Abstract
The structure and stability of families have long stood as key predictors of juvenile delinquency. Boys from “broken homes” experience a higher prevalence of juvenile delinquency than those from intact families (Rebellon 2002, Wells and Rankin 1991). Unresolved is whether the consequences of frequently disrupted family contexts endure to shape criminal trajectories into adulthood. Long-term influence may also be indirect. Life course criminologists credit family formation during the transition to adulthood, and particularly marriage, for redirecting men’s criminal trajectories, but children who experience repeated changes in family structure are more likely to experience precarious starts to their own eventual family formation. Using data from the Panel Study of Income Dynamics and its two child-centered supplemental studies (N=1,127), we find that the experience of repeated family structure change is associated with higher rates of arrest and incarceration during early adulthood for white men but not for black men. This association is partially mediated by a slower transition to marriage among men who experienced three or more changes in family structure during childhood.
Introduction
The study of criminal behavior once centered on juvenile delinquency. In the absence of longer-running longitudinal data, much research focused cross-sectionally on the teenage years – a portion of the life course during which criminal behavior peaks. The family thus played an important role in understanding youthful offending. Criminologists have argued that an assortment of characteristics ranging from the size of the family to parents’ childrearing methods matter (Farrington 2011). No characteristic has been as extensively researched as family structure, however (Wells and Rankin 1991). Popularized by Bowlby’s (1951) World Health Organization report, the conclusion that “broken homes” cause crime has been a sturdy one. Meta-analyses consistently find that children who are raised in homes in which at least one biological parent is absent face a higher prevalence of adolescent delinquency (Rebellon 2002, Wells and Rankin 1991).
With the maturation of life course studies, criminological attention has moved beyond the teenage years to understand criminal behavior over the entire life course. In particular, criminologists have sought to explain why some offenders persist in, while other offenders desist from criminal activity as they enter adulthood. Again, the role of the family has emerged as pivotal in steering criminal trajectories. Life course criminologists contend that marriage, in particular, has the potential to serve as a turning point, redirecting criminal trajectories (Laub and Sampson 2006, Sampson and Laub 1995).
Yet for all of their emphasis on the long view of criminal behavior, life-course criminologists have been relatively short-sighted in their view of the family. We know little about whether childhood family experiences endure to impact adult criminal outcomes. Moreover, there is an increasing awareness of the ways childhood family instability undermines transition to adulthood experiences including one’s own family formation (Fomby and Bosick 2013, Hofferth and Goldscheider 2010). This suggests that if early family instability does not directly contribute to adulthood criminality, it may fuel adult crime by undermining the transition-to-adulthood experiences that might otherwise curtail it.
In this paper we use the long-running US Panel Study of Income Dynamics (PSID) and its two child-centered supplements, which have followed a nationally-representative cohort of youth since 1997 to early adulthood in 2015, to ask whether the experience of repeated changes in mother’s union status during childhood are predictive of men’s higher likelihood of arrest or incarceration during early adulthood (ages 18–26). Furthermore, we consider whether any such association is mediated by differences in the pace of men’s entry into marriage and their higher risk of union dissolution depending on prior exposure to maternal union instability. We consider whether these pathways operate differently for white and black men, an expectation based on research demonstrating that frequent family instability is less consequential for the life course of trajectories of black compared to white young adults (Fomby, Mollborn and Sennott 2010).
Literature review
The life course literature has long acknowledged the role of family formation in redirecting criminal paths. Men who as juveniles were involved in crime and delinquency are less likely to persist in criminal involvement if and when they become married (Sampson and Laub 1995). Criminally active men who remain single are, in contrast, more likely to continue offending well into adulthood. Increasingly, scholars have sought to contend with the complicated lived experiences that render modern family life more complex and varied than in the past, including divorce, cohabitation, nonresidential parenthood, and multipartner fertility. Findings consistently indicate that cohabitation is related to desistance, but its not as strongly associated as marriage (Forrest 2014). Moreover, men who experience divorce or separation experience upticks in their criminal activity, particularly when not living with a spouse (Horney, Osgood and Marshall 1995). Work focusing on the association between parenthood and offending has arrived at less consistent findings, though men’s desistance generally appears to be unmotivated by fatherhood (Blokland and Nieuwbeerta 2005).
There is an increasing tension hinted at when the work in this area calls out analyses for using “old” data. What feels antiquated may not be the data itself. After all, longitudinal and life course researchers appreciate that the aging of data is inherent to prospective longitudinal data. More problematic is that the idea of stable marriage grows increasingly antiquated, especially in contexts in which criminal behavior is most likely. The median age at first marriage for women has increased from 24 in 1990 to 27 today. Black Americans marry at lower rates than whites in every age group, and particularly so during their late twenties, when 115.6 out of every 1000 white women but only 43 out of every 1000 black women marry (Aughinbaugh, Robles and Sun 2013, Raley, Sweeney and Wondra 2015). Thus the argument that marriage causes desistance strikes a conservative chord, and is ill-matched to the most problematic offenders who are unlikely to marry at rates similar as offenders who transitioned to adulthood in decades past (Sampson and Laub 1995).
The evolving historical context of marriage reinforces the selection processes that have hampered causal claims about the effect of marriage on criminal outcomes. Clearly, marriage is not randomly assigned; there are forces shaping who becomes married and who does not. Sampson and colleagues (2006) addressed selection issues using a counterfactual approach and found continued support for a marriage effect. That marriage remains effectual is important. Still, we contend that it is also important to understand the selection processes themselves.
The family literature recognizes the impact of repeated family structure change on the children who experience it. Above and beyond having been raised in a single-parent household, or having experienced parental divorce, experiencing family instability appears to undermine educational outcomes, problematizes one’s transition to adulthood, and contributes to risky and delinquent behavior (Fomby and Cherlin 2007, Fomby et al. 2010, Fomby 2013, Fomby and Bosick 2013, Lee and McLanahan 2015, Wu and Martinson 1993). The extent to which these setbacks persist into adulthood is less clear.
Static theories of criminal persistence suggest a direct route from early disadvantages that would include family instability to later criminal behavior. On the other hand, early family instability may produce problems in adulthood through a process of cumulative disadvantage. “Dynamic” theories recognize that setbacks accumulate throughout the life course and grant a causal connection between earlier and later problems (Sampson and Laub 1997). Early family instability is associated with disrupted transition to adulthood experiences of the children who experience it. Thus, early family instability may influence adult offending indirectly by disrupting the very transitional experiences thought to curtail criminal behavior.
