Unpacking the Linkages between Single Parent Households and Early Adolescent Adjustment

Abstract

Living with an unmarried mother is consistently associated with adjustment issues in adolescence, but these associations can vary by both time and place. Following life course theory, this study applied inverse probability of treatment weighting techniques to data from the National Longitudinal Survey of Youth 1979 Children and Young Adults study (n = 5,597) to estimate various treatment effects of family structures through childhood and early adolescence on internalizing and externalizing dimensions of adjustment at age 14. Young people who lived with an unmarried (single or cohabiting) mother during early childhood and adolescence were more likely to drink and reported more depressive symptoms by age 14 than those with a married mother, with particularly strong associations between living with an unmarried mother during early adolescence and drinking. These associations, however, varied according to sociodemographic selection into family structures. They were strongest for youth who more closely resembled the average adolescent living with a married mother.

Introduction

Within the literature on family structure disparities, social scientists have paid special attention to unmarried mothers and their adolescents, who often have elevated rates of adjustment issues that arise in reaction to economic and social stressors and forecast more challenging transitions into adulthood (McLanahan and Sandefur 1994; Amato 2010; Johnston et al. 2020). Despite this ample attention, there is still more to learn about this linkage through innovative conceptual and methodological approaches, such as by emphasizing its variability more than its generality (Turney 2015) and by comparing how it differs between externalizing and internalizing symptoms (Cavanagh and Fomby 2019).

In this spirt, this study attempted to understand when and for whom associations between living with an unmarried mother and adolescent adjustment do and do not apply. Our life course perspective dynamically unpacks and contextualizes this family structure disparity (Crosnoe 2021), considering how the adjustment of adolescents with unmarried mothers—single as well as unmarried but cohabiting—might vary along three dimensions: 1) the timing of living in this family structure, 2) the degree to which mothers’ and adolescents’ backgrounds “fit” the typical sociodemographic profile of this family structure; and 3) and whether adjustment issues are internalized or externalized. We explored this variation by applying a causally informed statistical approach (inverse probability of treatment weighting with heterogeneous causal effects; Morgan and Todd 2008) to intergenerational panel data from the National Longitudinal Survey of Youth 1979 Children and Young Adults study (NSLY79-CYA).

Such research follows conceptual innovations related to life course theory and other perspectives that focus on the moderation of linkages between family circumstances and adolescent adjustment by time and place. Notably, they are coupled with methodological innovations rarely used in this literature that center selection into family structure as a moderator rather than a confounder. The insights gleaned from this coupling are intended to better identify which young people are most vulnerable in the short and long term while presenting a new approach to answering critical questions about family contexts of adolescent adjustment.

The Adjustment of Adolescents Living with Unmarried Mothers

Many adolescents live with an unmarried mother who is single, usually through non-marital fertility, union dissolution, or both. Another large share of adolescents lives with an unmarried mother who is partnered, often cohabiting with the child’s father. In general, living with an unmarried mother is associated with poorer adjustment during adolescence relative to living with a married mother (Crosnoe and Cavanagh 2010; McLanahan 2004).

The explanation for these associations typically connects two interrelated mechanisms that both center on increased exposure to strains (e.g., economic anxiety, conflict) and reduced opportunities for mobility and safety that directly stress adolescents or indirectly matter through disrupted parenting (Breivik, Olweus, and Endresen 2009). In the socialization path, this family structure itself creates such developmentally significant forces. For example, not having a partner to help with household expenses and routines can create practical (e.g., money, time) or socioemotional (e.g., distress, distraction) constraints on a single mother’s ability to consistently engage with an adolescent in ways that help them feel secure, monitor their peer associations, and discourage their risk-taking (Hemovich, Lac, and Crano 2011; Turner, Irwin, Jr., and Millstein 1991; Barrett and Turner 2005; 2006; Hardaway and Cornelius 2014; Kendig and Bianchi 2008). In the selection path, this family structure arises from other sociodemographic circumstances that can also set such forces in motion regardless of family structure. For example, younger maternal age, lower education, unstable employment, health problems, and other factors may mean that a woman is more likely to become a single mother (either through divorce or non-marital fertility), but, whether she does or not, these background factors interfere with such consistent engagement with adolescents described above (Lichter, Qian, and Mellott 2006; Lichter, Sassler, and Turner 2014; Liu, Elliott, and Umberson 2010; Raley and Sweeney 2020; Wildsmith and Raley 2006; Williams and Finch 2019).

These previous examples of socialization and selection concern unmarried mothers who do not have second parental figure in the home. Unmarried but partnered mothers do have a second parental figure in the home, but the socialization and selection paths of this unmarried mother household are more akin to single mother households than married mother households. With less clear roles for parenting, family management, and economic support in cohabiting relationships than married relationships, a second parental figure often matters less to children and adolescents when that person is not married to the mother. At the same time, many of the same social and economic disadvantages that select women into single motherhood also increase the likelihood of entering informal (vs. formal) partnerships (Brown 2004; Dunifon and Kowaleski-Jones 2002; Kalil, Ryan, and Chor 2014; Manning and Lamb 2003). Thus, although the experiences of adolescents living with single mothers may differ from those living with cohabiting mothers, the differences of both relative to adolescents living with married mothers are likely to be greater.

Of course, the adjustment issues of adolescents living with unmarried mothers, through socialization or selection, take many forms. Social scientists typically focus on adjustment issues that are internalized, such as depressive symptoms, or externalized, such as alcohol use (Brown and Rinelli 2010; Cavanagh 2008; Hoffmann 2002; Lu 2019). Both tend to emerge and/or increase during adolescence and, when they occur at the start of adolescence, suggest that educational, health, and interpersonal trajectories into adulthood may be at risk. They also differ in important ways. Consequently, comparing them can illuminate why living in unmarried mother households is developmentally significant (Crosnoe 2021). Heightened depressive symptoms point to circumstances that trigger emotional distress, which leads to social withdrawal and raises the risks of enduring mental health struggles and their consequences. Increased alcohol use can also point to emotional distress, with drinking as a form of self-medication, but it more frequently suggests circumstances that reduce social and interpersonal controls on adolescents’ developmental propensity for risk-taking within peer associations, regardless of distress. Relative to drinking, therefore, depressive symptoms raise the specter of such dangers as suicidal ideation and are likely more reactive to psychological intervention. Relative to depressive symptoms, however, drinking raises the specter of such dangers as accidents and may be more reactive to social intervention (Bjarnason et al. 2003; Cho, Hallfors, and Iritani 2007; Daryanani et al. 2017; Flewelling and Bauman 1990; Heron 2019; Hussong et al. 2015; Schulenberg et al. 2014).

