Contributions of individual acculturation and neighborhood ethnic density to variations in Hispanic children's respiratory health in a US–Mexican border metropolis

Abstract

Background

We used an expanded conceptualization of ethnic density at the neighborhood level, tailored to Hispanic majority communities in the USA, and a robust measure of children's acculturation at the individual level, to predict Hispanic children's respiratory health.

Methods

We conducted a cross-sectional survey of 1904 children in 2012 in El Paso, TX, USA. One thousand one hundred and seven Hispanic children nested within 72 census tracts were analyzed. Multilevel logistic regression models with cross-level interactions were used to predict bronchitis, asthma and wheezing during sleep.

Results

A neighborhood-level ethnic density factor was a non-significant risk factor while individual-level acculturation was a significant risk factor for the three outcomes. Pest troubles and not having been breastfed as an infant intensified the positive association between ethnic density and bronchitis. Increases in ethnic density intensified the odds of wheezing in sleep if the child was not low birth weight or was not economically deprived.

Conclusions

Results suggest that increasing individual-level acculturation is detrimental for US Hispanic children's respiratory health in this Hispanic majority setting, while high ethnic density neighborhoods are mildly risky and pose more significant threats when other individual-level factors are present.

Introduction

Hispanic immigrant populations in the USA experience better health outcomes than expected given their social marginality.¹ A lack of acculturation to US norms is the most often cited explanation for this Hispanic health paradox² and evidence suggests that this explanation applies to children's asthma.^3,4 While the concept of acculturation has been thoroughly critiqued,⁵ when immigrants do not adopt the mainstream attitudes and behaviors of US culture (or acculturate^3,6), they appear to derive some protection from illness.^1,7

The Hispanic health paradox has also been attributed to the health benefits of residence in ethnically dense Hispanic neighborhoods.^8–15 While the majority of evidence supports a co-ethnic density advantage for Latino/as,^14,16 there is some evidence that Hispanic ethnic density can be a health risk.^17–19 A limitation of the Hispanic ethnic density literature has been the one-dimensional operationalization of ethnic density as the percentage of the population within areal units that is Hispanic or foreign-born. In communities with highly heterogeneous Hispanic populations and/or diverse multi-ethnic immigrant communities, these variables do not adequately capture ethnic similarities and distinctions among residents. While social support and reduced discrimination are key hypothesized mechanisms for ethnic density's protective effect,¹⁴ co-ethnic discrimination among Latinos based on citizenship, nativity and English language proficiency^20–22 complicates this assumption. In communities with substantial Hispanic populations, there is typically significant intra-Hispanic diversity, making ‘percent Hispanic’ a poor indicator of ethnic cohesion. Such is the case in El Paso, TX, USA, where over 80% of the over 800 000 residents are Hispanic and only eight census tracts have a population in which fewer than half of the residents are Hispanic.

Given the growth and increasing diversity of the Hispanic population in the USA, there is the need for new insight regarding the roles of acculturation and ethnic density in the Hispanic health paradox. Separate studies have demonstrated the effects of acculturation and ethnic density on health outcomes; however, the relative influence of both acculturation and ethnic density has rarely been examined. It is unclear whether the effects of ethnic density found in prior studies will remain significant when accounting for individual-level acculturation, instead of just nativity, which is most commonly used¹⁴ (see Patel et al.²³ for an exception using language acculturation). To address these limitations, we examine the influence of individual children's levels of acculturation measured using the Cultural Life Style Inventory²⁴ and neighborhood ethnic density—using an improved measure appropriate for heterogeneous Hispanic majority communities—on respiratory health outcomes for a representative sample of El Paso (Texas, USA) Hispanic school children. Additionally, we examine how individual risk factors interact with ethnic density to clarify specific limits to the Hispanic health paradox, which has rarely been done (for exceptions, see Refs 9,11,16).

