Multiple-Partner Fertility and Cohort Change in the Prevalence of Half-Siblings

Abstract

Introduction

The American family has experienced profound transformations since the mid-twentieth century fueled by growth in cohabitation, divorce, and nonmarital childbearing. Today, less than one-half of children will grow up with two married, biological parents (Carlson and Meyer 2014; Manning et al. 2014). Instead, children experience complex family systems as parents break up, repartner, and have additional children with new partners. The resulting families—defined by multiple-partner fertility, or parents with children by different partners—have important implications for child development and family stratification (Bronte-Tinkew et al. 2009; Dorius and Guzzo 2013; Fomby and Osborne 2016; Fomby et al. 2016; Halpern-Meekin and Tach 2008). Children’s exposure to multiple-partner fertility may have become more common over time, but data limitations have made it difficult to ascertain its changing prevalence in the wake of rapid family change. Here, we aim to fill this gap in the literature by documenting cohort change in a key indicator of children’s exposure to multiple-partner fertility: the presence of half-siblings.

The presence of half-siblings is a defining feature of multiple-partner fertility from the perspective of children. Although family scholars have documented the extent of maternal and paternal multiple-partner fertility in the U.S. population (Dorius 2012; Guzzo and Furstenberg 2007; Manlove et al. 2008; Thomson et al. 2014), prior estimates of multiple-partner fertility understated children’s exposure because usually they included data about the fertility of only one parent or only siblings living in the same household at a point in time (Guzzo and Dorius 2016; Stykes and Guzzo 2019). Most previous research has also focused either on a single cohort or change over a short period, which obscures long-run cohort change. Although prior work offers a strong foundation about the demographic correlates of multiple-partner fertility, these limitations have made it difficult to ascertain children’s cumulative exposure to half-siblings, how much it varies across demographic groups and economic strata, or how much it has changed in the wake of the second demographic transition.

We build on prior work by using data from the National Longitudinal Surveys of Youth (NLSY), which collect the information required to create cumulative measures of children’s exposure to half-siblings for two cohorts that span a period of great family change in the United States. In an advance over previous studies estimating multiple-partner fertility, we use nationally representative longitudinal data to measure both mothers’ and fathers’ fertility and both resident and nonresident half-siblings over the full course of childhood, yielding more complete measures and spanning a longer historical period than prior literature. In contrast with most previous research, we estimate exposure to half-siblings for cohorts of children instead of measuring fertility for cohorts of parents. First, we calculate children’s cumulative exposure to half-siblings across two cohorts and document the shifting bases of stratification. Then, we use demographic decomposition techniques to assess how compositional changes in the U.S. population and changes in subpopulation risks of experiencing multiple-partner fertility have fueled cohort change in children’s exposure to half-siblings.

Background

Multiple-partner fertility is the product of broad patterns of family change under the second demographic transition marked by delays in fertility and marriage; the rise of cohabitation, divorce, and other alternatives to marriage; and a growing disconnect between childbearing and marriage (Lesthaeghe 1995). Fueled both by ideational and economic change, the forces driving the second demographic transition have led to two types of family settings: one that reflects gains in resources to children via delayed childbearing, union stability, and material resources; and one that reflects losses in resources to children via nonmarital childbearing, union instability, and economic stagnation (McLanahan 2004; McLanahan and Jacobsen 2014). As a result, the second demographic transition has contributed to growing inequality in familial resources in the United States—what some have called “diverging destinies” of children (McLanahan 2004) or a “two-tiered family system” (Furstenberg 2014). Thus, the second demographic transition likely contributes to the intergenerational reproduction of socioeconomic and racial disparities (McLanahan 2009; McLanahan and Percheski 2008), particularly given that multiple-partner fertility is also correlated across generations (Lappegård and Thomson 2018).

Studying the prevalence of half-siblings—an indicator of multiple-partner fertility—is motivated in part by concerns over its implications for children in the United States (Klerman 2007). Multiple-partner fertility is a facet of complex (i.e., nonnuclear) family structures and family instability, both of which are associated with worse developmental, socioemotional, and behavioral outcomes for children net of other characteristics (Amato 2005; Cavanagh and Huston 2006, 2008; Fomby and Cherlin 2007; Langton and Berger 2011; Lee and McLanahan 2015; McLanahan and Sandefur 1994; Osborne et al. 2012). Studies investigating the impact of half-siblings independent of parents’ relationship instability and union status have also suggested that the presence of half-siblings can be detrimental to children’s and adolescents’ behavioral and educational outcomes (Bronte-Tinkew et al. 2009; Dorius and Guzzo 2013; Fomby et al. 2016; Halpern-Meekin and Tach 2008), even when children live with both biological parents (Gennetian 2005; Ginther and Pollak 2004).

The presence of half-siblings may negatively impact children because parents’ legal and social obligations to each other and to their children are less institutionalized than they are in standard nuclear families (Cherlin 1978; Sweeney 2010). The lack of confluence in biology, residence, and legal recognition that results from half-siblings might weaken reciprocal webs of obligation and support among family members, which are codified and legitimated in family law and social policy (Meyer et al. 2011; Stewart 2006). Thus, multiple-partner fertility can result in the dilution of parental economic and social resources when spread across multiple households (Manning and Smock 1999, 2000; Tach et al. 2010) as well as heightened familial role ambiguity that undermines parental authority, relationship quality, and investments in children (Stewart 2005). The presence of half-siblings is also associated with relationship instability and heightened levels of parental stress that may contribute to lowering children’s psychological well-being and achievements (Cherlin 1978; Halpern-Meekin and Tach 2008; Harcourt et al. 2015).

