Today's young adults have diverse union experiences; some enter enduring marital or cohabiting unions at young ages, but many delay or dissolve their unions or remain single. Childhood family instability—defined as parents' transitions into or out of romantic coresidential unions—offers one explanation for why some people are more likely than others to enter and exit unions. We evaluate whether this family instability hypothesis—a union-specific version of the general hypothesis that instability affects people across multiple life domains—can explain Black and White young adults' union formation and dissolution. Using data from the Panel Study of Income Dynamics' Transition into Adulthood Supplement (birth cohorts 1989–1999), we find that the marginal effects of childhood family instability on cohabitation and marriage are weaker for Black than for White youth. Further, Black–White differences in childhood family instability's prevalence are small. Consequently, novel decompositions that account for racial differences in instability's prevalence and marginal effects reveal that childhood family instability contributes little to Black–White inequality in young adults' union outcomes. Our results challenge the generalizability of the family instability hypothesis across racialized groups in the union domain. Explanations for Black–White differences in young-adult marriage and cohabitation reside beyond childhood family dynamics.
Families are dynamic. More than one third of today's young adults experienced family instability during childhood, defined as transitions into or out of parents' romantic coresidential unions (Brown et al. 2016).1 Researchers have hypothesized that these transitions affect children in many domains, including their socioemotional, academic, and family-building behaviors (Cavanagh and Fomby 2019). We focus on family-building behaviors, interrogating the generalizability of one specific family instability hypothesis: that experiencing parental union transitions during childhood causes young adults to enter and exit cohabitations and marriages, replicating their parents' complex union patterns (Cavanagh et al. 2008; Fomby and Bosick 2013; Thorsen 2017). Understanding these entries and exits among today's youth is important because by young adulthood's end, at age 29 (Manning 2020:802), their relationship biographies are more variable than those of earlier cohorts. More have formed and dissolved cohabitations, fewer have married, and many have remained single throughout young adulthood (Manning 2020; Zhang and Ang 2020). Understanding inequalities in union entries and exits across subpopulations is also important because they predict inequalities in health and economic well-being (McLanahan and Percheski 2008).
Black children are more likely than White children to experience family instability (Brown et al. 2016). The family instability hypothesis implies that this difference will generate racial inequalities in young-adult unions. Yet, this implication has not been tested. Nevertheless, public discourse often references Black–White differences in childhood family disruption to explain racial inequalities in unions and other outcomes, including poverty, educational attainment, and incarceration (Fremstad et al. 2019; Haskins 2009). Misattributing racial inequalities in unions to childhood family instability risks supporting White supremacy by misdirecting policy dollars toward increasing family stability and by stigmatizing family experiences more common among non-White populations owing to historical and contemporary exclusion and oppression (Williams 2021). Consequently, we test the implication of the family instability hypothesis that because Black children experience more family instability than White children, this instability should generate more union entrance and exit among Black than among White young adults. In so doing, we also address an open puzzle in the literature on family inequalities: Why are descriptive racial inequalities in union entrance inconsistent with the family instability hypothesis? Entry into cohabitation and marriage is not more common among Black than among White young adults today, despite the former's greater exposure to childhood family instability (Bloome and Ang 2020; Hemez 2018).2 We explore four explanations for this inconsistency.
First, childhood family instability's effects may be weaker for Black than for White people's unions, as has been shown for nonmarital birth (Wu and Martinson 1993). Second, childhood family instability's causal effects may be small for all people, similar to how childhood family structure's causal effects are small relative to its descriptive associations (McLanahan et al. 2013). Third, different qualities of childhood family instability (e.g., its direction into vs. out of parental unions) might differently affect young-adult unions, and the qualities commonly experienced by Black children might be less consequential than those commonly experienced by White children. Fourth, racial differences in instability's prevalence may be too small to meaningfully shape racial differences in young-adult unions.
By evaluating these four explanations, we make three contributions to the literature. First, we assess childhood family instability's effects on young-adult union formation and dissolution and whether these effects are equally large among racialized groups that have been socially constructed into dominant versus subordinate positions within the hierarchical system of race relations (Seamster and Ray 2018). We extend prior research on racially heterogeneous effects (Fomby and Cherlin 2007) by considering new outcomes and using more rigorous methods to account for group- and time-varying selection into childhood family instability. Second, we interrogate whether qualitative differences in family instability explain racially heterogeneous effects, extending research documenting Black and White children's diverse family experiences (Burton and Jayakody 2001). We distinguish the effects of three qualities of family instability: timing (experienced during early childhood [before age 5] vs. later childhood), volume (experienced during later childhood with vs. without instability in early childhood), and direction (experienced as parental union entrance vs. exit) (Cavanagh et al. 2008; Lee and McLanahan 2015). Third, we consider childhood family instability's population-level effects, accounting for racial differences in instability's prevalence and not simply its marginal effects, using a novel decomposition. Although racial differences in childhood family instability exist, they may not be very consequential in aggregate. In such a case, childhood family instability should not be considered a major axis of stratification between racialized groups.
In short, the family instability hypothesis regarding union outcomes may rest on faulty assumptions about the homogeneity of instability's effects across groups (when it may be less consequential for Black than for White people), the homogeneity of instability experiences across groups (when heterogeneous experiences may generate heterogeneous effects on young-adult unions), or the magnitude of instability differences across groups (which may be too small to cause meaningful disparities). We address these issues using data from the Panel Study of Income Dynamics' Transition into Adulthood Supplement. These data are ideal for understanding unions among today's young adults (born in 1989–1999) because they include full cohabitation and marriage histories linked to rich information about childhood family dynamics and characteristics from birth. Our results shed new light on the causes and consequences of family inequalities.
