, this paper explores the relationship between self-identifying as indigenous and/or afro-descendant on child under-5 mortality (n=20,770), stunting (n=15,828), wasting (n=15,827), and anemia (n=13,294). Rural-urban risk analysis suggest that indigenous and/or afro-descendent respondents have higher risk of under-5 mortality, stunting, wasting, and anemia. The same pattern is observed for cross-country risks, particularly for Bolivia and Colombia. Results from logistic multilevel regression models suggest that, even after controlling for geographic, socioeconomic, individual, reproductive, healthcare, and nutritional variables, self-identifying as indigenous and/or afro-descendant is associated with a higher risk of child stunting and wasting, but not necessarily a higher risk of under-5 mortality and anemia. While previous research has largely focused on the protective role of maternal education, results from this study suggest that paternal education, as well as, individual characteristics and early reproductive decisions, play a significant role in child health outcomes. My findings imply that efforts to improve child health in Latin America should account for ethnicity and/or race, since minority ethnoracial groups have higher risk of childhood morbidity in the region. In addition, these efforts should accompany education for both men and women, as well as, information about the effects of reproductive decisions on their children's health.

Highlights

• •Multilevel logit models for under-5 mortality, stunting, wasting, and anemia.
• •Tetrachoric factor analysis to construct socioeconomic and healthcare indices.
• •Indigenous and/or afro-descendent have higher morbidity risk in Latin America.
• •Ethnicity and/or race significantly associated with stunting and wasting.
• •Paternal education, individual, and reproductive characteristics significant.

1. Introduction

Throughout history, the region of Latin America has been characterized as one of the most ethnically and racially heterogeneous regions of the world. With an estimated population of 652 million people, most descend from three major ethnoracial groups: indigenous peoples (40 million), direct descendants of peoples inhabiting this region when European colonizers arrived in the fifteenth century; afro-descendent (120 million), direct descendants of Africans slaves forcibly brought to the region during and after the colonial period; and Europeans, direct descendants of largely Spanish and Portuguese immigrants (Perreira & Telles, 2014; Ribando, 2005, p. 25). Drawing from complex colonial and nation-building histories from the 15th century, Latin America has experienced substantial variation in the trajectories of ethnoracial groups, which has defined the region's demographic composition, representations of identity, assimilation processes, and changing definitions of ethnoracial classifications (Telles & Bailey, 2013; Telles & Torche, 2019). Unlike ethnicity and race elsewhere, ethnoracial classifications in Latin America have been substantially fluid, resulting particularly from the historical nation-building efforts to unite, divided black, indigenous, white, and mixed-race populations through mestizaje, or racial and cultural mixing ideologies (Telles & Bailey, 2013).

However, indigenous and/or afro-descendent groups in Latin America are extremely diverse, characterized by variety of cultures, identities, languages, traditions, faiths, and beliefs. The total number of indigenous groups is estimated between 655 and 826 (Davis-Castro, 2020, p. 34) and afro-descendent groups, while less fragmented, include black (negro/preto), mixed-black (mulatto), mixed-indigenous-black (zambo/chino/garifuna), and mixed-indigenous-black-white (pardo) groups (Telles et al., 2015). Despite this diversity, both indigenous and/or afro-descendent groups have historically been placed similarly at the bottom of the uneven class structure and racial and ethnic discrimination and exclusion continue to significantly determine their livelihoods. Indigenous and afro-descendent people suffer similar problems of economic, social, cultural and political inequality, compared to non-indigenous and/or non-afro-descendent groups, which reproduces and perpetuates socioeconomic, educational, health, and political inequities (Bello & Rangel, 2002). Despite this, little is known about ethnicity and/or race, and child health outcomes in this region, particularly, in terms of the variation across and within countries. In Latin America, scarcity in research on ethnoracial health disparities is explained by long-held beliefs that socio-economic status, rather than ethnoracial differences, structure inequality (Telles, 2006).

Theories on the social determinants of health have argued that the social status of ethnicity and race, as a “social rather than genetic” entity, contributes to disparities in risk exposure, access to resources, and health outcomes (Zuberi, 2001). Through underlying social and demographic processes, ethnicity and race contribute to differences in health that disadvantage ethnoracial minorities. Literature on the United States has documented persistent and pronounced health disparities between and within ethnoracial groups, with these groups experiencing earlier mortality, higher morbidity, and worse overall health (Vega & Rumbaut, 1991; Williams & Sternthal, 2010). The limited empirical research that does exist for Latin America has documented that indigenous and/or afro-descendent groups fare worse in terms of mortality and morbidity compared to non-indigenous and/or non-afro-descendent groups (Casas et al., 2001). For indigenous groups, infant mortality is 3.5 times higher in Panama (Flores & Mojica, 1992), life expectancy is 29 years lower for men and 27 years lower for women in Honduras (Rivas, 1993), child mortality is more than 2.5 times higher in Mexico, maternal mortality is 83% higher in Guatemala (Instituto Nacional Indigenista, 1997), and morbidity is two times higher in Bolivia (Pan American Health Organization, 1999).

Latin America is a good empirical case to study these relationships because countries in this region share close geographic proximity, as well as centuries of ethnolinguistic, geopolitical, and historically communal legacies (Beals, 1953; Inglehart & Carballo, 1997). Also, across countries, the historical configurations of boundaries of identity through national mestizaje projects, as well as the historical institutionalization of inequality through phenotypic markers of color-, culture-, and linguistics-coded ethnicity and/or race are quite similar (Telles & Bailey, 2013). This allows for fairer comparisons of health inequalities among and between ethnoracial minority groups. Building on this research gap, I use Demographic and Health Survey (DHS) data for Bolivia, Colombia, Guatemala, and Peru, between 1986 and 2015, to explore the relationship between ethnicity and/or race and under-5 mortality, stunting, wasting, and anemia among children. First, I describe relative risks by ethnicity and/or race and across urban-rural regions. Second, I conduct logistic multilevel regression models to evaluate the association of ethnicity and/or race and child health outcomes. Finally, I demonstrate the extent to which certain proximate factors—geographic, socioeconomic, individual, reproductive, healthcare, and nutrition—may moderate the association.

2. Data and methods

2.1. Data

This analysis uses pooled cross-sectional DHS data for Bolivia, Colombia, Guatemala, and Peru. DHS is a publicly-available, nationally-representative survey of women, collected by ICF International in collaboration with host country governments. The standardized DHS questionnaires, across countries and waves, allow for easy comparisons for a wide range of indicators in the areas of population, health, and nutrition. DHS uses a stratified cluster-sampling design to randomly select women ages 15–49 within households and clusters (Croft et al., 2018, p. 168). To account for homogeneity due to the non-simple random sample (i.e., nonindependence) and under- or over-sampling of different strata during sample selection (i.e., unequal selection probabilities), I adjust for sample cases with sampling weights (Hahs-Vaughn et al., 2011). As a result, I can confidently estimate standard errors and unbiased parameter estimates, as well as, present population-based estimates that account for differential probability of selection into the survey.

