Unmarried parents have less stable unions than married parents, but there is considerable debate over the sources of this instability. Unmarried parents may be more likely than married parents to end their unions because of compositional differences, such as more disadvantaged personal and relationship characteristics, or because they lack the normative and institutional supports of marriage, thus rendering their relationships more sensitive to disadvantage. In this article, we evaluate these two sources of union instability among married, cohabiting, and dating parents following the birth of a shared child, using five waves of longitudinal data from the Fragile Families and Child Wellbeing Study. Using discrete-time event history models, we find that demographic, economic, and relationship differences explain more than two-thirds of the increased risk of dissolution for unmarried parents relative to married parents. We also find that differential responses to economic or relationship disadvantage do not explain why unmarried parents are more likely to end their unions than married parents.
In 1950, nonmarital births accounted for less than 5 % of all births in the United States, but by 2009, more than 40 % of births were to unmarried mothers (Martin et al. 2011; Ventura 2009; Ventura and Bachrach 2000). Despite this striking change in the family context of childbearing, the vast majority of unmarried mothers continue to have their children within romantic relationships. About one-half of nonmarital children are born to cohabiting couples, and another 30 % start life with parents who are not living together but are romantically involved (Kennedy and Bumpass 2008). Thus, only one in five children born outside of marriage in the early twenty-first century were truly born to a “single” parent outside of a romantic relationship.
Are changes in the American family therefore more apparent than real? The answer depends on the degree to which unmarried couples who share children behave like married couple families over time—whether they stay together or remain active in their children’s lives. On this score, nonmarital unions bear little resemblance to marriages. Although U.S. divorce rates are high relative to other industrialized countries, nearly two-thirds (65 %) of all children born within marriage still live with both parents by age 15, while only about one in five (22 %) of those born to cohabiting couples do so (Andersson 2002). For children whose parents were dating but not living together when they were born, chances that the parents will remain together are even lower (Center for Research on Child Wellbeing 2007).1 Furthermore, fathers’ contact with their children drops off precipitously after unmarried parents break up (Tach et al. 2010)—so much so that by age 15, as few as one in five children born to unmarried parents report any meaningful contact with their fathers (Argys and Peters 2001).
Nonmarital births are far more common among those Americans already disadvantaged by virtue of their race, ethnicity, or socioeconomic status (SES). McLanahan and Percheski (2008) argued that nonmarital childbearing is a major force perpetuating U.S. inequality because of the negative consequences of union instability. In their review of the literature on unmarried parents—whom they call “fragile families”—they showed that break-ups, which are often followed by new relationships and subsequent children, increase maternal stress and mental health problems, reduce the quality of mothers’ and fathers’ parenting, reduce paternal investments, and ultimately produce negative outcomes for children.
What explains the fragility of unions among unmarried couples with children relative to married couples? Surprisingly little direct research is available on the causes of dissolution among unmarried parents; most prior work has focused on the determinants of dissolution for married couples or childless cohabiters. There are two potential reasons why dissolution rates may be higher for unmarried parents than they are for married parents. First, differences in the composition of the population of unmarried parents—such as greater economic and relationship disadvantage—may explain their greater relationship instability. Second, because of differences in the institutional supports and normative expectations for marital and nonmarital unions, the predictors of dissolution may differ for married and unmarried parents. If marriage is protective, marital unions may have a lower risk of dissolution in the face of poor economic performance and relationship troubles than do comparable nonmarital relationships.
In this article, we compare sources of union instability between married and unmarried parents using five waves of longitudinal survey data on roughly 4,000 unmarried couples and 1,000 married couples in the Fragile Families and Child Wellbeing Study who had a birth between 1998 and 2000. We use discrete-time event history models to estimate the risk of dissolution for married and unmarried parents and calculate how much of the increased risk of dissolution among unmarried parents relative to married parents can be explained by differences in the composition of the marital and nonmarital populations. We then test whether the predictors of dissolution differ for married and unmarried parents.
Previous Research on Union Dissolution
Nonmarital unions in the United States are typically short lived (Bumpass and Lu 2000; Bumpass and Sweet 1989; Lichter et al. 2006). Using data from the National Survey of Family Growth (NSFG), Kennedy and Bumpass (2008) found that more than one-half (56 %) of all cohabiting unions ended within the first two years, and just 14 % remained in nonmarital cohabiting relationships after five years. Policymakers and academics interested in child well-being and the reproduction of inequality are primarily concerned about the stability of the subset of nonmarital unions that involve children. Although the unions of cohabiters with and without children have fewer normative and institutional supports than marital unions, cohabiters with children face a different set of considerations and constraints than childless cohabiters when choosing to end their unions because they must also consider the welfare of their children.
Little representative longitudinal information about nonmarital unions involving children was available until the introduction of the Fragile Families and Child Wellbeing Study in the late 1990s. Based on a representative sample of a cohort of children born from 1998 to 2000 in medium to large U.S. cities, this study has found that unmarried parents were often romantically involved and were optimistic about the future of their unions at the time of their shared child’s birth (Center for Research on Child Wellbeing 2002).2 Yet, by the time their children were about 5 years old, a majority of these relationships had ended. Fewer than 20 % of the parents had married, whereas more than two-thirds had ended their relationships. In contrast, fewer than one in five of the unions among married parents had ended.
