Household Instability and Girls’ Teen Childbearing

Abstract

More than one third of U.S. children spend part of their childhood living with extended family members. By age 18, nearly 40% of U.S. children experience a household change involving a nonparent. Research has found that having extended family or nonrelatives join or leave children's households negatively affects children's educational attainment. I argue that we need new ways of theorizing, conceptualizing, and measuring household changes and their effects on children. I use the Panel Study of Income Dynamics and marginal structural models with inverse probability of treatment weighting to estimate the association between household changes involving parents and nonparents and teen childbearing among girls. I find that experiencing household changes involving nonparents and parents during childhood is associated with a significantly higher probability of having a child as a teenager than experiencing no changes. In addition, the association between changes involving parents and teen childbearing is statistically indistinguishable from the association between changes involving nonparents and teen childbearing, suggesting that household composition shifts involving nonparents can be as disruptive to girls as those involving parents.

Introduction

Recent estimates reveal that approximately one third of U.S. children spend part of childhood living with extended family members (Cross 2018). Extended family households are characterized by frequent changes (Pilkauskas 2012) that could expose children in these households to increased instability in caregiving, material resources, and social support (Perkins 2017; Raley et al. 2019). A narrow focus on parents and the nuclear family, however, has left social scientists with an incomplete understanding of how changes in household composition involving nonparents affect children's outcomes.

Given that the share of children who live with extended family and nonrelatives is growing (Harvey et al. 2021), more research on the consequences of these arrangements is needed. Households that include extended family or nonrelatives are incompletely institutionalized (Cherlin 1978; Sweeney 2010), meaning they do not have a common script for the roles and responsibilities of individuals living in these complex households. I focus on the disruption children experience when they are exposed to changes in household composition involving extended family and nonrelatives and highlight the dynamic nature of many children's households rather than estimate the effects of living with extended family and nonrelatives in shared households at any given point (see Harvey 2020). When a nonparent joins or leaves a child's household, all household members must adjust and reorganize to accommodate this change. This reorganization can be stressful for children. Without institutionalized roles and expectations for nonparent household members and a structure for maintaining these relationships outside the home, children could face uncertainty in incorporating new household members or compensating for their absence.

The negative consequences of parental divorce and repartnering for children's school engagement and behavior and for adolescents’ delinquency and earlier and riskier sexual activity are well documented (Brown 2006; Capaldi and Patterson 1991; Cavanagh and Fomby 2019; Fomby and Cherlin 2007; Fomby and Osborne 2010). Few studies, however, have examined changes in household composition involving extended family members and nonrelatives, a key source of instability for children. Broadening the focus to coresident extended family and nonrelatives is necessary because traditional measures of family instability involving parents exclude the nearly 20% of children who live stably with one or two parents throughout childhood but experience changes in household composition involving nonparents. In addition, because some children experience changes involving parents and nonparents, focusing only on changes involving parents means overlooking key household changes that affect nearly 40% of children (Perkins 2019). An emerging literature examines the association between changes in household composition and children's outcomes, finding that exposure to changes involving nonparents negatively affects children's cognitive outcomes and educational attainment (Mollborn et al. 2012; Perkins 2019). These studies have not yet established how broad the negative consequences of changes in household composition may be and whether the children for whom household changes are negatively associated with cognitive outcomes and education are the same ones who experience negative associations in other domains.

Building on prior studies, I examine extended family and nonrelative household members to estimate the association between nonparent changes and child outcomes in a different domain: teen childbearing. Assessing a new domain provides a novel way to think about the connections between household changes and adolescent development and considers the breadth of disadvantages and barriers to social mobility resulting from household changes.

I use data from the Panel Study of Income Dynamics (PSID) and marginal structural models to examine the relationship between experiencing household composition changes and having a child as a teenager. Changes in household composition could reflect underlying instability in a child's home environment. This instability may drive girls into romantic relationships or risky behavior, and some youth may see parenthood as an opportunity to create a sense of stability. I show that household composition changes involving all combinations of parents and nonparents are associated with higher odds of teen childbearing among girls than experiencing no changes in household composition. Household composition changes are associated with a higher probability of teen childbearing among White and Black girls compared with no household changes. Household changes in early and middle childhood (between birth and age 12) are generally as predictive of teen childbearing as exposure to household changes from birth through age 17. Experiencing nonparent changes may be just as disruptive to children and adolescents as experiencing parent changes. These findings highlight the importance of negative consequences that we miss by focusing only on family structure instability.