The Family Instability Hypothesis
The family instability hypothesis anticipates that children are affected by changes in their parents' romantic coresidential unions (Cavanagh and Fomby 2019; Hadfield, Amos et al. 2018). Within family demography, Wu and colleagues proposed the family instability hypothesis to explain why living with an unpartnered parent during childhood predicted women's early nonmarital childbearing (Wu 1996; Wu and Martinson 1993). According to this hypothesis, the family-structure transitions preceding or following spells of living with an unpartnered parent—not the single-parent structure itself—cause early nonmarital childbearing because disruptions to family systems generate stress, uncertainty, and family-relationship ambiguity, thereby leading adolescents to initiate sexual relationships. This hypothesis challenged explanations based on parenting practices in single-parent families, such as socialization toward nonmarital sex (Hofferth and Goldscheider 2010; Wu and Martinson 1993; Wu and Thomson 2001).
Childhood Family Instability and Young-Adult Union Inequalities
Wu and colleagues' work provided a novel explanation of nonmarital childbearing's intergenerational persistence. By shifting the conceptual model from family structure to family dynamics, this work also initiated inquiry into the intergenerational transmission of family instability. Researchers began investigating whether parents' union transitions (usually through divorce and remarriage) contributed to children's union transitions, elevating their risks of entering and exiting romantic coresidential relationships. The resulting research confirmed that childhood family instability predicted more frequent union formation and dissolution during early adulthood (by age 30; Amato and Patterson 2017) and across the life course (Kamp Dush et al. 2018; Wolfinger 2000). Both union formation and dissolution are centrally important explananda for the theory that children grow up to replicate their parents' union instability. Nevertheless, some studies document instability’s effects on young-adult union formation only, hypothesizing that these effects indirectly illuminate dissolution patterns because early union entry is a risk factor for subsequent dissolution (Fomby and Bosick 2013; Thorsen 2017). Although cohabitation and marriage remain distinct types of unions in the United States, researchers have found that childhood family instability positively predicts both types, perhaps because both provide young adults with pathways toward independent households, separate from the complex dynamics in their parental homes (Ryan et al. 2009; Teachman 2003).
The aforementioned literature was generally inattentive to whether the relationships between childhood family instability and later union formation and dissolution held across racialized groups. Yet, research on other outcomes demonstrated that family instability's effects were weaker for Black than for White people (for a review, see Fomby 2022). Wu and colleagues (1993, 1996) documented that family instability's effects on nonmarital childbearing were weaker for Black women than for White women. Family instability also more weakly predicts externalizing and risk-taking behavior, diminished academic achievement, and early school-leaving for Black than for White youth (Fomby and Cherlin 2007; Fomby et al. 2010; Lee and McLanahan 2015; Perkins 2019). In a meta-analysis of research on parental divorce's effects on 15 measures of adult well-being, Amato and Keith (1991) concluded that these effects were weaker for Black than for White people.
Why might family instability's effects on union outcomes be weaker for Black than for White young adults? One explanation is that the mechanisms generating these effects operate differently in Black and White families, either muting instability's effects for Black young adults or amplifying them for Whites. For example, family stress theory suggests that disruptions trigger emotional, financial, and social hardships that lead children to use cohabitation or marriage to exit parental homes full of complicated interpersonal dynamics (Hill 1949). Yet, family stress explains instability's effects on many outcomes less effectively among Black than among White youth, arguably because interpersonal and institutional racism expose Black people to compounding adversities that reduce each additional stressor's impact (Brand et al. 2019; Fomby et al. 2010; McLoyd et al. 2000). Alternatively, Black children's more frequent contact with extended kin may offset parents' diminished social control following union transitions, which might otherwise push youth toward early and unstable unions (McLoyd et al. 2000; Sarkisian and Gerstel 2004). Instability's effects also might be weaker among Black than among White people because White people face more stigma for this instability or have fewer skills to cope with it (McLanahan and Percheski 2008:269; Perkins 2019:530). People have more difficulties coping with events that they do not expect to face (Brand et al. 2019).
Two other factors imply weaker associations between childhood family instability and young-adult unions among Black than among White people. Selection into childhood family instability might be weaker among Black people, and the qualities of family instability that Black and White children experience may differ. We expand on these points below.3
Identifying Childhood Family Instability's Effects
Outcome differences between people who do versus do not experience childhood family instability could reflect instability's causal force, but they also could reflect differences that drive selection into instability. Identifying the causal part of these outcome differences is challenging, particularly when selection might differ across groups (Fomby et al. 2021; Raley and Sweeney 2021:89). For example, preexisting economic differences between people who do versus do not experience instability may be smaller among Black than among White people (Perkins 2019). Instability-associated differences would then reflect smaller economic inequalities among Black than White people (and instability's effects would be less overstated among Black people). Ignoring selection differences could lead instability's association with young-adult outcomes to look weaker among Black people even if its causal effect were not. We integrate selection differences in terms of instability's observed covariates via group-stratified models. This approach improves upon research that assumes homogeneous selection but remains vulnerable to unobservable characteristics, particularly those characteristics that are not highly correlated with our observed covariates. Beyond racial differences in selection, we incorporate time-varying selection to avoid conditioning on observed colliders and causal paths from early instability to young-adult outcomes while adjusting for observed confounders (Hernán et al. 2000).
Qualities of Childhood Family Instability
We disentangle three qualities of instability: timing (in early vs. later childhood), volume (in later childhood with vs. without early childhood instability), and direction (union entrance vs. exit). The qualities that Black children experience more often than White children could be less consequential than those more commonly experienced by White children (Burton and Jayakody 2001; Jarret and Burton 1999). This is the third explanation for the discordance between the family instability hypothesis' predictions and observed racial differences in young-adult unions. It expands the first (that the discordance reflects weaker instability effects among Black than among White people) by specifying that the heterogeneous effects might reflect different instability qualities.