Determinants of Divorce
The determinants of divorce have changed little over the last half-century (Teachman 2002). A large literature has consistently found that young age at marriage, low education and income, less religiosity, premarital conception and birth, premarital cohabitation, and prior marriages are associated with an increased risk of divorce (Furstenberg 1990; Raley and Bumpass 2003; Sweeny and Phillips 2004; Teachman 2004; Teachman et al. 2000; White 1990). Historically, African Americans were more likely to divorce than whites, although these differences have diminished somewhat over time (Teachman et al. 2000). Married couples of lower economic standing are more likely to divorce, regardless of whether economic standing is measured by education, income, or earnings (Martin 2006; McLanahan 2004). Evidence on the effect of women’s earnings and employment on martial stability is mixed (Brines and Joyner 1999).
Relationship quality also figures strongly in the literature on the determinants of divorce. More positive reports of relationship quality along a variety of indicators are related to a lower risk of divorce or separation (Booth and Amato 1994; Gager and Sanchez 2003; Gottman and Levenson 1992; White and Booth 1991). Amato and Rogers (1997) used marital problems reported in 1980 to predict divorces over the next two years. They showed that marital difficulties were a significant independent predictor of divorce, and that marital problems mediated some of the effect of more distal variables, such as age, church attendance, and the experience of parental divorce. Gottman (1994) also found that negative couple communication patterns were an important predictor of divorce. Prospective narratives from individuals who eventually divorce commonly feature relationship troubles, such as extramarital sex, communication difficulties, incompatibility, not spending enough time at home, and disagreements over money (Bloom et al. 1985; Burns 1984; Cleek and Pearson 1985; Kitson and Sussman 1982; Spanier and Thompson 1987). Childbearing is associated with declines in marital quality but increases marital stability in the initial years after the child’s birth (Cherlin 1978; Waite and Lillard 1991).
Dissolution Among Unmarried Couples with Children
Much of what we know about the determinants of dissolution among unmarried couples with children comes from recent in-depth qualitative work that follows unmarried parents—both mother and father—from the time of the birth onward. These studies have found that although economic factors figure strongly in couples’ accounts of why they are hesitant to marry each other, narratives of relationship dissolution seldom feature struggles over money.3 Instead, those whose relationships fail nearly always point to relationship quality and serious behavioral problems of one or both partners as the primary cause (Gibson-Davis 2007; Gibson-Davis et al. 2005; Hill 2007; Reed 2007; but see Waller 2008).4 Qualitative studies, however, cannot rule out the possibility that relationship troubles and personal problems are driven in part by less-proximate economic problems.
A growing body of research has identified the importance of relationship and economic factors for the transition to marriage among unmarried parents (Gibson-Davis 2007, 2009; Gibson-Davis et al. 2005; Sassler 2004; Smock and Manning 1997; Smock et al. 2005; Watson and McLanahan 2011), but only a few nationally representative studies have examined the determinants of dissolution among unmarried parents. These studies produce mixed evidence about the importance of economic and relationship characteristics for the stability of nonmarital unions. In a pooled analysis of married and cohabiting parents, Osborne et al. (2007) found that differences in union duration, demographic characteristics, education, and economic resources explained about one-third of the increased risk of dissolution for cohabiters relative to married parents, but relationship quality and family complexity explained little of cohabiters’ increased risk of dissolution. Carlson et al. (2004) found that positive economic factors encouraged marriage among unmarried parents one year after a child’s birth, but positive relationship factors—such as a lack of gender mistrust and relationship conflict, as well as perceived partner supportiveness—preserved the stability of both marital and nonmarital unions one year after the child’s birth. Finally, in an analysis of all cohabiters (both parents and nonparents), Lichter et al. (2006) found that women’s and partners’ employment predicted marriage but not dissolution.
Institutional and Normative Supports
In marriage, the legal contract, the institutionalized expectation of sustained (if not lifelong) commitment, and the costs of marital dissolution provide safeguards against relationship dissolution (Teachman et al. 1991). In contrast, unmarried couples have fewer legal supports, have weaker and sometimes conflicting normative expectations for sustained commitment, and face fewer stigma costs to ending their unions. Indeed, research on cohabiters (which historically did not distinguish between those with and without children) has found that they hold more individualistic views of marriage and report less dedication to their partners (Axinn and Thornton 1992; Cohan and Kleinbaum 2002; Rhoades et al. 2009; Thomson and Colella 1992).
It is difficult to observe the different institutional and normative forces operating on marital and nonmarital unions, particularly in survey data, but they may be inferred through differences in the strength of predictors of dissolution. If marriage is protective, then the associations between measures of relationship risk—such as low SES or poor relationship quality—and subsequent dissolution should be weaker for married parents than for similar unmarried parents. If this is the case, such differences in responses to risk may help explain why unmarried parents have less stable unions than married parents.
