Children enter the crucial transition to school with sociodemographic disparities firmly established. Domain-specific research (e.g., on poverty and family structure) has shed light on these disparities, but we need broader operationalizations of children’s environments to explain them. Building on existing theory, this study articulates the concept of developmental ecology—those interrelated features of a child’s proximal environment that shape development and health. Developmental ecology links structural and demographic factors with interactional, psychological, and genetic factors. Using the Early Childhood Longitudinal Study, Birth Cohort (ECLS-B), this study conducts latent class analyses to identify how 41 factors from three domains—namely, household resources, health risks, and ecological changes—cluster within children as four overarching developmental ecologies. Because it documents how numerous factors co-occur within children, this method allows an approximation of their lived environments. Findings illuminate powerful relationships between race/ethnicity, parental age, socioeconomic background, and nativity and a child’s developmental ecology, as well as associations between developmental ecology and kindergarten cognition, behavior, and health. Developmental ecology represents a major pathway through which demographic characteristics shape school readiness. Because specific factors have different implications depending on the ecologies in which they are embedded, findings support the usefulness of a broad ecological approach.
The transition to school is a critical point in the life course because academic success at this time is highly correlated with achievement in middle and high school and beyond (Butler et al. 1985; Entwisle et al. 2004; Weller et al. 1992). Yet, children enter school with major socioeconomic, racial/ethnic, and other disparities already in place (Entwisle et al. 2004). Only recently, as nationally representative longitudinal survey data from early childhood have become available, has a large body of research begun to trace the initial development of these disparities, with emerging literatures focusing on distinct aspects of young children’s proximal environments, such as family structure changes and experiences of poverty (Cavanagh and Huston 2006; Duncan et al. 1998; Fomby and Cherlin 2007; NICHD Early Child Care Research Network 2005).
Theorists have made conceptual strides with constructs such as capital and habitus, but a broader empirical perspective is needed to understand how various features combine to create those proximal environments. A typical research approach is to isolate the effects of one feature, such as hours spent in nonparental childcare or residential mobility, while controlling for others. However, that one feature may have different implications depending on the multitude of other factors in which it is embedded. Policy solutions that ignore this possibility may have limited success. Thus, a complementary approach is needed that approximates children’s lived everyday environments holistically. Kurt Lewin’s words echo this goal: “To explain social behavior it is necessary to represent the structure of the total situation and the distribution of the forces in it” (1939:868). Existing theoretical perspectives provide the conceptual underpinnings for broadly modeling children’s proximate contexts. This study’s goal is to integrate these existing perspectives by articulating an overarching conceptualization of a child’s developmental ecology. Narrower in scope than other theoretical approaches, such as Bronfenbrenner’s (1979) ecological model of child development and the school transition model (Entwisle et al. 2004), developmental ecology is defined as interrelated features of a child’s proximal social environment that are distinct from but influence children’s social interactions and individual characteristics. A developmental ecology is affected by social structural and demographic factors and shapes interactional and psychological factors, building our understanding of the emergence of developmental and health disparities in the life course. It represents a meso level of analysis that is less well studied than the more macro and micro levels. This concept builds on previous theoretical developments with the goal of facilitating empirical measurement.
Following children from birth to kindergarten in the Early Childhood Longitudinal Study, Birth Cohort (ECLS-B), this study conducts latent class analyses to operationalize the most prevalent developmental ecologies experienced by U.S. children born in 2001, using interrelated features from three domains: household resources, health risks, and changes in social environment. Because it shows how a wide variety of factors co-occur within children, this method allows an approximation of their lived environments. Findings illustrate powerful relationships between demographic factors (race/ethnicity, parental age, socioeconomic background, and nativity) and a child’s developmental ecology, as well as between developmental ecology and children’s cognition, behavior, and health. Developmental ecology represents a major pathway through which demographic factors shape school readiness. Because specific factors have different implications depending on the ecologies in which they are embedded, the study’s findings support the usefulness of a broader ecological approach for understanding school readiness.
Theory and Background
The Concept of Developmental Ecology
Building on insights from several influential theoretical perspectives, this study introduces the concept of a child’s developmental ecology, defined as interrelated features of children’s everyday proximal environments that impact their development and health. A child’s developmental ecology is shaped by more distal influences (such as the community’s social organization and economic circumstances) and demographic factors (such as parental age, socioeconomic background, nativity, and race). Developmental ecology in turn provides a setting for a child’s individual characteristics, perceptions, and interpersonal interactions with family members, teachers, and peers. This study acknowledges the importance of macro-level influences on children’s development and health, but like others studying similar issues (e.g., Becker 1991; Elder 1974), it retains a primary focus on more proximal environments. Developmental ecology is not intended to include children’s interpersonal interactions (such as parenting styles) or individual factors (such as personality or biology). Instead, it measures interrelated features of the meso-level settings that shape interactions and individuals. Examples include the resources, physical features, and health risks in children’s households, childcare settings, and immediate neighborhoods, as well as changes that children experience in these environments. A family’s current socioeconomic resources are part of a child’s developmental ecology, but the socioeconomic background of the parents is not. In operationalizing developmental ecology here, I include just three interrelated domains that are important aspects of children’s everyday environments: household resources, household health risks that impinge on the child, and repeated changes in the child’s environments. Extensive research supports the importance of these domains for school readiness. A child with a benign developmental ecology across a variety of interrelated domains should score more highly on measures of cognitive and behavioral development and health than a child with a compromised developmental ecology.