This paper advances the literature by taking a longer view of the role of family on criminal outcomes in adulthood. We bring a dynamic measure of family instability to the criminological literature in order to understand whether the association between early family instability and misbehavior persists into adulthood, and whether this association is mediated by men’s own family formation experiences.
Further, we consider whether this approach is equally useful for describing the probability of offending in early adulthood among non-Hispanic white and black men. A substantial literature has demonstrated that family instability is more strongly associated with externalizing behavior, delinquency, and early transition to adulthood experiences among white adolescents and young adults compared to their black peers. Little research has sought to explain racial differences in the influence of union instability on behavior, but some research suggests that black youth may have more social protection through enduring relationships with extended kin, neighbors, and peers following family disruption compared to white adolescents; there is also some evidence that economic stress may overwhelm the effects of repeated family structure change on behavior, and economic stress occurs more often in black compared to white families (Fomby et al. 2010). Given the disproportionately high incarceration rates among black young adults in the United States and the popular perception that family structure in black children’s families plays a causal role in criminal behavior leading to arrest and incarceration, it would be an important contribution to demonstrate the absence of any such association if none is present.
Methodological background
Cross-sectional studies characteristically struggle to establish temporal order, making causal connections difficult to infer. Mature longitudinal studies more successfully track the temporal order of life events, but leave researchers the task of capturing this order in their analysis.
To capture the diverse experiences navigating the transition to adulthood period, researchers have employed methods including latent class analysis (e.g. Macmillan and Eliason 2003, Osgood, Ruth, Eccles et al. 2005) and conjunctive analysis of case configurations (Doherty and Cwick 2016). These methods allow researchers to identify and describe the most typical role configurations and summarize the typical pathways through adulthood. These strategies are useful in describing conceptually related events in the life course. Analyses become more complicated as they move toward establishing causal order between life experiences.
The most straightforward analyses are those which examine experiences in one period of life on experiences in later life. Macmillan’s (Macmillan and Hagan 2004) use of the National Youth Survey to identify a “chain-like” sequence in which victimization experienced in adolescence diminishes educational self-efficacy at 18, which ultimately undermines socioeconomic attainment in early adulthood. In studies of the relationship between transition-to-adulthood experiences and crime, researchers recognize that these events and behaviors are occurring simultaneously. In order to sort out sequence, some impose temporal order by censoring the data into distinct age periods. Bosick (2015), for instance, dodges the issue of overlapping events by modeling transition-to-adulthood in early adulthood and examining the association between these pathways and later adulthood crime.
Other strategies include modeling risky and criminal behavior as part of the transition to adulthood, where both transitional experiences and criminal activity are simultaneously taken as indicators of a latent construct (Massoglia and Uggen 2010). Uggen and Janikula (Uggen and Janikula 1999) have used event history analysis to capture how time-varying voluntary labor experiences influence the duration until arrest. Similarly, Uggen (2000) tests for job-treatment effects by randomly assigning over 3,000 people with arrest histories to either a control or treatment group. He use an event history approach in order to aid in determining the sequence of work and criminal behavior. Horney, et al (Horney et al. 1995) use hierarchical linear modeling to capture the month-to-month impact of life circumstances (e.g. living with a romantic partner) on criminal behavior.
Data and Method
We use data from The Panel Study of Income Dynamics (PSID) and two of its supplemental studies, the Child Development Supplement (CDS) and the Transition into Adulthood Supplement (TAS). PSID began in 1968 as a nationally-representative sample of approximately 4,800 households. Original respondents and their descendants have been followed annually until 1997 and biennially since then. To maintain population representativeness, a sample refresher in 1997 added approximately 500 households headed by immigrants who had entered the United States since 1968. At each wave, the household head or the spouse or cohabiting partner of the head reports on family household composition, employment, earned and unearned income, assets, debt, educational attainment, expenditures, housing characteristics, and health and health care in the household. In 2015 (the most recent wave available), the study collected information on almost 25,000 individuals in approximately 9,000 households.
The PSID Child Development Supplement (CDS) is a longitudinal study of children’s development in family, school, and neighborhood context designed to identify and articulate the circumstances and characteristics of childhood experience that are predictive of status attainment and well-being across the life course (McGonagle, Schoeni, Sastry et al. 2012). The first wave of CDS, conducted in 1997, collected information on up to two eligible children aged 0 to 12 years per PSID household through interviews with children’s primary and secondary caregivers and with older children and through assessments and interviewer observations. Children and their caregivers were re-interviewed in 2002 and 2007, or until children reached age 18.
Beginning in 2005, the Transition to Adulthood Supplement (TAS) absorbed children from the original CDS cohort when they reached age 18 or left high school and has continued to interview respondents biennially. TAS is conducted as a telephone survey interview and collects information on transition-to-adulthood events such as school enrollment and completion, employment, cohabitation, marriage, and childbearing, as well as information about health, attitudes, expectations, social relationships, and illegal behavior and contact with the criminal justice system. In 2015, TAS respondents were between 18 and 27 years old. The youngest respondents (born 1996–97) participated in TAS for the first time in 2015, while older respondents participated in as many as five waves.1
Our analytic sample includes young men from the 1997 CDS cohort who identify racially as non-Hispanic white or non-Hispanic black, who responded to at least one wave of the Transition to Adulthood Supplement between 2005 and 2015, and who provided information on whether they had ever been arrested, on probation, or incarcerated (N=1,127). We exclude young adults from the 1997 immigrant refresher for two reasons. First, our analysis includes an indicator of parents’ history of contact with the criminal justice system, which was reported in 1995, prior to the inclusion of the immigrant refresher; and second, our analysis focuses on non-Hispanic white and black young adults, who appear only in small numbers in the immigrant refresher. Young adults who were not observed at least once in TAS were younger on average, more often non-white, and from families with lower household income at birth. We also exclude information about young adults at age 27 because small cell sizes contribute to unreliable estimators when predicting outcomes at that age. Weighted statistics are representative of young adults aged 18 to 26 years who were born between 1985 and 1997 and whose families have resided in the United States since 1968.