Data for this analysis comes from all available survey waves for: Bolivia (1989, 1994, 1998, 2003, 2008), Colombia (1986, 1990, 1995, 2000, 2005, 2010, 2015), Guatemala (1987, 1995, 1998–1999, 2014–2015), and Peru (1986, 1991–1992, 1996, 2000, 2004–2006, 2007–2008, 2009, 2010, 2011, 2012). I include these four countries because they have substantial ethnic and/or racial minority populations (Montenegro & Stephens, 2006), particularly in this dataset (i.e. Bolivia: 64.53%; Colombia: 86.61%; Guatemala: 57.15%; Peru: 17.01%). For some outcomes, not all women have available data (stunting and wasting not available for Bolivia and anemia not available for Colombia), so each outcome is different. My total samples are: under-5 mortality (n (level-1)=20,770; n (level-2)=3953), stunting (n (level-1)=15,828; n (level-2)=3372), wasting (n (level-1)=15,827; n (level-2)=3372), and anemia (n (level-1)=13,294; n (level-2)=2474).

2.2. Measurements

2.2.1. Outcome variables

The child health outcomes of interest are under-5 mortality, stunting, wasting, and anemia. Under-5 mortality indicates whether or not a woman has ever had a child die between the ages of 0–60 months. Stunting indicates whether or not a child's height-for-age Z-scores (HAZ) fall more than two standard deviations below the median height-for-age curve (World Health Organization, 2006). Wasting indicates whether or not a child's weight-for-height Z-scores (WHZ) fall more than two standard deviations below the median weight-for-height curve (World Health Organization, 2006). Finally, anemia (collected using HemoCue portable hemoglobin meters) indicates whether or not a child's blood hemoglobin level is less than 11 g per deciliter (g/dl) (World Health Organization, 2011). I constructed these dichotomous outcomes respecting clinical thresholds and using multiple variables that were originally continuous and/or categorical in the surveys. Dichotomization has been identified as optimal for a variable's strongest effects and simplifying the presentation of results for a wider audience (Farrington & Loeber, 2000).

2.2.2. Independent variables

Ethnicity and/or race is measured as a dichotomous variable, indicating whether or not a mother is indigenous and/or afro-descendant. I used the language spoken at home as proxy for indigenous and/or afro-descendant self-identification (Afro-descendant, Aymara, Quechua, Guarani, Garifuna, Maya, Xinca), which has been the primary marker of ethnoracial identity used in the past (Telles & Torche, 2019). While other research ideally recommends using multiple self-identified measures, interviewer-ascribed phenotypic classifications, and multiple sub-categories of race and ethnicity (Perreira & Telles, 2014; Telles et al., 2015), DHS data does not collect measures of race and ethnicity to create multiple sub-categories with sufficient statistical power for this analysis, so I followed previous precedent for dichotomization (Psacharopoulos & Patrinos, 1994, pp. 32–4566).

I controlled for several other factors that potentially confound my analyses. Maternal education is the single most important factor in explaining differentials in child health outcomes (Caldwell, 1979; Young et al., 1983). I include it in an interval scale (0, 1–3, 4–6, 7–9, ≥10 years), but also conducted initial analyses with both categorical or continuous measures, which do not change the direction or the significance of the associations (available from the author). To control for differences in temporal, living, and environmental conditions, I include a categorical variable for survey year (1986–2015), a dummy variable for type of residence (rural/urban), and a categorical variable for country (Bolivia, Colombia, Guatemala, Peru). To control for individual, partner, and household characteristics, I include a categorical variable for household wealth (poorest, poorer, middle, richer, and richest) and continuous variables for partner's education (0−23) and maternal age (13–49).1 Finally, to control for reproductive behavior, I control for maternal age at-first-birth (15–19, 20–34, 35+), birth parity (first, second or third, fourth or higher children), and birth interval (>2, 2–4, 4+).

I constructed socioeconomic and healthcare indices to assess how they moderate the relationship between ethnicity and/or race and child health outcomes. First, I selected variables that seemed to measure the underlying construct. All variables were coded as dichotomous (yes/no) and ranked by ascending order. Then, I performed tetrachoric factor analysis—the preferred method to describe variability for dichotomous measures—to determine how well each set of variables factored together, omitting obvious outliers. The household environment index measures the presence of consumer durables in a household (radio, television, telephone, refrigerator, bicycle, motorcycle, and car), as well as, overall living conditions (electricity and non-dirt floor). The prenatal care index includes receiving any prenatal care, prenatal care from a skilled professional, first prenatal care visit within 6 months, and 4+ prenatal care visits during pregnancy. The postnatal care mother index includes receiving any postnatal care from a skilled professional, postnatal care within 24 hours of delivery, and postnatal check within 2 days of delivery. Finally, the postnatal care child index includes receiving any postnatal care from a skilled professional, postnatal care within 24 hours of birth, and postnatal check within 2 days of birth. Table 1 shows sample proportions, factor loadings, and Cronbach's alphas (α) for the indices. Internal reliability of the four measures is above the α ≥ 0.70 threshold used in the social sciences (Nunnally & Bernstein, 1994).

Table 1.

Description of variables included in indices, by outcome (Source: author's calculations of Demographic and Health Surveys data for Bolivia, Colombia, Guatemala, and Peru, 1986–2015).

Measure	Sample proportion	Factor loading	α
Household environment index
Own a radio	0.780	0.345	0.7310
Own a television	6.782	0.927
Own a telephone	0.165	0.784
Own a refrigerator	0.372	0.902
Own a bicycle	0.290	0.300
Own a motorcycle	0.117	0.486
Own a car	0.082	0.589
Has electricity	0.748	0.899
Has nondirt floor	0.635	0.756
Prenatal care index
Received any prenatal care	0.150	0.8354	0.8873
Received prenatal care from a skilled provider	0.403	0.2035
Received prenatal care in the first 6 months of pregnancy	0.939	0.6522
Received four or more prenatal care visits	0.733	0.9959
Postnatal care mother index
Received postnatal care within 24 hours	0.049	0.9832	0.9175
Received postnatal care within 2 days	0.078	1.0000
Received postnatal care from a doctor or nurse	0.066	0.9832
Postnatal care child index
Received postnatal care within 24 hours	0.180	0.9987	0.9795
Received postnatal care within 2 days	0.201	1.0000
Received postnatal care from a doctor or nurse	0.189	0.9987

Finally, to control for child nutrition, I constructed three main dichotomous feeding indicators for infants and young children: Minimum Dietary Diversity (MDD), Minimum Meal Frequency (MMF), and Minimum Acceptable Diet (MAD) (World Health Organization, 2008). MDD measures whether a child is fed from 4 or more food groups (grains, roots and tubers, legumes and nuts, dairy products, flesh foods, eggs, vitamin-A rich fruits and vegetables, and other fruits and vegetables). MMF measures whether a child is fed solid, semi-solid, or soft foods (including milk feeds for non-breastfed children) the minimum number of times or more (2 times for breastfed infants 6–8 months, 3 times for breastfed children 9–23 months, 4 times for non-breastfed children 6–23 months). Finally, MAD measures whether a child receives a minimum acceptable diet (at least 2 milk feedings, MDD, and MMF).

