Most literature on female employment focuses on the intersection between women’s labor supply and family events such as marriage, divorce, or childbearing. Even when using longitudinal data and methods, most studies estimate average net effects over time and assume homogeneity among women. Less is known about diversity in women’s cumulative work patterns over the long run. Using group-based trajectory analysis, I model the employment trajectories of early Baby Boom women in the United States from ages 20 to 54. I find that women in this cohort can be classified in four ideal-type groups: those who were consistently detached from the labor force (21 %), those who gradually increased their market attachment (27 %), those who worked intensely in young adulthood but dropped out of the workforce after midlife (13 %), and those who were steadily employed across midlife (40 %). I then explore a variety of traits associated with membership in each of these groups. I find that (1) the timing of family events (marriage, fertility) helps to distinguish between groups with weak or strong attachment to the labor force in early adulthood; (2) external constraints (workplace discrimination, husband’s opposition to wife’s work, ill health) explain membership in groups that experienced work trajectory reversals; and (3) individual preferences influence labor supply across women’s life course. This analysis reveals a high degree of complexity in women’s lifetime working patterns, highlighting the need to understand women’s labor supply as a fluid process.
Structural constraints, like patterns of socialization, define the limits of human action; but within these limits, a range of human responses is possible.Kathleen Gerson, Hard Choices
Forty years into the gender revolution, we know a great deal about the correlates of women’s employment behavior, particularly at the intersection of work and family. Starting in the 1970s, American women—including mothers—greatly increased their labor supply, experiencing fewer and shorter work interruptions, as fertility declined and social attitudes shifted in favor of their participation in paid work (Joesch 1994; Rindfuss et al. 1996). Many studies have investigated short-term effects of work or family transitions: for example, most of the motherhood wage penalty literature (cf. Avellar and Smock 2003; Budig and England 2001; Waldfogel 1998) follows women for a period of about 10 years, averaging effects at either side of the transition to motherhood and de facto assuming a constant effect of childbearing on wages. Research exploring outcomes longitudinally (Joshi 2002; Kahn et al. 2014; Phipps et al. 2001; Rake 2000; Sigle-Rushton and Waldfogel 2007) has traditionally taken a variable-centered approach, emphasizing net associations between variables. This approach is helpful for estimating net effects between variables, but it can obscure the cumulative and reciprocal consequences of life events in context as interactive experiences that shape trajectories and create long-term dependencies. In this article, I provide a long-term, person-centered exploration of the ideal-type employment trajectories of a cohort of American women, summarizing their actual work and family experiences from ages 20 to 54. I investigate whether those trajectories were characterized by continuity or discontinuity, and why.
Female labor supply is adaptable, adjusting over the life course as women juggle work with other aspects of their lives. For a significant number of women (and an increasing number of men), work is not a continuous uninterrupted status but rather an activity that can be put on hold temporarily in order to prioritize other life pursuits, such as raising a family. Accounting for this complexity calls for both longitudinal data and a method that is sensitive to the timing of events, the path dependence they impose, and the possibility of reversals.
Previous research, using a variable-centered approach, has sought to reduce complexity by estimating the net impact of certain predictors (such as age, education, marriage, or fertility) on women’s employment: its implicit goal—not always achievable because of design, measurement, or data limitations—has been to reveal causal relationships among variables. I take a complementary, person-centered approach, seeking to reduce complexity by identifying patterns of typicality in women’s employment experiences over the life course. The main goal here is not causal but descriptive: to explore patterns of convergence in women’s long-term labor supply, and to relate these to a series of observable traits. In other words, instead of asking the variable-centered question of how women’s traits combine and compete to explain variation in their employment outcomes, I ask three interrelated person-centered questions aimed at unveiling heterogeneity in women’s employment experiences. First, are some working trajectories observed with higher typicality than others? Second, do these trajectories reveal a pattern of continuity or discontinuity? And finally, how did the actual lives and experiences of women in different trajectory groups differ?
Answering these questions requires a very long window of observation, spanning women’s entire working lives. In recent years, the older members of the Baby Boom cohort (born between 1946 and 1964) reached retirement age, completing their working careers. Women in this cohort pioneered some of the most important processes underpinning the gender revolution, such as the rise in female employment rates and the decline and postponement of fertility. They became more attached to the labor force, and their employment profiles became increasingly similar to those of men (Spain and Bianchi 1996). They are also the first generation for which rich longitudinal data are available, including educational, professional, fertility, and relationship histories.
Using data from the National Longitudinal Survey of Young Women, NLS-YW (a cohort born between 1944 and 1954, and interviewed between 1968 and 2003) and group-based trajectory analysis (Jung and Wickrama 2008; Muthén and Muthén 2000), I explore the working profiles of early Baby Boomers between the ages of 20 and 54, and find that their employment experiences can be summarized in four ideal-type groups (group size): consistently detached (21 %), increasingly attached (27 %), increasingly detached (13 %), and consistently attached (40 %). I then explore the association between membership in these groups and a variety of factors, including some found most frequently in qualitative studies—for example, work preferences, feelings of discrimination, job satisfaction, attitudes toward mothering, and external constraints (own health, health of family members, and husbands’ lack of support). Thus, I provide a novel empirical assessment of prior—mostly qualitative—studies of women’s work–family preferences, experiences, and narratives (cf. Blair-Loy 2006; Damaske 2011; Gerson 1986).