A Life Course Approach to the Adjustment of Adolescents Living with Unmarried Mothers

The ways that selection into and socialization within unmarried mother families lead to internalizing and externalizing symptoms of adolescents illustrate life course theory’s principle of linked lives. This principle advocates for studying the interplay of the interpersonal and institutional pathways of parents with the developmental pathways of their children over time. It emphasizes how parents’ own experiences affect their children by shaping the structural and ecological conditions in which children are growing up (Elder, Shanahan, and Jennings 2015); in this case, how the partnership experiences of mothers—and the sociodemographic circumstances from which they arise—create the conditions in which their children think, feel, and behave (McLanahan and Bumpass 1988). Importantly, life course theory also suggests that the nature of linked lives can be highly variable, as, like most life course phenomena, they are likely to be temporally dynamic and socially contextualized (Elder et al. 2015). Following such a theoretical approach, this study builds on the large literature on the associations between unmarried mothers and adolescent adjustment by focusing on a key temporal dimension of variability in linked lives (by the timing of young people’s experience of living in such a family structure) and a key contextual dimension of this variability (by the degree to which the circumstances of adolescents match the typical pattern of sociodemographic selection into this family structure).

First, in terms of timing, there may be sensitive periods (Ben-Shlomo and Kuh 2002) in which family structure is particularly significant for adolescent adjustment. During a child’s first years, the stress and strain of living in an unmarried mother household—or the circumstances that lead to it—may interfere with mother-child attachment, which has long-ranging implications for emotional development (Bowlby 1969). Any early effects on young children, through attachment or not, can also set in motion a process of cumulative disadvantage, given the tendency towards path dependence in developmental trajectories (Pasqualini, Lanari, and Pieroni 2018). During early adolescence, youth begin spending more unsupervised, unstructured time with peers (Cavanagh and Huston 2008), which fosters peer pressure for risk-taking in general and substance use experimentation in particular (Crawford and Novak 2002; Chassin et al. 2004). Adolescents who live in an unmarried mother household during this period may be especially vulnerable to this social aspect of development, as the stress and strain of this experience—or the circumstances triggering it—create greater immediate conduits to alcohol use. Those conduits include greater needs for social integration, interference with adult supervision, and amplification of the developmentally normative increase in conflict with parents (Krohn, Hall, and Lizotte 2009; Wagner et al. 2010; Weaver and Schofield 2015). Thus, relative to middle childhood, early childhood and early adolescent experiences of living in single mother households would be sensitive periods for adjustment in early adolescence, with the former more likely to manifest in depressive symptoms (emotional) and the latter in alcohol use (social).

Second, in terms of contextualization, the sociodemographic patterns that select adolescents into living with unmarried mothers do not simply influence adjustment; they may moderate associations between family structure and adjustment (Lee and McLanahan 2015). For this contextualization, we consider how adolescents in unmarried mother households may have especially poor outcomes when they look (sociodemographically) more like a typical adolescent in a married mother household. In a form of social comparison, this disparity might result from the greater vulnerability of living with an unmarried mother in social spaces where it is less common, with both adolescents and their mothers feeling more stress in the face of any accompanying stigma or isolation, and the lesser vulnerability of doing so in spaces where it is more common, with both more likely to develop adaptive strategies and access social support (Bernardi and Boertien 2016; Crosnoe, Johnston, and Cavanagh 2021). The social nature of this comparison might be more important for domains most connected to interpersonal activity (e.g., drinking). As evidence, the associations between unmarried mother households and adolescent adjustment vary across schools as a function of the proportion of fellow students with married parents (Cavanagh and Fomby 2012). In the case of saturated disadvantage (Hannon 2003), the experience of living with an unmarried mother among those who are most sociodemographically likely to do so would be more redundant with the accompanying sociodemographic disadvantages. As evidence, unmarried mother households tend to be less developmentally significant for groups with more entrenched histories of socioeconomic marginalization (Cavanagh and Fomby 2019; Cross 2020; Lee and McLanahan 2015). Relative to social comparison, such saturated disadvantage would perhaps be less likely to differ between more and less social domains of adjustment. Thus, associations between unmarried mother household and adolescent adjustment may be stronger for adolescents who least fit the sociodemographic profile of youth in unmarried mother households and weaker for those who most fit this profile, with some differences across internalizing and externalizing dimensions of adjustment. The modeling strategy we describe below—inverse probability of treatment weighting—offers a parsimonious way to test this potential moderating role of sociodemographic selection.

Study Aims and Hypotheses

Following this life course elaboration of the linked lives of unmarried mothers and their adolescents, this study’s overarching goal was to examine variability in the linkage between living with an unmarried mother through the early life course and adolescents’ adjustment and to compare such variability across domains of adjustment. This overarching goal led to two focal sets of hypotheses: 1) the association between living with an unmarried (especially single) mother and early adolescent adjustment would be strongest when this family structure was experienced in early childhood or early adolescence (vs. middle childhood) and when adolescents showed the least resemblance to the sociodemographic profile of the typical unmarried mother household (vs. most); and 2) the association of maternal partnership status in early childhood with adolescent adjustment would be greater for depressive symptoms (vs. alcohol use) while the association of this status in early adolescence—and its moderation by sociodemographic profile—would be greater for alcohol use (vs. depressive symptoms).

These two sets of hypotheses were tested simultaneously with a causally informed modeling technique that estimates the “effect” of living with an unmarried mother for the general population of youth, the subpopulation of youth who lived with an unmarried mother, and the subpopulation of youth who lived with a married mother. The treatment in this study is living with an unmarried mother and these estimates are known, respectively, as the average treatment effect (ATE), the average treatment effect on the treated (ATT), and the average treatment effect on the untreated (ATU). For the first set of hypotheses, comparing differences in the ATEs at each age assessed variation by timing of living with an unmarried mother and comparing differences in the ATUs and ATTs assessed moderation by sociodemographic profile. For the second set of hypotheses, comparing differences between the ATEs during early childhood and early adolescence for both alcohol use and depressive symptoms assessed differences in sensitivity of these outcomes to maternal partnership status experienced during particular ages. Comparing the differences between the ATTs and ATUs for both alcohol use and depressive symptoms assessed differences in the sensitivity of these outcomes to differences in the significance of living with an unmarried mother for youth from differing sociodemographic profiles.