In sum, nonmarital unions involving children are considerably less stable than marital unions, yet previous studies have not determined whether this relative instability is due to differences in the compositional characteristics of marital and nonmarital parents or whether the determinants of relationship dissolution differ for married and unmarried parents. In the analyses that follow, we use five waves of longitudinal data from the Fragile Families and Child Wellbeing Study to estimate how much of the disparity in dissolution between married and unmarried parents is due to their different personal and relationship characteristics. We then examine whether the economic, behavioral, and relationship predictors of dissolution differ for married and unmarried parents. If marriage is protective, we might expect these predictors to be significantly weaker for married parents than for unmarried parents.
Data and Measures
The Fragile Families and Child Wellbeing Study follows a cohort of nearly 5,000 children born in 20 U.S. cities between 1998 and 2000. The study interviewed mothers and fathers at the time of the child’s birth and again after about one year, three years, five years, and nine years. The survey contains a large oversample of nonmarital births and, when weighted, the data are representative of births in U.S. cities with populations larger than 200,000. In an advance over many previous household surveys used to study family life, both the mother and father were interviewed at each follow-up, regardless of their relationship status.
For each survey wave, our analyses are based on a subsample of the 4,898 children in the Fragile Families Study whose biological parents were in a romantic relationship at the time of their birth (N = 4,244), whether marriage (N = 1,187), cohabitation (N = 1,783), or nonresident dating (N = 1,274). For subsequent waves, we restrict the sample to children whose mother responded to the survey and answered the question about her relationship status with the father.5 Our analyses pool observations across survey waves, and a couple contribute observations to the sample at each wave until the mother reports that they are no longer in a romantic relationship; after that survey wave, they exit the sample. This yields a sample size of 3,740 at one year; 2,595 at three years; 1,936 at five years; and 1,510 at nine years; for a total pooled sample of 9,781 couple-wave observations.
The main dependent variable in our study is the mother’s report of her relationship status with the child’s father at each follow-up survey wave collected around the child’s first, third, fifth, and ninth birthdays. We measured whether the biological parents are married, cohabiting (living together all or most of the time), dating (romantically involved but not living together all or most of the time), or in no romantic relationship. In our analysis of union dissolution, we coded whether the couple is in any romantic relationship (married, cohabiting, or dating) versus in no romantic relationship. We used mothers’ reports of relationship status at each survey wave because fathers have higher rates of attrition that are systematically related to relationship status.6
Our independent variables consist of time-constant measures taken from the baseline survey and time-varying measures taken from the survey wave prior to when relationship outcomes are assessed so that they occur temporally prior to any change in relationship status. Unless noted otherwise, we took fathers’ measures from fathers’ surveys and mothers’ measures from mothers’ surveys. Our measures of economic characteristics include each parent’s education, employment, and earnings. Mother’s and father’s education are measured at the time of the birth and were coded as a series of dummy variables for less than high school, high school diploma or GED, some college, and college or higher. We used the mother’s report of the father’s education if his report was missing. Father’s employment was measured as a dummy variable indicator for whether he reported working for pay in the week prior to the survey; we used the mother’s report of the father’s employment if his report was missing. We measured mother’s employment as a dummy variable indicator for whether she worked for pay for at least one month in the previous year, allowing for temporary exits from the labor force related to pregnancy and maternity leave. Mother’s and father’s annual earnings were derived from their reports of wages and weeks worked in the past year (in Waves 1–3) or total annual earnings (in Wave 4); after examining the goodness of fit of several functional forms, we used the natural log of annual earnings in our models. Finally, we included dummy variable indicators for whether the mother received Temporary Assistance to Needy Families (TANF) in the previous year and whether the mother currently receives a housing assistance subsidy.
Behavioral and Health Characteristics
We measured several behavioral and health characteristics of each parent. Father ever in jail or prison is a dummy variable coded as 1 at each survey wave if either the mother or father reported that he had ever been in jail or prison.7 We also included a dummy variable indicator for fathers that equals 1 if the father reported that his drinking or drug use interfered with work or relationships in the past year. We used the mother’s report of the father’s incarceration and drug use if his report was missing. We included a parallel dummy variable for the mother’s own drug use. We measured self-reported health for both the mother and father on a five-point scale: poor (1), fair, good, very good, or excellent (5). Finally, we included a measure of mother’s reported religious affiliation at the baseline survey—Catholic, other (non-Catholic) Christian, other (non-Christian) religion, or no religion—and a time-varying measure of how often mothers typically attend religious services on a scale ranging from never (1) to once per week or more (5).
Relationship quality was measured as a four-item scale at each survey wave. On a scale from never (1) to often (3), the mother was asked how often the father is fair and willing to compromise, expresses love and affection, insults or criticizes her (reverse coded), and encourages her to do things that are important to her. A standardized scale of these items was created, with higher values indicating higher-quality relationships. The reliability of this scale was .60 at the baseline survey. We also created a parallel scale for fathers based on their answers to the same set of questions about their relationships with the mothers.
A mother was coded as having experienced domestic violence at each survey wave if she indicated that the child’s father had hit or slapped her in the past year. Unfortunately, no question in each wave of the Fragile Families survey asked mothers whether their partners have ever been unfaithful. We do, however, have an indicator of mother’s distrust of men at the baseline interview, which was measured by the average to two items: “Men cannot be trusted to be faithful,” and “In a dating relationship, a man is largely out to take advantage of a woman.” The response options for these two questions range from 1 (strongly disagree) to 4 (strongly agree).
