Fast Track Intervention Effects on Family Formation

Abstract

The present study examines whether the Fast Track intervention, a 10-year randomized controlled trial with children at risk for conduct problems, affects family formation in adulthood, as indexed by partnerships, parenthood, and family structure and whether the intervention effect differs across participants’ gender and race/ethnicity. Participants included 891 children (intervention n = 445; control n = 446; 69% male; 51% Black, 47% White) who were recruited in kindergarten and followed to age 32 or 34 (80% participation of still-living participants), when they reported on their romantic partnerships, parenthood, and family structure. Controlling for numerous covariates that are related to family formation, intervention participants were more likely than those in the control group to be married rather than single and to have a larger number of children; the intervention and control groups did not differ on cohabitation status, age at first marriage, whether they had ever divorced, their likelihood of being a parent, the age at which they first became a parent, the spacing of births, family structure (partnered or not, with or without children), or in whether they were residentially independent of their parents and grandparents. Intervention effects were not moderated by gender, but race/ethnicity moderated the effect of the intervention on the probability of having any children and the number of children. These findings suggest that several elements of family formation may remain unchanged by an intervention that changes many other behavioral and psychological trajectories of participants.

Family formation patterns affect population size, labor market supply, and demand for social institutions (Poston & Bouvier, 2010). In addition, family formation and family structure predict a myriad of individual outcomes across developmental stages, from childhood and adolescence (Buehler, 2020) through adulthood and old age (Thomas et al., 2017). Consequently, understanding patterns of family formation remains an important goal for family scholars and psychologists. Adding to this line of research, the present study examines whether a behavioral intervention with children at high risk for developing conduct problems affects family formation in adulthood, as indexed by partnerships, parenthood, and family structure and whether the intervention effect differs across participants’ gender and race/ethnicity. To the best of our knowledge, ours is the first study to evaluate whether an early intervention has spillover effects on a variety of family formation outcomes in adulthood. However, in a review of early childhood interventions and teenage pregnancy prevention, two of four interventions studied significantly reduced teenage pregnancy but were not related to number of children or marriage by early adulthood (Arnold & Coyne, 2020).

The Fast Track randomized controlled trial tested the efficacy of an early intervention to prevent adult psychopathology and improve well-being in early-starting conduct problem children. Kindergartners in three cohorts (1991–1993) who screened at high risk for conduct problems were randomized into either a 10-year intervention, or matched control group (Dodge et al., 2015). The Fast Track intervention comprehensively addressed childhood risk factors for conduct problems by targeting children’s intrapersonal, interpersonal, and academic skills, and their parents’ parenting skills and behaviors through two intervention phases: elementary (grades 1–5) and secondary school (grades 6–10) (details provided in Conduct Problems Prevention Research Group [CPPRG], 2020).

The FT intervention had wide-ranging effects, including reduced aggression, delinquency, risk-taking, hazardous substance use, incarceration, and risky sexual behavior and improvement of academic skills through young adulthood (CPPRG, 2002, 2004, 2010, 2011; Dodge et al., 2015; Godwin & CPPRG, 2020). Although FT did not directly target family formation behavior, the significant effects of the intervention could “spill over” to affect family formation processes for participating children. In this paper we test this hypothesis by examining direct effects of assignment to the FT intervention on partnership, parenthood, and family structure. As a conceptual framework, we consider several hypothetical pathways for the impact of FT on family formation: an indirect personal and interpersonal skills effect, an indirect behavioral effect, and an indirect socioeconomic effect; although these pathways frame our expectation of the impact of FT on family formation, they are not tested in this manuscript.

First, FT might have impacted family formation patterns of the participating children by improving children’s personal and interpersonal skills (e.g., self-regulation, self-esteem, social competence) connected to the ability to form and sustain secure and stable relationships, which might have led to higher rates of partnerships and parenthood among the intervention group compared to the control group (Moilanen & Manuel, 2017; Xia et al., 2018). Consequently, the prevalence of cohabitation and marriage would be higher and the prevalence of divorce would be lower for the intervention group. In addition, the incidence of parenthood would be higher in the intervention group. Finally, the proportion of the participants living with a partner or a partner and children would be higher in the intervention group.

Second, FT might have affected family formation indirectly by improving underlying behavioral patterns of aggression, delinquency, and risk-taking that can be associated with lower rates of partnerships (Pietromonaco & Beck, 2015) and higher rates of early childbearing (Barrett et al., 2015) and short birth spacing (Crittenden et al., 2009). Accordingly, we would expect that by reducing externalizing behaviors, the FT intervention would increase participants’ probability of entering partnerships (cohabitation and marriage), partnership duration (fewer divorces), age of first birth, and birth spacing. We would also expect a higher proportion of individuals living with a partner or living with a partner and children in the intervention group.

Third, FT might have affected family formation indirectly through improving participants’ socioeconomic situation in adulthood. The FT intervention demonstrated effects on several predictors of socioeconomic status, including improving academic skills and decreasing incarceration and hazardous substance use through young adulthood (CPPRG, 2010; Dodge et al., 2015; Godwin & CPPRG, 2020). Because socioeconomic status in adulthood affects the likelihood of union formation, divorce, and childbearing (Karney, 2021), FT effects on socioeconomic conditions in adulthood may spill over into family formation.

