A vast amount of literature has documented negative associations between family instability and child development, with the largest associations being in the socioemotional (behavioral) domain. Yet, prior work has paid limited attention to differentiating the role of the number, types, and sequencing of family transitions that children experience, as well as to understanding potential heterogeneity in these associations by family structure at birth. We use data from the Fragile Families and Child Wellbeing Study and hierarchical linear models to examine associations of family structure states and transitions with children’s socioemotional development during the first nine years of life. We pay close attention to the type and number of family structure transitions experienced and examine whether associations differ depending on family structure at birth. For children born to cohabiting or noncoresident parents, we find little evidence that subsequent family structure experiences are associated with socioemotional development. For children born to married parents, we find associations between family instability and poorer socioemotional development. However, this largely reflects the influence of parental breakup; we find little evidence that socioemotional trajectories differ for children with various family structure experiences subsequent to their parents’ breakup.
A large literature has documented associations between family structure experiences and child development. Current evidence indicates that relative to children spending their entire childhood in a married two-biological-parent family, those experiencing single-parent and social-parent (married or cohabiting stepparent) families, as well as those experiencing family structure transitions, fare worse on a host of developmental outcomes, with the largest associations being in the socioemotional (behavioral) domain (Lee and McLanahan 2015; McLanahan et al. 2013; Waldfogel et al. 2010). However, prior literature has paid limited attention to differentiating the role of the specific types of transitions that particular children have experienced, as well as to understanding potential heterogeneity in these associations by early family structure experiences—most notably, family structure at birth.
It is well known that there is differential selection into family structure at birth, such that socially and economically disadvantaged individuals are disproportionately likely to have unmarried births (Berger and Bzostek 2014; Hofferth and Goldscheider 2010). Unmarried births, whether within or outside cohabitation, are in turn associated both with subsequent family instability and poorer developmental outcomes for children (McLanahan 2004, 2011; McLanahan and Jacobsen 2015). Furthermore, the specific type(s) and sequence of family structure transitions that a child experiences will, to some extent by definition, vary according to the family context into which the child is born. For these reasons, it is important to fully consider diverse pathways into various family structure states and transitions when estimating relations between family structure experiences and child outcomes. Although most prior studies have controlled for parental marital and/or coresidential status at birth, few examined whether associations between subsequent family structure experiences and child outcomes vary by family structure of origin. Better understanding of the role of diverse pathways into particular family experiences may provide crucial information regarding whether and under what circumstances associations between family structure experiences and child development may have causal implications, as well as which children’s developmental trajectories are likely to be influenced by which types (and sequences) of family structure transitions.
We use data from the Fragile Families and Child Wellbeing Study (FFCW) to examine associations of family structure experiences with children’s socioemotional development (behavior problems) during the first nine years of life. FFCW comprises a U.S. population-based sample of children born in large urban areas (many to unmarried parents) between 1998 and 2000. FFCW children and their families are relatively disadvantaged compared with the U.S. population as a whole. The FFCW study is a valuable resource for examining family structure transitions, given that family instability is particularly common among socioeconomically disadvantaged groups. Indeed, the FFCW data offer the opportunity to assess how family structure experiences may matter for children facing many sources of disadvantage in their lives. At the same time, these data are limited in that estimates may not be generalizable to nonurban and more-advantaged children and families.
Our analyses use a hierarchical linear modeling (HLM) approach, in which estimates are identified using both within- and between-child variation, and in which we incorporate several strategies to minimize bias due to social selection. We also pay close attention to both the number and type(s) of family structure transitions that children experience and also examine whether there is heterogeneity in associations between family structure experiences and child socioemotional development by whether children are born to married, cohabiting, or noncoresidential parents. Given that most U.S. children—and most disadvantaged children, in particular—will not spend their entire childhood living with both of their biological parents, it is important to identify how diverse birth circumstances and family experiences are likely to influence child development and well-being. Such knowledge may have implications for better understanding intergenerational transmission of inequality (McLanahan 2004; McLanahan and Jacobsen 2015; McLanahan and Percheski 2008).
Our analyses extend prior research in two additional ways. First, whereas most previous studies have primarily compared the well-being of children living in stable two-biological-parent families to that of children with other family structure experiences, we employ different counterfactual groups, depending on a child’s family structure at birth. In this way, we compare outcomes for children born into the same family contexts but who experience different family structure trajectories. For children born to cohabiting or married parents, we compare those experiencing family structure transitions with those experiencing only a stable two-biological-parent family. Whereas most previous research has also compared the family structure experiences of children born to single mothers with those of children experiencing only a stable (often married) two-biological-parent family, we focus on differences in socioemotional development among children born to noncoresidential parents, and compare those who experience subsequent family structure transitions with those who experience only a stable single-mother family. This strategy better accounts for initial selection into family type and also reflects that pathways into particular family experiences are likely to be distinct, as may be their influence on child well-being. Finally, our approach allows us to examine differences in child outcomes associated with (the state of or transition into) parental absence (or union dissolution), social parent presence (or parental repartnering), and (potentially) parental breakup with a social parent, as well as the sequencing of such experiences.
Even after accounting for family structure at birth and controlling for early family structure experiences (in our case, from birth to age 3), estimates may be biased by unobserved factors that are associated both with later family structure experiences (in our case, between ages 3 and 9) and with children’s later developmental trajectories. For this reason, in some specifications of our models, we also include a falsification test (Magnuson and Berger 2009; Osborne et al. 2012) that effectively adjusts for (unobserved) initial differences in children’s socioemotional adjustment (at age 3) that are associated both with later family structure states and transitions and with children’s socioemotional adjustment over time (from age 3 to 9).
