Developing Health Lifestyle Pathways and Social Inequalities across Early Childhood

Abstract

Lifestyles are a long-theorized aspect of social inequalities that root individual behaviors in social group differences. Although the health lifestyle construct is an important advance for understanding social inequalities and health behaviors, research has not theorized or investigated the longitudinal development of health lifestyles from infancy through the transition to school. This study documented children’s longitudinal health lifestyle pathways, articulated and tested a theoretical framework of health lifestyle development in early life, and assessed associations with kindergarten cognition, socioemotional behavior, and health. Latent class analyses identified health lifestyle pathways using the US Early Childhood Longitudinal Study – Birth Cohort (ECLS-B; N≈6,550). Children’s health lifestyle pathways were complex, combining healthier and unhealthier behaviors and changing with age. Social background prior to birth was associated with health lifestyle pathways, as were parents’ resources, health behaviors, and non-health-focused parenting. Developing health lifestyle pathways were related to kindergarten cognition, behavior, and health net of social background and other parent influences. Thus, family context is important for the development of complex health lifestyle pathways across early childhood, which have implications for school preparedness and thus for social inequalities and well-being throughout life. Developing health lifestyles both reflect and reproduce social inequalities across generations.

Introduction

As specific health behaviors have proven resistant to change, the concept of health lifestyles has gained currency among researchers. Embedded in social group-based identities, norms, and structural influences (Christensen and Carpiano 2014; Cockerham 2005), health lifestyles give rise to constellations of health behaviors within individuals. In this perspective, seeking to change a single behavior without targeting the underlying health lifestyle may not be effective. Most health lifestyles research has focused on adulthood, but early life is important for establishing behaviors that both persist into adulthood and have distinct implications for later life health regardless of adult health behaviors. Early childhood is also an important focal point for understanding how social advantage may be transmitted intergenerationally through the formation of lifestyles during a life phase when parents are highly influential but children are developing agency over some health behaviors.

The sole extant health lifestyle study in early childhood has found that parents shape health lifestyles at age 4 and these lifestyles have implications for early health and development (Mollborn, James-Hawkins, Lawrence, and Fomby 2014). Across middle childhood and early adolescence, earlier health lifestyles shape later ones in a dynamic but cumulative process, and family influences are joined by peers and schools (Mollborn and Lawrence 2018). Adolescent and young adult health lifestyles have implications for health outcomes in early adulthood (Burdette, Needham, Taylor, and Hill 2017; Lawrence, Mollborn, and Hummer 2017).

What is missing is a longitudinal investigation of health lifestyle development across the earliest years, when children leave the womb and slowly gain independence over their health behaviors, embedded in social contexts and monitored by adults, before making the transition to school. Our study is the first to model health lifestyle development across this period. Our conceptual framework of family influences on early childhood health lifestyle development acknowledges the interplay between structure and agency (Cockerham 2005) but focuses on structure, as recent theorizing has suggested is needed (Cockerham 2013). We coin the term “developing health lifestyle pathways” to characterize longitudinal measures of health lifestyles across time points. These pathways may be associated with children’s kindergarten readiness in multiple domains, which has fundamental implications for individuals’ lives.

This study addresses three research questions: (1) What are the complexities in developing health lifestyle pathways—both con/discordance in the healthfulness of a child’s different behaviors, and longitudinal con/discordance in healthfulness across age? (2) How do families influence children’s developing health lifestyle pathways? (3) Do developing health lifestyle pathways matter for children’s kindergarten readiness, net offamily factors?

Theory and background

Health lifestyles and early childhood

Although large literatures have examined social influences on health behaviors (Pampel, Krueger, and Denney 2010; Short and Mollbom 2015) and their links to later life outcomes (Glewwe, Jacoby, and King 2001; Lansford et al., 2002; Lansford et al. 2007), empirical tests of health lifestyle theories to better understand how behaviors are interrelated and undergirded by social phenomena are relatively new. The idea of lifestyles as collectively based everyday manifestations of structural inequalities, with social categories such as class shaping the lifestyles available to people, goes back as far as Max Weber (1978) and has been further developed by other theorists (e.g., Abel 1991; Sobel 1981). Lifestyles are an interplay between structure and agency because individuals often choose among multiple lifestyle options available to their social category. Health lifestyles, as developed by Cockerham (2005, 2013) and others (e.g., Frohlich, Corin, and Potvin 2001; Korp 2008), refer to constellations of health behaviors that are interrelated because of underlying social identities, normative processes, and shared understandings of health. For example, a high school athlete may have a team-based health lifestyle that goes beyond health behaviors specific to their sport to encompass discordantly healthful patterns of substance use, diet, sleep, and sexual activity. The group-based identity underlying this lifestyle may result in an individual’s behaviors resisting change unless group norms change or the person changes groups. The health lifestyles literature among adults is increasingly extensive (e.g., de la Flaye, DAmico, Miles, Ewing, and Tucker 2014; Freeh 2012; Saint Onge and Krueger 2017; Sinha 1992; Stefansdottir and Vilhjalmsson 2007).

Our study advances prior work by conceptualizing and assessing the longitudinal development of health lifestyles in early childhood and the intergenerational factors that shape them. Flealth lifestyles in childhood are uniquely important for understanding the intergenerational transmission of health behaviors and social advantages and disadvantages (Mollborn and Lawrence 2018). Fiealth lifestyle research on adults focuses on “achieved” health lifestyles, conceptualized as a blend of individual agency and structural constraints (Cockerham 2005; Mollborn et al. 2014). But through what processes do people adopt these health lifestyles? Children mostly start with a “received” health lifestyle over which parents and adults have an outsized influence—yet even then, an infant may refuse to sleep, eat, or play despite the caregiver’s best efforts. Previous research on early childhood health lifestyles has conceptualized them as a mix of child health behaviors that are controlled by others (e.g., secondhand smoke exposure, infant sleep position, or physical activity during school) and those controlled by the child, a caregiver, or a combination of both (e.g., screen time, sugar or produce consumption, or wearing a helmet; Mollborn et al. 2014). Understanding how social contexts shape the development of children’s health behaviors is important to health theorists (Christensen 2004). Children may learn behaviors directly from parents or be exposed to social contexts that encourage similar behaviors to those of their parents (De Genna et al. 2006). Conceptualizing early childhood health lifestyles in this way reflects current theorizing on the human life course, which emphasizes that researchers should avoid strict distinctions between individual agency and the influence of those with whom a person’s lives are linked—in fact, agency and linked lives are intrinsically interrelated, and the agency of people who are not considered to be “developmentally normal [adults]” is still important to theorize and study (Landes and Settersten 2019).

Because health lifestyles produce distinctions between social classes that benefit elites, parents and children may be motivated to engage in certain health behaviors in order to enact their social group memberships (Messner 2000) rather than because of the behaviors’ health benefits alone. These practices of distinction likely apply to children as well as adults, as evidenced by research on topics such as advantaged parents’ selection of child care providers (Augustine, Cavanagh, and Crosnoe 2009) and summer activities (Chin and Phillips 2004) and selective use of off-label medication to enhance children’s academic achievement (King, Jennings, and Fletcher 2014). Yet not all health lifestyle development is deliberate. Like adult health lifestyles, children’s health lifestyles are not always shaped consciously by parents or children, as health behaviors become an embodied aspect of people’s habitus (Bourdieu 1986; Cockerham 2013). Young children learn many health behaviors, such as hygiene practices, diet, and health care seeking, that they will engage in for the rest of their lives. This makes childhood a particularly important focus for health lifestyles research.

Complexities in children’s developing health lifestyle pathways

Previous research on health lifestyles has often found that they are discordant across dimensions within individuals—healthful and unhealthful behaviors coexist in the same person at one time point and across time (e.g., Burdette et al. 2017; Mollborn and Lawrence 2018; Saint Onge and Krueger 2017). Thus, prevalent health lifestyles in a population cannot be arrayed along a unidimensional continuum of healthfulness. Here, we articulate potential reasons why social and developmental contexts are likely to yield discordance in health lifestyles.¹

A person’s health behaviors can be discordant for a variety of potential reasons. First, group-based identities underlying a health lifestyle may encourage some behaviors and discourage others for reasons unrelated to healthfulness, such as group norms (Cockerham 2005). A young child who plays on a soccer team may get lots of exercise but also eat ice cream after every match. Second, children’s preferences arising from these identities may differ from adult caregivers’ preferences, and when child and adult preferences do not match, the degree of adult monitoring will likely be important for the resulting health behaviors of the child. Third, norms and routines in nonparental care settings such as childcare or preschool can shape health behaviors. Fourth, some health behaviors (e.g., health care seeking, consumption of produce, participation in athletics and dance) are costlier for families than others (e.g., playing outdoors, consumption of inexpensive snack foods). Fifth, widespread racial neighborhood segregation in the United States (Massey and Denton 1993) can lead to discrimination and violence that can make neighborhoods differentially available to young children for outdoor play and can shape adult monitoring. These varied contexts, along with other circumstances, support a mix of healthier and unhealthier behaviors within children.