Intervention effects might differ for boys and girls. Higher socioeconomic status, in terms of educational attainment and salary, is positively associated with union formation and parenthood for men but negatively associated with union formation and childbearing for women (Mayol-Garcia et al., 2021; Monte & Ellis, 2014; Monte & Knopp, 2019). Marked gender differences in family formation patterns have been documented with men entering unions and transitioning to parenthood later than women (U.S. Census Bureau, 2021) as well as men being more likely than women to never marry (U.S. Census Bureau, 2021) and remain childless through age 55 (Valerio et al., 2021). Differences in family formation patterns by race/ethnicity also have been documented, with Black and Hispanic women, on average, giving birth at younger ages and being less likely to be married at the time of becoming a parent than White women (Monte & Ellis, 2014), and Black and Hispanic families, on average, being more likely than White families to live in multigenerational households (Cross, 2018).

At age 25, random assignment to intervention had a positive impact on externalizing, internalizing, crime, and well-being outcomes for both males and females, with no significant difference in magnitude across groups (Dodge et al., 2015). The only significant gender x intervention interaction effect was for participants’ spanking of their offspring, with a decline in spanking for intervention participants for both males and females but a larger effect for males. By age 34, mothers who had been assigned to the FT intervention as children had lower depression symptoms, alcohol problems, drug problems, corporal punishment use, and food insecurity compared to control group mothers, effects that were not found for fathers (Rothenberg et al., in press).

At age 25, random assignment to intervention had a positive impact on externalizing, internalizing, crime, and well-being outcomes for both White and Black participants, with no significant interaction effect. Due to documented variation in family formation patterns by gender and race/ethnicity in other samples and variation by gender in other domains in the present study, in this study we tested for gender and race/ethnicity differences in partnerships, parenthood, and family structure.

The Present Study

The present study tests whether the Fast Track intervention received during childhood affects family formation of the children when they become adults and whether the intervention effect differs by gender or by race/ethnicity. Our preregistration documents are available through the Open Science Framework (https://osf.io/dcbfv/?view_only=7d5c9f9ff9fa45dd8a9c38898361f21c). In these analyses, we did not have a priori directional hypotheses but addressed the following specific questions. First, did the intervention affect cohabitation status, whether participants married, their age of marriage, and whether they divorced? Second, did the intervention affect whether participants became a parent and, if so, at what age? Third, did the intervention affect how many children participants have and how closely their biological children are spaced? Fourth, did the intervention affect their current family structure: single without children, single with children, partnered without children, partnered with children? Fifth, did the intervention affect whether participants live in a household with a parent or grandparent? Finally, in addressing these questions, we test whether the effects of the intervention on family formation differ by gender or race/ethnicity.

Method

Participants

High-risk elementary schools (n = 55) were selected for study participation based on neighborhood crime and poverty rates in four geographic areas: Durham, NC; Nashville, TN; rural Pennsylvania; and Seattle, WA. Within site, schools were matched on demographics and randomly assigned to intervention or control conditions. Kindergarteners were recruited from three consecutive cohorts (1991–1993) based on a two-step screening procedure (Lochman & CPPRG, 1995). First, teachers completed the Teacher Observation of Classroom Adaptation-Revised measure (Werthamer-Larsson et al., 1991) for 9,594 kindergarteners, yielding an aggressive behavior sum score. Then, parents of children scoring in the top 40% within cohort and site were solicited to complete a 22-item instrument based on the Child Behavior Checklist (Achenbach, 1991) to capture behavior problems at home. Standardized teacher and parent scores were combined to create a severity-of-risk screen score (CPPRG, 2020).

Children within cohort and site were rank ordered based on this screen score, and study children were recruited starting with the highest risk until designated sample sizes were reached within sites, cohorts, and conditions, yielding a sample of 891 children (intervention n = 445; control n = 446; see Supplemental Figure 1 for CONSORT flow chart). In the first year of the study, the average child age was 6.58 years (SD = 0.48). Sixty-nine percent of the sample was male; 51% was Black, 47% White, and 2% other race/ethnicity. Of the 2% of sample in the “other” category, 10 reported “other.” Among the 13 who reported more specific race/ethnicity information, 9 reported Hispanic, 3 reported Native American, and 1 reported Asian. At baseline, 33% of the sample lived with both biological parents, 43% lived with a single, non-cohabiting biological parent, 9% lived with a biological parent and her/his partner, and 8% lived with a biological parent and her/his new spouse. The remaining 8% lived with a non-biological parent including adoptive/foster parents, grandparents, other relatives, and unrelated guardians. The median years of education for mothers and fathers was 12. The average number of children in the household (including the participant) was 3.03 (SD = 1.50). Only one child per family participated in the study. No robust statistically significant differences between the intervention and control sample have been found across variables capturing child behaviors, cognitive and non-cognitive attributes, parenting behaviors including warmth and harsh parenting, and family socioeconomic status (CPPRG, 2004).