In addition, given evidence that boys’ developmental outcomes appear to be more sensitive to parental divorce and single motherhood than girls (Cavanagh and Fomby 2019) and that boys tend to have more externalizing than internalizing responses to environmental challenges (Rosenfield, Lennon, and White 2005), we tested for possible gender differences in the experiences of family structure for both outcomes. These models also indirectly tested for variation in the heterogeneous causal effects themselves: for example, if boys are in general more sensitive to living with unmarried mothers than girls, then the differences between the ATUs, ATTs, and ATEs might be smaller for boys than girls because boys, regardless of sociodemographic background, are more affected by this type of household. Thus, we expect that the ATE estimates would be larger for boys than for girls regarding alcohol use; the ATE estimates for depressive symptomology would be larger for girls than for boys; and the differences among the ATU, ATT, and ATE estimates would be smaller for boys than for girls.

Material and Methods

Data

The National Longitudinal Survey of Youth 1979 Child and Young Adult (NLSY79-CYA) sample includes 11,521 children born to the original National Longitudinal Survey of Youth 1979 mothers. Beginning in 1986, the mothers of the NLSY79-CYA youth completed a biennial Child survey for children younger than 10 and youth at least 10 years old completed their own Child survey. Starting in 1994, youth at least 15 years old completed a biennial Young Adult survey. Thus, the sample represents the experiences of children and adolescents in the U.S. from the 1980s through the mid-2010s. We first limited the sample to 6,263 youth with a maternal survey at birth and their own survey at age 14 or 15 (depending on interview timing). We then excluded 639 youth whose mothers were married at the child’s birth, then divorced, and then remarried before the child’s age 14 survey, so that the experience of living with a married mother would not be combined with the experience of living with a remarried parent. The final analytical sample included 5,597 youth, and the time points in the data were approximately infancy (age 0) and ages 2, 4, 6, 8, 10, 12, and 14.

Measurement

Adjustment in early adolescence.

For early externalizing symptoms, a dichotomous variable measured any alcohol use at the time of youth’s age 14 survey (any alcohol use at age 14 or before vs. no such use). For early internalizing symptoms, depressive symptomology was measured by seven items from the Center for Epidemiological Studies-Depression (CES-D) short form measure at youths’ age 14 survey. Youth rated how often in the past week they experienced the following symptoms: poor appetite, trouble focusing, feeling depressed, feeling that everything was an effort, restless sleep, feeling sad, and feeling like they could not “get going”. Responses to each question were 0 (rarely or none of the time), 1 (some or a little of the time), 2 (occasionally or a moderate amount of time), and 3 (most or all of the time). These responses to the seven questions were summed to create a score ranging from 0 to 21.

Maternal partnership through childhood and adolescence.

We measured family structure according to mother’s self-reported partnership status (dummy variables for married, cohabiting, and single) at the time of each survey. At each of the surveys completed by the child’s mother (annually between 1979 and 1993; every other year between 1994 and 2014), mothers reported if her spouse and/or partner lived in the household. If she reported a spouse in the household, regardless of whether or not she reported a partner, she was coded as married. If she reported a partner and no spouse, she was considered cohabiting. A mother who reported no spouse and no partner was coded as single. Maternal partnership was measured at child ages 0, 2, 4, 6, 8, 10, 12, and 14.

Sociodemographic covariates.

Time-invariant covariates included the mother’s race/ethnicity (white, Black, American Indian/Alaska Native, multi-racial, Hispanic, or other race), nativity (born outside of the U.S. or not), education at the child’s birth (dummy variables for less than high school, high school, some college, more than college), age at first birth (19–42), and substance use during pregnancy (1 = engaged in alcohol, marijuana, or cigarette use in the year prior to the child’s birth, 0 = no such engagement) as well as the child’s gender and birth year. We included additional time-invariant dichotomous covariates capturing facets of mothers’ backgrounds that have been linked to unmarried motherhood: whether her own parents had a high school education, whether she lived in the South at age 14, whether she lived in a rural area at age 14, and whether she lived with both parents at age 14 (Lichter et al. 2006; Lichter et al. 2014; Wildsmith and Raley 2006). Four additional covariates assessed mothers’ exposure to adverse childhood experiences (Williams and Finch 2019): whether she grew up in a household with someone with mental illness, grew up in a household with someone with substance abuse, experienced physical abuse, and/or experienced emotional neglect. Time-varying covariates measured at every time point were household size (number of adults and children in the household), poverty status (a dichotomous indicator of a total family income below the federal poverty line for that year and household size), and mother’s employment status (any employment including military service or unemployed/out of the labor force).

Plan of Analyses

This study estimated the associations between living with a single or cohabiting mother at ages 0, 2, 4, 6, 8, 10, 12, and 14 and the odds of early adolescent alcohol use and extent of depressive symptoms in a series of regression models with inverse probability (IP) weighting. These models addressed the dimension of variability (i.e., by timing of living with an unmarried mother) by calculating the associations between living with a single or cohabiting versus married mother at each age and adolescents’ adjustment outcomes at age 14. The IP weighting used in these models addressed the second dimension of variability (i.e., fit with the typical sociodemographic profile of a family structure) by reweighting individuals according to how much they sociodemographically resembled the average adolescent with a given maternal partnership status. IP weighting expands on previous research on family structure and adolescent development by allowing exploration of how the socializing role of family structure differs by a youth’s likelihood of selection into that family structure. Borrowing the language of experimental design, the logic of this approach is best explained by the various treatment effects it produces (Morgan and Winship 2008).

The general idea of inverse probability of treatment weighting is to simulate the random assignment in a hypothetical experiment such that the treated and untreated groups resemble one another. In this case, there were two comparisons of treated vs. untreated groups: youth who lived with a single vs. married mother at each age and youth with a cohabiting vs. single mother at each age.

Each hypothetical experiment could take one of three forms. First, youth from the general population could be randomly assigned to live with a single or married mother. With this type of randomization, the sociodemographic distribution of the treated (single mother) and untreated (married mother) groups should be the same. Any difference in outcomes between these two groups could be interpreted as the average treatment effect (ATE), or the effect of living with a single mother among all youth. The ATE estimates can be thought of as analogous to more standard regression models; the difference being that the ATE estimates are derived from models that use weighting to control for confounds, while regression models use adjustment to control for confounds. Second, an experiment could randomly assign to the treatment groups only youth who currently live with a single mother. The sociodemographic distribution of the treated and untreated groups will still be similar across the two groups, but the sociodemographic characteristics of the youth drawn from a single mother family in this experiment will differ from those of the youth drawn from the general population in the first experiment. Differences in outcomes could be interpreted as the average treatment effect among the treated (ATT), or the average effect of living with a single mother among those who actually live with a single mother. Third, an experiment could only randomize to the treatment group youth from married mother families. Again, the sociodemographic distribution of the treated and untreated groups will be similar but still differ from the youth in the first and second experiments. Differences in outcomes between the treated and untreated groups in this experiment could be interpreted as the average treatment effect among the untreated (ATU), or the average effect of living with a single mother among those who actually live with a married mother.