Although most studies to date have focused on coresident half-siblings—in part because of difficulties in identifying nonresident siblings (Dorius and Guzzo 2013; Fomby et al. 2016; Halpern-Meekin and Tach 2008)—nonresident half-siblings also influence the transmission of social and economic resources from parents to children. In particular, prior research has found that nonresident fathers’ contributions of time and money to biological children from past relationships decline after they have children with new partners (Cherlin and Seltzer 2014; Furstenberg 2014; Manning and Smock 1999; Manning et al. 2004). Although most scholars have focused on the challenges involved with parenting under conditions of family complexity, others have noted potential strengths and resilience that may stem from a larger extended kin network (Furstenberg 2014; Sanner et al. 2018). Still others have suggested that the negative consequences of family complexity found in most empirical studies may be due to the selection of disadvantaged individuals into such family forms (Lee and McLanahan 2015).

Existing Prevalence Estimates

Researchers have used three types of data to identify the prevalence of multiple-partner fertility in the population: men’s or women’s fertility data, relationship rosters from household surveys, or more complete fertility and family data that are limited to disadvantaged populations. Each of these data sources has distinct advantages and disadvantages for measuring multiple-partner fertility, which leads to varying estimates of its pervasiveness (Guzzo 2014; Guzzo and Dorius 2016; Stykes and Guzzo 2019). In addition to data limitations, discrepancies arise because of differences in the operationalization of the measure, units of analysis, age span of respondents, and use of cross-sectional versus longitudinal data (for a detailed review, see Guzzo and Dorius 2016).

Identifying all of a child’s half-siblings requires information about both parents’ fertility, but most nationally representative surveys do not collect this information. From a child’s perspective, estimates of half-siblings based on fertility data typically count the half-siblings produced by only one parent. For example, estimates of the prevalence of male multiple-partner fertility typically range from 8 % of men to 17 % of fathers aged 15–44 (Guzzo 2014; Guzzo and Furstenberg 2007; Scott et al. 2013). Underreporting of men’s fertility is also a well-known issue for survey data (Grall 2013; Joyner et al. 2012). Stykes and Guzzo (2019) found that about 12 % to 13 % of all women aged 15–44, and 22 % to 24 % of mothers in that age range, experienced multiple-partner fertility (see also Dorius 2012; Guzzo 2014; Thomson et al. 2014). Estimates based on fertility data must often make assumptions about paternity for those with multiple non-union births, which can lead to bias when nonmarital fertility is common (Stykes and Guzzo 2019). Even in the absence of this bias, fertility estimates from only one parent still do not represent children’s complete exposure to multiple-partner fertility.

Another way to estimate multiple-partner fertility is through detailed relationship matrices available in some nationally representative household surveys, such as the Survey of Income and Program Participation (SIPP). These data show that 7.5 % of mothers had children by more than one partner in 2008 (Evenhouse and Reilly 2010), and 11.1 % of children had half-siblings (Manning et al. 2014). Household rosters underestimate multiple-partner fertility considerably because they account for only those living in the same household as the respondent. Stykes and Guzzo (2019) found that estimates using the SIPP rosters capture only one-quarter of the women with self-identified multiple-partner fertility (also see Monte 2017). Although scholars have advocated for direct measures about multiple-partner fertility in surveys (Guzzo and Dorius 2016), Stykes and Guzzo (2019) found that a direct question about multiple-partner fertility, which was included in the SIPP in 2014, substantially underestimates women’s multiple-partner fertility, perhaps because of social desirability bias.

Other data sources offer more complete fertility information from both mothers and fathers, but they are drawn from disadvantaged samples that limit their generalizability to the national population. In the Fragile Families and Child Wellbeing Survey, a sample that is representative of urban births, 60 % of unmarried parents and 21 % of married parents reported having children from a previous partnership at the time of the focal child’s birth (Carlson and Furstenberg 2006). Using a sample of families in the Wisconsin child support enforcement system, which includes children under age 18 who live with mothers and have identified fathers, Cancian and Meyer (2011) estimated that 31 % of mothers were in couples in which at least one partner has multiple-partner fertility. Using other Wisconsin administrative data covering every child who received a referral or whose parent contacted the state child support agency—capturing almost all nonmarital births in the state—Cancian et al. (2011) found that 60 % of firstborn children of unmarried mothers had at least one half-sibling by age 10. The Wisconsin administrative data in the latter study underscores the importance of having fertility information for both mothers and fathers: 23 % of children had half-siblings only on their father’s side, 18 % had them only on their mothers’ side, and 19 % had half-siblings from both parents. Without data on both parents’ fertility, estimates of children’s exposure to half-siblings would be significantly lower.

Previous studies have underscored the wide variability in estimates of the prevalence of multiple-partner fertility (Guzzo 2014; Guzzo and Dorius 2016) and documented different forms of bias in estimates across approaches and data sets (Stykes and Guzzo 2019). The present study builds on this literature by focusing on estimates for cohorts of children (via half-siblings) rather than cohorts of parents (via multiple-partner fertility). We also provide a more complete accounting of children’s exposure to multiple-partner fertility via half-siblings by including both mother’s and father’s fertility, both resident and nonresident siblings, and measuring exposure cumulatively over the course of childhood rather than at a point in time. Finally, building on the abbreviated temporal coverage of prior research, we generate estimates for two cohorts that span a period of great family change in the United States and document the changing bases of stratification in children’s exposure to multiple-partner fertility.

Determinants of Multiple-Partner Fertility

The predictors of multiple-partner fertility are remarkably consistent across studies, pointing to the intersection of socioeconomic disadvantage and family formation. Mothers and fathers who start their fertility at young ages are significantly more likely to go on to have children with multiple partners (Carlson and Furstenberg 2006; Guzzo and Furstenberg 2007b; Manlove et al. 2008), in large part because parents who are younger at the time of their first birth tend to have less stable romantic unions than do those who start childbearing at older ages (Guzzo and Hayford 2012; McLanahan and Beck 2010) and are thus at risk of childbearing with multiple partners for a longer period. In the United States, nonmarital relationships involving children are significantly less stable than marital relationships: by age 15, 34 % of children born to married parents and 73 % of children born to unmarried parents have witnessed their parents’ unions dissolve (Andersson et al. 2017). When American parents’ relationships end, 49 % enter a new cohabiting or marital relationship within 3 years (Andersson et al. 2017), and more than 75 % do so within 10 years (Andersson et al. 2017; Bramlett and Mosher 2002).