We contribute to the literature through our choice of health conditions, as we study children's lifetime diagnosed asthma, bronchitis in the past year, and wheezing in sleep in the past year. To our knowledge, bronchitis and wheezing in sleep have never been studied as health outcomes in ethnic density studies. Even in the burgeoning acculturation/health literature, children's acculturation has never been examined as a predictor of bronchitis or wheezing in sleep, since most studies focus on asthma. These two conditions are highly relevant in Hispanic communities, as children are likely to suffer from undiagnosed or poorly managed asthma^25,26 (and therefore wheeze at night) and bronchitis²⁷ (possibly connected to an overuse of antibiotics, accessed without a prescription in Mexico).²⁸

Methods

Study design and subjects

Individual-level data for El Paso children were collected through a population-based, cross-sectional, observational mail survey that was approved by our university's Institutional Review Board. The closed-ended questionnaire was sent to all primary caretakers of fourth/fifth grade students in the El Paso Independent School District (EPISD). The EPISD is located within the city of El Paso, TX, USA, which is a Mexican-majority metropolis located on the US–Mexico border; a detailed description of the city is available elsewhere.²⁷ The survey was administered to obtain the highest achievable response rates.²⁹ All survey materials were provided to households in English and Spanish. Mailings were sent in three waves during May of 2012.

Ultimately, 6295 primary caretakers were provided surveys at their home address and 1904 were returned completed for a 30.2% response rate. Eighty-two percent of respondents were mothers and 10% were fathers. 95% of children were aged 9–11 years (mean age = 10.36; standard deviation = 0.767). Even lower survey response rates can yield representative samples.^30–32 Our sample is generally representative of the EPISD student population across all grades.³³ The percent male and percent Hispanic are nearly identical (49.9 versus 51.4% and 82.2 versus 82.6%, respectively); the sample has a lower percentage of economically disadvantaged children than the EPISD as a whole (60.4 versus 71.1%).

Based on home address, residential locations for the children living in El Paso were geocoded. Inclusion criteria were (i) having Hispanic ethnicity, (ii) living at or within 1 mile of the current residence for 12 months or more, and (iii) living in a 2010 census tract (i.e. our operational definition of a neighborhood) that intersected the EPISD boundary with at least one other participating child. This is because at least two cases per cluster are required for the multilevel model. Thus, 1107 Hispanic children residing in 72 census tracts were included in the analysis.

Dependent variables

The questionnaire asked about if the child had bronchitis during the past 12 months,³⁴ if the child had a lifetime asthma diagnosis³⁵ and whether the child had wheezed during sleep in the last 12 months.³⁵ Wheezing during sleep is reflective of undiagnosed or poorly controlled asthma. Table 1 reports descriptive statistics for these three variables.

Table 1.

Descriptive statistics for individual-level (Hispanic children, n = 1107) and neighborhood-level (census tracts, n = 72) variables

Dichotomous variables	Coding	Total N	Yes	No			Percent missing
Individual level
Bronchitis	1 = bronchitis in last 12 months, 0 = no bronchitis in last 12 months	1082	108	974			2.26
Lifetime asthma diagnosis	1 = asthma, 0 = no asthma	1072	161	911			3.16
Wheeze in sleep	1 = wheeze in sleep in last 12 months, 0 = no wheeze in sleep in last 12 months	1084	76	1008			2.08
Obese	1 = obese, 0 = not obese	784	180	605			29.18
Male	1 = male, 0 = female	1079	529	550			2.53
Economic deprivation	1 = free meals, 0 = no free meals	1002	521	481			9.49
Trouble with pests	1 = troubled by one or more pest in the home, 0 = not troubled by pests	1107	487	620			0
Mother's respiratory problems	1 = mother has asthma or allergies, 0 = mother has neither asthma nor allergies	1107	365	742			0
Heath insurance	1 = continuously insured for the last year, 0 = not continuously insured for the last year	1093	929	164			1.26
Breast fed	1 = breastfed, 0 = not breastfed	1091	753	338			1.45
Low birth weight	1 = <2500 g, 0 = 2500 g or greater	925	83	842			16.44
Continuous variables		Total N	Mean	SD	Min.	Max.	Percent missing	Comp. loading
Individual level
Child's acculturationa		1038	−0.1	0.96	−2.6	1.35	6.23	NA
Neighborhood level
Ethnic density factorb		72	0	1	−2.24	2.14	0	NA
Variables comprising the ethnic density factor
% Spanish-speaking Hispanic households isolated by language		72	26.03	13.72	0.00	61.10	0	0.939
% Spanish-speaking households reporting speaking English less than very well		72	26.03	13.71	0.00	61.12	0	0.939
% Households speaking Spanish at home		72	69.58	20.19	12.31	100	0	0.910
% Foreign-born out of the total population		72	27.15	10.80	3.46	51.18	0	0.928
% Non-citizen out of the total population		72	16.13	8.29	1.82	36.68	0	0.922
% Hispanic/Latino of any race		72	76.55	18.54	16.40	99.00	0	0.909
Reversed mean level of education for Hispanics		72	12.81	2.30	17.18	8.19	0	0.855
Reversed median household income in the past 12 months (in 2011 inflation-adjusted dollars)		72	39 080	25 276	122 791	11 584	0	0.801

^aThis factor was created using all 1901 children in our dataset, and this paper reports results only for the 1107 qualifying Hispanic children. For that reason, the mean is slightly <0 and the standard deviation slightly <1. The component loadings and individual items are presented in Table 2.

