How Education Shapes Women's Work and Family Lives Across Race and Ethnicity

Abstract

Drawing on life course and intersectional approaches, this study examines how education shapes the intertwined domains of work and family across race and ethnicity. By applying multichannel sequence analysis and cluster analysis to the National Longitudinal Survey of Youth 1979, we identify a typology of life course trajectories of work and family and test for the interactive associations of race and ethnicity with college education for different trajectory types. While our results show statistically significant and often sizable education effects across racial and ethnic groups for most of the work‒family clusters, they also suggest that the size and direction of the education effect vary widely across groups. Educational attainment plays an outsize role in shaping Black women's work‒family lives, increasing their access to steady work and partnerships, while educational attainment primarily works to increase White women's participation in part-time work. In contrast, Latina women's work‒family trajectories are less responsive to their educational attainment. In combination, the racialized role of education and persistent racial and ethnic gaps across the education distribution yield unequal patterns in work‒family strategies among Black, Latina, and White women.

Introduction

In the United States, educational attainment stratifies women's labor market outcomes and family transitions (Cherlin 2014; DiPrete and Buchmann 2013; Goldin 2021). Although women's educational attainment increased dramatically in the last half century, White women are almost twice as likely to receive bachelor's degrees as Black and Latina women (DiPrete and Buchmann 2013). These latter two groups are also systematically excluded both from the best employment opportunities and from some family forms, such as steady partnership or partnered fertility (Barnes 2016; Branch 2011; Browne and Misra 2003; Cross 2023). Educational attainment differences by race and ethnicity could explain these disparities in women's work and family outcomes. Yet some evidence suggests that education varies in shaping work and family outcomes for Black, Latina, and White women (Clarke and McCall 2013; Cross 2018; Damaske et al. 2017), suggesting that even when Black and Latina women achieve education levels comparable with those of White women, they will not experience the same patterns as they combine work, marriage, and fertility across the life course.

Surprisingly little research has explored how education shapes work‒family patterns differently for Black, White, and Latina women. We take both intersectional and life course perspectives to investigate this question. An intersectional perspective posits that education, race, and ethnicity shape women's access to work and family forms and how they reconcile these intertwined life domains (Clarke and McCall 2013; Damaske 2011; Flippen 2014). A life course perspective further suggests that race, ethnicity, and education may shape the transitions, timing, and sequencing of work and family patterns (Elder 1998; Moen 2001; Moen and Chermack 2005; Moen and Han 2001). Recent longitudinal studies have suggested that education, race, and ethnicity have long-term consequences for employment patterns and family forms (Aisenbrey and Fasang 2017; Damaske and Frech 2016; Fasang and Aisenbrey 2022; Killewald and Zhuo 2019), although these studies have emphasized the importance of race and ethnicity net of education (rather than in combination with education).

We take advantage of rich longitudinal data on women's employment, fertility, and partnership histories from the National Longitudinal Survey of Youth 1979 (NLSY79), a nationally representative cohort sample of U.S. adults born between 1957 and 1964, to answer the following questions. What are common patterns of work and family across women's life courses? To answer this first question, we use multichannel sequence analysis and cluster analysis to construct a typology of life course trajectories of work and family for Black, Latina, and White women between the ages of 23 and 50. We next ask, do the associations between college education and these trajectories vary by race and ethnicity and, if so, how? To answer this question, we use multinomial logistic regression models with coarsened exact matching to test for the interactive associations of the measure of race and ethnicity together with college education for different trajectory types.

Conceptual Framework: Life Course and Intersectional Perspectives on Women's Work and Family

Taking a life course perspective, we conceptualize work and family as spheres that interlock over time. Many studies have focused on the effects of family events on women's employment, measuring employment using either current measures of labor force participation or specific employment transitions (e.g., England et al. 2004; Felmlee 1984; Montez et al. 2014). While rich and informative, this research cannot capture the reciprocal and cumulative connection between the work and family spheres over the longer life course. Because most women still take on the bulk of caregiving and housework (Bianchi and Milkie 2010; Hook 2010), they face the difficult challenges of reconciling their labor force participation with their family circumstances (Blair-Loy 2009; Collins 2019; Damaske 2011; Stone 2007). For example, women who withdraw from the labor market when entering motherhood will accumulate less early work experience, a handicap to securing employment later in life (Willson 2003). In contrast, women in demanding professions may be left with less time to dedicate to their families but have more resources to outsource caregiving and housework (Gerstel and Clawson 2014; Schneider and Hastings 2017).

We also take an intersectional approach, which requires us to wrestle with “categorical complexity,” or the expectation that within each category (e.g., education) there are social groups (college-educated or not college-educated) and then even more detailed groups (White college-educated, Black college-educated, Latina college-educated) (Branch 2011; Hancock 2007; McCall 2005). In the United States, education, race, and ethnicity appear to play an outsize role in determining which women work, form and maintain lasting partnerships, and achieve their fertility goals (Barber et al. 2021; Barnes 2016; Damaske 2011; Dow 2019; Higginbotham 2001). Late baby boom women, as captured by the NLSY79 cohort, experienced simultaneously greater access to college education and the labor market and, for less educated women, diminished access to good jobs and to membership in nuclear families (Branch and Hanley 2014; Furstenberg 2007; Wilson 2011).

