shows weighted descriptive statistics for the analytic sample. Looking first at socioeconomic and demographic characteristics, the sample is largely non-Hispanic white, especially among the sample of marital first births; minorities, particularly non-Hispanic blacks, are under-represented due to the sample restriction that the first birth occur within a cohabiting or marital union. Just over three-fourths of the women lived with both biological parents at age 14, with substantially higher proportions of women in the marital birth sample having lived with both parents than in the cohabiting births sample. About 30% of women reported that their mother's education was high school or less, about 40% reported that their mother had a high school degree, and about 30% reported that their mother had some college or higher. Among the combined sample, 79% of the women themselves had a high school degree at the time of their first birth. Women in the marital first birth sample were more educationally advantaged than women in the cohabiting first birth sample, as indicated by both their mother's and their own education status.
Weighted Descriptive Statistics for Women with a Cohabiting or Marital First Birth
Turning now to relationship characteristics and history, 7% of women had cohabited with a different partner and 3% had been married to a different partner prior to their first-birth union. 7% were partnered with men who had been married before. More women who were cohabiting at their first birth had cohabited in the past and had a partner who had been previously married, while more women who were married at their first birth had themselves been married before. The majority of women in the sample (83%) were married at the time of birth, with 56% having not cohabited with their partner prior to marriage and 27% married at birth but having cohabited prior to marriage with their partner. On average, the couples had been together in a coresidential relationship just under 3 years prior to their first birth; as would be expected, those with a cohabiting first birth had been together a shorter time period (about 21 months) than those with a marital first birth (about 37 months). There were about 8 years of observation on average (not shown) between the first birth and the relationship's end or time of the survey. By the end of the period of observation, about a third of the relationships had dissolved. This varied by the type of relationship at birth, with two-thirds of cohabiting relationships dissolving compared to only a quarter of marital relationships.
Finally, looking at the fertility characteristics, women were on average 24 years old at first birth, with married mothers being about 3 ½ years older (25 years) than cohabiting mothers (21.4 years). About one-fifth had conceived their child prior to the start of coresidence (e.g., their first birth occurred 7 months or less after the start of coresidence), but this was more likely to be the case for cohabiting births (33%) than marital births (19%). In light of the relatively long average duration of relationships at the time of birth, this suggests that our sample has substantial variability in relationship status and strength prior to first births – some couples were coresiding in response to a pregnancy, while others (primarily married couples) had been together for a long time. Two-thirds of women reported that both she and her partner had intended their first birth (40% among cohabiting women and 72% among married women), while 7% reported that both she and her partner did not intend to get pregnant with their first child (22% among cohabiting women and 5% among married women). The remaining one-fourth of the women reported disagreement between themselves and their partner on whether the birth was intended or not (38% among cohabiting women and 24% among married women). By the end of the period of observation, about 63% of women had had a subsequent birth in the same union as their first birth (of the women without a second birth in the same union, 45% of relationships had dissolved and 55% were censored at the time of survey, not shown). 43% of women reported only intended subsequent births, 2% reported only unintended subsequent births, 10% reported only subsequent births where she and her partner disagreed upon their intentionality, and 8% had a combination of different types of births – intended, unintended, and/or disagreed-upon births. Having intended subsequent births only was more common among those with a marital first birth (46%) than those with a cohabiting first birth (30%), whereas more cohabitors reported not having any subsequent births (50%) than married women (34%) by the end of the period of observation.
Discrete-time event history results
Couples with unintended births are likely to have other characteristics associated with instability. We turn to multivariate event history models to account for some of these correlated characteristics. details the results from the logistic regression of socioeconomic, demographic, relationship, and fertility variables on the stability of women's coresidential unions (combining cohabitation and marriage). Results are presented in the form of odds ratios. As the dependent variable measures whether the relationship dissolved or not, a number less than one indicates a decreased risk of dissolution and a number greater than one indicates an increased risk of dissolution in a given person-month.