Dependent variables
Our analysis predicts TAS respondents’ contact with the criminal justice system in each year between age 18 and age 26 as a function of family structure history in childhood and own family formation in early adulthood. At each wave, respondents are asked in separate questions whether they have ever been arrested or in jail for an offense and if so, whether the event occurred once or more than once. (To protect respondent confidentiality, the actual number of occurrences beyond one is not available.) In follow-up questions, respondents report their age at the first event and at the most recent event. The same question format is administered to respondents at each wave; hence, over time, respondents may provide discrepant reports of the timing of the same event. In the absence of any external validation, we use all available information and thus may overestimate event occurrence. From these reports, we construct dichotomous indicators of whether a respondent experienced each type of contact with the criminal justice system at each age.
Key independent variables
Our attention to young adults’ family structure history is focused on their exposure to maternal union instability, or a mother’s repeated changes in union status when a young adult was between 0 and 17 years old. To construct a mother’s union history, we relied on five sources of information. First, for all adults age 15 or older ever observed in a PSID household, the marriage history file provides an accounting of the start and end dates of each marriage, a unique identifier for each spouse, and, where relevant, the reason the marriage ended (death, divorce, or separation). Second, the birth history file includes information about each live birth experienced by those 12 years and older,2including each child’s birth date and the unique identifier of the child’s other biological parent. Third, at each interview, the household roster identifies couples in the household who are married to or cohabiting with each other through the marital pairs indicator. Fourth, the household roster also gathers information on the move-in dates of all new household members who have joined and the move-out dates of all former household members who have departed since the last interview. Finally, the Family Identification Mapping System (FIMS) links each individual ever observed in a PSID household to the unique identifiers for their biological and adoptive parents, grandparents, great-grandparents, and siblings wherever those individuals have also been observed in a PSID household.
We use FIMS to identify the biological or adoptive mother and father of each TAS participant and apply this information to link each parent back to their own marriage and birth history in order to determine whether a young adult’s mother was married to her child’s biological/adoptive father or to another partner at each year of the young adult’s life up to age 17. If the marriage history indicated that the mother was unmarried in a given year, we used the marital pairs indicator from that year’s PSID household roster to determine whether she was in a cohabiting relationship with the child’s father or with another partner. Finally, we used the move-in and move-out date information to determine when a mother’s cohabiting partner entered and left a young adult’s household in order to determine her cohabitation status in off-years after PSID moved to biennial interviewing.
From this information, we constructed a three-category measure of mother’s union status at each year of a young adult’s life from birth to age 17: unpartnered; married to or cohabiting with a child’s biological father; or married to or cohabiting with another partner. (Sample sizes were insufficient to allow us to break out marital and cohabiting unions into separate categories.) From this annual accounting, we constructed a summed indicator of the total number of maternal union status changes a child experienced to age 17. Approximately 42 percent of children experienced at least one change in maternal union status and 20 percent experienced 3 changes or more. We topcoded the count variable at 3 or more changes in union status. We treat this topcoded variable as categorical, assessing whether the likelihood of arrest or jail is different on average for young adults who experienced 1, 2, or 3+ changes in maternal union status compared to those who experienced no changes in union status. Our primary focus is on young adults who experienced three or more union status transitions. This approach allows us to capture nonlinear associations between maternal union instability and child outcomes and is consistent with the conceptualization of family instability as an experience of frequent or chronic change to which family systems struggle to adapt (Fomby and Mollborn 2017). To isolate the association of maternal union instability with each outcome, we also control for maternal union status at birth and in late adolescence (i.e., age 16 or 17 in most cases) (Wu and Martinson 1993).
We consider whether the experience of three life course transition events mediates any observed association between exposure to maternal union instability in childhood and the likelihood of arrest or incarceration in each year during early adulthood. At each TAS wave, respondents report whether they are currently or have ever been married, and if so, at what age the union began. If the marriage has ended, respondents also report their age at union dissolution. Respondents who are cohabiting with a romantic partner at interview report the start date of the current cohabiting union. From these reports, we construct time-varying measures of marital and cohabitation status at each age from 18 to 26. (Note that cohabiting unions that begin and end between biennial interviews are not reported. The same issue arises in the core PSID interview.) Respondents also report on whether they have ever had any biological children and if so, at what age the first child was born. From these reports, we constructed a time-varying measure of parenthood status. In the time-varying measures of union status, a young adult may move into or out of the state of being in a union over time. In contrast, in our coding scheme, a young adult cannot exit the status of being a parent. In our multivariate analyses, indicators of marriage, cohabitation, and parenthood are lagged one year.
Control variables
A variety of circumstances and characteristics also may be associated with young adults’ prior exposure to maternal union instability and with events in early adulthood, including contact with the criminal justice system and union and family formation. To isolate the association between the variables of interest in our empirical model, our multivariate analyses include statistical controls for sociodemographic characteristics including young adult gender and age, mother’s age at the child’s birth, and the family income-to-needs ratio in 1997 (the year CDS began). Because we anticipate that the association of maternal union instability with criminal offending in early adulthood will differ for white and black young adults, we included a dichotomous indicator for race and interacted this with the categorical measure of maternal union instability in our multivariate models.
A substantial body of research has demonstrated that exposure to parents’ relationship instability is associated with children’s and adolescents’ externalizing behavior and delinquency. We included reports of young adults’ behavior measured at two points in time. First, at each wave of CDS, primary caregivers completed the Behavior Problems Index (Peterson and Zill 1986) an inventory of occasional or frequent child behaviors that are predictive of psychological disorders and poor social adjustment. We used the 17-item BPI externalizing behavior subscale from the earliest CDS wave available (wave I for children who were 3 or older in 1997, and wave II otherwise). Second, at each TAS wave, young adults report how often they have been in a physical fight or have damaged public or private property in the last six months. We dichotomized each item at 2 occurrences or more vs. none or one and used the earliest report available – typically reported when young adults are between 18 and 20 years old – to create a two-item index describing frequent illegal behavior in late adolescence/early adulthood.3To account for the possibility that parents who have criminally offended may be more likely both to experience union instability and to have children who offend, we included an indicator of whether either parent had ever been arrested or charged with a crime as of the 1995 PSID interview. Models also adjusted for whether the young adult’s family was originally included in the SRC (general population) sample or SEO (low-income) oversample and whether the young adult’s family participated in the Core PSID interview consistently when the young adult was growing up.