Black children are more likely than White children to experience family instability early, before age 5 (Brown et al. 2016). For the third explanation to hold in terms of timing, then, instability's effects must be weaker during early than later childhood. Two theories support this explanation. Resource loss due to family instability might be particularly consequential during later childhood because it pushes youths out of college and into unions (McLanahan and Bumpass 1988). Likewise, child–parent conflict might be particularly consequential during later childhood, given high adolescent stress and limited recovery time before young-adult outcomes (Branje 2018).
Yet, other theories suggest that instability's effects may be stronger during early than later childhood. Considering resource loss again, loss at young ages might place children on adverse developmental trajectories (Schoon et al. 2012). Considering stress, because young children are more dependent on their caregivers than older children, they may be more affected by changes in caregivers' well-being (Cavanagh and Fomby 2019). Instability in early and later childhood also could be equally consequential because economic and emotional shocks affect both younger and older children.
In short, theory regarding timing is ambiguous. Empirical evidence is also inconsistent across outcomes. Prior research suggests that early and later instability are equally consequential for outcomes such as substance use and academic achievement, but later instability is more consequential for outcomes such as teen dating and depression (Cavanagh et al. 2008; Cavanagh and Fomby 2019).
Family instability's volume also may shape young-adult unions. Prior research on the number of family transitions shows that each transition matters either as much as or less than the first (Cavanagh et al. 2008; Wu and Martinson 1993; but see Bzostek and Berger 2017).
However, these findings rely on counts of family transitions, which are confounded with their timing: higher order transitions occur later in life. To avoid this confounding, we measure instability's volume via the addition of transitions in later childhood on top of transitions in early childhood (vs. transitions in later childhood only). We disentangle transitions' effects in early versus later childhood (a timing effect) from transitions' effects in one versus both stages (a volume effect). By our definition, Black children likely experience lower volumes of family instability than White children, given their lower mean number of transitions among those experiencing at least one (Brown et al. 2016). Thus, for the third explanation to hold in terms of volume, lower volumes should be less consequential than higher volumes for young-adult unions.
Childhood family instability can also be characterized by its direction. Black children are more likely to experience entrances into parental unions, whereas White children are more likely to experience parental union exits (Brown et al. 2016). For the third explanation to hold in terms of direction, then, the effects of parental union entries must be weaker than the effects of exits.
Prior research supports this expectation for other outcomes. For childhood and adolescent socioemotional outcomes, parental union formation has smaller effects than parental union dissolution (Bzostek and Berger 2017; Lee and McLanahan 2015). Yet for young-adult marriage and cohabitation, we expect parental union formation and dissolution to be equally consequential because of role modeling and family stress. Research suggests that both parental repartnering and divorce hasten young-adult union formation, particularly cohabitation (Amato and Kane 2011; Sassler et al. 2009; Teachman 2003; Thornton 1991).
Childhood Family Instability's Prevalence and Population Aggregated Effects
Black and White children experience different qualities of family instability. However, overall, Black children are more likely than White children to experience any family instability (Brown et al. 2016; Raley and Wildsmith 2004). This prevalence difference could generate racial differences in young-adult outcomes, even if instability's marginal effects were small. The logic is simple: small marginal effects, when experienced by a large share of people, scale up to a large population effect at the aggregate level.
Racial differences in the prevalence of two-parent families are large, reflecting institutional and interpersonal racism's effects on family life (Burton et al. 2010). Today's Black young adults spent approximately 33% of childhood with two biological parents, compared with 78% for White young adults (Cross 2020). Consequently, differences in childhood family structure distributions may contribute to sizable racial inequalities in adult outcomes (McLanahan and Percheski 2008:269) because of family structure's role as a mediator of racism's effects. However, differences in childhood family instability distributions may not contribute to sizable racial inequalities in adulthood because these distributions may be fairly similar across racialized groups. Black children are particularly likely to live with stably single parents and unlikely to experience parental divorce (Brown et al. 2016). Thus, racial differences in childhood family instability may be too small to generate substantial racial differences in young-adult outcomes.
Data and Measures
We use data from the Panel Study of Income Dynamics (PSID) and its Transition into Adulthood Supplement (TAS) to assess childhood family instability's role in Black and White young adults' unions. These data are ideal for three reasons. First, the TAS's 2017 wave (TAS 2017) captures the timing of first cohabitation and marriage. Second, the data include an oversample of families within 200% of the federal poverty level in 1967, ensuring adequate representation of African Americans. Third, and most importantly, by combining PSID and TAS data, we follow people prospectively from birth.
The PSID began in 1968 with a nationally representative sample of 4,802 U.S. families whose descendants through birth or adoption have been followed since. We limit our study to non-Hispanic Black and White respondents because of the small number of respondents from other groups.4 TAS-2017 supplemented the core PSID interview to collect in-depth information about respondents aged 18–29 (born in 1989–1999). We study the TAS-2017 marriage and cohabitation histories, supplemented with core PSID data beginning in 1985 (to capture prebirth covariates for the oldest cohort). Our results should generalize to the population of young adults in this cohort who descended from families residing in the United States in 1968.5 Our sample includes 1,817 (882 Black and 935 White) of 2,526 respondents who completed the TAS-2017 survey. We omit those who were not non-Hispanic Black or White (n = 381), those in the 1997 immigrant refresher (n = 19),6 and those missing data on our measures (n = 309). Multiply imputing missing data leaves our conclusions unchanged.
To capture union formation, we measure respondents' ages in months at first cohabitation, first marriage, and first union (cohabitation or marriage). Unions include same-sex and different-sex partners. To capture union dissolution, we construct a three-category measure of the first-union outcome: (1) intact as of the last observation; (2) ended, and the respondent did not repartner; and (3) ended, and the respondent repartnered. For cohabitation, we disaggregate further: (1a) cohabitation remains intact and (1b) cohabitation transitioned to intact marriage. We construct these categorical measures because the TAS-2017 instrument does not capture first-cohabitation end dates for those cohabiting more than once. Thus, we cannot measure age at first cohabitation/union dissolution.