Multiple-partner fertility is also linked to socioeconomic status (SES) in large part because SES shapes the aforementioned timing and relationship contexts of fertility. Women who come from lower-SES families and who have low levels of education themselves have children at significantly younger ages, and they are much more likely to do so in the context of nonmarital and less stable relationships (Amato 2010; Manning et al. 2004; Raley and Bumpass 2003; Smock and Greenland 2010). Multiple-partner fertility is also stratified by parental race/ethnicity. Guzzo and Furstenberg (2007) estimated that 15 % of black men had children with multiple partners, compared with 6 % of non-Hispanic white men and 9.5 % of Hispanic men. A great deal of the racial/ethnic variation in multiple-partner fertility can be attributed to differences in SES and the proximate fertility and family-formation behaviors discussed earlier (Guzzo and Furstenberg 2007). However, family scholars have pointed to the unique historical and institutional experiences of slavery and discrimination in shaping kinship relations within African American communities (Clayton et al. 2003; Patterson 1998; Stack 1974; Tucker 2000). They have also noted distinct fertility patterns among some immigrant populations (Raley et al. 2004), which may influence the prevalence of multiple-partner fertility among racial/ethnic groups with large shares of foreign-born residents, including Asians and Latinos (Carlson and Furstenberg 2006; Manlove et al. 2008). Given that multiple-partner fertility is more common among disadvantaged parents and that many scholars argue that it could diminish children’s life chances, it may contribute to social and economic stratification along racial and class lines (McLanahan and Percheski 2008).

Trends in Multiple-Partner Fertility

The rapid pace of family change over the past half-century as part of the second demographic transition has altered fertility and family-formation behaviors (Lesthaeghe and Neidert 2006; Sobotka 2008), some of which may have increased the prevalence of multiple-partner fertility and others of which may have worked in the opposite direction. The age of first marriage increased substantially, and a larger share of the population never marries (Raley et al. 2015), which means that more adults are potentially at risk for having children outside marriage for a growing portion of the life course (Finer 2007; Ventura 2009; Ventura and Bachrach 2000).

The age of first birth has also increased substantially among more-educated women, fueled by the growing availability of contraception and increased educational attainment and labor force participation. These forces have lowered women’s total fertility over time (Sweeney and Raley 2014). Taken together, rising age at first birth and declining total fertility over the past several decades may have depressed the overall prevalence of multiple-partner fertility, particularly for more-educated women.

At the same time, the age of first birth has not increased as much for less-educated women despite rising ages of first marriage, indicating a greater disconnect between marriage and childbearing among lower-SES women than among higher-SES women in the United States (Edin and Kefalas 2011; Gibson-Davis 2009). As a result, nonmarital childbearing has grown precipitously among the less-educated but barely budged for the most-educated (McLanahan 2004). The growing share of children born outside marriage has likely increased low-SES children’s exposure to family instability over time (Kennedy and Bumpass 2008; Manlove et al. 2010), thereby increasing their likelihood of multiple-partner fertility. Collectively, the aforementioned trends suggest widening socioeconomic differences in fertility and family-formation decisions (Furstenberg 2014; McLanahan 2004).

Racial/ethnic dimensions of family change are partly a reflection of socioeconomic inequality, but some racial/ethnic differences in family formation and stability persist even within socioeconomic strata (Raley et al. 2015; Sweeney and Raley 2014), suggesting additional historical, social, and institutional forces at work. For example, total fertility rates for black and white women have converged, and fertility has declined for Latino women. Despite this convergence, teen birth rates remain twice as high for black and Latino women relative to white women (Sweeney and Raley 2014), nonmarital and solo childbearing are more common for black women than for their white counterparts within the same educational strata (Sweeney and Raley 2014), and the timing and prevalence of marriage have diverged considerably for black and white women (Raley et al. 2015).

Even though trends in multiple-partner fertility may shed light on the stratification of children’s family experiences, it is not clear how macroeconomic and sociocultural changes have influenced its prevalence. To date, the few studies addressing trends in multiple-partner fertility have produced conflicting findings. Using the National Survey of Family Growth to examine cohort trends in male fertility, Guzzo and Furstenberg (2007) found that rates of age-specific multiple-partner fertility are higher among younger cohorts than older cohorts, but Manlove et al. (2008) found a decline in multiple-partner fertility from older to younger cohorts using the same data. Researchers who have used the SIPP to examine temporal change (rather than cohort change) between the early 1990s and 2010 found little change in the point-in-time prevalence of residential multiple-partner fertility over this period (Evenhouse and Reilly 2010; Manning et al. 2014).

The Present Study

Multiple-partner fertility likely affects a substantial number of children and has important implications for their access to parents’ material and social resources, but prior studies provided only a partial enumeration of childbearing across partnerships and understated children’s exposure to multiple-partner fertility. As a result, we have an incomplete understanding of children’s exposure to half-siblings, how exposure has changed across cohorts in the wake of the second demographic transition, and how exposure varies by race and SES. In this article, we address these questions by drawing on data from two nationally representative cohorts of the NLSY to calculate children’s cumulative exposure to multiple-partner fertility and how it has changed during a period of great family change in the United States. We then document the shifting bases of social stratification in children’s exposure to half-siblings. Finally, we use demographic decomposition techniques to assess the influence of compositional changes and changes in subpopulation risks on cohort change in children’s exposure to multiple-partner fertility in the United States.