^bThis factor was created by analyzing American Community Survey (2007–11) estimates from all El Paso County census tract (n = 161).

Individual-level independent variables

We created a child's acculturation index by factor analyzing the 11 items from the shortened Cultural Life Style Inventory,²⁴ modified to allow the caretaker to report on the child. This scale captures multiple dimensions of cultural preferences and practices, which allows for a more thorough consideration of acculturation than the use of other proxies (e.g. nativity). The items and response options are presented in Table 2. To create the factor, we used maximum likelihood extraction and direct oblim rotation methods in SPSS. Oblique methods like direct oblim allow factors to correlate, theoretically rendering more accurate results than orthogonal methods like Varimax.³⁶ A second factor was generated but not analyzed, because all loadings were <0.45.³⁶ The eigenvalue of the first factor was 5.9 and it explained 53.5% of the variance.

Table 2.

Children's acculturation factor: questions, responses and factor loadings for Cultural Life Style Inventory (CLSI) questions

CLSI question	CLSI response options	Comp. 1	Comp. 2
What language does the child use when she or he speaks with her or his brothers and sisters?	(1) Only in Spanish; (2) more Spanish than English; (3) Both in English and Spanish about equally; (4) more in English than Spanish; (5) only in English	0.800	0.327
What language does the child use when he or she speaks with his or her parents or primary caretakers?		0.880	0.377
What language does the child use when he or she speaks with his or her closest friends?		0.845	0.420
What kind of radio stations does the child listen to?	(1) Only Spanish speaking; (2) mostly Spanish speaking; (3) both English and Spanish speaking about equally; (4) mostly English speaking; (5) only English speaking	0.547	0.106
What kind of TV programs does the child watch?		0.813	0.336
What kind of magazines, websites, books or newspapers does the child read?	(1) Only in Spanish; (2) more Spanish than English; (3) both in English and Spanish about equally; (4) more in English than Spanish; (5) only in English	0.805	0.334
In what language does the child pray?		0.800	0.327
In what language are the jokes with which the child is familiar?	(1) All are in Spanish; (2) more are Spanish than in English; (3) some are in English and some are in Spanish about equally; (4) more are in English than in Spanish; (5) all are in English	0.880	0.377
What is the ethnic background of the child's closest friends?	(1) All are Hispanic (Mexican, Mexican American, Latino, etc.); (2) most are Hispanic; (3) both non-Hispanic and Hispanic about equally; (4) most are non-Hispanic (white Anglo, African American, Asian American, etc.); (5) all are non-Hispanic (white Anglo, African American, Asian American, etc.)	0.845	0.420
When the child goes to social functions such as parties, dances, picnics or sports events, what is the ethnic background of the people that the child tends to go with?	(1) Always Hispanics; (2) mostly with Hispanics; (3) both with non-Hispanics and Hispanics about equally; (4) mostly with non-Hispanics; (5) always with non-Hispanics	0.547	0.106
What types of national or cultural holidays (such as Fourth of July and Dieciséis de Septiembre) does the child typically celebrate?	(1) Only Mexican holidays; (2) mostly Mexican holidays; (3) both American (USA) and Mexican holidays about equally; (4) more American (USA) holidays; (5) only American (USA) holidays	0.813	0.336

Apart from child's acculturation, individual-level attributes known to be associated with children's respiratory health are included as control variables. See Table 1 for descriptive statistics. They include sex³⁷ and economic deprivation, which was operationalized as qualifying for free/reduced price meals at school and constructed using guidelines from the Food and Nutrition Service of US Department of Agriculture and survey data about the number of persons living in the child's home and the household's total annual income. In terms of medical history, we account for obesity,³⁸ based on Center for Disease Control guidelines and parent-reported height, weight, age and sex; low birth weight,³⁹ which was created using reported birth weight and International Classification of Diseases guidelines; if the child was ever breastfed;⁴⁰ and maternal respiratory problems.⁴¹ We use continuous insurance coverage over the last 12 months to measure access to care. We adjust for in-home environmental conditions using pest exposure.⁴²