The Institutional Review Boards of the participating universities approved all project procedures. Parents completed written informed consent, and children provided oral assent. Parents (and children after grade 5) were compensated for completing interviews, and parents from the intervention sample received payment for program participation. Initial program participation was extremely high with 96% of parents and 98% of children attending at least 1 group session during grade 1. Among participating families, 79% of parents and 90% of children completed 50% or more of the group sessions provided. Although participation declined over time, due largely to residential moves, at least 80% of youth continued to participate during the adolescent phase of the intervention.

At approximately age 34, participants were interviewed to gather information about family formation used in the present study (n = 568, 67% of the living participants). To increase the sample size, if age 34 data were missing, data collected at age 32 were used. T-tests and chi-square difference tests comparing the age 34 sample (n = 568) and the sample with age 32 data only (n = 117) across 25 pre-intervention variables (capturing child behaviors and attributes as well as parenting behaviors of the participants’ parents) and demographic characteristics including intervention status revealed only 2 statistically significant differences for gender and site. The age 32 only sample was more likely to be male, more likely to be from Nashville, and less likely to be from Durham. Comparing this combined sample (n = 685, 80% of living participants) to the original sample (n = 891) across the same 25 variables revealed statistically significant differences for gender and site. The participants with adult data were more likely to be female and less likely to be from Nashville. Given the few significant differences among these tests, we concluded that attrition did not systematically alter the representativeness of the samples. Little’s Test (Little, 1988), however, revealed that data are not missing completely at random (χ² = 1,994.53, dof = 1405, p < 0.01). Therefore, full information maximum likelihood estimation was used to adjust for missing data. To improve precision in calculating the impact of intervention, these 25 pre-intervention and demographic variables were included in the statistical models as they might predict family formation patterns.

Measures and Procedures

Pre-intervention and demographic control variables.

The following 25 pre-intervention and demographic variables were included as control variables: parent depression, child hostile attributions, child aggressive behavior, parent’s appropriateness of discipline, parent’s family satisfaction, parent’s friendship satisfaction, parent’s physical punishment, parent’s verbal punishment, parental stress, socioeconomic status, child’s oppositional behavior, child’s externalizing behavior, parent’s warm and harsh discipline, child’s social competence, child’s letter and word identification score, child’s emotion recognition questionnaire score, parent’s warmth, child’s competent responses to social problem solving, neighborhood danger, Wechsler Intelligence Scale for Children score, child’s externalizing behavior risk screen score, race/ethnicity (Black/non-Black), gender, geographic site, and cohort.

Family formation measures.

Family formation was measured across three categories: romantic partnership status (current and life course patterns), parenthood (current and life course patterns), and current family structure using self-reported data. Current romantic partnership status was self-reported as single, living with partner, or married. Life course patterns of partnership were captured by indicators of never being married, ever divorced, and age at first marriage. Age at first marriage (in years) was based on data from age 32 and earlier data collections.

Current parenthood was measured using two variables: an indicator for having any children and number of children. These outcomes include biological and non-biological children to most accurately capture the parenting/family structure experience; sensitivity analyses were performed on similar outcomes capturing only biological children. Descriptively, among respondents with at least one living biological child under the age of 18, 58% of parents lived with all of their biological children 100% of the year, 77% lived with their biological children 50% or more of the time, 87% of parents actively parented their biological children (defined as living with the child for 20% of the year or more or seeing the child at least weekly and self-reporting involvement with parenting the child), 8% of parents have at least one biological child they never see, and 8% of parents have at least one biological child placed for adoption or in foster care. Parenthood across the life course was measured by age at first child’s birth (in years) based on earliest reported age across all waves of survey responses and minimum spacing between biological children.

Current family structure was measured using a categorical variable describing partnership status and parenthood simultaneously: single without children, single with children, partnered without children, and partnered with children. Distinguishing between married and partnered participants with and without children yielded categories with relatively small sizes that prevent statistical models from converging properly. Finally, an indicator for co-residence with parents or grandparents was created based on participant reports of living with a parent and/or grandparent based solely on age 34 data.

Analysis Plan

Multinomial and logistic regression models were used to test the main effect of Fast Track intervention status on each categorical and binary outcome. For count outcomes, including number of children and minimum years between children, negative binomial models were estimated to account for over-dispersion. Given that age at first marriage and first birth were reported in years rather than days, discrete-time survival analyses were performed. All models were estimated in Mplus using full information maximum likelihood and standard errors were clustered by kindergarten school because the intervention was implemented at the school level. The models also controlled for pre-intervention and demographic covariates. Given that data were combined across two waves of data collection, age at interview was included as a covariate. Given that 5% of participants were incarcerated at the time of the interview, potentially disrupting family formation, models also controlled for incarceration at the time of interview. There is no evidence of multicollinearity among the covariates as evidenced by the low variance inflation factors (mean VIF = 1.40, maximum VIF = 2.3). We ran sensitivity analyses without the pre-intervention covariates, and the substantive results remained unchanged. Therefore, the reported results include all covariates, as planned in our pre-registration documentation.