Data and Method

Data and Analytic Samples

We use the 1979 and the 1997 cohorts from the National Longitudinal Surveys of Youth (NLSY79, NLSY97). The NLSY79 cohort is a nationally representative sample of 12,686 youth born between 1957 and 1964 who were first interviewed in 1979 when they were 14–21 years old. The NLSY97 is a second nationally representative longitudinal survey of 8,984 youth comprising birth cohorts 1980–1984 who were first interviewed (along with a parent) when they were 12–17 years old in 1997. Both NLSY data sets contain information that identifies multiple-partner fertility from both maternal and paternal fertility and from both residential and nonresidential siblings.¹

To make the NLSY79 cohort comparable with the NLSY97 cohort, we first exclude the military and economically disadvantaged oversamples (N = 2,923). Then, we restrict the sample to the 8,815 respondents still in sample as of the 1993 survey wave, when the survey collected information on the presence and demographic characteristics of up to 12 siblings. We also exclude 355 respondents who had not given information on the age of one or more of their siblings. We include only those siblings born before the respondent turned age 18 for comparability across surveys. The NLSY79 asked questions about respondents’ biological relatedness to each sibling in 2006, 2010, and 2012, so we drop 1,218 respondents who did not have data on their relationship to any of their siblings because of attrition in all three survey waves, as well as 83 respondents who did not provide information on their relationships to all their siblings. This yields a final analytical sample for the NLSY79 of 7,159 youth.

The NLSY97 initial sample was composed of 8,984 respondents. We use information collected in the first survey round on the resident and nonresident rosters, which ask respondents to list and provide demographic characteristics for all household members on the resident roster and all nonresident family members on the nonresident roster. Half-siblings are listed as a family relationship category on both the resident and nonresident rosters. We drop 83 respondents who were categorized as “mixed race” so that our racial categories are consistent with the 1979 cohort race definitions (described in greater detail later). Our final sample contains 8,901 youth.

We use multiple imputation with chained equations with 50 data sets to impute values for covariates with missing information in either cohort (i.e., mother’s education, mother’s age at first birth, and marital status at respondent’s birth).² For analysis of both the NLSY79 and NLSY97 cohorts, we use weights created in the NLSY custom weighting program to correct for nonresponse and attrition, which produces weighted statistics that are nationally representative.³ In the online appendix (Tables A1 and A2), we report analyses that assess potential bias due to nonrandom attrition, item nonresponse, and imputation. We find that changes to the analytic sample yield very small changes (of less than 1 percentage point) to the estimates of interest and do not alter our substantive conclusions. We also validate the representativeness of our samples by benchmarking summary statistics to decennial censuses and vital statistics for the whole population (shown in Table 1) and separately by child race/ethnicity (shown in Table A3, online appendix).

Measures

Presence of Full and Half-Siblings

For the NLSY79 cohort, we determine presence of full and half-siblings through questions in the 2006, 2010, and 2012 survey waves that asked whether the respondent and each of his/her siblings shared the same mother and whether they shared the same father. We categorize siblings who shared the same mother and the same father as full siblings. Siblings who shared only the same mother or the same father—but not both—are categorized as half-siblings. Siblings with missing information on relatedness are excluded from the sample, as are siblings who are not biologically related (e.g., stepsiblings or adoptive siblings) because they are not measured comprehensively in either cohort.⁴ We also use information on the respondent’s and siblings’ ages to restrict the sample to siblings who were born before the respondent’s 18th birthday to ensure consistency with measures in the NLSY97 cohort (as described later).

We determine the presence of siblings in the NLSY97 cohort through a variable that indicates the respondent’s relationship to each person listed on the household and nonhousehold rosters. This variable includes categories for full and half-siblings (on the mother’s and father’s side) on both rosters. The household roster was updated in each survey wave, but the nonhousehold roster was obtained only in the baseline 1997 survey, when the respondents were aged 13–17. Because of this, we are missing information on nonresident siblings who were born after 1997 for respondents who were aged 13–16 at the time of the baseline survey. To recover this information, we use the youth who were age 17 at the time of the 1997 survey to estimate the age-specific shares of new full- and half-siblings born between ages 13 and 17 and impute additional siblings for those who were aged 13–16 at the time of the 1997 wave using the hot deck method (Andridge and Little 2010; Mander and Clayton 2007); see the online appendix for details.⁵

Race

We categorize the respondent’s race as non-Hispanic white, non-Hispanic black, Hispanic, and other race. The “other” group may include American Indian/Eskimo/Aleut, Asian/Pacific Islander, and “something else.” We keep respondents of “other” race in both surveys for some descriptive analyses, but we exclude them from the decomposition analyses given that the composition of this group likely changed a lot over time, deeming the category inconsistent across cohorts. In the NLSY97, youth were also categorized as “mixed race” (N = 83). We drop these cases because of lack of correspondence with the racial categories available in the NLSY79.

Mother’s Education

We measure mother’s education based on a question asked in the first round of each survey about the number of years of schooling completed by respondents’ biological mothers. When this information is missing, we use yearly information collected in the household roster to fill in mother’s education. Then, we recode mother’s years of schooling into four categories: (1) less than high school (< 12 years); (2) high school diploma (12 years); (3) some college education (13–15 years); and (4) college degree or more (16 or more years).

Mother’s Age at First Birth

With the NLSY79 survey, we generate mother’s age at first birth using mother’s age at the year of birth of the respondent’s oldest sibling with the same mother, which was collected in the 1988 survey wave. If the resulting age at first birth is implausible (younger than 11), we replace their age at first birth with age at youth’s birth (N = 231). In the first round of the NLSY97 survey, mothers were asked directly about their age at first birth.

Mother’s Marital Status at Respondent’s Birth

The NLSY79 does not identify marital status at the parent’s first birth or at the respondent’s birth, so we follow prior studies by proxying mother’s marital status at respondent’s birth based on living arrangements when respondent was 0 years old (Bloome 2017). In 1988, NLSY79 respondents were asked to retrospectively report their relationship to coresident adults at each age from 0 to 17. Because rates of nonmarital childbearing and cohabitation were quite low among the parents of children in birth cohorts 1957–1964 (Ventura and Bachrach 2000), living with both biological parents at age 0 is highly correlated with them being married. In the NLSY97, we rely on a relationship history roster asked of the responding parent at the baseline survey, and we compare the dates of marriage spells with the respondent’s birth year to determine mother’s marital status at the respondent’s birth.