Our multivariate regression models address three questions: First, is there an association between exposure to maternal union instability and contact with the criminal justice system in early adulthood? Second, does any such association differ by race? Third, do union formation and childbearing mediate any observed association? To address these questions, we used logistic regression to estimate the log-odds of arrest or incarceration each year from age 18 to age 26 as a function of exposure to maternal union instability, the interaction of that exposure with young adult race, own family formation events, and background characteristics. Our data are organized in person-year format with each individual contributing up to nine records. Models are clustered on individuals’ unique identifiers to adjust for non-independence of observations. We explored random effects models as an alternative specification to account for this non-independence and to exploit the panel nature of the data. The panel-level variance component (rho) was not statistically significant from zero, indicating that variation in the probability of experiencing the outcome across years is unimportant in the model and that the random effects model was no different from the pooled model used here. For each outcome, we present two model specifications, one including only our indicators of maternal union status change, race, and background characteristics and one that also accounts for whether a respondent was married, cohabiting, or a parent in the preceding year. To facilitate interpretation of results, we discuss estimated probabilities of experiencing arrest or incarceration by level of maternal union status change and race.
Coefficients from nested logistic regression models cannot be directly compared to assess mediation. In order to assess the impact of accounting for a young adult’s recent union formation or dissolution and parenthood, we present results from the KHB method for decomposing total effects (or associations) into direct and indirect effects. This method compares the full model with a reduced model that substitutes the mediators with the residuals of those mediators from a model in which the mediators are regressed on the key variable of interest (Breen, Karlson and Holm 2013).
Results
Table 1 summarizes weighted descriptive statistics for the dependent variables, key variables of interest, and control variables overall and by race from the person-year file. Statistically significant group differences by race (p<.05) are starred. Overall, 4.8 percent of men were arrested in each year and 1.5 percent were incarcerated. Arrest and incarceration probabilities were about twice as high for black compared to white men. Considering the key variables of interest, approximately one in five men experienced three or more changes in maternal union status. One-sixth of white men and nearly one-third of black men experienced such union instability. In young adulthood, approximately 4 percent of men were married, and 3.6 percent were cohabiting in each year, and 11 percent had become a father. The probability of marriage was lower and the probability of fatherhood higher for black men compared to white men. For both white and black men, approximately 12 percent had a parent who had been arrested or charged with a crime by 1995. Black men had higher caregiver-reported externalizing behavior and more frequent risky behavior (fighting and property damage) in early adulthood compared to white men.
Table 1.
Weighted descriptive characteristics, non-Hispanic white and black men age 18–28, 2005–2015 PSID Transition into Adulthood Supplement
| Overall | Non-Hispanic white |
Non-Hispanic Black | ||||
|---|---|---|---|---|---|---|
| Mean | SD | Mean | SD | Mean | SD | |
| Outcomes | ||||||
| Arrested in year t | 0.048 | 0.212 | 0.040 | 0.152 | 0.077 | 0.414 * |
| In jail in year t | 0.015 | 0.121 | 0.013 | 0.087 | 0.024 | 0.238 * |
| Maternal union status change | ||||||
| 0 changes | 0.580 | 0.491 | 0.635 | 0.374 | 0.368 | 0.750 * |
| 1 change | 0.099 | 0.296 | 0.090 | 0.223 | 0.131 | 0.524 * |
| 2 changes | 0.133 | 0.338 | 0.118 | 0.250 | 0.191 | 0.611 * |
| 3+ changes | 0.189 | 0.389 | 0.157 | 0.282 | 0.310 | 0.719 * |
| Young adult characteristics | ||||||
| Non-Hispanic black (vs. NH white) | 0.207 | 0.403 | 0.000 | 1.000 | ||
| Married in year t | 0.041 | 0.197 | 0.048 | 0.166 | 0.014 | 0.186 * |
| Cohabiting in year t | 0.036 | 0.185 | 0.034 | 0.140 | 0.044 | 0.321 * |
| Had first child by year t | 0.112 | 0.313 | 0.072 | 0.201 | 0.265 | 0.686 * |
| Age at year t | 21.190 | 2.391 | 21.162 | 1.863 | 21.299 | 3.775 |
| Family background | ||||||
| Family income-to-needs in 1997 | 3.198 | 2.190 | 3.574 | 1.710 | 1.762 | 2.343 * |
| Mother’s age at birth | 27.604 | 5.674 | 28.037 | 4.338 | 25.946 | 9.112 * |
| Mother’s union status at birth | ||||||
| Married | 0.764 | 0.422 | 0.868 | 0.263 | 0.366 | 0.749 * |
| Never married | 0.196 | 0.395 | 0.099 | 0.232 | 0.568 | 0.770 * |
| Widowed/divorced/separated | 0.033 | 0.177 | 0.025 | 0.122 | 0.060 | 0.369 * |
| Unknown | 0.007 | 0.086 | 0.008 | 0.068 | 0.006 | 0.124 |
| Mother’s union status in late adolescence | ||||||
| Partnered with child’s father | 0.598 | 0.487 | 0.661 | 0.367 | 0.356 | 0.745 * |
| Unpartnered | 0.244 | 0.427 | 0.201 | 0.311 | 0.408 | 0.764 * |
| Partnered with other | 0.158 | 0.363 | 0.138 | 0.268 | 0.236 | 0.660 * |
| Prior behavior | ||||||
| Either parent ever arrested/charged | 0.123 | 0.326 | 0.124 | 0.256 | 0.116 | 0.498 |
| Externalizing behavior (0–17, CDS-I or II) | 5.937 | 3.964 | 5.812 | 3.028 | 6.416 | 6.651 * |
| Risk-taking in early adulthood ((0–2) | 0.206 | 0.441 | 0.178 | 0.308 | 0.316 | 0.895 * |
| Sample information | ||||||
| SRC sample | 0.876 | 0.328 | 1.000 | 0.017 | 0.402 | 0.762 * |
| SEO sample | 0.124 | 0.328 | 0.000 | 0.017 | 0.598 | 0.762 * |
| Observed to age 16/17 | 0.968 | 0.175 | 0.967 | 0.138 | 0.970 | 0.264 |
| Number of indviduals | 1127 | 586 | 541 | |||
| Number of records | 6768 | 3418 | 3350 | |||
Group differences are statistically significant at p<.05.