Three Approaches to Women’s Employment Behavior
Theoretical approaches to women’s employment behavior can be grouped in three streams. First, gender socialization theories—also known as theories of gendered selves—emphasize the role of internalized attitudes and preferences in the everyday process of “doing gender” (Butler 1988; Chodorow 1999; West and Zimmerman 1987). According to these theories, women learn gender during childhood and adolescence by internalizing, from their social milieu, feminine traits, gender-typed attitudes, and behavioral predispositions (such as nurturing and relational skills, communal thinking, or empathy).
Second, structural theories of gender inequality highlight the factors that impinge on women’s ability to choose and act freely: women’s behavior is influenced by their location at the intersection of existing social structures, such as the gender system and the labor market. Most of the outcomes that are often attributed to behavioral or attitudinal differences between men and women are, according to these theories, a reflection of (1) the structural positions they occupy as a result of their race, sex, class, and so forth; and (2) the ways in which society allocates market rewards across the social structure (Reskin 1988). These theories emphasize that when subjected to comparable social conditions (e.g., similar expectations about domestic roles or parenting norms) and given equal access to opportunities (e.g., professional mentorship, opportunities for promotion), most men and women behave in largely similar ways (Kanter 1977; Risman 1998).
Third, the developmental approach constitutes a sort of middle ground between socialization and structural theories. According to this view, agency and structure, choice and constraint, are interwoven such that social action is a blend of constrained choice and chosen constraint. On the one hand, subjective attitudes, preferences, expectations, and internalized gender roles provide a normative framework that lends meaning to the available alternatives for action. On the other hand, institutions and structures limit, in quantity or quality, the available options that an individual is presented, depending on her location within economic, social, political, and gender systems. This hybrid approach has become the standard in the qualitative literature seeking to understand women’s work–family narratives and normative schemes (Blair-Loy 2006; Damaske 2011; Gerson 1986). Because it combines elements from socialization and structural theories, I consider this the better approach to account for the complexity and fluidity of women’s market behavior across the life course.
Pulls and Pushes, Structure and Agency
Women’s work and family experiences are fluid. In her landmark study of women’s employment trajectories, Gerson (1986) found that most of her interviewees (a sample of 63 women aged 27 to 37 when interviewed in 1978 and 1979) reached adolescence with a dominant preference for their adult lives: 45 % of them wanted to work for pay, while the others planned an adult life centered on motherhood and homemaking. However, as these women recounted their experiences, four distinct groups emerged, depending on whether they had fulfilled their teenage plans. Of those who grew up expecting to work for pay, only 40 % managed to be employed consistently. Among those expecting to stay at home, only one-third did just that, and two-thirds ended up working for pay.
From these narratives, Gerson identified a series of forces pulling women toward the market or pushing them toward domesticity. Regardless of their intentions earlier in life, women who ended up working for pay were more likely to report high levels of job satisfaction, to narrate episodes of financial strain, or to have experienced marital instability. Conversely, women who ended up focusing on motherhood were more likely to think of childcare as a rewarding activity; have achieved some financial security and partnership stability; and say that they had felt ill-treated or discriminated against at work (Gerson 1986).
This variability between women’s stated preferences and their employment trajectories illustrates the tension between gender socialization and structural location. Risman (1998) used longitudinal data from the Washington State Career Development Study to predict married women’s work hours, from their high school graduation to their early 30s. She found evidence for both socialization and structural mechanisms. On the one hand, early preferences for work or family were significant predictors of future work intensity; on the other hand, adult experiences such as childbearing and women’s own earning potential were also associated with midlife employment. Socialization and structural forces are arguably behind differences in women’s ability to fulfill their work–family preferences across countries with distinct welfare regimes and levels of public support for working mothers (Gash 2008; Yerkes 2013).
More recently, England (2010) combined structural and socialization explanations to reassess progress in gender equality in the United States, describing the gender revolution as “uneven and stalled,” particularly after the mid-1990s. This description generated controversy among feminist scholars opposed to the proposition that gender inequality originates from women’s own choices (cf. Bergmann 2010; McCall 2011; Reskin and Maroto 2010). England’s response to her critics captures well the position I take in this article (England 2011:116–117):
Among sociologists of gender . . . aversion exists to explanations that assert a causal role for socialized preferences on the supply side of labor markets. I respectfully disagree; I believe that continuous gendered socialization affects taken-for-granted assumptions (e.g., which jobs we even consider), identities, and preferences. Outside social forces change our insides. Rather than eschewing socialization explanations in fear that they will be used to blame the victim, I believe we should point out that people did not choose the constraining social forces that formed their preferences, identities, and assumptions . . . and that even if they chose their jobs, they were not always aware of and certainly do not prefer the low pay in those jobs.
Life Course Theory: Events in Context
The life course is a sequence of socially defined events and roles that the individual enacts over time (Giele and Elder 1998). Life course theory treats events as transitions that modify people’s statuses, identities, or roles (Quadagno 2007); as such, events are characterized by their timing, duration, spacing, and sequencing. For instance, the effects of childbearing on women’s employment is moderated by the age of the mother (timing), the time elapsed since last birth (duration), the position and ordering of this and other children (spacing), and whether motherhood precedes or follows marriage (sequencing). In the long haul, some life events can be reverted, but most carry consequences even after they are “undone.” For example, the loss of a child cannot be equated with childlessness, nor can divorce be equated with singlehood—even though, statically, they are characterized by the absence of children or a husband, respectively.
The lasting effect of events is best summarized in trajectories, which are pathways that emerge over the life course with some typicality (Hynes and Clarkberg 2005). As such, trajectories not only summarize the cumulative experiences of a group of individuals with respect to some observable outcome; when correctly identified, they can also unveil the mechanisms by which events combine to generate an outcome, reveal underlying processes of cumulative disadvantage, and illustrate the long-term consequences of events for people in different structural locations.