Family structure experiences matter for child development because they influence children’s caregiving environments, including the levels of parenting and economic resources available to or invested in them and the nature of their relationships with their caregivers. Biological/legal ties and stable family structures are thought to be best for children’s development by resulting in larger, higher-quality, and more consistent relationships and investments, both because stable families and (married) two-biological-parent families tend to be more socioeconomically advantaged than other families and because parents therein have the greatest incentives to invest in children (Berger and McLanahan 2015; Carlson and Berger 2013). When family structures change, family resources, parental investments, and children’s caregiving environments are likely to also change. Indeed, the primary mechanisms that are thought to link family structure experiences to child development are economic resources, parental time and attention, and family conflict and stress (Amato 2005). At the same time, the consequences of the transitions that children experience may depend on the context in which they occur, including early family structure experiences and the sequencing of transitions, such as whether children experiencing single- or social-parent families were born to single mothers or experienced parental breakup during childhood, potentially followed by additional transitions. We discuss each of these issues in turn.
Family structure transitions, regardless of type, may negatively influence child development by disrupting family roles and routines and potentially leading to changes in residence, parental employment, and social support, thereby resulting in stress and conflict for parents and children alike (Cavanagh and Huston 2008; Fomby and Cherlin 2007; Hetherington 1999). Even in the best circumstances, family structure transitions are likely to involve some degree of stress for the children and adults involved. Moreover, accumulated stress and lack of consistency, which are associated with repeated transitions, may be particularly harmful for children (Amato 2010; Hetherington 1989). A growing body of research provides support for this hypothesis, demonstrating negative associations between the presence or number of transitions that a child has experienced with cognitive—and, particularly, socioemotional—well-being (Beck et al. 2010; Cavanagh and Huston 2008; Cooper et al. 2011; Fomby and Cherlin 2007; Lee and McLanahan 2015; Magnuson and Berger 2009; Osborne and McLanahan 2007).
Types of Transitions
Associations of family structure transitions with child development may vary by type(s) of transition(s), given differential implications regarding changes in economic resources, parental time and attention, and family conflict and stress. Yet, a priori expectations regarding the relative magnitude and direction of specific types of transitions are not always unambiguous. The dissolution of a child’s biological parents’ union is often associated with decreased economic resources and parental time and attention available to the child, as well as high (if not increased) levels of stress. In turn, parental breakup has consistently been linked to adverse outcomes for children (Lee and McLanahan 2015; Magnuson and Berger 2009; Mitchell et al. 2015; Osborne et al. 2012), although there is variation by parental relationship quality (Amato 2005; Fomby and Osborne 2010). However, the ways in which parental reconciliation and repartnering (with a social parent) might influence child development are less clear. Whereas parental reconciliation may be accompanied by increased parenting quality and economic resources, how it may affect stress and conflict is ambiguous. Parental reconciliation may be associated with reduced stress and conflict if parents have resolved the issues that led them to break up in the first place; it may be associated with increased stress and conflict if those issues continue to be problematic. Likewise, parental reconciliation may be a source of comfort and security for children, or it may be accompanied by ongoing insecurity about the future consistency of their family relationships and living arrangements. Empirical evidence has suggested that parental reconciliation is positively associated with maternal well-being (Osborne et al. 2012). However, the evidence regarding associations with child well-being is mixed (Liu and Heiland 2012; Magnuson and Berger 2009; Mitchell et al. 2015).
A priori expectations regarding parental repartnering are also ambiguous. We focus on maternal repartnering because the vast majority of children reside with their mothers half-time or more when their parents break up (Kreider and Ellis 2011). Maternal repartnering may increase access to financial and parenting resources. It may also lead to increases in maternal happiness and reductions in maternal stress and depression (Osborne et al. 2012), which could positively influence child well-being. At the same time, a new partner might disrupt established family roles and routines and lead to increased conflict in the family—for example, if maternal repartnering generates or exacerbates existing tensions with a child’s nonresident biological father, or if the child resists accepting a new authority figure and/or competes with the new partner for the mother’s time and attention (Hetherington and Stanley-Hagan 1999; Marsiglio and Hinojosa 2010). Recent empirical evidence suggests that maternal repartnering is associated with improvements in mothers’ (and their children’s) economic resources (Dewilde and Uunk 2008; Jansen et al. 2009; Osborne et al. 2012), unmarried mothers tend to repartner with men who have greater economic capabilities (education and earnings potential) than their child(ren)’s biological fathers (Bzostek et al. 2012), (particularly married) social fathers are often quite involved with children (Berger et al. 2008), and social father involvement is positively associated with child well-being (Berger and McLanahan 2015; Bzostek 2008). Evidence also suggests that after an initial adjustment period, many social parents develop positive, nurturing relationships with children (Marsiglio and Hinojosa 2010; White and Gilbreth 2001). Of course, maternal repartnering may positively influence maternal well-being while also having a neutral or even negative influence on child development, even if social fathers exhibit desirable parenting behaviors.
For the most part, research has found negative associations between maternal repartnering and child development (Amato 1994, 2005; Brown 2004; Coleman et al. 2000; Hetherington and Jodl 1994; Hofferth 2006; Manning and Lamb 2003; Marsiglio and Hinojosa 2010; Thomson et al. 1994). However, these findings predominately reflect comparisons with children experiencing only a stable two-biological-parent family; it is less clear whether developmental outcomes between children whose mothers repartner differ from those who experience only a stable single-mother family (Chan et al. 2015; Fomby et al. 2016; Lee and McLanahan 2015; Ryan and Claessens 2013; Ryan et al. 2015), which we argue is a better counterfactual.