Mother’s pro-marriage attitudes were measured by the average response to two statements about the importance of marriage asked during the first survey wave: “It is better for a couple to get married than to just live together,” and “It is better for children if their parents are married.” Mother’s traditional attitudes were measured by the average response to two statements: “The important decisions in the family should be made by the man of the house,” and “It is much better for everyone if the man earns the main living and the woman takes care of the home and the family.” A parallel measure of father’s traditional attitudes was also created using the father’s responses to the same statements during the first survey wave. The response options for all these items range from 1 (strongly disagree) to 4 (strongly agree). These measures are comparable with those used in Carlson et al. (2004).
We included previous fertility measures for whether the couple had prior shared children before the birth of the focal child, whether the mother had a prior child with a different partner, and whether the father had a prior child with a different partner, relative to the focal child being the first child for both partners. The previous fertility measures were derived from the mother’s and father’s reports in the one-year survey of whether they had children with someone other than the focal child’s parent. We also included a measure of relationship duration in years measured at the time of the focal child’s birth. This measure was created based on the mother’s response to the question, “How long have you known the baby’s father?”8
We included a time-constant measure of relationship commitment measured at the baseline survey, including mother’s reports of whether the father provided financial support during the pregnancy, whether the father visited the mother in the hospital, and whether the child was given the father’s last name. We also included a measure of whether the mother reported that the baby’s father suggested that she get an abortion.
Because we were concerned that some relationship quality and commitment measures may be highly correlated with one another, we examined the inter-item correlations and performed a factor analysis of these measures (Table 6 in the appendix). The factor analysis identified that, with the exception of the abortion measure, the relationship commitment measures loaded on the same factor; thus, we summed them into a single measure of relationship commitment, ranging from 0 (father did not provide financial support or visit mother, and child did not take father’s last name) to 3 (all three occurred). The factor analysis also revealed that mother’s and father’s reports of relationship quality loaded onto the same factor, so we combined these two measures into a single measure of relationship quality.
We measured mother’s race as white, black, or other race; we measured mother’s ethnicity as Hispanic or non-Hispanic. The mother was coded as foreign-born if she reported being born outside the United States. Mother’s and father’s ages were measured at the time the child was born. We also included a dummy variable indicator for whether the child is male, and a dummy variable for childhood family structure indicating whether the mother lived in a two-parent family when she was 16. We experimented with including the measures of both father’s and mother’s background characteristics in our models, but mother’s and father’s measures are highly correlated, so we included only mother’s background measures in our regression analyses. Results were substantively similar if father’s background characteristics or the average of mother’s and father’s characteristics were included instead. All background measures were taken from the Wave 1 baseline survey administered shortly after the child was born.
Item nonresponse for our independent variables was generally low—in most cases, at less than 5 %. The items for which nonresponse was higher include whether the father was employed (8 %) and father’s earnings (20 %). Multiple imputation by chained equations was conducted using Stata’s mi commands for missing values on all independent variables (Royston 2004). The imputation models included variables reported by mothers and fathers that are associated with either the dependent variable of interest, relationship dissolution, or the likelihood of having missing data (Allison 2002).9 Results presented are based on results from 20 imputed data sets; analyses using single imputation or listwise deletion produced similar results.
In subsequent models, we add vectors of control variables to Eq. (1). The full model includes m time-constant control variables measured at the baseline survey (unless otherwise noted previously) and n time-varying predictors measured at t – 1, the wave prior to the potential dissolution. We enter the covariates sequentially, starting with the most exogenous. Following Carlson et al. (2004) and Osborne et al. (2007), we add demographic controls, followed by economic characteristics, then behavioral characteristics, and finally relationship characteristics. We use the full model with all controls to determine how much compositional differences among married, cohabiting, and dating parents explain of the baseline disparities in the odds of dissolution. We also use this model to estimate how much less stable married parents’ unions are if they have the characteristics of the average cohabiting or dating couple.
Our second research question asks whether the predictors of dissolution differ for married and unmarried parents. To test this, we estimate a fully interacted model in which each variable in Eq. (1) is interacted with a dummy variable for nonmarital birth. We pool cohabititng and dating parents into a single nonmarital category because supplemental analyses indicated no differrences between those groups. The main coefficients for each variable in this specification thus represent the predicted association with dissolution for married couples, and the interaction terms indicate whether that association differs significantly for unmarried couples. This is equivalent to running two separate models—one for married couples and one for unmarried couples—which we also report because the separate models indicate whether the coefficients for unmarried parents are significantly different from zero. We thus obtain information on the significant predictors of dissolution for married and unmarried parents, and whether the predictors differ significantly between married and unmarried parents. If the institutional and normative supports of marriage are protective, we expect the associations of covariates with future dissolution will be significantly weaker for married parents than they are for unmarried parents. Note that this model estimates the associations between covariates and dissolution observed after couples have had a marital or nonmarital birth; it does not account for the nonrandom selection of couples into marital or nonmarital births or differences between married and unmarried parents that we do not observe in our data.