In short, inverse probability of treatment weighting reweights the treated and untreated groups so as to sociodemographically resemble the full population of youth, the treated group, or the untreated group, to estimate the ATE, ATT, and ATU, respectively (Cole and Hernán 2008; Hernán and Robins 2020). If the effect of living with an unmarried mother did not differ by youths’ sociodemographic likelihood of actually living with an unmarried mother, then the ATE, and ATT, and ATU would reveal similar estimates. On the other hand, if the background characteristics associated with living with an unmarried mother moderated the effect of living with an unmarried mother, then the ATE, ATT, and ATU will differ (Sato and Matsuyama 2003). For example, if the ATU of living with a single mother was stronger than the corresponding ATT, we could infer that the effect of living with a single mother is stronger for those least sociodemographically likely to live with a single mother. In other words, the extent to which the ATT, ATE, and ATU differ indicates that the socializing role of a family structure on early drinking or depressive symptomology differs by youths’ sociodemographic likelihood of living in that family structure.

Estimating models with inverse probability of treatment weights required two steps. The first step was to construct the inverse probability of treatment weights. Weights were created for each family structure comparison, or treatment (living with a single vs. married and living with a cohabiting vs. married) at each age (0, 2, 4, 6, 8, 10, 12, and 14) and for each type of treatment effect (ATE, ATT, and ATU). To do so, we derived the predicted probabilities of being in each treatment group (e.g., having a single vs. married mother) at each age from logistic models including all time-invariant sociodemographic covariates; concurrent measures of family poverty, household size, and mother’s employment status; and all prior measures of mothers’ partnership status, family poverty, household size, and employment status. The equation for calculating these predicted probabilities took the form of $ln\left[\frac{p_t}{1-p_t}\right]={\beta }_{0,t}+{\beta }_c{x}_c+{\beta }_t{x}_t+{\beta }_{t-2,t-4,\dots ,t-n}{x}_{t-2,t-4,\dots t-n}$ where $\ln \left[\frac{p_t}{1-p_t}\right]$ is the log odds of being in the treatment group at age t, β_0,t is the intercept, x_c is the vector of time-invariant covariates, x_t is the vector of concurrent covariates measured at age t, and x_{t−2,t−4,...t−n} is the vector of covariates measured at each previous age. After calculating the 16 sets of predicted probabilities (2 comparisons x 8 ages), the weights for each treatment effect and comparison were created by setting the weights either to 1 or to a function of the predicted probability of being in the treatment group depending on the type of treatment (see Figure 1 and Morgan and Todd 2008). For example, for the average treatment effect on the treated (ATT) of living with a single vs. married mother at age 6, youth would be reweighted such that youth who lived with a single mother at age 6 (x_sm6=1) would receive a weight of 1 and youth who lived with a married mother at age 6 (x_sm6=0) would receive a weight of [p_sm6/(1 − p_sm6)], where p_sm6 is the predicted probability of living with a single vs. married mother derived from the logistic regression model. For the average treatment effect, we used a stabilized weight in which the numerator represented the probability of treatment without covariates and the denominator represented the probability of treatment with covariates to reduce the influence of individuals with particularly large or small values of p.

Figure 1.

Construction of Inverse Probability Weights

Note: p = conditional predicted probability of being in the treated group. ATE = average treatment effect, ATT = average treatment effect on the treated, ATU = average treatment effect on the untreated. For the average treatment effects, we stabilized the weights by replacing the numerator with the unconditional probability of treatment.

The second step was to use logistic and ordinary least squares models to estimate the treatment effects of family structure at each age on the risk of the two outcomes, respectively. Each model estimated the effect of one of two comparisons (living with a single vs. married or cohabiting vs. married) at one of eight ages (0, 2, 4, 6, 8, 10, 12, and 14) on one of two outcomes at age 14, using one of three weighting schemes (ATE, ATT, and ATU). For example, the model for the ATT of living with a single vs. married mother at age 6 on drinking at age 14 would take the form $ln\left[\frac{p}{1-p}\right]={\beta }_{sm0}+{\beta }_{sm6}{x}_{sm6}$ where $ln\left[\frac{p}{1-p}\right]$ is the log odds of drinking at age 14, β_sm0 is the intercept, x_sm6 is a binary variable indicating that youth lived with a single (vs. married) mother at age 6, and β_sm6 is the coefficient of interest, the effect of living with a single vs. married mother at age 6. Youth would be reweighted according to the ATT weights described in the previous paragraph. These models were repeated for each comparison at each age, each type of treatment effect, and each outcome.

As shown in Table 1, missing data were minimal (< 5% per variable) except for mothers’ adverse childhood experiences (approximately 8% per ACE), substance use during pregnancy (approximately 9% for cigarettes and alcohol, 47% for marijuana), and poverty status (17%). The poverty variables were based on respondents’ answers to income questions, which have a disproportionately high amount of non-response compared to other types of questions (U.S. Bureau of Labor Statistics n.d.). The ACE questions were only in the 2012 and 2014 surveys, so they have a higher-than-average amount of missingness due to attrition. Marijuana usage during pregnancy had an especially high rate of missingness because it was not asked of mothers who gave birth prior to 1986. To address missing data, 25 datasets were produced by chained multiple imputation equations prior to the calculation of weights (White, Royston, and Wood 2011). Weights were calculated within each multiply imputed dataset, multiplied by the NLSY custom sampling weights to adjust for oversampling and differential attrition, and trimmed at the 99^th percentile to improve precision of the final estimates (Lee, Lessler, and Stuart 2011).

Table 1.