Mother’s Total Fertility

In both the NLSY79 and the NLSY97, we calculate mother’s total fertility when the child is 17 by summing all resident and nonresident biological children on the mother’s side that were enumerated by respondents.

Methods

First, we estimate children’s cumulative exposure to full and half-siblings by age 17 across racial and socioeconomic subgroups. Similar to previous work on mothers’ fertility (Stykes and Guzzo 2019), we compare these estimates with those in the literature derived either from mothers’ fertility or household rosters in order to assess the magnitude of bias generated by using fertility information from only one parent or only resident children. For comparability with previous work (Guzzo and Dorius 2016), we produce two sets of estimates: (1) the proportion of children who have at least one half-sibling, and (2) the proportion of children with any siblings who have at least one half-sibling. The first estimate indicates prevalence in the population, and the second assesses the risk of having a half-sibling for children (i.e., the prevalence among those who are eligible to experience parental multiple-partner fertility).

The compositional changes and changes in subpopulation risks can be calculated using either the 1979 or the 1997 cohort as the reference group. Because it is not clear which group should be chosen as the reference group for most decomposition analyses, scholars have suggested that the reference group should be derived from a pooled sample. We report results using the pooled sample of the two cohorts⁸ (Oaxaca and Ransom 1994), and we use normalized regressions to properly estimate risks for categorical variables⁹ (Gardeazabal and Ugidos 2005; Yun 2005).

Results

Compositional Shifts Across Cohorts

Table 1 shows the demographic composition of respondents in both cohorts, benchmarked by data from comparable cohorts using decennial censuses and vital statistics. The proportion of children who are white declined, but the proportion of children who are Hispanic nearly doubled. The share of respondents’ mothers without a high school diploma declined by almost one-half, and the share of mothers with some postsecondary schooling more than doubled. Across cohorts, the median age of respondents’ mothers’ first birth also increased by about two years, and mothers’ mean total fertility declined by about one child. The starkest changes occurred in mothers’ marital status. Whereas 93 % of the NLSY79 cohort had mothers who were married at the time of their birth, only 80 % of youth in the NLSY97 did so. These trends reflect broader national trends, as indicated by characteristics from comparable cohorts in the decennial census and vital statistics.¹⁰ Table A3 in the online appendix reports descriptive statistics separately by child race/ethnicity.

Cohort Change in Sibling Complexity

Table 2 shows the cohort-specific proportions of children who have full or half-siblings by their 18th birthday. The share of children with any siblings by their 18th birthday declined from 95 % in the 1979 cohort to 90 % in the 1997 cohort (panel A). The average number of siblings also declined during this period, from 3.2 to 2.0. Full siblings constitute the majority of all siblings, and their numbers declined as well, from 2.7 to 1.5 (panel B). Nearly 90 % of the children in the 1979 cohort had at least one full sibling, falling to about 80 % of children in the 1997 cohort.

Finally, and most importantly for the present study, the prevalence of half-siblings increased by 5 percentage points across cohorts (panel C). The share of children with a half-sibling in the 1979 cohort was quite high: one in five children (20.4 %) born between 1957 and 1964 had a half-sibling by the time they turned age 18. Exposure to half-siblings also increased significantly across cohorts, and more than one in four children (25.5 %) born between 1980 and 1984 had at least one half-sibling by their 18th birthday. Among children with any siblings, the prevalence of half-siblings increased by 7 percentage points (from 21.5 % to 28.4 %).

Our estimates of the prevalence of half-siblings are much higher than existing nationally representative estimates in the literature because previous estimates typically included only siblings living in the same household (in the case of half-sibling estimates) or assessed only one parent’s fertility history (in the case of multiple-partner fertility estimates). Many estimates also identified the presence of siblings at a point in time, whereas our estimates measure the cumulative prevalence of half-siblings by age 18.

To demonstrate the importance of data from both parents, and from both coresident and nonresident locations, we reestimate the prevalence of half-siblings for youth in each cohort using information on only one parent’s fertility and information on only either coresidential or nonresidential siblings. Results presented in Table 3 suggest that maternal fertility measures miss almost one-quarter of families with half-siblings, and residential sibling measures miss almost one-half of families with half-siblings. Specifically, if our estimates of children’s exposure to half-siblings included only residential half-siblings in the 1997 cohort, the prevalence would be only 13.2 % (9.4 + 3.8) of children versus 25.5 %. Similarly, if we count only half-siblings on the mother’s side, our estimate of the share of children with half-siblings is almost 6 percentage points lower for the 1997 cohort at 19.6 % (16.7 + 2.9) versus 25.5 %, and 3.5 percentage points lower for the 1979 cohort at 17 % (14.5 + 2.5) versus 20.4 %. The estimates of children’s experiences of half-siblings on the mother’s side are similar to previous estimates of maternal multiple-partner fertility produced using the NLSY (Dorius 2012; Dorius and Guzzo 2013).