Toward a Theory of Household Instability

The theory of instability and change is often used to explain how parental relationship dissolution and repartnering result in adverse outcomes for children (Capaldi and Patterson 1991; Wu and Martinson 1993). Social control theory, socialization and social learning theories, economic change theory, and family stress theory all contribute to the instability and change hypothesis: family structure changes may result in less supervision, shifting role models and expectations, and reduced resources (social, emotional, and economic), with important consequences for children as they age into adolescence and even adulthood (Cavanagh and Fomby 2019). Adolescent childbearing is one such important consequence.

Although family structure and family stability are correlates of teen pregnancy, research has focused more on parents and their romantic partners than on extended family and nonrelatives. Beginning with single changes in family structure and expanding to incorporate multiple disruptions involving parents and their romantic partners, research has found that changes in teens’ residential environments are associated with adolescent romantic relationships and earlier onset of sexual activity. Experiencing parental divorce in childhood is also associated with an increased risk of having a nonmarital birth (Cherlin et al. 1995). In addition, family instability, defined as multiple changes in mothers’ romantic partners, is associated with adolescents forming new and more romantic relationships (Cavanagh et al. 2008; Ivanova et al. 2014) and earlier sexual onset and earlier nonmarital birth (Fomby et al. 2010; Goldberg et al. 2017). Two or more changes in family structure before age 13 are associated with teen pregnancy among girls born in the 1980s in the United States (Smith et al. 2018). These studies examined changes in household composition involving parents—that is, family instability. In contrast, I argue that we should examine household instability, including extended family and nonrelatives who enter and leave children's households, accounting for changes involving parents and nonparents.

Broadening the focus from parents to include extended family and nonrelatives is crucial given how many children spend time living with people other than their parents and siblings. In 2018, 15% of American children lived in shared households that included adults other than their parents, parents’ romantic partners, and adult siblings (Harvey et al. 2021). The share of children living in shared households increased between 1996 and 2018, driven by increases in children living with grandparents (Harvey et al. 2021; Pilkauskas and Cross 2018). A study following children over time found that 57% of Black children and 20% of White children live with grandparents, aunts, uncles, or other relatives at some point before they turn 18 (Cross 2018). Extended family households are not only a safety-net strategy among low-income families; they are common among low-income families of all races and ethnicities and among Black and Hispanic families with high socioeconomic status (Cross 2018; Raley et al. 2019). An even larger share of children live in shared households at some point during childhood when nonrelatives are also considered (Pilkauskas et al. 2014). Living with extended family or nonrelatives, then, is common for U.S. children.

Shifting the focus from household structure to instability is an essential next step for research on shared households. Nearly 40% of children experience household changes involving extended family and nonrelatives, changes that are missed by measures of family instability comprising only changes in parents (Perkins 2019). Because living with a grandparent may also affect children (Amorim 2019; DeLeire and Kalil 2002; Dunifon and Kowaleski-Jones 2007; Harvey 2022; Sun and Li 2014), disruptions that children may experience from household composition changes involve grandparents, other extended family, and nonrelatives. In addition, the dynamic nature of children's households, rather than the effects of living with extended family and nonrelatives at any one point, must be considered.

Childhood exposure to nonparent changes in household composition is negatively associated with short-term cognitive outcomes among preschool-age White and Black children (Mollborn et al. 2012). Children and adolescents who experience changes involving nonparents are also less likely to graduate from high school and enroll in postsecondary education (Perkins 2019). In this article, I build on research demonstrating the adverse consequences of changes in household composition and consider the effects on teen childbearing.

Household Contexts of Adolescent Sexuality and Childbearing

Predictors of Teen Childbearing

Why might household changes involving nonparents be associated with teen childbearing? Individual, family, and contextual characteristics predict teen childbearing. These factors include conflict and violence in romantic relationships (Barber et al. 2018; Erdmans and Black 2015); experiencing poverty during childhood (Smith et al. 2018) and living in a neighborhood characterized by concentrated poverty (Wodtke 2013); and other contextual characteristics beyond one's neighborhood, such as unemployment rates, religiosity, and access to family planning resources (Mollborn 2017).

Research revealing mechanisms underlying the association of family and residential context with adolescent sexuality and childbearing may apply to broader household changes. Better relationship quality between adolescents and their mothers, adolescents and other adults, and long-term residence with extended family attenuate the negative association between family instability and nonmarital sex and nonmarital birth (Fomby et al. 2010). Nonparent changes in household composition could affect parent–child relationships. Changes in household roles when extended family or nonrelatives join or leave the household could be a distraction or an emotional, financial, or social stressor that reduces a parent's focus on the child (Cavanagh and Fomby 2019; Harvey 2022). Further, if long-term residence with extended family is a stabilizing force that compensates for instability involving parents, changes involving extended family members could counteract the advantage and push adolescents toward romantic relationships or risky behavior. Depression and relationship seriousness are associated with a stronger desire for a pregnancy and less desire to avoid one (Weitzman et al. 2017). Young women who demonstrate a lack of efficacy are less likely to practice consistent contraception with sexual partners (England et al. 2016). Early parenthood may also be intentional. Living with extended family or in multigenerational households may provide a network of caregivers for children of teenagers and that network, along with concerns about the health profile of kin, may justify early childbearing among some teens (Geronimus 1996).