On the whole, then, whereas the preponderance of empirical evidence suggests that relative to living in an undisrupted (married) two-biological-parent family, maternal repartnering is associated with modest reductions in child well-being, it remains possible that (single) mothers and their children may benefit when a mother enters into a stable long-lasting partnership. These benefits may be particularly apparent when children in social-father families are compared with those in single-mother families and, as we discuss later, when children have not also experienced their parents’ breakup.
Heterogeneity by Family Structure at Birth: Social Selection, Family Structure Trajectories, and Child Well-being
There is considerable selection into family structure at birth in ways that are likely to be associated with children’s endowments, subsequent family structure transitions, and subsequent child development (Berger and Bzostek 2014; Hofferth and Goldscheider 2010). Moreover, family structure at birth is inextricably linked to children’s pathways into particular family structures and the types and amount of family instability they experience. As such, it is critically important to consider the different contexts in which children experience family structure changes, as well as the divergent pathways through which they come to experience a given family structure.
Selection into family structure at birth is important. Approximately 43 % of all births to American women under age 40 are nonmarital (Manning et al. 2015). Whereas nearly 60 % of these births are to cohabiting couples (Manning et al. 2015), cohabiting unions are strikingly unstable in the United States, and most will eventually dissolve, many within the first few years of the birth (Andersson 2002). Girls growing up in a single-mother family have been shown to be more likely to become a single parent themselves (Hofferth and Goldscheider 2010). Nonmarital births in the United States are also disproportionately common among socioeconomically disadvantaged groups (Manning et al. 2015). The individuals who are most likely to experience a nonmarital birth are also most likely to experience limited economic resources, economic instability, family structure instability, and lower-quality family relationships, each of which is associated with adverse developmental outcomes for children (McLanahan 2004; McLanahan and Jacobsen 2015; McLanahan and Percheski 2008; McLoyd 1990), independent of parental marital status at birth. Associations between family structure experiences and child outcomes tend to be attenuated considerably when researchers adjust for selection factors. The most rigorous studies have tended to leverage within-child variation to rule out unobserved time-invariant factors, which further attenuates, but does not eliminate, these associations (see McLanahan et al. 2013).
The higher level of disadvantage in FFCW is likely to have important implications for our findings. For example, lower levels of income and education in the sample may connote higher levels of financial strain, which may be a major stressor that could fully or partially outweigh the consequences of family structure transitions. In addition, higher levels of economic hardship—especially common for families with unmarried births—may place FFCW families at particular risk of family structure instability. These factors, coupled with evidence suggesting that the negative associations of family structure transitions with child development are larger for advantaged families (Ryan and Claessens 2013; Ryan et al. 2015), imply that levels of both family structure instability and behavior problems may be greater in our sample than in a nationally representative sample, whereas associations between family structure transitions and child socioemotional development may be larger in a nationally representative (or more-advantaged) sample than in ours.
Beyond social selection, children’s family structure trajectories are likely to vary by the family structure into which they are born because family structure at birth is inextricably tied to subsequent family structure (in)stability; particular type(s) and sequencing of family structure experiences; and, potentially, associations between their family structure experiences and developmental trajectories. For example, family structure instability for a child born to cohabiting or married biological parents necessarily includes the dissolution of their parents’ union. Moreover, maternal repartnering for children born to married or cohabiting parents necessitates prior parental breakup, thereby reflecting at least two separate transitions. In contrast, maternal repartnering for children born to a single (neither married nor cohabiting) mother will not typically involve parental breakup, although some such children’s mothers and fathers will have formed postbirth cohabiting or marital unions, some of which will have subsequently dissolved. Children’s development may, in turn, also vary by the specific family structure trajectories they experience. For example, the association between maternal repartnering and child development may differ by whether a child also experienced parental breakup.
Drawing from life course theory, Ryan and Claessens (2013) and Ryan et al. (2015) posited that family instability may have a lesser influence on children born to unmarried and socioeconomically disadvantaged parents than their married and more advantaged counterparts for two primary reasons. First, because family structure transitions are more common among unmarried and disadvantaged families, they may be perceived as more normative and predictable, resulting in less conflict and stress. Second, transitions among such families may engender smaller changes in resources and parental investments than transitions among married and more-advantaged families.
Recent empirical evidence has suggested that the consequences of family instability for children may, indeed, depend on the context in which family structure experiences occur. Three recent studies are particularly relevant to our work: Ryan and Claessens (2013), Ryan et al. (2015), and Lee and McLanahan (2015). Each of these studies employed high-quality longitudinal data and leveraged both within- and between-child variation to identify associations. Each also explicitly modeled specific types of transitions and considered potential heterogeneity by socioeconomic (dis)advantage.
Ryan and Claessens (2013) found that (1) family structure experiences in early childhood (prior to age 3) are more strongly associated with subsequent socioemotional development than are later family structure experiences; (2) family structure transitions are more negatively associated with socioemotional development for children born to married parents than those born to unmarried parents; and (3) union dissolution is more negatively associated with socioemotional development than is maternal repartnering with a social father. However, they did not explicitly consider parental reconciliation into a coresident union or relationship dissolution for social-father families. Nor did they examine potential heterogeneity by parental cohabitation status at birth for children born to unmarried parents.