The first aspect of a child’s developmental ecology studied here is the household’s material and social resources. Socioeconomic resources in early childhood have important implications for school readiness. By kindergarten, children from the lowest socioeconomic quintile have reading, mathematics, and general knowledge scores that are a standard deviation below the scores of children in the highest quintile (Burkam et al. 2004). Other material and social resources—such as food security, medical care, and physical environments—are also important for understanding children’s development, and they may improve children’s school readiness directly or indirectly (Gershoff et al. 2007; Guo and Harris 2000; Mayer 1997). The data set used here permits a multifaceted conceptualization of resources.
A second aspect of developmental ecology is health risks in the form of health behaviors that impinge on the child, including diet, sleep habits, secondhand smoke, violence exposure, and safety precautions. These factors have been linked to health outcomes (Bair-Merritt et al. 2006; Cook et al. 2004; Reilly et al. 2005; Sadeh et al. 2003; Silverstein et al. 2006). Research on health lifestyles emphasizes that health behaviors do not occur in a vacuum but tend to cluster in ways that have implications for individuals’ health and identities (Cockerham 2005; Laaksonen et al. 2003). Limited previous work on health lifestyles in early childhood suggests that they have important consequences for children’s school readiness (Mollborn et al. 2014).
A third important facet of a child’s developmental ecology is repeated changes in children’s proximal environments. Some recent studies have focused on dynamic, time-dependent processes shaping child development (DiPrete and Eirich 2006; Duncan et al. 1998; Wagmiller et al. 2006). Even before they enter school, children have begun to experience transitions that affect their development. Parents’ partnerships and households dissolve and re-form, families move to new homes, childcare arrangements change, and parents enter or leave the labor force. Repeated changes in family and social environments can compromise school readiness (Cavanagh and Huston 2006; Cooper et al. 2009; Joshi and Bogen 2007; NICHD Early Child Care Research Network 2002). Whether a change itself is positive or negative, the turbulence to the family system may be detrimental if the child lacks time to recover between transitions. The repeated entry and exit of a parent’s romantic partners from a household has deleterious consequences for many areas of children’s development, with explanations including mother-partner conflict, parenting behavior, maternal stress, father involvement, and unequal investments in children in blended families (Cavanagh and Huston 2006; Cooper et al. 2009; Fomby and Cherlin 2007; Osborne and McLanahan 2007). Although these largely interactional processes are undoubtedly important, the picture they provide is incomplete. The reported effects of repeated change in mothers’ unions may be a marker for the consequences of more wide-ranging instability. Changes in multiple domains of children’s environments have not been evaluated together in early childhood. I propose an ecological perspective that considers concurrent and cascading environmental changes throughout early childhood.
The concept of developmental ecology has been informed by three important theoretical perspectives. First, the school transition model (Entwisle et al. 2004) identifies social structural circumstances that influence children’s life circumstances, which in turn affect three proximal influences on children’s cognitive preparedness for school: (1) social psychological factors (interpersonal interactions), (2) experiential factors (experiences other than family relationships), and (3) personal factors (child attributes). Crosnoe’s (2006) extension of the model incorporates health as a key component of school readiness.
The second theoretical influence—in which the term “developmental ecology” was first introduced in reference to children—is Bronfenbrenner’s (1979) ecological model of child development. Emphasizing the importance of social environments for expanding developmental psychologists’ understanding of children’s outcomes, Bronfenbrenner’s model has a multilevel focus on intraindividual factors, “multiperson systems of interaction” (1979:21), the proximal settings of interactions, and sociocultural influences. My study’s developmental ecology concept is narrower, emphasizing what Bronfenbrenner calls “microsystems,” or the environments in which children’s interactions with others take place. Isolating these environments from both more proximal and distal influences increases theoretical precision, allowing integration with other midrange theory.
Finally, the concept of developmental ecology is similar in scope to Lewin’s (1939) idea of a child’s “life-space.” Also intended as a way of integrating disparate social influences, a life space as defined by Lewin is subjective and bounded by a child’s perceived experiences. In contrast, a developmental ecology is objectively defined by measurable factors in a child’s social environment. But like a life space, a developmental ecology is limited to the environment that is most proximal to a child and to the interpersonal interactions she experiences.
Other constructs, such as capital and habitus (Bourdieu 1986a, b) and articulations of how they play out in children’s everyday lives (e.g., Lareau 2003), are also related to developmental ecology. However, developmental ecology focuses not on interpersonal interactions or individual dispositions but instead on children’s immediate settings within which interactions and dispositions play out. Ethnographic approaches to these settings, such as Lareau’s (2003), have begun to empirically capture the breadth of interrelated factors in them. This study’s operationalization of developmental ecology mirrors that holistic approach, using nationally representative quantitative data.