Much like remarriage (Cherlin 1978) and stepfamilies (Sweeney 2010), shared households—despite their increasing prevalence (Pilkauskas and Cross 2018)—are incompletely institutionalized because there are few guidelines for the expectations that members have of each other. When forming a shared household, members may encounter and resolve problems unknown to other family types. For example, who is responsible for disciplining children in an extended family or nonrelated household, whose schedule and standards for housekeeping prevail, and what kinship terms are appropriate for coresident extended family or nonrelatives? When shared households dissolve, how do the remaining family members distribute the responsibilities of departing household members and compensate for the skills and resources removed? What happens to relationships formed during coresidence when household members leave? Negotiating these transitions and reorganizing the new household, whether bigger or smaller, takes time and can be stressful for children and other household members.

Differences by Racialized Group and Age

Shared households and household changes involving nonparents are more common among non-White than White children (Cross 2018; Perkins 2017). As a result, household composition changes may be more or less salient across racialized groups. Exposure to nonparent changes in household composition is negatively associated with short-term cognitive outcomes among preschool-age White and Black children but not Latino children (Mollborn et al. 2012). It is also associated with educational attainment: White children exposed to nonparent household change are less likely to graduate from high school and enroll in postsecondary education, but Black children experience no significant negative effect (Perkins 2019). Transitions in a mother's romantic partner predict earlier sexual initiation and nonmarital childbearing among White adolescents but are less predictive among Black adolescents (Fomby et al. 2010), suggesting differences by racialized group in the association between nonparent changes and teen childbearing. Family structures and family processes differ across and within racialized groups (Cross 2020, 2023; Williams and Baker 2021). Historical and present-day structural racism and heteropatriarchy make naive any comparisons of the effects of family structure for White and Black children that do not consider the structural conditions under which family processes unfold for different groups (Cross et al. 2022). In this article, I estimate the association between household changes and teen childbearing for White and Black girls, the two racialized groups best represented in the PSID.

In addition to racialized group, the age at which children experience household instability could matter for adolescent and longer term outcomes. Girls whose father was always absent or left between ages 6 and 13 had sex earlier than different-age siblings whose fathers were always present in the home (Ryan 2015). Sustained exposure to a poor neighborhood during adolescence may have a bigger effect on the probability of becoming a teen parent than exposure earlier in childhood (Wodtke 2013). Thus, whether the association between changes in household composition and teen childbearing differs by the age when children experienced the changes may illuminate which mechanisms are critical in linking household change to sexual activity and childbearing among adolescents.

Racialized group and age are two dimensions along which exposure to and the consequences of changes in household composition may vary. Coresiding with extended family and nonrelatives and experiencing changes in these arrangements may indicate or correlate with disadvantage, making it critical to consider the various economic or social status characteristics that predict changes in household composition and teen childbearing.

Data and Variable Construction

Data and Measures

I use data from the Panel Study of Income Dynamics to explore the link between changes in household composition and teen childbearing. The PSID contains data from a nationally representative sample of families surveyed since 1968. The original sample included approximately 4,800 families containing roughly 18,000 individuals, and the study's genealogical design adds children and grandchildren of original sample members as they form their own independent households. More than 80,000 individuals have contributed information to at least one study wave (PSID 2019).

For this study, I use data from the 1968–2019 waves of data collected annually from 1968 to 1997 and biennially since 1997. Females enter my sample in the first year they appear in the PSID, soon after their birth, and I observe them for at least 20 years to track changes in household composition during childhood and whether they had a child as a teenager. My sample, therefore, includes Black and White girls born between 1968 and 1999.

I use fertility history data collected from every individual to determine who experienced a teen birth. The teen childbearing outcome variable is coded as 1 for individuals who reported a first child's birthdate before they turned 20 (Hamilton et al. 2020) and 0 otherwise. The youngest and oldest sample members were, respectively, 20 and 50 years old by 2019, the most recent year of data in my analysis. Approximately 24% of my sample was at least 45 years old by 2019. Sample individuals were old enough by 2019 to have complete data on their childbearing experience as teenagers, although some have many remaining years of fertility. Births are a conservative fertility measure that does not account for pregnancy losses and abortions. The teen birth rate declined dramatically during the observation period and was at a record low in 2020 (Osterman et al. 2022). Early childbearing, however, may still disrupt social capital investment and truncate educational investment for young women, which could be especially consequential as college education becomes increasingly important for economic independence (Su and Musick 2023).¹