Figures 1 to 3 provide an assessment of our expectation that men’s contact with the criminal justice system and their own family formation trajectories will differ by maternal union history and race. Figures 1 and 2 summarize the incidence of arrest and incarceration at each age by race and by two categories of maternal union status history: having experienced no changes or having experienced three or more changes. At each age, white young adult men who experienced three or more changes in maternal union status were arrested or incarcerated more often compared to same-race peers who experienced no changes in maternal union status. No such disparity appears for black men. Figure 3 describes the proportion of men who were married at each age for the same groups. Marriage rates are low among all men until about age 23. At age 24, those who experienced no maternal union status change pull away from those who experienced 3 changes or more among white men. Among black men, marriage rates remain low, but those who experienced frequent maternal union status change are somewhat more likely to be married compared to their peers who experienced no union status change.
Figure 1.
Incidence of arrest at each age by race and maternal union status history
Figure 3.
Proportion of men who were married at each age by race and maternal union status history
Figure 2.
Incidence of incarceration at each age by race and maternal union status history
Table 2 presents results from multivariate models predicting arrest and incarceration. For each outcome, the baseline model includes indicators of maternal union status change, race, the interaction of race and maternal union status change, and control variables. The full model adds lagged measures of the young adult’s marital status, cohabitation status, and parenthood status at each age. Table 3 reports estimated probabilities from each model, varying exposure to maternal union status change and race and holding all other variables constant at their means.
Table 2.
Coefficients and robust standard errors from logistic regression models estimating log-odds of arrest and jail, non-Hispanic white and black men age 18–28, 2005–2015. PSID Transition to Adulthood Supplement.
| Arrest | Jail | |||||||
|---|---|---|---|---|---|---|---|---|
| Baseline | Full | Baseline | Full | |||||
| B | SE | B | SE | B | SE | B | SE | |
| Maternal union status change (vs. none) | ||||||||
| 1 change | 0.230 | 0.318 | 0.380 | 0.320 | 0.205 | 0.529 | 0.490 | 0.534 |
| 2 changes | 0.240 | 0.318 | 0.336 | 0.343 | 0.038 | 0.566 | 0.470 | 0.589 |
| 3+ changes | 0.815 | 0.262 ** | 0.858 | 0.290 ** | 0.863 | 0.519 † | 1.041 | 0.550 † |
| Non-Hispanic black (vs. NH white) | 0.500 | 0.261 † | 0.548 | 0.287 † | 0.423 | 0.490 | 0.579 | 0.526 |
| Age at year t | 0.007 | 0.021 | 0.010 | 0.029 | 0.025 | 0.037 | −0.018 | 0.050 |
| Union status change * race | ||||||||
| 1 change * black | −0.102 | 0.378 | −0.318 | 0.391 | −0.184 | 0.665 | −0.578 | 0.684 |
| 2 changes *black | −0.172 | 0.348 | −0.316 | 0.367 | 0.177 | 0.621 | −0.312 | 0.642 |
| 3 changes * black | −0.891 | 0.286 ** | −0.902 | 0.311 ** | −0.857 | 0.572 | −1 .087 | 0.596 † |
| Family background | ||||||||
| Family income-to-needs in 1997 | −0.122 | 0.037 ** | −0.090 | 0.040 * | −0.210 | 0.079 ** | −0.172 | 0.088 † |
| Mother’s age at birth | 0.009 | 0.012 | 0.016 | 0.013 | 0.008 | 0.023 | 0.001 | 0.023 |
| Mother’s union status at birth (vs. married) | ||||||||
| Never married | 0.180 | 0.147 | 0.174 | 0.152 | 0.193 | 0.258 | 0.198 | 0.269 |
| Widowed/divorced/separated | 0.238 | 0.243 | 0.080 | 0.257 | 0.179 | 0.422 | 0.189 | 0.398 |
| Unknown | −0.259 | 0.536 | −0.217 | 0.455 | −0.376 | 1.092 | −0.259 | 0.949 |
| Mother’s union status in late adolescence | ||||||||
| Unpartnered | −0.048 | 0.165 | 0.013 | 0.178 | 0.039 | 0.304 | −0.107 | 0.336 |
| Partnered with other | −0.286 | 0.201 | −0.210 | 0.212 | −0.101 | 0.353 | −0.407 | 0.362 |
| Prior behavior | ||||||||
| Either parent ever arrested/charged | −0.004 | 0.173 | 0.055 | 0.180 | −0.010 | 0.326 | 0.041 | 0.313 |
| Externalizing behavior (0–17, CDS-I or II) | −0.008 | 0.013 | −0.011 | 0.014 | 0.002 | 0.024 | −0.013 | 0.026 |
| Risk-taking in early adulthood ((0–2) | 0.521 | 0.099 *** | 0.436 | 0.107 *** | 0.585 | 0.177 ** | 0.579 | 0.180 ** |
| Sample information | ||||||||
| SEO sample (vs. SRC) | 0.068 | 0.196 | 0.024 | 0.209 | −0.082 | 0.314 | −0.189 | 0.328 |
| Observed to age 16/17 | 0.157 | 0.292 | −0.043 | 0.334 | 0.126 | 0.608 | −0.105 | 0.606 |
| Own family formation | ||||||||
| Married in year t−1 | −2.235 | 1.025 * | −0.464 | 0.758 | ||||
| Cohabiting in year t−1 | 0.187 | 0.275 | −0.497 | 0.614 | ||||
| Had first child by year t−1 | 0.504 | 0.136 *** | 0.789 | 0.230 ** | ||||
| Intercept | −3.526 | 0.677 *** | −3.762 | 0.849 *** | −4.878 | 1.162 *** | −3.647 | 1.411 † |
| Number of records | 6768 | 6754 | ||||||
| Wald chi−2 | 91.720 | 100.900 | 59.91 | 81.210 | ||||
| psuedo log-likelihood | −1435.920 | −1175.880 | −598.890 | −505.090 | ||||
p<.01
p<.05
p<.10
Table 3.