Neighborhood-level independent variable

To improve on previous studies of Hispanic ethnic density which have utilized neighborhood proportion foreign-born or proportion Hispanic, we drew on previous studies of neighborhood acculturation.^43,44 These studies employed census data to create area-based proxies that are sensitive to similarities and distinctions within US Hispanic populations. For example, in the first study of this kind, Espinoza de los Monteros et al.⁴⁴ combined measures of the percentages of foreign-born individuals, foreign-born individuals who arrived within 10 years prior to the census and Spanish-speaking households who reported speaking English less than very well. Building from that and more recent work on ethnic enclaves,⁴⁵ we factor analyzed a series of highly relevant American Community Survey 5-year estimates (2007–11) at the census tract level (Table 1). We again used maximum likelihood extraction and direct oblim rotation methods.³⁶ Only one factor was extracted; its eigenvalue was 6.5 and it explained 81.2% of the variance (Table 1).

Multiple imputation

Multiple imputation (MI) was applied to the individual-level dataset to address missing values and non-response bias, and the multiply imputed data were analyzed using HLM7 software. MI involves creating multiple sets of values for missing observations using a regression-based approach. It is used to avoid the bias that can occur when missing values are not missing completely at random⁴⁶ and is appropriate for self-reported survey data.⁴⁷ In SPSS, 10 imputed datasets were specified to increase power and 200 between-imputation iterations were used to ensure that the resulting imputations were independent of each other. HLM7 software accommodates the MI procedure by performing separate analyses on each dataset and then pooling results across analyses. Statistics for percent missing of each variable are included in Table 1.

Analytic strategy

We employed hierarchical logistic regression modeling using HLM7 software to assess the influence of neighborhood-level and individual-level risk factors on individual-level binary-dependent variables. HLRM is preferable to traditional logistic regression in this case, because ignoring the hierarchical structure of data causes aggregation bias and leads to incorrect inferences.⁴⁸ First, we predicted the three dependent variables using the individual- and neighborhood-level variables. Second, we added cross-level interactions between the neighborhood-level predictor and all individual-level predictors. Then, we conducted a sensitivity analysis, using percent foreign-born and then percent Hispanic, instead of the neighborhood ethnic density factor. According to variance inflation factor, tolerance, and condition index criteria,⁴⁹ inferences from the models are not affected by multicollinearity.

Results

Table 3 reports main effects. The mother having respiratory problems (odds ratio, OR = 2.6; confidence interval, CI = 1.8–3.9) and higher levels of child's acculturation (OR = 1.4; CI = 1.1–1.8) increased the odds of a child having bronchitis in the past year. The neighborhood Hispanic ethnic density factor was positive but not significant (OR = 1.2; CI = 0.9–1.6). In the asthma model, the mother having respiratory problems and child's acculturation were again significant and positive (OR = 1.7; CI = 1.3–2.4 and OR = 1.4; CI = 1.2–1.8, respectively), as was being obese (OR = 1.5; CI = 1.1–2.1), being male (OR = 1.4; CI = 1.1–1.8) and having health insurance (OR = 2.7; CI = 1.4–5.2). The neighborhood ethnic density factor was positive but non-significant (OR = 1.1; CI = 0.9–1.3). For wheezing in sleep, the mother having respiratory problems (OR = 2.5; CI = 1.5–4.6) and higher levels of child's acculturation (OR = 1.6; CI = 1.1–2.3) were significant. The ethnic density factor was again positive and non-significant (OR = 1.3; CI = 0.9–1.8).

Table 3.