Unweighted Sample Statistics for Study Variables

	Frequency or Mean (SD)	% Missing
Drank alcohol by age 14	27.01%	2.63%
Depressive symptomology score at age 14	4.33 (3.56)	4.43%
Waves in single parent family	30.05%	2.40%
Waves in married parent family	63.49%	2.40%
Waves in cohabiting parent family	6.46%	2.40%
Family structure at child’s birth
Married parents	63.37%	2.41%
Cohabiting parent	6.99%	2.41%
Single parent	29.64%	2.41%
Average age at marriage dissolution among those with married mother at birth	8.58 (3.99)	32.44%
Waves in poverty	24.23%	17.14%
Waves with employed mother	59.72%	3.18%
Family size across waves	4.36 (1.28)	2.40%
Female	49.10%	0.00%
Mother’s race/ethnicity
Non-Hispanic White	43.93%	0.00%
Non-Hispanic Black	31.12%	0.00%
Non-Hispanic American Indian/Alaskan Native	0.70%	0.00%
Non-Hispanic other race	4.32%	0.00%
Non-Hispanic multiple races	0.88%	0.00%
Hispanic	19.05%	0.00%
Mother’s age at birth	26.61 (5.44)	0.00%
Mother’s education at birth
Less than high school	21.30%	2.47%
High school	42.33%	2.47%
Some college	20.90%	2.47%
College or more	15.46%	2.47%
Mother drank alcohol during pregnancy	42.66%	8.58%
Mother smoked cigarettes during pregnancy	27.90%	8.68%
Mother used marijuana during pregnancy	2.37%	46.54%
Mother’s parents finished high school	60.42%	3.59%
Mother was foreign-born	8.27%	0.00%
Mother lived in South at age 14	36.88%	4.22%
Mother lived in rural area at age 14	20.34%	0.41%
Mother lived with two parents at age 14	66.38%	0.25%
Mother’s adverse childhood experiences
Mental illness in household	9.18%	7.95%
Substance use in household	20.45%	8.02%
Physical abuse	16.38%	8.08%
Emotional neglect	19.22%	8.24%
n	5,597

Results

The analytical sample of adolescents was 44% non-Hispanic white, 31% Black, 0.7% American Indian/Alaskan Native, 4% other race/ethnicity (including Asian), 0.9% multi-racial, and 19% Hispanic, and it was 49% female. By age 14, 27% of youth in the sample had drunk alcohol. The mean depressive symptomology was 4.33 (out of 21). At birth, 63% of youth lived with a married mother, and 37% lived with an unmarried mother (7% lived with a cohabiting mother; 30% lived with a single mother). Across childhood, at each age, between 59% and 67% of young people lived with a married mother, between 26 and 35% lived with a single mother, and between 6 and 7% lived with a cohabiting mother. See Table 1 for full descriptive statistics for the sample as a whole.

In order to highlight the sociodemographic differences between youth living with married, cohabiting, or single mothers, Table 2 displays the distribution of the study variables by four mutually exclusive family structure history types based on maternal partnership status: always married and never cohabited, always married or cohabiting, ever single and never cohabited, and ever single and ever cohabited. Compared to youth who lived with a consistently married mother, all other youth were more likely to drink alcohol by age 14, had greater depressive symptomology at age 14, and lived in greater poverty. Their mothers were less likely to be non-Hispanic white, employed, and a high school graduate. Youth in families with mothers who had histories of being both single and cohabiting in particular were more likely to drink, had greater depressive symptomology, and lived in greater poverty compared to youth with consistently married mothers. As demonstrated by Table 2, the distribution of covariates differed across the subpopulations of youth living in different family structures.

Table 2.

Unweighted Sample Statistics, by Mother’s Partnership History

	Always married, never cohabited	Always married or cohabiting	Ever single, never cohabited	Ever single, ever cohabited
Drank alcohol by age 14*	22.75%	27.92%	27.52%	35.52%
Depressive symptomology score at age 14*	4.02 (3.40)	4.16 (3.54)	4.51 (3.69)	4.88 (3.64)
Waves in single mother family*	0.00%	0.00%	62.67%	48.47%
Waves in married mother family*	100.00%	51.29%	37.33%	21.12%
Waves in cohabiting mother family*	0.00%	48.71%	0.00%	30.40%
Family structure at child’s birth*
Married mother	100.00%	12.90%	39.45%	19.29%
Cohabiting mother	0.00%	87.10%	0.00%	27.35%
Single mother	0.00%	0.00%	60.55%	53.36%
Waves in poverty*	6.07%	29.60%	38.10%	43.14%
Waves with employed mother*	64.63%	57.33%	57.12%	53.87%
Family size across waves*	4.44 (1.05)	4.55 (1.24)	4.28 (1.45)	4.30 (1.41)
Female
Mother’s race/ethnicity*	48.10%	48.39%	50.00%	50.79%
Non-Hispanic White	65.70%	42.58%	24.41%	29.06%
Non-Hispanic Black	11.78%	24.52%	52.01%	42.98%
Non-Hispanic American Indian/Alaskan Native	0.22%	0.00%	0.77%	1.22%
Non-Hispanic other race	5.22%	5.81%	3.73%	2.93%
Non-Hispanic multiple races	0.65%	0.00%	1.30%	1.34%
Hispanic	16.44%	27.10%	17.77%	22.47%
Mother’s age at birth*	27.88 (4.90)	28.00 (4.93)	25.40 (5.78)	24.60 (5.22)
Mother’s education at birth*
Less than high school	10.07%	27.74%	27.47%	36.64%
High school	38.35%	52.90%	45.18%	46.94%
Some college	23.74%	14.19%	20.57%	14.83%
College or more	27.84%	5.16%	6.78%	1.59%
Mother drank alcohol during pregnancy*	46.51%	40.85%	37.63%	45.85%
Mother smoked cigarettes during pregnancy*	19.44%	43.26%	30.08%	43.01%
Mother used marijuana during pregnancy*	0.70%	5.94%	3.02%	4.32%
Mother’s parents finished high school*	72.87%	53.29%	49.75%	50.19%
Mother was foreign-born	8.84%	7.10%	7.05%	7.08%
Mother lived in South at age 14*	33.18%	31.37%	43.21%	37.72%
Mother lived in rural area at age 14*	23.33%	20.26%	20.56%	14.22%
Mother lived with two parents at age 14*	78.69%	69.48%	56.63%	52.52%
Mother’s adverse childhood experiences
Mental illness in household	9.64%	11.92%	8.25%	9.55%
Substance use in household*	20.43%	19.87%	18.35%	25.99%
Physical abuse*	12.92%	23.33%	16.13%	24.67%
Emotional neglect*	15.59%	16.67%	20.11%	27.18%
n	2,318	155	1,688	819

^*Significantly different at p < 0.05 across family structure history types.