Although our estimates of exposure to half-siblings provide more complete coverage than prior estimates, our estimates are still likely conservative. Our results suggest that respondents may be less likely to report half-siblings on their fathers’ side, mirroring the established issue of underreporting of paternal fertility in surveys (Grall 2013; Joyner et al. 2012; Rendall et al. 1999). For example, a much smaller share of the youth reported having a half-sibling on the father’s side than on the mother’s side: about 9 % (5.8 + 2.9) reported paternal half-siblings in the 1997 cohort, and 6 % (3.5 + 2.5) did so in the 1979 cohort. These estimates suggest a higher share of reported half-siblings are born to mothers than to fathers, which does not corroborate estimates based on administrative data that found a similar prevalence of multiple-partner fertility for disadvantaged men and women (Cancian et al. 2011). Previous studies have documented the underestimation of father’s multiple-partner fertility through household rosters, incomplete reports from mothers, and father’s low response rates (Grall 2013; Joyner et al. 2012; Rendall et al. 1999). Our study suggests that father’s multiple-partner fertility may also be underestimated through children’s reports. We summarize this with a measure of the ratio of maternal half-siblings to paternal half-siblings in Table 3, which indicates a ratio of roughly 2 to 1. Perhaps NLSY youth are less aware, or less likely to report the existence, of half-siblings on their fathers’ side because of nonresidence, lack of contact with fathers, or feeling less close to paternal half-siblings. Although our estimates are not immune to the pervasive measurement issues associated with paternal fertility, they still demonstrate the value of using information from both parents and from both residential and nonresidential sources; underreports of paternal half-siblings suggest that the prevalence of half-siblings is likely even larger than what is reported here. Notably, the underreporting of half-siblings on the fathers’ side relative to the mothers’ side appears to occur in both cohorts, which suggests that it might have less influence on our interpretation of trends for the general population than it has on levels. Table 3 also indicates, however, that underreports of paternal half-siblings may vary across social groups; they are most likely to be captured among white children or children of the most-educated mothers.

To test whether our substantive conclusions about the changing prevalence of half-siblings hold in the absence of this potential differential underreporting, we conducted the analyses presented here using only measures of maternal fertility, which likely include less error. These robustness checks (available upon request) suggest that our conclusions about trends in the prevalence of half-siblings among the general population and by SES and race are not substantially affected by potential underreporting of half-siblings on the father’s side.

Changing Bases of Social Stratification

Next, we assess the changing bases of stratification in exposure to half-siblings across several characteristics. Table 4 shows a large racial/ethnic difference in the presence of half-siblings in the 1979 cohort. Nearly 44 % of black children had half-siblings, and they were almost three times as likely to have half-siblings as white children (16 %) and twice as likely as Hispanic children (24 %). Over time, these racial gaps narrowed significantly. The proportion of children with a half-sibling increased by 8 percentage points for white children, from 16 % to 24 %. It also increased slightly (by 2 percentage points) for Hispanic children. The proportion of children of “other race” with a half-sibling also decreased over time, likely because of changes in the shares of each racial group included in this composite measure. And, perhaps most notably, the proportion of black children with a half-sibling decreased by 9 percentage points, from 44 % to 35 %. Additional analyses benchmarking demographic characteristics of non-Hispanic black children in the NLSY with data from the census suggest that differential selection of non-Hispanic black children across NLSY cohorts cannot explain this trend (see online appendix, Table A3). This trend also holds when we examine only maternal half-siblings (Table 3).

The socioeconomic gradient in multiple-partner fertility, as measured by maternal education, has also shifted over time. Among the 1979 cohort, multiple-partner fertility was strongly concentrated among the least-educated, with more than 30 % of children whose mothers had less than a high school diploma having half-siblings. Exposure to multiple-partner fertility was significantly lower among children whose mothers had a high school diploma (17 %), some college (13 %), or a four-year college degree (11 %). Over time, the prevalence of multiple-partner fertility has grown across all parts of the educational distribution, but it has grown the most among the middle of the educational distribution: by 11 percentage points for mothers with a high school diploma and by 13 percentage points for mothers with some college. This shifted the educational cleavage of multiple-partner fertility from having high school diploma versus not to having four-year college degree versus not.

Exposure to multiple-partner fertility also varies by maternal age at first birth. In the 1979 cohort, 38 % of children whose mothers had a teen birth, 19 % of children born to mothers who were young adults (aged 18–24), and 12 % of children born to older mothers (aged 25 or older) had half-siblings. Multiple-partner fertility has increased over time for mothers of all ages, although it has increased the most for young adult mothers (by 10 percentage points). By the 1997 cohort, 42 % of children whose mothers had teenage births, 30 % of children whose mothers were young adults at the time of first birth, and 15 % of children with older mothers had half-siblings.

Regarding shifts in nonmarital births, just 17 % of children in the 1979 cohort whose parents were married at their birth were exposed to multiple-partner fertility by their 18th birthday, compared with fully 59 % of children whose parents were unmarried. The likelihood that children of married parents were exposed to multiple-partner fertility increased across cohorts: in the 1997 cohort, 21 % of children with married parents at birth had half-siblings by their 18th birthday. This increase is likely due to the rising divorce rate over this period. By contrast, exposure to multiple-partner fertility declined among children with unmarried parents between 1979 and 1997: in 1997, 44 % of children whose parents were unmarried were exposed to multiple-partner fertility during childhood. This pattern likely occurred because of the declining selectivity of nonmarital childbearing as it expanded over this period. There is still a large cleavage in exposure to half-siblings by parental marital status, but it has narrowed over time.

Finally, exposure to half-siblings also varies by maternal total fertility. In the 1979 cohort, about 14 % to 16 % of children whose mothers had one¹¹ to three children had a half-sibling, compared with 25 % of those whose mothers had four or more children had half-siblings. The 1997 cohort shows greater variation in the share of children who have half-siblings among those whose mothers had one, two, or three children (11 %, 19 %, and 27 %, respectively). Although exposure to half-siblings increased over time for all groups, it increased the most for children whose mothers had three or four children (by 11 and 16 percentage points, respectively).

Demographic Decomposition of Cohort Change in Sibling Complexity

The composition of the U.S. population has changed markedly between the 1979 and 1997 cohorts. The importance of shifting population composition for trends in children’s exposure to multiple-partner fertility is unclear, however, because the increasing prevalence of half-siblings may be due to compositional shifts or to changes in the associations between population characteristics and half-siblings (i.e., subpopulation risks). Table 5 shows the results of logistic regression models predicting presence of half-siblings in a pooled sample of the 1979 and the 1997 cohorts. All models include a dummy variable that shows how the log-likelihood of having half-siblings differs in the 1997 cohort relative to 1979 cohort. Changes in this coefficient across models that control for different population characteristics indicate how changes in population composition account for growing sibling complexity.