We present descriptive statistics for three union-formation outcomes (forming a cohabitation, a marriage, and either union type) and four union-dissolution outcomes (dissolving a cohabitation, a marriage, and either union type, plus transitioning a cohabitation into marriage). We present regression results for two union-formation outcomes (forming a cohabitation and forming a marriage) and one union-dissolution outcome (dissolving any union). Too few respondents entered multiple unions to permit analysis of that outcome.
In each year, we capture childhood family instability with indicators for whether and how the union status of the child's coresident parent(s) changed from the prior year (Panel Study of Income Dynamics 2020).7,8 We do not count transitions of parental cohabitations into marriages with the same partner as parental union formations or dissolutions. Table A1 (online appendix) contains detailed summary statistics.
We call instability experienced between birth and age 4 early childhood instability and instability experienced between age 5 and entry into the risk set for young-adult cohabitation and marriage later childhood instability.9 We select age 5 as the cut point between early and later childhood based on prior research showing that family influences are particularly strong before age 5 because children spend little time in school before then (Cavanagh and Fomby 2019).
We adjust for factors generating childhood family instability to disentangle instability's effects from the effects of instability-generating factors. We select a rich set of covariates on the basis of a comprehensive literature review. Many are commonly used in recent studies (e.g., income) and some are logical extensions (e.g., wealth). We measure characteristics at birth (adjusting for selection into early childhood instability) and age 5 (adjusting for selection into later childhood instability).10 We do not include covariates such as respondents' educational attainment or preunion childbearing because they are intermediate outcomes between childhood instability and young-adult unions, making it inappropriate to control for them in estimating total effects (Robins et al. 2000).
We measure the following for each respondent at birth: gender (male or female); racialized group (Black or White); birth cohort (1989–1999); mother's and father's ages (separately; <20, 20–24, 25+, or missing); living arrangement (living with two, one, or zero parents); religion (Catholic, other Christian, or non-Christian); household employment (either head or spouse employed, neither employed/head not employed and no spouse, or missing); household structure (household headed by mother/father or someone else); region (South or non-South); parents' highest education level (less than high school, high school, some college, or bachelor's degree or more); income (averaged across the five years before birth; trimmed at the top and bottom 2% of the distributions within head's age, gender, and survey year to reduce outliers' influence; transformed with the inverse hyperbolic sine); wealth (averaged, trimmed, and transformed like income); homeownership (yes or no); receipt of Aid to Families with Dependent Children (AFDC) or Temporary Assistance for Needy Families (TANF) (yes or no); and other welfare receipt, including Social Security, Supplemental Security Income, unemployment, workers' compensation, public housing, or food stamps (yes or no). We interact income and living arrangements, income and wealth, and homeownership and living arrangements to balance the covariates' distributions between people “treated” and “untreated” with family instability.
We measure the preceding variables that vary substantially over time again at age 5: household living arrangements and employment; income and wealth (averaged between the child's birth and age 4 and transformed); and homeownership, AFDC/TANF receipt, and other welfare receipt. We also measure whether parental education increased between the child's birth and age 4 and extended-family coresidence (lived in a family between birth and age 4 with grandparents or other relatives vs. did not). Table A2 (online appendix) contains summary statistics.
Methods for Estimating Marginal Effects
To understand how young-adult unions respond to the “treatment” of childhood family instability, we use marginal structural models (MSMs). MSMs are appropriate for estimating the effects of treatments that unfold over time because they adjust for time-varying covariates that may affect and be affected by time-varying treatments via weighting rather than conditioning, thereby avoiding biases due to overcontrolling or collider stratification (Hernán et al. 2000, 2002). (See section A1 of the online appendix for explanations of these biases.)
The analysis proceeds in two stages. First, we estimate inverse probability of treatment weights (IPTWs) (discussed in section A1 of the online appendix). The goal is to estimate IPTWs that can be used to reweight the sample to represent a counterfactual population in which treatment (parental union transition) at each time is unconfounded by observed covariates; people whose covariate histories are underrepresented or overrepresented in their current treatment group are up-weighted or down-weighted, respectively, to balance covariates across groups. Second, we use the IPTWs to weight models that connect our treatments and outcomes, as we discuss later. These models are called MSMs. Causal inference is impossible without assumptions (Imbens and Rubin 2015). A key assumption for MSMs is no unobserved confounders; the probability of assignment to each treatment group at each time depends on past treatment and measured covariate history but not unobserved confounders.11 Using IPTWs in MSMs removes bias from observed confounders. Unobservable characteristics will be balanced across treatment groups only to the extent that they correlate with observable characteristics. Unobservable characteristics that are not highly correlated with the rich set of observed confounders used to estimate the IPTWs undermine causal identification.
We use the IPTWs in the MSMs for our outcomes. For the cohabitation and marriage outcomes, we use discrete-time event-history models predicting the hazard h of first cohabiting or marrying at age t (in months) for person i of racialized group r via logistic regressions:
where is the jth predictor. We stratify the regressions by racialized group to allow the parameters to vary across groups. Our predictors include an intercept, a cubic in age (and lower order terms), linear cohort, linear age interacted with linear cohort, gender, gender interacted with cohort, indicators for early and later childhood family instability, and interactions between those indicators and gender. We include gender interactions because women typically form unions earlier than men (Manning 2020), but we do not anticipate gender differences in instability's effects (see footnote 3). We weight the regressions with the IPTWs to adjust for selection while not conditioning on observed colliders or controlling away observed paths between early instability and young-adult unions that flow through confounders of the association between later instability and young-adult unions. To obtain interpretable results, we use our discrete-time event-history estimates to calculate the cumulative probabilities of ever cohabiting or marrying by age 28.8, the oldest age we observe (see section A1, online appendix) (Bloome and Ang 2022). Because racial differences in instability's effects are similar across ages (as discussed later), considering younger ages leaves our results unchanged.