2.3. Analysis

I use a two-level multilevel logit approach, whereby individual women units (level-1) are nested within survey cluster units (level-2), respecting the hierarchical design of DHS data.2 My multilevel logit models include a random intercept at the cluster-level—to capture heterogeneity among clusters—and fixed effects for all other individual-level coefficients. Compared with single-level regression analysis that assumes that all individuals are independent, this methodology accounts for the fact that individuals in the same cluster may have similar characteristics. More technically, multilevel models correct for biases in parameter estimates, standard errors, confidence intervals, and significance tests, resulting from clustering, and estimate robust variance and covariance of random effects (Guo & Zhao, 2000). I chose a logit approach because my dependent variables are dichotomous:

$$\log \left[\frac{{\mathbf{P}}_{\mathbf{i}\mathbf{j}}}{1-{\mathbf{P}}_{\mathbf{i}\mathbf{j}}}\right]={\mathbf{\beta }}_0+{\mathbf{\beta }}_1{\mathbf{X}}_{\mathbf{i}\mathbf{j}}+\cdots +{\mathbf{\beta }}_{\mathbf{k}}{\mathbf{X}}_{\mathbf{k}}+{\mathit{u}}_{\mathit{j}}+{\mathbf{e}}_{\mathbf{i}\mathbf{j}},$$ (1)

where i are the level-1 and j the level-2 units; ${\mathbf{P}}_{\mathbf{i}\mathbf{j}}/(1-{\mathbf{P}}_{\mathbf{i}\mathbf{j}})$ is the probability of the binary child health outcome ${\mathbf{Y}}_{\mathbf{i}\mathbf{j}}$ under-5 mortality, stunting, wasting, anemia for woman i in cluster j; I define the probability of the response equal to one as ${\mathbf{P}}_{\mathbf{i}\mathbf{j}}=\mathbf{P}\mathbf{r}({\mathbf{Y}}_{\mathbf{i}\mathbf{j}}=1)$ and let ${\mathbf{P}}_{\mathbf{i}\mathbf{j}}$ be modeled using a logit link function; $\mathbf{\beta }$ is the corresponding fixed coefficient and ${\mathbf{X}}_{\mathbf{i}\mathbf{j}}$ is an explanatory variable for woman i in cluster j; ${\mathbf{u}}_{\mathbf{j}}$is the random effect at cluster j, allowing for differential intercepts for cluster-level observations; and the error term, ${\mathbf{e}}_{\mathbf{i}\mathbf{j}}$, is the individual-level residual for individual i of cluster j. Thus, this equation expresses the log of the odds of child mortality, stunting, wasting, and anemia, as a linear function of the set of explanatory variables previously mentioned.

The multilevel logistic models were estimated in two stages. First, I estimated a baseline model with ethnicity and/or race to observe the association of this factor with the risk of each outcome, in the absence of other associations. In this baseline model, the ethnicity and/or race coefficient served as a basis of comparison to measure whether the introduction of other factors, in subsequent models, moderated the ethnoracial effect. I assessed this in two manners: first, how the magnitude of the ethnoracial coefficient changed with the introduction of other factors, and second, how the statistical significance of the ethnoracial coefficient changed as well. Second, I estimated subsequent models by adding geographic, socioeconomic, individual, reproductive, healthcare, and nutritional controls, to see how the effect of ethnicity and/or race is moderated, until a full model was assessed with all variables and controls included.

3.0. Results

3.0.1. Sample characteristics

Table 2 provides descriptive statistics. Approximately 13% of children died before 5, 27% of children are stunted, 6% of children are wasted, and 59% of children have anemia. On average, 45% of women live in rural areas and 56% in urban areas. In addition, 35% of women self-identify as indigenous and/or afro-descendant and 66% self-identify as non-indigenous and/or non-afro-descendant. Maternal education is still low, with approximately 8% of mothers reporting zero years of education, 15% 1–3 years, 27% 4–6 years, 14% 7–9 years, and 37% 10 and more years. Pertaining to other control variables, on average husbands’ report 8 years of education and respondents are 30 years old. Approximately, 54% of women gave birth between ages 15–19, 46% between 20 and 34, and 0.42% at 35 and above. In addition, 66% of women bore their second or third child and 34% their fourth or more child. Finally, 14% waited less than two years between births, 37% two and four years, and 49% more than four years.

Table 2.

Descriptive statistics: proportions and means of key variables, by outcome (Source: author's calculations of Demographic and Health Surveys data for Bolivia, Colombia, Guatemala, and Peru, 1986–2015).

Measure		Under-5 Mortality	Stunting	Wasting	Anemia	Average
Total sample (unweighted N)		20,770	15,828	15,827	13,294
		12.63	26.69	6.44	59.33
Main independent variables
Ethnicity and/or race
	Indigenous and/or afro-descendant	38.08	37.37	37.37	25.03	34.46
	Non-indigenous and/or non-afro-descendant	61.92	62.63	62.63	74.97	65.54
Country
	Bolivia	27.23	.	.	67.90	23.78
	Colombia	5.94	12.37	3.57	.	5.47
	Guatemala	12.05	43.60	11.93	47.76	28.84
	Peru	12.12	25.86	5.56	62.65	26.55
Type of residence
	Rural	43.63	43.71	43.71	46.95	44.50
	Urban	56.37	56.29	56.29	53.05	55.50
Individual and socioeconomic variables
Years of education
	0 years	7.31	7.26	7.26	8.47	7.58
	1–3 years	14.96	14.36	14.36	15.62	14.83
	4–6 years	26.29	26.34	26.34	26.91	26.47
	7–9 years	14.29	14.57	14.57	13.90	14.33
	10+ years	37.14	37.46	37.46	35.10	36.79
Household wealth
	Poorest	25.40	25.60	25.60	24.64	25.31
	Poorer	24.85	25.42	25.42	25.69	25.35
	Middle	21.45	21.28	21.28	21.48	21.37
	Richer	17.04	16.59	16.59	16.90	16.78
	Richest	11.27	11.12	11.12	11.29	11.20
Household environment index		0.59	0.62	0.62	0.56	0.60
Husband's education		8.57	8.51	8.51	8.06	8.41
Respondent's age		30.08	30.03	30.03	30.30	30.11
Reproductive variables
Age at first birth
	15–19	53.38	54.13	54.13	53.68	53.83
	20–34	46.15	45.44	45.44	45.96	45.75
	35+	0.47	0.42	0.42	0.36	0.42
Birth parity
	Second or third	65.22	67.00	67.00	63.58	65.70
	Fourth or higher	34.78	33.00	33.00	36.42	34.30
Birth interval
	>2 years	14.15	13.36	13.36	13.44	13.58
	2–4 years	37.82	36.31	36.31	38.06	37.13
	4+ years	48.02	50.32	50.32	48.50	49.29
Healthcare variables
Prenatal care index		1.41	1.42	1.42	1.42	1.42
Postnatal care mother index		0.10	0.12	0.12	0.15	0.12
Postnatal care child index		0.41	0.48	0.48	0.56	0.48
Nutritional variables
Minimum dietary diversity (MDD)		54.20	61.32	61.31	58.61	58.86
Minimum meal frequency (MMF)		21.58	23.16	23.16	21.94	22.46
Minimum acceptable diet (MAD)		4.12	4.76	4.76	4.79	4.61

3.0.2. Ethnic and racial disparities by region and country

Table 3 presents rural-urban proportions and absolute differences and Fig. 1 also presents relative risks of under-5 mortality, stunting, wasting, and anemia, by indigenous and/or afro-descendent self-identification. Respondents who self-identify as indigenous and/or afro-descendant, and who live in rural areas compared to urban areas, have 2.27-times higher risk of under-5 mortality, 2.83-times higher risk of stunting, 2.20-times higher risk of wasting, and 3.63-times higher risk of anemia. While the risk is also higher for non-indigenous and/or non-afro-descendant respondents in rural areas, it is much lower than that of indigenous and/or afro-descendant respondents. The same analysis was conducted across countries, which is available as supplementary electronic results (Table 8S and Figs. 2S–5S).