Odds Ratios from Logistic Regression of Birth Intendedness on Union Dissolution among Women with a Coresidential (Cohabiting or Marital) First Birth in the 2002 NSFG
Model 1 shows the unconditional association of first birth intentionality with union dissolution for all women with a coresidential (cohabiting or marital) first birth, controlling for relationship duration after the first birth to account for independent exposure risk. As hypothesized, an unintended or disagreed-upon birth increases the likelihood of union dissolution. A birth that is considered unintended by both partners increases the odds of dissolution fivefold relative to an intended birth, while disagreement doubles the odds. Union dissolution is significantly more likely after an unintended birth (by 32%) than after a disagreed-upon birth as well (not shown). The likelihood of dissolution is elevated in the two years following a birth (OR=1.28) but subsequently declines over time. Model 2 adds in socioeconomic, demographic, and union characteristics. Although the elevated chances of dissolution seen in Model 1 are sharply attenuated by controlling for other characteristics – indicating that selection on observable characteristics (particularly union type) explains much of the higher likelihood of dissolution after an unintended birth – first birth intentionality is nonetheless an important predictor of dissolution. Compared to women who reported that they and their partner intended their first birth, having an unintended first birth or disagreeing with their partner about birth intentionality is associated with a significantly higher odds of dissolution, even in the presence of socioeconomic, demographic, and relationship controls. When the respondent reported that both she and her partner did not intend the birth, the odds of dissolution are about 81% higher than if the birth was intended. Among couples with disagreement on intentionality (meaning at least one person considered the birth intended), the odds of dissolution are significantly higher than among couples in which the first birth was intended, by about 30%. Significance tests (not shown) demonstrated that the difference in the likelihood of dissolution between unintended births and disagreed-upon births is also statistically significant, with the odds of dissolution being about 40% higher if the birth was unintended by both partners than if it at least one partner reported the birth was intended, as expected in Hypothesis 1.
Relationship type is the strongest predictor of subsequent union stability among parents, even more so than intentionality. Women who were cohabiting at birth (regardless of whether they had subsequently married or not) have odds of dissolution about four times higher than women who were married at birth and had not cohabited prior to marriage. Women who cohabited prior to marriage but had a marital birth also have an elevated risk of dissolution compared to women who had a marital first birth and did not cohabit prior to marriage. Women who had prior cohabitations or marriages had elevated odds of dissolution (OR=1.42 and OR=1.62, respectively).
Generally, other socioeconomic and demographic characteristics are not associated with union dissolution, though the risk of dissolution was lower for foreign-born Hispanic women relative to non-Hispanic white women (OR=0.72). The lack of significant socioeconomic and demographic predictors of dissolution seems surprising given previous findings of variation in union stability. This result occurs primarily because socioeconomic and demographic characteristics are strongly related to first birth circumstances (particularly intentionality and union status at first birth), so limiting our sample to coresidential first births and controlling for circumstances at the time of birth accounts for most variation in stability. Finally, it is worth noting that the odds of dissolution decrease with union duration, are inversely related to the woman's age at birth, and increase for women who had their first births in more recent years.
Model 3 adds information on subsequent fertility and intentionality to the first birth measures in Model 1 (the unconditional model). Two things are of note here. First, adding measures of subsequent fertility improves model fit, indicating that subsequent fertility and intentionality is an important independent predictor of union stability. In particular, relative to women who have only intended subsequent births (the modal category), women who do not have a second birth are about 71% more likely to experience relationship dissolution. (Of course, couples who break up are no longer at risk for a second birth together. Because measures of fertility are time-varying, and models account for time elapsed since the first birth, our models capture effects of fertility on dissolution and not the reverse causal direction.) Women with only unintended subsequent births are 2.77 times as likely to experience dissolution than women with only intended births, and women with disagreed-upon births are 1.62 times as likely to experience dissolution, net of first birth intentionality. Second, the association between the odds of dissolution and first birth intentionality remains large and significant, with the odds of dissolution 3 times as high for an unintended first birth relative to an intended first birth and about 1.6 times as high for a disagreed-upon first birth, even controlling for subsequent fertility.
Model 4 adds socioeconomic and demographic characteristics. The effects of the socioeconomic, demographic, and union formation variables change little compared to Model 2. As such, we again focus our discussion of results on fertility intentionality. The magnitude of the association between first birth intentionality and union dissolution is only minimally attenuated when adding indicators of higher-order fertility to other controls (Model 4 vs. Model 2). Women with an unintended or disagreed-upon first birth remain significantly more likely to experience relationship dissolution, by about 74% and 22%, respectively. That is, the association between first birth intendedness and relationship dissolution does not appear to be explained by either subsequent childbearing (or lack thereof) or socioeconomic, demographic, or union characteristics. Further, the association between subsequent fertility and dissolution seen in Model 3 is only attenuated slightly by the inclusion of socioeconomic, demographic, and union variables in Model 4, suggesting higher-parity births have a strong, independent effect on union stability. Further, in models not shown, where we interacted first and second birth intentionality, we found that any combination of fertility and intentionality other than a first intended birth followed by only subsequent intended births increased the risk of union dissolution. Multiple unintended births, though relatively rare, were particularly detrimental to union stability.