The three domains of developmental ecology analyzed here are expected to be interrelated. Developmental ecology may be an important pathway through which major demographic dividing lines that predate the child’s birth—such as race/ethnicity, socioeconomic background, maternal nativity, and teen parenthood—shape school readiness. The relationships between these demographic factors and school readiness have been established (Crosnoe 2006; Entwisle et al. 2004; Mollborn and Dennis 2012). School readiness—measured using the domains of cognition, behavior, and health that have been shown to affect children’s successful school transitions and longer-term outcomes (Crosnoe 2006; Entwisle et al. 2004; Halonen et al. 2006; Weller et al. 1992)—is a strong predictor of academic achievement and adult socioeconomic outcomes (Duncan et al. 1998; Duncan et al. 2007a, b). Health and education in childhood influence socioeconomic attainment and health in adulthood (Haas 2007; Hayward and Gorman 2004; Palloni 2006). This study’s hypotheses are illustrated by a conceptual model in Figure 1.
Hypothesis 1: Major demographic disadvantages (minority race/ethnicity, mother’s foreign birth, mother’s low socioeconomic background, and teen parenthood) predict compromised developmental ecology.
Hypothesis 2: Compromised developmental ecology in early childhood negatively predicts kindergarten readiness.
Hypothesis 3: Developmental ecology is a primary pathway through which demographic disadvantages shape children’s kindergarten readiness.
I do not expect developmental ecology to fully mediate links between demographic characteristics and child outcomes or to perfectly predict kindergarten readiness. Even if all aspects of a child’s developmental ecology were captured here, complementary influences shape school readiness. The ecological perspective can be integrated with theories on residential segregation, labor force participation, kinship, and other extrahousehold processes, as well as interactional, psychological, and genetic processes internal to the child and family.
Household resources, health risks, and changes to a child’s environment are all multifaceted, highly interrelated phenomena. To represent them as such, I conducted latent class analyses (LCA) (Jung and Wickrama 2008; Vermunt and Magidson 2002), which classify cases into subgroups, or classes, according to predominant patterns in multivariate, categorical data (Clogg et al. 1995). The researcher derives latent classes from the observed data rather than imposing classes on the data a priori. LCA allows the inclusion of many variables while retaining parsimony, and it distills naturally occurring interactions among variables in the latent classes rather than requiring each indicator to have an isolated effect on the outcome. LCA is therefore useful for identifying how numerous developmental ecology factors co-occur in a population.
Whereas a typical LCA identifies patterns in a particular domain (such as household resources or health risks), this study needed to model patterns first within, and then across, three different domains. Previous researchers have addressed similar empirical goals in related but distinct types of statistical models. For example, Luyckx and colleagues (2008) estimated sets of latent trajectories over time for college students’ identity formation and adjustment, and then conducted bivariate chi-squared tests to estimate the relationship between these two sets of categories. Fanti and Henrich (2010) predicted latent trajectories of children’s externalizing and internalizing problem behaviors, and then predicted the probability of membership in multiple classes across these two domains. My approach more closely follows Fanti and Henrich’s but uses LCA. I first estimated a set of latent classes for each developmental ecology domain (first-order latent classes), and then conducted a second-order LCA using the first-order latent classes as inputs. This latter analysis modeled developmental ecology by combining the three domains.
Data came from the Early Childhood Longitudinal Study, Birth Cohort (ECLS-B), which surveyed and directly assessed about 10,600 U.S. children born in 2001 from infancy through the fall of kindergarten (U.S. Department of Education 2007).1 As the first nationally representative U.S. longitudinal study of early childhood, it yields the best available data for predicting multiple dimensions of school readiness and assessing young children’s developmental ecologies, including repeated changes over time. The sample was drawn from all 2001 births registered in the National Center for Health Statistics vital statistics system, using a clustered (96 counties and county groups), list frame design.
This study used data from all survey waves, collected at approximately 11, 24, and 52 months of age and also in the fall of the kindergarten year, at an average of 66 months of age. The primary parent, who was almost always the biological mother, completed interviews in person. For budgetary reasons, the kindergarten wave was conducted on a random subsample of about 85 % of the children who had completed the previous wave (Snow et al. 2009). Weighted response rates for the parent interview were 74 %, 93 %, 91 %, and 92 % to 93 %, respectively, at each wave. The sample was restricted to children with valid weights in kindergarten. I retained these approximately 6,450 children by conducting multiple imputation in Stata. (Twenty imputations filled in missing data on all independent and dependent variables informed by all analysis variables, developmental ecology indicators, and earlier child outcomes.) LCA in SAS automatically retained all eligible cases.2 All analyses adjusted for nonresponse and disproportionate sampling using probability weights to make findings representative of U.S. children born in 2001; LCA in SAS adjusted for clustering within primary sampling units, and all other analyses in Stata adjusted for clustering and stratification.
I included as many items as possible in the initial LCA of household resources, health risks, and ecological changes. Except for ecological changes, which covered all of early childhood, all indicators were measured at age 4.5 and were dichotomous unless noted.3 Because data collection from fathers had low response rates, the main focus was on maternal factors.