Estimated probability of arrest or jail by maternal union status history and race, baseline and full models
| Arrest | Jail | |||
|---|---|---|---|---|
| Baseline | Full | Baseline | Full | |
| Non-Hispanic white, no change | 0.038 | 0.035 | 0.011 | 0.010 |
| Non-Hispanic black, no change | 0.061 | 0.059 | 0.017 | 0.018 |
| Non-Hispanic white, 1 change | 0.047 | 0.050 | 0.014 | 0.016 |
| Non-Hispanic black, 1 change | 0.069 | 0.062 | 0.018 | 0.017 |
| Non-Hispanic white, 2 changes | 0.048 | 0.048 | 0.012 | 0.016 |
| Non-Hispanic black, 2 changes | 0.065 | 0.060 | 0.021 | 0.021 |
| Non-Hispanic white, 3 changes | 0.082 | 0.078 | 0.027 | 0.028 |
| Non-Hispanic black, 3 changes | 0.057 | 0.056 | 0.017 | 0.017 |
Baseline models indicate that white young adults who experienced three or more maternal union status changes are significantly more likely to be arrested (p<.01) and marginally more likely to be incarcerated (p<.10) in a given year compared to white peers who experienced no maternal union status change net of covariates. Interaction terms indicate that while black men have an elevated likelihood of being arrested or incarcerated compared to white men overall, the probability of these events for black men is unrelated to their history of maternal union status change. This pattern of findings is consistent with prior research on the association of family structure instability with white youths’ behavioral trajectories and with research suggesting that this association is largely absent for black youth.
The full models describe the association between frequent maternal union status change and each outcome when a young adult’s own family formation events are taken into account. The main effect of frequent family change in childhood remains strongly significantly associated with the log-odds of arrest and marginally associated with the log-odds of incarceration. The interaction term associated with frequent maternal union status change for black young adults is marginally significant. Having been married in the prior year is associated with lower log-odds of arrest, while having a child is associated with higher log-odds of both arrest and incarceration. Living in a cohabiting union in the prior year is not significantly associated with either outcome.
Tables 3 and 4 provide two practical interpretations of the regression results summarized in Table 2. Table 3 presents estimated probabilities of arrest and jail for white and black young adults at each level of maternal union status change in childhood. White young adults who experienced no maternal union status change had predicted probabilities of arrest and jail in each year of .038 and .011 respectively in the baseline model. These probabilities were reduced trivially in the full model. The probability of arrest increased by about one-third for white youth who experienced one or two maternal union status changes compared to none in both the baseline and full models, but for those who experienced 3 or more changes, the probabilities of arrest (.082) and jail (.027) were more than twice as high in the baseline model, suggesting that the association of family change with criminal offending is better characterized by a threshold model than by a linear model. The probability of arrest for this group was reduced to .078 in the full model, but the probability of jail was not significantly affected. Black young adults had higher predicted rates of arrest and jail compared to whites except among white young adults who experienced three changes or more in maternal union status. Overall, changes in maternal union status did not substantially alter the predicted probabilities of arrest and incarceration for black men.
Table 4.
KHB decomposition of total effects of 3+ maternal union status changes for white young adult men into indirect and direct effects as mediated by family formation behavior in early adulthood
| Marriage | Cohabitation | Birth | |
|---|---|---|---|
| Arrest | |||
| Baseline | 0.884 * | 0.777 * | 0.788 * |
| Full | 0.779 * | 0.779 * | 0.779 * |
| Difference | 0.104 * | −0.002 | 0.009 |
| Percent change | −12 % | 0% | −1 % |
| Jail | |||
| Baseline | 0.675 | 0.613 | 0.621 |
| Full | 0.604 | 0.604 | 0.604 |
| Difference | 0.070 * | 0.008 | 0.017 |
| Percent change | −10 % | −1 % | −3 % |
Table 4 provides a decomposition of the total effect of three or more changes in maternal union status on the likelihood of arrest or jail for white youth into an indirect effect, or the portion that is attributable to mediators, and a direct effect, or the portion that remains unexplained by mediation. The underlying regression models are weighted but not clustered on the unique identifier as are the models summarized in Table 2. We assess the mediating effect of the lagged indicators for young adults’ marital, cohabiting, and parent status in separate decomposition models that control for the other family formation events and all covariates included in the full model. Men who were married in the preceding year were less likely to be arrested or in jail in a given year compared to men who were not married, and this association significantly mediated the association between frequent maternal union status change and each outcome. Relating this to our conceptual model, white men who experienced three or more changes in maternal union status were less likely to marry in early adulthood, and this disparity compared to other men partially explains their higher probability of arrest and time in jail. Cohabitation status and parenthood did not mediate the association between maternal union status change and either outcome.
Discussion
Nearly one in five contemporary young adults experiences three or more changes in maternal union status in the course of childhood. A substantial literature has documented that exposure to frequent union instability is associated with increased behavior problems and delinquency across the course of childhood and adolescence, but to date, little research has investigated whether this association extends to criminal offending in early adulthood. On the one hand, this relationship may not persist: most young offenders stop engaging in delinquent and criminal behavior when they enter adulthood, so any association between family structure and behavior in childhood may not be consequential for behavior in adulthood. Alternatively, family instability may be associated with persistent offending through a number of potential pathways, including through selection mechanisms, such that parents who are likely to have criminally offended themselves are more likely both to experience union instability and to raise children who become criminal offenders; through a process of cumulative disadvantage by which the experience of each additional change in family structure compounds children’s risk of delinquent or criminal behavior over the life course; or through a more indirect process whereby children who experience frequent family change are less likely to enter stable unions themselves and thus remain more likely to criminally offend compared to peers who enter early marriage.
Our analysis considered these competing alternatives to explain patterns of contact with the criminal justice system for contemporary white and black young adult men who experienced varying levels of maternal union status instability in childhood. We found that for white men, but not for black men, the experience of three or more changes in maternal union status is positively associated with the likelihood of arrest and incarceration through age 26. The probability of arrest in a given year was .082 and the probability of incarceration was .027 for this group in the baseline model, more than double that for young white men who experienced no family structure change. This relationship holds net of parents’ own criminal offending history and indicators of early child behavior problems, suggesting that the association is not attributable to selection mechanisms or to problem behavior established early in life. White men’s lower odds of marriage explain about 12 percent of the association, indicating that some portion of the relationship is indirect and attributable to men’s generally more precarious transitions to adulthood when they experience frequent family structure change. Yet much of the direct association remains after taking these mechanisms into account.