In the next step of the analysis, the inverse probability of treatment weights allowed for an estimation of treatment effects across the entire population of youth as well as each subpopulation (i.e., those with a married, single, or cohabiting mother at each age). After creating the inverse probability of treatment weights, we confirmed that the ATE weights sufficiently balanced the baseline covariate distributions across youths’ family structure at age 0 using standardized differences in means (Austin and Stuart 2015). See Supplementary Figures S1–S2 for comparisons of the covariate balance before and after weighting. See Supplementary Figures S3–S6 for a graphical display of how the distribution of covariates changes across each of the three weighting schemes.

Examining Maternal Partnership Status and Early Adolescent Drinking

The first set of hypotheses was concerned with variation in the associations of maternal partnership status and adolescent adjustment by timing and sociodemographic background. We first compare these basic associations for each of the two outcomes (drinking and depressive symptoms) before exploring the aforementioned variability in these associations for each outcome. We start with the externalizing dimension of adolescent adjustment (alcohol use).

The initial component of this first set of hypotheses was that living with an unmarried mother during early childhood (ages 0–4) and early adolescence (12–14) would be most strongly associated with adolescent adjustment compared to other ages. Figures 2 and 3 display the odds ratios of alcohol use by age 14 for youth living with a single vs. married mother and for youth living with a cohabiting vs. married mother, respectively. Each line represents the ATE, ATU, or ATT created through IP weighting. We tested this first hypothesis based on the ATE estimates, which reweighted the sample to sociodemographically resemble the overall population of youth. Living with a single mother at age 10 and 14 predicted 80% and 57% greater odds, respectively, of drinking at age 14 compared to youth living with a married mother at those ages. Living with a cohabiting mother at birth and ages 6, 12, and 14 predicted between 85% and 292% greater odds of drinking at age 14 compared to youth living with a married mother at those ages.

Figure 2.

Treatment Effects of Living with a Single vs. Married Mother on Alcohol Use by Age 14, by Age

Note: ATE = average treatment effect, ATT = average treatment effect on the treated, ATU = average treatment effect on the untreated.

Figure 3.

Treatment Effects of Living with a Cohabiting vs. Married Mother on Alcohol Use by Age 14, by Age

Note: ATE = average treatment effect, ATT = average treatment effect on the treated, ATU = average treatment effect on the untreated.

The second component of the first set of hypotheses was that living with an unmarried (single or cohabiting) mother would matter most to adolescents who showed the least (vs. most) sociodemographic resemblance to youth in the typical unmarried mother household. It was tested using the ATT and ATU estimates shown in Figures 2 and 3. Starting with the comparison of youth who lived in single vs. married mother households in Figure 2, the ATU and ATT estimates reweighted the sample to sociodemographically resemble the population of youth living with married and unmarried mothers, respectively. According to ATU estimates, youth who lived with a single mother at birth and at age 10 had 51% and 88% greater odds of drinking at age 14 than those with married mothers at those ages, respectively. The ATT estimates revealed somewhat weaker associations between living with a single mother and drinking, although, according to these estimates, youth with a single vs. married mother at age 8 had 55% greater odds of drinking at age 14. In Figure 3, the ATU and ATT estimates reweighted the sample to sociodemographically resemble the population of youth living with married and cohabiting mothers, respectively. According to the ATU estimates, living with a cohabiting vs. married mother at birth, ages 2, 6, 10, and 12 was associated with between 66% and 191% greater odds of drinking at age 14. In contrast, the ATT estimates were all nonsignificant.

Examining Maternal Partnership Status and Early Adolescent Depressive Symptoms

Next, we turn to testing the first set of hypotheses for the internalizing dimension of adjustment (depressive symptoms). Again, we first used the ATE estimates to test the component of this set of hypotheses regarding the timing of living with an unmarried mother. They are shown in Figures 4 and 5. Compared to drinking, youths’ depressive symptoms at age 14 were somewhat less sensitive to family structure growing up. As shown in Figure 4, youth who lived with a single vs. married mother at age 6 on average scored 1.2 points greater on the depressive symptoms scale at age 14, with no differences between youth living with a single vs. married mother at any other age. There were no significant differences in depressive symptoms at age 14 for youth who lived with a cohabiting vs. married mother at any age (see Figure 5).

Figure 4.

Treatment Effects of Living with a Single vs. Married Mother on Depressive Symptomology at Age 14, by Age

Note: ATE = average treatment effect, ATT = average treatment effect on the treated, ATU = average treatment effect on the untreated.

Figure 5.

Treatment Effects of Living with a Cohabiting vs. Married Mother on Depressive Symptomology at Age 14, by Age

Note: ATE = average treatment effect, ATT = average treatment effect on the treated, ATU = average treatment effect on the untreated.

To test the component of the first set of hypotheses that living with an unmarried mother would be most consequential for youth who showed the least (vs. most) sociodemographic resemblance to youth in the typical unmarried mother household, we used the ATT and ATU estimates shown in Figures 4 and 5. When youth were reweighted to resemble those who lived with a married mother (the ATU estimates), living with a single mother at birth, age 6, and age 14 was associated with 0.5, 1.0, and 0.8 more points on the depressive symptoms scale, respectively. When youth were reweighted to resemble those who lived with a single mother, living with a single vs. married mother at any age was not associated with depressive symptoms at age 14. All ATU and ATT estimates for living with a cohabiting vs. married mother at any age were all nonsignificant, as were the ATE estimates.

These results partially supported the first component of the first set of hypotheses regarding sensitive periods of early childhood and early adolescence in the association between living with an unmarried mother and adolescent adjustment, with the evidence particularly strong for a sensitive period of early adolescence for drinking. Living with an unmarried mother during middle childhood, however, was also associated with some elevated risks. These results supported the second component of the first set of hypotheses stating that the association between living with an unmarried mother and adolescent adjustment would be stronger for those who least sociodemographically resembled the typical adolescent with an unmarried (particularly single) mother. For both drinking and depressive symptoms, the role of living with a single mother was stronger in the ATU estimates than the ATT estimates; for drinking, the differences between living with a cohabiting vs. married were also stronger in the ATU estimates than the ATT estimates.

As a sensitivity analysis, we compared the associations of living with a single vs. cohabiting mother and each adjustment outcome (results not shown; available upon request). Youth who lived with a single, compared to cohabiting, mother at age 12 had 52% lower odds of drinking, but the two groups did not significantly differ in depressive symptoms at age 14.