Household resources were represented in three areas. Socioeconomic resources were measured by household income-to-needs ratio (income as a percentage of the federal poverty line adjusted for household size, with missing values imputed by ECLS-B staff and coded as below the poverty line, near poverty at between 100 % and 200 % of the poverty line, or above); household assets (car ownership, homeownership, investments or retirement savings, and having a checking or savings account); food insecurity (uncertain food provision in the household); uninsured child (no private or government-supported health insurance); and maternal educational attainment (less than a high school diploma, high school diploma, some college, and bachelor’s degree or higher). Social support was measured by the biological father’s coresidence; the number of other coresident children (conceptualized as a drain on social support and coded as 0, 1, or 2 or more); any coresident grandparent (whether a support or a drain); any coresident other adult (whether a support or a drain); and whether a household member besides the study child had special needs (another drain on social support). Everyday activities that require or provide resources were captured through mother not working for pay, mother not enrolled in school, mother currently married, and childcare/preschool type (none, Head Start, other center-based care, kin-based care, or other noncenter care).
Health risks were represented in five areas; see Mollborn et al. (2014) for more details. Diet was measured through household food insecurity (as described earlier); consumption at least daily of milk, soda/sugary drinks, sweet snacks, and salty snacks; and consumption of five or more servings of fruits and vegetables per day. Secondhand smoke exposure was represented by having a family member who smoked inside the home. Safety was operationalized by the presence of a working smoke detector in the home, always wearing a helmet during activities such as skating and biking, always sitting in a car seat or booster seat, always sitting in the back seat of a vehicle, and having an unlocked gun in the home. Sleep was represented through insufficient sleep (fewer than 10 hours per night) and late bedtime (after 10 p.m.). Violence was measured by indicators of the child having witnessed a violent act in the home (such as property destruction or physical fights) and the child having been the victim of violence in the home.
Ecological changes were represented by counts of changes that children had experienced throughout early childhood. I subsequently recoded each count variable into three categories, with cutoffs close to the first and third quartiles. A first set of variables captured a child’s situation at each wave and identified changes between waves, ranging from 0 to 3 total changes: changes in the mother’s coresident spouse or partner (coded as 0, 1, or 2–3), coresident grandparents (none versus any), coresident other adults (none versus any), the number of coresident children (coded as 0, 1, or 2–3), and type of childcare (based on the five categories detailed earlier and coded into 0–1, 2, or 3). Maternal paid work hours and childcare hours were first coded into three categories (0, 1–29, or 30 or more hours per week); then changes from one category to another were counted and recoded into 0, 1, or 2–3 transitions for mother’s paid work and 0–1, 2, or 3 transitions for childcare hours. Residential moves were tracked as the number of moves since the previous wave, ranging from 0–25 and recoded into 0, 1–2, and 3 or more moves.
Children’s ages at the Kindergarten and Wave 1 assessments were included as controls. Analyses also included background factors from before the child’s birth—such as maternal educational experiences, social disadvantage, and exposure to physiological risks in utero—that could be related both to developmental ecology and to child outcomes. The parent’s own report was included when possible, with missing data filled in from other sources. Child gender and race/ethnicity were constructed by ECLS-B. An indicator captured whether the child was born to a mother or father who was younger than 20. Mothers who started receiving prenatal care after the first trimester or not at all were identified, as was the child’s birth weight. Maternal alcohol (at least one drink per week) and tobacco (any) consumption during the third trimester of pregnancy were measured. Dichotomous variables indicated whether the mother was foreign-born, she lived with both biological parents until age 16, she had been born to a teen mother, she ever repeated a grade in school, and her family ever received welfare when she was between the ages of 5 and 16. Finally, Wave 1 cognitive and behavior assessment scores and parent-reported child health (very good or excellent versus others) were included to control for the possibility that early development and health shaped subsequent resources, health risks, or ecological changes.
Analyses focused on six kindergarten outcome variables, from the year the child first enrolled in kindergarten. The two cognitive outcomes (early reading and math) were the result of one-on-one child assessments adapted from reputable assessment batteries developed for other child development studies or for the ECLS-B.4 The cognitive scales are an improvement over assessments of young children that were available in the past (Rock and Stenner 2005). The early reading assessment was a 35-item test covering age-appropriate areas, such as phonological awareness, letter recognition and sound knowledge, word recognition, and print conventions (ECLS-B–reported reliability = .84). The early math two-stage assessment was routed after the first stage, depending on the child’s score and evaluated number sense, geometry, operations, counting, pattern understanding, and measurement (ECLS-B–reported reliability = .89). I standardized the ECLS-B–constructed scale scores for both variables.
Behavior was measured by creating separate scales from parent and teacher reports: parents and teachers are familiar with different dimensions of children’s behavior and have different frames of reference. Higher scores represent more positive behavior. Children’s parent-reported behavior was captured by a standardized index of 24 items that asked the parent how frequently the child behaved in certain ways, using a 5-point scale ranging from “never” to “very often” (Cronbach’s alpha = .86). Sample questions asked how often the child is physically aggressive or acts impulsively and also how often the child shares belongings or volunteers to help other children. The standardized teacher-reported behavior measure came from the kindergarten teacher and included some of the same questions asked of parents. Sample items asked how often the child displayed positive emotion, social interaction, and cooperative behavior.