In this article, I do not systematically evaluate the three theoretical perspectives presented earlier. Instead, I draw on them because they point to the relevant factors shaping women’s employment experiences, providing the background for my empirical work. I also use these theories to show how person-centered approaches, by unveiling heterogeneity among women and across the life course, can lend support to these theoretical approaches—at least partially—for some subgroups, at some stages in life.
Moderators of the Relationship Between Fertility and Employment
Bearing children is the strongest single factor depressing women’s labor force participation (Spain and Bianchi 1996) and lifetime earnings (Sigle-Rushton and Waldfogel 2007). For adult women, most of the time not spent in the labor force (or sleeping or engaged in self-care) is devoted to domestic work (childcare and housework), which increases significantly with the arrival of offspring (Bianchi 2000). The impact of fertility on employment is moderated by a number of factors.
Life Course Moderators
Early childbearing often involves the disruption of formal education (Hofferth et al. 2001) and impinges on human capital accumulation, affecting work experience and wages as early as the mid-20s (Klepinger et al. 1999). As a result, early mothers pay a larger wage penalty than other women (Taniguchi 1999). Longer work interruptions increase the chances of downward occupational mobility (Aisenbrey et al. 2009) and reduce earnings (Baum 2002; Phipps et al. 2001). These negative effects can be perceived almost two decades after the birth of a child (Shapiro and Mott 1994), particularly for women with three of more children (Kahn et al. 2014). The sequencing of marriage and motherhood has had shifting effects over time: in the late 1970s, single mothers were marginally more likely to be employed than married mothers, but this relationship reversed in the 1980s and converged again in the late 1990s (Cohen and Bianchi 1999).
In short, the life course literature suggests that the timing, sequencing, and duration of events should be particularly important in early adulthood because the simultaneous launching of professional and family careers results in a time squeeze that hinders human capital investments (Clarkberg and Moen 2001; Milkie et al. 2004). I expect early marriage, early and out-of-wedlock first births, and higher completed fertility to be associated with a slow start in women’s careers.
Work and family experiences can either pull women toward the market or push them toward domesticity. Discrimination is frequently invoked to explain differences in market performance between mothers and childless women (Budig and England 2001; Waldfogel 1997). These effects, which are difficult to test in observational studies, have been confirmed in semi-experimental designs (Correll et al. 2007).
Husbands’ gender attitudes might act as an external constraint, influencing the distribution of household chores (Baxter et al. 2008; Bianchi et al. 2000; Evertsson 2014). Women married to husbands with liberal gender attitudes in the mid-1980s worked at higher rates than other women (Smith 1985), but both husbands’ attitudes and wives’ employment behaviors adjusted early in most marriages to minimize conflict (Spitze and Waite 1981); where they didn’t, they increased marital dissatisfaction and divorce (Amato and Booth 1995). Although the direction of the link between marital breakup and female employment is unclear (Greenstein 1990; Sayer and Bianchi 2000; Schoen et al. 2002), married women often exhibit lower employment rates than their divorced peers (Drobnic et al. 1999; Jeon 2008; Smock et al. 1999). Personal health limitations as well as those of relatives can hinder labor force participation. Beyond the nuclear family, care demands from parents and grandchildren increase over time, particularly for women, who later in life take up caring obligations more often than men do (Kahn et al. 2011).
Thus, I hypothesize that negative workplace experiences, lack of support from husbands, and health limitations will push women toward nonmarket activities, whereas divorce will pull them toward paid work.
Socialization, Attitudes, and Preferences
Individuals grow up to develop different work and family orientations, which shape employment behaviors (Hakim 2002). Work expectations are associated with human capital accumulation and post-school occupational decisions (Shaw and Shapiro 1987). Young adult women who expect to be loosely attached to the workforce are employed at lower rates, earn lower wages, and reach less-prestigious occupations at midlife than those with strong work expectations (García-Manglano 2012).
However, preferences are unlikely to remain fixed over time: women adjust their work hours in response to marriage and childbearing (Drago et al. 2006), and some weaken their commitment to work during the first years after giving birth (Evertsson 2013). In practice, preferences are fluid in two ways. First, women might want different things for different life stages, as in the case of those who plan to drop out of the labor force after they have children. Second, preferences may change as a result of actual life experiences if, for instance, women find paid work less satisfying than expected. Not surprisingly, women who report negative work experiences and job dissatisfaction are more likely to leave the labor force, and to stay out of it, after childbearing (Böckerman and Ilmakunnas 2009; McRae 1993; Stone 2007).
I expect that women’s work and family preferences will influence both levels and trends of labor force attachment: dissatisfaction with work will push women toward domesticity, and dissatisfaction with mothering toward higher market involvement.
Other Sociodemographic Moderators
More-educated women tend to be more strongly attached to paid work (Goldin 2006; Goldin and Katz 2008). Women married to men at the bottom and top of the income distribution are less likely to be employed than those married to men with average incomes (Cotter et al. 2007). The association between race and employment is unclear: women of racial minority groups have long exhibited higher employment rates than white women, but this has changed in recent decades (England et al. 2004).