Because work and family operate as reciprocal and cumulative endeavors in women's lives, we expect work‒family trajectories to develop differently depending on race, ethnicity, and education. This approach allows us to integrate important intersectional and qualitative findings that racial, ethnic, and class inequalities lie not only in work‒family outcomes at specific ages but also in life course pathways (Clarke 2011; Damaske 2011).

The Intersection of Race, Ethnicity, and Education in Work and Family Patterns

Longitudinal studies point to the importance of education, race, and ethnicity for the timing and sequencing of White women's work patterns and transitions. Once Damaske and Frech (2016) controlled for education, they did not find racial and ethnic differences in women's trajectories of steady full-time work, working hours that increase with age, or lack of paid employment. They did find that reentry employment trajectories, characterized by a withdrawal from full-time work around age 30 followed by a return to full-time work by 45, are more prevalent among White women than among Black and Latina women, and that Latina and White women (but not Black women) are more likely to reduce workforce participation during their prime childbearing years. Across racial and ethnic groups, college-educated women experience steadier and longer employment spells (Alon and Tienda 2005; Damaske and Frech 2016), while women of color with no college face greater job volatility and longer spells of nonemployment at younger ages, which, in turn, lead to lower labor force attachment (LFA) across the life course (Alon and Haberfeld 2007).

Studies of work and family trajectories have found that Black and Latina mothers stay out of the labor force for shorter periods after childbirth and exhibit stronger and steadier LFA after motherhood; White mothers are more likely to take part-time employment (Hynes and Clarkberg 2005; Killewald and Zhuo 2019; Lu et al. 2017). A study analyzing multichannel trajectories of family transitions and occupational prestige in the United States noted that, net of several socioeconomic status measures and compared with White men, Black women are less likely to be in work‒family clusters that combine partnered parenthood, two or more children, and medium- or high-prestige occupations, and also less likely to be in clusters that combine childlessness, partnership, and medium-prestige occupations (Aisenbrey and Fasang 2017).

Still, much of the research on the intersection of race, ethnicity, and education, which we review in the following, focuses on singular work‒family outcomes, such as marriage, fertility, or employment.

The Educational, Racial, and Ethnic Gradients of Marital Patterns

Scholarship on U.S. families usually examines education,¹ race, and ethnicity as independent from one another in predicting differences in fertility and marriage and finds strong educational, racial, and ethnic gradients in family patterns.² Much of the literature on race and ethnicity, education, and marriage has focused on specific components of the life course, particularly disparities in marital transitions, although there has also been consideration of marital timing and sequencing. Black women marry at lower rates than do White and Latina women, who have relatively similar marital rates (Raley et al. 2015; Sweeney and Phillips 2004), and these gaps persist across all education groups (Aughinbaugh et al. 2013; Raley et al. 2015). Among those who do marry, timing varies: White and Latina women marry earlier than Black women, and higher educational attainment is associated with later age at first marriage (Aughinbaugh et al. 2013). There has also been some consideration of sequencing: if married, Black women experience more marital disruption than do White or Latina women, and White women are more likely to remarry than either Black or Latina women (Raley et al. 2015; Sweeney and Phillips 2004). The risk of marital disruption for both White and Black women decreases as their education levels increase, although the same is not true for Latina women (Raley et al. 2015). Premarital births increase the risk of marital disruption for White women, but not for Black women (Sweeney and Phillips 2004).

The Educational, Racial, and Ethnic Gradients of Fertility Patterns

There are racial and ethnic differences in fertility levels: Latina women have the highest fertility, followed by Black women, whose fertility is slightly higher than that of White women (Aughinbaugh and Sun 2016). Importantly, the negative association between education and fertility is strongest for Black women, followed by Latina and then White women (Aughinbaugh and Sun 2016; Smock and Schwartz 2020). There are also important racial and educational disparities in the timing of childbearing. Black women's childbearing is concentrated earlier in life than White women's, and college-educated women have children later than do women with a high school degree or less (Aughinbaugh and Sun 2016; McLanahan 2004; Smock and Schwartz 2020). Finally, there are racial and educational differences in the sequencing of births: among the college-educated, Black women experience the highest likelihood of unpartnered births (Clarke 2011; Musick et al. 2009).

The Educational, Racial, and Ethnic Gradients of Work Patterns

Much of the research on women's work patterns has focused on disparities in labor force participation, the timing of workforce transitions, and workplace sequencing. Across education levels, women of color face barriers to entry into paid work, lower wages, and more discriminatory practices in the labor market (Branch 2011; Branch and Hanley 2014; Roscigno et al. 2012). Timing matters early in women's work trajectories: non-college-educated Black and Latina women accumulate less work experience at the outset of their careers than do similarly educated White women (Alon and Haberfeld 2007; Willson 2003). Education, race, and ethnicity also influence the sequencing of work: less educated Black and Latina women experience lower LFA and longer spells of nonemployment than do their White peers (Alon and Haberfeld 2007; England et al. 2004; Goldin 2006, 2021; Landivar 2017; Percheski 2008).