Another type of discordance within an individual’s developing health lifestyle pathway can arise from changes over time. Capturing longitudinal dynamics in young children’s health and health behaviors can be challenging but is important for understanding later life (Goosby, Cheadle, and McDade 2016). Children’s early health behaviors likely shape their later ones in a partially path-dependent process through the development of habitus (Bourdieu 1986; DiPrete and Eirich 2006; Montez and Hayward 2011). Yet health lifestyles presumably also change as a child ages. Although earlier lifestyles probably matter for later ones, we expect new influences to arise as children develop and gain agency and their social contexts and identities change. For example, young children increasingly receive nonparental care with age (NICHD Early Child Care Research Network 2005), and nonparental care providers shape child health behaviors including feeding, physical activity, and sleep. Group care contexts such as preschool also facilitate peer influences. Corsaro (2003) documented preschoolers’ elaborate and influential peer cultures, and three year olds acknowledge and enforce peer group norms (Koymen et al. 2014). Extrafamilial contexts in early childhood are particularly tied to parents’ willingness or ability to expend financial and other resources for child care or enrichment activities. Increased mobility and maturity also facilitate engagement with extracurricular activities, health care providers, and neighborhood contexts. Because measuring a health lifestyle at a single time point in early childhood misses other time points when development, social contexts, agency, and identities may be starkly different, the influence of that cross-sectional health lifestyle on the child will likely be smaller than the influence of a longitudinally measured health lifestyle. In other words, cumulative longitudinal pathways capturing health lifestyles at multiple points in time should be more strongly related to kindergarten readiness than a single time point. Measuring longitudinal health lifestyles across early childhood is a challenge because children’s health behaviors differ substantially between infancy and kindergarten age; our study puts forth one potential analytical strategy.

Family influences on developing health lifestyles

Although there is substantial change across early childhood in children’s developing health lifestyles, we expect intergenerational processes to shape these lifestyles early on and throughout this period. Intergenerational transmission processes occur via family factors (e.g., resources and modeling of behaviors) and the continuity of social contexts (e.g., neighborhoods or school quality) across generations (De Genna, Stack, Serbin, Ledingham, and Schwartzman 2006). We examine the familial intergenerational processes that shape children’s developing health lifestyles. Parents can shape children’s health behaviors by facilitating or impeding behaviors, modeling behaviors, and transmitting beliefs and attitudes towards behaviors (Lau, Quadrel, and Hartman 1990; Patrick and Nicklas 2005; Salvy, de la Haye, Bowker, and Hermans 2012). Although we focus on parents here, nonparental care providers may also shape children’s health lifestyles through similar mechanisms.

Parents’ social background—by which we mean sociodemographic characteristics preceding a child’s birth—is expected to influence children’s health lifestyles by structuring “life chances” (Cockerham 2005). Race/ethnicity, nativity, parents’ childhood socioeconomic status, and teen parenthood status measure social background, which affects health and status attainment in the next generation (Barr 2015; Geronimus and Korenman 1993; Levine, Pollack, and Comfort 2001; Park and Myers 2010). These factors represent social group memberships, which are associated with health lifestyles (e.g., Saint Onge and Krueger 2017).

Other parent-related processes also matter for children’s health lifestyles, and they often co-occur and can compound and accumulate in their effects. The first is parents’ resources during the child’s first years of life. Socially advantaged people have more resources and are better able to deploy them for parenting and health improvement (Lareau 2011; Link and Phelan 1995). We expect parents’ resources to shape children’s developing health lifestyles (Augustine, Prickett, and Kimbro 2017; Prickett and Augustine 2016).

Second, parents’ health behaviors may be associated with their children’s lifestyles. There are intergenerational similarities in health behaviors (Abella and Heslin 1984; Wickrama, Conger, and Elder 1999), and parents’ health behaviors are associated with children’s life outcomes (Bauldry et al. 2016). Many health behaviors that comprise early health lifestyles are either directly role modeled by parents—such as diet and sleep—or carried out by parents but impinge on the child’s body—such as secondhand smoke exposure. Role modeling is an important aspect of socialization into behaviors (De Genna et al. 2006). Parents may model behaviors that they desire for their children because they see children’s behaviors as manifestations of their own identities (Lee 2008; McCarthy, Edwards, and Gillies 2000).

Third, non-health-focused parenting practices may shape children’s health lifestyles. Parenting practices are heavily shaped by parents’ social contexts and background (Davison, Jurkowski, and Lawson 2013; Lareau 2011; Prickett and Augustine 2016) and influence children’s health (Augustine et al. 2017). Socially advantaged parents engage in “intensive parenting” (Hays 1996) to maximize children’s future prospects, and disadvantaged parents experience structural constraints on their parenting efforts (Lareau 2011). For example, an intensive parenting style involves high levels of parental monitoring and close attention to coaching children for interactions with adults such as health care providers (Lareau 2011). These seemingly non-health-focused parenting practices should result in different child health behaviors and different understandings of appropriate health lifestyles.

Measuring health lifestyles and kindergarten readiness

A flurry of recent research has applied latent class analyses to operationalize health lifestyles, mostly among adults (e.g., De Vries et al. 2008; Krueger, Bhaloo, and Rosenau 2009; Laska, Pasch, Lust, Story, and Ehlinger 2009; Sodergren et al. 2014). Latent class analyses have gained in popularity by allowing predominant, inherently categorical profiles of health behaviors within individuals to arise from the data. Latent class analyses can also model predictors and consequences of profiles (McAloney, Graham, Law, and Platt 2013). Some studies of other life stages have measured health lifestyles at multiple time points (Burdette et al. 2017; Lawrence et al. 2017; Mollborn and Lawrence 2018), but combined longitudinal representations of health lifestyles are rare (excepting Daw, Margolis, and Wright 2017). One study analyzed health lifestyles at age 4½ predicting development a year later (Mollborn et al. 2014), but none has captured early childhood health lifestyles longitudinally.

Recent research has established associations between people’s health lifestyles and their health, mortality, school readiness, and school achievement in different stages of the life course (Burdette et al. 2017; Mollborn et al. 2014; Mollborn and Lawrence 2018; Saint Onge and Krueger 2017). For young children, health behaviors in areas such as diet, safety, sleep, and physical activity have wide-ranging implications for socioemotional behavior, cognition, and health. Considering these behaviors together in a single lifestyle may be even more powerful for understanding their effects (Saint Onge and Krueger 2017). We therefore expect children’s developing health lifestyle pathways to shape their cognitive, behavioral, and health-related kindergarten readiness, which is important because of its substantial implications for later health and socioeconomic attainment (Entwisle, Alexander, and Olson 2004; Palloni 2006; Weller, Schnittjer, and Tuten 1992). These domains of kindergarten readiness are intertwined, building on each other to shape later development and academic achievement. Evidence that health lifestyle pathways shape kindergarten readiness would suggest that they play a role in intergenerational transmission of health and social advantage within families.

Summary

Building on existing theory and findings, we expect to find three empirical patterns in our data that reflect our research questions. These expectations are not falsifiable hypotheses because in some cases, statistical significance tests are not applied. First, we expect that children’s developing health lifestyle pathways will be discordant in the healthfulness of their behaviors, both within a given time point and over time. Second, we expect that parents’ social background, resources, health behaviors, and non-health-focused parenting practices will be associated with children’s developing health lifestyle pathways. Third, we expect that more healthful developing health lifestyle pathways will be related to higher kindergarten readiness, net of sociodemographic and parental factors.

Data and Methods

Data

We used the Early Childhood Longitudinal Study-Birth Cohort (ECLS-B; U.S. Department of Education, 2007). Sampling weights at each wave adjust results to be nationally representative of the US birth cohort of 2001. A clustered, list frame design obtained a probability sample of births (see Snow et al. 2009 for additional sample design details). The ECLS-B first collected parent interviews and child assessments when infants were ten months old (Wave 1) then followed up at about ages 2 (Wave 2), 4 (Wave 3), and 5 (Wave 4). Children who had not yet started kindergarten participated in Wave 5. We used information from either Wave 4 or 5, whenever the child was enrolled in kindergarten (called Wave K). Weighted response rates were 74%, 93%, 91%, 92%, and 93% for parent interviews at each wave, respectively. Most children lived with a biological parent, and the responding “primary” parent was usually the mother. The Wave 1 sample included about 10,200 children with valid sampling weights, Wave 2 about 8,950, Wave 3 approximately 8,000, and Wave K 6,550 (dataset restrictions required rounding all reported Ns to the nearest 50). For analyses that examined an individual wave, we used all available observations for that wave to best represent nationally representative health lifestyle classes at that particular age. Second-order latent class analyses and analyses predicting kindergarten outcomes were necessarily limited to the 6,550 children who participated in Wave K, with sampling weights adjusting for nonresponse and attrition across waves.

Measures

Health lifestyles.

For each wave, we used a set of age-appropriate measures of children’s health behaviors to inform health lifestyles. Indicators differed by wave because of rapidly changing early childhood development. We included all available health behaviors at each wave, defined as practices or exposures in the child’s body. This operationalization focuses on the child and does not attempt to adjudicate whether, for example, the number of hours that a child sleeps is under the control of the parent who sets a bedtime versus the child who wakes up before the parent would prefer. Attempting such adjudication would be impossible with these survey data, introducing potential bias based on assumptions we would make about child agency that might or might not be correct. Our definition is in line with recent life course theorizing that argues that attempting to separate a person’s agency from the influences of the people with whom their life is linked is misguided (Fandes and Settersten 2019). Our definition, which follows Mollbom and colleagues (2014), thus includes all available measures of child health behaviors without making decisions on the basis of the child’s control or caregiver’s intent. Measured behaviors fall under the categories of diet/nutrition, sleep, safety, physical activity, exposure to substances, and health care. Other people’s health behaviors and measures that do not impinge directly on the child’s body are excluded. Thus, for example, we include a child’s experience of food insecurity as a health lifestyle indicator of a child’s diet, but household receipt of SNAP is not because we do not know whether it resulted changes to the child’s diet.