Married parents have more advantaged economic, behavioral, and relationship characteristics than cohabiting and dating parents around the time of their child’s birth. Table 1 shows that many married parents are white, but the racial composition of cohabiters is more evenly mixed between blacks, whites, and Hispanics; most dating parents are black. Unmarried mothers and fathers were younger at the birth of the focal child and have less education, lower employment rates, and lower earnings than their married counterparts. Self-reports of drug and alcohol use affecting daily life are less than 10 % for all relationship types, but they are higher for unmarried parents than they are for married parents. More than one-third of unmarried fathers have been incarcerated, compared with less than 10 % of married fathers. Relationship durations are longer for married parents, and more than one-half of married parents had children together before the birth of the focal child, compared with about one-third of cohabiting parents and one-quarter of dating parents. Unmarried parents, however, were much more likely to have had children with different partners prior to the focal child. Finally, indicators of commitment during the pregnancy and relationship quality at the time of the birth were high for all parents, especially those who were married or cohabiting.
Table 2 traces the survival of parental relationships following a child’s birth and reveals high levels of instability for unmarried parents relative to married parents. In this sample, 27 % of marital relationships ended by the child’s ninth birthday, compared with 53 % of cohabiting relationships and 81 % of dating relationships. In fact, 49 % of dating relationships ended within the first year after the child’s birth. Although few cohabiters remained stably cohabiting by the child’s ninth birthday (12 %), about one-third of cohabiting couples married by that time (35 %). In contrast, just 13 % of dating parents married by the child’s ninth birthday, and only 6 % remained in stable nonmarital relationships. Thus, stable nonmarital relationships are quite rare, with the overwhelming majority ending by the child’s ninth birthday, although couples cohabiting at the child’s birth are much more likely than dating couples to end their nonmarital unions through marriage.
Compositional Explanations for Union Dissolution
We examine how much of these line unadjusted differences in union dissolution can be explained by compositional differences in the marital and nonmarital populations in Tables 3 and 4. Table 3 presents coefficients from the full discrete-time event history logit model of the transition to union dissolution following a child’s birth, and Table 4 summarizes these results for the key covariates of interest and converts the coefficients to odds ratios. Model 1 of Table 3 presents the log odds coefficients from the baseline model, including dummy variable indicators for survey wave and for couples who were cohabiting or dating at the time of the child’s birth, relative to married parents. The positive and significant coefficients for the three-, five-, and nine-year wave dummy variables indicate that the likelihood of dissolution for a married couple (the omitted reference category) increases at each wave after the child’s birth.
At the one-year survey, the odds of union dissolution for a cohabiting couple are 5.88 times the odds of divorce for a married couple (see Table 4 for odds ratios). The negative coefficients on the interaction terms between cohabitation and survey wave in Table 3 indicate that the greater odds of dissolution for cohabiting parents relative to married parents decline somewhat over time, to 3.92 by the three-year survey and to 2.75 by the nine-year survey. The chances of dissolution are even higher for dating parents, whose odds of union dissolution by the child’s first birthday are 20.59 times the odds for married parents. Like cohabiters, however, the increased risk of dissolution for dating parents declines over time, although differences remain large. By the three-year survey, the odds of dissolution are 8.67 times higher for dating parents than for married parents, declining to 5.65 times higher by the nine-year survey.
Model 2 adds demographic and background control variables. Consistent with previous research, we find that black parents are significantly more likely to end their unions than white parents, while foreign-born parents are significantly less likely than native-born parents to end their unions. Additionally, mothers who grew up in intact families were less likely to end their unions, as were mothers who were older at the time of the child’s birth. The addition of the demographic and background characteristics explains 19 % of the disparity in dissolution between cohabiting and married parents in the first year after the child’s birth, about 25 % of the disparity in dissolution by three and five years, and about one-third of the disparity in dissolution by nine years after the child’s birth (Table 4, Model 2). For dating parents, demographic differences account for 16 % of the heightened risk of dissolution relative to married parents within the first year, and 24 % to 33 % at later years (Table 4, Model 2).
Model 3 adds parental economic characteristics to the model. Mothers and fathers with a college degree were significantly less likely to end their unions. Paternal earnings reduced the risk of dissolution, but maternal employment and maternal TANF receipt both increased the likelihood of dissolution. Accounting for demographic differences had already explained about 20 % of the difference in dissolution between cohabiters and married parents by the child’s first birthday, but the addition of the economic characteristics explains another 13 % of the dissolution disparity between the two groups within the first year (from 19 % of the gap with just demographic controls to 32 % of the gap with demographic and economic controls in Table 4, Model 3). Economic differences explain another 22 % of the difference by the fifth birthday, and more than one-half (58 %) of the disparity by the ninth birthday. The addition of economic characteristics also explains another 10 % to 19 % of the disparity in dissolution between dating and married parents (depending on the wave). Together, differences in the demographic and economic composition of the cohabiting and married populations explain about one-third of the gap in dissolution within the first year and one-half of the gap by five years; for dating parents, demographic and economic composition also explain 26 % to 51 % of the gap.
Model 4 adds behavioral controls to the model. Father’s incarceration is significantly associated with the likelihood of union dissolution, but no other predictors are significant, net of the other covariates in the model. The addition of the behavioral controls explains only a small additional amount of the disparity in dissolution rates (<10 %) net of economic and demographic controls.