To assess whether the impact of intervention on family formation differs for men and women, we estimated multigroup models by gender using the knownclass feature in Mplus. Testing moderation within a multigroup framework allows for heterogeneity in the relations between outcomes and covariates and the residual variances for men and women (Muthén et al., 2016). The modeling strategy, therefore, more accurately captures gender differences in family formation previously found in the literature. The null model allowed the coefficients for all covariates to vary freely by gender except the intervention coefficients while the alternative model freely estimated all coefficients (including intervention) for women and men. A likelihood ratio test determined the significance of moderation by gender. The indicator for incarceration was excluded from these models because only one female participant was incarcerated.

Similarly, we examined whether the impact of intervention of family formation differs by race/ethnicity by estimating multigroup models by race/ethnicity (Black/non-Black). The indicator for incarceration was included in these models.

Transparency and Openness

This study’s design and analyses were preregistered on the Open Science Framework (https://osf.io/dcbfv/?view_only=7d5c9f9ff9fa45dd8a9c38898361f21c). We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study, and we follow APA Style Journal Article Reporting Standards (Kazak, 2018). Data and analysis code are available upon request from the corresponding author, and research materials are available at fasttrackproject.org.

Results

Partnership

As seen in Table 1, a greater proportion of the intervention sample was married compared to the control sample at the time of interview. Twenty-seven percent of the intervention sample was married, 18% was living with a partner, and 55% reported being single in adulthood; only 22% of the control sample was married, 20% was living with a partner, and 58% was single. The multinomial logit model revealed that the odds of being married relative to being single in adulthood were 1.52 times greater for the intervention participants relative to the control (p = .02). The intervention effect on the probability of living with a partner relative to being single was not statistically significant. Supplemental Table 1 provides the full set of results.

Table 1:

Main Effects of Intervention on Family Formation

	Intervention (n=346) Proportion/ Mean (SD)	Control (n=335) Proportion/ Mean (SD)	Intervention Effect OR/b (95%CI)
Partnership
Current Partnership Status
Single	0.55	0.58	-
Living with Partner	0.18	0.20	1.07 (0.72, 1.59)
Married	0.27	0.22	1.52 (1.06, 2.20)*
Never Married	0.64	0.66	0.83 (0.60, 1.15)
Ever Divorced	0.11	0.14	0.74 (0.47, 1.16)
Age at First Marriagea,b	25.12 (3.59)	23.62 (3.76)	0.05 (−0.24, 0.34)
Parenthood
Any Children	0.74	0.71	1.29 (0.88, 1.88)
Number of Childrenc	1.93 (1.81)	1.67 (1.55)	0.18 (0.03, 0.33)*
Age at First Birthb,d	21.74 (4.50)	21.65 (4.53)	0.087 (−0.09, 0.27)
Minimum Years between Childrenc,e	2.64 (2.65)	3.08 (2.73)	−0.163 (−0.38, 0.06)
Family Structure
Current Family Structure
Single - No Children	0.20	0.22	-
Single with Children	0.35	0.36	1.21 (0.75, 1.94)
Partnered - No Children	0.07	0.07	1.24 (0.61, 2.54)
Partnered with Children	0.38	0.35	1.50 (0.96, 2.35)
Co-residence with Parents/Grandparentsf	0.20	0.14	1.45 (0.86, 2.47)

^a.Collected at age 32 only and means based on married participants: intervention n=112, control n=108

^b.Analyzed within a discrete-time survival analysis model – beta coefficients reported.

^c.Analyzed within a negative binomial model – beta coefficients reported.

^d.Minimium age at first birth reported over lifetime: intervention n=268, control n=256

^e.Sample limited to partipants with children: intervention n=154, control n=137

^f.Collected at age 34 only: intervention n=287, control n=281

^*p < .05.

As a sensitivity analysis, we examined more detailed partnership variables that were available at age 32 but not at age 34: Married and living together with spouse, Living with a partner, Engaged and living together, Engaged and not living together, Have one girlfriend/boyfriend/dating one person regularly, Dating one or more people (at least once a week), Dating one or more people (one or more times a month), Not dating and have no girlfriend/boyfriend, Divorced, Separated from spouse, Widowed. Among the participants at age 32 who reported being single (n = 328), 37% reported being in a committed relationship (Engaged and not living together/Have one girlfriend/boyfriend/dating one person regularly), 8% reported dating one or more people at least once a week or one or more times a month, and 55% reported not dating or being divorced/separated/widowed. Using data from age 32 only, an alternative variable capturing partnership status was created with the following categories: Married, Living with partner, In a committed relationship but not living with partner, Single/dating but not dating one person regularly. At age 32, 24% of the intervention sample was married, 28% was living with a partner, 18% reported being in a committed relationship but not living with their partner, and 30% reported being single; only 20% of the control sample was married, 26% was living with a partner, 19% reported being in a committed relationship but not living with their partner, and 34% were single/dating but not dating one person regularly. The multinomial logit model revealed that the odds of being married relative to being single/dating but not dating one person regularly at age 32 were 1.62 times greater for the intervention participants relative to the control (OR = 1.62, CI = 1.01, 2.61, p = .05). The intervention effects on the probability of living with a partner and the probability of being in a committed relationship but not living with the partner relative to being single were not statistically significant. This alternative outcome variable replicated the original findings: intervention increased the probability of being married relative to being single/not in a committed relationship. Intervention was not significantly related to the probability of being in other partnership groups.