To some extent, the relatively small attenuating effect of marriage on exposure to maternal union instability is to be expected. Disparities in marriage probabilities between those who experienced no family change themselves and those who experienced three or more changes did not begin to emerge until men were in their mid-20s, so would not explain differences in contact with the criminal justice system at earlier ages. But even the relatively modest influence of marriage is striking. We note that while prior work has argued that marriage has a causal role in desistance from crime, we cannot entirely eliminate self-selection into marriage as a competing explanation. For the birth cohort considered, observed marriages occurred relatively early: during the period of observation (2005–15), men’s median age at first marriage increased from 27 to 29 years (U.S. Census Bureau 2017), and prior work has demonstrated that family instability in childhood is associated with lower rates of marriage in early adulthood overall (Fomby and Bosick 2013). Thus, the group of young men who experienced frequent maternal union status change and who subsequently married at a relatively young age may be self-selected either on traits that also make them less likely to engage in criminal behavior or more likely to be responsive to marriage as a social control mechanism compared to their peers.
Notably, other family formation events including cohabitation and the birth of a first child did not diminish the likelihood of arrest or incarceration. In fact, the birth of a first child positively predicted both arrest and incarceration. We do not interpret this as a causal relationship; rather, we expect that early fatherhood, and what is most often nonmarital fatherhood in this population, is indicative of prior disadvantage or diminished educational and employment opportunities in early adulthood that may in turn contribute to more frequent offending. Accounting for current employment did not weaken the association between having a child and the outcome variables, but we will investigate other factors such as educational attainment and current school enrollment and nonresidence with children that may refine or explain this relationship.
Conclusion
The majority (58%) of PSID respondents experienced stable family arrangements from birth through adolescence. Those who experienced highly unstable arrangements characterized by three or more changes in maternal union status were more likely to be black. Most black children experienced at least one family structure change, with a third (31%) experiencing three or more.
Prior literature suggests these unstable family backgrounds would have negative implications for adolescent risk behavior and transition-to-adulthood experiences (e.g. Fomby and Bosick 2013). The findings presented in this article suggest that these unstable family contexts shape criminal outcomes even in adulthood. Our findings predict that whites who experience a single change in maternal union status during childhood are significantly more likely to be arrested in adulthood. Those who experience high levels of family instability (three or more transitions) are not only more likely to be arrested, but are also significantly more likely to experience incarceration in adulthood. These findings point to a need for criminologists to move beyond the classic, yet static focus on “broken homes” toward measures of family instability, which more accurately capture the dynamic nature of modern family contexts.
Despite black children’s disproportionate exposure to family structure transitions, these changes are not significantly related to their interactions with the criminal justice system in adulthood. Black men who grow up in unstable families are not more likely to experience arrest or incarceration than black men from stable family structures. This is not to suggest family instability does not impact blacks in the long term. Like white men, their experiences with family structure appear to shape their own family formation during their transitions to adulthood. Those who experience a single family structure transition, or two family structure changes are significantly less likely to marry or cohabitate, respectively, than those from stable family structures. Further, the greater number of family structure transitions they experienced in childhood, the more likely it is that they will become fathers in early adulthood.
Consistent with the life course criminology literature, marriage during the transition to adulthood surfaces as particularly consequential for adult criminal outcomes. Indeed, our findings suggest marriage attenuates the relationship between early family structure instability and adulthood arrest and incarceration. Put differently, men who marry are less likely be arrested or incarcerated in adulthood, but selection into marriage is not random. Childhood experiences with family instability, among other factors, shape those experiences. Thus we observe an accumulation of disadvantage wherein children from unstable family contexts are more likely to become involved with the criminal justice, and less likely to experience life events like marriage that might otherwise curtail that involvement.
This analysis contributes to research in life course criminology by drawing upon long-running, multigenerational data that includes prospective information on mothers’ and children’s experience. However, the study does have some limitations. First, we cannot infer a causal relationship between either exposure to maternal union status change or one’s own family formation behavior and subsequent probabilities of offending. We have controlled for factors that are theoretically related to each of the key variables of interest, but there is always the possibility that observed associations are driven by unobserved factors. In supplemental analyses, we tested other plausibly related covariates such as parental education, parent and young adult religiosity, and young adults’ delinquent behavior in adolescence, and young adult employment but did not find that these substantially improved model fit or mediated the association between family instability and criminal offending.
Nevertheless, this paper encourages a fuller focus on the role of families in shaping criminal behavior over the life course. Rather than analyzing early childhood family experiences as separate from one’s own family formation, we advocate taking a long-view of family influence. Doing so allows us to understand the multiple avenues by which intergenerational forces shape criminal behavior.
Footnotes
Children born or adopted into PSID families become respondents to the biennial Core PSID interview themselves when they establish economic independence in a household separate from their parents’. By 2015, the majority of young adults from the original CDS sample who were 28 years or older had become Core PSID respondents themselves. As a result of that transition and to contain study costs, young adults who were age 28 or older in 2015 were not interviewed in TAS in that year.
As of 2015, the birth history is collected only from PSID household members age 15 and older.
Other items such as illegal substance use, driving under the influence, and thrill-seeking, were unrelated either to maternal union instability or the outcome measures and were excluded from the young adult behavior index.
Contributor Information
Stacey Bosick, University of Colorado Denver.
Paula Fomby, University of Michigan.