Consistent with the notion that family instability may have fewer consequences for children in less advantaged families, Ryan et al. (2015) found virtually no statistically significant differences in behavioral problem trajectories for low-income children who experience family structure transitions relative to low-income children experiencing stable family structures. In contrast, among high-income families, they found that children moving from two-biological-parent to single-mother households exhibit greater behavior problems than those experiencing stable two-biological-parent families. They also found a significant reduction in behavior problems for children moving from single-mother to social-parent families. Again, they did not specifically consider parental reconciliation, relationship dissolution for social-father families, or heterogeneity by parental cohabitation at birth.
Finally, Lee and McLanahan’s (2015) results indicate that transitions out of a two-parent family are negatively associated with child development, particularly with respect to behavior problems, and that these associations are largest for white children. They found less consistent evidence regarding transitions into a two-parent family, although such transitions appear to be particularly negative for Hispanic children. A primary limitation is that the study did not differentiate two-biological-parent families from mother–social-father families, but rather considered all two-parent families as a single group. In addition, the study assessed only concurrent associations between family structure experiences and child outcomes in the same period; it did not investigate associations between family structure trajectories and children’s developmental trajectories. Finally, it did not explicitly estimate associations of early family structure experiences with subsequent developmental outcomes (although they differenced these effects out of their within-child analyses).
Our study builds on and extends these recent analyses in several ways. Like Ryan and Claessens (2013), we estimate separate models by marital status at birth. However, we separate children born to cohabiting parents and those born to single mothers rather than combining these groups. In addition, our FFCW sample is, on the whole, relatively less advantaged than their (National Longitudinal Sample of Youth; NLSY) sample. Our FFCW sample allows for analyses among a sample of children who are most likely to experience family instability, albeit while limiting the generalizability of our results vis-à-vis the U.S. population as a whole. We also focus on a wider range of family structure transition types, including parental reconciliation and relationship dissolution for social-father families. Finally, we estimate some HLM models that include a novel falsification test (Magnuson and Berger 2009; Osborne et al. 2012), which provides an especially rigorous means of accounting for selection into family instability. Whereas we use the same data as Lee and McLanahan (2015), we extend their analyses by differentiating two-biological-parent families from mother–social-father families, estimating associations between family structure trajectories and children’s developmental trajectories, and explicitly modeling associations of early family structure experiences with socioemotional development. Our falsification test also provides an additional rigorous adjustment for selection.
Data and Methods
Our data are drawn from FFCW, a longitudinal birth cohort study of 4,897 children born between 1998 and 2000 in 20 U.S. cities (see Reichman et al. 2001). Interviews were conducted shortly after the focal child’s birth and when the child was approximately 1, 3, 5, and 9 years old. Because nonmarital births were oversampled by a 3-to-1 ratio relative to marital births, FFCW families are, on average, more disadvantaged than would be the case in a nationally representative sample. Of particular relevance to our analyses, sample families are disproportionately likely to include a single or social parent and to experience family structure transitions. Wagmiller (2010) compared sample characteristics of FFCW with those of the Early Childhood Longitudinal Study-Birth Cohort (ECLS-B), which includes a nationally representative sample of all U.S. births to women ages 15 and over in 2001. As expected, he found that FFCW families were more likely to be African American and had lower levels of earnings, income, and parental education. He therefore recommended that all analyses using FFCW data adjust for parental education, race/ethnicity, and household income. We follow this recommendation. Nonetheless, the higher level of disadvantage in FFCW is likely to have important implications for our findings.
We used multiple imputation to impute values for all variables with missing data in the full FFCW sample. Specifically, we imputed 10 complete data sets using Stata’s mi commands. Retaining children with missing data is important because these children and their families systematically differ from those with complete data. After conducting the imputations, we excluded children who did not live primarily with their mother at a given observation point (approximately 3.6 % of the sample). This resulted in a potential analysis sample of 14,132 to 14,178 child-wave observations of 4,787 to 4,812 children per data set (across the 10 data sets). Following Von Hippel’s (2007)1 recommendation, we then deleted all cases with imputed outcome data.2 Our final analysis sample consists of 9,308 to 9,317 child-wave observations of 3,222 to 3,223 children. Rates of imputed data in this sample were relatively low, at less than 1 % for all variables except child low birth weight (LBW) (3 %) and family structure (5 %, 4 %, 4 %, and 11 % at ages 1, 3, 5, and 9, respectively).
Child Socioemotional Development
Our outcome measures assess child behavior problems at approximately ages 3, 5, and 9 using the aggressive, withdrawn, and anxious/depressed subscales of the Child Behavior Checklist (CBCL) (Achenbach and Rescorla 2000), which was completed by the child’s mother at each interview. The aggressive behaviors subscale (alpha = .88, .84, and .89 at ages 3, 5, and 9, respectively) includes items such as the extent to which the child is cruel, bullying, or mean to others; physically attacks people; and has temper tantrums or a hot temper. The withdrawn behaviors subscale (alpha = .67, .62, and .70) includes items such as the extent to which the child refuses to talk; is unhappy, sad, or depressed; and is withdrawn and doesn’t get involved with others. The anxious/depressed behaviors subscale (alpha = .64, .70, and .78) includes items such as the extent to which the child cries a lot; feels worthless or inferior; and is nervous, high-strung, or tense.3 Each measure was standardized in three-month child-age intervals to have a mean of 0 and a standard deviation of 1.