Similar proportions of both the intervention and control samples reported never being married (64% of intervention and 66% of control) and ever divorced (11% of intervention and 14% of control). Based on logistic regressions, the intervention was not statistically significantly related to either outcome. Although the average age at first marriage among the intervention group was slightly higher than that in the control group (25.1 years compared to 23.6 years), the discrete-time survival analysis revealed that intervention status was not significantly related to age at first marriage.

Parenthood

At the time of the interview, 74% of the intervention sample and 71% of the control sample reported having a child. The logistic regression model revealed that intervention was not significantly related to the probability of having a child. A sensitivity analysis revealed that intervention was not significantly related to the probability of having a biological child. Among intervention participants the average number of children was 1.93, and the average among control participants was 1.67. The negative binomial model revealed that intervention was significantly related to number of children with intervention increasing the expected number of children by 1.20 (p = .02). A sensitivity analysis revealed that intervention was significantly related to the number of biological children with intervention increasing the expected number of biological children by 1.25 (p = .004). The average age at first birth was the same for the intervention and control samples (21.7 years). A discrete-time survival analysis revealed that intervention status was not significantly related to age at first birth. Although the minimum years between birth of the participants’ children was somewhat lower among the intervention sample than control (2.64 years compared to 3.08), the intervention effect was not statistically significant based on the negative binomial model.

Family Structure

Among the intervention sample, 20% reported being single without any children, 35% reported being single with children, 7% reported a partner but no children, and 38% reported a partner and children. The proportions were similar in the control sample (22%, 36%, 7%, and 35%, respectively) with a slightly higher proportion reporting single and no children and a slightly lower proportion reporting being partnered with children. The negative binomial model testing the intervention effect on the probability of being partnered with children relative to being single with no children was not significant. Although 20% of the intervention sample compared to 14% of the control sample reported living with a parent or grandparent, the intervention effect was not statistically significant based on the logistic regression.

Moderation of Intervention Effects by Gender and Race/Ethnicity

Table 2 provides the proportions and means by gender, the intervention effects when all coefficients are freely estimated by gender, and the likelihood ratio tests of moderation by gender. There was no statistically significant evidence of moderation by gender. Supplemental Table 2 provides the full set of results for the moderation models.

Table 2:

Moderation of Intervention Effects by Gender

	Among Female Participants			Among Male Participants			LL Ratio Test of Moderation x2(dof), pvalue
	Proportion/Mean (SD)		Intervention Effect OR/b (95%CI)	Proportion/Mean (SD)		Intervention Effect OR/b (95%CI)
	Intervention (n=106)	Control (n=120)		Intervention (n=240)	Control (n=215)
Partnership
Current Partnership Status							1.01(2), 0.60
Single	0.55	0.59	-	0.55	0.58	-
Living with Partner	0.17	0.20	0.95 (0.36, 2.52)	0.19	0.20	1.14 (0.69, 1.89)
Married	0.28	0.21	2.31 (0.93, 5.72)	0.26	0.22	1.38 (0.88, 2.16)
Never Married	0.61	0.66	0.63 (0.32, 1.25)	0.65	0.66	0.92 (0.60, 1.39)	0.24(1), 0.63
Ever Divorced	0.14	0.14	0.80 (0.41, 1.58)	0.11	0.14	0.64 (0.33, 1.21)	0.66(1), 0.42
Age at First Marriagea,b	24.82 (3.73)	23.35 (4.16)	0.12 (−0.34, 0.58)	25.27 (3.54)	23.82 (3.48)	0.04 (−0.31, 0.39)	0.08(1), 0.78
Parenthood
Any Children	0.82	0.83	1.13 (0.55, 2.33)	0.70	0.64	1.34 (0.86, 2.09)	0.17(1), 0.68
Number of Childrenc	2.25 (1.78)	1.83 (1.40)	1.32 (1.06, 1.57)	1.78 (1.82)	1.573 (1.62)	1.11 (0.92, 1.31)	2.04(1), 0.15
Age at First Birthb,d	21.37 (4.35)	21.04 (4.51)	−0.04 (−0.33, 0.25)	21.92 (4.57)	22.09 (4.51)	0.20 (−0.04, 0.44)	2.02(1), 0.16
Minimum Years between Childrenc,e	2.81 (2.64)	3.31 (2.61)	0.81 (0.58, 1.05)	2.54 (2.67)	2.93 (2.80)	0.93 (0.67, 1.19)	0.56(1), 0.46
Family Structure
Current Family Structure							0.41(3), 0.94
Single - No Children	0.14	0.13		0.22	0.27
Single with Children	0.42	0.45	0.92 (0.41, 2.07)	0.32	0.30	1.25 (0.74, 2.13)
Partnered - No Children	0.05	0.04	1.00 (0.27, 3.75)	0.08	0.08	1.21 (0.60, 2.45)
Partnered with Children	0.40	0.37	1.24 (0.48, 3.20)	0.38	0.34	1.39 (0.84, 2.33)
Co-residence with parents/grandparentsf	0.19	0.10	2.65 (0.97, 7.27)	0.20	0.16	1.16 (0.64, 2.12)	1.42(1), 0.23