Results of the Hierarchal Logistic Regression Models (HLRM) predicting three respiratory outcomes among El Paso school children: main effects

Fixed effect	A. Bronchitis				B. Asthma				C. Wheeze in sleep
	Coeff	P-value	OR	CI	Coeff	P-value	OR	CI	Coeff	P-value	OR	CI
Intercept	−3.244	<0.001	0.039	(0.017, 0.089)	−3.355	<0.001	0.035	(0.016, 0.075)	−3.401	<0.001	0.033	(0.014, 0.081)
Neighborhood level
Ethnic density	0.182	0.250	1.199	(0.877, 1.639)	0.045	0.644	1.046	(0.861, 1.272)	0.274	0.087	1.315	(0.960, 1.800)
Individual level
Obesity	−0.028	0.915	0.972	(0.579, 1.631)	0.428	0.006	1.535	(1.132, 2.082)	0.104	0.616	1.109	(0.740, 1.663)
Male	0.146	0.403	1.157	(0.822, 1.630)	0.322	0.005	1.379	(1.104, 1.723)	0.309	0.105	1.362	(0.937, 1.979)
Economic deprivation	0.151	0.534	1.163	(0.722, 1.876)	0.209	0.320	1.232	(0.816, 1.862)	0.248	0.334	1.282	(0.775, 2.120)
Pest	0.191	0.257	1.211	(0.869, 1.686)	0.271	0.094	1.311	(0.954, 1.801)	0.170	0.389	1.185	(0.805, 1.744)
Mom's Resp.	0.962	<0.001	2.618	(1.751, 3.915)	0.567	<0.001	1.763	(1.278, 2.434)	0.904	<0.001	2.468	(1.503, 4.053)
Insured	0.220	0.451	1.246	(0.703, 2.206)	0.989	0.003	2.689	(1.388, 5.209)	0.064	0.844	1.066	(0.566, 2.006)
Breastfed	0.106	0.617	1.112	(0.734, 1.684)	−0.075	0.651	0.928	(0.670, 1.284)	0.035	0.881	1.036	(0.653, 1.642)
Low birth weight	0.233	0.288	1.262	(0.821, 1.941)	0.386	0.071	1.471	(0.967, 2.238)	−0.008	0.98	0.992	(0.517, 1.903)
Acculturation	0.321	0.010	1.378	(1.080, 1.759)	0.351	0.001	1.421	(1.153, 1.751)	0.473	0.009	1.605	(1.127, 2.285)

Table 4 reports interaction results. In the bronchitis model, two interaction terms were significant. Reporting pest troubles intensified the positive association between ethnic density and bronchitis. If the child was not breastfed, increases in the ethnic density factor were associated with greater risk; if the child was breastfed, increases in the ethnic density factor were protective. In the asthma model, there were no significant interaction terms. When predicting the odds of wheezing in sleep, there were two significant interaction terms. Increases in the ethnic density factor intensified the odds of wheezing in sleep if the child was not low birth weight or was not economically deprived.

Table 4.

Results of the Hierarchal Logistic Regression Models (HLRM) predicting three respiratory outcomes among El Paso school children: interaction effects