Health was measured using two dichotomous variables. The broader operationalization used the primary parent’s report of the child’s general health status (very good or excellent compared with good, fair, or poor). Asthma diagnosis was coded as 1 if the parent reported at any wave that a medical professional had diagnosed the child with asthma.5
I conducted LCA using PROC LCA in the SAS statistical software package. Latent class analysis uses categorical items and assumes underlying discrete groups, or “classes,” of respondents. The underlying groups identified are a function of the items chosen. Analyses were conducted using all eligible cases. The best-fitting number of classes was determined using the Bayesian information criterion (BIC) and the Akaike information criterion (AIC). First, three separate sets of latent classes were created from the indicators for each domain (household resources, health risks, and ecological changes). I selected the most parsimonious model with a good fit as indicated by little incremental improvement in BIC or AIC for higher-order solutions: five classes for household resources and for health risks, and six classes for ecological changes.6 I assigned each child a probability of membership in each class, and I assigned the class with the highest probability of membership to each child. This process of modal assignment is less ideal than using continuous membership probabilities, but it was the only feasible approach for second-order LCA and is likely to result in conservative coefficient estimates. See Table 1 for posterior probabilities and population shares for the three domain-specific LCAs.
Second-order LCA used the three categorical variables representing 16 latent classes to create an overarching set of latent classes representing each child’s developmental ecology. A four-class solution emerged as the best fit according to both BIC and incremental improvements in AIC. Table 2 reports the class-conditional response probabilities for these developmental ecology classes, and Table 3 displays means of the control variables and child outcomes for each class with significance tests comparing means with the “advantaged” class. The one-word class labels are intended as mnemonic devices.
Next, descriptive analyses compared means for each class on the sociodemographic and outcome variables. Multinomial logistic regression models predicted the likelihood of membership in developmental ecology classes on the basis of sociodemographic variables. Further multivariate analyses predicted children’s kindergarten readiness (using ordinary least squares (OLS) regressions for cognition and behavior and binary logistic regressions for health) by developmental ecology class and control variables. Finally, graphs presented analyses of developmental ecology classes as mediators of prebirth demographic influences on kindergarten readiness.
Developmental Ecology Latent Classes
Class 1 (advantaged) made up the largest proportion of the sample at 43 % (see Table 2). Its mnemonic comes from the greater resources and reduced health risks and ecological changes generally experienced by these children. The children’s families tended to have high levels of assets and educational attainment, and very few lived in or near poverty. The biological father was nearly always coresident and married to the mother, and 78 % of the children attended center-based preschool. More than three-quarters (78 %) of children in Class 1 belonged to the “consistently positive” health risk class, which represented low overall health risk. These children experienced the highest stability in all developmental contexts except for childcare changes. Control variables in Table 3 also indicated lower risks from background factors. With the exception of maternal alcohol consumption, these children were significantly overrepresented among the more privileged categories for every background variable. Their outcomes were significantly better than those of other classes for every kindergarten outcome, with cognitive scores about 0.4 of a standard deviation above the mean.
Class 2 (vulnerable), a stark contrast to the “advantaged” class in its relatively lower resources and greater health risks and ecological changes, made up 22 % of the sample. These children were likely to live in or near poverty with single mothers whose educational attainment was low and assets were few; 35 % of their households experienced food insecurity. Extended households and household members with special needs were also relatively prevalent. The nutrition/sleep problems and food insecurity/violence/smoking health risk classes were strongly overrepresented, with just 14 % of children not falling into a problematic health risk class. Their environments were characterized by considerable change, with the highest instability of any class for coresident child and other adult transitions, moves, and childcare type transitions. Beyond these developmental ecology risks, Table 3 shows that the children in this class had high background risk. Their mothers were the most likely to have smoked during pregnancy and received inadequate prenatal care, and they were the most frequently born with low birth weight. “Vulnerable” children had compromised kindergarten outcomes. Their reading and math scores were one-half of a standard deviation below the sample mean, their teacher-reported behavior scores were one-third of a standard deviation below the sample mean, and 27 % had been diagnosed with asthma.
Class 3 (traditional) made up 20 % of the sample (see Table 2). The mnemonic for this class comes from its family structure, maternal work, and childcare patterns, which resemble those of a stereotypical family from decades past. These children tended to live near poverty at preschool age, with the biological father coresident and married to the mother, who was typically high school–educated. About one-half the children had nonworking mothers and did not attend preschool or childcare. These children were mostly represented in health risk classes that had neither the best nor the worst developmental implications. Their ecological change histories fell in the middle for every domain except childcare, which evidenced high stability. “Traditional” children had significantly compromised early reading and math scores at approximately one-quarter of a standard deviation below the sample mean, and overall health was the lowest of any class.
Finally, Class 4 (nontraditional) made up 15 % of the sample. In terms of household structure, childcare, and maternal activities, these children experienced newer, increasingly prevalent family forms. “Nontraditional” children tended to live with single mothers, and their socioeconomic resources were second highest after those of “advantaged” children. The children often lived in extended households and attended center-based preschool, and 85 % of their mothers worked. These children fell near the middle in many health risks, but their sleep patterns were as problematic as those of “vulnerable” children. They experienced the most partner, grandparent, maternal work, and childcare hours transitions. “Nontraditional” children were sometimes demographically similar to the “traditional” class, with less risky backgrounds than those in the “vulnerable” class. They had the second-highest reading, math, parent-reported behavior, and overall health scores. Their teacher-reported behavior scores and asthma prevalence were second lowest.