I operationalize changes in household composition using categorical measures and count measures. First, a categorical measure assigns children to one of four mutually exclusive groups on the basis of household changes experienced from baseline to age 17:² (1) involving only parents and stepparents, (2) involving only nonparents (i.e., siblings ages 25 and older and child and adult extended family and nonrelatives), (3) involving both parents and nonparents, and (4) no household changes.³ I classify changes involving cohabiting partners of parents as nonparent changes rather than parent changes because cohabiting partners are not identified consistently across PSID waves. My results are robust to classifying changes involving cohabiting partners of parents as parent changes (see online appendix E). Second, I create three count variables, one for each type of change. The count measure follows prior studies focusing on the number of changes in family structure that children experience rather than whether a mother or father exits or enters a child's family (Cavanagh and Fomby 2019).

I code changes in household composition as a change rather than classifying them as “good” or “bad” changes. For example, a grandparent coming to live with a child may contribute financially and provide childcare and social support (Amorim 2019), whereas a grandparent leaving a child's home may remove those resources. Alternatively, a grandparent joining a child's household because they need support could mean fewer resources for coresident children if parents and other adults are now caring for an older adult. Knowing only the relationship of the person to the child and whether they enter or exit the household rarely enables the determination of the nature of a change. Further, it is not always possible to isolate independent associations of changes involving specific relations who join and leave children's households. Vignettes from the data (presented later) illustrate that children often experience changes involving parents and grandparents or grandparents and other nonparents at the same time. As I explain in the Methods section, my estimation strategy controls for household economic resources, household size, the number of children in the household, and other characteristics to account for changing household resources co-occurring with changing household composition. Conceptually, I consider household composition changes to be disruptions in children's environments that may be stressful and require a period of adjustment and adaptation, regardless of whether they are “good” or “bad.” I return to this issue in interpreting and discussing my results.

From descriptive statistics in Table 1, I present vignettes that illustrate typical experiences in the four categories of household change. They represent multiple individuals drawn from the PSID to depict typical trajectories of household change that sampled individuals experienced. I base my descriptions of these trajectories on my analysis of publicly available data, but to avoid data disclosure, I create composite profiles that do not represent specific individual PSID respondents.

Childhood Characteristics by Type of Household Change

No Change

Children who experience no household changes have a parent head of household. For most of childhood, this parent is employed, although some experience brief unemployment spells. Typically, their parents are married, but a few have stably unmarried parents. Most of these children have household incomes above the poverty line for most of childhood.

Change Involving Parents

Most children in this category experience one or two changes involving parents. They have more disruption in their household than children who experience no changes, but they generally experience fewer changes than do children in the categories of household change that involve nonparents. Amanda, a composite representative example, lived with her sibling and married parents when she was first observed by the PSID. Her father has a high school diploma and has taken some college classes, but he does not have a bachelor's degree. He works full-time, and the family's income was roughly $40,000 (in 2020 dollars) when Amanda was born. Their income was somewhat unstable when Amanda was young, first increasing to $80,000 but then dipping back toward the poverty line. When Amanda was 13, her father moved out, and her mother was the head of the household. Amanda's mother remarried, and her new husband moved in when Amanda was 16.

Change Involving Nonparents

Most children in this category experience two to four changes involving nonparents; they typically experience more changes in household composition than children experiencing only parent changes. Jennifer, a composite representative example, lived with her mother, grandmother, and two aunts at baseline. Her grandmother was the household head; she did not finish high school and is employed, but the family's income hovered around the poverty line when Jennifer was young. When she was 3, Jennifer left her grandmother's house with her mother, who cared for Jennifer full-time and did not work outside the home. When Jennifer was 9, her cousin came to live with them for a couple of years and left when Jennifer was 11.

Change Involving Both Parents and Nonparents

Children in this category experience more instability than children in the other categories of household change. During childhood, most children in this category experience one or two changes involving just parents, one or two changes involving just nonparents, and one change involving parents and nonparents simultaneously. At baseline, Nicole, another composite example, lived with her single mother, who has a bachelor's degree and works full-time. The family's income was three times the poverty line, approximately $50,000 in 2020 dollars, and fluctuated during Nicole's childhood. When Nicole was 1, her father moved in. When she was 6, her aunt, uncle, and three cousins joined the household. One year later, Nicole's father, aunt, uncle, and cousins left the household. Nicole's mother's boyfriend moved in when Nicole was 10 and moved out when Nicole was 13. At age 16, Nicole was living with her mother and had her own child.