The unconditional model (Model 1) shows a positive and significant time trend in exposure to half-siblings, which is consistent with the results presented in Table 3 (an increase of 5 percentage points). The cohort coefficient is only slightly smaller in Model 3 after controls for race and ethnicity are included (0.285 vs. 0.326), indicating that changes in the racial composition of the population explain little of the trend in sibling complexity across cohorts. When we control for compositional shifts in mother’s education (Model 2), mother’s age at first birth (Model 4), and mother’s total fertility (Model 6), the coefficient in the cohort dummy variable becomes even larger. This suggests that compositional shifts in mother’s education, age at first birth, and total fertility contributed to reducing exposure to multiple-partner fertility over time. In the absence of declining total fertility and rising maternal education and age at first birth, cohort increases in children’s exposure to half-siblings would have been even larger than what occurred. Model 5 indicates, on the other hand, that compositional changes in nonmarital childbearing explain a large portion of the increase in exposure to multiple-partner fertility over time as the coefficient in the cohort dummy variable decreases substantially from 0.326 to 0.099.

The models presented in Table 5 constrain the associations of population characteristics with sibling complexity to be constant across cohorts, but these associations may vary. Table 6 reports results from the full model (Model 7 of Table 5) estimated separately by cohort, which allows the associations of covariates to vary over time. Models in Table 6 indicate differences in subpopulation risks between the 1979 and the 1997 cohorts. For instance, although the probability of half-siblings was higher for non-Hispanic black children than non-Hispanic white children in the 1979 cohort net of other variables in the model, this difference is no longer statistically significant in the 1997 cohort net of the other covariates in the model. These results confirm the descriptive trends in Table 4, which show a narrowing of black-white differences in sibling complexity.

The shifting effects of maternal education on sibling complexity across cohorts are more nuanced. In the 1979 cohort, children of mothers with less than high school education were significantly more likely to have half-siblings than children of mothers with a high school diploma as their highest educational attainment. By the 1997 cohort, the difference between mothers without a high school diploma and mothers with either a high school diploma or some college had narrowed, while differences between mothers with a four-year college degree and everyone else had increased. In other words, the educational gradient in sibling complexity remains, but the cleavage in the educational distribution has shifted toward college graduates versus others, which also confirms descriptive trends portrayed in Table 4. Table 6 also indicates that mother’s age at first birth and marital childbearing remain significantly and negatively associated with exposure to sibling complexity over time, although the association became weaker for both variables. Finally, the association between mothers’ total fertility and sibling complexity remained positive and became stronger over time.

Our decomposition analysis in Table 7 brings together the two strands of analysis in Tables 5 and 6 by calculating the amount of change in children’s exposure to half-siblings over time that is attributable to compositional shifts or changes in subpopulation risks. The pooled decomposition results (Model 1) suggest that the vast majority of the increase in exposure to half-siblings is due to changing subpopulation risks (i.e., rates) rather than due to shifts in population composition. If composition alone had changed between 1979 and 1997, the change in exposure to half-siblings would have been less than what occurred.¹² Conversely, if only subpopulation risks of exposure to half-siblings had changed (leaving composition the same in 1979 and 1997), exposure to half-siblings would have increased by even more (7.7 percentage points) than what occurred (6.2 percentage points).

Of all the compositional shifts that occurred across cohorts, changes in nonmarital childbearing were by far the most strongly associated with increasing exposure to half-siblings. If only the proportion of nonmarital births had changed across cohorts, children’s exposure to half-siblings would have grown by 2.1 percentage points between 1979 and 1997. By contrast, compositional shifts in mother’s education, total fertility, and age at first birth reduce the trend toward sibling complexity. Virtually none of the existing change in sibling complexity over time is explained by changes in the racial composition of the population. Thus, changes in the composition of the population over time have had countervailing consequences for children’s exposure to half-siblings over time.

Most changes in subpopulation risks of exposure to sibling complexity have worked to increase exposure to half-siblings over time, particularly for race, nonmarital childbearing, and mother’s total fertility. This finding resonates with the cohort-specific regressions shown in Table 6. The likelihood of having a half-sibling has increased a great deal for the most prevalent social groups: non-Hispanic white children, mothers who were married at the respondent’s birth, and mothers who have two children.

As shown in Table 4, the direction of cohort change in sibling complexity differed across racial groups. The proportion of white children with half-siblings increased by about 8 percentage points; by contrast, the proportion of black children with half-siblings decreased by 9 percentage points, and the prevalence of half-siblings for Hispanic children remained relatively stable. We reestimate the decomposition model separately for non-Hispanic white and black children in order to understand the sources of this change. The results for the full population are reflected in the non-Hispanic white population, with most growth in exposure to half-siblings due to changes in subpopulation risks rather than changes in population composition. Despite the decreasing trend in the prevalence of half-siblings over time for non-Hispanic black children, they represent the only group for whom compositional shifts would have contributed to increases in the prevalence of half-siblings. In fact, this observed decrease among non-Hispanic black children is solely driven by the intercept (–0.476), which suggests that it must be explained by trends in variables not included in our models. Although a full exploration of these additional factors is beyond the scope of the present study, we speculate about potential sources of change in the Discussion section.

Discussion

Multiple-partner fertility is a central dimension of family stratification in the United States: it is more common among disadvantaged parents, and it may limit children’s life chances by reducing parental resources or increasing familial stressors (Bronte-Tinkew et al. 2009; Carlson and Furstenberg 2006; Fomby et al. 2016; Guzzo and Furstenberg 2007b; Manlove et al. 2010). Exploiting underutilized sibling data from two nationally representative cohorts, we identify a remarkably high cumulative prevalence of half-siblings among youth born in the mid-twentieth century: about one in five children born between 1957 and 1964 had a half-sibling by their 18th birthday. Exposure to half-siblings increased even more across cohorts, with one in four children born between 1980 and 1984 having a half-sibling by their 18th birthday. Our estimates are larger than prior estimates of children’s exposure to multiple-partner fertility because we incorporate information on the fertility of both parents, include both residential and nonresidential siblings, and estimate cumulative rather than point-in-time exposure. These findings shed light on the widespread prevalence of a type of family complexity that may bring additional challenges to families in a context where nonbiological and nonresidential family relationships are not fully institutionalized within American society.