Fixed effect	A. Bronchitis				B. Asthma				C. Wheeze in sleep
	Coeff	P-value	OR	CI	Coeff	P-value	OR	CI	Coeff	P-value	OR	CI
Intercept	−2.042	<0.001	0.130	(0.101, 0.166)	−2.221	<0.001	0.108	(0.082, 0.143)	−2.300	<0.001	0.100	(0.074, 0.136)
Neighborhood level
Ethnic density	0.103	0.410	1.109	(0.865, 1.422)	0.045	0.781	1.046	(0.761, 1.437)	0.362	0.058	1.437	(0.988, 2.089)
Individual level
Obesity	0.006	0.944	1.006	(0.847, 1.196)	0.279	0.007	1.322	(1.082, 1.614)	0.035	0.662	1.036	(0.883, 1.214)
* Ethnic density	0.055	0.614	1.056	(0.852, 1.310)	0.060	0.603	1.061	(0.845, 1.333)	−0.048	0.584	0.953	(0.800, 1.136)
Male	0.083	0.230	1.086	(0.948, 1.245)	0.251	<0.001	1.286	(1.121, 1.475)	0.176	0.014	1.192	(1.037, 1.371)
* Ethnic density	0.036	0.658	1.036	(0.883, 1.217)	−0.071	0.353	0.931	(0.799, 1.085)	−0.076	0.209	0.927	(0.823, 1.044)
Economic deprivation	0.103	0.235	1.109	(0.934, 1.316)	0.142	0.275	1.152	(0.891, 1.489)	0.156	0.097	1.169	(0.971, 1.406)
* Ethnic density	−0.047	0.622	0.954	(0.788, 1.154)	0.022	0.875	1.022	(0.778, 1.342)	−0.275	0.018	0.759	(0.606, 0.953)
Pest	0.103	0.095	1.108	(0.982, 1.251)	0.151	0.118	1.163	(0.961, 1.408)	0.082	0.292	1.086	(0.930, 1.268)
* Ethnic density	0.126	0.035	1.134	(1.009, 1.274)	0.019	0.852	1.019	(0.834, 1.245)	0.040	0.621	1.041	(0.886, 1.222)
Mom's Resp.	0.382	<0.001	1.465	(1.225, 1.753)	0.443	<0.001	1.557	(1.253, 1.933)	0.467	<0.001	1.595	(1.317, 1.932)
* Ethnic density	0.026	0.816	1.026	(0.824, 1.278)	0.183	0.138	1.201	(0.941, 1.533)	0.063	0.582	1.065	(0.848, 1.338)
Insured	−0.072	0.431	0.930	(0.775, 1.116)	0.257	0.022	1.292	(1.040, 1.607)	−0.049	0.665	0.953	(0.762, 1.190)
* Ethnic density	0.039	0.629	1.039	(0.887, 1.218)	0.074	0.490	1.077	(0.870, 1.333)	−0.037	0.751	0.964	(0.764, 1.216)
Breastfed	0.118	0.144	1.125	(0.959, 1.319)	−0.105	0.280	0.900	(0.742, 1.092)	0.031	0.759	1.031	(0.845, 1.258)
* Ethnic density	−0.142	0.045	0.867	(0.755, 0.996)	−0.126	0.191	0.882	(0.729, 1.067)	0.000	0.997	1.000	(0.799, 1.250)
Low birth weight	−0.001	0.992	0.999	(0.849, 1.176)	0.321	0.023	1.379	(1.048, 1.815)	0.064	0.546	1.066	(0.863, 1.317)
* Ethnic density	0.019	0.800	1.019	(0.879, 1.181)	−0.090	0.540	0.914	(0.683, 1.223)	−0.253	0.013	0.777	(0.638, 0.945)
Acculturation	0.142	<0.001	1.152	(1.063, 1.249)	0.210	<0.001	1.234	(1.103, 1.382)	0.149	0.001	1.160	(1.064, 1.266)
* Ethnic density	−0.009	0.806	0.991	(0.918, 1.069)	−0.085	0.184	0.918	(0.808, 1.043)	0.076	0.091	1.079	(0.988, 1.179)

The main effects models (Table 3) fit significantly better (P < 0.001) than the null models (not shown) for each of the three dependent variables. In comparing the interaction models (Table 4) with the main effects models, significantly better fit (P = 0.002) was recorded for wheezing in sleep, but not for the other two outcomes variables (P > 0.50). Results of those two models are still discussed because of their conceptual and theoretical relevance.

The sensitivity analysis demonstrated that the main effects were nearly identical using the ethnic density factor, percent Hispanic or percent foreign-born. In the bronchitis interaction model, there were no significant interaction effects using percent Hispanic or percent foreign-born, but there were two using the factor. In the asthma interaction model, interactions with pest and male and percent foreign-born were significant, but no interactions with the factor or percent Hispanic were significant. In the wheezing interaction model, the interaction with economic deprivation was significant for percent Hispanic and the factor, but not percent foreign-born. The interaction with male was significant for percent foreign-born, but not for the factor or percent Hispanic.

Discussion

Main findings

This study highlights the broader importance of child's acculturation as a risk factor for respiratory health conditions by demonstrating that its reach extends beyond asthma to include bronchitis and wheezing in sleep. The effect of the child's acculturation was largely independent of neighborhood ethnic characteristics and acculturation was a more important determinant than neighborhood ethnic density. The strongest predictor for all respiratory outcomes was maternal respiratory problems.

Results from this study suggest the hypothesis that accounting for individual-level acculturation may explain away the protective main effect of ethnic density found in other studies. We found ethnic density to be a moderate (non-significant) risk factor in all six models presented here. For asthma, we found that the ethnic density factor switched signs from negative (protective factor) to positive (risk factor) with the addition of the child's acculturation variable (model without acculturation not shown), although the neighborhood ethnic density term was not significant in either model. A previously published ethnic density study in El Paso found that percent foreign-born in the census tract was a significant and negative predictor of the odds of a child wheezing in the past year, accounting for parent nativity among other variables.¹¹ For comparison, we predicted child's wheezing in the past year, using the same suite of variables employed in this paper (models not shown). We found that the ethnic density factor has the same sign switch (from negative to positive) after the addition of child's acculturation, although neighborhood ethnic density was not significant in either model. This suggests that a robust measure of individual-level acculturation is needed in ethnic density studies.