Table 3.

Minority-majority ethnic and/or racial relative risk of under-5 mortality, stunting, wasting, and anemia, by type of residence (Source: author's calculations of Demographic and Health Surveys data for Bolivia, Colombia, Guatemala, and Peru, 1986–2015; under-5 mortality N = 20,770, stunting N = 15,828; wasting N = 15,827, and anemia N = 13,294).

Ethnicity and/or race	Minority ethnic and/or racial group				Majority ethnic and/or racial group
	Proportion outcome		Absolute difference	Relative risk	Proportion outcome		Absolute difference	Relative risk
	Urban	Rural	Rural-urban	Minority rural/urban	Urban	Rural	Rural-urban	Majority rural/urban
Measure	(1)	(2)	(2)–(1)	(2)/(1)	(3)	(4)	(4)–(3)	(4)/(3)
Under-5 mortality	12.16	27.65	15.49	2.27	25.69	34.50	8.81	1.34
Stunting	9.68	27.43	17.75	2.83	25.03	37.86	12.83	1.51
Wasting	13.68	30.12	16.44	2.20	20.29	35.90	15.61	1.77
Anemia	5.65	20.51	14.86	3.63	41.33	32.51	−8.82	0.79

Note: For the purpose of simplification, minority group is defined as indigenous and/or afro-descendant and majority group as non-indigenous and/or non-afro-descendant.

Fig. 1.

Minority-majority ethnic and/or racial relative risk of under-5 mortality, stunting, wasting, and anemia, by type of residence (Source: author's calculations of Demographic and Health Surveys data for Bolivia, Colombia, Guatemala, and Peru, 1986–2015; under-5 mortality N = 20,770, stunting N = 15,828; wasting N = 15,827, and anemia N = 13,294).

3.1. Individual and household results

3.1.1. Under-5 mortality

Table 4 presents the odds ratio results of the multilevel logit model predicting under-5 mortality among children in Bolivia, Colombia, Guatemala, and Peru, between 1986 and 2015 (n (level-1)=20,770; n (level-2)=3953). Model 1 includes only the underlying factor of interest—ethnicity and/or race—and temporal and geographic controls. Self-identifying as indigenous and/or afro-descendant is associated with increasing risk of under-5 mortality, but the association is not significant throughout the models (p-value≥0.050). Conversely, living in a rural area, compared with an urban area, is associated with 117-percent greater odds (1-exponent of the log odds) of under-5 mortality (p-value≤0.000), but this association loses significance across the models. Results also indicate heterogeneity in the association between ethnicity and/or race and under-5 mortality across countries. Living in Colombia, Guatemala, or Peru (compared to Bolivia), is more protective (more risk-reducing) for under-5 mortality than living in Bolivia.

Table 4.

Results of multilevel logit model for the odds of under-5 mortality in Latin America (Source: author's calculations of Demographic and Health Surveys data for Bolivia, Colombia, Guatemala, and Peru, 1986–2015; N (level-1) = 20,770; N (level-2) = 3953).

Variables	Model 1					Model 2					Model 3					Model 4					Model 5
	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient		95% C·I.	95% C·I.	S.D.
			Low	High				Low	High				Low	High				Low	High				Low	High
Year	0.942	***	0.917	0.968	0.013	0.932	***	0.907	0.959	0.013	0.941	***	0.914	0.968	0.014	0.944	***	0.917	0.972	0.014	0.945	***	0.917	0.973	0.014
Ethnicity and/orrace
Indigenous and/or afro-descendant (ref. = not)	1.175		0.903	1.528	0.158	1.149		0.887	1.487	0.151	1.069		0.821	1.391	0.144	1.061		0.815	1.382	0.143	1.060		0.814	1.381	0.143
Geographic factors
Rural (ref. = urban)	2.170	***	1.831	2.573	0.189	1.042		0.849	1.280	0.109	1.053		0.847	1.308	0.117	1.045		0.842	1.297	0.115	1.043		0.841	1.295	0.115
Colombia (ref. = Bolivia)	0.237	***	0.177	0.318	0.035	0.273	***	0.201	0.370	0.042	0.341	***	0.247	0.469	0.056	0.333	***	0.241	0.460	0.055	0.338	***	0.242	0.472	0.058
Guatemala (ref. = Bolivia)	0.543	***	0.380	0.776	0.099	0.482	***	0.333	0.696	0.091	0.602	**	0.408	0.887	0.119	0.588	**	0.389	0.889	0.124	0.591	**	0.387	0.902	0.127
Peru (ref. = Bolivia)	0.479	***	0.380	0.604	0.057	0.574	***	0.455	0.725	0.068	0.535	***	0.419	0.683	0.067	0.549	***	0.421	0.715	0.074	0.557	***	0.424	0.731	0.077
Ethnicity x type of residence (ref.=not ethnic x urban)
Not ethnic x rural	2.170	***	1.831	2.573	0.189	1.042		0.849	1.280	0.109	1.053		0.847	1.308	0.117	1.045		0.842	1.297	0.115	1.043		0.841	1.295	0.115
Ethnic x urban	1.175		0.903	1.528	0.158	1.149		0.887	1.487	0.151	1.069		0.821	1.391	0.144	1.061		0.815	1.382	0.143	1.060		0.814	1.381	0.143
Ethnic x rural	2.552	***	2.107	3.091	0.249	1.205		0.972	1.492	0.132	1.184		0.946	1.483	0.136	1.179		0.942	1.477	0.135	1.175		0.938	1.470	0.135
Socioeconomic factors
Mother's education (ref. = 10+ years)
0 years						3.793	***	2.865	5.023	0.543	1.467	*	1.069	2.012	0.237	1.466	*	1.069	2.012	0.237	1.457	*	1.062	1.999	0.235
1–3 years						2.852	***	2.229	3.648	0.358	1.417	**	1.084	1.853	0.194	1.414	**	1.082	1.849	0.193	1.407	**	1.076	1.839	0.192
4–6 years						2.060	***	1.665	2.548	0.224	1.304	*	1.033	1.645	0.155	1.301	*	1.031	1.641	0.154	1.298	*	1.029	1.638	0.154
7–9 years						1.774	***	1.387	2.268	0.223	1.436	**	1.086	1.899	0.205	1.436	**	1.086	1.898	0.204	1.434	**	1.086	1.894	0.204
Household wealth index						0.839	***	0.761	0.925	0.042	0.904		0.814	1.004	0.048	0.905		0.814	1.005	0.049	0.907		0.816	1.007	0.049
Household environment index						1.168		0.847	1.612	0.192	0.968		0.702	1.334	0.159	0.975		0.705	1.347	0.161	0.977		0.707	1.350	0.161
Husband's education						0.953	***	0.934	0.972	0.009	0.981		0.961	1.001	0.010	0.981		0.961	1.001	0.010	0.981		0.961	1.002	0.010
Individual and reproductive factors
Respondent's age											1.054	***	1.038	1.071	0.008	1.054	***	1.038	1.071	0.008	1.054	***	1.038	1.071	0.008
Age at first birth											0.607	***	0.519	0.709	0.048	0.607	***	0.519	0.709	0.048	0.607	***	0.519	0.710	0.048
Birth parity											4.057	***	3.310	4.973	0.421	4.051	***	3.304	4.967	0.421	4.035	***	3.292	4.945	0.419
Birth interval											0.679	***	0.604	0.763	0.040	0.680	***	0.605	0.764	0.040	0.679	***	0.604	0.763	0.041
Healthcare factors
Prenatal care index																0.883		0.535	1.457	0.226	0.884		0.536	1.457	0.225
Postnatal care mother index																0.985		0.745	1.302	0.140	0.984		0.745	1.301	0.140
Postnatal care child index																0.935		0.755	1.158	0.102	0.938		0.756	1.163	0.103
Nutritional factors
Minimum dietary diversity (MDD)																					0.979		0.835	1.147	0.079
Minimum meal frequency (MMF)																					0.880		0.721	1.075	0.090
Minimum acceptable diet (MAD)																					1.118		0.761	1.642	0.219
Random effect (cluster-level)	0.725		0.599	0.878	0.071	0.658		0.532	0.814	0.071	0.744		0.596	0.928	0.084	0.744		0.596	0.928	0.084	0.742		0.595	0.924	0.083
N (level-1)	20,770					20,770					20,770					20,770					20,770
N (level-2)	3953					3953					3953					3953					3953