Person-Centered Approaches to Longitudinal Data
The growing availability of longitudinal data has boosted interest in person-centered approaches, which identify groups of individuals sharing particular attributes, manifested in diverse developmental trajectories. Alternatively, variable-centered approaches describe associations between variables, seeking to estimate their contribution toward an outcome of interest (Laursen and Hoff 2006; Stemmler 2014). Both variable- and person-centered approaches seek to understand reality by reducing complexity, but they do this in fundamentally different—albeit complementary—ways. Variable-centered approaches reduce complexity by estimating net relationships between variables: they explore the ways in which a set of independent variables compete or combine to explain variation in a response variable. After complexity has been reduced (i.e., post-estimation), variable-centered approaches allow for the re-creation of complex scenarios through prediction. Predicted outcomes can be calculated for a virtually infinite number of ideal-type individuals, resulting from any combination of traits: “amounts” for continuous variables, and “categories” for discrete variables.
Alternatively, person-centered approaches reduce complexity by seeking commonalities in the experiences of individuals with respect to a given outcome of interest. In other words, they explore whether certain outcomes can be observed in a given population with some typicality. To achieve this, they sort people into ideal-type clusters: within group, similarities with respect to the outcome of interest are maximized; and across groups, similarities are minimized (i.e., differences are maximized). With longitudinal data, these methods group people with similar long-term outcomes. When complexity has been reduced through the identification of ideal-type trajectories for a given outcome, complex scenarios can be re-created—in this case, by estimating whether membership in different trajectory groups is associated with particular experiences or attributes.
Group-Based Trajectory Analysis
Group-based trajectory analysis (GBTA; also known as latent class growth analysis) is a person-based method of analysis suited for the exploration of developmental outcomes over time (Jung and Wickrama 2008; Nagin 1999, 2009). It belongs to the family of mixture models, which allow for the identification of distinct subpopulations within a population. GBTA is widely used in the fields of criminology and psychology. In recent years, demographers have used GBTA to model fertility preferences across the life course (Hayford 2009); women’s market behavior in the months immediately before and after having a child (Hynes and Clarkberg 2005); and men’s pathways into marriage, fatherhood, and employment (Dariotis et al. 2011).
Bayesian and Akaike information criteria (BIC and AIC, respectively) can be used to assess model fit under different functional forms and numbers of groups. Likelihood ratio tests are also available for the comparison of nested GBTA models, differing only in the number of groups. After complexity has been reduced by the estimation of ideal-type trajectory groups and the allocation of individuals into these groups, multinomial logistic regression is used to estimate the association between trajectory groups and their relevant covariates.
Time-Varying and Time-Invariant Covariates in GBTA
It is important to understand the role played by time-varying and time-invariant covariates in GBTA. As a person-centered approach, GBTA seeks a twofold objective: its primary goal is to place individuals into a parsimonious set of ideal-type subpopulations, relative to some outcome of interest; its secondary goal is to understand what led people to follow a particular trajectory, exploring differences between people across groups. This distinction is key given that the primary goal is defined across time (stylizing a time-varying outcome into representative subgroups), whereas the secondary objective is static (explaining membership into this or that trajectory group). Hence, variables can enter a GBTA model in two ways: (1) as time-varying covariates, they modify the trajectory components—such as intercept and slope—directly shaping the longitudinal outcome of interest (primary goal); and (2) as time-invariant covariates, they predict static trajectory group membership (secondary goal).
In this article, the outcome of interest is employment, which varies across time. My primary goal is to summarize women’s lifetime labor supply into ideal-type working trajectories. Thus, only employment is entered as a time-varying covariate, which might seem counterintuitive given that these are panel data. However, using time-varying covariates would substantially redefine the outcome and modify the nature of the analysis. This is how the outcome would be modified: if time-varying measures of education, fertility, and income were included, GBTA would produce “employment-net-of-education-and-fertility-and-income” trajectories, making their interpretation fuzzy. This would also modify the nature of the analysis, making it variable-centered—that is, one that aims at estimating net effects between covariates.
For these reasons, I enter independent variables as time-invariant covariates; they summarize traits, experiences, attitudes, and behaviors that spanned women’s adult life course.3 Despite this limitation, time-invariant covariates can be time-sensitive: they can be defined to incorporate temporality across the life course. For example, “timing of first birth” (categorical) will be coded as “childless; teen mother; early 20s; late 20s,” providing temporal information on a woman’s life course experience in this realm.
Data and Measures
Data and Sample
I use all 22 waves from the National Longitudinal Survey of Young Women (NLS-YW), between 1968 and 2003. The NLS-YW includes information on 5,159 women born between 1944 and 1954 and first interviewed at 14 to 24 years old. The survey was discontinued in 2003, with respondents aged 49 to 59. NLS-YW data are well suited for a longitudinal exploration of women’s employment outcomes because they include women’s complete educational, partnership, fertility, and employment histories. Additional information is available on women’s subjective feelings of discrimination, job satisfaction, attitudes toward childcare, and work–family orientations (the latter only from baseline to age 35).
To minimize the impact of teenage employment (possibly part-time or during the summer) and early retirement (which might respond to various motivations and circumstances), I focus on women between the ages of 20 and 54, resulting in the loss of 199 women who attrited in their teens.4 Of the remaining 4,960 cohort members, 28 never provided information about one or more employment covariates, leaving an analytic sample of 4,932 early Baby Boomers—more than 95 % of the original sample. GBTA allows each woman to contribute to the analysis during the years in which she provides valid data, even if she is lost to attrition in later waves. In the NLS-YW, retention rates are relatively high, with 94 % of women contributing at least five (and 83 % at least 10) waves of data.