Wave 1 (age 10 months) behaviors included four items capturing diet/nutrition, two sleep, one substance exposure, and one physical activity (Table 1 lists health lifestyle indicators by wave). Dichotomous measures indicated the child’s fulfilment of age-appropriate nutritional recommendations: drinking breast milk for six or more months, not going to bed with a bottle containing a liquid other than water, and not eating solid foods before four months old. We also included a dichotomous measure of child food insecurity in the household, which impacts feeding practices. Two measures captured recommended sleep positions: if the child consistently slept on her back as a newborn, then at 3 months old. A dichotomous measure represented the child’s exposure to secondhand smoke inside the home. Four categories represented how often the child played outdoors: never to a few times per month; a few times per week; once per day; or more than once per day.

Table 1.

Weighted Health Lifestyle Indicator Means across Second-Order Latent Classes

	Overall sample	Healthful concordant	Healthful but TV/outdoors/junk food	Discordant	Unfavorable concordant	Unfavorable low cost
Population share		19%	17%	30%	17%	17%
Wave 1
Bottle to bed (not water)°	0.27	0.11	0.14	0.71	0.59	0.57
Breastmilk for 6+ months°	0.32	0.57	0.45	0.29	0.11	0.15
Solid food prior to 4 months°	0.23	0.10	0.16	0.23	0.38	0.34
Child food insecurity°	0.11	0.03	0.04	0.11	0.17	0.20
Sleep position on back (newborn)°	0.59	0.77	0.74	0.53	0.47	0.46
Sleep position on back (3 months)°	0.57	0.72	0.72	0.52	0.44	0.44
Someone smokes in home°	0.11	0.00	0.01	0.11	0.20	0.25
Frequency of outside play
Never/rarely/few times per month	0.23	0.17	0.22	0.26	0.27	0.22
Few times per week	0.30	0.32	0.32	0.31	0.27*	0.26 *
Once per day	0.26	0.30 *	0.28	0.24*	0.26	0.26
More than once per day	0.21	0.21	0.17	0.19	0.20	0.26
Wave 2
Safe seat (car/backseat)°	0.95	0.99	0.99	0.96	0.87	0.92
Regular soda consumption°	0.26	0.06	0.13	0.29	0.40	0.40
Bottle to bed°	0.17	0.10	0.08	0.20	0.17	0.28
Child food insecurity°	0.09	0.02	0.04	0.11	0.15	0.15
Someone smokes in home°	0.10	0.01	0.00	0.09	0.22	0.25
Spanking°	0.45	0.25	0.37	0.46	0.71	0.51
Regular dinner
0-4 days per week	0.24	0.11	0.16	0.24	0.44	0.28
5-6	0.22	0.26	0.29	0.21	0.19	0.16
7	0.54	0.63	0.55	0.55	0.37	0.56
Frequency of outside play
Never/rarely/few times per month	0.18	0.16	0.18	0.19	0.21	0.16
Few times/week-once/day●	0.60	0.60	0.59	0.60	0.60	0.61
More than once per day●	0.22	0.23	0.23	0.20	0.20	0.23
Average TV watching per day
0-<1 hour	0.15	0.25	0.10	0.15	0.13	0.13
1-<2	0.33	0.40 *	0.40 *	0.32	0.28*	0.22 *
2-<3	0.24	0.24	0.32 *	0.23	0.23	0.21 *
≥3	0.27	0.11	0.17	0.30	0.36	0.44
Wave 3
Child food insecurity°	0.13	0.01	0.03	0.15	0.20	0.27
Someone smokes in home°	0.12	0.00	0.01	0.12	0.20	0.27
Athletics/dance°	0.36	0.64	0.57	0.24	0.29	0.15
Insufficient sleep/late bedtime°	0.23	0.12	0.13	0.23	0.40	0.30
Safe seat°	0.79	0.95	0.85	0.83	0.59	0.68
Spanking°	0.34	0.11	0.29	0.37	0.53	0.41
Victim of violence°	0.02	0.00	0.01	0.02	0.02	0.04
Always wears helmet°	0.47	0.68	0.46	0.46	0.30	0.39
Dentist/doctor within recommendations°	0.57	0.70	0.65	0.53	0.55	0.48
Regular dinner
0-4 days per week	0.34	0.19	0.32	0.32	0.56	0.40
5-6	0.31	0.41 *	0.37*	0.31*	0.24*	0.22 *
7	0.34	0.41	0.31	0.37	0.20	0.37
Frequency of outside play
Never/rarely/few times per month	0.15	0.10	0.14	0.18	0.19	0.14
Few times/week-once/day	0.68	0.75 *	0.70	0.67	0.67	0.61 *
More than once per day	0.17	0.15	0.16	0.14	0.14	0.25
Average TV watching per day
0-<1 hours	0.03	0.08	0.01	0.02	0.01	0.01
1-<2	0.10	0.20 *	0.10	0.10	0.05 *	0.05 *
2-<3	0.26	0.44 *	0.26	0.24	0.18*	0.15 *
≥3	0.61	0.28	0.64	0.64	0.76	0.78
Computer use
Never	0.45	0.24 *	0.30*	0.46	0.55*	0.69 *
1-2 times per week	0.34	0.51 *	0.45*	0.32	0.24*	0.19 *
≥3	0.21	0.25 *	0.25 *	0.22	0.21	0.12 *
Soda consumption
None	0.28	0.57	0.26	0.27	0.11	0.13
Less than once per day	0.42	0.38*	0.49 *	0.49 *	0.43	0.24 *
Once or more	0.30	0.05	0.25	0.24	0.46	0.63
Fast food consumption
None	0.24	0.40	0.13	0.26	0.11	0.23
Less than once per day	0.65	0.57*	0.75 *	0.68*	0.70*	0.55 *
Once or more	0.11	0.03	0.13	0.05	0.19	0.22
Fruit and vegetable once per day°	0.92	0.95	0.94	0.91	0.87	0.92
Milk consumption
None	0.15	0.13	0.12	0.14	0.20	0.16
Less than once per day	0.15	0.15	0.15	0.17 *	0.16	0.12 *
Once	0.31	0.30	0.31	0.34 *	0.27 *	0.27 *
Twice or more	0.40	0.42	0.41	0.35	0.37	0.46
Sweets once or more per day°	0.47	0.39	0.73	0.25	0.51	0.67
Salty snacks once or more per day°	0.31	0.09	0.52	0.13	0.41	0.62
Wave K
Child food insecurity°	0.12	0.01	0.02	0.13	0.17	0.24
Fruit and vegetable once per day°	0.91	0.94	0.95	0.91	0.83	0.94
Sweets once or more per day°	0.46	0.43	0.73	0.21	0.46	0.75
Salty snacks once or more per day°	0.31	0.18	0.43	0.11	0.32	0.70
Insufficient sleep/late bedtime°	0.22	0.12	0.13	0.23	0.37	0.27
Someone smokes in home°	0.11	0.00	0.02	0.10	0.18	0.27
Athletics/dance°	0.49	0.82	0.75	0.33	0.46	0.21
Safe seat°	0.71	0.90	0.79	0.72	0.40	0.65
Spanking°	0.24	0.09	0.16	0.21	0.45	0.34
Victim of violence°	0.02	0.01	0.01	0.02	0.04	0.01
Always wears helmet°	0.45	0.62	0.49	0.46	0.26	0.38
Dentist/doctor within recommendations°	0.84	0.92	0.92	0.82	0.80	0.76
Regular dinner
0-4 days per week	0.33	0.18	0.26	0.30	0.61	0.38
5-6	0.32	0.39*	0.42 *	0.31	0.23 *	0.24*
7	0.35	0.43	0.32	0.40	0.16	0.38
Average TV watching per day
0-<1 hours	0.03	0.08	0.01	0.02	0.01	0.01
1-<2	0.10	0.20 *	0.10	0.10	0.05 *	0.05 *
2-<3	0.26	0.44 *	0.26	0.24	0.18*	0.15 *
≥3	0.61	0.28	0.64	0.64	0.76	0.78
Computer use
Never	0.43	0.43	0.41	0.45	0.32 *	0.50 *
1-2 times per week	0.37	0.37	0.39	0.38	0.36	0.34 *
≥3	0.20	0.20	0.20	0.18*	0.32 *	0.16 *
Soda consumption
None	0.27	0.52	0.24	0.27	0.10	0.15
Less than once per day	0.46	0.42*	0.56 *	0.54*	0.52*	0.20 *
Once or more	0.27	0.05	0.20	0.18	0.38	0.65
Fast food consumption
None	0.27	0.38	0.17	0.28	0.15	0.30
Less than once per day	0.64	0.60*	0.76 *	0.67*	0.70*	0.48 *
Once or more	0.09	0.02	0.06	0.04	0.15	0.21
Milk consumption
None	0.16	0.12	0.12	0.14	0.29	0.13
Less than once per day●	0.17	0.18	0.16	0.17	0.15	0.16
Once	0.32	0.32	0.32	0.36 *	0.26 *	0.31
Twice or more	0.35	0.38	0.40	0.32	0.29	0.40
Physical education
<20 minutes per week	0.11	0.11	0.09	0.14	0.04	0.14
20-<60	0.38	0.44 *	0.44 *	0.36	0.31 *	0.36
60+	0.51	0.45	0.47	0.50	0.65	0.50

Source: ECLS-B

Notes: Accounts for complex survey design. N≈6,550.