Family Structure States and Transitions
Our analyses incorporated a series of indicators of maternal (cohabiting or marital) unions at each survey wave to capture information about the family structure states and transitions experienced by focal children from birth to age nine. We separated the sample into three groups: biological parents married at child’s birth; biological parents cohabiting at child’s birth; and biological parents not coresident at child’s birth. For each of these three subsamples, we predicted initial (age 3) levels of child behavior problems based on whether, between birth and age 3, a child (1) was in a stable biological-father family; (2) was in a stable single-mother family; (3) experienced his or her biological parents breaking up; (4) experienced his or her biological parents repartnering (i.e., moving (back) in together); (5) experienced his or her mother partnering with another man (social father); and (6) experienced his or her mother breaking up with a social father. The transitions measures are not mutually exclusive because children can experience more than one of these events.
We used both the total number and the specific types of family structure states and transitions (in separate models) experienced between ages 3 and 9 to predict changes in children’s outcomes during that time period (the slope terms in our HLM models). For the specification focused only on the number of family structure transitions, to capture potential nonlinearities, we used dummy variables indicating that a child experienced one, two, or three or more transitions between ages 3 and 9, with no transitions serving as the reference category.
For analyses of associations between particular types of family structure states and transitions and child socioemotional development, we focus on five indicators of specific types of family structure states and transitions that a child may have experienced during the six-year period between ages 3 and 9: ever transitioning into a two-biological-parent family; ever transitioning into a social-father family; ever transitioning into a single-mother family; residing in a stable two-biological-parent family; residing in a stable social-father family; and residing in a stable single-mother family. Residing in a stable single-mother family from age 3 to 9 was the reference category in models for children whose parents were noncoresident at their birth; residing in a stable two-biological-parent family (married or cohabiting) was the reference category for children whose parents were coresident when the child was born.
We used an extensive set of covariates (shown in upcoming Table 2) to predict initial levels of behavior problems at age 3. These include whether the child was female, whether the child was born LBW, the mother’s race (black, Hispanic, and other race/ethnicity, with white as the reference category), the mother’s educational attainment at the time of the birth (less than a high school education and greater than a high school education, with a high school diploma or GED as the reference category), whether the mother was born in the United States, whether the mother lived with both of her biological parents as a teen, the mother’s age at the child’s birth, whether the child was the mother’s first birth, whether the mother worked in the year before the child’s birth, whether the mother received Temporary Assistance for Needy Families (TANF) in the year before the child’s birth, the mother’s household income (natural logarithm) in the year before the child’s birth, the number of children in the mother’s household at the time of the birth, the number of adults in the mother’s household at the time of the birth, whether either parent considered abortion upon learning of the pregnancy, and indicators for city of residence at the time of the birth.
To predict changes in children’s outcomes between ages 3 and 9, we used a parsimonious set of covariates. We employed this strategy because we aimed to estimate the full associations between family structure states or transitions and child well-being and therefore wanted to avoid controlling for any of the mechanisms (employment, income, family size, and so on) through which family structure experiences may influence child well-being. Including such factors would likely bias our family structure states and transitions estimates toward zero. Thus, we controlled only for the following time-invariant covariates when predicting developmental trajectories: child gender, maternal race/ethnicity, maternal education, and maternal age at the child’s birth.
to indicate that the initial level (BP0i) of behavior problems is predicted both by family structure experiences between the focal child’s birth and age 3 (FS0i) and, for models that include the falsification test, by family structure experiences between ages 3 age 9 (FSti), as well as the extensive set of covariates (ECOVS0i). The subsequent linear slope (BP1i) in behavior problems is predicted by family structure experiences between ages 3 and 9 (FSti) and the parsimonious set of time-invariant covariates (PCOVS0i).
The key parameter of interest is Bt1, which represents the per year difference in the rate of change in behavior problems (between ages 3 and 9) that is associated with a given family structure state or transition, relative to the rate of change in the outcome for children who resided in a stable (two-biological-parent or single-parent) family structure throughout the period.
We estimated two separate versions of Eq. (1a). In the first, we used only those family structure experiences that occurred at or prior to the age 3 interview (FS0i) and a rich set of covariates to predict the initial level (intercept) of the outcome. In the second, we also included the (subsequent) family structure states and transitions (FSti) that occurred between the age 3 and 9 interviews to predict the intercept. Although these experiences occurred after the initial level of the behavior problems was measured, this strategy (a falsification test) allows us to rigorously adjust for social selection by accounting for whether children who experienced particular family structure states or transitions had greater (or fewer) behavior problems than reference group children prior to these experiences (Magnuson and Berger 2009; Osborne et al. 2012). We present results from models with and without the falsification test because whereas the falsification test offers a rigorous adjustment for selection into family structure experiences, it may also constitute overcontrolling in the sense that if children’s behavior problems are causally related to later family structure transitions (Hawkins et al. 2007), we would not want to remove this effect from the estimated association between those family structure transitions and subsequent behavior problems. We view estimates from the two models as upper and lower bounds of the associations of interest.
In these models, initial levels of the outcomes are differenced out of the equations, and both within- and between-child variation are used to identify associations for the slope estimates. Thus, in interpreting our results, we primarily focus on associations of family structure experiences with trajectories (slopes) in child behavior problems rather than differences in their initial levels. These estimates could still be biased by the omission of unobserved time-varying characteristics or by persistent characteristics that have time-varying effects. We present weighted descriptive statistics and unweighted HLM results.