Main Variables: Employment and Time
The modeling of longitudinal trajectories using GBTA rests on two variables, which constitute the building blocks of growth curves: a time-varying outcome and a time variable. The outcome here is employment (dummy variable, with 1 = employed), available at each survey year. Women are considered employed if when asked, “What were you doing most of last week?,” they responded they were working for pay or were self-employed, part-time or full-time; women with a job but who didn’t work in the previous week (e.g., on maternity leave) were coded as employed. Time is measured in years of age. The analytic data set is structured in the wide format: one record per woman, including age and employment at each survey year, followed by the independent variables.
Life course events. The temporal dimension of family events is captured by a categorical variable for first birth timing (childless, teen mother, early 20s, and late 20 or later; this cohort’s median age at first birth falls in the early 20s, earlier than that of more recent cohorts), as well as four dummy variables to characterize other family experiences: (1) women who married before age 25; (2) those who never married; (3) unmarried first motherhood, if first birth happened before first marriage; and (4) women who had three or more children (which, in addition to high fertility, serves as a proxy for the duration of the childrearing phase: that is, spending many years with young children at home).
External constraints. Workplace discrimination is measured with a dummy variable that activates when a woman reports, in more than 50 % of the occasions in which she was observed employed, that she felt discriminated because of her race, age, sex, or nativity.6 Two dummy variables capture family constraints: ever divorced, and having a husband who frequently expressed opposition to her employment. Two dummy variables flag women facing frequent health situations that limited the amount or type of work they could do: one for their own health, and one for the health of her relatives. For details on the definition of “frequent” for these variables, see Table 5 in the appendix.
Attitudes and preferences. Work plans7 come from women’s answer to the question, “What would you like to be doing when you are 35 years old?,” which was asked between the ages of 14 and 34. I aggregate women’s answers into a distribution of work plans, with those always saying that they wanted to “work for pay” at the upper end, and those who always expressed a different preference (i.e., “looking after home/family” or “doing something else”) at the lower end of the distribution. Then I break this work-preference distribution in terciles: high, mixed, and low work expectations. Hence, this variable can be interpreted as a woman’s relative position, within her birth cohort, in the intensity of her midlife work plans. Two dummy variables capture women’s subjective enjoyment of paid work and childcare. Job satisfaction was asked every survey year, using a four-item scale (like it very much, like it fairly well, dislike it somewhat, dislike it very much). Enjoyment of childcare was assessed for all women, regardless of motherhood status, using a five-item response scale for the question, “How do you feel about taking care of children?” (like it very much, like it somewhat, not care either way, dislike it somewhat, dislike it very much). In both cases, I flag women who were below the cohort median in the distribution of satisfaction with these activities: that is, the least satisfied within this cohort.
Sociodemographic controls. Human capital is measured as the highest grade completed at age 25, in four categories: less than high school, high school graduate, some college, and college graduate or more. I control for husband’s earning capacity (adjusted to constant 1990 dollars, broken down in quartiles), and women’s race/ethnicity, with a dummy variable for non-Hispanic whites.
Figure 1 presents average employment rates for all women in the NLS-YW cohort, from ages 20 to 54. Rates increased from 41 % to 52 % in the respondents’ early 20s, remained relatively flat in their late 20s, and grew consistently during their 30s and early 40s, reaching a lifetime high at age 43, with 71 % of women working for pay. Overall participation rates for this cohort decreased after age 43, with 60 % employed at age 55. This profile fits previous studies showing that the working trajectories of recent cohorts of American women lost the traditional M-shaped profile of older cohorts (with a trough around the late 20s), becoming similar to men’s—an inverted U—but at a lower level, peaking between 70 % and 75 % in the mid-40s (Spain and Bianchi 1996).
However, cohort averages mask a high degree of heterogeneity. In nested GBTA models, I tested different numbers of employment groups (from two to six) and functional forms (linear, quadratic, and cubic). I settled for four trajectories, for substantive reasons: (1) adding more trajectories duplicated existing groups into parallel curves, with similar functional forms; (2) reducing the number of groups resulted in the merger of two of the existing trajectories, entailing a substantial loss of information given that the four model-predicted groups have distinct shapes. After four groups were chosen, BIC and AIC statistics confirmed that quadratic, rather than cubic, polynomials provided a more parsimonious fit.8
Figure 2 reveals a reasonable level of agreement between the actual employment trajectories of women in each of the four groups (panel a) and the four model-predicted employment profiles resulting from GBTA (panel b). Women differed in their employment patterns in two important ways: their baseline orientation, and their posterior tendency. Women were split almost in one-half by the mid-20s, with 52 % of them in the two groups that exhibited an early focus on work, and the remaining 48 % in trajectories characterized by low attachment to paid work in early adulthood. However, early trajectories did not necessarily imply continuity after midlife: either by choice or in response to external factors, 40 % of women experienced dramatic turnarounds over time, losing their early focus on employment (13 %) or otherwise increasing labor supply after a slow start (27 %).
These patterns, with two baseline preferences (market work vs. domesticity) and four possible trajectories (continuous vs. discontinuous) are strikingly similar to the four groups identified by Gerson (1986), with one important difference: Gerson’s baseline orientations came from women’s own recollections (and potential reinterpretations) of their teenage preferences, but these were not directly observed at the time, as they are in my study.
Characteristics of the Four Employment Groups
Table 1 presents means and proportions for the full NLS-YW cohort (first column) and for each of the four employment groups identified earlier. I name groups according to their employment profile (group size in parentheses) and present summary statistics on the sociodemographic traits and adult experiences characterizing the lives of its members.