For the union-dissolution outcome, we use a similar approach, modeling the outcome using regressions weighted by IPTWs and stratified by racialized group. However, because data limitations force us to use a binary dissolution measure (which ignores dissolution timing), we use logistic regressions on person-level data. Our predictors are the same as for our union-formation outcomes, except we exclude age indicators (including the age–cohort interaction) because we lack information on dissolution timing. The event-history models are superior to the simple binary models because they adjust for right-censoring, which is common in our data. At the last observation, many respondents were too young to have entered or exited unions. Failure to account for this censoring in our dissolution analysis biases our results toward the experiences of people who formed unions at young ages and people who were born earlier.
Key Marginal Effects
These MSMs using IPTWs are sufficient to estimate childhood family instability's marginal effects (and racial differences therein) under standard assumptions (Hernán et al. 2001). We estimate several marginal effects. We begin with the total effects in each racialized group r (Black or White) and gender g (men or women):
where is the potential union outcome for person i when experiencing childhood family instability during both early and later childhood; is the potential union outcome when experiencing childhood instability during neither early nor later childhood; and is the expectation operator. We estimate these expectations for each racialized-by-gender group from our MSMs.12
We estimate several additional effects to distinguish the influence of instability's timing and volume (see section A1 of the online appendix for formal definitions). We estimate early childhood effects (, the effect of experiencing instability only during early childhood) and conditional later childhood effects (, the effect of experiencing instability during later childhood after experiencing instability during early childhood). These two effects sum to the total effects. Distinguishing them illuminates how instability's timing matters. However, the contrast between early and conditional later childhood effects also contains information about instability's volume. To disentangle timing and volume, we examine two additional effects. Life-stage effects () capture the difference between the effect of family instability only in later childhood () versus only in early childhood (). Life-stage effects reveal whether instability is more consequential if it occurs during early versus later childhood. Compounding effects () capture the difference between the effect of instability in later childhood with versus without instability during early childhood. Compounding effects provide insight into instability's volume, revealing whether instability in later childhood is more impactful if it occurs with versus without early childhood instability.
We examine how each of our key marginal effects (, , , , , and ) differs between racialized groups for each gender (e.g., ). If the estimated differences are large, then the data suggest that the family instability hypothesis regarding young-adult union outcomes is not generalizable across racialized groups.
Population Aggregated Effects
Childhood family instability's population aggregated effects capture how much childhood family instability contributes to racial differences in young-adult union outcomes, incorporating its marginal effects and prevalence. We distinguish three contributions to the racial difference in each union outcome.
The first contribution is an aggregated rate effect. It quantifies how much of the outcome difference is driven by racial differences in marginal effects. The second contribution is an aggregated composition effect. It quantifies how much of the outcome difference is driven by racial differences in the prevalence of different types of childhood family instability. These two contributions sum to instability's population aggregated effect on the racial difference in each outcome. The third contribution is a residual, called remaining differences. It quantifies how much of the outcome difference is not causally associated with instability (based on scaling up the marginal effects, whose causal interpretation rests on the above-discussed assumptions, particularly regarding unobservable characteristics).
Formally, let equal the racial difference in the probability of outcome in gender . This descriptive difference decomposes into the three quantities discussed earlier:
The aggregated rate effect is
where , and are the racial differences in the total, early, and life-stage marginal effects of childhood family instability on outcome in gender , and is the probability of experiencing childhood family instability pattern among White people of gender . For example, is the probability of experiencing family instability only during later childhood among White women.13 Define as the racial difference in the probability of experiencing childhood family instability pattern , . The aggregated composition effect, then, is
The remaining differences are
To the extent that our MSMs overestimate childhood family instability's marginal effects (e.g., inappropriately attributing outcome differences to childhood family instability when they are caused by unmeasured confounders), we will underestimate the remaining differences and overestimate the population aggregated effect. To capture our estimates' uncertainty, we use parametric bootstrapping (see section A1, online appendix).
Black–White Differences in Childhood Family Instability's Marginal Effects
During childhood, approximately 32.1% and 38.6% of White and Black people, respectively, experienced parental union transitions (Table 1). Reflecting group differences in parental living arrangements at birth (panel A), Black children were more likely than White children to experience a parental union formation during at least one childhood life stage (panel D), and White children were more likely than Black children to experience a parental union dissolution (panel E). If a transition occurred, it was more likely to happen in early childhood only for Black than for White children (panel C) and was more likely to reoccur among White than among Black children (panel G). During young adulthood, White people were more likely to marry and less likely to dissolve unions than Black people; cohabitation rates were similar across groups (Table 2 and Figure 1; see section A2 of the online appendix for further discussion).
To connect these childhood and young-adult patterns, we use MSMs weighted by IPTWs to balance our “treated” and “untreated” samples. Our weights accomplish this balancing goal; see Figure 2 for the any parental union transition treatment (formation and dissolution treatments display the same pattern). Once weighted, people who did versus did not experience childhood family instability look similar in terms of their background covariates. The weighted versus unweighted differences are smaller for Black than for White people. Less balancing is needed for Black than for White people because the former's selection into instability is less severe (see also section A1, online appendix).
Table 3 displays the marginal effects by gender of experiencing any type of childhood family instability on ever cohabiting by age 29 (panel A), ever marrying by age 29 (panel B), or dissolving a first union by the last observation in 2017, given union formation (panel C).