Methods

Marginal Structural Models

To account for dynamic selection into and out of different household structures involving changes in household composition, I use marginal structural models and inverse probability of treatment (IPT) weighting (Robins et al. 2000). These models control for baseline and time-varying characteristics of children's households related to both changes in household composition and teen childbearing and address bias from time-varying covariates that can influence and be influenced by time-varying predictors.⁴ Changes in household composition involving nonparents are the key predictor. These changes can occur at any point and multiple times during childhood. Household changes involving nonparents early in childhood may affect household characteristics, such as poverty and parental employment later in childhood, which may affect teen childbearing. Marginal structural models with IPT weighting isolate the association between changes in household composition and teen childbearing by adjusting for confounding factors that occur before a household change but do not adjust for the values of the same confounders that occur after a household change and could mediate the relationship between household change and teen childbearing. Prior work on developmental context and teen childbearing used marginal structural models to estimate the causal effects of cumulative exposure to neighborhood disadvantage on teen parenthood (Wodtke 2013). I estimate the stabilized IPT weight separately for White and Black girls. I describe its construction in online appendix A.

With the stabilized IPT weight, I run a weighted logistic regression model with teen childbearing as the outcome:

In this equation, p represents the group of children who, between ages 0 and 17, experienced change involving only parents, np represents children who experienced change involving only nonparents, and b represents children who experienced changes involving both parents and nonparents. The reference group is children who experienced no changes involving parents or nonparents. Coefficients for these variables, the lowercase gammas in Eq. (1), represent the association between exposure to household changes and teen childbearing. Using stabilized IPT weights reintroduces correlations between the household composition variables and the baseline covariates, so the outcome model includes controls for the baseline covariates from the prediction model (x₀).

In addition to the categorical specification, I use a separate set of models to predict teen childbearing using counts of household changes experienced during childhood:

Equation (2) represents the model for parent change, where p represents the number of changes involving only parents that girls experience. The coefficient, represented by the lowercase delta, represents the association between each additional change involving parents and teen childbearing. The reference group in this model is girls who experienced no household changes. I use equivalent models to predict changes involving nonparents and changes involving both parents and nonparents. I separate these models by type of change so that girls experiencing changes involving parents, for example, are compared with girls experiencing no changes rather than a mixed group with some girls experiencing no changes and others experiencing changes involving nonparents or involving both parents and nonparents.

Sample Restrictions

I start with the 4,047 Black or White females who have a positive PSID longitudinal individual weight and who I observe from age 0 or 1 until at least age 17 to track their household composition and changes from early childhood through adolescence. I exclude 457 children who were missing from the survey for more than two years because they missed at least one wave of data collection and I cannot observe changes in household composition. For the 3,590 children who remain, I construct complete, dynamic household rosters and identify the household members who leave and join their households. After estimating the IPT weight for the household change that each child experiences at every wave, I exclude another 169 children who had missing values for covariates in the estimation model and are thus missing a weight in at least one wave. (All individuals must have a weight at every wave to accurately represent their exposure to household change during childhood and adolescence.) Of the remaining 3,421 individuals, I exclude 249 individuals born after 1999. Although I observe these individuals to at least age 17 by 2019, I do not observe their final teenage years and therefore do not know whether they had a child before age 20. Finally, I exclude an additional nine individuals for whom I have complete household rosters up to age 17 but do not have information on whether they had a child between ages 17 and 20.⁵ My final sample for the models estimating the association between changes in household composition and teen childbearing contains 3,163 Black and White girls:⁶ 1,197 who experienced no changes in household composition, 442 who experienced parent changes, 701 who experienced nonparent changes, and 823 who experienced changes involving parents and nonparents.

Results

Table 1 presents descriptive statistics from my sample of 3,163 Black and White girls, weighted with the PSID longitudinal weight to account for sampling design and attrition. The sample includes girls born between 1968 and 1999, approximately 14% of whom gave birth as a teenager. Nearly half of the sample children experienced no changes involving parents or nonparents during childhood, whereas 16% experienced changes involving parents only, 17% experienced changes involving nonparents only, and 20% experienced changes involving parents and nonparents. Therefore, 53% of the sample experienced at least one change in household composition involving a parent, extended family, or nonrelative during childhood. The household change counts show that, on average, children experienced more changes involving nonparents (0.93 at ages 0–17) than changes involving parents (0.53). For children exposed to at least one change in household composition, 0.99 changes involved only parents, 1.75 involved only nonparents, and 0.30 involved both parents and nonparents.

Table 1 also presents mean values for baseline characteristics of the child, their household head, and household used to estimate the IPT weight. At baseline, 83% lived with married parents, and 92% lived in a parent-headed household. Most (85%) lived with household heads who were White at baseline. The sample is fairly evenly split in terms of the head's educational attainment, with 22% having less than a high school diploma, 33% having a high school diploma, 21% having some college but no degree, and 24% having at least a bachelor's degree. More than half (58%) lived in an owned home at baseline. The average household size at baseline was just over four, with an average of two children in the household.