^a.Collected at age 32 only and means based on married participants: Female intervention n = 38, control n = 43; Male intervention n = 74, control n = 65

^b.Analyzed within a discrete-time survival analysis model – beta coefficients reported.

^c.Analyzed within a negative binomial model – beta coefficients reported.

^d.Minimum age at first birth reported over lifetime among parents: Female intervention n = 89, control n =107; Male intervention n = 179, control n = 149

^e.Sample limited to participants with more than one child: Female intervention n = 59, control n = 54; Male intervention n = 95, control n = 83

^f.Collected at age 34 only: Female intervention n = 97, control n = 109; Male intervention n = 190, control n = 172

Table 3 provides the proportions and means by race/ethnicity, the intervention effects when all coefficients are freely estimated by race/ethnicity, and the likelihood ratio tests of moderation by race/ethnicity. There was evidence of statistically significant moderation by race/ethnicity for the intervention effect on the probability of having any children and the number of children. Supplemental Table 3 provides the full set of results for the moderation models. At the time of the interview, 72% of non-Black intervention participants and 69% of non-Black control participants reported having a child while 75% of Black intervention participants and 74% of Black control participants reported having a child. Participation in the intervention increased the odds of having a child by 1.79 (p = .05) among non-Black participants but was not significant among Black participants. A sensitivity analysis revealed that race/ethnicity did not significantly moderate the impact of intervention on the probability of having a biological child. Among non-Black participants, intervention participants had an average of 1.87 children and control participants had an average of 1.44 children. Among Black participants, intervention participants had an average of 1.98 children and control participants had an average of 1.91 children. The multigroup negative binomial models revealed that race/ethnicity significantly moderated the intervention effect on number of children. Among non-Black participants, intervention was associated with a 1.45 increase in the expected number of children (p = .002). Among Black participants, intervention was not significantly related to number of children. A sensitivity analysis estimating the number of biological children also revealed significant moderation of the intervention effect by race/ethnicity. Among non-Black participants, intervention was associated with a 1.53 increase in the expected number of biological children (p < .001). Among Black participants, intervention was not significantly related to number of biological children.

Table 3:

Moderation of Intervention Effects by Race/Ethnicity

	Among Non-Black Participants			Among Black Participants			LL Ratio Test of Moderation x2(dof), pvalue
	Proportion/Mean (SD)		Intervention Effect OR/b (95%CI)	Proportion/Mean (SD)		Intervention Effect OR/b (95%CI)
	Intervention (n=163)	Control (n=171)		Intervention (n=183)	Control (n=164)
Partnership
Current Partnership Status							2.87(2), 0.24
Single	0.44	0.46		0.65	0.71
Living with Partner	0.17	0.21	0.73 (0.40, 1.35)	0.20	0.19	1.45 (0.83, 2.53)
Married	0.40	0.33	1.33 (0.83, 2.13)	0.15	0.10	1.86 (0.86, 4.00)
Never Married	0.46	0.51	0.78 (0.51, 1.19)	0.79	0.80	0.98 (0.53, 1.81)	0.72(1), 0.40
Ever Divorced	0.18	0.2	0.78 (0.47, 1.29)	0.06	0.09	0.38 (0.12, 1.23)	1.58(1), 0.21
Age at First Marriagea,b	24.40 (3.49)	23.35 (3.95)	0.06 (−0.29, 0.41)	26.69 (3.35)	24.37 (3.12)	−0.03 (−0.70, 0.63)	0.06(1), 0.80
Parenthood
Any Children	0.72	0.68	1.79 (1.00, 3.21)*	0.75	0.74	1.06 (0.59, 1.92)	191(*), 0.00
Number of Childrenc	1.87 (1.75)	1.44 (1.34)	1.45 (1.12, 1.79)**	1.98 (1.87)	1.91 (1.71)	1.02 (0.83, 1.22)	7.48(1), 0.01
Age at First Birthb,d	22.46 (4.71)	22.72 (4.71)	0.20 (−0.09, 0.48)	21.2 (4.27)	20.67 (4.14)	0.04 (−0.21, 0.28)	0.69(1), 0.41
Minimum Years between Childrenc,e	2.79 (2.64)	3.57 (2.67)	0.73 (0.54, 0.91)	2.51 (2.67)	2.68 (2.72)	1.02 (0.74, 1.29)	2.78(1), 0.10
Family Structure
Current Family Structure							3.67(3), 0.30
Single - No Children	0.17	0.22		0.22	0.23
Single with Children	0.27	0.25	2.05 (0.89, 4.75)	0.43	0.47	0.95 (0.48, 1.84)
Partnered - No Children	0.10	0.10	1.42 (0.55, 3.61)	0.04	0.04	1.59 (0.14, 18.0)
Partnered with Children	0.46	0.44	1.88 (0.98, 3.60)	0.31	0.26	1.52 (0.75, 3.08)
Co-residence with parents/grandparentsf	0.22	0.12	2.12 (0.99, 4.55)	0.17	0.16	1.09 (0.48, 2.46)	1.51(1), 0.22