Notes: *p < 0.05, **p < 0.01, ***p < 0.001.

Statistically significant coefficient at p < 0.05 are bolded. Reference category is given in parentheses.

The interaction between ethnicity and/or race and rural-urban residence is associated with 155-percent greater odds of under-5 mortality for indigenous and/or afro-descendant respondents in rural areas, but the effect loses significance across models. Subsequent models control for additional proximate factors—socioeconomic, individual, reproductive, healthcare, and nutrition—present similar findings to the baseline model (Models 2–5). Socioeconomic factors, such as mother's education, plays a significant protective role in diminishing the risk of under-5 mortality, but the strength weakens with the introduction of individual and reproductive factors. Other controls are also initially protective, but lose significance with the introduction of individual and reproductive factors (Model 3). Unexpectedly, results indicate that healthcare and nutritional factors are not protective of under-5 mortality (Models 4–5).

3.1.2. Stunting

Table 5 presents the odds ratio results of the multilevel logit model predicting stunting among children in Colombia, Guatemala, and Peru, between 1986 and 2015 (n (level-1)=15,828; n (level-2)=3372). Self-identifying as indigenous and/or afro-descendant is associated with a higher risk of stunting, which remains significant throughout the models (Models 1–5). Even after accounting for all proximate factors, indigenous and/or afro-descendent women have 62-percent greater odds of having a child stunted, than non-indigenous and/or non-afro-descendent mothers (p-value≤0.000). Across the models, living in a rural area, living in Guatemala or Peru (compared to Colombia), lower maternal education, and higher parity, are associated with higher risk of stunting (Models 2–4). Surprisingly, the interaction between ethnicity and/or race and rural-urban residence is associated with higher risk of stunting for non-indigenous and/or non-afro-descendant respondents living in rural areas (Models 1–5). Generally, household wealth, household environment, husband's education, and higher birth interval are associated with lower risk of stunting (Models 2–5). Like under-5 mortality, healthcare factors do not play a protective role for stunting, but nutritional factors do. However, while MAD is associated with lower risk of stunting, MDD and MMF are counterintuitively associated with a higher risk (Model 5).

Table 5.

Results of multilevel logit model for the odds of stunting in Latin America (Source: author's calculations of Demographic and Health Surveys data for Colombia, Guatemala, and Peru, 1986–2015; N (level-1) = 15,828; N (level-2) = 3372).

Variables	Model 1					Model 2						Model 3						Model 4						Model 5
	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient			95% C·I.	95% C·I.	S.D.	Coefficient			95% C·I.	95% C·I.	S.D.	Coefficient			95% C·I.	95% C·I.	S.D.
			Low	High				Low	High					Low	High					Low	High					Low	High
Year	0.953	***	0.926	0.980	0.014	0.947	***	0.921	0.974	0.014	0.950		***	0.924	0.977	0.014	0.951		***	0.924	0.979	0.014	0.952		***	0.925	0.980	0.014
Ethnicity and/or race
Indigenous and/or afro-descendant (ref. = not)	1.580	***	1.247	2.003	0.191	1.580	***	1.252	1.993	0.187	1.562		***	1.237	1.971	0.186	1.580		***	1.249	1.997	0.189	1.622		***	1.280	2.055	0.196
Geographic factors
Rural (ref. = urban)	3.073	***	2.629	3.591	0.245	1.263	**	1.055	1.513	0.116	1.267		**	1.056	1.519	0.117	1.264		**	1.054	1.517	0.118	1.281		**	1.068	1.538	0.119
Guatemala (ref. = Colombia)	6.621	***	5.261	8.333	0.777	5.696	***	4.398	7.377	0.752	5.564		***	4.288	7.220	0.740	6.423		***	4.745	8.694	0.992	6.708		***	4.936	9.115	1.050
Peru (ref. = Colombia)	2.991	***	2.415	3.705	0.327	2.804	***	2.251	3.494	0.314	2.772		***	2.221	3.459	0.313	2.892		***	2.239	3.734	0.377	2.917		***	2.253	3.776	0.384
Ethnicity x type of residence (ref.=not ethnic x urban)
Not ethnic x rural	3.073	***	2.629	3.591	0.245	1.263	**	1.055	1.513	0.116	1.267		**	1.056	1.519	0.117	1.264		**	1.054	1.517	0.118	1.281		**	1.068	1.538	0.119
Ethnic x urban	1.580	***	1.247	2.003	0.191	1.580	***	1.252	1.993	0.187	1.562		***	1.237	1.971	0.186	1.580		***	1.249	1.997	0.189	1.622		***	1.280	2.055	0.196
Ethnic x rural	2.784	***	2.309	3.357	0.266	1.184		0.973	1.442	0.119	1.184			0.972	1.441	0.119	1.192			0.978	1.452	0.120	1.210			0.992	1.476	0.123
Socioeconomic factors
Mother's education (ref. = 10+ years)
0 years						2.362	***	1.836	3.039	0.304	2.178		***	1.676	2.831	0.291	2.166		***	1.666	2.816	0.290	2.234		***	1.716	2.910	0.301
1–3 years						1.966	***	1.593	2.426	0.211	1.875		***	1.501	2.344	0.213	1.869		***	1.494	2.337	0.213	1.920		***	1.533	2.404	0.220
4–6 years						1.692	***	1.418	2.020	0.153	1.669		***	1.387	2.009	0.158	1.676		***	1.393	2.018	0.159	1.683		***	1.396	2.028	0.160
7–9 years						1.263	**	1.040	1.535	0.125	1.279		**	1.046	1.565	0.132	1.283		**	1.049	1.569	0.132	1.293		**	1.057	1.582	0.133
Household wealth index						0.817	***	0.751	0.889	0.035	0.828		***	0.759	0.902	0.036	0.831		***	0.762	0.906	0.036	0.822			0.754	0.896	0.036
Household environment index						0.564	***	0.432	0.736	0.077	0.582		***	0.446	0.761	0.079	0.589		***	0.451	0.770	0.081	0.569			0.434	0.745	0.078
Husband's education						0.969	***	0.952	0.985	0.008	0.972		***	0.955	0.988	0.008	0.972		***	0.955	0.988	0.008	0.971			0.954	0.987	0.009
Individual and reproductive factors
Respondent's age											1.008			0.996	1.021	0.007	1.008			0.996	1.021	0.007	1.005			0.992	1.018	0.007
Age at first birth											1.059			0.924	1.214	0.074	1.060			0.925	1.215	0.074	1.071			0.934	1.228	0.075
Birth parity											1.192			1.019	1.394	0.095	1.189		*	1.016	1.391	0.095	1.235		***	1.055	1.446	0.099
Birth interval											0.779		***	0.716	0.847	0.033	0.780		***	0.718	0.849	0.033	0.787		***	0.724	0.856	0.034
Healthcare factors
Prenatal care index																	1.201			0.735	1.962	0.301	1.196			0.728	1.965	0.303
Postnatal care mother index																	0.784			0.599	1.028	0.108	0.801			0.612	1.050	0.110
Postnatal care child index																	0.957			0.800	1.146	0.088	0.946			0.790	1.133	0.087
Nutritional factors
Minimum dietary diversity (MDD)																							1.425		***	1.260	1.610	0.089
Minimum meal frequency (MMF)																							1.202		*	1.029	1.404	0.095
Minimum acceptable diet (MAD)																							0.478		***	0.343	0.668	0.081
Random effect (cluster-level)	0.793		0.664	0.948	0.072	0.763		0.630	0.925	0.075	0.763			0.628	0.926	0.076	0.760			0.626	0.922	0.075	0.768			0.631	0.935	0.077
N (level-1)	15,828					15,828					15,828						15,828						15,828
N (level-2)	3372					3372					3372						3372						3372