Adding measures of relationship characteristics in Model 5 explains an additional 8 % to 16 % of the dissolution disparity between cohabiters and married parents within the first year, and an additional 12 % to 14 % of the disparity between dating and married parents. Here, distrust of men is significantly associated with an increased likelihood of dissolution. Prior fertility also matters: prior shared children promote relationship stability, but father’s prior multipartner fertility undermines relationship stability. Both relationship quality and relationship commitment are strong predictors of relationship stability. Taken together, the full model estimated in Model 5 shows that compositional differences explain more than two-thirds of the increased risk of dissolution for cohabiting and dating parents relative to married parents by the time the focal child is 9 years old.11
Models 1–5 measure relationship status at the time of the focal child’s birth, but unmarried couples may change their relationship status in the years following the child’s birth. Model 6 of Table 3 adds interactions between relationship status at baseline and relationship status at the survey wave t – 1 (the wave prior to when dissolution is measured). The negative interaction term between cohabiting at baseline and married at prior wave indicates that cohabiters who marry after the child’s birth have more stable unions than cohabiters who remain in cohabiting relationships. In contrast, cohabiters who transition to dating relationships are significantly more likely to end their unions than are stable cohabiters. Dating parents who transition to either cohabitation or marriage are significantly less likely to end their unions than are parents who remain stably dating.
Figure 1 provides a visual summary of the results of Tables 3 and 4. It plots the baseline predicted probability of union dissolution for each survey wave for married, cohabiting, and dating parents from Model 1 (solid lines) against the predicted probability of dissolution for the average couple in the sample accounting for the full set of covariates in Model 5 (dotted lines), with all dichotomous controls set to the omitted reference category and all continuous controls set to their sample grand means. The gaps between the solid and dotted lines in the figure show how much the probability of dissolution changes if we assign that relationship type the characteristics of the average couple in the Fragile Families sample instead of the average characteristics of their own relationship type.
The probability of dissolution within the first year after a child’s birth is about 50 % for the average dating couple but just 26 % for dating couples with the characteristics of the average couple in the Fragile Families sample. These disparities are further mitigated between the child’s third and ninth birthday, with just a 25 % chance of dissolution for dating parents with the characteristics of the average parent. Adjusting for compositional differences does less to reduce the probability of dissolution for cohabiters because their characteristics are more similar to the average couple. Their chances of dissolution drop from about 20 % to just less than 15 % within the first year after the child’s birth, and from 25 % to 17 % by the child’s ninth birthday. The likelihood of dissolution within the first year for married couples with the characteristics of the average parent remains low at about 5 % and increases to about 10 % by five years. Thus, the average couple has a 15 % chance of dissolution between their child’s fifth and ninth birthday if they were married at the child’s birth, a 17 % chance of dissolution if they were cohabiting, and a 25 % chance if they were dating. The closeness of the dotted lines, relative to the solid lines, shows that accounting for compositional differences explains much of the difference in the probability of dissolution across relationship types.
Figure 2 offers an alternative thought experiment, showing how much higher the probability of dissolution is for married parents when they have the characteristics of the average cohabiting couple or the average dating couple, instead of the characteristics of the average married couple. The probability of dissolution for the average married couple is about 5 % by the first year survey and rises to about 9 % between the five- and nine-year surveys. If that couple had the demographic, economic, and relationship characteristics of the average cohabiting couple instead of the characteristics of the average married couple, their probability of dissolution would increase to 19 % between the child’s fifth and ninth birthdays. And if the couple had the characteristics of the average dating couple, the probability of dissolution would rise to 29 % between the five- and nine-year survey waves. Thus, if married couples were as disadvantaged in their economic, behavioral, and relationship characteristics as the average dating couple, they would be about three times more likely to divorce between their child’s fifth and ninth birthdays.
Differences in the Predictors of Relationship Dissolution
We now turn to our second research question: whether the predictors of relationship dissolution differ for married and unmarried parents.12 Table 5 presents the results from two models. The first includes demographic and economic covariates; the second includes the full set of demographic, economic, behavioral, and relationship covariates. In Table 5, we present results from separate regressions for married and unmarried parents that show whether the coefficients differ significantly from zero. We also estimate a fully interacted model to test whether the coefficients for married and unmarried parents are significantly different from each other.13
The chances of union dissolution increase over time after the child’s birth for married couples, but the chances of union dissolution typically decline over time for unmarried couples, particularly for dating couples (consistent with the trends in Fig. 1 and Table 3). The associations between demographic characteristics and the log odds of dissolution are largely similar for married and unmarried parents, except that racial differences in the chances of dissolution are larger for unmarried parents than for married parents. In contrast, many economic characteristics are more strongly associated with dissolution for married parents than they are for unmarried parents. Both mother’s and father’s earnings and education are more strongly associated with relationship stability for married parents than for unmarried parents. In contrast, earnings, education, and employment are by and large not significantly associated with dissolution for unmarried parents, net of demographic controls.