^a.Collected at age 32 only and means based on married participants: Non-Black intervention n = 77, control n = 78; Black intervention n = 35, control n = 30

^b.Analyzed within a discrete-time survival analysis model - beta coefficients reported.

^c.Analyzed within a negative binomial model - beta coefficients reported.

^d.Minimum age at first birth reported over lifetime among parents: Non-Black intervention n = 115, control n =122; Black intervention n = 153, control n = 134

^e.Sample limited to participants with more than one child: Non-Black intervention n = 73, control n = 61; Black intervention n = 81, control n = 76

^f.Collected at age 34 only: Non-Black intervention n = 137, control n = 146; Black intervention n = 150, control n = 135

^*p < .05

^**p < .01.

Discussion

This study examined whether the Fast Track intervention, a 10-year intervention that was implemented with children at high risk of developing conduct problems, affected family formation when the participants were adults. Compared to the control group, the intervention group was more likely to be married rather than single at age 34 and to have more children (biological and non-biological). The control group and intervention group did not differ in their cohabitation status, age at first marriage, whether they had ever divorced, their likelihood of being a parent, the age at which they first became a parent, the spacing of births, their family structure (partnered or not, with or without children), or in whether they were co-resident with their parents or grandparents. The effects of the Fast Track intervention on partnership, parenthood, and family structure did not differ for males and females but did differ by race/ethnicity.

Overall, we found more evidence for similarities than differences between the intervention group and control group in family formation in terms of partnerships, parenthood, and family structure. However, the two differences we found, with the intervention group more likely than the control group to be married and have more children, are both indicators of more traditional family formation patterns. Although marriage and parenthood have become increasingly uncoupled over historical time (Hayford et al., 2014), marriage and parenthood remain connected for more educated and higher SES families, although even among college-educated young adults, first births are increasingly occurring prior to marriage (Cherlin, 2021). Family formation patterns have become an important part of social mobility, as married parents are better able to build wealth and pass this wealth on to their children compared with single parents (Gibson-Davis & Hill, 2021). Marriage also confers health benefits compared to singlehood and cohabitation (Carr & Springer, 2010), so an intervention that increases the likelihood of marriage may have both economic and health benefits.

Race/ethnicity also moderated the effect of the intervention. For non-Black participants, the intervention significantly increased the probability of having a child, and being assigned to intervention was also associated with having more children and more biological children; the intervention did not affect these outcomes for Black participants. Because only 2% of the sample reported being in a racial/ethnic group other than “Black” or “White,” the findings for the non-Black participants are driven by the White participants. Using data from nationally-representative samples, differences in family formation patterns by race/ethnicity continue to be found, even controlling for education (Conwell & Doran, 2021).

Previous research has suggested that interventions that prevent externalizing behavior can disrupt the intergenerational transmission of antisocial behavior, particularly for women (Rothenberg et al., 2018). However, our findings suggest that these gender differences do not extend to family formation itself. Partnership, parenthood, and family structure as defined in the present study involve the presence, timing, and combination of social roles. Gender differences may be more apparent when examining behaviors, such as caregiving, that occur in these social roles (Kerr & Capaldi, 2019).

An interesting question is why the intervention did not affect more of the family formation outcomes. Previous research has found that family formation patterns are stable across generations (e.g., Lansford et al., 2019). That is, children in the control and intervention groups may have followed similar patterns to family formation because their parents followed similar patterns. Thus, it is possible that although spillover effects from an intervention designed for a different purpose might have been expected through at least three pathways (i.e., affecting social skills that increase the likelihood of forming and maintaining social relationships; improving underlying behavioral patterns of aggression, delinquency, and risk-taking that predict family formation; and affecting adult socioeconomic outcomes in ways that are related to family formation), interventions specifically targeting family formation would be needed if the goal were explicitly to change partnerships, parenthood, and family structure.

It is also important to situate the findings in historical time and consider possible cohort effects. Major demographic shifts in family formation have occurred over the last several decades (Smock & Schwartz, 2020). In the United States and many other countries, the average age at first marriage has increased, and larger proportions of both men and women are choosing to remain unmarried (U.S. Census Bureau, 2021). Likewise, the average age at first becoming a parent has increased, and larger proportions of men and women are choosing not to become parents (Frejka, 2017; Mathews & Hamilton, 2016; OECD, 2018). Rates of divorce, remarriage, and cohabitation have also changed, and parenthood has been increasingly uncoupled from marriage (Hayford et al., 2014). Consequently, it is possible that any hypothesized effects of the FT intervention on family behavior were subsumed by larger structural changes in family formation patterns in the United States. In addition, specific historical events when participants in the present study were in their early 20s, such as the Great Recession from 2007 to 2009, might have mitigated intervention effects on family formation patterns (Guzzo, 2022). Thus, the findings should also be considered in light of the historical period in which they are situated.