Evidence of how education shapes women's work is somewhat mixed. In a cross-sectional study of women's work‒family context between 1976 and 2011, Montez et al. (2014) found that compared with more highly educated peers, White women with a high school degree or less were less likely to be employed, marry, or have children or to combine these roles, while educational differences were narrow among Black women and mixed among Latina women. Moreover, there is some evidence that Latina immigrant women receive fewer occupational returns to increasing education levels (Flippen 2014). Yet England et al. (2004) found that, once they accounted for racial and ethnic differences in educational attainment, college-educated Black and Puerto Rican women work a few more weeks per year than do White women. Studies that focus solely on mothers' labor force participation found that (net of education) White mothers have lower participation than Black and Latina mothers and that the negative relationship between motherhood and employment is weaker for women of color (England et al. 2004; Florian 2018; Landivar 2013, 2017).

Theoretical Expectations

Work and Family Trajectories Across the Life Course

In sum, a growing literature has highlighted the varied trajectories of women's employment across the life course (Aisenbrey and Fasang 2017; Damaske and Frech 2016; Killewald and Zhuo 2019; Lu et al. 2017). Research has consistently found that a majority of women, including mothers, participate in full-time employment across their main childbearing and working years (Damaske and Frech 2016; Killewald and Zhuo 2019)—but also that a substantial plurality of women have lower labor force attachment, with long periods of reduced or no work during prime childbearing and child-rearing years, either with or without a partner (Aisenbrey and Fasang 2017; Damaske and Frech 2016; Weisshaar and Cabello-Hutt 2020). Building on this scholarship, we anticipate four broad patterns of work‒family life courses, two of which are defined by high labor force attachment and two by lower labor force attachment, with motherhood and partnership further defining each pattern: “steady” work with motherhood, both with and without partnership; “steady” work without motherhood, both with and without partnership; “pulled-back” work, which involves spells out of the labor force or working part-time, with both motherhood and partnership; and “interrupted” work, which consists of labor force participation interrupted by repeated spells of unemployment, with motherhood and without partnership (Damaske 2011:15).

• Hypothesis 1a—Steady Work With Motherhood (H1a): We anticipate a steady work and motherhood trajectory characterized by continuous full-time work across the life course combined with motherhood; this trajectory may be further differentiated by the timing, sequencing, and partnership context of motherhood.

• Hypothesis 1b—Steady Work Without Motherhood (H1b): We further expect that there will be at least one trajectory characterized by continuous overwork (>50 hours per week) across the life course without motherhood; this trajectory may be further differentiated by the timing and sequencing of partnership.

• Hypothesis 1c—Pulled-Back Work With Partnered Motherhood (H1c): We anticipate at least two trajectories of lower labor force participation combined with partnered motherhood, including one of low labor force attachment with early partnered motherhood and one of part-time work with delayed partnered motherhood.

• Hypothesis 1d—Interrupted Work With Single Motherhood (H1d): We anticipate one or more trajectories of interrupted work that most likely will be paired with early and single motherhood.

The Role of Education in Work‒Family Trajectories Across the Life Course by Women's Race and Ethnicity

A life course perspective suggests that college education will predict how both work and family events unfold jointly across the life course. An intersectional perspective further complicates this relationship, leading us to anticipate that college education will matter differently for Black, White, and Latina women across both work and family domains.

• Hypothesis 2a—Steady Work With Motherhood (H2a): The association between college education and steady work with motherhood will be positive, but the strength of the association will vary by race and ethnicity and cannot be fully anticipated.

• Hypothesis 2b—Steady Work Without Motherhood (H2b): The association between college education and steady work without motherhood will be positive and will be largest for Black women, especially single Black women.

• Hypothesis 2c—Pulled-Back Work With Partnered Motherhood (H2c): The association between college education and pulled-back work will be negative, but the strength of the association will vary by race and ethnicity and cannot be fully anticipated.

• Hypothesis 2d—Interrupted Work With Single Motherhood (H2d): The association between college education and interrupted work with single motherhood will be negative, and largest for Black women and smallest for Latina women.

Data and Methods

Data and Sample

We use data from the National Longitudinal Study of Youth 1979, which is a nationally representative sample of 12,686 young men and women born between 1957 and 1964 (late baby boomers) who were first surveyed in 1979, when respondents were aged 14–22. This dataset is ideal to model jointly life courses of work and family because it follows an entire cohort of women from young to middle adulthood, allowing us to capture most family events and related work transitions. The NLSY79 initial sample is composed of three independent probability samples: the cross-sectional sample (noninstitutionalized civilians), a set of supplemental samples (oversamples of civilian Hispanic or Latino, Black, and economically disadvantaged non-Blacks/non-Hispanics), and a military sample. Our starting sample size is composed of 6,283 women. We exclude the military sample of women (n = 456), as well as the oversample of economically disadvantaged White women because it was discontinued in 1989 (n = 901). We also exclude women who died (n = 18) or were lost to attrition (n = 39) before age 23. After exclusions, our final sample is composed of 4,869 women: 1,462 Black women, 963 Latina women, and 2,444 White women. To account for the supplemental oversamples and selective attrition, we apply the survey weights provided by the NLSY79 to all descriptive statistics and empirical analysis. We focus on the age range 23–50 because it captures women's prime working age and the years when most partnership and fertility events occur and ensures that most women had completed their education before trajectory measurement.