In analyses presented in , we tested whether birth intentionality affects stability differently in cohabiting versus marital unions, showing Models 2 and 4 (models with controls) presented in disaggregated by relationship status at first birth. The first two columns show the results for first births in cohabiting unions (including women who marry after the birth). Focusing on birth intentionality, first birth intentionality increases the odds of dissolution by about a third in Model 2 (without controls for subsequent fertility) but becomes non-significant in the presence of higher-order fertility indicators and socioeconomic, demographic, and relationship variables. Higher-order fertility itself is associated with dissolution, but the magnitude of the association is fairly small. Women who have no second birth are 40% more likely to experience the dissolution of their first birth union relative to women with intended subsequent births, and only unintended subsequent births in the first-birth union increase the likelihood of dissolution by 83%. Relatively few socioeconomic and demographic variables are associated with union stability among women who were cohabiting at their first birth, though women who had not transitioned to marriage have about 25% lower odds of dissolution than women who transitioned to marriage. This is somewhat counterintuitive, but mirrors Manning's (2004)
findings that children have little effect on cohabitation stability but a destabilizing effect on couples who transition from cohabitation to marriage.
Odds Ratios from Logistic Regression of Birth Intendedness on Union Dissolution among Women with a First Birth in the 2002 NSFG, by Union Type at First Birth
Looking at Models 2 and 4 for marital unions reveals a different picture. Here, contrary to Hypothesis 3 (where we expected a strong negative association for cohabitations but not marriages), first and higher-order births are quite strongly related to union dissolution, with unintended first and subsequent births independently increasing the likelihood of dissolution. In Model 2, which includes only first birth intentionality, the odds of marital dissolution are 3.7 times as high after an unintended first birth relative to a intended first birth, with disagreement increasing the odds of dissolution by about 50%. The inclusion of higher-order fertility variables attenuates the magnitude of the first-birth variables somewhat, but they remain large and statistically significant. Even controlling for higher-order births, an unintended first birth increases the likelihood of dissolution threefold. Further, couples who have only unintended subsequent births are 4.9 times as likely to experience dissolution relative to those who only have intended subsequent births. Disagreement on first and higher-parity births increase the odds of dissolution as well, by 39% and 55%, respectively.
Clearly, then, the increased risk of dissolution for unintended and disagreed-upon births seen in is largely driven by the effect on marriages. It may be that cohabiting unions are so inherently unstable that fertility (and intentionality) affects stability differently than it does for marriage - recall that in the combined models, relationship type is the strongest predictor of instability by far, with individuals who were cohabiting at first birth far more likely to dissolve than those who were married at first birth. It is also worth noting that some of the findings regarding the demographic and union variables from the combined models are significant only for the married subsample. Foreign-born Hispanics are significantly less likely to experience dissolution, and prior cohabitation and marriage increases the chances of divorce, but these associations are only present for women who were married at their first birth.
shows results from fixed-effects analyses of relationship dissolution after intended and unintended births (Model 5). Recall that only time-varying characteristics can be included in these models, and as a result coefficients are estimated based on changes in the characteristic. The coefficients for our central independent variables, birth intentionality, can be interpreted as the difference in the odds of dissolution in birth intervals following an unintended or disagreed-upon birth relative to intervals following an intended birth, the reference category. All stable characteristics of women and their relationships - including unobserved characteristics as well as variables included in previous models, such as the couple's relationship status at the first birth, whether the first birth was legitimated, whether married couples cohabited before marriage, the age at the start of coresidence, family background, etc. - are accounted for in this model.
Odds Ratios from Fixed-Effects Regression of Intendedness of Most Recent Birth on Union Dissolution among Women with a Coresidential First Birth in the NSFG
Contrary to hypothesis 3, fixed-effects models show a large positive association between unintended fertility and relationship dissolution. The odds of dissolution are 3.42 times higher after an unintended birth than an intended birth, and this association is statistically significant (p<.001). The association shown in Models 2 and 4 is not attenuated when accounting for stable characteristics; in fact, the coefficient is larger in the fixed effects specification. The coefficient may be larger because unobserved characteristics not accounted for in Models 2 and 4 suppress the true association. In addition, fixed-effects models estimate subject-specific coefficients, rather than population-averaged coefficients, which tend to be larger in magnitude (Teachman 2011
). The association between couple disagreement about birth intentionality and dissolution is also positive, and about the same magnitude as in Model 2 above (OR = 1.26). However, because this coefficient is estimated based only on couples with more than one birth of different intentionalities, this model has less statistical power and the coefficient is not statistically significant (p=.21). Overall, Model 5 confirms the basic finding in the models above that unintended births negatively impact union stability. The association between unintended fertility and relationship dissolution is not purely the result of selection based on stable individual and couple characteristics.
As noted above, this type of analysis can produce biased coefficient estimates for characteristics that vary monotonically with time. For example, couples transition from cohabitation to marriage, but not from marriage to cohabitation, so the coefficient for cohabitation during the month only varies in one direction. The negative coefficient for cohabitation in the model may result from this bias - since couples only transition to marriage if their cohabiting relationship does not dissolve, the odds of dissolution during marriage are necessarily greater for these couples. However, this coefficient is also consistent with the finding from the models for couples cohabiting at the first birth that dissolution rates are higher for those who marry after the birth than those who remain cohabiting.