Comparing Variability in the Associations of Maternal Partnership Status with Internalizing and Externalizing Dimensions of Adolescent Adjustment

The second set of hypotheses was that early childhood would be a more sensitive period for maternal partnership status in the internalizing domain of depressive symptoms (vs. the externalizing domain of alcohol use) and that early adolescence would be a more sensitive period for maternal partnership in the externalizing domain of alcohol use (vs. the internalizing domain of depressive symptoms). We also hypothesized that sociodemographic moderation of associations between maternal partnership status and adjustment would be greater for drinking than for depressive symptoms.

The initial component of this second set of hypotheses was tested by comparing the ATE estimates of living with an unmarried mother at ages 0–2 and 12–14 for depressive symptoms vs. alcohol use. According to Z tests (Clogg, Petkova, and Haritou 1995), the differences between the ATE coefficients representing the effect of living with an unmarried mother at ages 0–2 on drinking vs. depressive symptoms were not statistically significant, nor were the differences in coefficients at ages 12–14. Thus, there was no evidence for different sensitive periods for alcohol use vs. depressive symptoms, although in general maternal partnership status was more strongly associated with adolescent drinking than with depressive symptoms.

The remaining component of this second set of hypotheses was tested by comparing the differences between the ATT and ATU estimates for depressive symptoms vs. drinking. The larger the differences between the two estimates, the greater the moderation by sociodemographic background. For alcohol use, the differences between the ATT and ATU estimates for living with a single mother at birth and living with a cohabiting mother at ages 10 and 12 were significant (p < 0.05). For depressive symptoms, the ATU and ATT estimates of living with a single mother at age 6 were significantly different (p < 0.05). Thus, these results partially supported expectations that the association between maternal partnership status and drinking would be more sensitive to differences in sociodemographic background than depressive symptoms, although both outcomes showed moderation by sociodemographic background. Furthermore, a lack of statistical significance does not necessarily rule out heterogenous causal effects by sociodemographic background (Morgan and Todd 2008). On average, the ATU estimates for living with a single or cohabiting mother at each age on drinking and depressive symptoms at age 14 were over four times larger in magnitude than the ATT estimates, suggesting considerable moderation by sociodemographic background.

Probing the Sensitivity of Results

Given research on gender differences in links between family structure and developmental trajectories (Chen and Jacobson 2012) and gender differences in responses to stress among adolescents (Rosenfield et al. 2005), we re-estimated the ATE models with interaction terms by gender and family structure at each age. Living with a cohabiting vs. married mother at age 6 was associated with greater depressive symptoms for girls than boys. These results partially supported the expectation that girls’ depressive symptomology would be more responsive to family structure than boys, but not the expectation that boys’ alcohol use would be more responsive to family structure than girls’ use. Given that the treatment effects largely did not differ by gender, there was also no evidence that the differences between the ATU, ATT, and ATU estimates were larger or smaller for boys vs. girls.

Because many youth do not drink regardless of family structure status even though family structure may influence the amount of drinking for those with an underlying susceptibility to drink, we estimated alternative negative binomial models for alcohol use at age 14. These models employed an outcome variable counting the number of days drank in the past month at age 14 rather than a binary outcome variable of any drinking at age 14. These alternative models revealed substantively similar results as those presented above (results available upon request), with slight differences in the ages at which the focal family structure variables were significant.

To gauge whether results might differ depending on the seriousness of the adjustment issue, we also conducted logistic models predicting any binge drinking (5+ drinks in one sitting) in the past month (results available upon request). Approximately 3% of youth reported binge drinking in the past month. There were no significant differences in any of the three weighting schemes or ages when comparing youth who lived with a single vs. married mother, perhaps due to the small number of youth who binge drank. Youth who lived with a cohabiting vs. married mother during early and middle childhood and at age 14, however, were significantly more likely to binge-drink, with stronger associations for those most likely to live with a married mother. This pattern was in line with the earlier findings that the association between living with an unmarried mother and adolescent adjustment was stronger for those who were least sociodemographically likely to live with an unmarried mother.

Discussion

This study applied IP weighting to longitudinal data to contextualize the association between living with an unmarried parent and early adolescent adjustment within developmental time, sociodemographic background, and dimensions of externalizing vs. internalizing behavior. IP treatment weighting is not commonly used by family scholars, but this study demonstrates how social science researchers can use this method to tackle core questions about how commonly studied aspects of the developmental significance of family experiences vary by sociodemographic background.

Beginning with timing, most significant associations between family structure and adolescent outcomes occurred during pre- or early adolescence and involved adolescent drinking rather than depressive symptoms. Recall that we tested the timing hypotheses using the average treatment effects in which youth were reweighted to resemble the general population of youth, and thus these effects are generalizable to youth in general rather than only to those living in a specific family structure. This pattern of timing was true for both the single vs. married mother comparison as well as the cohabiting vs. married mother comparison for drinking but not for depressive symptomology. It suggests that more proximal mechanisms may be at play in the association between unmarried mother families and adolescent drinking, such as declining parental monitoring during the developmental period in which youth gain more independence from adults and spend more time with their peers (Hoffmann 2017). The fact that living with a cohabiting vs. married mother was also associated with a greater risk of drinking suggests that less parental monitoring in an unmarried parent household is likely not the only mechanism linking this family structure to greater adolescent risk-taking. Although a second parent in the household may increase monitoring, supervision, and household resources regardless of partnership status, formal vs. informal partnerships tend to bring greater stability, resource-sharing, and role clarity, all of which can benefit youth (Brown 2004; Manning and Lamb 2004). A cohabiting parent could introduce family instability and role ambiguity that increases inter-family conflict and youths’ time apart from parents, thus diminishing any potential gains in adult supervision (Manning 2015).

Notably, there were also significant effects of living with a single mother during infancy (age 0) and middle childhood (ages 6–10), primarily for drinking. The significant association for living with a single mother in the first year of life are in line with previous research on the timing of exposure to family structure and instability (Cavanagh and Fomby 2019), suggesting that patterns of attachment during this period of intensive development could have rippling effects into adolescence. Taken together, these results hint at a sensitive period during early adolescence but also highlight that specific mechanisms may operate differently at different ages to link family structure with adolescent development. In particular, more research is needed to unpack the specific mechanisms linking family structure experiences in middle childhood with later outcomes.