References
- Aughinbaugh Alison, Robles Omar and Sun Hugette. 2013. ”Marriage and Divorce: Patterns by Gender, Race, and Educational Attainment.” Monthly Labor Review (October). doi: 10.21916/mlr.2013.32. [DOI] [Google Scholar]
- Blokland Arjan AJ and Nieuwbeerta Paul. 2005. “The Effects of Life Circumstances on Longitudinal Trajectories of Offending.” Criminology 43(4):1203–40. [Google Scholar]
- Bowlby John. 1951. Maternal Care and Mental Health, Vol. 2: World Health Organization; Geneva. [PMC free article] [PubMed] [Google Scholar]
- Breen Richard, Kristian Bernt Karlson and Anders Holm. 2013. “Total, Direct, and Indirect Effects in Logit and Probit Models.” Sociological Methods & Research 42:164–91. [Google Scholar]
- Doherty Elaine Eggleston and Cwick Jaclyn M.. 2016. “Unpacking the Complexity of Life Events and Desistance: An Application of Conjunctive Analysis of Case Configurations to Developmental and Life Course Criminology.” Journal of Developmental and Life-Course Criminology 2(1):45–63. doi: 10.1007/s40865-015-0023-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Farrington David P. 2011. “Families and Crime.” Crime and public policy:130–57. [Google Scholar]
- Fomby P and Cherlin AJ. 2007. “Family Instability and Child Well-Being.” American Sociological Review 72(2):181–204. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fomby Paula, Mollborn Stefanie and Christie A Sennott. 2010. “Race/Ethnic Differences in Effects of Family Instability on Adolescents’ Risk Behavior.” Journal of Marriage and Family 72(2):234–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fomby Paula. 2013. “Family Instability and College Enrollment and Completion.” Population Research and Policy Review 32(4):469–94. doi: 10.1007/s11113-013-9284-7. [DOI] [Google Scholar]
- Fomby Paula and Bosick Stacey J.. 2013. “Family Instability and the Transition to Adulthood.” Journal of Marriage and Family 75(5):1266–87. doi: 10.1111/jomf.12063. [DOI] [Google Scholar]
- Fomby Paula and Mollborn Stefanie. 2017. “Ecological Instability and Children’s Classroom Behavior in Kindergarten.” Demography 54(5):1627–51. doi: 10.1007/s13524-017-0602-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Forrest Walter. 2014. “Cohabitation, Relationship Quality, and Desistance from Crime.” Journal of Marriage and Family 76(3):539–56. [Google Scholar]
- Hofferth SL and Goldscheider F. 2010. “Family Structure and the Transition to Early Parenthood.” Demography 47(2):415–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Horney Julie, Osgood D. Wayne and Ineke Haen Marshall. 1995. “Criminal Careers in the Short-Term: Intra-Individual Variability in Crime and Its Relation to Local Life Circumstances.” American sociological review 60(5):655–73. [Google Scholar]
- Laub John H. and Sampson Robert J.. 2006. Shared Beginnings, Divergent Lives. Cambridge, Massachusetts and London, England: Harvard University Press. [Google Scholar]
- Lee D and McLanahan S. 2015. “Family Structure Transitions and Child Development: Instability, Selection, and Population Heterogeneity.” American Sociological Review 80(4):738–63. doi: 10.1177/0003122415592129. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Macmillan Ross and Eliason Scott R.. 2003. “Characterizing the Life Course as Role Configurations and Pathways” Pp. 529–54 in Handbook of the Life Course, edited by Mortimer JT and Shanahan MJ. New York: Kluwer Academic/Plenum Publishers. [Google Scholar]
- Macmillan Ross and Hagan John. 2004. “Violence in the Transition to Adulthood: Adolescent Victimization, Education, and Socioeconomic Attainment in Later Life.” Journal of Research on Adolescence 14(2):127–58. [Google Scholar]
- Massoglia Michael and Christopher Uggen 2010. “Settling Down and Aging Out: Toward an Interactionist Theory of Desistance and the Transition to Adulthood.” American Journal of Sociology 116(2):543–82. doi: 10.1086/653835. [DOI] [PMC free article] [PubMed] [Google Scholar]
- McGonagle Katherine A., Schoeni Robert F., Sastry Narayan and Freedman Vicki A.. 2012. “The Panel Study of Income Dynamics: Overview, Recent Innovations, and Potential for Life Course Research.” Longitudinal and life course studies 3(2):188. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Osgood, D. Wayne, Gretchen Ruth, Jacquelynne S. Eccles, Janis E. Jacobs and Bonnie L. Barber 2005. “Six Paths to Adulthood: Fast Starters, Parents without Careers, Educated Partners, Educated Singles, Working Singles, and Slow Starters” Pp. 320–55 in On the Frontier of Adulthood, edited by Settersten RA Jr., Furstenberg FF Jr. and Rumbaut RG. Chicago & London: The University of Chicago Press. [Google Scholar]
- Peterson James L. and Zill Nicholas. 1986. “Marital Disruption, Parent-Child Relationships, and Behavior Problems in Children.” Journal of Marriage and Family 48:295–307. [Google Scholar]
- Raley R. Kelly, Megan M. Sweeney and Danielle Wondra 2015. “The Growing Racial and Ethnic Divide in U.S. Marriage Patterns.” The Future of children / Center for the Future of Children, the David and Lucile Packard Foundation 25(2):89–109. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rebellon Cesar J. 2002. “Reconsidering the Broken Homes/Delinquency Relationship and Exploring Its Mediating Mechanism (S).” Criminology 40(1):103–36. [Google Scholar]
- Sampson Robert J and John H Laub. 1997. “A Life-Course Theory of Cumulative Disadvantage and the Stability of Delinquency.” Developmental theories of crime and delinquency 7:133–61. [Google Scholar]
- Sampson Robert J, John H Lauband Christopher Wimer. 2006. “Does Marriage Reduce Crime? A Counterfactual Approach to within‐Individual Causal Effects.” Criminology 44(3):465–508. [Google Scholar]
- Sampson Robert J. and Laub John H.. 1995. Crime in the Making: Pathways and Turning Points through Life. Cambridge: Harvard University Press. [Google Scholar]
- Census Bureau US. 2017. “Table Ms-2. Estimated Median Age at First Marriage, by Sex: 1890 to the Present.” edited by ms2.xls. Washington, DC: Department of Commerce. [Google Scholar]
- Uggen Christopher and Janikula Jennifer. 1999. “Volunteerism and Arrest in the Transition to Adulthood.” Social forces 78(1):331–62. [Google Scholar]
- Uggen Christopher. 2000. “Work as a Turning Point in the Life Course of Criminals: A Duration Model of Age, Employment, and Recidivism.” American sociological review 65(4):529–46. [Google Scholar]
- Wells L Edwardand Joseph H Rankin. 1991. “Families and Delinquency: A Meta-Analysis of the Impact of Broken Homes.” Social problems 38(1):71–93. [Google Scholar]
- Wu LL and Martinson BC. 1993. “Family-Structure and the Risk of a Premarital Birth.” American Sociological Review 58(2):210–32. doi: 10.2307/2095967. [DOI] [Google Scholar]