^°l=yes, 0=no.

^●No means were significantly different from sample mean. Gray=significantly healthier than sample mean, red=significantly unhealthier than sample mean, white=not significantly different from sample mean for that indicator, highest/lowest values are bolded.

^*Significantly different from sample mean for rows not color coded because healthfulness of behavior is less well established.

Wave 2 (age 2) behaviors included four diet/nutrition measures, two safety, one substance exposure, and two physical activity. Wave 2 included similar measures of taking a bottle to bed and child food insecurity, and a new dichotomous measure of usually drinking soda or sweetened fruit drinks with meals or snacks. We measured dinner served at a regular time as 0-4, 5-6, or 7 days per week. Dichotomous variables captured whether the child consistently sat in a car seat and in the back seat if they rode in a car, and whether the child was spanked in the last two weeks. Secondhand smoke exposure and outdoor play were coded as in Wave 1. A variable captured daily average hours of television watching (none to less than one hour, one to less than two, two to less than three, or three or more).

Wave 3 behaviors (age 4½) included eight diet/nutrition measures, four safety, one substance exposure, one sleep, four physical activity, and one health care usage. Like earlier waves, we included child food insecurity and the number of days dinner was regularly served. Additional diet measures included average consumption of sugar-sweetened beverages (none, less than once per day, and once or more per day), fast food (none, less than once per day, and once or more per day), and milk (none, less than once per day, once per day, twice or more per day). A dichotomous measure captured whether the child ate a daily average of at least one fruit and one vegetable. Measures indicated whether the child ate a sweet snack (e.g., cookies, cake) or salty snack (e.g., potato chips, pretzels) once or more per day. Vehicle safety indicated whether the child was in the proper car seat or used a seat belt in the back seat for those four years and older. Spanking was included as in Wave 2. Safety measures captured whether the parent reported the child had ever been a victim of violence, and whether the child always wore a helmet if they bicycled or skateboarded. Physical activity included frequency of outside play and average television watching as in earlier waves, a dichotomous measure of participation in athletics or dance, and computer use (never, once or twice a week, three or more times per week). We measured inadequate sleep as not having a regular bedtime, bedtime after 10 pm, or sleeping less than 10 hours per night. Finally, a measure indicated whether the child both visited the doctor for a well-child visit and ever visited the dentist in the last year.

Wave K (age 5½) behaviors were measured as in Wave 3 with one exception. Because outdoor play was not available, we captured physical education (PE) time per week as reported by the kindergarten teacher (less than 20 minutes, 20-59, or 60 or more).

Parental influences.

Social background was measured in infancy (Wave 1) but referred to sociodemographic factors determined prior to birth: child race/ethnicity (mutually exclusive dichotomous measures of Non-Hispanic White, non-Flispanic Black, Flispanic, and other race/ethnicity including multiracial), maternal nativity (US or foreign-born), whether any parent was under age 20 at the child’s birth (yes/no), and a continuous measure of the (usually maternal) grandmother’s years of educational attainment. All parent resource measures were from Wave 1, in the child’s infancy. Socioeconomic status was represented by income-to-needs ratio (the ECLS-B-constructed ratio of household income to the year- and household size-specific poverty threshold), maternal education (years recoded from degrees), and a household assets scale (car, bank account, investments, home ownership, and nonsubsidized housing). Dichotomous variables captured food stamp (SNAP) receipt, receipt of parenting help or advice, and whether a father (biological or social) or a grandparent lived with the child. A variable measured the number of household members under the age of 18. We captured maternal paid work status as not working, part time (less than 40 hours), versus full time (40 or more). Parent health behavior measures came from Wave 1. Dichotomous variables represented maternal smoking and binge drinking (defined as four drinks in one sitting) in the last month. Although self-reported continuous body mass index (BMI) is not a health behavior, we used it as an imperfect proxy for maternal diet and exercise.

Non-health-focused parenting practice measures came from Wave 1. First, a constructed, continuous indicator represents the mother’s score on the Nursing Child Assessment Teaching Scale (NCATS), measuring parenting behaviors supporting cognitive and socioemotional growth. Second, an indicator represents an abbreviated, adapted version of the FIOME scale (see Snow et al. 2009 for more details), which combined interviewer-observed and parent-reported parent behaviors that create a positive learning environment for the child (averaged across ten behaviors). Third, a variable whether the parent responded that the statement, “You can spoil a tiny baby by picking him up every time he cries” as closest to her own ideas (as opposed to “You cannot spoil a tiny baby by picking him up every time he cries.”). Last, the Knowledge of Infant Development Inventory (KIDI) score reflects correct parental responses to 18 questions about child development. Greater values for the NCATS, HOME, and KIDI measures indicate more positive parenting, whereas a higher value on the “spoil” variable indicates less positive parenting.

Child outcomes.

For child well-being, we used five kindergarten measures: reading, math, parent-reported and teacher-reported socioemotional behavior, and parent-reported health status. Child assessments were conducted for early reading (ECLS-B reported reliability = 0.84) and math (ECLS-B reported reliability = 0.89), using standardized scale scores with higher values reflecting more favorable behavior. ECLS-B asked parents and teachers about the child’s behavior. The parent-reported behavior scale standardized responses to 25 items on a five-point scale (Cronbach’s alpha = 0.87), with 22 standardized items for the teacher scale (Cronbach’s alpha = 0.93). Health status compared children whose parent reported their general health as very good or excellent to good, fair, or poor.

Controls.

To adjust for potential selection bias, we also analyzed Wave 1 control variables that may be simultaneously related to predictors, health lifestyles, and kindergarten readiness: child age in months, gender (male/female), birth weight (very low < 1500 grams, low ≥ 1500 but < 2500 grams, or normal), birth order (e.g., 1 if the mother’s first child, 2 for second child), cognitive score, interviewer-rated behavior, parent-reported health (very good/excellent vs. good/fair/poor), and asthma diagnosis. Analyses of kindergarten readiness outcomes also included child age at the kindergarten wave and number of hours in nonparental care at Wave 3.

Analyses

We used latent class analysis (LCA) to identify health lifestyle classes at each wave. LCA uses a structural equation modeling approach to identify a categorical latent trait (health lifestyles) from observed variables. LCA assumes that the variables are independent after accounting for membership into the categorical latent trait. Fit statistics, including AIC, BIC, and G², and theoretical assessment of the meaning of the classes identify the best-fitting number of classes. For these results, we identified as best-fitting solutions those that represented the lowest BIC (which rewards parsimony), contained large enough latent classes to provide sufficient statistical power, had acceptable average posterior probabilities, and made theoretical sense (see Table 2 for fit statistics).

Table 2.

Fit Statistics for First-and Second-Order Latent Class

# classes		G2	BIC	AIC	# classes		G2	BIC	AIC
Wave 1 (N≈10200)	2	2249	2443	2291	Wave 2 (N≈8950)	2	2839	3085	2893
	3	1327	1623	1391		3	2675	3048	2757
	4	936	1333	1022		4	2546	3047	2656
	5	838	1337	946		5	2478	3106	2616
	6	763	1363	893		6	2411	3166	2577
	7	727	1429	879		7	2352	3235	2546
	8	689	1492	863		8	2295	3305	2517
	9	654	1559	850		9	2244	3381	2494
	10	626	1633	844		10	2193	3458	2471
	11	596	1704	836		11	2147	3539	2453
	12	571	1781	833		12	2107	3626	2441
Wave 3 (N≈8000)	2	73975	74488	74089	Wave K (N≈6550)	2	55293	55794	55407
	3	72463	73236	72635		3	53972	54728	54144
	4	72002	73035	72232		4	53552	54562	53782
	5	71633	72928	71921		5	53160	54425	53448
	6	71381	72936	71727		6	52855	54375	53201
	7	71175	72991	71579		7	52634	54409	53038
	8	70979	73055	71441		8	52445	54475	52907
	9	70799	73136	71319		9	52280	54564	52800
	10	70620	73218	71198		10	52127	54666	52705
	11	70459	73317	71095		11	51966	54760	52602
	12	70293	73412	70987		12	51827	54875	52521
Second order (N≈6550)	2	1782	2054	1844
	3	1458	1871	1552
	4	1218	1771	1344
	5	1050	1744	1208
	6	955	1790	1145
	7	868	1843	1090
	8	787	1903	1041
	9	722	1978	1008
	10	671	2068	989
	11	625	2162	975
	12	580	2258	962

Source: ECLS-B

Notes: Accounts for complex sampling design.