Missing Data Imputation

We impute missing data using multiple imputations with chained equations (M = 10).³ Our approach combines sequence imputation with multiple imputation so that missing data for both the outcome variables and predictor variables can be imputed jointly (Halpin 2016; see Lu et al. 2017 for a similar approach). For a detailed discussion on the missing data imputation, see the online supplementary materials (appendix A). After imputing the missing data using the mi impute chained command in Stata, we have complete monthly trajectories of work and family as well as their predictors for 4,869 women, which results in a sample size of 58,690 across 10 imputed datasets. We apply the multichannel sequence analysis to the 10 imputed datasets. Then we estimate the multinomial logistic model, pooling estimates across the 10 imputed datasets using Rubin's rules (Rubin 1987).

Work‒Family Trajectories: Multichannel Sequence Analysis and Cluster Analysis

To address our first research question, we use multichannel sequence analysis followed by cluster analysis to identify common patterns of bidimensional life course trajectories, treating work and family as two separate life course domains (Aisenbrey and Fasang 2017; Gauthier et al. 2010).

Step 1: Defining the Work and Family Sequences

The first step of multichannel sequence analysis consists of creating bidimensional life course sequences, one for women's work domain and one for their family domain. We start by defining the possible states of the work and family domains, which are measured from the first calendar month of age 23 to the last calendar month of age 50, resulting in 336 consecutive work and family states for each woman in the sample. Table 1 summarizes the distribution of these states.

Month-by-Month Work Sequences

Studies show that women's varying levels of work intensity have distinct predictors and matter differently for how women reconcile work and family (Damaske and Frech 2016; Killewald and Zhuo 2019; Weisshaar and Cabello-Hutt 2020). Drawing on these findings, we specify nine monthly work states to capture variation in work intensity and distinguish between being out of the labor force and being unemployed (for a similar strategy, see Killewald and Zhuo 2019):⁴not working, that is, not working or unknown whether out of the labor force or unemployed; out of the labor force; unemployed; employed, unknown hours; marginally employed, that is, employed and working fewer than 20 hours per week; part-time, that is, employed and working at least 20 hours but fewer than 35 hours per week; full-time, that is, employed and working between 35 and 49 hours per week; overworked, that is, employed and working 50 or more hours per week; and deceased, that is, data missing owing to respondent's death.

Month-by-Month Family Sequences

Previous scholarship emphasizes the importance of partnership, fertility, timing, and parity in women's work and family lives (Aisenbrey and Fasang 2017; Florian 2018; Kahn et al. 2014; Muller et al. 2020). Therefore, we specify the family states to reflect distinct combinations of partnership status and number of children (recorded in the household roster variable). We distinguish seven family states: single and childless; single and at least one child; partnered and childless; partnered and one child; partnered and two children; partnered and at least 3 children; and deceased, that is, data missing owing to respondent's death.

Step 2: Multichannel Optimal Matching

The next step of sequence analysis is to construct a measure of dissimilarity between these sequences. We do this by calculating multichannel optimal matching distance (OM) (Abbott 1995; Gauthier et al. 2010). Multichannel optimal matching is an extension of optimal matching for a single life course domain, in which the dissimilarity measure between two sequences corresponds to the minimum total costs of transforming a sequence into another, using three possible operations: insertion, deletion, or substitution (Abbott 1995; Gabadinho et al. 2011). Insertion and deletion are assigned a flat cost, which is conventionally known as the indel cost. Substitution costs correspond to the cost of substituting one monthly state for another. In our case, this would correspond to the cost of substituting one work state (e.g., full-time) for another (e.g., overwork). Each possible substitution forms a substitution cost matrix, in which each pair of discrepant states is assigned a cost. The total dissimilarity between two sequences is the sum of these costs across all months in which the sequences differ. Multichannel optimal matching extends classic optimal matching by constructing a dissimilarity matrix for each domain under consideration—here, work and family domains—and computing a final distance matrix by summing the distances of each domain. This final matrix of dissimilarities locally aligns distinct domains' life course trajectories simultaneously.

In our application, we have two life domains: work and family. For the work domain, we construct a 9 × 9 substitution cost matrix, and for the family domain, we construct a 7 × 7 substitution cost matrix. We use constant costs, which means that we assign a substitution cost of 2 to any transformation of one status into another and an indel cost of 1. Optimal matching with a flat cost of 2 is more sensitive to duration (how long women spend in each family and work state) and temporality (when women enter a specific work or family state), and less sensitive to random perturbation, than other dissimilarity measures (Studer and Ritschard 2016). As a robustness check, we replicate the analysis using transition rates as transformation costs and find very consistent results.