The four developmental ecology categories that emerged from the data thus do not merely represent increasing levels of risk on the same underlying dimension. Instead, a multitude of developmental ecology factors combine in ways that are often unintuitive, such as the fairly high levels of ecological instability and relatively positive developmental outcomes that characterize “nontraditional” children. Some children who come from similar demographic backgrounds have very different developmental ecologies—specifically, the “traditional” and “nontraditional” classes—that are associated with different levels of school readiness. A child’s developmental ecology cannot be predicted from her sociodemographic background alone. Moreover, the same factor can look indistinguishable in two developmental ecologies, yet its implications for school readiness can be very different depending on the developmental ecology in which it is embedded. Two developmental ecologies (“vulnerable” and “nontraditional”) both had high levels of partner transitions in the household, coexisting with other types of high instability. However, the high partner instability was combined with very different levels of resources and health risks, so ultimately, the “vulnerable” children were much worse off in kindergarten.
Hypothesis 1: Major Demographic Disadvantages Predict Compromised Developmental Ecology
Table 4 reports coefficients from multinomial logistic regression analyses predicting the likelihood of developmental ecology class membership. The results reinforce the finding that powerful processes sort children from different sociodemographic groups into different developmental ecologies. Race/ethnicity strongly predicted developmental ecology class. Black and Hispanic children had the highest likelihood of being in the “vulnerable” class compared with “advantaged,” with relative risk ratios of 29 and 11, respectively, compared with white children. Black and Hispanic children also had significantly higher odds of being in the “traditional” and “nontraditional” classes. Children with foreign-born mothers were most likely to be in the “traditional” class and were significantly less likely than others to be in the “nontraditional” class.
Socioeconomic background was another predictor of developmental ecology. Children whose mothers had received welfare benefits in childhood were twice as likely to be in either the “vulnerable” or the “traditional” class as in the “advantaged” one. Children of teen parents were fully 42 times as likely to be in the “vulnerable” as the “advantaged” class, 19 times as likely to be in the “nontraditional” class, and 9 times as likely to be in the “traditional” class. Each of the hypothesized demographic dividing lines significantly predicted children’s developmental ecology classes as expected, supporting Hypothesis 1.
Hypothesis 2: Favorable Developmental Ecology Predicts Kindergarten Readiness
Table 5 presents multivariate analyses predicting children’s kindergarten readiness by developmental ecology. I estimated linear regression models for continuous outcomes (early reading and math and teacher-reported behavior; parent-reported behavior was analyzed supplementally), and binary logistic regressions predicted the likelihoods of being in very good or excellent health and having received an asthma diagnosis. The bivariate Model 1 for each outcome found that children in all other classes had significantly compromised outcomes in every domain compared with “advantaged” children. The “vulnerable” class had by far the most compromised outcomes for every measure except health status (p < .05 in pairwise comparisons to every other class). Children from the “nontraditional” class had significantly higher reading and math scores and higher odds of very good health than those from the “traditional” class, but their teacher-reported behavior scores were significantly lower and their odds of asthma significantly higher than those of the “traditional” class.
Each of these relationships was somewhat reduced by the introduction of controls in Model 2, but very few lost significance. These findings support Hypothesis 2. The consistent developmental advantages experienced by children in the “advantaged” class and nearly consistent disadvantages for the “vulnerable” class are evident, as is the variable pattern of disparities between children in the “traditional” and “nontraditional” classes. The consistency of the findings across the cognitive, behavioral, and health domains is apparent.
The magnitudes of these differences in Model 1 of Table 5 are noteworthy when compared with major sociodemographic predictors of school readiness: developmental ecology classes mattered much more than these sociodemographic factors for children’s cognitive and behavioral outcomes and about as much for health. The leftmost sets of bars in Fig. 2 display comparable coefficients and odds ratios from bivariate analyses for several major sociodemographic factors that have strong relationships with school readiness: teen parenthood, socioeconomic background, race/ethnicity (here, black and Hispanic compared with white), and maternal nativity. The leftmost bars in Fig. 2 can be compared directly with the coefficients/odds ratios in Table 5, Model 1. The gap between the “advantaged” and “vulnerable” developmental ecology classes was 0.85 of a standard deviation for kindergarten reading scores in Table 5, Model 1, compared with less than 0.50 of a standard deviation for teen parenthood, socioeconomic background, race/ethnicity, and nativity. Although the sociodemographic disparities were slightly higher for math scores, they were still considerably smaller than the 0.92 of a standard deviation developmental ecology disparity. The developmental ecology gap of 0.57 of a standard deviation for teacher-reported behavior was double or more the coefficient size for the sociodemographic variables. Developmental ecology indicators alone explained 12 % of the variation in children’s kindergarten reading scores, 15 % for math, and 5 % for teacher-reported behavior.