For each wave, we assigned each individual to the class for which they had the highest probability of membership (see Appendices A–C). We then used each child’s health lifestyles from each wave as indicators in a new, second-order LCA model (for examples of this approach, see Keller and Kempf 1997; Klingzell et al. 2016; Mollborn 2016; Roberson, Norona, Lenger, and Olmstead 2018; Roberson, Norona, Zorotovich, and Dirnberger 2017; Wetzel, Carstensen, and Bohnke 2013). Note that LCA has misclassification bias that is compounded in the second-order models, a limitation of this approach. The resulting classes thus represented predominant health lifestyle pathways across early childhood, or “developing health lifestyles.” In contrast to growth curve models or latent class growth analysis, this approach allows health lifestyles to emerge independently at each wave. This is particularly important given the differences in behaviors at each of the time points. We then assigned each child to the developing health lifestyle pathway in which she had the highest probability of membership. The mean probability of pathway assignment was 0.74. Studies find that this approach generally produces downwardly biased estimates when latent classes are used in subsequent models (Bolck, Croon, and Hagenaars 2004; Vermunt 2010), but methodological validation of our specific analyses is beyond the scope of this study. Given our wide range of covariates and multiple outcomes, a single-step approach to estimating classes and relationships was not feasible.

Descriptive statistics identified health lifestyle pathway-related patterns of sociodemographic traits and other characteristics collected at Wave 1. Most of the conceptual model was assessed using multinomial logistic regression models that predicted developing health lifestyle pathways using social background; controls; and parent resources, health behaviors, and non-health-focused parenting practices. Ordinary least squares (OLS) regression and logistic regression predicted child outcomes at the kindergarten wave based on developing health lifestyle pathways, controls, and other conceptual model components.

The cross-sectional LCA analyses retained all individuals with a valid weight for each wave, using full information maximum likelihood (FIML) to account for missingness. The longitudinal LCA creating developing health lifestyle pathways retained individuals with a valid kindergarten weight (TKT≈6,550) through FIML. The subsequent descriptive statistics and regression models used multiple imputation to retain the full sample. We used a Monte Carlo Markov Chain approach, creating 20 datasets. All dependent and independent variables informed imputation together with auxiliary variables: Wave 3 reading and math scores and teacher-reported behavior, and which year the child enrolled in kindergarten. An average of 2.8% of values were imputed. Wave-specific LCAs used that wave’s weights; all other analyses accounted for weights, clustering, and strata to make the results nationally representative for the kindergarten cohort of 2001 births. All results discussed are p<05 unless otherwise noted.

Results

What are the complexities in developing health lifestyle pathways?

The second-order latent class analysis resulted in five longitudinal developing health lifestyle pathways (see Table 1), derived from the best-fitting first-order class solutions: four Wave 1 and 2, five Wave 3, and six Wave K classes (see Table 3). We expected a child’s health lifestyle to be discordant in the healthfulness of its behaviors, both within a single time point and across age. Tables 1 and 3 use color coding to indicate health behaviors and lifestyles that are significantly more and significantly less healthful than the sample mean, and the pathways with the highest and lowest prevalence are bolded. The findings show that within each of the five health lifestyle pathways, children had a discordant mix of healthful and unhealthful behaviors within and across ages. First, the “healthful concordant” pathway was the most concordant pathway within and across waves, with only the school-controlled health behavior of physical education falling below the mean (Table 1). Children in this pathway were heavily drawn from the most healthful first-order lifestyle class in the first three waves, but in kindergarten only about one third of children came from the healthiest first-order latent class (Table 3).

Table 3.

Class-Conditional Response Probabilities of Five-Class Model from Second-Order Latent Class

	Overall sample	Healthful concordant	Healthful but TV/outdoors/junk food	Discord ant	Unfavorable concordant	Unfavorable low cost
Population share		19%	17%	30%	17%	17%
Wave 1
Mostly healthful except sleep°	0.20	0.22	0.23	0.24	0.12	0.14
Generally unfavorable°	0.20	0.05	0.06	0.21	0.37	0.35
Generally healthful°	0.33	0.61	0.55	0.27	0.14	0.11
Unfavorable food/smoking°	0.27	0.12	0.16	0.29	0.37	0.39
Wave 2
Middle°	0.14	0.14	0.20 *	0.15	0.15	0.04 *
Middle, unfavorable food/TV°	0.19	0.06 *	0.07*	0.24*	0.18	0.37 *
Generally unfavorable°	0.19	0.01	0.04	0.18	0.46	0.29
Generally healthful°	0.48	0.79	0.69	0.43	0.21	0.29
Wave 3
Mid. low-cost°	0.10	0.06*	0.03*	0.19 *	0.01 *	0.16*
Generally unfavorable°	0.17	0.02	0.04	0.04	0.37	0.49
Generally healthful°	0.24	0.90	0.03	0.17	0.01	0.04
Middle°	0.35	0.00 *	0.31*	0.55 *	0.52*	0.26*
Healthful/mid except junk food°	0.14	0.02*	0.59 *	0.05*	0.10*	0.05 *
Wave Kindergarten
Healthful except junk food°	0.17	0.30*	0.48 *	0.05*	0.08*	0.00 *
Unfavorable/safety°	0.16	0.01	0.06	0.15	0.57	0.00
Unfavorable/junk food/TV°	0.16	0.03	0.09	0.03	0.22	0.56
Generally healthful°	0.09	0.29	0.06	0.08	0.00	0.02
Healthful/middle°	0.19	0.34 *	0.22*	0.26*	0.05*	0.03 *
Unfavorable/middle, low-cost°	0.23	0.03 *	0.11*	0.43 *	0.08*	0.40*

Source: ECLS-B

Notes: Accounts for complex survey design. N≈6,550.

^°1=yes, 0=no.

^●No means were significantly different from sample mean. Gray=significantly healthier than sample mean, red=significantly unhealthier than sample mean, white=not significantly different from sample mean for that indicator, highest/lowest values are bolded.

^*Signiiicantly different from sample mean for rows not color coded because healthfulness of behavior is less well established.

Second, children in the “healthful but TV/outdoors/junk food” pathway largely had behaviors that were more healthful than average, but several behaviors in the areas of screen time, outdoor play, and diet (particularly junk food intake) were significantly less healthful than average at one or more waves (Table 1). The first two lifestyle pathways started out almost uniformly healthful at Waves 1 and 2, although “healthful concordant” usually had somewhat more favorable behaviors than “healthful but TV/outdoors/junk food” (see Tables 1 and 3). By Wave 2 and increasing over time, the “healthful but TV/outdoors/junk food” pathway started to diverge from the “healthful concordant” pathway for these unhealthy behaviors. Thus, although the “healthful but TV/outdoors/junk food” pathway included many healthful behaviors, it was discordant both within and across ages.

Third, the largest, “discordant” pathway represented almost one third of children. Children’s averages on some diet, safety, and daily routine measures were significantly more healthful than the sample mean, many behaviors fell near the mean, and at least one measure in each domain was significantly less healthful than the sample mean. The “discordant” developing health lifestyle pathway came from first-order classes that overall had fairly average levels of healthfulness in children’s behaviors (see Table 3). Table 1 shows that generally unfavorable behaviors in infancy developed into mostly average behaviors at Wave 2, then into a mix of favorable and unfavorable behaviors at Waves 3 and K. This relative improvement with age in children’s health behaviors stands in contrast to children in the “healthful but TV/outdoors/junk food” pathway, whose behaviors became less consistently healthful with age. Yet even more than that pathway, it displays high levels of discordance within and across ages.

In the “unfavorable concordant” pathway, many behaviors across several domains were significantly less healthful than the sample mean, and many others fell near the mean. Only physical education, a school-controlled health behavior, was significantly more healthful than the mean. The “unfavorable concordant” pathway was mostly in a relatively average class at Wave 3, then returned to unhealthful classes at Wave K, particularly with regards to safety. These inconsistencies over time did not follow a linear trend with age, as those in the “discordant” and “healthful but TV/outdoors/junk food” pathways did.

The fifth, “unfavorable low cost” pathway was similar to the “unfavorable concordant” pathway in the prevalence of less healthful behaviors. But behaviors were more discordant in the “unfavorable low cost” pathway: Children were consistently above the sample mean in their time spent in outdoor play, regular meals, and a few other diet measures (Table 1). Like the “unfavorable concordant” pathway, children in “unfavorable low cost” disproportionately came from unhealthful first-order classes at Waves 1 and 2 (Table 3). Starting at Wave 3, the “unfavorable low cost” pathway was overrepresented in “low cost” (with low engagement in health behaviors that require financial resources, many of which captured extrafamilial contexts) and “junk food/TV” unfavorable classes. Taken together, these findings show considerable discordance within and across age in children’s developing health lifestyle pathways.

How do parents influence children’s developing health lifestyle pathways?

Table 4 presents descriptive information about children in each developing health lifestyle pathway, who differed systematically in expected ways. Generally although not consistently, as the health behaviors within a specific pathway generally became less favorable (although all lifestyle classes were discordant in their healthfulness), the children in it had less advantaged parental social backgrounds. Parents of children in the “healthful concordant” pathway, closely followed by “healthful but TV/outdoors/junk food,” had the highest levels of social advantage (e.g., higher proportion White, lower proportion teen parents, higher grandmother education) and resources. The “discordant” pathway tended to fall in between, with both “unfavorable” pathways but particularly the “unfavorable low cost” pathway representing greater disadvantage. Maternal smoking and BMI (though not binge drinking) followed the same pattern, as did most non-health-focused parenting measures and all kindergarten outcomes.

Table 4.