Step 3: Cluster Analysis

For the third step of the sequence analysis, we use cluster analysis, based on the final dissimilarity matrix produced in the second step, which allows us to identify the common work‒family patterns in our sample. Following Studer (2013), we compare different numbers of clusters and a range of clustering methods. We first explore visually different cluster solutions (2‒20) using hierarchical clustering (the Ward method). To determine the most appropriate number of clusters, we consider four cutoff criteria of partition quality: the Average Silhouette Width weighted, the Hubert's Gamma, the Point Biserial Correlation, and the Hubert's C (see Hennig and Liao 2013). The cutoff criteria support a solution with eight work‒family clusters (see Figure C2 in the online appendix C). We compare the partition quality of eight-cluster solutions that use three different clustering methods: the Ward method, the partition around medoids (PAM) algorithm, and a combination of the Ward method and the PAM algorithm (Studer 2013). Our final solution is an eight-cluster solution using a combination of the Ward and PAM clustering methods. The sequence analysis and cluster solution were conducted in R using the TraMineR package (Gabadinho et al. 2011) and the WeightedCluster package (Studer 2013).

Predicting Work‒Family Trajectory Membership by College Education and the Measure of Race and Ethnicity

Multinomial Logistic Regression Model and Coarsened Exact Matching

To show how the associations between college education and different work‒family life courses vary by women's race and ethnicity, we combine multinomial logistic regression models with coarsened exact matching (CEM). CEM is a matching procedure that compares an outcome of interest between individuals from two groups who have similar observable characteristics (Blackwell et al. 2009; Iacus et al. 2012). In our application, this translates into comparing college-educated and non-college-educated women who share similar background characteristics. While CEM gets us closer to causality, it is not a magic bullet and does not account for unobserved differences between college-educated and non-college-educated women that may shape both their likelihood of attending college and their subsequent work‒family life courses. Thus, although hereafter for simplicity we use the term “effects,” our results should still be interpreted as associations rather than causal effects.

The CEM procedure involves three steps: (1) selecting and “coarsening” control variables (i.e., transforming continuous variables into categorical ones grouped into few categories); (2) creating strata in which all women have the same values on all control variables; and (3) discarding any observations for which we do not have a comparison group. The CEM procedure generates a weight, which can then be used in a regression to ensure balance in control variables between college-educated and non-college-educated women. CEM offers several advantages, including being compatible with multiply imputed data and nonlinear regression models such as multinomial logistic regression models. In our application, we match women according to early-life individual and family characteristics. The matching procedure results in a final sample of 4,610 matched women out of 4,869 women.⁵

The outcome variable $πio=jπio=J$ corresponds to the odds that a woman i falls into the cluster outcome j as opposed to the baseline cluster outcome J. Our main explanatory variables are education and race and ethnicity, as well as their interaction, which allows for the effects of women's education to vary by race and ethnicity. The regression model is weighted by the product of the CEM weights and sample weights. Because the CEM weights account for all the control variables included in the matching procedure, the multinomial logistic regression models include only birth year as a control variable (⁠$Zi$⁠).

Variables

Table 2 summarizes the proportions of all the covariates.

Race and Ethnicity

We measure race and ethnicity using the NLSY79 race and ethnicity variable, which has the following categories: Black (non-Hispanic-Black), Latina (Hispanic), and White (non-Hispanic non-Black).

Education

Following Alon and Tienda (2005), we measure education (as reported by the respondent at age 23) as a binary indicator that equals 1 for college-educated women (at least one year of college) and 0 for high-school-educated women (12th grade or less).

Control Variables

To establish a clear temporal order for the work‒family trajectories, we use in the CEM procedure only control variables measured at age 23 or before. We measure maternal role modeling using whether the respondent had a working mother at age 14 (1 = mother worked when respondent was 14). We also include the respondent's U.S. nativity (1 = born in the United States), family structure at age 14 (1 = lived with both biological parents at age 14), rural residency at age 14 (1 = lived in a rural area at age 14), number of siblings (1 = 3+ siblings), mother's education (1 = high school or more), father's education (1 = high school or more), and Armed Forces Qualification Test (AFQT) percentile score (1 = bottom quartile, 2 = bottom-middle quartile, 3 = top-middle quartile, 4 = top quartile).

Sensitivity Analysis

To assess the robustness of the multinomial logistic findings, we compare our findings against the following alternative specifications: (1) excluding sample weights; (2) excluding CEM weights and variables; and (3) using a four-year college degree as a cutoff point for education. Our results are broadly consistent with these alternative specifications (see the online supplementary materials, appendix E).

Results

Describing Common Patterns of Women's Work and Family Trajectories

Using sequence and cluster analysis, we identify eight work‒family clusters that fall into the broad patterns of work‒family life courses hypothesized earlier (see Figure 1).⁶

H1a: Steady Work With Motherhood

Three clusters share steady, full-time LFA across the life course but differ in the timing and parity of motherhood, as well as its partnership context.