Results from logit models with teen childbearing as the outcome and categorical household change variables, weighted with the IPT weight, show that exposure to change involving parents only, change involving nonparents only, and change involving both parents and nonparents are all associated with significantly higher odds of having a child as a teenager. Table 2 presents coefficients from two sets of models. In the first set, household change is observed only through age 12; in the second set, household change is observed through age 17. (Online appendix D presents all model coefficients.) In the full sample, the coefficients for change in household composition experienced through age 12 are very similar in direction, magnitude, and significance to the coefficients for change in household composition experienced through age 17. Household changes during childhood and adolescence (by age 17) are associated with significantly higher log odds of having a child as a teenager relative to the reference group of no changes, as demonstrated by a coefficient of 0.725 for change involving parents, 0.879 for change involving nonparents, and 0.702 for change involving both parents and nonparents. Wald tests show that the three coefficients for the category of change are not significantly different from one another, suggesting that the association between nonparent change and teen childbearing is indistinguishable in size from the association between parent change and teen childbearing.

These results suggest that the predicted probability of having a child is .167 for girls who experience parent-only household change through age 17. Comparative figures are .189 for girls who experience nonparent change, .164 for girls who experience parent and nonparent change, and .088 for girls who experience no household composition change.

Table 2 also presents coefficients from fully interacted models that estimate the association between types of change in household composition and teen childbearing separately for Black and White girls (following Long and Mustillo 2021). Among Black girls, change involving parents and change involving nonparents are significantly associated with higher log odds of teen childbearing relative to the reference group who experienced no changes. The coefficient for change involving both parents and nonparents is positive and marginally significant (p = .068). Among White girls, all three change types are significant at the .05 level. These results are robust to classifying parents’ cohabiting partners as parents instead of nonparents (see online appendix E). Figure 1 shows Black and White girls’ predicted probability of having a child as a teenager by household change status. The predicted probability of teen childbearing for Black girls is .231. Among Black girls who experienced no changes in household composition, the predicted probability is .171 and ranges from .251 to .300 across the three categories of household change. Among White girls, the predicted probability of teen childbearing is .106 and ranges from .073 for those experiencing no household changes to .143 among those experiencing parent and nonparent changes. Across all categories of household change, the predicted probability of teen childbearing is significantly higher among Black girls than White girls. Differences in the predicted probability of teen childbearing across the three categories of household change and the no-change reference group among Black girls, however, are not significantly different from the respective differences among White girls. Accordingly, these estimates do not suggest that the association between household change and teen childbearing differs for Black versus White girls.

Table 3 shows coefficients representing change in log odds of childbearing from IPT-weighted logit models with counts of household changes as predictors. The specification presented in Table 2 predicts teen childbearing using mutually exclusive categories of household change compared with a reference group experiencing no household changes. In Table 3, I separate the models by type of change to precisely compare types of changes with no household changes. The six coefficients in each panel are from six separate models restricted to children experiencing no household changes or one category of household change. The first coefficient in the top left (0.176) is from a model restricted to girls who experienced only changes involving parents or no changes through age 12, and the coefficient for parent change represents the increase in the log odds of teen childbearing associated with each change involving a parent experienced through age 12 relative to the reference group with no experience of household changes involving parents or nonparents. The coefficient for change in parents experienced through age 17 (0.323) is significant at the .05 level. Two rows lower, the coefficient of 0.141 suggests that relative to girls experiencing no changes, each change involving nonparents by age 17 is associated with a significantly higher likelihood of becoming a teen parent. In the models comparing change involving both parents and nonparents with no household changes, the coefficient for parent and nonparent changes experienced through age 12 (0.953) is positive and significant (p < .05), and the coefficient for changes experienced through age 17 (0.498) is marginally significant (p = .088).

As in Table 2, Table 3 shows coefficients from fully interacted models estimating the association between household change counts and teen childbearing separately for Black and White girls. For changes to age 17, five of the six coefficients are positive, and three are statistically significant. The coefficient for changes involving both parents and nonparents among White girls is negative but very imprecisely estimated. Predicted probabilities of teen childbearing are significantly larger for Black girls experiencing all types of change, but there is no evidence of a significant interaction between racialized group and household change.

Discussion

Children's families and households and the changes and instability in them have consequences for long-term child well-being. Using PSID data for a sample of girls born between 1968 and 1999, I find that experiencing household change involving parents, nonparents, and both parents and nonparents during childhood and adolescence is significantly associated with teen childbearing. Insofar as teen parenthood is disadvantageous to girls, this finding is consistent with recent studies documenting the deleterious effects of changes in household composition on children's educational attainment (Perkins 2019). My results support the conclusion that changes in household composition have negative consequences beyond education.