Consistently Detached (21 %). Most women in this group remained detached from paid work throughout adulthood: only one in five were employed at any given age. Their family formation patterns were disproportionately precocious (37 % of them were teen mothers), nontraditional (33 % had out-of-wedlock first births, and 33 % never married), and intensive (46 % had three or more children). With respect to external constraints, they had mixed experiences: even though, on average, they divorced less often than their peers (33 %), they were more likely to have a husband who opposed their employment (50 %) and were confronted with health problems at relatively high rates, be it their own health (43 %) or that of their relatives (19 %). They were not particularly work-oriented, with nearly one-half of them repeatedly expressing preferences for domesticity. More often than their peers, they felt discriminated against at work, dissatisfied with their jobs, and satisfied with mothering. Even though they were less educated than other women, a relatively high proportion of them married men with earnings in the top and bottom quartiles: this variable is highly bimodal for this group.
Increasingly Attached (27 %). Women in this group remained loosely attached to employment in their 20s, gradually joined the labor force during their 30s, and reached high employment rates (over 80 %) after age 40. Their traits and experiences were average for their cohort, following a traditional housewife path in young adulthood: 87 % of them married, many of them by age 25 (81 %); becoming mothers in the early 20s (51 %); and going on to have three or more children (44 %). Virtually none remained childless. They married husbands with traditional views regarding their employment at a time when work opportunities were expanding for women. Maybe for this reason, 50 % of these women eventually divorced—more than in any other group. Their attitudes and preferences were average for this cohort, slightly skewed toward domesticity: only 26 % had strong work expectations. Most of them graduated from high school and married men with earnings above the median. Non-Hispanic whites are overrepresented in this group.
Increasingly Detached (13 %). The smallest of all, this group includes women who exhibited high employment rates in early adulthood—peaking around 75 % between the ages of 25 and 35—but who later left the labor force at dramatic rates, with fewer than one in four employed after age 50. Their family experiences were characterized by late fertility relative to their cohort peers, with 33 % having their first child in their late 20s. Tellingly, they felt discriminated at work more often than women in other groups, which could be related to their declining labor supply. They reported having relatives with bad health more often than women who ended up working at high rates; they had average work expectations, and many didn’t enjoy caring for children. This was the most highly educated group: 43 % graduated from college or pursued higher education. Their husbands were almost evenly distributed across the earnings distribution, and their racial/ethnic composition was average for this cohort.
Consistently Attached (40 %). This most-numerous group consists of those strongly attached to the labor force, with employment rates over 80 % (peaking above 90 %) for most of their adult lives. They were less likely to marry and married later than other groups; remained childless at an astonishing rate (40 %); and when having children, they rarely had three or more (14 %). Most women in this group married men who supported their employment choices; divorce rates were average for this cohort. Relative to their peers, they rarely experienced conditions that might have swayed them away from employment: only 11 % reported frequent feelings of discrimination, only 24 % experienced health limitations, and only 7 % saw their employment limited by the health of a relative. Many women in this group (42 %) consistently expressed plans to work for pay, and few expressed dissatisfaction with work. They were unlikely to drop out of school (6 %), and more than one-half attended college. Those who married did not have husbands with particularly high earnings.
In summary, descriptive results suggest that the following factors (common to the Increasingly Detached and the Consistently Attached groups) facilitated employment in early adulthood: late childbearing, low fertility rates, strong work orientations, pursuit of higher education, and husbands with slightly better-than-average incomes (but not exceptionally high) who supported their employment choices. Opposite experiences were associated with an early detachment from market work. Interestingly, the two groups (Consistently Detached and Increasingly Detached) who ended up working at low rates often reported feeling discriminated against at work and dealing with health problems.
I now turn to the second stage of the GBTA, in which multinomial logistic regression is used to investigate associations between group membership and a variety of factors. I build multivariate models9 around the three stories that emerge from the earlier description: (1) baseline orientations (the level of attachment to the labor force in early adulthood: strong vs. weak), and divergence from these baselines (2) toward increasing employment rates among those weakly attached at first and (3) toward nonmarket pursuits among those initially focused on market work.
Factors Associated With High Employment in Young Adulthood
Women in the Consistently Detached and Increasingly Attached groups were loosely attached to the workforce in early adulthood, with fewer than one in three members of these groups working for pay before age 30. In contrast, women in the Consistently Attached and Increasingly Detached groups shared early trajectories of strong attachment to the labor force—as high as 75 % by their late 20s. What lies behind these early differences in labor force attachment? Table 2 presents results from models predicting membership in the two groups exhibiting strong early workforce attachment relative to the two loosely attached groups.
Multivariate results confirm the picture described earlier. High rates of labor force participation in early adulthood are associated with later and lower fertility and marriage rates; with fewer external constraints from discrimination, unsupportive husbands, or bad health; with stronger work expectations, higher job satisfaction, and greater dislike of childcare; with higher education; and with earnings that are neither very high nor very low. Experiencing a divorce is also associated with stronger early attachment to the labor force.
Factors Associated With Increasing Attachment to the Labor Force
Unlike Consistently Detached women, who exhibited very low employment rates (approximately 20 %) for most of their adult lives, women in the Increasingly Attached group gradually entered the labor market over time. Table 3 presents results from models predicting membership in the latter group, as opposed to the group that remained largely detached from paid work.
Most of the factors that explained early workforce attachment also explain a later entry into the market among those loosely attached at first: having fewer children; experiencing fewer instances of discrimination or relatives with ill health; having a supportive husband, or divorcing; and expressing stronger work expectations. However, women’s levels of satisfaction with work or with the task of caring for children (which had predicted baseline levels of attachment) were largely irrelevant to late entries into the market. In the case of childcare, this result could be due to the fact that enjoyment of childcare questions were asked early in the NLS-YW; conversely, job satisfaction questions are available throughout the study, which seems to suggest that trajectories, once initiated, are difficult to reverse based solely on personal preferences.