Working and partnered early mothers (18% of the sample) enter parenthood (coupled with partnership) early and have, on average, about two children. Women in this cluster experience steady full-time employment from ages 23 to 50, spending more than 71% of the months observed working at least full-time and only 14% not working.

Working and partnered later mothers (9% of the sample) differ from the preceding group in two important respects: they delay both partnership and motherhood, and most have only one child. On average, they spend only 16% of the observation months (from ages 23 to 50) not employed and 69% working at least full-time.

Similarly, working and single mothers (17% of the sample) spend 68% of the months between ages 23 and 50 working full-time and only 13% not working. But for most of that time they are single with children living in their household; while most enter a partnership at some point in time, these partnerships are short-lived.

H1b: Steady Work Without Motherhood

Two clusters share steady, full-time LFA without children. They differ in the intensity of their work hours and their partnership status.

Working and partnered women (9% of the sample) partner later, and at age 50 most are childless; all display continuous full-time employment from early adulthood to midlife.

Overworked and single women (11% of the sample) remain childless and, in general, unpartnered up to age 50. Employed full-time and continuously from early to midlife, they work the longest hours and are most likely to be overworked, spending 18% of the observation window working more than 50 hours per week.

H1c: Pulled-Back Work With Partnered Motherhood

Two clusters exhibit pulled-back work with partnered motherhood; they differ in LFA timing and intensity, as well as in the timing and parity of partnered motherhood.

Stay-at-home and partnered mothers (16% of the sample) have an early onset and high level of fertility: a large majority have at least three children, most born within a partnership. In line with an early onset of fertility, we also observe children leaving the household when the women enter their 40s. These women spend, on average, 33% of their time between ages 23 and 50 out of the paid workforce, usually synchronizing their fertility transitions with their labor force participation. While there is some heterogeneity in women's work trajectories, some women in this cluster start to work as children start leaving the home.

Part-time working and partnered mothers (10% of the sample) average only two children and experience a later onset of partnered fertility, concentrated in their late 20s. While they differ somewhat in LFA, they are the women most likely to work part-time. Again, their work pattern appears closely associated with childbearing, with a higher rate of withdrawal from full-time work between ages 23 and 30 and a return to part-time work above age 35.

H1d: Interrupted Work With Single Motherhood

Single mothers with interrupted work (10% of the sample) bear children early and have low LFA. By age 23, three quarters of women in this cluster are already mothers. While they differ somewhat in partnership status, most are single mothers, and their partnerships are short-lived. Again, because of their early onset of fertility, we observe children moving out of the household when women are in their 40s. These women are out of the labor force much of the time, with short spells of employment, and they are two to three times as likely to be unemployed as women in any other work‒family cluster.

How Does College Education Shape Women's Work and Family Trajectories by Race and Ethnicity?

Hypotheses 2a–2d specify how college education is associated with each expected pattern of work‒family life course by women's race and ethnicity. In the following, we first describe how we interpret interactions in nonlinear models, and then focus on those associations.

We apply recent methodological recommendations on effect heterogeneity in nonlinear models to address whether and how the effects of college education on women's work‒family trajectories vary by race and ethnicity (Bloome and Ang 2020; Mize 2019; Mize et al. 2019). We present all results on the probability scale.⁷ We first present the average predicted probabilities of each work‒family cluster for the six combinations of the race and ethnicity variable and the college education variable (Figure 2). We then show the average marginal effects (AMEs) of college education by race and ethnicity for each work‒family cluster, and whether these effects differ by race and ethnicity (contrast of margins) (Table 3). All average predicted probabilities and AMEs are calculated for each woman in the sample and averaged out by race and ethnicity.⁸

H2a: Steady Work With Motherhood

The association between college education and steady work is inconsistent: it is positive for partnered women, but negative for unpartnered ones, and it varies by race and ethnicity. In the working and partnered clusters of both early and later mothers, the AMEs of college education are positive and statistically significant for Black women only and differ from those of White (but not Latina) women. In contrast, the AMEs of college education for working and single mothers are moderate and negative for all women in this group and reach statistical significance for Black and White (but not Latina) women.

H2b: Steady Work Without Motherhood

The association between college education and steady work without motherhood is positive and varies by race and ethnicity only among single women. In contrast to our theoretical expectations, we find null and nonsignificant associations between college education and the probability of a working and partnered but childless life course for all women, with no variation across race and ethnicity. In striking contrast, the association between college education and being overworked and single is positive and statistically significant for all women, but the AME is three times as great for Black women as for White women.

H2c: Pulled-Back Work With Partnered Motherhood

In contrast with our hypothesis, the associations between college education and pulled-back work with partnered motherhood are mostly not statistically significant and do not vary by race and ethnicity. We find one exception: the AME of college education for White women in the part-time work and partnered mothers cluster is positive and statistically significant.

H2d: Interrupted Work With Single Motherhood

The association between college education and being in this work‒family cluster is negative and statistically significant for all women but varies in effect size by race and ethnicity. The negative effect of college education is five times as large for Black women as for White women and twice as large as for Latina women, and these differences are statistically significant.