Model 2 adds measures of behavioral and relationship characteristics. Among married parents, the only significant behavioral predictor of dissolution is maternal drug use. In contrast, the only significant predictor of dissolution for unmarried parents is whether the father was incarcerated. (The magnitudes of the incarceration coefficients are similar for married and unmarried parents, but the standard error is larger for married parents, rendering the coefficient insignificant.) Self-reported health was not significantly associated with dissolution for either married or unmarried parents, net of the other measures in the model.
Both relationship quality and commitment are significant predictors of stability, but relationship quality is a significantly stronger predictor of stability for married parents than it is for unmarried parents. Prior fertility also matters, with fathers’ prior multi-partnered fertility undermining union stability for both married and unmarried parents.
In sum, economic factors, like education and earnings, were stronger predictors of marital union stability than nonmarital union stability. Relationship quality matters for both groups, but it is a significantly stronger predictor of stability for married parents than it is for unmarried parents. These findings do not support the hypothesis that the normative and institutional benefits of marriage are protective in the face of relationship or economic challenges. Many of the predictors of dissolution do not differ between married and unmarried parents. Those that do differ—mainly economic and relationship characteristics—are usually significantly stronger predictors of dissolution for married parents than for unmarried parents. We cannot make strong causal claims based on the models presented here because our results may be driven by factors associated with union dissolution that are correlated with unobservable characteristics differently for parents with marital births than for those with nonmarital births. However, the associations we observe here suggest that differential responses to economic or relationship disadvantage do not explain why unmarried parents are more likely to end their unions than married parents.
Unmarried parents have considerably less-stable unions than married parents. In this article, we examined the sources of instability for married, cohabiting, and dating parents following the birth of a shared child. We found that compositional differences explained more than two-thirds of the increased risk of dissolution for cohabiting and dating parents relative to married parents. If married parents were as disadvantaged as the average dating couple, they would be about three times more likely to end their unions by their child’s ninth birthday.
We also examined whether the predictors of dissolution differed for married and unmarried parents and found that the association between SES and subsequent dissolution and that between relationship quality and subsequent dissolution were stronger for married parents than for unmarried parents. In other words, the likelihood of relationship dissolution increased at a faster rate for married parents as economic standing or relationship quality declined. In fact, measures of SES were rarely significant predictors of dissolution for couples following a nonmarital birth. In comparison, relationship quality was the strongest predictor of nonmarital union stability, but it was an even stronger predictor of marital stability.
Instead of providing evidence that the institutional and normative supports of marriage protect couples against economic or relationship hardships, we find that, if anything, the stability of marital unions is more sensitive to relationship and economic conditions after a child’s birth than the stability of nonmarital unions. Thus, differential responses to economic or relationship disadvantage do not explain why unmarried parents are more likely to end their unions than are married parents. We focus on unmarried parents here because of their relevance for child well-being and public policy; these results may not apply to childless couples because parents face a different set of constraints and considerations than nonparents when choosing to end their unions.
Why are economic standing and relationship quality more strongly associated with dissolution for married parents than for unmarried parents? One clue comes from a body of qualitative data suggesting that unmarried couples with children perceive a “marriage bar”: that is, they have higher standards for marriage than they have for nonmarital relationships. They express hesitancy to marry when they face economic problems or relationship difficulties, and believe they would be less likely to stay together if they exposed their relationships to the higher standards of marriage rather than the more flexible standards associated with nonmarital unions. These couples claim that certain conditions—such as extreme economic marginality, frequent conflict, involvement in crime, incarceration, or even infidelity (Edin and Kefalas 2005; Edin and Reed 2005; Edin et al. 2004; Gibson-Davis 2009; Gibson-Davis et al. 2005; Reed 2006; Waller 2002)—can be dealt with in a nonmarital union but would virtually mandate a divorce if they were married. Our findings provide suggestive evidence in favor of this “marriage bar” hypothesis, which yields predictions opposite those of the institutional hypothesis.
Our results should not be interpreted as a definitive test of this “marriage bar” perspective, however, because the differential predictors of dissolution we observe for married and unmarried parents may be due partly to differences between these two groups that we are unable to observe in our data. Our results could be biased if these unobserved characteristics are differentially correlated with the predictors in our models for married and unmarried parents. Although our models describe how married and unmarried parents behave following the birth of a child, they should not be interpreted as causal effects because couples select into marital and nonmarital unions in nonrandom ways that are not fully accounted for in these models.
The Fragile Families Study has a more extensive battery of control variables than those typically available in nationally representative data sets, but it is still unlikely that we measure all the characteristics that differ between married and unmarried parents, and we likely measure some of the observed variables with error. Qualitative studies of unmarried parents have found that infidelity and sexual jealousy, mistrust, substance abuse, and domestic violence are the most common reasons mothers gave for ending their relationships (Edin and Kefalas 2005; Hill 2007; Reed 2007). These experiences are likely to be underreported in our survey data, which may weaken their estimated coefficients in our analysis; furthermore, infidelity was not measured in the all waves of the survey. The compositional differences we observe in this study should therefore be considered conservative estimates, or lower bounds, on how much of the disparities in union dissolution are explained by compositional factors. Even as conservative estimates, our results show that compositional differences explain most of the elevated dissolution rates among unmarried parents, while differences in the predictors of dissolution are not the primary explanations for these disparities (at least in the large urban areas covered by the Fragile Families study).