Weighted Means for Multivariate Analysis Variables for Population and by Second-Order Latent Classes

	Population	Healthful concordant (19%)	Healthful but TV/outdoors/junk food (17%)	Discordant (30%)	Unfavorable concordant (17%)	Unfavorable low cost (17%)
Controls (Wave 1)
Age Wave 1 (months)	10.46	10.42	10.49	10.42	10.60*	10.45
Male°	0.51	0.48*	0.48*	0.53	0.53	0.52
Birthweight: Normal°	0.92	0.93	0.94	0.92	0.92	0.92
Low°	0.06	0.06	0.05	0.07	0.07	0.07
Very low°●	0.01	0.01	0.01	0.01	0.01	0.01
Birth order	2.04	1.86*	1.97*	2.07	2.09	2.23*
Cognitive score	0.19	0.23	0.23	0.18	0.24	0.11*
Behavior score●	0.11	0.09	0.15	0.11	0.09	0.08
Very good/excellent health°	0.88	0.93*	0.90*	0.87*	0.87*	0.87*
Asthma°	0.05	0.04*	0.04*	0.05	0.06	0.08*
Social Background (Wave 1)
Race: Non-Hispanic White°	0.54	0.75*	0.72*	0.46*	0.42*	0.38*
Non-Hispanic Black°	0.14	0.05*	0.07*	0.13	0.28*	0.22*
Hispanic°	0.25	0.14*	0.14*	0.34*	0.23	0.34*
Non-Hispanic other°●	0.07	0.07	0.07	0.08	0.08	0.06
Foreign-bom mother°	0.19	0.16*	0.11*	0.25*	0.14*	0.24*
Teen parent°	0.13	0.03	0.06	0.14	0.19	0.22
Grandmother education (years)	11.42	13.01	12.63	10.93	10.90	9.91
Resources (Wave 1)
Household members under 18	2.14	1.88*	2.00*	2.18	2.23*	2.42*
Grandparent in residence°	0.16	0.04*	0.11*	0.19*	0.21*	0.23*
No parenting help/advice°	0.11	0.09	0.13	0.11	0.11	0.12
Father in residence°	0.80	0.95	0.89	0.80	0.67	0.68
Mother’s education (years)	13.21	15.19	14.40	12.79	12.27	11.51
Income-to-needs ratio	2.85	4.91	3.92	2.29	1.93	1.41
Asset scale	0.68	0.85	0.82	0.65	0.58	0.52
Receipt of food stamps°	0.20	0.06	0.06	0.20	0.33	0.35
Mother’s paid work: None°	0.47	0.41*	0.43*	0.48	0.42*	0.58*
Part-time°	0.26	0.31*	0.32*	0.23*	0.27	0.19*
Full-time°	0.27	0.28	0.25	0.29	0.31*	0.23*
Mother health behaviors (Wave 1)
Smokes°	0.19	0.05	0.11	0.21	0.28	0.29
Binge drinks°●	0.09	0.07	0.09	0.09	0.10	0.08
BMI	26.93	25.27	26.20	27.29	27.92	27.96
Parenting measures (Wave 1)
NCATS	34.64	36.27	35.55	34.21	33.81	33.47
HOME scale	0.08	0.37	0.33	0.01	−0.18	−0.10
Do not pick up crying child	0.35	0.18	0.23	0.37	0.48	0.52
Parenting knowledge score	5.09	5.92	5.71	4.90	4.67	4.30
Outcomes (Wave K)
Reading	−0.02	0.46	0.23	−0.13	−0.17	−0.47
Math	0.01	0.55	0.30	−0.12	−0.18	−0.45
Parent-reported behavior	0.07	0.37	0.21	0.09	−0.17	−0.20
Teacher-reported behavior	0.03	0.29	0.22	−0.03	−0.12	−0.18
Very good/excellent health°	0.87	0.94	0.91	0.87	0.83	0.78

Source: ECLS-B Notes: Accounts for complex survey design. N≈6,550.

^°1=yes, 0=no.

^●No means were significantly different from sample mean. Gray=significantly more favorable than sample mean, red=significantly less favorable than sample mean, white=not significantly different from sample mean for that indicator.

^*Significantly different from sample mean for rows not color coded.

Multinomial logistic regression models, presented in Table 5, investigated the second research question. We expected children’s developing health lifestyle pathways to be shaped by parents’ social background, resources, health behaviors, and non-health-focused parenting practices. Among the social background measures, only the teen parent indicator was not associated with significant differences between multiple pathways. Social background measures differed relatively little between the two “healthful” pathways and between the two “unfavorable” pathways. It is therefore interesting to examine (as we do below) whether these pairs of lifestyles, in which otherwise quite similar children have different patterns of health behaviors over time, have distinct implications for child development. As expected, every non-health-focused parenting and parental health behavior indicator and most resource indicators was significantly related to membership in developing health lifestyle pathways.

Table 5.

Relative Risk Ratios from Multinomial Logistic Regression Models Predicting Early Childhood Lifestyle Classes, Comnared to “Healthful Concordant”

	Healthful but TV/outdoors/junk food		Discordant		Unfavorable concordant		Unfavorable low cost
Controls (Wave 1)
Age Wave 1	1.06		0.99		1.04		1.09
Male	1.01		1.31	*	1.27		1.26	+
Birthweight (normal)
Low	0.84		0.88		0.76		0.69	*
Very low	0.58	*	0.91		0.76		0.60	+
Birth order	1.08		1.09		1.09		1.15
Cognitive score	0.90		0.99		0.99		0.78	+
Behavior score	1.08		1.09		1.08		1.13
Very good/excellent health	0.66	+	0.63	*	0.65	+	0.73
Asthma	0.62		0.52	+	0.46	+	0.58
Social background (Wave 1)
Race/ethnicity (White)
Black	1.26		2.04	**	3.47	***	2.41	**
Hispanic	0.92		1.69	**	1.15		1.14
Other	1.49	+	1.94	**	2.32	**	1.66	*
Foreign-born mother	0.52	**	0.65	*	0.31	***	0.45	**
Teen parent	0.80		1.16		1.26		1.24
Grandmother education	0.99		0.97		0.94	*	0.93	*
Resources (Wave 1)
Household members under 18	1.00		1.01		1.01		1.02
Grandparent in residence	1.90	*	2.33	**	2.24	**	2.54	**
No parenting help/advice	1.51	+	1.08		0.90		0.99
Father in residence	0.64	+	0.93		0.82		0.84
Mom education	0.91	**	0.84	***	0.78	***	0.72	***
Income-to-needs	0.95	*	0.85	***	0.85	**	0.74	***
Asset scale	1.24		0.63		0.51	+	0.41	*
Food stamps	0.45	*	0.80		1.00		0.88
Mom work status (full time)
Not working	1.12		0.73	+	0.57	**	0.80
Part time	1.14		0.66	*	0.71	+	0.64	*
Mom’s behaviors (Wave 1)
Smokes	1.64	*	2.97	***	3.40	***	3.52	***
Binge drinker	1.30		1.54	+	1.66	*	1.56
BMI	1.02		1.02	*	1.03	*	1.03	*
Parenting (Wave 1)
NCATS	0.97	+	0.96	**	0.95	**	0.96	*
HOME scale	0.98		0.80	**	0.71	***	0.83	*
Do not pick up crying child	1.11		1.23		1.51	*	1.67	**
Knowledge count	0.95		0.91	*	0.87	**	0.83	***
Constant	7.90	+	306.25	***	754.41	***	1219.30	***

Source: ECLS-B

Notes: Accounts for complex sampling design. N≈6,550.

Do developing health lifestyle pathways matter for children’s kindergarten readiness?

Table 6 summarizes results from multivariate models predicting kindergarten cognition, socioemotional behavior, and health. The table presents coefficients from linear regression models predicting early reading and math and parent- and teacher-reported socioemotional behavior and from binary logistic regression models predicting the likelihood of very/good excellent compared to good/fair/poor health. As expected, health lifestyle pathways were associated with all kindergarten readiness outcomes after adjusting for controls, social background, and parents’ resources, health behaviors, and parenting practices. Table 6, Model 3’s fully adjusted early reading scores were about 0.4 standard deviations lower for children in the “unfavorable low cost” pathway compared to “healthful concordant.” The “healthful but TV/outdoors/junk food,” “discordant,” and “unfavorable concordant” pathways had early reading scores that were about 0.2 standard deviations lower than “healthful concordant.” Early math scores followed a similar pattern with similar magnitudes, except that the “discordant” and “unfavorable concordant” pathways had scores that were 0.3 standard deviations lower than “healthful concordant.” For parent-reported behavior, the two unfavorable pathways had similar scores at about 0.4 standard deviations below the “healthful concordant” pathway, while the “healthful but TV/outdoors/junk food” and “discordant” pathways were about 0.15 standard deviations lower than “healthful concordant.” Teacher-reported behavior was 0.20 standard deviations below “healthful concordant” for “unfavorable low cost,” 0.14 for “discordant” and “unfavorable concordant,” (p< 10 for “unfavorable concordant”) and showed no significant difference between the two healthful pathways. For health indicators, the disparities were different. Parents of children in the “unfavorable low cost” pathway were 2.3 times (exp[0.85]) as likely to report good/fair/poor health compared to very good/excellent, and children in the “unfavorable concordant” pathway were about twice as likely, compared to “healthful concordant.”

Table 6.