Turning to the issue of variability by sociodemographic background, the IP weighting allowed for an exploration of the heterogeneity of the associations between living in a single or cohabiting mother family and the risk of drinking and depressive symptomology. Recall that the ATT estimates describe the effect of the treatment (i.e., living with an unmarried mother at a given age) among the subpopulation of youth who actually experienced the treatment, whereas the ATU estimates describe the effect of the treatment among the subpopulation of youth who did not experience the treatment (i.e., living with a married mother at a given age). Because these subpopulations differ by their background characteristics and experiences, differences in the ATU and ATT estimates can be interpreted as how associations vary by these background characteristics and experiences (i.e., selection processes). For both dimensions of adjustment outcomes, the risk of living with a an unmarried vs. married mother was generally stronger for youth who sociodemographically resembled those who lived with a married mother (i.e., the ATU estimates) and weaker for those who sociodemographically resembled youth who lived with a unmarried mother (i.e., the ATT estimates), regardless of the timing of this family structure. For example, Black youth who had experienced poverty throughout childhood and whose mothers had less education more sociodemographically resembled the average youth who lived with an unmarried mother, and these youth exhibited the weakest association between mothers’ partnership statuses and their drinking at age 14. This finding is consistent with previous research suggesting that family instability matters most to the development of the children least likely to experience it (Brand et al. 2019; Cavanagh and Fomby 2019; Cross 2020; Fomby and Cherlin 2007) and that the benefit of marriage for child development is limited to those who are sociodemographically likely to be married (Ryan 2012; Wasserman 2020).

The differences between the ATT and ATU estimates also highlight the importance of youths’ family structure history for the developmental significance of living with an unmarried mother. Youth who lived with an unmarried mother at a given age were more likely to have lived with an unmarried mother at younger ages; in other words, the ATT estimates pertained to these youth, while the ATU estimates pertained to youth who were more likely to have lived with a married mother throughout childhood and experienced parental divorce as young adolescents. Youth who had consistently lived with an unmarried mother may not experience that mother’s partnership status as stressful, obviating the need to turn to drinking as a coping mechanism or as source of peer integration. Conversely, youth who had always lived with married biological parents (and likely lived in communities where married parents were the norm) might have experienced a parental divorce as particularly disruptive, triggering early drinking. These findings also suggest that subgroup norms about single parenthood matter for youths’ emotional experiences of family structure. Future research should explore more in depth the social psychological mechanisms linking societal or group norms to the role of family structure in adolescent development.

Assessing differences in the associations of living with an unmarried mother and externalizing vs. internalizing dimensions of adolescent development helps to elucidate more precisely how and when living with an unmarried mother could factor into adolescent development. A greater sensitivity of drinking (an externalizing symptom) vs. depressive symptomology (an internalizing symptom) to family structure would suggest that the mechanisms linking family structure to adolescent development are more strongly connected with peers than mental distress. The differences in the strength of these associations between family structure and each of the two outcomes also may further differ by the timing of living with an unmarried mother and the sociodemographic background of youth. For example, if living with an unmarried mother is consequential for youth’s development because a lack of a second parent in the household leads to less supervision and/or youth in these households are more vulnerable to peer pressure due to a greater need for social integration, then early adolescence would be a sensitive period for externalizing behaviors like drinking but not necessarily for internalizing behaviors like depressive symptomology. Although the differences in the associations between living with an unmarried mother at any age and drinking vs. depressive symptomology did not reach statistical significance, adolescent drinking was generally more sensitive to living with an unmarried mother than depressive symptoms, particularly in early adolescence. Another type of variability is by sociodemographic profile. If youth who are least like the average adolescent with an unmarried parent are more likely to have negative outcomes when living with an unmarried parent due to additional social vulnerability, then this vulnerability may be more likely to manifest in social settings as externalizing behavior rather than signs of internalization. For example, youth who feel out-of-place because their family structure is not like their peers may be more likely to turn to drinking due to pressure or desire for social integration. Our results supported this line of thinking, demonstrating that the variation in the association between living with an unmarried mother and adolescent development by youth’s sociodemographic background was greater for drinking than depressive symptoms. Taken together, these results suggest that youth react to living with an unmarried mother in highly social ways that may put them at greater risk of accidents, substance use disorders, and criminal justice involvement.

In addition to posing new questions for future research, this study offers a methodological model for answering them that has many advantages over typical covariate-adjusted regression. First, this approach does not require the assumption of linear association between the covariates and the outcome as regression adjustment does (Thoemmes and Ong 2016). Second, the different weighting schemes allow for the detection of heterogeneity in the association of maternal partnership status with early adjustment depending on youths’ family structure history and sociodemographic characteristics (Morgan and Todd 2008), going beyond previous work that tends to focus on effect heterogeneity by a single variable (e.g., race or gender). Third, IP weighting uses formally defined estimands that demonstrate more clearly the potential effect of an intervention in a target population (e.g., the general population vs. the population who tends to experience the exposure or not). These latter two advantages are particularly important for developing and assessing policy. For example, the ATT estimates suggest that marriage does not strongly benefit youth from families who are sociodemographically likely to be unmarried (e.g., impoverished and with lower levels of education), thus encouraging marriage among these families without additional economic or parenting support will likely not improve adolescent current and future prospects (Antecol and Bedard 2007; Brown 2010).

While building on the findings of this study, future research will need to address some of its limitations. For example, the nature of the sample—which included children born over many years to a cohort of women born around the same time—and our restriction of the analytical sample to children who experienced adolescence in the most recent decades meant that we disproportionately focused on children born to older mothers and largely excluded children born to teenage mothers. The sample design of the NLSY79-CYA also excluded youth who lived with single fathers or in non-parental homes. Because our measurement of family structure did not take into account family structure experiences other than remarriages between waves, we may have underestimated family instability that occurred between waves among youth living in single or cohabiting mothers. Our estimates of the developmental risks of living with a single or cohabiting mother are thus conservative. Another limitation is that it is unlikely that the inverse probability of treatment weights captured all relevant aspects of youth’s sociodemographic background that predict living with an unmarried mother, despite our best efforts at including as many theoretically motivated predictors as possible. Thus, some part of the socializing role of family structure we saw in the ATE, ATT, and ATU estimates may in fact have reflected unmeasured selection processes that affect both mothers’ partnership status and adolescent development. Lastly, sample size limitations prevented consideration of further heterogeneity in the role of maternal partnership status based on the biological status of fathers.

Overall, this study has laid the groundwork for future research on family contexts of adolescent adjustment using a powerful method that is becoming more popular in the social and behavioral sciences. Further research can address additional questions such as: through what mechanisms do single parent families increase the risk of drinking and depressive symptoms among the youth who are statistically least likely to live in these family structures? What are alternative ways to support adolescents at risk of drinking and depressive symptoms besides intervening on family structure? In these ways, this research has potential to bridge the silos between different disciplinary approaches to the study of families and children.