Strengths, Limitations, and Future Directions

This study had several important strengths, including the long-term follow-up through age 34 to examine the effects on family formation in adulthood of a 10-year intervention that began when participants were in kindergarten. The findings should be considered in light of the study’s limitations. First, if males were unaware of children they had fathered, they may have underreported their parenthood compared to females. Men have been found to self-report 80–90% of births documented in Vital Statistics and by the Census Bureau (Joyner et al., 2012), suggesting that the self-reports used in the present study would have captured the large majority of parenthood experiences. Second, we documented family formation through age 34, but with increasing average ages at the time of first marriage and parenthood, we cannot conclude that the participants who had not married or become parents by age 34 would never marry or become parents. In national samples of Americans, the average age at the time of first birth was 23.1 for women and 25.5 for men (Centers for Disease Control and Prevention, 2017), and the median age at first marriage was 28.6 for women and 30.4 for men in 2021 (U.S. Census Bureau, 2021). Thus, by following the sample through age 34, we would have captured the majority but not all first marriages and births. Third, family structure is fluid over time (e.g., Gillespie & Lei, 2021). We assessed a snapshot of family structure at age 34 in terms of whether participants were partnered or not and living with children or not as well as whether they were living in a household with their own parents or grandparents. As with the age of marriage and parenthood variables, these snapshots do not imply that participants will never live in a particular family structure but only their current family structure. Fourth, we recognize that other factors we controlled for but did not investigate substantively in the present study, such as SES, could have moderated the relation between participation in the Fast Track intervention and family formation.

In addition to future research to address these limitations, an important future direction will be to test models that include indirect effects to test mechanisms through which early intervention might have long-term effects on family formation patterns in adulthood. Another future direction will be to understand how childhood interventions in one generation affect not only family formation but also the quality of romantic partnerships and parent-child relationships when these children form families in adulthood, as well as the quality of these adult children’s relationships with their parents and grandparents when they continue to reside together. A new area of research suggests that intervention with one generation can continue to pay dividends in the next generation by improving behavioral and academic outcomes for offspring of the original intervention participants (Hill et al., 2020; Rothenberg et al., 2022). Thus, examining not only family structure but also family processes and relationship quality is central to future research on childhood intervention effects on adult families.

Implications for Policy and Practice

Social policies sometimes explicitly try to shape family formation or structure. For example, the one child policy in China (Alpermann & Zhan, 2019), the zero population growth strategies in the 1960s and 1970s in the United States (Davis, 1973), current family support legislation meant to encourage fertility in low-fertility countries such as Italy (Luci-Greulich & Thévenon, 2013), and the advent of no-fault divorce laws that made it easier for couples to divorce in the United States (Nakonezny et al., 1995) all illustrate ways that social policies can shape individuals’ experiences with partnerships, parenthood, and family structure. Other social policies do not explicitly target family formation or structure but can affect individuals’ opportunities and decisions related to family formation and structure indirectly. For example, fertility declines in many countries during times of recession when unemployment and economic uncertainties are high (Matysiak et al., 2021). Our findings that an intervention that was not designed to affect family formation but that affected many other areas of development in adulthood suggest that countries with goals related to family formation would be better served by policies that directly address those goals rather than more indirect approaches.

Interventions that target family formation have most often been delivered as teen pregnancy prevention programs that aim to delay parenthood (Lugo-Gil et al., 2018). Clearly, delaying parenthood is only one component of family formation, although an important one. Our findings suggest that, as with policies, interventions that are effective in changing a targeted set of outcomes do not necessarily generalize in their effects to other outcomes not specifically targeted in the intervention, even if the intervention changes potential behavioral and socioeconomic precursors to that outcome. The specificity principle in developmental science holds that specific developmental outcomes are dependent on specific experiences of individuals with specific characteristics in specific cultural contexts at specific points in time (Bornstein, 2019). This level of specificity highlights the importance of articulating clear theories of change in the development and implementation of interventions with particular target outcomes.

Conclusions

We conclude that the Fast Track intervention, a 10-year intervention with kindergarteners who were at risk for conduct problems, had few effects on partnerships, parenthood, and family structure when the original child participants became adults. Although intervention participants were more likely than those in the control group to be married rather than single and to have a larger number of children, the intervention and control group did not differ on cohabitation status, age at first marriage, whether they had ever divorced, their likelihood of being a parent, the age at which they first became a parent, the spacing of births, their family structure (partnered or not, with or without children), or in whether they were co-resident with their parents or grandparents. These findings suggest that several elements of family formation may remain unchanged by an intervention that changes many other behavioral and psychological trajectories of participants.