Why Some Women Became Increasingly Detached From Work at Midlife
Among women with strong attachment to the workforce during their 20s and early 30s, some became Increasingly Detached after age 35, bottoming at rates below 25 % after age 50, while others remained employed. Table 4 explores the factors associated with this process of withdrawal from paid work.
These results show a clear pattern. Life course events, attitudes, and preferences had only a limited association with shrinking employment rates; only later births (as opposed to remaining childless) and not having strong work expectations explained part of the decline in labor supply after midlife. However, constraining circumstances appear more relevant here: reporting frequent feelings of discrimination at work, having an unsupportive husband (and not divorcing), and health limitations (personal or from relatives) are all factors that increase the chances that a woman belongs to the group that cut back on employment after midlife. Education mildly protected women from decreasing employment, but having a top-earning husband increased the chances of labor force retrenchment.
In the introduction to this article, I asked three questions about the actual working experiences of Baby Boom women in the United States. First, are some working trajectories observed with higher typicality than others? Second, do these trajectories reveal a pattern of continuity or discontinuity? And finally, how did the actual lives and experiences of women in different trajectory groups differ? Using group-based trajectory analysis and data from women born between 1944 and 1954, observed from ages 20 to 54, this article reveals the following:
Baby Boom American women followed one of four ideal-type working trajectories: 21 % remained Consistently Detached from the labor force throughout their adult lives; 27 % became Increasingly Attached to paid work as they aged; 13 % were Increasingly Detached from the workforce, cutting back on employment after more than a decade working at high rates; and 40 % remained Consistently Attached to market work.
These findings reveal that paid work for this cohort was very fluid: for roughly 40 % of the women in my sample (those Increasingly Attached and Increasingly Detached), employment was not a continuous state but rather an endeavor to which they devoted changing amounts of effort over time, depending on a variety of circumstances.
Finally, multinomial models investigated the socioeconomic and demographic traits associated with membership in the four employment groups. I first explored the factors associated with a strong baseline level of attachment to paid work in early adulthood. Second, among those loosely attached to the market early in life, I investigated the factors leading to growing employment rates. Finally, among those strongly attached to the market in young adulthood, I explored the factors associated with retrenchment from paid work.
I found that family events were most influential earlier in adulthood, splitting the sample in two and explaining why some women (who had many children, or who had them early) remained loosely attached to the labor force in their 20s. However, life course events were less useful for explaining later drops in employment among women who worked intensely from ages 20 to 35. External constraints proved particularly important for labor force dropout at midlife. Subjective feelings of discrimination predicted declining employment rates, as did marrying a husband with unsupportive views about women’s work, or having to deal with situations of bad health. Divorce correlated with reversals toward higher employment at midlife. Attitudes toward work and home life differentiated Consistently Detached (stay-at-home) and Consistently Attached women such that those in the first group were more likely to express dissatisfaction with work and satisfaction with childcare, which, as subjective assessments, could follow from post-hoc rationalization. Work–family preferences were robustly related to work profiles: repeatedly expressing intentions to work for pay had lasting effects, pulling women toward the market across the life course.
Numerous studies have explored the correlates of women’s work–family behaviors since the gender revolution of the 1970s. The quantitative literature has mostly adopted a variable-centered approach, investigating net effects of life course transitions (e.g., marriage, childbearing) and other traits (e.g., education, health) on employment. Although useful for examining causality and average effects, variable-centered approaches are less apt for the exploration of population heterogeneity and for unveiling the cumulative, interactive nature of work and family experiences, which in real life seldom take place in ceteris paribus, watertight compartments. The qualitative literature offers a richer, more nuanced view of the narratives, normative schemas, and motivations behind women’s employment decisions and outcomes. However, these studies have their own measurement problems, from recall errors to the post-hoc rationalization and reinterpretation of events.
In this article, I have sought to complement the existing literature with a person-centered, longitudinal exploration of women’s labor supply. Using group-based trajectory analysis, I extracted commonalities among the wide spectrum of employment experiences of a generation of American women, born between 1944 and 1954 and observed from ages 20 to 54. Four stylized work trajectories emerged, characterized by: detachment from the labor market (21 %), increasing levels of participation in paid work (27 %), retrenchment from the market after midlife (13 %), and consistently high employment rates throughout adulthood (40 %).
This article illustrates how person-centered analyses can further our understanding of complex social outcomes by reducing complexity along a different axis (subpopulations with different experiences of the same outcome) than variable-centered analyses (net effects between variables). Additionally, by keeping a focus on the long-term experiences and outcomes of their subjects, they can reveal underlying dependencies and connections between the work and family domains, thus dialoging with, and potentially validating, insights gained in qualitative studies.
Three characteristics of the employment histories of Baby Boom women in the United States became salient in this study: heterogeneity, discontinuity, and adaptability. Heterogeneity, because in a social context of growing female employment (both across decades and along the life course), I found subpopulations with diverse work experiences, at times contrary to the secular trend toward growing employment. Discontinuity, because an important proportion of women (40 % of the sample) experienced trajectory reversals: some became increasingly attached to the labor force, and some grew increasingly detached over time. Adaptability, because employment patterns adjusted to experiences, preferences, and constraints: the timing of marriage and fertility influenced human capital accumulation and the launching of professional careers; external limitations such as workplace discrimination, unsupportive husbands, or bad health correlated with loose attachment to the labor force; and preferences for market work over domestic work translated into higher labor supply throughout adulthood.