The Role of Race and Ethnicity

Because the foregoing findings suggest an enduring role of race and ethnicity, we also present the AMEs of race and ethnicity by college education for each work‒family cluster, and whether these effects differ by education (contrast of margins) (Table 4).

Steady Work With Motherhood

The racial and ethnic differences in steady work with motherhood depend on the partnership context. Among partnered women, the stronger educational gradient for Black women results in small and not statistically significant racial and ethnic differences among college-educated women. Among non-college-educated women, racial gaps persist; Black women are less likely to experience steady work with partnered motherhood than either Latina or White women. Among unpartnered women, whether college educated or not, Black women are most likely to combine steady work with motherhood, followed by Latina women and White women.

Steady Work Without Motherhood

In the two clusters of women who work steadily without motherhood, the statistically significant racial and ethnic differences center on Black‒White differences and vary by partnership context. Partnered Black women are less likely than partnered White women to be working, across education levels. Only among unpartnered, college-educated women are Black women more likely than White women to be overworked.

Pulled-Back Work With Partnered Motherhood

Stay-at-home and partnered mothers are in general less likely to be Black, but this racial and ethnic difference reaches statistical significance only for Black‒Latina differences among non-college-educated women and Black‒White differences among college-educated women. In contrast, across education levels, White women are consistently most likely to be in the part-time work and partnered mothers cluster, followed by Latina women and Black women.

Interrupted Work and Single Motherhood

Across college education, Black women are most likely to be in this cluster, followed by Latina women and White women. Racial and ethnic differences persist across education groups in this work‒family cluster. Yet the larger educational gradients for Black women (and to a lesser degree for Latina women) in this cluster lead to smaller racial and ethnic differences among college-educated women.

Discussion

Younger baby boomers (like those in this study) entered the labor market more than three decades ago, and most women in this cohort have worked continuously and full-time. But race and ethnicity and college education have intersected in heterogeneous ways to shape work and family trajectories across the life course. While our results show statistically significant and often sizable education effects across racial and ethnic groups for most of the work‒family clusters, they also suggest that the size and direction of the returns of education vary widely across groups. There are two clusters in which the effects of education are significant and consistent for all three racial and ethnic groups: college education reduces all three groups' likelihood of being single mothers with interrupted work and increases their likelihood of being single with steady overwork. But significant disparities by race and ethnicity remain in these clusters even with the benefit of college education. Moreover, there are two clusters for which the effects of education are null across racial and ethnic groups, but for which, instead, we find evidence of persistent and statistically significant racial and ethnic differences among both college-educated and non-college-educated women. In the following, we discuss the implications of our findings for each racial and ethnic group.

Educational attainment plays an outsize role in shaping Black women's lives. Our findings suggest that structural constraints shape the work‒family trajectories of Black women more than they do for other women. Black women are widely overrepresented in clusters that are characterized by either interrupted or full-time work and single motherhood, and are underrepresented in clusters that combine steady employment with marriage and childbearing. These findings are suggestive evidence of the consequential historical exclusion of Black Americans from the marriage institution and employment structures (Higginbotham 2001; Hill 2003) and the continued racism in marriage and labor markets (Clarke 2011; Ray 2019; Williams 2019). Yet returns to education narrow some of the racial and ethnic disparities. College education increases Black women's likelihood of being in both the working and partnered early mothers cluster and the working and partnered later mothers cluster and reduces the likelihood that they will experience interrupted work and single motherhood. College education also increases Black women's overrepresentation in the overworked singles cluster. Overall, our research suggests that college education increases Black women's participation in most pathways that involve steady work. In contrast to previous research, we also find that college education increases their participation in pathways that combine steady work with stable partnerships, narrowing racial and ethnic gaps for the college educated.

The persistent large racial and ethnic disparities in most of the clusters despite the beneficial returns to college education for Black women may be explained by existing research. College-educated Black women face additional burdens when combining work and motherhood, such as knowing that they must combat structural racism as they parent (Dow 2019), that their spouses and communities may need their involvement in ways that are not asked of women of other races (Barnes 2016), and that their spouses may face greater barriers to continuous employment and to higher wages, which may shape their ability to coparent (Glauber 2007; Gonalons-Pons et al. 2021). Institutional constraints in Black women's family lives (see Cross 2023), the lack of family-supportive policies in the United States, and how racialized organizations institutionalize and reify existing racial inequalities (e.g., Ray 2019) are likely to explain these differences.

Latina women's work‒family trajectories appear to be less responsive to their educational attainment, with college education affecting only their likelihood of overwork and singlehood or interrupted work and single motherhood. This finding aligns with and extends Flippen and Parrado's (2015) finding that immigrant Mexican women's work trajectories are responsive to their educational attainment in Mexico, but not once the women are in the United States. However, this does not mean that theories of Latina familism can adequately explain why Latina women's work‒family pathways appear less responsive to college education. While previous findings have emphasized motherhood as a consequential factor affecting Latina women's employment (e.g., Alon and Tienda 2005; England et al. 2004), our typology of trajectories draws a more complex picture. In our sample, Latina mothers are more likely to experience early partnered parenthood, but with either steady nonemployment (stay-at-home and partnered mothers) or steady employment (working and partnered early mothers). And Latina women in these two work‒family clusters hold very different levels of resources. The Latina stay-at-home and partnered mothers are the most socioeconomically vulnerable group in our sample. In contrast, the Latina working and partnered early mothers are more advantaged than the average Latina woman in our sample.