We observed two opposing patterns across the significant interaction effects, which should be tested in larger datasets in other contexts. While some individual-level risk factors intensified the effect of ethnic density on bronchitis, other risk factors attenuated the positive effect of increasing ethnic density on wheezing in sleep.

First, having trouble with pests and not having been breastfed as an infant intensified the risky effect of increasing ethnic density on bronchitis. These findings demonstrate synergy between risk factors and a ‘double jeopardy’. While the main effect and correlation (not shown) support the conclusion that breastfeeding is not very closely linked to bronchitis in this sample, breastfeeding seems to act indirectly, serving to attenuate the risk of increasing Hispanic ethnic density. These findings for bronchitis support the conclusion that residing in ethnically dense neighborhoods is more of a health threat for children with underlying risk factors.¹¹

Second, ethnic density was associated with greater odds of wheezing in sleep for children who were not low birth weight and not economically deprived. Results suggest that ethnic density provides a small protective cover against night wheezing for some vulnerable children. In the case of the economic deprivation finding, it seems as if poorer children's families are better able to tap the benefits of social support structures found in poor Hispanic immigrant enclaves.¹¹ The benefits of ethnic density for children of low birth weight in terms of night wheezing are surprising and need further investigation. The protective nature of ethnic density for these two groups of children aligns with the majority of the Hispanic ethnic density literature.¹⁴

What is already known

Living in ethnically dense Hispanic neighborhoods may be a reason behind the Hispanic health paradox.^8–15 While asthma has been rarely examined in ethnic density studies, higher percentages of foreign-born residents in the neighborhood were associated with lower odds of Hispanic children's wheezing and this effect was even stronger for children of foreign-born caretakers in El Paso, TX, USA.¹¹ Among adults in Chicago, there was no significant direct effect of percent foreign-born Latino neighborhood composition on odds of asthma, but foreign-born adults were significantly less likely to have asthma if they lived in heavily immigrant neighborhoods as opposed to neighborhood with few immigrants.⁹ The majority of evidence supports a co-ethnic density advantage for Latino/as,^14,16 but Hispanic ethnic density has been operationalized one-dimensionally. While social support and reduced discrimination are key hypothesized mechanisms for ethnic density's protective effect,¹⁴ co-ethnic discrimination among Latinos based on citizenship, nativity and English language proficiency^20–22 complicates this assumption.

What this study adds

Results suggest that optimism about the apparent resilience of US Hispanic children living in high ethnic density neighborhoods should be tempered, at least in Hispanic majority communities like El Paso. Future studies should control for individual-level acculturation, as we did here, to more accurately isolate the effects of neighborhood ethnic characteristics. This study introduced a novel and more nuanced way of operationalizing ethnic density, borrowing from the neighborhood acculturation literature.⁴⁴ The sensitivity analyses revealed that interaction results are somewhat sensitive to how ethnic density is operationalized, suggesting that a theoretically informed concept is most appropriate. Our measure combined ethnicity, nativity, citizenship, language proficiency/deficiency and SES. Such a measure will be relevant in US cities with large, rapidly growing and heterogeneous Hispanic populations, like Los Angles, New York, Houston, Chicago, Dallas, Miami and Phoenix,⁵⁰ where ‘percent Hispanic’ and ‘percent foreign-born’ are too crude for capturing neighborhood ethnic similarities and distinctions.

Limitations

We used parent-reported height/weight to calculate obesity, which may lead to inaccuracies.⁵¹ The population size of the tracts ranged from 744 to 9399, with a mean of 4654. This size is not ideal to capture ‘everyday lived environments’,⁵² but we could not use census block groups due to the wide margins of error in the ACS estimates and the lack of necessary variables from the 2010 census. While our acculturation scale is based on Berry's well-respected model of acculturation, it does not include a measure of acculturative stress.⁵ We are also missing health behavior variables, which are important explanations for the Hispanic health paradox.²

Funding

This work was supported by Award Number P20 MD002287-05S1 from the National Institute of Minority Health and Health Disparities (NIMHD) and the Environmental Protection Agency (EPA). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIMHD or EPA.