The total effects are consistent with our first explanation for why Black young adults do not cohabit or marry more than White young adults, despite experiencing more childhood family instability: instability's effect is weaker among Black than among White people. Instability's total effect on cohabitation (marriage) is 12.91 (14.80) percentage points larger among White women than among Black women and 19.19 (12.68) points larger among White men than among Black men (Table 3, columns 1 and 4). These racial differences are fairly stable across ages (Figure A1, online appendix). Family instability pushes White people toward cohabitation and marriage much more than Black people. All these effect differences are large and positive. Most of their 95% confidence intervals' masses lie above 0, although small negative values are included in three of the four intervals. For union dissolution, in contrast, instability's total effects are positive but very uncertain for both White and Black people; racial differences appear negligible. In sum, instability's total effects are much larger among White than among Black people for cohabitation and marriage (as indicated by the tall positive bars in Figure 3) but slightly smaller among White than among Black people for union dissolution. Results are consistent with the first explanation for our puzzle (for two outcomes).
However, the results are not unequivocably consistent with the second explanation, which posits that instability's effects are small for all people. Childhood family instability's total effects on cohabitation and marriage might be extremely large, despite being smaller for Black than for White people (Table 3). For example, we estimate that the total effect on cohabitation among Black women is 7.7 percentage points; the top of our 95% interval is 19.1 percentage points (Table 3, panel A, column 3). Overall, 68.8% of Black women cohabited by age 29 (Table 2). Thus, the total effect could be one quarter of Black women's overall cohabitation rate. Likewise, for the marriage outcome, large positive effects relative to baseline rates are possible for White and Black people. However, for Black people, large negative effects are also possible: for outcomes as rare as young-adult marriage among Black people, intervals are wide. For the union-dissolution outcome, both large negative and large positive effects are plausible for White and Black people. Conditioning on union entrance reduces our sample size and generates uncertain estimates. In short, instability's effects could be large for young-adult union experiences among both Black and White people, although effects are smaller among Black people.
Black–White Differences in the Effects of Different Qualities of Family Instability
Larger childhood family instability effects among White than among Black people could reflect that some qualities of this instability have larger effects than others and that the qualities with the largest effects are most often experienced by White people. We find only partial support for this hypothesis.
Figure 4 highlights Black and White women's cohabitation, but the results hold for all three union outcomes and all four gender-by-racialized groups. Qualitative differences cannot explain racial differences in childhood family instability's effects because for all timings, volumes, and directions of instability, effects are larger (Figure 4's bars are taller) among White women than among Black women. For example, in terms of timing, both early and later childhood instability are more consequential for White women's cohabitation than for Black women's (compare the first two bars' heights across panels). Further, within each gender-by-racialized group, most qualities have similar effects, as explicated in the next three subsections.
Black children who experience a parental union transition are more likely than their White peers to experience it in early childhood only (Table 1). Thus, weaker effects of transitions in early childhood than later childhood would be necessary to explain the overall weaker effects of transitions among Black than among White people. Life-stage effects, which subtract early childhood effects from later-only effects, provide partial support for this hypothesis for cohabitation only.
The positive life-stage effects on cohabitation for Black women and men (Table 3, panel A, columns 3 and 6) indicate that parental union transitions had weaker effects in early childhood than in later childhood. Early childhood instability was approximately 5–6 percentage points less consequential than later childhood instability for their young-adult cohabitation. In contrast, the point estimates for White women and men are negative (panel A, columns 2 and 5), suggesting slightly stronger effects of early transitions than later transitions on White people's cohabitation. For both the marriage and union-dissolution outcomes, life-stage effects are negative for Black people. For these outcomes, parental union transitions during early childhood were more consequential than transitions during later childhood. In sum, the positive life-stage effects among Black people for the cohabitation outcome provide some support for the hypothesis that the racial difference in instability's overall effects can be explained by racial differences in its timing. Taken as a whole, though, the results do not provide strong support for this hypothesis.
High volumes of instability (with transitions in later childhood added to transitions in early childhood) are less common among Black than among White children (Table 1). Thus, another hypothesis for why Black young adults are less likely than their White peers to form unions is that high volumes are more consequential than low volumes. Yet, the results do not strongly support this hypothesis (Table 3).
For cohabitation, high volumes of instability are not much more or less consequential than low volumes. For White people, transitions have diminishing returns: the effect of adding a transition in later childhood to a transition in early childhood (the conditional later childhood effect) is smaller than the effect of experiencing a transition only in later childhood (the later-only effect). The compounding effect, which subtracts the later-only effect from the conditional later childhood effect, is thus negative (e.g., −14.3 = 2.9 – 17.2; Table 3, panel A, column 2). For Black people, the compounding effect is slightly positive, suggesting that instability's returns are mildly increasing (panel A, columns 3 and 6). For marriage, the compounding effects are negative for Black and White people (panel B). High volume does not appear to be more consequential than low volume for marriage, and the volume hypothesis therefore is not supported. The same is true for union dissolution for Black people (panel C).
White children are more likely than Black children to experience parental exits out of relationships, and Black children are more likely to experience entrances (Table 1). Thus, if parental union exits are more consequential than entrances for young-adult unions, these direction differences could explain instability's stronger effects among White than among Black people. However, we find little support for this hypothesis.
For young-adult cohabitation, marriage, and union dissolution (panels A–C in Table A3, online appendix), we report the early childhood effects of parental union formation (row 1 within each panel) and dissolution (row 2). We highlight these early childhood effects because they do not implicitly pool the effects of parental union formation and dissolution (unlike total effects or conditional later childhood effects). For young-adult cohabitation, parental union formation and dissolution have similarly sized effects for White people. Formation increases the probability of cohabitation by 16.2 and 19.8 percentage points among White women and men; corresponding increases for dissolution are 15.7 and 19.2 percentage points. Dissolution has larger effects than formation for Black people, but these dissolution effects are smaller in magnitude than the dissolution effects for White people. For young-adult marriage, parental union dissolution is less consequential than parental union formation. For young-adult union dissolution, parental union dissolution is less consequential than parental union formation for Black people but more consequential than parental union formation for White people. In sum, the direction hypothesis, which requires stronger dissolution than formation effects, is not strongly supported.