Notes: *p < 0.05, **p < 0.01, ***p < 0.001.

Statistically significant coefficient at p < 0.05 are bolded. Reference category is given in parentheses.

3.1.3. Wasting

Table 6 presents the odds ratio results of the multilevel logit model predicting wasting among children in Colombia, Guatemala, and Peru between 1986 and 2015 (n (level-1)=15,827; n (level-2)=3372). Like stunting, self-identifying as indigenous and/or afro-descendant is associated with higher risk of wasting, which remains significant throughout the models (Models 1–5). Compared to stunting, however, indigenous and/or afro-descendent self-identification has 1.5-times stronger effect on wasting, that is, 158-percent greater odds of having a child wasted (p-value≤0.000). While living in a rural area is initially associated with higher risk of wasting, the association loses significance with the introduction of socioeconomic factors (Model 2). Like stunting, the interaction between ethnicity and/or race and rural-urban residence is associated with higher risk of wasting for indigenous and/or afro-descendant respondents living in urban areas (Models 1–5). Also, living in Guatemala or Peru (compared to Colombia), lower maternal education, and parity, are associated with higher risk of wasting (Models 2–3). On the other hand, household wealth, household environment, husband's education, birth interval, and maternal postnatal care are associated with lower risk of wasting (Models 2–5). In contrast to stunting, nutritional factors are not significantly associated with wasting (Model 5).

Table 6.

Results of multilevel logit model for the odds of wasting in Latin America (Source: author's calculations of Demographic and Health Surveys data for Colombia, Guatemala, and Peru, 1986–2015; N (level-1) = 15,827; N (level-2) = 3372).

Variables	Model 1					Model 2					Model 3						Model 4						Model 5
	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient			95% C·I.	95% C·I.	S.D.	Coefficient			95% C·I.	95% C·I.	S.D.
			Low	High				Low	High				Low	High					Low	High					Low	High
Year	0.986		0.943	1.031	0.023	0.981		0.936	1.028	0.023	0.985		0.939	1.033	0.024	0.990			0.942	1.039	0.025	0.990			0.943	1.040	0.025
Ethnicity and/orrace
Indigenous and/or afro-descendant (ref. = not)	2.511	***	1.729	3.646	0.478	2.559	***	1.798	3.642	0.461	2.519	***	1.762	3.600	0.459	2.562		***	1.784	3.681	0.474	2.578		***	1.794	3.704	0.477
Geographic factors
Rural (ref. = urban)	2.786	***	2.172	3.574	0.354	1.089		0.819	1.448	0.158	1.097		0.823	1.463	0.161	1.082			0.810	1.447	0.160	1.085			0.812	1.450	0.160
Guatemala (ref. = Colombia)	4.038	***	2.845	5.730	0.721	3.160	***	2.116	4.721	0.647	3.169	***	2.110	4.759	0.658	3.903		***	2.480	6.141	0.903	3.904		***	2.476	6.155	0.907
Peru (ref. = Colombia)	2.428	***	1.722	3.424	0.426	2.211	***	1.551	3.151	0.400	2.030	***	1.415	2.913	0.374	2.326		***	1.537	3.520	0.492	2.331		***	1.540	3.529	0.493
Ethnicity x type of residence (ref.=not ethnic x urban)
Not ethnic x rural	2.786	***	2.172	3.574	0.354	1.089		0.819	1.448	0.158	1.097		0.823	1.463	0.161	1.082			0.810	1.447	0.160	1.085			0.812	1.450	0.160
Ethnic x urban	2.511	***	1.729	3.646	0.478	2.559	***	1.798	3.642	0.461	2.519	***	1.762	3.600	0.459	2.562		***	1.784	3.681	0.474	2.578		***	1.794	3.704	0.477
Ethnic x rural	3.208	***	2.394	4.300	0.479	1.339		0.994	1.803	0.204	1.329		0.985	1.795	0.204	1.332			0.985	1.801	0.205	1.340			0.991	1.811	0.206
Socioeconomic factors
Mother's education (ref. = 10+ years)
0 years						2.547	***	1.742	3.724	0.493	2.026	***	1.337	3.070	0.430	2.018		***	1.329	3.064	0.430	2.027		***	1.335	3.078	0.432
1–3 years						1.688	***	1.206	2.362	0.289	1.457	*	1.024	2.074	0.263	1.455		*	1.019	2.078	0.265	1.460		*	1.022	2.086	0.266
4–6 years						1.495	**	1.099	2.036	0.235	1.408	*	1.019	1.946	0.233	1.427		*	1.028	1.980	0.238	1.422		*	1.025	1.975	0.238
7–9 years						1.236		0.871	1.754	0.221	1.258		0.882	1.795	0.228	1.270			0.889	1.814	0.231	1.271			0.889	1.817	0.232
Household wealth index						0.796	***	0.700	0.906	0.053	0.803	***	0.706	0.913	0.053	0.809		***	0.711	0.921	0.053	0.809		***	0.712	0.920	0.053
Household environment index						0.568	***	0.374	0.862	0.121	0.564	***	0.368	0.862	0.122	0.578		**	0.376	0.889	0.127	0.572		**	0.373	0.879	0.125
Husband's education						0.956	***	0.930	0.982	0.013	0.965	**	0.938	0.992	0.014	0.965		**	0.938	0.992	0.014	0.965		**	0.938	0.992	0.014
Individual and reproductive factors
Respondent's age											1.025	**	1.005	1.045	0.010	1.025		**	1.005	1.045	0.010	1.024		**	1.004	1.045	0.010
Age at first birth											1.005		0.819	1.235	0.105	1.009			0.821	1.240	0.106	1.009			0.821	1.240	0.106
Birth parity											1.360	**	1.067	1.733	0.168	1.356		**	1.064	1.729	0.168	1.370		**	1.074	1.747	0.170
Birth interval											0.801	***	0.701	0.915	0.054	0.806		***	0.704	0.921	0.055	0.806		***	0.705	0.922	0.055
Healthcare factors
Prenatal care index																2.109			0.923	4.819	0.889	2.125			0.930	4.852	0.895
Postnatal care mother index																0.678		*	0.463	0.994	0.132	0.685		*	0.467	1.004	0.134
Postnatal care child index																0.852			0.640	1.135	0.124	0.849			0.637	1.130	0.124
Nutritional factors
Minimum dietary diversity (MDD)																						1.114			0.931	1.333	0.102
Minimum meal frequency (MMF)																						0.992			0.748	1.317	0.143
Minimum acceptable diet (MAD)																						0.744			0.421	1.315	0.216
Random effect (cluster-level)	0.777		0.586	1.030	0.112	0.692		0.493	0.971	0.120	0.706		0.505	0.985	0.120	0.709			0.508	0.991	0.121	0.705			0.504	0.987	0.121
N (level-1)	15,827					15,827					15,827					15,827						15,827
N (level-2)	3372					3372					3372					3372						3372