White women's work‒family patterns appear more responsive to their educational attainment than do Latina women's patterns, although differently so from those of Black women. Both college-educated and non-college-educated White women are disproportionately likely to be partnered. They are overrepresented in clusters that combine partnership and motherhood with either no or part-time employment, the pattern that is widely believed to allow for “balance” between work and family (see Stone and Lovejoy 2019); college education increases their participation in the part-time work cluster and increases racial and ethnic disparities among the college educated in this cluster. Moreover, the women in this cluster enter parenthood later, most often after a period of full-time employment, a finding that supports Stone and Lovejoy's (2019) contention that patterns of “opting out” followed by “opting back in” constitute a “paradox of privilege”: college-educated women (particularly White women) benefit from normative breadwinner‒homemaker models that privilege breadwinner husbands' earning potential and decrease women's investment in their own careers. College education also decreases the likelihood that White women will be working and single mothers, a group that exhibits large race and ethnicity disparities among both the college-educated and non-college-educated. We cannot disentangle whether White women combine work and family in distinct ways because of constraints or preferences (indeed, there is reason to suspect that both may come into play; see Damaske 2011); future research should further investigate this issue.

This study has two major limitations. First, our findings are specific to younger baby boomers and cannot be fully generalized to the work‒family experience of more recent cohorts of American women. Still, longitudinal data are necessary to understand the interplay between work and family over the life course, and many of our findings are likely to remain relevant for younger cohorts. Future work should explore generational differences, but social class, as measured by education, continues to be predictive of family formation trends (Ruggles 2015) and economic outcomes (Pfeffer 2018). Also, given the many institutional barriers to marriage (Subramanian et al. 2018) and persistent racism in labor markets (Quillian et al. 2017), it is unlikely that the structural barriers faced by Black women in our sample would be much different today. Similarly, our temporally limited data cannot speak directly to the work‒family experience of more recent waves of immigrant women and younger cohorts of Asian and Latina native-born women, but some of our key findings appear to be echoed in recent work using data from newer immigration waves. For example, Lu and colleagues (2017), using the 1996–2008 panels of the Survey of Income and Program Participation, found that native women of color and long-residing immigrants exhibit stronger LFA after childbirth than do White women. Taken together, recent research suggests that we need more refined measures of racial and ethnic identities, as well as immigrant status, to understand access to different work‒family life courses.

A second limitation is that we had to simplify the empirical analysis to strike the right balance between parsimony and complexity. Thus our control variables could not include time-varying variables (which would have changed simultaneously with the work and family sequences). We also simplified some aspects of women's family lives by not distinguishing between cohabitation and marriage. These simplifications allowed us to obtain a more conservative number of clusters and avoid overfitting our data. Several robustness checks showed that different sets of control variables and a distinction between marriage and cohabitation did not produce substantially different results.

Our study draws a complex picture of the distinctive roles played by education, race, and ethnicity in shaping the work‒family lives of women across the life course. Racialized returns to education, combined with persistent racial and ethnic gaps across the education distribution, produce unequal patterns in work‒family strategies among Black, Latina, and White women. While Black women are widely underrepresented in work‒family trajectories that combine steady employment with partnership and childbearing, college education does appear to narrow these racial and ethnic disparities. Latina women's work and family trajectories appear less responsive to their educational attainment; for them, partnership and motherhood may be commonly paired either with steady employment or with exits from the labor market. Finally, part-time employment to reconcile work with partnered motherhood is a strategy adopted almost exclusively by White women, especially those with higher education. Overall, our results highlight the importance of combining an intersectional lens with a life course approach. The dichotomous patterns for Black and White women, in particular, suggest that college education may result not only in access to steady employment opportunities over the life course, but also in access to trajectories that privilege women's partnerships and childbearing.

Acknowledgments

This work has benefited from useful discussions with Nicola Barban, Diederik Boertien, Jonathan Daw, Diane Felmlee, Jennifer Glick, Valarie King, Katherine E. Maich, Molly Martin, Sarah E. Patterson, Susana Quirós, Maria Sironi, Ashton Verdery, Sander Wagner, and participants at the 2018 annual meeting of the Population Association of America, the PRI Gender Working Group, and the CREST Sociology Seminars. We thank Nancy Mann for her superb copyediting work. The authors gratefully acknowledge funding from the Eunice Kennedy Shriver National Institute of Child Health and Human Development to the Population Research Institute at The Pennsylvania State University for Population Research Infrastructure (P2CHD04102) and Family Demography Training (T-32HD007514) grants.

Notes

References