We conclude by highlighting the need for additional research on the sources of instability within nonmarital unions. In our models, very few of the “usual suspects” predicted the likelihood of dissolution for unmarried parents following a child’s birth, other than relationship quality and commitment, and the explanatory power of these models was low relative to the models for married parents. Qualitative research provides intriguing insights into alternative explanations that are not currently measured in standard surveys, such as differing norms governing the conditions under which relationships should be maintained or dissolved. Surveys should construct items informed by these qualitative insights. Future work should also employ methods that can estimate causal effects rather than simply describing populations as we did in this article (Duncan 2008). This is an important area for future research, given that family instability has long-term consequences for the well-being of children.
Earlier versions of this article were presented at the annual meeting of the American Sociological Association in San Francisco, CA, and at the annual meeting of the Population Association of America in Detroit, MI. The authors thank Paula England, Michael Rosenfeld, Anastasia Snyder, Stewart Tolnay, Bradford Wilcox, and four anonymous reviewers for helpful suggestions and comments. Tach acknowledges support from the Robert Wood Johnson Foundation Health & Society Scholars Program.
In contrast, cohabitations among parents in Scandinavian countries (Sweden, Norway, and Finland) are nearly as stable as American marriages (Andersson 2002: Table 5).
When interviewed just after their child is born, two-thirds of mothers and 75 % of fathers believed that it was better for children if their parents were married. Their plans for their relationships at the time of their child’s birth reflected this view: more than 80 % said they were romantically involved, nearly three-quarters said there was at least a fifty-fifty chance they would marry each other, and more than one-half said that the probability of marrying each other was either good or certain (Center for Research on Child Wellbeing 2002, 2007).
Ironically, dissolutions of cohabiting relationships lead to declines in income for both men and women. After break-ups, men who cohabited suffer from moderate declines in economic standing, and the declines for their ex-partners are much sharper. This is especially true for African Americans and Hispanics. These effects are similar to those of formerly married men and women (Avellar and Smock 2005).
This work is consistent with other qualitative studies that are retrospective rather than longitudinal (Edin and Kefalas 2005; Waller 2002).
In the Fragile Families data, nonresponse and attrition were higher for unmarried mothers and fathers than for married parents. At baseline, 87 % of eligible unmarried mothers agreed to participate in the survey, and 75 % of the fathers were interviewed. At subsequent survey waves, response rates for unmarried mothers were 90 % at one year, 88 % at three years, 87 % at five years, and 76 % at nine years. Mothers who dropped out of the study were more likely to be white or Latino, were less likely to be married to the father when the child was born, and had lower average SES (Cooper et al. 2007).
In a comparison of mothers’ and fathers’ reports of relationship status at the one-year follow-up (when missing data are least for both mothers and fathers), we found that mothers and fathers agreed on their relationship status in 87 % of cases in which we had information reported by both the mother and father. Part of this disparity is due to the fact that some mothers and fathers were interviewed on different dates, sometimes with lags of several months between their respective survey dates.
This is measured for the first time at the one-year survey, so both the baseline and one-year measures come from the one-year survey.
We used this measure because it was asked of all survey respondents. A question about romantic relationship duration was asked only of couples still in romantic relationships at the one-year follow-up, which excludes couples who broke up within the first year. The correlation between these two measures for respondents with nonmissing data on both was about .65.
This includes parents’ relationship status at baseline, mother’s and father’s employment and educational characteristics, mother’s race, child’s gender, mother’s and father’s history of drug use and incarceration, and mother’s and father’s reported relationship quality.
Couples exit our sample after they dissolve their unions, and they do not reenter the sample if they rekindle their relationship at a later survey wave.
Because the married and unmarried parent populations differ markedly in their racial composition, racial differences are possible in the associations between relationship status and dissolution. In subsequent analyses (not shown, available upon request), we tested whether race moderated the association between relationship status and dissolution. We found that in a model including all covariates, the associations between relationship status at the birth (cohabiting or dating vs. married) and subsequent dissolution did not differ significantly across white, black, and Hispanic mothers. We performed a similar analysis to test whether the associations between relationship status and dissolution varied by the duration of the relationship (which measures how long each mother had known her partner at the time of the child’s birth)—less than five years versus more than five years—and also found that the associations between relationship status and dissolution did not differ significantly by prior union duration in the full model.
As a first step toward examining the causes of dissolution, we examined mothers’ self-reported reason for dissolution. The overwhelming majority of married and unmarried mothers who break up list “relationship reasons” as the primary reason, which includes everything from “don’t get along” to “too young.” Relatively few list economic hardships, drug use, or incarceration as the primary causes of dissolution. Of course, this measure may not be an accurate account of the conditions that cause couples to break up. First, there are many disparate reasons lumped together under the broad label of “relationship issues.” Second, although “relationship reasons” may be the proximate cause of dissolution, economic or behavioral conditions could be more-distal factors that influence the quality of relationships, which in turn influences of relationship dissolution.
We pool cohabiting and dating parents into a single unmarried group because preliminary analyses established that their predictors were substantively and statistically similar (results not shown; available upon request).