I find that exposure to changes involving parents and their romantic partners is associated with higher odds of teen parenthood (Fomby et al. 2010; Smith et al. 2018). In addition, the association between nonparent changes and teen childbearing is similar to, and statistically indistinguishable from, the association between parent changes and teen childbearing. Results from four model specifications increase confidence in the conclusion that experiencing changes involving parents and nonparents is associated with a higher probability of teen childbearing than experiencing no household changes. Tables 2 and 3 show results from models including household changes experienced through age 12 and through age 17 as predictors of teen childbearing. In Table 2, the coefficients across these models are very similar in magnitude and significance (e.g., 0.618 vs. 0.725 and 0.637 vs. 0.642), suggesting that changes in early and middle childhood are just as predictive of later outcomes as changes through adolescence. Early experiences with household change matter for longer term outcomes, suggesting that the negative consequences of experiencing instability in one's developmental context can manifest through longer term mechanisms and are not wholly an immediate response to household change.

The category versus count specifications also provide insight into new theorizing on nonparent household change. The categories of parent change and nonparent change include girls who experience multiple changes of the same type. As the vignettes of Amanda, Jennifer, and Nicole illustrate, the categorical conceptualization of household change represents types of change, but change type masks the number of changes: children who experience nonparent changes and changes involving both parents and nonparents are exposed to more changes overall than children who experience only parent changes. One puzzling result is that the coefficient for experiencing a change involving both parents and nonparents in the count model through age 17 is not significantly positive. In the categorical model, girls are considered to have experienced a change in parent and nonparent if they experienced at least one change involving a parent and one change involving a nonparent at any time in childhood. These changes could have happened in the same wave or in different waves. In the count model, the coefficient for change in parents and nonparents represents instances in which children experienced parent and nonparent change in the same wave. Experiencing two types of change within a year or two may have an even bigger impact than one type of change alone. Alternatively, two types of change in the same wave could offset each other: what a child loses (or gains) with the departure of one household member, she might gain (or lose) with the arrival of another. The confidence interval for this coefficient is large, and most of the interval's mass lies above 0, including positive values that are even bigger than the significant coefficients for parent change and nonparent change. Therefore, this coefficient's lack of significance could result from imprecision (a small group that experiences both types of changes in the same wave) and the variety of positive and negative circumstances that could precede and follow two changes occurring in a short time frame. Experiencing the two types of change separately would require adjusting to changes in household roles and expectations twice and could be more disruptive than experiencing both changes within a relatively short period.

These results should prompt further investigation of the mechanisms underlying disruptive effects of household changes involving extended family and nonrelatives for children and adolescents. The instability and change perspective argues that family instability involving parents’ romantic relationships can weaken a child's sense of security and lead to ambiguity in household roles, relationships, and expectations about the child's behavior. As a result, family instability contributes to worse outcomes for children and adolescents in social and emotional behavior; academic performance and cognitive functioning; and sex, fertility, and union formation (Cavanagh and Fomby 2019). Household instability may operate through some of the same family stress mechanisms that connect family instability to child and adolescent outcomes (Hadfield et al. 2018). Thus, social scientists must conceptualize and measure household changes more broadly rather than focus narrowly on family instability.

Limitations of this analysis suggest next steps for future research. I aimed to estimate the independent association between nonparent changes in household composition and the adolescent outcome of teen childbearing, which itself has implications for an individual's transition to adulthood and longer term well-being. I was intentionally inclusive in documenting a child's exposure to changes in household composition across childhood and modeling the likelihood of teen childbearing as a function of those changes, but doing so limits my ability to tease out independent associations between a grandparent joining a household or a family friend leaving the household.⁷ I seriously considered how to isolate the association of different relations coming and going from children's homes. This issue, however, is not one of just limited or imprecise data. Ultimately, as shown by Nicole's experience, family processes do not unfold in a way that facilitates the estimation of precise independent associations. Instead, many children experience concurrent changes in household composition. Thus, accounting for the overall level of instability, as my four categories of household change accomplish, is not only a parsimonious approach but also one that reflects children's lived experiences. Changes involving grandparents versus other extended family members, or kin versus nonkin, may be differently associated with children's outcomes because multigenerational shared households tend to last longer than shared households formed with other extended family or nonkin (Harvey 2020). I encourage qualitative researchers to consider these potential differences to learn more about the processes involved in forming and dissolving shared households and inform theory development and future quantitative research on household change.