This article has important limitations. First, I analyzed an older cohort, whose work and family experiences may not resemble those of contemporary American women.10 Early Baby Boomers are particularly interesting because they spearheaded the gender revolution: they were the first generation to be employed in large numbers, they experienced declining fertility rates, and they were the first to combine work and family in high proportions (Goldin 2004). However, family change has accelerated in recent decades. Important features of contemporary family life, such as the erosion of normative scripts for the transition to adulthood (Aassve et al. 2007; Shanahan 2000), the increasing variability in the timing of marriage and parenthood by education (Cohen and Bianchi 1999; Martin 2000), and the rise in cohabitation (Brown et al. 2008; Bumpass and Lu 2000) might have made the experiences of recent cohorts of women more heterogeneous and complex than those of their older peers (Aassve et al. 2007). Increasing job instability and greater economic vulnerability, particularly among young adults (Kahn et al. 2013; Levy 1998), might have added to the forces destabilizing women’s long-term careers.
Second, the method used here is eminently descriptive. Person-centered analyses do not have the estimation of causal effects as their primary goal. Thus, they do not rest as heavily as variable-centered approaches on assumptions of exogeneity and temporal sequencing between cause and effect. However, because endogeneity and temporal overlap are never desirable qualities in statistics, I ran robustness tests (available upon request) in which I defined endogenous variables as occurring earlier in life: work–family preferences between ages 20 and 25 (instead of ages 20 to 34) and feelings of discrimination at work, job satisfaction, and enjoyment of childcare between ages 20 and 34 (instead of ages 20 to 54). These alternative definitions did not produce substantively different results, but the relationships described in this article should be deemed associational rather than causal.
Third, I have relied heavily on subjective reports, which are often driven by social desirability and post-hoc rationalization. Still, a well-established body of literature stresses the importance of preferences and attitudinal orientations in shaping behavior. Additionally, the use of subjective information allowed me to approximate factors, such as discrimination, which (because they are difficult to measure objectively) have remained black-boxed in previous observational studies.
This article highlights the importance of looking at female market outcomes over the long run, moving beyond static characterizations of women’s work–family outcomes, and allowing patterns of dissimilarity to emerge and compound. This should lead to a more fluid assessment of women’s long-term strategies, and those of their husbands and families—moving, as Moen and Sweet (2004) put it, from a dichotomous “work–family” paradigm to one of “flexible careers.”
I am grateful to Suzanne M. Bianchi and Joan R. Kahn for mentorship; to Frances Goldscheider, Alexandra Killewald, Spencer L. James, and Philip N. Cohen for helpful feedback; and to Ozan Aksoy for statistical assistance.
For the empirical work, I use the Stata traj plugin, developed by Jones and Nagin (2012).
Because I model the probability of being employed at a given age (dummy variable), this section presents the logistic specification of GBTA.
Using time-invariant covariates might raise endogeneity concerns, which are otherwise unavoidable in this type of analysis given that GBTA is not primarily aimed at elucidating causality. To assuage this concern, I ran alternative models (available upon request) in which key independent variables were defined early in adulthood. Results were substantively similar; I provide further comments on these models in the Discussion section.
I considered dropping more generally all observations in which women were enrolled in school, but decided against it because (1) higher education is often combined with employment, and the data don’t allow us to decipher which of the two activities was given priority at each point in time; and (2) this is a study of women’s labor supply, and the possibility that education might have lead some women to take up low-paying or part-time work is less relevant than the fact that they were employed.
Table 5 in the appendix lists all the variables used, including the years in which each of them was available.
A minority of women were never observed to be working for pay between the ages of 20 and 54. These “undefined” cases are given the value 0, because workplace discrimination was not a relevant issue for them. The same decision was made for the dummy variables “being dissatisfied with work” and “having a husband who opposes her employment” for women never observed to be employed and married, respectively. Analyses, available upon request, in which “undefined” values were imputed using three alternative methods (grouping into a separate category, multiple imputation, and listwise deletion) produced consistent results.
Following prior literature (Hayford 2009; Ryder and Westoff 1971), which found these concepts to be empirically similar, I use the terms “expectations,” “plans,” “intentions,” “orientations,” and “preferences” interchangeably throughout this article.
This model with four quadratic trajectory groups was replicated using Mplus (Muthén and Muthén 2014), allowing me to confirm that the results reported here are not the result of local maxima and that my models achieved a satisfactory entropy score (0.71). Mplus scripts are available upon request.
The results shown here correspond to a common multinomial logistic model. However, to better illustrate differences between pairs of groups, I switched reference categories and split the results in different tables. For the full model, including all possible groups comparisons, see Table 6 in the appendix.
Unfortunately, recent NLS surveys did not collect data on key variables, such as husband’s opposition to their wives’ employment, women’s own work plans and expectations, satisfaction with the role of caring for children, and the health of relatives. Roughly similar analyses of the employment trajectories of women in the NLSY-79 cohort—whose older members are in their early 50s as of this writing—reveal the existence of similar groups, with slightly different compositions: (1) Consistently Detached (16 %); (2) Increasingly Attached (25 %); (3) Increasingly Detached (22 %); and (4) Consistently Attached (36 %). These results are available upon request. However, because of the lack of key covariates, conducting a fully comparable analysis of the traits associated with group membership is not possible using more recent NLS cohorts.