Notes: *p < 0.05, **p < 0.01, ***p < 0.001.

Statistically significant coefficient at p < 0.05 are bolded. Reference category is given in parentheses.

3.1.4. Anemia

Table 7 presents the odds ratio results of the multilevel logit model predicting anemia among children in Bolivia, Guatemala, and Peru, between 1986 and 2015 (n (level-1)=13,294; n (level-2)=2474). Like under-5 mortality, self-identifying as indigenous and/or afro-descendant is not significantly associated with anemia throughout the models (p-value≥0.050). While living in a rural area is initially associated with higher risk of anemia, the association reverses with the introduction of socioeconomic factors and becomes protective (Models 2–5). For the first time in this analysis, country of residence does not play a protective role for anemia (Models 1–5) and the interaction between ethnicity and/or race and rural-urban residence is associated with a lower risk of anemia for non-indigenous and/or non-afro-descendant respondents in rural areas (Models 1–5). Maternal education, household wealth, and mother's age are associated with lower risk of anemia (Models 2–3). Surprisingly, child postnatal care is associated with a higher risk of anemia (Models 4–5), as are maternal age-at-first-birth and birth parity (Models 3–5). Similarly, to stunting, nutritional factors play a counterintuitive role. While MMF is associated with lower risk of anemia, MDD and MAD are associated with a higher risk (Model 5).

Table 7.

Results of multilevel logit model for the odds of anemia in Latin America (Source: author's calculations of Demographic and Health Surveys data for Bolivia, Guatemala, and Peru, 1986–2015; N (level-1) = 13,294; N (level-2) = 2474).

	Model 1					Model 2						Model 3						Model 4						Model 5
Variables	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient		95% C·I.	95% C·I.	S.D.	Coefficient			95% C·I.	95% C·I.	S.D.	Coefficient			95% C·I.	95% C·I.	S.D.	Coefficient			95% C·I.	95% C·I.	S.D.
			Low	High				Low	High					Low	High					Low	High					Low	High
Year	0.941	***	0.915	0.967	0.013	0.931	***	0.904	0.958	0.014	0.935		***	0.909	0.962	0.014	0.929		***	0.902	0.956	0.014	0.931		***	0.904	0.958	0.014
Ethnicity and/or race
Indigenous and/or afro-descendant (ref. = not)	0.888		0.705	1.118	0.104	0.872		0.690	1.102	0.104	0.877			0.693	1.110	0.105	0.890			0.701	1.129	0.108	0.885			0.698	1.121	0.107
Geographic factors
Rural (ref. = urban)	1.231	***	1.068	1.417	0.089	0.806	**	0.684	0.949	0.067	0.809		**	0.687	0.952	0.067	0.820		**	0.697	0.966	0.068	0.813		**	0.691	0.957	0.068
Guatemala (ref. = Bolivia)	0.702		0.470	1.048	0.143	0.783		0.521	1.176	0.163	0.739			0.491	1.113	0.154	0.767			0.501	1.174	0.167	0.679			0.440	1.049	0.151
Peru (ref. = Bolivia)	1.081		0.785	1.488	0.176	1.162		0.841	1.605	0.191	1.231			0.888	1.706	0.205	1.123			0.799	1.577	0.195	1.031			0.725	1.464	0.185
Ethnicity x type of residence (ref.=not ethnic x urban)
Not ethnic x rural	1.231	***	1.068	1.417	0.089	0.806	**	0.684	0.949	0.067	0.809		**	0.687	0.952	0.067	0.820		**	0.697	0.966	0.068	0.813		**	0.691	0.957	0.068
Ethnic x urban	0.888		0.705	1.118	0.104	0.872		0.690	1.102	0.104	0.877			0.693	1.110	0.105	0.890			0.701	1.129	0.108	0.885			0.698	1.121	0.107
Ethnic x rural	1.698	***	1.431	2.015	0.148	1.126		0.930	1.365	0.110	1.126			0.929	1.366	0.111	1.142			0.940	1.389	0.114	1.118			0.920	1.359	0.111
Socioeconomic factors
Mother's education (ref. = 10+ years)
0 years						1.243		0.977	1.583	0.153	1.323		*	1.028	1.701	0.170	1.329		*	1.032	1.711	0.171	1.344		*	1.044	1.730	0.173
1–3 years						1.339	***	1.082	1.658	0.146	1.376		***	1.102	1.718	0.156	1.388		***	1.112	1.733	0.157	1.390		***	1.114	1.734	0.157
4–6 years						1.249	**	1.055	1.478	0.107	1.253		**	1.052	1.492	0.112	1.262		***	1.060	1.503	0.112	1.270		***	1.065	1.515	0.114
7–9 years						1.180		0.979	1.424	0.113	1.167			0.964	1.414	0.114	1.171			0.966	1.418	0.115	1.163			0.959	1.411	0.115
Household wealth index						0.840	***	0.773	0.913	0.036	0.860		***	0.791	0.936	0.037	0.860		***	0.790	0.936	0.037	0.859		***	0.789	0.935	0.037
Household environment index						1.039		0.792	1.363	0.144	1.083			0.825	1.421	0.150	1.061			0.808	1.394	0.148	1.062			0.809	1.393	0.147
Husband's education						0.990		0.971	1.009	0.010	0.989			0.970	1.008	0.010	0.990			0.971	1.008	0.010	0.991			0.973	1.010	0.010
Individual and reproductive factors
Respondent's age											0.970		***	0.957	0.983	0.007	0.970		***	0.957	0.983	0.007	0.969		***	0.956	0.982	0.007
Age at first birth											1.208		***	1.061	1.375	0.080	1.208		***	1.061	1.375	0.080	1.206		***	1.058	1.374	0.080
Birth parity											1.307		***	1.110	1.539	0.109	1.313		***	1.115	1.546	0.109	1.310		***	1.112	1.543	0.109
Birth interval											0.952			0.874	1.037	0.042	0.950			0.872	1.035	0.042	0.948			0.870	1.034	0.042
Healthcare factors
Prenatal care index																	0.867			0.520	1.446	0.226	0.879			0.526	1.468	0.230
Postnatal care mother index																	1.062			0.841	1.341	0.126	1.031			0.818	1.299	0.122
Postnatal care child index																	1.212		*	1.012	1.451	0.112	1.206		*	1.007	1.445	0.111
Nutritional factors
Minimum dietary diversity (MDD)																							1.152		*	1.015	1.307	0.074
Minimum meal frequency (MMF)																							0.693		***	0.596	0.807	0.054
Minimum acceptable diet (MAD)																							1.955		***	1.451	2.632	0.297
Random effect (cluster-level)	0.697		0.581	0.836	0.065	0.670		0.556	0.807	0.064	0.672			0.558	0.809	0.064	0.676			0.561	0.814	0.064	0.669			0.555	0.807	0.064
N (level-1)	13,294					13,294					13,294						13,294						13,294
N (level-2)	2474					2474					2474						2474						2474