Family Structure Transitions
Figure 1 illustrates the extent to which families experienced transitions between the focal child’s birth and the age 9 interview. Overall, 50 % of children experienced at least one transition, with 18 % experiencing one, 16 % experiencing two, and 16 % experiencing three or more transitions;4 44 % of children spent the first nine years of childhood living with both of their biological parents, and 6 % lived with only their mother. Family structure transitions were especially common for children born to unmarried parents. Whereas 29 % of children born to married parents experienced one or more transitions, this was true of 71 % of children born to cohabiting parents, and 76 % of children whose parents were not coresiding at their birth. More than one-quarter of all children born to unmarried parents experienced three or more family structure transitions by age 9. Furthermore, 12 % of children whose parents were married at their birth, 30 % of those whose parents were cohabiting at their birth, and 49 % of those whose parents were not coresiding at their birth had lived with at least one social-father by age 9 (not shown).
Figure 2 depicts the frequency of family structure states and transitions that children experienced between ages 3 and 9. The estimates total to more than 1 because the transition categories are not mutually exclusive. On the whole, 47 % of children lived in a stable, two-biological-parent family from age 3 to 9, 10 % lived in a stable single-mother family, and 0.3 % lived in a stable social-father family. Forty-three percent of children experienced one or more transitions (25 % experienced one, 12 % experienced two, and 6 % experienced three or more transitions; not shown in Fig. 2). With regard to particular types of transitions, 9 % experienced a transition to a two-biological-parent family, 24 % experienced a transition to a single-mother family, and 21% experienced a transition to a social-father family.5
Socioemotional Development and Family Characteristics
Table 1 shows mean levels of behavior problems at ages 3 and 9 by family structure at birth (columns 1–3) and by family structure stability (columns 4–6). Children born to married parents exhibit fewer behavior problems at both ages than children born to unmarried parents. Whereas differences between children whose parents were married and those whose parents were noncoresidential at their birth tend to be largest and most likely to attain statistical significance, children born to cohabiting parents generally exhibit fewer behavior problems than those born to noncoresidential parents and greater behavior problems than those born to married parents. These differences are more pronounced for aggressive and anxious/depressed problems than for withdrawn behavior. With respect to family structure stability, the raw data suggest that the most pronounced differences in behavior problems (at both age 3 and age 9) are between children experiencing family structure transitions and those experiencing only a stable two-biological-parent family.
Given evidence that child behavior problems at ages 3 and 9 differ by family structure at birth, we present in Table 2 descriptive statistics for children who did and those who did not experience one or more family structure transitions between ages 3 and 9, by family structure at birth. We find no statistically significant (p < .05) differences in behavior problems at either age 3 or age 9 between children who did and those who did not experience family structure transitions within each of these categories. Together, the results presented in Tables 1 and 2 suggest that differences in mean behavior problems are more pronounced between children born into different family structures than between children within each group who did and who did not experience family transitions.
Turning to family characteristics, the raw data indicate that early childhood family structure transitions are positively associated with later family structure transitions. Comparing the two columns for each of the three groups, we see that children who experience family structure instability between ages 3 and 9 are also disproportionately likely to have experienced instability prior to age 3. In addition, for children born to noncoresident parents, the raw data reveal very few differences in the sociodemographic characteristics of those who did and those who did not experience family structure transitions between ages 3 and 9. This may partly reflect that the vast majority of these children experienced transitions. In contrast, among children whose parents were cohabiting or (especially) married at their birth, those experiencing transitions were less socially and economically advantaged than children in stable family structures.
Table 3 presents results from HLM models in which we examined the association between experiencing any family structure transition (vs. living in a stable family structure) and child behavior problems from ages 3 to 9. We show estimates from two models for each outcome. The first model adjusts only for family structure experiences prior to age 3 in predicting the age 3 intercept; the second also adjusts for subsequent family structure experiences (the falsification test). The coefficients are interpreted as the standard deviation (SD) difference in the intercept—or the per year SD difference in the slope—that is associated with a particular family structure experience.
Panel A of Table 3 shows results for children born to parents who were not coresiding at their birth. The intercept estimates from Model 1 reveal that children whose parents moved in together during their first three years of life had lower initial (age 3) levels of aggressive and anxious/depressed behavior problems than those who experienced a stable single-mother family. However, this was true only for children who experienced parental union formation without subsequent dissolution (i.e., whose parents did not subsequently break up). The intercept results from Model 2 indicate that children born to noncoresident parents who went on to experience three or more family structure transitions between ages 3 and 9 had fewer aggressive, withdrawn, and anxious/depressed behavior problems at age 3 than did children who did not go on to experience family structure transitions. In other words, we find evidence that children born to noncoresiding parents who subsequently experienced multiple family structure transitions (which typically began with maternal repartnering) started out with fewer behavioral problems. Again, this could be interpreted as reflecting either selection into family structure transitions (based on unobservable characteristics that are correlated with child behavior problems) or the effect of early child behavior on subsequent family structure transitions.
The slope estimates from Model 1 suggest no significant differences in behavior problem trajectories between children who were in stable single-mother families and those who experienced one or more transitions between ages 3 and 9. The slope estimates from Model 2 suggest that having experienced one transition is associated with an increase in aggressive behavior problem of 0.02 SDs per year, or roughly 0.14 SDs (0.024 SDs per year × 6 years) in total between ages 3 and 9, whereas experiencing three or more transitions is associated with approximately 0.32 SDs greater aggressive behavior problems by age 9 (both marginally significant at p < .10). We find no other statistically significant differences between any of the coefficients.