Coefficients from OLS and Binary Logistic Regression Models Predicting Kindergarten Readiness Outcomes

	Reading			Math			Parent-reported behavior
	1	2	3	1	2	3	1	2	3
Health lifestyles (Healthful concordant)
Healthful but TV/outdoors/junk food	−0.27 ***	−0.25 ***	−0.17 **	−0.30 ***	−0.27 ***	−0.19 ***	−0.17 ***	−0.17 ***	−0.15 **
Discordant	−0.60 ***	−0.45 ***	−0.24 ***	−0.68 ***	−0.51 ***	−0.30 ***	−0.29 ***	−0.24 ***	−0.16 **
Unfavorable concordant	−0.65 ***	−0.50 ***	−0.23 **	−0.75 ***	−0.56 ***	−0.29 ***	−0.55 ***	−0.53 ***	−0.40 ***
Unfavorable low cost	−0.92 ***	−0.70 ***	−0.39 ***	−0.99 ***	−0.74 ***	−0.43 ***	−0.58 ***	−0.53 ***	−0.41 ***
	Teacher-reported behavior			Very good/Excellent Health
	1	2	3	1	2	3
Health lifestyles (Healthful concordant)
Healthful but TV/outdoors/junk food	−0.09	−0.08	−0.04	−0.35	−0.31	−0.26
Discordant	−0.33 ***	−0.28 ***	−0.14 *	−0.78 ***	−0.42 *	−0.29
Unfavorable concordant	−0.43 ***	−0.33 ***	−0.14 +	−1.13 ***	−0.85 ***	−0.68 **
Unfavorable low cost	−0.49 ***	−0.41 ***	−0.20 *	−1.47 ***	−1.05 ***	−0.85 ***

Source: ECLS-B

Notes: Accounts for complex sampling design. N≈6,550. Model 1 controls for age at Wave K; Model 2 includes age and all controls; Model 3 includes age, all controls, resources, parent health behaviors, and parenting (see Table 5 for descriptions).

The kindergarten readiness analysis yielded additional key findings. First, different developing health lifestyle pathways mattered for different outcomes—which may have distinct long-term developmental implications. For example, children in the “healthful concordant” pathway unsurprisingly and consistently had the most favorable outcomes. The “unfavorable low cost” pathway always had the most unfavorable kindergarten reading and math scores, but for parent-reported behavior and child health it was joined by the “unfavorable concordant” pathway at statistically indistinguishable levels. The “healthful but TV/outdoors/junk food” pathway was indistinguishable from “healthful concordant” for teacher-reported behavior and child health, but for reading, math, and parent-reported behavior, its magnitude was closer to the “discordant” and sometimes even the “unfavorable concordant” pathway than to “healthful concordant.”

Second, further supporting our expectations, demographically similar children with distinct developing health lifestyles had quite different levels of kindergarten readiness. For early reading and math, differences within the two healthful pathways and within the two unfavorable pathways were close to 0.2 standard deviations in fully adjusted models. For teacher-reported behavior and health, though, the two healthful and two unfavorable pathways were statistically indistinguishable, respectively. Together, these findings suggest that developing health lifestyle pathways are associated with specific domains of kindergarten readiness in important ways that are distinct from social background.

Finally, supplemental analyses (not shown) provided further support for the usefulness of a developing health lifestyle pathways approach. We found that developing health lifestyle pathways’ implications for children’s kindergarten readiness (Table 6, Model 3) were much more powerful than effect sizes for cross-sectional health lifestyles in the first two years of life. Compared to age 4, effect sizes for developing health lifestyle pathways were larger for all outcomes except parent-reported behavior, for which they were similar. This underscores the importance of operationalizing health lifestyles as they unfold across ages in order to avoid underestimation of their implications.

Conclusion

Although individual health behaviors have been studied, little is known about how families and social inequalities shape the development of health lifestyles across early childhood, a life phase when parents are highly influential but children are developing agency over some health behaviors. We used US nationally representative data from the Early Childhood Longitudinal Study – Birth Cohort to model predominant pathways of health lifestyles across the first five years of life, test theoretical ideas about complexities in and influences on these pathways, and estimate their implications for kindergarten cognition, socioemotional behavior, and health.

Our first research question focused on complexities in developing health lifestyle pathways, both in terms of discordant behaviors and changes across age. Each predominant early childhood health lifestyle pathway included a mixture of favorable and unfavorable health behaviors, which we argue reflects complexities in children’s developmental contexts. This discordance in the healthfulness of behaviors within individuals echoes findings from other stages of the life course (e.g., Burdette et al. 2017; Mollbom and Lawrence 2018; Saint Onge and Krueger 2017). Pathways also changed substantially with age, which we expected given rapid shifts in children’s contexts and development across early childhood and given considerable changes in individuals’ health lifestyles across time in later stages of the life course (Burdette et al. 2017; Mollbom and Lawrence 2018). Future research, perhaps starting with qualitative work, should document why these changes happen and what processes trigger their timing in early childhood.

The second research question emphasized parental influences on children’s developing health lifestyle pathways. Parents’ social background prior to the child’s birth, including socioeconomic status and race/ethnicity, was related to developing health lifestyle pathways. Parents’ resources, health behaviors, and non-health-focused parenting practices were also associated with these pathways. Although previous research has established that parental social background and resources are associated with young children’s health lifestyles (Mollborn et al. 2014), non-health-focused parenting practices have not previously been studied in early childhood, nor have parent health behaviors been studied as predictors of young people’s health lifestyles. Our analyses show that non-health-focused parenting practices and parent health behaviors are distinct predictors of health lifestyles above and beyond social background and parent resources. Future data capturing both parental and nonparental care contexts can better articulate their evolving roles in these processes.

Analyses addressing our third research question found that longitudinally measured health lifestyles had important associations with kindergarten readiness, net of parents’ social background, health behaviors, non-health-focused parenting practices, and resources. These longitudinal operationalizations were more useful predictors than cross-sectional lifestyle measures, and sociodemographically similar children who had different developing health lifestyle pathways ultimately had quite different levels of kindergarten readiness. These combined implications speak to the power of a longitudinally oriented health lifestyle pathways approach for understanding human development. The mechanisms linking developing health lifestyles to various domains of kindergarten readiness were not studied here but will be important when refining future theoretical models. Health lifestyles may impinge directly on health, cognition, and socioemotional behavior, or they may be part of overarching lifestyles that include broader sets of behaviors and influences shaping child development. The link between health lifestyles and wider parenting projects and child behaviors is a promising route for further study.

Taken together, our findings support a longitudinal approach to children’s health lifestyles that expects discordance in a child’s health lifestyle both within a single time point and across time. Earlier lifestyles seem to matter for later ones to some extent, but substantial changes across age suggest that the dynamics of children’s developmental contexts, identities, development, and agency are also important. Future research should more fully capture the contexts experienced by children, and changes in these contexts, to articulate the processes through which age shapes health lifestyle pathways. For example, experiences with discrimination, neighborhood segregation, or family structure changes may be important. Further articulating and measuring these complexities can help build theory on health lifestyles across the life course.

This study is a first step in articulating the longitudinal development and implications of early childhood health lifestyles. Its limitations—which are shared by other health lifestyle studies—include using observational data, which may suffer from omitted variable bias. The ECLS-B data are from 2001-2007, and newer data could portray health lifestyles in a more recent cohort. The two-stage LCA introduces additional classification error compared to one stage, a limitation that future methodological research should investigate. As LCA is an active area of methodological research, we anticipate that a consensus around best practices will be useful in further developing the theoretical and empirical insights offered here. Our broad characterization of behaviors comprising health lifestyles precludes in-depth study of specific behavioral domains, such as diet or sleep. A lack of measures of child preferences, group-based identity measures, or who instigated a child’s behavior limits our conclusions. More attention to the role of nonparental child care for health lifestyles is also warranted, and a similar theoretical model could be tested using data that better combine parental and nonparental care influences.

This study has implications for policy. First, because of the discordance in health lifestyles, children who are identified as healthy based on a single health behavior measure may be struggling in other domains. Interventions—which often focus on modifying a single health behavior such as physical activity—should adopt a broader focus that encompasses a variety of health behaviors and identifies areas of risk and social contexts that may be shaping the health lifestyle in which that risk is occurring. This approach may lead policymakers toward “upstream” interventions that could be more effective at addressing broader contexts and lifestyles, such as improving neighborhood safety or (especially for children in the “unfavorable low cost” pathway) supplementing young families’ financial resources for athletic activities and health care.

Second, because longitudinal examinations of health lifestyles reveal important patterns that would be missed even with repeated cross-sectional analysis, identification of children for policy intervention should rely on a longitudinal view rather than on a snapshot of children’s health behaviors. Similarly, interventions may need a longer time horizon to be more effective. Bolstering family supports across early childhood, rather than for shorter periods, may well be important in these efforts.

Third, health lifestyles are a distinct phenomenon with developmental implications, so focusing on group-level identities to intervene in overarching lifestyles rather than specific behaviors may be a promising route to behavioral change. Facilitating group-level conversations about, for example, what kindergartners and their parents within a school want their children’s health behaviors to look like may be a more fruitful intervention than attempting to change individual people’s minds in isolation.

Finally, intervening in early childhood may be particularly important for shaping health lifestyles throughout life and into subsequent generations. This is true both because of the considerable continuity in health lifestyles across ages, and because health lifestyle pathways have implications for school readiness, which in turn is strongly associated with later academic and socioeconomic achievement (Entwisle et al. 2004). Thus, early interventions into health lifestyles could pay off throughout the life course for both socioeconomic attainment and health.