We also identify shifting racial and educational gradients in children’s exposure to sibling complexity. A strong racial/ethnic gradient in exposure to half-siblings existed in 1979, but this has narrowed significantly over time as the prevalence of half-siblings increased substantially for white children but declined for black children. Large educational gradients are evident for both cohorts; but whereas the most significant dividing line used to be whether a mother had a high school diploma, the more significant cleavage now is whether the mother has a college degree. These trends mirror broader national trends in the narrowing of racial/ethnic differences in family formation and the concomitant widening inequality based on college credentials (Cherlin 2014; McLanahan 2004; Raley et al. 2015). These trends also suggest that single-partner fertility has become more concentrated among the most-educated mothers—those with a four-year college education.

Despite these trends, the shifting racial and educational composition of the U.S. population cannot explain the growth in sibling complexity. Instead, trends in sibling complexity signal broader social changes that occurred in the wake of the second demographic transition, resulting from changes in the patterns of marriage, cohabitation, and childbearing. These trends have had countervailing consequences for children’s exposure to multiple-partner fertility, however. Nonmarital childbearing is the single strongest compositional change associated with an increase in exposure to half-siblings in our decomposition models, whereas changes in mother’s age at first birth and total fertility would have led to declines in exposure to half-siblings. Indeed, prior research has identified that early childbearing and nonmarital childbearing are key correlates of multiple-partner fertility for men and women because they predict both less stable relationships and more rapid repartnering (Guzzo 2014).

One surprising finding to emerge from our analyses is that exposure to half-siblings increased by 8 percentage points for non-Hispanic white children, while the prevalence of half-siblings decreased by 9 percentage points for non-Hispanic black children and changed very little for Hispanic children. These findings do not seem to be driven by sampling bias or measurement error. Our decomposition analysis suggests that white mothers’ subpopulation risks of exposure became more similar to those of black mothers over time. Thus, patterns associated with half-siblings are no longer confined to black families but now also occur more frequently among whites. The effects of disadvantageous characteristics are still greater for racial minorities, however, and results indicate that the prevalence of half-siblings would not have decreased for black children if not for secular changes among black families that are not captured by the standard sociodemographic characteristics included in our models. Uncovering the unobserved sources of this change is a ripe area for future research and may include factors such as access to contraception and health care, changing social norms, the growth of a black middle class, and labor market dynamics (Burton et al. 2009; Kearney and Levine 2015; Lichter et al. 1991; Sweeney 2002).

Our work on the historical and changing prevalence of half siblings invites a broader scholarly conversation about the changing meaning of multiple-partner fertility and half-siblings over time. Much of the scholarship on multiple-partner fertility has been partly motivated by concerns over the implications for family processes and child development. Scholars have argued that negative outcomes for children in complex families may occur because parents’ legal and social obligations to each other and to their children are less institutionalized than they are in standard nuclear families in the United States (Cherlin 1978; Sweeney 2010). Our findings invite scholars of family stratification to revisit these hypotheses now that multiple-partner fertility has become more common across a range of demographic groups. Have such changes resulted in accumulating disadvantages for children because of greater instability and fewer resources? Or have complex family systems become less deleterious in contexts where such family forms are more common and may therefore receive greater formal and informal legitimacy and recognition? Our findings also invite further subgroup analysis of the institutionalization hypothesis given differences in the prevalence of the experience across subgroups historically and today. Finally, future research might consider the implications of half-siblings from a child’s perspective in terms of social and economic exchange over the life course, much as family scholars investigated the meaning of stepsiblings as an emerging family form a generation ago (Beer 1989; Cherlin 1978).

Despite the value of using nationally representative data to observe half-siblings across multiple cohorts, the NLSY data also have limitations, inviting further study of the measurement of siblings specifically and family complexity more broadly. First, we are able to observe only one form of sibling complexity—half-siblings—because the NLSY does not collect data from both cohorts about other types of siblings (e.g., step or adoptive) in cumulative or comprehensive ways. Second, we are also unable to include other relevant covariates (e.g., parents’ work status, income, immigration status) in our models because they could not be created consistently across cohorts. The lack of comprehensive data on fathers and nonresident parents is particularly unfortunate. Finally, our ability to measure the presence of siblings depends on the participant youth considering someone a sibling, and our results indicate that some half-siblings may be left out of these accounts, particularly paternal half-siblings. We attempt to estimate the extent of this bias and do not believe that it accounts for the cohort trends or subgroup variation in half-siblings that we find in this article, but the true extent of this bias is unknown and constitutes a ripe area for future data collection efforts, along with investigations of paternal fertility measures more broadly.

Growth in children’s exposure to sibling complexity has implications for the measurement of childhood family structure, which often neglects multiple-partner fertility from both parents or from siblings who do not share the same household. The dearth of high-quality fertility data suggests a clear need for better data and measures that include the presence of siblings from both parents and better data on nonresident family members from a longitudinal perspective. Without such data, studies of the effects of family structure will continue to misclassify children in complex families, which will downwardly bias estimates of the effects of family structure on children’s outcomes. Better measurement of sibling complexity would also shed light on the shifting bases of stratification in family processes. Although educational gradients persist, racial differences are narrowing, and changing patterns of family formation have countervailing influences on sibling complexity. Continued examination of racial and socioeconomic differences in family life will help to uncover the changing institutionalization of nontraditional family forms as well as the evolving mechanisms by which family processes exacerbate or mitigate inequality across generations.

Acknowledgments

We would like to thank Kelly Musick, Vida Maralani, Megan Sweeney, Elizabeth Wildsmith, and anonymous reviewers for their thoughtful feedback on earlier versions of this article. This work was supported by the William T. Grant Foundation.

Notes

References

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.