Considering both initial levels of and trajectories in behavior problems, for example, these results indicate that children born to noncoresidential parents who experienced three or more family structure transitions between ages 3 and 9 have roughly 0.02 SDs (−0.30 + 0.32) more behavior problems at age 9 than their counterparts who experienced a stable single-mother family between ages 3 and 9. This is not a substantively important difference. Thus, the results from both specifications (Models 1 and 2) yield the same conclusion: no meaningful association exists between family structure transitions and socioemotional development for children born to noncoresidential parents.
For children born to cohabiting parents, we find that experiencing two or more transitions is associated with increased aggressive behavior problems in Model 1. However, we find no differences between any of the intercept or slope coefficients after the falsification test is included in the model (Model 2). This finding suggests that children born to cohabiting parents exhibit similar levels of behavior problems, regardless of their subsequent family structure experiences, after initial differences in child behavior problems—which may be associated with subsequent family structure transitions—are controlled. Depending on whether the falsification test is interpreted as rigorously adjusting for selection or as overcontrolling, these results suggest that experiencing multiple transitions has a modest or null association with child behavior problems for children born to cohabiting parents.
For children born to married parents, we find consistent evidence that experiencing one family structure transition (most frequently parental breakup) between ages 3 and 9 is associated with greater (aggressive and withdrawn) behavior problems. Children who were born to married parents, were still living with both parents at age 3, and experienced one family structure transition between ages 3 and 9 exhibited approximately 0.27 SDs greater aggressive behavior problems and 0.20 SDs greater withdrawn behavior problems at age 9 than children who resided in a stable two-biological-parent family between ages 3 and 9. We find no other statistically significant differences between any of the coefficients.
Table 4 presents slope coefficients from HLM models focusing on specific types of family structure states and transitions. Among children born to noncoresidential parents, we find no statistically significant differences in behavior problem trajectories between those who lived in a stable single-mother family and those who experienced other family structure states or transitions between ages 3 and 9. Likewise, we find no statistically significant differences in behavioral trajectories among children experiencing any of the other family structure states and transitions.
For children born to cohabiting parents, we find that those whose parents reunited when their child was between ages 3 and 9 after having previously broken up (as indicated by entering a biological-father family between ages 3 and 9) had a greater increase in aggressive (0.19 SDs by age 9) and withdrawn (0.23 SDs) behavior problems than children who resided in a stable two-biological-parent family from ages 3 to 9 (both marginally significant at p < .10). They also had a greater increase in withdrawn behavior problems than children whose mothers repartnered with a social father between ages 3 and 9. There were no other statistically significant differences by family structure experiences.
For children born to married parents, the results from Model 2 demonstrate that those whose mothers repartnered with a social father and those who entered a single-mother family (experienced their mother’s breakup with either their biological or social father) exhibited greater growth in aggressive behavior problems than did children residing in a stable social-father family from age 3 to age 9. In addition, children who were born to married parents and went on to experience a stable single-mother family from ages 3 to 9 (whose parents had broken up prior to age 3, and whose mother did not repartner) exhibited greater growth in withdrawn behavior problems than those who resided in a stable two-biological-parent family. Finally, those who experienced maternal repartnering with a social father between ages 3 and 9 exhibited greater growth in anxious/depressed behavior problems than those who lived stably with both of their biological parents.6
Our analyses yield new information about family instability and its associations with child socioemotional development among relatively disadvantaged urban families. First, we document that children born to unmarried urban parents overwhelming experience family structure transitions by age 9, with approximately one-half experiencing two or more transitions and approximately one-quarter experiencing three or more transitions by that time. Furthermore, few differences exist in the number of transitions experienced, on average, by children born to cohabiting parents and those born to noncoresident parents, despite the fact that the types and sequencing of such transitions differ.
Second, we show that children who experience early family structure transitions (by age 3) are also disproportionately likely to experience later transitions. In other words, early family instability appears to predict later family instability.
Third, we find larger differences in both initial (age 3) and later (age 9) behavior problems based on the child’s family structure at birth than based on the family structure transitions children experienced within each of these groups. That is, family structure at birth appears to be more closely associated with child socioemotional development than does family structure instability (after family structure at birth is taken into account). This finding may imply that differential selection into family structure at birth is a substantial driver of observed correlations between family instability and child socioemotional outcomes.
Fourth, our HLM results for children born to cohabiting or noncoresident parents reveal virtually no substantively meaningful evidence that subsequent family structure experiences are associated with socioemotional development through age 9. The only consistent exception to this pattern is that children born to cohabiting parents who subsequently break up and then reconcile exhibit modestly greater behavior problems than children whose parents consistently lived together. Prior research has paid scant attention to how on-again/off-again cohabitation between children’s biological parents may be linked to child well-being. This topic deserves future attention.
Finally, we find associations between family instability and poorer socioemotional development for children born to married parents. However, these associations largely reflect the influence of parental breakup; we find little evidence that socioemotional trajectories differ for children with various family structure experiences subsequent to their parents’ breakup. For the most part, our results are quite consistent across both our standard HLM models and HLM models that include a falsification test that effectively adjusts for differences in children’s early socioemotional adjustment that are associated with both later family structure states and transitions and socioemotional adjustment over time, although we find slight differences in effect sizes and statistical significance in a few cases.
Our analyses have several limitations. First, because family structure information was gathered only at (and about) the time of each FFCW interview, rather than continuously between interviews, we may be missing relationship transitions occurring between interviews and thus may be underestimating the instability that sample children experience.7 Second, our measures were drawn solely from mothers’ reports. To the extent that systematic differences exist in mothers’ reporting of their relationship status or their children’s behavior problems, our results will be biased. Third, statistical power was limited in instances in which relatively few children experienced a particular family structure trajectory. Nonetheless, we were struck by the paucity of statistically significant results from a wide variety of statistical comparisons. Fourth, because it is not clear whether or how associations between family structure experiences and child behavior problems may differ between the (relatively disadvantaged) FFCW sample and a more representative sample, we caution that our results cannot be generalized to nonurban or more-advantaged groups. Finally, associations between family structure experiences and child outcomes may be subject to additional heterogeneity through factors such as child’s age at the time of a particular transition, child’s sex, child’s race/ethnicity, and nonresident father involvement (for children not living with their father). We were unable to test for such heterogeneity because of the small cell sizes, given our emphasis on both examining differences in associations based on family structure at birth and on modeling the full range of family structure states and transitions that children could have experienced. Nonetheless, it is possible that other types of heterogeneity matter. On the whole, our results do not stand in contrast to prior literature indicating that, on average, children in stable, two-biological-parent families fare better on a wide range of outcomes than those in all other family types (Amato 2005, 2010; Waldfogel et al. 2010). Rather, our results suggest that at least with respect to socioemotional development, these associations are heterogeneous by family structure at birth, which is closely linked to subsequent family structure experiences, and also that specific types and sequencing of such experiences may matter. Our results are quite consistent with recent work that found larger negative associations of parental breakup than other types of family structure transitions with child outcomes (Lee and McLanahan 2015). They are also consistent with findings from Ryan and Claessens’ (2013) and Ryan et al.’s (2015) analyses of NLSY data (a more representative sample, but also an older data set, than ours), which indicated that associations between family instability and child outcomes are strongest among children born to married and more-advantaged families.
Together, these findings provide support for theories that family instability may have a lesser influence on disadvantaged than advantaged children, potentially either because family structure transitions are more common, predictable, and normative or because they are associated with smaller changes in resources, parental investments, stress, and conflict among the former. Furthermore, our results suggest that after analyses are disaggregated by family structure at birth and early family structure experiences are fully accounted for, little support exists for arguments that family instability, in general, has an adverse causal effect on child socioemotional development, with the potential exception of parental breakup for children born to married parents. We are unable to assess whether this reflects social selection, differences in child endowments at birth (Harris 2011), or early family structure experiences. Unfortunately, many facets of child development cannot be consistently measured in ways that allow for estimation of trajectories starting at birth.
It will be important to assess whether future research using different data sources, age groups, and a longer time horizon can replicate our results. If so, then we argue that family scholars should perhaps rethink the potential implications of family structure and associated instability for child well-being. Perhaps, for example, the influence of family structure (net of essential characteristics such as access to financial resources) has shifted over time, as nonmarital births and “nontraditional” family forms have become normative family experiences, especially in disadvantaged populations.
The Fragile Families and Child Wellbeing Study is funded by NICHD grant numbers R01HD36916, R01HD39135, and R01HD40421, as well as a consortium of private foundations and other government agencies. This research was supported by NICHD Grant No. K01HD054421 (to Berger) and by the Robert Wood Johnson Scholars in Health Policy Research Program (for Bzostek), as well as by funding from the Institute for Research on Poverty and the Waisman Center (NICHD Grant No. P30 HD03352) at the University of Wisconsin–Madison. We are grateful to Anne Solaz, Laurent Toulemon, seminar participants at the French National Institute for Demographic Studies (Institut national d'études démographiques; INED), and participants at the 2012 American Sociological Association and 2013 Population Association of America annual meetings for helpful comments on earlier versions of this article.
Graham (2009) and Johnson and Young (2011) recommended using all cases—including those with imputed outcome data—in statistical analyses. As such, we conducted supplemental analyses using fully imputed data. We also conducted supplemental analyses using listwise deletion (complete case analyses). Results from these supplemental analyses are available from the authors upon request. Whereas there were some differences in the size and significance of estimates produced by each strategy, the direction and overall pattern of each set of estimates from the supplemental analyses were quite consistent with those from our primary analyses. Our overall conclusions would not have changed regardless of which strategy was used.
We conducted a comparison of descriptive statistics for all variables used in our analyses based on (1) complete case data (listwise deletion), (2) fully imputed data, and (3) imputed data excluding cases for which outcomes were imputed. For the most part, the characteristics of the three samples are quite similar, and this is particularly true when comparing the complete case data with the fully imputed data. The most notable differences are that cases included in our analysis sample had slightly lower levels of behavior problems and slightly higher rates of stable biological-father family structure both between birth and age 3 and between ages 3 and 9. These results are available from the authors upon request.
The specific items included in these three subscales vary depending on the child’s age. Those listed here are drawn from the measures when the child was approximately age 5.
Children who were observed transitioning directly from one type of two-parent family to another across waves were coded as also having experienced a transition into a single-mother family.
Again, children who were observed transitioning directly from one type of two-parent family to another across waves were coded as also having experienced a transition into a single-mother family.
Coefficients for the full sets of control variables for Tables 3 and 4 are available from the authors upon request. On the whole, the covariates predict behavior problems in expected ways, such that social and economic disadvantage are associated with greater child behavior problems.
At the age 5 interview, mothers were asked to provide information about romantic relationships that they had formed and dissolved between the age 3 and age 5 interviews. These data suggest that very few mothers lived with more than one partner between those survey waves (Bzostek et al. 2012). Nonetheless, we have likely underestimated the number of transitions that some children experienced.