Population Aggregated Effects
All preceding results reference instability's marginal effects. Here, we consider its prevalence—specifically, that the Black–White difference in childhood family instability's prevalence might be too small to generate meaningful differences in young-adult outcomes. Table 1 provides initial support for this explanation. Black children were only roughly 6.5 percentage points more likely than White children to experience a parental union transition. Yet, Table 1 also reveals racial differences in instability's direction, timing, and volume. Could these differences generate large outcome differences?
Table 4 suggests that the answer is no. Panel A contains results by gender and outcome for our most general instability “treatment,” any parental union transition; panels B and C contain results for parental union formation and dissolution.
Each panel's first row reports the total racial difference in the column-specific outcome. For example, White women were roughly 10.4 percentage points more likely to cohabit by age 29 than Black women (panel A, first row and column). Each panel's second row reports the remaining difference. For example, of the total 10.4-point difference, 7.0 percentage points were not associated with childhood family instability's causal effects on cohabitation. The remaining 3.4 percentage points represent the population aggregated effect, the sum of the aggregated rate effect and aggregated composition effect (3.7 + [–0.3]). The population aggregated effect is driven by racial differences in instability's marginal effects (evidenced by the 3.7-point aggregated rate effect). Differences in instability's prevalence have little impact. If anything, they reduce outcome differences (by −0.3 here, because instability pushes people to cohabit and Black people are more exposed to this push).
Moreover, for all outcomes and parental transitions, the remaining differences are much larger than the aggregated rate or composition effects. Most racial differences in young-adult cohabitation, marriage, and union dissolution are not driven by childhood family instability. Point estimates suggest that between 0.8% and 42.4% of these racial differences are generated by childhood family instability (across outcomes, predictors, and genders); the mean is 16.7%. (The only effect exceeding one third is any parental union transition's effect on men's cohabitation difference.) Of these small portions, aggregated rate effects greatly exceed aggregated composition effects (see Figure A2, online appendix). These results support our fourth explanation: racial differences in exposure to childhood family instability are too small to generate large differences in union outcomes.
More than one third of today's young adults experienced transitions in their parents' romantic cohabiting partnerships during childhood. This family instability affects many life outcomes (Cavanagh and Fomby 2019). In the union domain, the family instability hypothesis predicts that childhood family instability shapes young adults' own family instability, pushing them to enter marital and cohabiting unions and dissolve these unions at young ages (Fomby and Bosick 2013; Thorsen 2017). Understanding the demographically dense period of young adulthood among today's youth is important because their relationship biographies differ from those of prior cohorts. These young-adult relationships affect individuals' later-life experiences and drive social change (Manning 2020; Tillman et al. 2019). Public conversations about Black–White differences in young adulthood sometimes attribute them to differences in childhood family instability (Haskins 2009). Because family instability is more common among Black than among White children, the family instability hypothesis predicts more union entry among Black than White young adults, but this prediction runs counter to fact. Black young adults are not more likely to marry or cohabit than Whites, although they are more likely to dissolve their unions (Hemez 2018; Manning 2020). Understanding this failed prediction is worthwhile because as young adults have delayed marriage and cohabitation, inequalities in these experiences have grown, potentially increasing inequalities in economic and health outcomes related to union formation (Bloome and Ang 2020; Zhang and Ang 2020).
We reviewed four explanations for the puzzling inconsistency between the family instability hypothesis and observed union-formation inequalities. Our results are consistent with the first explanation: childhood family instability is less consequential for Black than for White young-adult unions. Our results are also consistent with the fourth explanation: perhaps surprisingly, racial differences in childhood family instability's prevalence are small. Our results are inconsistent with the second explanation: instability's effects are not unequivocally small, although they are smaller among Black than among White people. Our results are also generally inconsistent with the third explanation: racial differences in instability's volume and direction cannot explain our puzzle; racial differences in instability's timing offer a partial explanation for cohabitation only. These qualitative differences in Black and White children's instability experiences cannot resolve the puzzle because most qualities have similar effects within each racialized group. Instead, resolution resides in the small magnitude of racial differences in childhood family instability's prevalence and its smaller effects on cohabitation and marriage among Black than among White people. Thus, childhood family instability explains little of the racial differences in young-adult union outcomes.
These findings offer three contributions to the literature. First, we documented the limited generalizability of the family instability hypothesis across Black and White young adults today in terms of their cohabitation and marriage experiences by age 29, extending prior research on racially heterogeneous effects to the new domain of young-adult unions. Second, we revealed that its limited generalizability cannot be understood via different qualities of instability. Finally, we quantified instability's minor contribution to population-level inequalities in young-adult unions.
Our findings have several substantive and methodological implications. Substantively, first, they imply that family dynamics have weaker intergenerational effects among Black than among White people. We speculate that these effects might weaken for both groups in the future, to the extent that changes in marriage and cohabitation make parental union transitions increasingly common and potentially less stressful and stigmatizing (Dronkers and Härkönen 2008; Raab 2017; but see Härkönen et al. 2021). The intersection of group-specific effects (e.g., for Black and White people) and context-specific effects (e.g., for societies with lower and higher instability prevalence) is an area ripe for research. Future investigators could consider whether contextual changes lead group-specific effects to converge. Second, our findings imply that group differences in childhood family instability should not be framed as important contributors to racial inequalities in union outcomes. From a policy perspective, interventions on childhood family instability are unlikely to yield dividends in terms of racial differences in marriage or cohabitation. Third, intuitions about family structure (i.e., who is part of the family unit) should not be applied rotely to family instability (i.e., changes in who is part of the family unit). Racial differences are large in family structure but not in family instability (Table 1).