My analysis suggests that the association between changes in household composition and teen childbearing is similar for Black and White girls. The predicted probabilities of teen childbearing based on fully interacted models are significantly higher for Black girls who experienced household change than for White girls, yet the differences between each category of household change compared with the reference group are not significantly different by racialized group. Research on the association between family instability and adolescent romantic relationships and sexual behavior found that the relationship is significant among White children but less consistently significant among Black children (Cavanagh and Fomby 2019; Fomby et al. 2010). Whether these differences arise because of the different methods used across studies or because different mechanisms link household instability to sexual behavior is an important area to explore in future research. Similar statistical associations between household changes and teen childbearing for Black and White girls do not necessarily mean that the forces leading to household changes or the consequences of teen childbearing are the same across groups. Comparisons across groups should consider the historical forces that make some family processes more or less protective for Whites versus Blacks (Cross et al. 2022).

Estimating the association between changes in household composition and teen childbearing by age and racialized group represents attempts to address potential heterogeneity in the consequences of household change for children. Recent research found heterogeneous effects of having a child as a teenager, with the negative effects of teen childbearing on college completion and earnings twice as strong for women who were unlikely to become pregnant as teenagers than for women who were likely to become pregnant as teenagers (Diaz and Fiel 2016). Exposure to changes in household composition is not uniformly distributed across children. Just as there are heterogeneous effects of teen pregnancy on later outcomes, there may be heterogeneous effects of household change on teen childbearing based on a child's propensity to experience household change (see Perkins [in press] on heterogeneous effects of household change on educational attainment). Future research should further investigate heterogeneity in the association between household change and teen childbearing by race, ethnicity, and social class. These findings, along with those of Diaz and Fiel (2016), suggest that sociologists may benefit from conceptualizing teen parenthood as a correlate of other indicators of disadvantage rather than an isolated event that itself has further negative consequences for the teen parent. More research is necessary to assess differences in the causes and consequences of teen parenthood in a context of declining teen childbearing, where the group of teens having children is ever more selective.

I present evidence in this article that experiencing household changes involving extended family and nonrelatives is associated with a higher probability of having a child as a teenager. My dynamic modeling technique using IPT weights and marginal structural models improves upon ordinary least-squares models by addressing time-varying confounding variables. The IPT weights improve the covariate balance between those who experience household change and those who do not, but the weights do not eliminate imbalance. Thus, I stop short of interpreting my results as causal estimates, which requires assuming that my prediction and outcome models include all confounding factors predicting household change and are also associated with teen childbearing and that there are no unobserved confounding variables. The models I use to estimate the IPT weights include several characteristics of the child, the household head, and the household at baseline and at every wave the child is observed. Additional characteristics, such as physical and mental health, are plausibly related to both household changes and teen childbearing, and I would have included such characteristics if the PSID had consistently asked about them. I also do not have information about the children's families and networks beyond the individuals who lived with them during childhood. Membership in a child's physical household is an arbitrary boundary of that child's family system (Cavanagh and Fomby 2019; Cross 2023), and characteristics of children's families and networks beyond their households are likely associated with changes in household composition and behavior in adolescence.

Recent scholarship has brought new attention to the composition of children's households and the large and growing proportion of children who live with individuals beyond parents and siblings. This focus is critical for more accurately describing the characteristics of American families. Just as studies of family instability shifted our focus from family structure to the dynamic nature of parents’ romantic relationships, the next logical step after accounting for the full composition of children's households is to consider how change and instability in those households matter for children's longer term outcomes. Building on emerging literature considering the consequences of children's exposure to changes in household composition, I examine the association between changes in household composition involving nonparents and the probability of having a child as a teenager. I argue that social scientists should develop new theories of household change and instability, with family instability being an important component but not the only one. Household change and instability include changes in household composition involving parents as well as nonparent household members and their partners. Traditional measures of family instability focusing only on parents and their romantic partners miss changes in household composition to which a substantial proportion of children are exposed. As scholarship continues to demonstrate that these disruptions in developmental context are consequential across multiple domains, it becomes ever more important to account for the full extent of change and instability children experience as they grow up.

Acknowledgments

For their excellent comments, I thank Megan Doherty Bea, Katharine Donato, Corey Fields, Becky Hsu, Yuki Kato, Rahsaan Mahadeo, Brian McCabe, Ann Owens, David Pedulla, Eva Rosen, Jennifer M. Silva, and Alix Winter. I thank my husband, Zeke, for dedicating prime work hours to caring for our twin toddlers so that I could dedicate time to this research while we were without childcare during the COVID-19 pandemic. The collection of data used in this study was partly supported by the National Institutes of Health (grants R01 HD069609 and R01 AG040213) and the National Science Foundation (SES 1157698 and 1623684).

Notes

References