Appendix A. Class-Conditional Response Probabilities from First-Order Latent Class Analyses (LCA), Waves 1 and 2

Wave 1 (10 months old)
	Population	Healthful	Mostly healthful, unfavorable sleep	Unfavorable food/ smoking	Generally unfavorable
Population share		28%	19%	31%	22%
Bottle to bed (not water)	0.28	0.09	0.12	0.41	0.48
Breastfed for 6+ months	0.31	0.61	0.56	0.07	0.08
Solid food prior to 4 months	0.23	0.07	0.14	0.35	0.36
Child food insecurity	0.11	0.04	0.11	0.12	0.19
Sleep position on back (newborn)	0.58	0.94	0.25	0.80	0.12
Sleep position on back (3 months)	0.55	1.00	0.05	0.85	0.03
Someone smokes in home	0.12	0.01	0.01	0.19	0.23
Frequency of outside play
Never/rarely/few times per month	0.22	0.18	0.22	0.26	0.23
Few times per week	0.31	0.32	0.30	0.32	0.30
Once per day	0.26	0.29	0.27	0.26	0.22
More than once per day	0.21	0.21	0.22	0.17	0.25
Wave 2 (2 vears old)		Generally healthful	Middle	Mid, unfavorable food/TV	Generally unfavorable
Population share		40%	17%	24%	19%
Safe seat (car/backseat)	0.95	0.99	0.98	0.96	0.84
Regular soda consumption	0.26	0.14	0.12	0.39	0.47
Bottle to bed	0.16	0.07	0.15	0.36	0.14
Child food insecurity	0.09	0.03	0.10	0.12	0.21
Someone smokes in home	0.1	0.01	0.07	0.13	0.30
Spanking	0.44	0.30	0.41	0.36	0.84
Regular dinner
0-4 days per week	0.24	0.07	0.34	0.26	0.42
5-6 days per week	0.22	0.22	0.37	0.16	0.17
7 days per week	0.54	0.69	0.30	0.58	0.41
Frequency outside play
Never/rarely/few times per month	0.18	0.12	0.34	0.19	0.16
Few times per week/ once per day	0.61	0.59	0.66	0.61	0.58
More than once per day	0.21	0.29	0.00	0.19	0.26
Average TV watching per day
0-<1	0.15	0.23	0.05	0.09	0.17
1-<2	0.33	0.40	0.43	0.22	0.21
2-<3	0.24	0.23	0.36	0.19	0.22
3+	0.27	0.14	0.17	0.49	0.40

Source: ECLS-B

Notes: Accounts for complex survey design. Wave 1 N≈10,200. Wave 2 N≈8950.

Appendix B. Class-Conditional Response Probabilities from First-Order Latent Class Analyses (LCA), Wave 3 (Age 4½)

	All	Generally healthful	Middle	Healthful/mid except junk food	Mid, low-cost	Generally unfavorable
Population share		0.22	0.33	0.15	0.13	0.17
Child food insecurity	0.14	0.00	0.16	0.02	0.24	0.28
Smoking in home	0.12	0.00	0.15	0.04	0.15	0.28
Athletics/dance	0.36	0.63	0.28	0.59	0.09	0.15
Insufficient sleep/late bedtime	0.23	0.12	0.29	0.16	0.17	0.35
Safe seat	0.79	0.92	0.76	0.86	0.85	0.57
Spanking	0.34	0.14	0.46	0.28	0.27	0.47
Victim of violence	0.02	0.00	0.02	0.00	0.04	0.03
Always wears helmet	0.48	0.64	0.36	0.50	0.60	0.36
Dentist/doctor within recommendations	0.58	0.70	0.55	0.69	0.43	0.50
Regular dinner
0-4 days per week	0.34	0.18	0.46	0.29	0.22	0.49
5-6 days per week	0.31	0.40	0.30	0.33	0.22	0.21
7 days per week	0.35	0.42	0.24	0.37	0.56	0.30
Outside play
Never/rarely/few times per month	0.15	0.09	0.18	0.12	0.20	0.17
Few times per week/ once per day	0.68	0.76	0.69	0.71	0.61	0.59
More than once per day	0.17	0.15	0.13	0.17	0.19	0.25
Average TV watching/day
0-<1 hours	0.03	0.08	0.02	0.00	0.04	0.01
1-<2	0.11	0.19	0.08	0.10	0.12	0.04
2-<3	0.25	0.42	0.23	0.26	0.19	0.13
3+	0.61	0.31	0.68	0.63	0.65	0.82
Computer use
1	0.45	0.24	0.52	0.26	0.56	0.69
2	0.33	0.50	0.28	0.45	0.24	0.17
3	0.22	0.26	0.20	0.30	0.20	0.14
Soda consumption
none	0.27	0.52	0.13	0.24	0.50	0.07
< 1 per day	0.42	0.44	0.58	0.47	0.27	0.18
1+ per day	0.30	0.04	0.28	0.28	0.23	0.75
Fast food
none	0.23	0.36	0.09	0.17	0.57	0.16
< 1 per day	0.65	0.61	0.82	0.69	0.41	0.50
1+ per day	0.11	0.03	0.08	0.14	0.02	0.34
Fruit/vegetable ≥1/day	0.92	0.95	0.86	0.93	0.94	0.95
Milk
None	0.15	0.13	0.21	0.09	0.06	0.18
< 1 per day	0.16	0.14	0.18	0.14	0.17	0.16
1+ per day	0.30	0.34	0.29	0.35	0.30	0.22
2+ per day	0.40	0.40	0.32	0.42	0.47	0.45
Sweets 1+ per day	0.47	0.34	0.22	0.85	0.36	0.85
Salty snacks 1+ per day	0.30	0.09	0.07	0.58	0.23	0.81

Source: ECLS-B

Notes: Accounts for complex survey design. N≈8,000.

Appendix C. Class-Conditional Response Probabilities from First-Order Latent Class Analyses (LCA), Wave K (Age 5½)

	Population	Generally healthful	Healthful/mid	Healthful except food	Unfavorable/mid, low-cost	Unfavorable/safety	Unfavorable/junk/TV
Population share		10%	18%	16%	23%	17%	16%
Child food insecurity	0.12	0.04	0.01	0.00	0.21	0.17	0.21
Fruit and vegetable once/day	0.91	0.98	0.89	0.96	0.93	0.82	0.93
Sweets once or more/day	0.46	0.41	0.01	0.94	0.27	0.35	0.95
Salty snacks ≥once/day	0.31	0.16	0.03	0.44	0.18	0.20	0.86
Insufficient sleep/late bedtime	0.22	0.09	0.22	0.11	0.19	0.46	0.32
Someone smokes in home	0.11	0.00	0.01	0.00	0.20	0.16	0.21
Athletics/dance	0.49	0.72	0.70	0.83	0.12	0.47	0.32
Safe seat	0.71	0.96	0.83	0.81	0.71	0.40	0.62
Spanking	0.24	0.08	0.11	0.12	0.23	0.43	0.39
Victim of violence	0.02	0.01	0.02	0.00	0.02	0.04	0.02
Always wears helmet	0.45	0.72	0.54	0.49	0.47	0.26	0.34
Dentist/doctor within recommendations	0.84	0.90	0.91	0.93	0.77	0.80	0.80
Regular dinner
0-4 days per week	0.33	0.14	0.30	0.21	0.26	0.59	0.43
5-6 days per week	0.32	0.31	0.38	0.46	0.30	0.24	0.22
7 days per week	0.35	0.55	0.33	0.33	0.44	0.18	0.35
Average TV watching per day
0-<1 hours	0.03	0.08	0.04	0.03	0.02	0.02	0.01
1-<2	0.1	0.37	0.17	0.16	0.13	0.08	0.07
2-<3	0.26	0.37	0.34	0.37	0.15	0.23	0.20
3+	0.61	0.19	0.45	0.44	0.69	0.66	0.72
Computer use total
1	0.43	0.41	0.41	0.43	0.49	0.33	0.47
2	0.37	0.39	0.41	0.41	0.33	0.36	0.35
3	0.2	0.21	0.18	0.16	0.18	0.31	0.18
Soda consumption
none	0.27	0.79	0.31	0.25	0.24	0.13	0.12
less than once per day	0.46	0.20	0.63	0.61	0.49	0.55	0.14
once or more per day	0.27	0.01	0.07	0.14	0.27	0.32	0.74
Fast food
none	0.27	0.72	0.13	0.17	0.33	0.17	0.26
less than once per day	0.64	0.24	0.84	0.82	0.63	0.70	0.45
once or more per day	0.09	0.04	0.02	0.01	0.04	0.13	0.28
Milk
none	0.16	0.13	0.15	0.09	0.13	0.30	0.13
less than once per day	0.17	0.17	0.23	0.14	0.14	0.17	0.15
once per day	0.32	0.32	0.29	0.33	0.36	0.30	0.30
twice or more per day	0.35	0.38	0.33	0.43	0.37	0.23	0.42
PE minutes per week
<20	0.11	0.08	0.10	0.12	0.20	0.03	0.11
20-<60	0.38	0.40	0.47	0.43	0.32	0.34	0.35
60+	0.51	0.52	0.43	0.44	0.48	0.63	0.54

Source: ECLS-B

Notes: Accounts for complex survey design. N≈6,550.