Methodologically, our findings suggest that researchers studying group differences in outcomes should incorporate group differences in key predictors' marginal effects and prevalence. Our novel decomposition illustrates one way to do so by extending the logic of traditional descriptive decompositions. A second methodological implication is that allowing for different selection across groups is important. Selection into childhood family instability is more extreme among White than among Black people (Figure 2). Our finding that this instability is also more consequential for White people adds to a growing literature documenting larger effects of social and demographic “treatments” (e.g., college attendance, parental divorce) on people who are less likely to encounter them (Brand et al. 2019; Brand and Xie 2010).14
Our study also has limitations. First, childhood family instability's estimated effects could reflect unobserved confounders. Others have found large family instability effects when accounting for unobservable characteristics (e.g., Ryan and Claessens 2013), while we rely on a rich set of observed covariates. We extend prior literature by documenting larger effects among White than among Black people after accounting for time- and group-varying selection on observable variables, and by documenting that childhood family instability explains only a small portion of racial differences in young-adult unions. Adjusting for more confounders could reduce this small size to zero. Second, we measure young-adult union dissolution dichotomously. If the PSID Transition into Adulthood Supplement adds a dissolution calendar, our results should be revisited. Third, to distinguish volume from timing (because higher order transitions occur later in life), we grouped children who, in one life stage, experienced one and multiple transitions. Future research could separate these groups if sample sizes permit.
Researchers could also explore alternative data sources to address another limitation: we studied only 1,817 people, making our results uncertain. Larger samples would generate tighter confidence intervals. We selected this sample because of its detailed prospective information on a contemporary cohort's experiences from birth onward (data sources such as the National Longitudinal Survey of Youth and Add Health lack this information). As the TAS sample grows with more PSID children entering young adulthood, researchers should revisit our work. They could also expand our work into older adulthood to determine whether instability's effects diminish beyond young adulthood.
Our findings suggest that researchers should look beyond childhood family dynamics to understand Black–White differences in young-adult unions and consider broader conduits of institutional and interpersonal racism. Examples include educational and employment discrimination (which limit access to well-paying jobs that some consider necessary for marriage) and dating-market discrimination (which limits partnership opportunities, particularly for Black women) (Clarke 2011; Hamilton et al. 2009). We do not quantify racism's effects. These effects are embedded in several parts of our results, including racial differences in instability's prevalence; these differences reflect many processes that, from a “fundamental causes” perspective, stem from racism (Cogburn 2019:738; Letiecq 2019). Quantifying racism's contribution to young-adult union inequalities is an important task moving forward.
Meanwhile, our analysis indicates that researchers and policymakers who would like to understand or intervene on racial differences in romantic coresidential unions—differences that, according to Stewart (2020:217), reflect “this country's most camouflaged civil rights issue”—should look beyond childhood family instability.
Authors are listed alphabetically. We gratefully acknowledge insightful comments from Demography's editors and reviewers, Susan Brown, Andrew Cherlin, Chantal Hailey, and Peter Rich; excellent research assistance from Meichu Chen; and generous support from NICHD research grant R01HD088506 (Fomby), project grant P01HD087155 (Bloome and Fomby), and center grant P2CHD041028, as well as National Social Science Foundation of China research grant 22CRK010 (Zhang).
We use the term instability to converse with prior literature, but see Hadfield, Ungar, and Nixon (2018) for critique.
Descriptive inequalities in union dissolution are not inconsistent with the family instability hypothesis; union dissolution rates are higher among Black than among White people (Manning 2020; Schweizer 2020). We investigate both union formation and dissolution because they are theoretically central to the hypothesis that childhood family instability replicates across generations (Amato and Patterson 2017; Thorsen 2017).
Many racial and gender inequalities intersect. We allow instability’s effects to differ by gender within racialized groups. However, we do not anticipate gendered effect heterogeneity because both men and women are influenced by their childhood families and form young-adult unions (Manning 2020). This gender similarity in union outcomes contrasts with other outcomes, such as externalizing behaviors, which differ across genders (Fomby and Mollborn 2017).
We obtain racialized group information from the respondent’s first self-report of their racial and ethnic categories; incorporating second reports leaves results unchanged.
We use TAS sample weights when modeling selection into childhood family instability, as detailed in section A1 of the online appendix.
The immigrant refresher lacks baseline covariates and complete prospective information on family instability for children entering the sample after birth.
Because the PSID moved from annual to biennial interviews in 1997, we impute off-interview years with information from the year before or after randomly.
We study the child’s coresident parent(s) because the survey design prevents us from observing union changes among nonresident parents who are not followable PSID family members.
Young adults enter the risk set at age 14 (the youngest age at which sample members entered unions).
If we do not observe a respondent at exact ages 0 or 5 (e.g., because those ages coincide with off years in the PSID’s biennial interview schedule), we impute covariate values for those ages from the prior year. If the prior-year data are also missing, we impute from the subsequent year. If that year’s data are also missing, we look two years prior and then two years subsequent. Our procedure differs slightly for two covariates, income and wealth (as discussed later).
Other assumptions include proper specification of the IPTW models and nonzero treatment probability for all in the reweighted sample (Robins et al. 2000).
For example, for the cohabitation outcome, is the contrast between two predicted probabilities of ever cohabiting by age 28.8 among people from racialized group and gender : one among people who experienced family instability in early and later childhood, and one among people who experienced no family instability during childhood. We generate both predicted probabilities from our IPTW-weighted discrete-time event-history models.
Our notation departs slightly from that of some other analyses. In our notation, pattern (x,y) represents event x occurring before y.
Racial differences in selection are large, although racial differences in marginal probabilities of experiencing childhood family instability are small.