Although comparisons of effect sizes across logistic regressions should be approached with caution, these results suggest that developmental ecology classes were about as important for children’s health as major sociodemographic factors. In bivariate models, children from the “vulnerable” and “traditional” classes were one-third as likely as the “advantaged” to be in very good health—a relationship that was similar to that between health status and Hispanic ethnicity and stronger than those between health status and other demographic factors. For the odds of having been diagnosed with asthma, the nearly threefold difference between the “vulnerable” and “advantaged” classes was similar to the black-white disparity and much stronger than other sociodemographic relationships. This operationalization of developmental ecology as a four-category variable was at least as powerful (and usually more so) for understanding kindergarten readiness as socioeconomic background, race/ethnicity, nativity, and teen parenthood.
Because of its parsimony (strong prediction of outcomes based on few categories), this classification of developmental ecology may be preferable to more complicated operationalizations. Supplemental analyses found that the proportion of variance explained in Model 2 of Table 5 was not much higher when the four-category measure was replaced with the 16 first-order latent classes or the 41 individual developmental ecology factors, and the four-category measure greatly reduced multicollinearity. Perhaps the most important advantage of the latent class approach, however, is that a specific factor can vary in its implications for development and health depending on the cluster of developmental ecology factors in which it is embedded. Without introducing very complicated and statistically infeasible interactions, this nuance would not be possible in a 41-factor multivariate model.
Hypothesis 3: Developmental Ecology Is a Primary Pathway Through Which Major Demographic Dividing Lines Shape Kindergarten Readiness
Measuring two developmental ecology domains at a single time point made the analyses of developmental ecology as a mediator of sociodemographic effects conservative: the explanatory power of developmental ecology when tracked across early childhood is likely to be higher. Even so, findings strongly supported Hypothesis 3. Figure 2 presents results from separate multivariate models for each demographic measure that first included only the sociodemographic factor in question; then added all controls used in Table 5, Model 2; and then introduced the four developmental ecology classes. Only statistically significant coefficients are displayed. These findings, in combination with results presenting the relationships between each sociodemographic factor and developmental ecology indicators in Table 4, fulfill Baron and Kenny’s (1986) mediation requirements.
Significant sociodemographic effects to be mediated are represented by the middle bars in Fig. 2, which show coefficients/odds ratios from models that included the relevant sociodemographic measure and all controls. The rightmost bars display significant coefficients/odds ratios after developmental ecology classes are included as mediators of the relevant sociodemographic measure. If there is no middle bar, there was no significant relationship between the sociodemographic factor and the outcome after controls were included. If there is a middle bar but no rightmost bar, developmental ecology fully mediated that sociodemographic factor. If the rightmost bar is smaller than the middle bar, the sociodemographic factor was partially mediated by developmental ecology.
Figure 2 shows that for every sociodemographic factor and every child outcome, net sociodemographic relationships represented by the middle bars were partially or fully mediated by developmental ecology classes in the rightmost set of bars. For teen parenthood, significant net relationships with early reading and teacher-reported behavior scores were fully mediated by developmental ecology, and most of the relationship with math scores was mediated. For mother’s welfare background, significant associations with children’s reading and math scores and asthma odds were fully mediated by developmental ecology, as was the relationship between the mother repeating a grade and teacher-reported behavior scores. The relationships between mother’s grade repetition and reading, math, and parent-reported behavior scores and odds of very good health were partially mediated by developmental ecology. Developmental ecology indicators fully mediated the black-white gap in reading and teacher-reported behavior scores and partially mediated the relationships with math scores, general health, and asthma diagnosis, as well as uncovering a previously suppressed positive relationship between black race and parent-reported behavior. The only fully mediated Hispanic-white gap was for teacher-reported behavior, but relationships with reading and math scores and health status were partially mediated. Developmental ecology indicators did little to mediate disparities between children of foreign-born mothers and others in parent-reported behavior and health status. Children of foreign-born mothers experienced a persistent advantage in asthma diagnosis odds.
This study builds on previous theories to articulate and operationalize the concept of developmental ecology, or interrelated features of children’s proximal social environments. Four starkly differentiated developmental ecology categories emerged from second-order LCA. Socioeconomic background, race/ethnicity, maternal nativity, and teen parenthood were strongly related to children’s developmental ecology. Developmental ecology classes predicted each of the six kindergarten outcomes. The strength of these relationships was as great as—and often substantially greater than—the associations between the major sociodemographic predictors and kindergarten readiness. Developmental ecology classes were a primary mediating pathway through which race/ethnicity, socioeconomic background, and teen parenthood influenced school readiness.
This study found notably consistent relationships between the four developmental ecology classes and cognitive, behavioral, and health outcomes. Disparate developmental domains were shaped by the same developmental ecology classes, supporting the promise of broad theoretical approaches to understanding child development. Yet, the sometimes different implications of developmental ecology classes even within domains—such as differences in health versus asthma rankings for the “traditional” versus “nontraditional” classes—simultaneously support the value of more specific research foci.
Research informing policy often isolates a single factor (or “policy lever”) in a child’s environment and studies its average effect on outcomes. This study suggests that such approaches may provide imperfect and inefficient solutions. Analyzing separate variables, or even interactions of two or three variables, does not capture the interrelatedness of many factors that is part of the developmental ecology construct. Examining how these factors cluster empirically clarifies that specific policy levers can have very different implications depending on the other factors with which they co-occur. For example, single motherhood is a common target for policies such as marriage promotion. Nesting union status within a broader developmental ecology complicates the standard story about the negative consequences of single motherhood: single motherhood has strikingly different implications depending on the developmental ecology in which it is embedded. Both the “nontraditional” and “vulnerable” classes were overwhelmingly composed of single-mother families, but the former group’s development was usually close to the national average. Thus, narrow policies aimed at reducing single motherhood miss the diversity of environments experienced by these children and may not be effective. In another example, repeated ecological changes were not always detrimental to children and sometimes predicted favorable kindergarten readiness depending on the developmental ecology in which they were embedded.
Developmental ecology is a major pathway through which sociodemographic factors such as race/ethnicity and socioeconomic background shape children’s development and health at kindergarten start. However, as the gray box in Fig. 1 shows, proximal unobserved mechanisms link developmental ecology to child outcomes. These mechanisms still need to be articulated. Guo and Harris (2000) found that cognitive stimulation and parenting style mediated the relationship between poverty and children’s intellectual development. These factors may also be shaped by developmental ecology. Alternatively, developmental ecology may influence early health, behavior, and cognition, which could then translate into later kindergarten readiness through cumulative disadvantage processes (Bast and Reitsma 1998; DiPrete and Eirich 2006). Aspects of a child’s personality or social interactions, perhaps interacting with genetic propensities, could also mediate the relationship between developmental ecology and school readiness. Future research necessitates increased cooperation among demographers, sociologists, behavioral geneticists, and psychologists to document the mechanisms through which distal processes influence more proximal ones to shape child development and health.
Developmental ecology may have interesting implications for social policy. Some factors that shape kindergarten readiness are more policy amenable than others. Changing a child’s socioeconomic background or race/ethnicity is not possible, but intervening in her developmental ecology is feasible. Key resources can be supplemented, health risks can be mitigated, and policies can support stability in children’s social environments. Evidence suggests that social programs can improve children’s health and school readiness (e.g., Barham 2012; Gertler 1994). Randomized interventions should build on this study’s findings to target interrelated clusters of factors that put children at risk. Some policies have already started to modify risk environments rather than specific risk factors. Beyond preschool, Head Start has wraparound offerings for parents, such as health care and employment services. Home nurse visitation programs for young mothers address multiple concerns in early life. And childcare subsidies, a seemingly narrow policy effort, simultaneously address children’s early educational needs and mothers’ employment and educational needs. A developmental ecology perspective implies that we should build on such efforts.
This study has laid out an initial conceptualization of developmental ecology and documented its empirical strength, but further theoretical development and operationalization are needed. There are likely to be other useful developmental ecology domains. Other promising aspects of developmental ecology that were not captured in the ECLS-B include resources outside the home but in children’s proximal settings (such as childcare providers and neighborhoods), physical features and health risks in childcare and neighborhood settings, children’s structural opportunities to interact with different people, and repeated changes in all these factors. New developmental ecology domains are added with the start of formal schooling, and it would be useful to study their interface with preexisting factors. Other time-related dynamics of children’s proximal environments should be studied besides the frequency of change; aspects of children’s developmental ecology may have different implications for school readiness depending on their timing. Or as researchers have found for poverty (Duncan et al. 1994; NICHD Early Child Care Research Network 2005), the duration of exposure to aspects of developmental ecology may have important implications beyond their presence at any given time point. Studying developmental ecology longitudinally would advance our knowledge. Newer data and methods may facilitate these kinds of investigations in the future, making early childhood an exciting area for both theoretical development and empirical analysis.
This research is based on work supported by grants from the National Science Foundation (SES 1061058 and SES 1423524). Research funds were also provided by the NIH/NICHD–funded CU Population Center (P2C HD066613). I thank Richard Jessor, Paula Fomby, Justine Tinkler, Laurie James-Hawkins, Elizabeth Lawrence, Joshua Goode, and Jeff Dennis for their contributions to this study.
To meet ECLS-B confidentiality requirements, all Ns have been rounded to the nearest 50.
Approximately 700 children received one or more imputed values for all analyses except those of teacher-reported behavior, for which an additional 1,500 children received imputed values.
I chose this wave because the measures of health risks were more comprehensive than in earlier waves, and resource measures were similar to those in earlier waves.
These included the Test of Early Mathematics Ability-3, the Peabody Picture Vocabulary Test, the Preschool Comprehensive Test of Phonological and Print Processing, the PreLAS 2000, and the sister study Early Childhood Longitudinal Study-Kindergarten Cohort (ECLS-K). See Snow and colleagues (2009) for more information on the kindergarten assessments.
Because this question was not asked at kindergarten start for the minority of children who started kindergarten in 2007, the 2006 indicator of asthma was used.
The lowest absolute BIC solution was 10 classes for resources, 8 classes for health risks, and 7 classes for ecological change, but these solutions often generated very small classes with little incremental improvement in fit from the selected solutions.