Notes: *p < 0.05, **p < 0.01, ***p < 0.001.

Statistically significant coefficient at p < 0.05 are bolded. Reference category is given in parentheses.

4.0. Discussion

This analysis uses DHS data for Bolivia, Colombia, Guatemala, and Peru, between 1986 and 2015, to explore the multilevel relationship between self-identifying as indigenous and/or afro-descendant and child health outcomes. I tested for the moderating effects of geographic, socioeconomic, individual, reproductive, healthcare, and nutritional variables. This analysis makes two important contributions to the literature. First, it provides an empirical assessment of persistent and pronounced child health disparities across ethnic and/or racial groups in Latin America. In concurrence with past studies, I found that self-identifying as indigenous and/or afro-descendant is associated with higher risk of stunting and wasting (Casas et al., 2001; Giuffrida et al., 2007). Most surprisingly, however, under-5 mortality and anemia are not, which challenges previous research on these two outcomes (Kuang-Yao Pan et al., 2010; Psacharopoulos & Patrinos, 1994, pp. 32–4566).

With previous research documenting that indigenous and afro-descendent people suffer higher levels of poverty and marginalization, precarious and difficult employment conditions, higher levels of illiteracy, lower access to formal education, worse overall health, and limited political participation and representation, my findings imply that efforts to improve child health in Latin America should account for ethnicity and/or race. This research shows that minority ethnoracial groups, such as indigenous and/or afro-descendent, have higher risk of childhood morbidity than do non-minority ethnoracial groups in the region. As pressure increases to improve children's health, as well as, to address ethnoracial health inequities in the developing world, it is increasingly important to truly understand this relationship given severe resource constraints. In addition, these efforts should also accompany education for both men and women, as well as, information about the effects of reproductive decisions on their children's health.

Second, this analysis assesses heterogeneity of this relationship across and within countries in Latin America. While previous research has mainly focused on country-level effects (Frost et al., 2005; Jokisch & McSweeney, 2011) and/or cross-country effects (Hatt & Waters, 2006; Heaton et al., 2005), I also identify and contribute to understanding the cross-regional effects (Van de Poel et al., 2007). In concurrence with past studies, multiple results from this analysis suggest significant cross-country and rural-urban differences. Self-identifying as indigenous and/or afro-descendent and residing in an urban area slightly protects from stunting and wasting, but does not protect from under-mortality and anemia. These results contradict previous research that has documented worse overall health outcomes for minority children (Shin, 2007). Consistent with previous research, maternal education maintains a strong effect on all four health outcomes, even after controlling for all other variables (Frost et al., 2005). However, socioeconomic variables do not have the same strong effect documented in the literature (Van de Poel et al., 2007). As has been documented in other regions, paternal education (Breierova & Duflo, 2004; Semba et al., 2008), as well as, individual characteristics and reproductive decisions (Heaton et al., 2005), play a more significant role in child health outcomes. Paternal education may in fact be important because fathers are often more educated than mothers in developing countries and given their higher status, may have more decision-making power regarding their children's health (Aslam & Kingdon, 2012). Finally, prenatal, postnatal, and feeding practices have mixed associations (De Onis et al., 2006; Ruel & Menon, 2002).

4.1. Limitations

Although this study has made a substantial set of contributions to understanding child health outcomes in Latin America, I acknowledge the following limitations and need for future research that builds on these findings. First, this research relies on self-reported data. Thus, some of these results might be an artifact of reporting bias, whereby respondents selectively choose to share and/or fail to recall certain information about current or previous experiences. For example, the variable I constructed for ethnicity and/or race relies exclusively on self-reported language spoken at home, which may be a conservative measure given that some individuals who self-identify as indigenous and/or afro-descendent no longer speak the languages While ideally, I should use multiple self-identified, interviewer-ascribed, and multiple sub-categories of race and ethnicity, DHS data does not collect these measures so I was forced to collapse both indigenous and/or afro-descendent self-identification into one variable.

Second, this research relies on cross-sectional data, so I am unable to evaluate how self-identifying as indigenous and/or afro-descendant impacts child health over the life course. Third, in an effort to make comparable analytical variables across countries and waves, I collapsed categorical responses, which may have led to the loss of significant information. However, as has been documented, one of the primary advantages of pooling datasets together is an increase in statistical power, which in turn, decreases the likelihood of errors from interviewer noise, poorly worded questions, data entry mistakes, and sampling variability. Finally, this analysis is limited to four countries in Latin America. Despite having the largest populations of indigenous and/or afro-descendant groups, it is important to emphasize that countries in the region have unique cultures, histories, and trajectories, so these results cannot be blindly generalized to other countries in the region.

More research is needed to fully assess the relationship between ethnicity and/or race and child health outcomes in Latin America. To assess generalizability and discuss causal mechanisms, we need additional cross-sectional and longitudinal data using novel indicators. In addition, to provide a more comprehensive picture of the unique experiences of diverse sub-groups of indigenous and/or afro-descendent groups, we need multiple self-identified measures, as well as interviewer-ascribed phenotypic classifications of race and ethnicity. Despite these limitations, results from this study clearly suggests that indigenous and/or afro-descendent respondents have higher risk of stunting and wasting in Latin America. In addition, while most research has previously focused on the protective role of maternal education, results from this study suggest that paternal education, individual characteristics, and reproductive decisions, play significant roles in child health outcomes. Given centuries of discrimination and exclusion, as well as, large populations of indigenous and/or afro-descendent groups in Latin America, we need to further study, understand, and assess the relationship between ethnoracial self-identification and child health outcomes to improve the precarious conditions of ethnoracial minorities in the region.

CRediT authorship contribution statement

Lucrecia Mena-Meléndez: Conceptualization, Methodology, Software, Formal analysis, Data curation, Writing - original draft, Writing - review & editing, Visualization.

Appendix A. Supplementary data

The following is the Supplementary data to this article: