U.S. Mothers’ Long-Term Employment Patterns

Abstract

Previous research on maternal employment has disproportionately focused on the immediate postpartum period and typically modeled either cross-sectional employment status or time until a specific employment transition. We instead conceptualize maternal employment as a long-term pattern, extending the observation window and embedding employment statuses in temporal context. Using data from NLSY79 and sequence analysis, we document five common employment patterns of American mothers over the first 18 years of maternity. Three typical patterns revolve around a single employment status: full-time (36 %), part-time (13 %), or nonemployment (21 %); the other two patterns are characterized by 6 (15 %) or 11 (14 %) years of nonemployment, followed by a period of transition and then full-time employment. Analyses of the immediate postpartum period cannot distinguish between the nonemployment and reentry groups, which have different employment experiences and different prematernity characteristics. Next, we describe how mothers’ human capital, attitudes and cultural models, family experiences, and race/ethnicity are associated with the employment patterns they follow, elucidating that these characteristics may be associated not only with how much mothers work but also the patterning of their employment. Our results support studying maternal employment as a long-term pattern and employing research approaches that address the qualitative distinctness of these diverse patterns.

Introduction

In the United States, motherhood is associated with a substantial decline in women’s paid labor (Boushey 2008; Gangl and Ziefle 2009), which in turn is a major contributor to mothers’ lower wages compared with childless women (Budig and England 2001; Gangl and Ziefle 2009). The difference between men and women in labor market experience also contributes to the gender pay gap (Blau and Kahn 1997; Wellington 1994). Thus, identifying common patterns of mothers’ employment, as well as what maternal characteristics predict these distinct patterns, can help us understand heterogeneity in women’s experiences at the work–family intersection that have consequences for their wage outcomes.

Yet, the patterns of mothers’ employment and how they vary with mothers’ economic and demographic characteristics are only partially understood (Lu et al. 2017). Prior quantitative research on maternal employment has typically considered a relatively narrow period of a few years surrounding the birth of a child (Desai and Waite 1991; Eggebeen 1988; Glass and Riley 1998; Hynes and Clarkberg 2005; Klerman and Leibowitz 1994; Leibowitz and Klerman 1995; Leibowitz et al. 1992; Lu et al. 2017; Wenk and Garrett 1992; Yoon and Waite 1994), providing a truncated picture of mothers’ employment. Using qualitative data, Damaske (2011) argued that a lifetime perspective—exploring how women work throughout their lives—is necessary to understand how class shapes women’s employment. We argue that the same is true for understanding how a broad range of characteristics is associated with mothers’ employment outcomes.

Our first contribution is to describe U.S. mothers’ long-term employment patterns, analyzing their employment for 18 years after the birth of a first child. If mothers’ employment does not return quickly to prematernity levels, mothers are likely to continue to fall further behind their childless peers in terms of accumulated experience and, thus, wages as their children age. Recognizing that the motherhood wage penalty will be shaped by mothers’ long-term as well as immediate postpartum employment, we extend the observation window to cover nearly two decades of mothers’ employment. By necessity, this means that we do not study the employment patterns of today’s new mothers. Instead, we describe the work experiences of a group of mothers currently at midlife, who have largely completed child-rearing, are near the peak of their lifetime wage trajectories, and are now approaching retirement age.

Analytic methods that employ snapshot measures of mothers’ employment status or examine a specific transition, such as first return to employment postpartum, cannot capture the full process and patterning of mothers’ careers (Lu et al. 2017). We consider long-term employment patterns holistically, allowing us to answer the following questions: Is part-time employment a common but temporary status, as mothers transition between nonemployment and full-time work? Or is it primarily a long-term strategy, employed by a distinct subset of mothers? Is the overall depressed rate of maternal employment driven primarily by a subset of mothers who are out of the labor market long-term, or do most mothers follow a similar pattern of a short withdrawal from the labor market just after the birth of their children and then a return to employment?

Our second contribution is to evaluate how conventional predictors of maternal employment—human capital, attitudes and cultural models, family experiences, and race/ethnicity—are associated with different employment patterns. We argue that conclusions about which mothers work more or less should not be based only on analysis of the first few years postpartum and that some characteristics may differentiate how women’s employment unfolds more strongly than how many hours of labor market experience they ultimately accumulate.

Background and Contributions

Describing Maternal Employment Patterns

Studies of maternal employment and its correlates have often measured maternal employment outcomes within several years postpartum (Desai and Waite 1991; Glass and Riley 1998; Hynes and Clarkberg 2005; Klerman and Leibowitz 1994; Leibowitz and Klerman 1995; Leibowitz et al. 1992; Lu et al. 2017; Wenk and Garrett 1992; Yoon and Waite 1994). This focus is reasonable, given the intensity of employment changes surrounding the transition to maternity and interest in how policies such as maternity leave affect the employment of new mothers. However, these analyses overlook substantial variation in long-term employment patterns and accumulated work experience; for example, they cannot distinguish between mothers who return to full-time employment after the observation window ends and those who never return to full-time employment. The omission is problematic because of the strong contribution of work experience to the motherhood wage penalty and gender wage gap (Blau and Kahn 1997; Budig and England 2001; Gangl and Ziefle 2009; Wellington 1994). By lengthening the focal period through the first 18 years of maternity, we examine maternal employment in longer-term perspective. In this way, we paint a fuller picture of how mothers combine child-rearing and employment across the life course. We then describe the characteristics of mothers that are associated with following particular employment patterns, recognizing that predictors of long-term employment patterns may or may not match those identified in research focused on new mothers. Understanding how mothers accumulate work experience and which mothers tend to accumulate more or less is important to furthering our understanding of the processes by which both gender wage inequality and the within-sex motherhood penalty are produced.

Prior scholarship that has considered women’s or mothers’ part-time and full-time employment separately has found distinct predictors of each (e.g., Drobnič 2000; Lu et al. 2017). We measure employment status categorically, recognizing variation in work intensity among employed mothers and considering unemployment and out of labor force as distinct statuses. To analyze maternal employment patterns holistically as a series of sequential, connected employment statuses, we use sequence analysis. Sequence analysis was popularized in the social sciences by Abbott and Hrycak (1990) and has been used to study, among other things, variation in women’s (Aassve et al. 2007; Anyadike-Danes and McVicar 2010; Simonson et al. 2011) or both sexes’ (Biemann et al. 2012; Han and Moen 1999; Malo and Muñoz–Bullón 2003) long-term employment patterns.¹ One strength of sequence analysis is that it can identify less-standard patterns (Aisenbrey and Fasang 2010), allowing us to describe heterogeneity in mothers’ employment experiences. Unlike prior research on long-term employment trajectories as patterned by age (e.g., Damaske and Frech 2016), we describe employment patterns following the birth of a first child, consistent with the strong effect of motherhood on women’s paid work (Boushey 2008; Gangl and Ziefle 2009).

Several previous studies have incorporated aspects of our proposed long-term sequence-based categorical approach to studying maternal employment. Vandenheuvel (1997) described mothers’ employment patterns for up to 10 years after the birth of a first child, categorizing mothers as (1) in the labor force in all years, (2) out of the labor force in all years, or (3) something in between. However, she did not distinguish length of labor force participation within the third group; furthermore, full-time employment, part-time employment, and unemployment were grouped together as in the labor force. Lu et al. (2017) and Hynes and Clarkberg (2005) analyzed mothers’ employment patterns as they unfold following a birth, but for at most two years after a birth.² Our approach integrates the advantages of both a detailed employment trajectory and a long period.

Identifying Social Correlates of Maternal Employment Patterns

After identifying common maternal employment patterns, we explore the maternal characteristics associated with each pattern. We consider four groups of characteristics widely studied in prior research: human capital, attitudes and cultural models, family experiences, and race/ethnicity. Our analyses describe how these characteristics are associated with maternal employment over a longer period than has previously been studied and perhaps differently with part-time versus full-time employment. We use theoretical perspectives on maternal employment in conjunction with prior empirical evidence to generate expectations about correlates of maternal employment patterns. However, we recognize that our analyses are not well suited to causal inference; our results describe correlates of maternal employment patterns, rather than generating causal estimates or assessing the relative validity of different theoretical perspectives.

Human Capital

We expect that mothers’ employment will be more intense in terms of work hours and more consistent across years when their economic rewards for work are higher, increasing the opportunity cost of labor force exit. By this logic, wages and measures of wage-earning potential, including education, work experience, and job tenure, should be positively associated with women’s or mothers’ employment. Prior research has generally, although not universally, found support for these predictions (Blau and Kahn 2007; Damaske and Frech 2016; Desai and Waite 1991; Eggebeen 1988; Glass 1988; Glass and Riley 1998; Hynes and Clarkberg 2005; Klerman and Leibowitz 1994; Landivar 2013; Leibowitz and Klerman 1995; Leibowitz et al. 1992; Lu et al. 2017; Shafer 2011; Taniguchi and Rosenfeld 2002; Wenk and Garrett 1992). Of course, associational evidence does not prove the theoretical argument. Women’s human capital and labor supply may be jointly determined by other factors, and their human capital may affect labor market decisions for reasons other than the opportunity cost of exit.

Attitudes and Cultural Models

The gender beliefs perspective highlights that women’s gender role beliefs and expectations about employment and motherhood shape their employment patterns (Damaske and Frech 2016). Prior research has generally found that women and mothers who have more traditional gender role attitudes (Damaske and Frech 2016; Glass and Riley 1998; Hynes and Clarkberg 2005) and who do not desire to be employed at age 35 (Damaske and Frech 2016; Desai and Waite 1991) are more likely to be nonemployed or less consistently employed. We likewise expect that net of economic circumstances, maternal employment will be less intense and less consistent for mothers with more traditional gender role attitudes, preferences for more children, and a preference for unpaid versus paid labor.

Family Experiences

Beyond women’s own attitudes and economic motivations for work, we expect that family context will also be associated with maternal employment patterns. Among married women, husbands’ higher earnings and longer paid work hours are both expected to depress maternal employment by reducing the demand for her paid work and increasing the demand for her unpaid work, respectively; the same factors should tend to lower employment for married mothers compared with unmarried mothers. Prior research has generally, although not universally, found the expected negative association between women’s or mothers’ employment and husbands’ (and other family members’) income or wages (Blau and Kahn 2007; Boushey 2008; Damaske and Frech 2016; Eggebeen 1988; Glass 1988; Glass and Riley 1998; Klerman and Leibowitz 1994; Leibowitz and Klerman 1995; Leibowitz et al. 1992; Lu et al. 2017; Shafer 2011; Taniguchi and Rosenfeld 2002; Wenk and Garrett 1992). Wives are also more likely to exit the labor force when their husbands work long hours (Cha 2010; Shafer 2011).

We expect that married mothers will be underrepresented in sequences characterized by persistent full-time employment. In the United States, among women currently not employed, married mothers of minor children have higher rates of labor force reentry to part-time employment but lower rates of reentry to full-time employment than their single, childless counterparts (Drobnič 2000). We expect that distinguishing between part-time and full-time employment may be particularly important for understanding how marriage is associated with mothers’ employment. As a measure of family context, we also consider age at maternity; later fertility is generally associated with less employment disruption (Hynes and Clarkberg 2005; Wenk and Garrett 1992).

Race/Ethnicity

Prior research has generally found that after other factors are controlled, African American and Hispanic mothers are less likely than white mothers to exit employment or reduce work hours (Landivar 2013; Lu et al. 2017; Vandenheuvel 1997). We expect that African American and Hispanic mothers will have higher likelihood of consistent full-time employment than white mothers, with other characteristics held constant.

Considerations and Summary

Throughout our analyses, we are attentive to the possibility that some characteristics may be associated not with higher or lower employment rates but rather with the pattern of employment. For example, a mother who is employed part-time at 1,000 hours per year in each of the first two years after the birth of a first child and a mother who withdraws completely from the labor force for a year following a first birth and then returns to full-time employment at 2,000 hours per year would accumulate the same total number of hours of paid work experience but through quite different employment paths. These two patterns may be typical for mothers with different characteristics, and they may have different long-term career consequences. Analyses that do not distinguish between part-time and full-time employment, that consider employment patterns as only quantitatively distinct (measured in total hours of accumulated work experience) rather than also categorically, or that do not consider employment patterns over a longer postpartum period will be unable to uncover these differences.

In summary, we challenge the traditional focus on the immediate postpartum period and the emphasis on quantitative measures of higher or lower employment levels (or faster or slower labor market entrances and exits). Instead, we place maternal employment in long-term perspective and identify discrete, qualitatively distinct employment patterns, allowing us to uncover key characteristics associated with each pattern. This fuller understanding of how mothers’ work is structured throughout their lives and which mothers tend to experience different employment patterns is crucial given the substantial economic consequences of these patterns for wage differences among mothers, between mothers and nonmothers, and between men and women.

Data and Methods

We use data from the 1979–2014 waves of the National Longitudinal Survey of Youth 1979 (NLSY79). NLSY79 surveyed a nationally representative sample of 12,686 men and women aged 14–22 in 1979 and has subsequently reinterviewed them annually or biennially, with a response rate above 77 % (Bureau of Labor Statistics 2017b, 2017c). Our results describe the experiences of mothers currently at midlife (ages 54–61 at the end of 2018).

Our target population is biological mothers who lived until their first child’s 18th birthday. We exclude women in subsamples discontinued by NLSY79, childless women observed through age 40, and those who died childless or less than 18 years after a first birth. Among women potentially in our target population, 4 % attrited before age 40 and were childless when last observed, 3 % became mothers less than 18 years prior to their 2014 NLSY79 interview, and 9 % attrited prior to their first child’s 18th birthday. After exclusions, our analytic sample used to identify common maternal employment patterns, which we refer to as the Extended Sample, includes 3,465 women. All subsequent analyses are weighted with NLSY79 custom weights for respondents in any survey year.³

Identifying Common Employment Patterns

We use optimal matching, a type of sequence analysis, to describe mothers’ long-term employment patterns. First, we define a month-by-month employment sequence for each mother. Second, we create a dissimilarity matrix that measures how different any two sequences in the data set are from each other. Third, we use the dissimilarity matrix to create clusters of sequences that follow similar patterns. These clusters represent common maternal employment patterns and become the outcome variable in subsequent analyses.

Month-by-Month Employment Sequences

NLSY79 constructs measures of respondents’ labor force status and total work hours in each week since January 1, 1978. These measures are based on respondents’ start and stop dates of jobs and periods not working, usual hours worked at jobs, and whether time not working was unemployed or out of the labor force (Bureau of Labor Statistics 2017a). We use the weekly measures to construct monthly maternal employment sequences that begin at the first calendar month after the birth of a first child and last 18 years (216 months).

In each week, we classify a respondent as (1) employed full-time (at least 35 hours) or active-duty military; (2) employed part-time (at least 20 but fewer than 35 hours); (3) marginally employed (fewer than 20 hours); (4) employed but with unknown work hours; (5) unemployed; (6) out of the labor force (OOLF); (7) not working, unknown whether out of labor force or unemployed; (8) on a within-job work gap (which can include maternity leave);⁴ or (9) missing. We define monthly employment status as the modal weekly status that month, tie-breaking in favor of the status that is most common in the sample.

The most common source of missing data is women who became mothers before 1978, when the weekly employment arrays began.⁵ In the next section, we describe how we incorporate missing employment data in the sequence analysis.

Constructing a Dissimilarity Matrix

We analyze employment sequences using optimal matching, which identifies the smallest possible cost of transforming one sequence into another (Abbott and Hrycak 1990). This cost, or dissimilarity, measures the distance between two sequences. For any month in which two sequences have discrepant statuses, the substitution cost matrix defines the cost of substituting one employment status for another. We measure the total dissimilarity between the two sequences as the sum of these costs across all months in which the sequences have discrepant statuses.⁶ Each pairwise cost in the substitution cost matrix reflects the degree of dissimilarity between the two statuses. The substitution cost is not determined by how easy it is for an individual to change from one status to another: it reflects how discrepant two employment statuses are from one another. Because they measure dissimilarity, cost matrices are symmetric (e.g., substitution from full-time to part-time is the same as vice versa). The scale of costs is arbitrary.

We use a custom cost matrix, shown in Table 1, that is based on the overarching logic that two employment statuses are more similar to one another when the number of hours spent in paid work is more similar. Three nonmissing statuses involve 0 paid work hours: OOLF, unemployment, and within-job work gaps. We consider them to be relatively similar to one another. However, among these three statuses, we view within-job work gaps as relatively closer to the employed statuses because, unlike for unemployment and OOLF, the connection to an employer is maintained. We consider employed statuses to be slightly more similar to unemployment than to OOLF because unemployed women remain in the labor force. We treat completely missing employment status as equally discrepant from all other statuses. When a respondent is employed with unknown work hours or is nonworking but it is unknown whether she is unemployed or OOLF, we assign costs to place the status in a middle ground among the possible true statuses. We recognize that our cost matrix is one of many reasonable choices, and alternative cost matrices produce similar cluster solutions (see online appendix text, Tables S7 and S8, and Fig. S1).

Identifying Clusters

After establishing a matrix of pairwise dissimilarities between sequences, we use weighted cluster analysis to identify groups of mothers with similar employment sequences. Following the discussion in Aisenbrey and Fasang (2010) of tools for validating cluster solutions after sequence analysis, we experiment with several clustering algorithms and numbers of clusters and consider both partition quality and construct validity when selecting a cluster solution. We select the k-medoids algorithm and five clusters (see the online appendix text and Figs. S2–S4). To generate a five-cluster solution, the algorithm begins with five randomly chosen sequences as medoids and assigns each sequence to the closest medoid. Medoids can be thought of as representative sequences for their clusters. The algorithm iterates by changing the selection of medoids, ultimately selecting the medoids and associated assignment of sequences to clusters that minimize the weighted sum of distances from each observation to the medoid of its assigned cluster (Reddy and Vinzamuri 2013). Because cluster solutions can be sensitive to the initial random draw of medoids, we use five random starts and select the best fit. We do not claim that there are exactly five ideal types of maternal employment patterns: five clusters is one reasonable choice based on model fit, analytical stability, and interpretability.

Modeling Employment Patterns

We use a weighted multinomial logit model to assess how human capital, attitudes and cultural models, family experiences, and race/ethnicity are associated with mothers’ employment patterns. We do not use post-first-birth traits to predict employment cluster. Our model is forward-looking, asking what characteristics before motherhood predict the employment pattern a woman will follow after becoming a mother.

To aid interpretability, we generate predicted probabilities of membership in each cluster for each observation, first setting the value of a given covariate to its reference value and then changing it to another value—for example, changing education from high school diploma to some college. For each individual, we subtract the predicted probability with the reference value for the covariate from the predicted probability with the new covariate value. Then we average these differences (with survey weights) across observations.⁷

We estimate models on four samples, each of which has different appropriate covariates, described in more detail later. In the Extended Sample, we include all 3,465 mothers and use covariates that can be measured prematernity for all mothers: education, marital status, and age at first birth; the respondent’s mother’s education and employment status; and race/ethnicity. The Post-1979 Sample includes only the 2,499 mothers whose first birth occurred at least 12 months after they were interviewed in 1979 but permits many additional covariates: prepregnancy measures of gender role attitudes, ideal number of children, preference for work at age 35, religious attendance, and paid work in the two years prior to the birth.

Compared with the Post-1979 Sample, the Extended Sample includes more teen mothers (28 % vs. 13 %) and mothers without a high school diploma (25 % vs. 14 %). Nonetheless, the cluster membership distribution is similar for both samples. To help identify patterns in the multivariable results that might be driven by using the more restricted Post-1979 Sample, rather than the Extended Sample, we estimate on both samples a simplified model that includes only those covariates that are available for the Extended Sample. We find similar patterns, with the exception of the results for age (see the online appendix, Tables S13–S15). Therefore, we use the Post-1979 Sample for our main multivariable results, but we draw from the results for the Extended Sample when discussing patterns by age.

In the Married Sample, we further restrict the Post-1979 Sample to the 1,627 mothers who were married in the month prior to the birth and who were asked in the calendar year of the birth questions about their spouse’s work hours and earnings in the previous calendar year. For these women, we include husbands’ earnings and work hours as additional covariates.

The Working Sample restricts the Post-1979 Sample to the 2,012 mothers who, in the most recent interview they completed at least a year prior to the birth, reported some jobs since the prior interview and reported valid wages for all jobs they held. For these women, we include wage and job tenure as additional covariates. We exclude mothers whose latest prematernity interview took place more than three years prior to the first birth.

All financial variables are adjusted for inflation to 2016 dollars. We multiply impute item-missing covariates (see the online appendix, Table S16, for item-missing rates).

Human Capital

We measure education as last reported by the respondent prior to the birth, generating four categories based on highest grade completed: no high school diploma (less than 12th grade), exactly a high school diploma (12th grade), some college (one to three years of college), and a four-year college degree or more (four or more years of college). For the Post-1979 Sample, we measure hours spent in paid work in (1) the calendar year before the birth, and (2) the calendar year two years before the birth. For each year, we use three categories: no paid work, less than full-time work (less than 1,820 annual hours, which translates to 35 hours per week), and full-time or more work (at least 1,820 annual hours).

In the Working Sample, we additionally consider prematernity hourly wage and job tenure (in weeks) as reported in the last survey wave that took place at least a year prior to the first birth. For respondents with multiple reported jobs since the last interview, we tie-break first in favor of a current job at the time of the interview, then the job with more weekly work hours, and then the job with longer tenure. Both variables are log-transformed and top- and bottom-coded at the 99th and 1st percentiles of the distributions.

Attitudes and Cultural Models

In all models, as a proxy for role modeling, we include a categorical measure of the respondent’s mother’s education, using the same categorization as for the respondent’s education, and a binary variable for whether the respondent’s mother was employed when the respondent was age 14. Prior research has indicated that daughters are more likely to be employed when their own mothers were employed (McGinn et al. 2018; Morrill and Morrill 2013). Mother’s education may also be a proxy for childhood economic advantage; Damaske and Frech (2016) found lower likelihood of steady employment for women whose mothers had less education. For respondents who did not live with their mother at age 14 but did live with a stepmother, we use the employment status of the coresidential mother figure. For both variables, we include a separate category indicating that the respondent did not live with her mother (or a stepmother) at age 14.

In the Post-1979 Sample, we have several additional variables, each measured in 1979. We include how many children the respondent thinks is ideal for a family, top-coded at 4. We include an indicator variable for whether the respondent wants to work at age 35. We use a measure of frequency of religious attendance as a proxy for cultural models of work and family to which the respondent may be exposed. We categorize attendance as less than monthly, monthly but less than weekly, or weekly or more.

NLSY79 asked respondents eight 4-point Likert-scale questions about their attitudes toward women’s roles. We use principal components analysis to identify statements that cohere together and then create an index of gender role attitudes by averaging responses to five statements (α = .75): a woman’s place is in the home, not in the office or shop; a wife who carries out her full family responsibilities doesn’t have time for outside employment; the employment of wives leads to more juvenile delinquency; it is much better for everyone concerned if the man is the achiever outside the home and the woman takes care of the home and family; and women are much happier if they stay at home and take care of their children. A higher score on the index indicates more traditional views.

Family Experiences

For all respondents, we measure age at first birth in categories: less than 20, 20–25, and over 25. We include an indicator variable marking whether the respondent is married in the month prior to the birth.

For women in the Married Sample, we measure husbands’ work characteristics as reported in the survey wave in the calendar year of the birth, when respondents report spouses’ work hours and earnings in the prior calendar year.⁸ Husbands’ work hours are measured in three categories based on usual weekly work hours: less than full-time (fewer than 35 hours), full-time (at least 35 but fewer than 50 hours), or overwork (50 or more hours). We distinguish overwork from full-time work given prior evidence that husbands’ overwork is associated with heightened likelihood of married mothers’ employment exit (Cha 2010; Shafer 2011). We also measure husbands’ annual earnings, transformed to their percentile in the weighted distribution of the Married Sample.

Race/Ethnicity

We measure race/ethnicity based on assignment by NLSY79 during household screening as Hispanic, non-Hispanic African American, and non-Hispanic women of other races (whom we call “white” for simplicity).⁹

Results

Maternal Employment Patterns

Figure 1 shows the weighted fraction of respondents in the Extended Sample in each employment status in each month from one year prematernity to the 18th year postpartum. Missing rates are elevated in the early years, which include pre-1978 months: 15 % one year prior to the first birth versus 4 % by the child’s second birthday and 1 % by the child’s fourth birthday. This pattern is important to keep in mind as we interpret the changes in known employment statuses over the same period.

Changes in employment are dramatic around the time of the first birth. The share of women employed full-time plunges from 51 % one year before maternity to 20 % in the first month postpartum; it rebounds to 34 % by the child’s first birthday but does not return to the prematernity level until more than a dozen years after the first birth. Immediately following the first birth, 15 % of mothers are in a within-job work gap, presumably on maternity leave, but this falls to 1 % a year after the birth. Part-time and marginal employment rates drop in the months preceding the first birth but are subsequently relatively stable; at the child’s first birthday, 15 % of mothers are employed part-time or marginally, compared with 18 % at the child’s 18th birthday.

The substantial change in women’s employment at the time of a first birth highlights the value of anchoring employment sequences to maternity.¹⁰ Fig. 1 also shows that a substantial share of the postpartum rise of full-time employment is overlooked when the observation window ends only a few years after the first birth.

Figure 2 shows the five medoids, or centers of clusters, of our cluster solution. Thirty-six percent of mothers are in the Full-Time cluster, the medoid of which is one month of within-job work gap followed by full-time employment in all subsequent months. Thirteen percent of the sample is in the Part-Time cluster, with a medoid characterized by part-time employment in every month. The medoid sequence of the Nonemployed cluster is OOLF in every month, and this group captures 21 % of the sample.

The final two clusters are Early Return (15 %) and Late Return (14 %). The medoid sequence of the Early Return cluster is approximately six years OOLF, followed by an approximately two-year transition that includes multiple statuses and then full-time employment for the next decade. The medoid of the Late Return cluster is 11 years OOLF and then approximately 5 years in part-time or marginal employment before entering full-time employment about 16 years after the first birth.¹¹ Analyses that focus only on the first several years after a first birth will not distinguish the Early Return, Late Return, and Nonemployed clusters.

Figure 3 shows all employment sequences, grouped by cluster, with each row reflecting the employment sequence of one mother. Each cluster is heterogeneous, but between-cluster variation in employment patterns is also clear.

As shown in Fig. 4, the clusters differ substantially in their labor market experiences over the first 18 years of motherhood. On average, mothers in the Full-Time cluster spend 80 % of their first 18 years of motherhood in full-time employment. Mothers in the Nonemployed cluster spend 78 % of their time nonemployed, either OOLF or unemployed. The Part-Time cluster spends 55 % of their time in part-time or marginal employment and substantial minorities of their time in full-time employment (25 %) and OOLF or unemployed (14 %). On average, the Early Return and Late Return groups each spend approximately three-quarters of their time in full-time employment or nonemployed, but their mix is different. The Early Return cluster spends 47 % of their time full-time employed, compared with 21 % for the Late Return cluster. The fact that even members of clusters whose medoid is dominated by a single status spend considerable minorities of their time in other employment statuses reveals the value of considering employment as a set of interconnected statuses, not a single snapshot: mothers in the same employment status at a given moment may be embedded in quite different long-term patterns.

In the Post-1979 Sample, the average total number of children born to mothers in each cluster ranges from 2.0 for women in the Full-Time cluster to 2.8 for mothers in the Nonemployed cluster (see Table 2). We do not use subsequent children to retrospectively predict an employment pattern that began prior to their births, although we recognize that women’s fertility will affect employment patterns in later years.

Social Correlates of Maternal Employment Patterns

Table 2 summarizes the human capital, attitudes and cultural models, family experiences, and race/ethnicity of mothers in each of the clusters, using the Post-1979 Sample. Between-group differences in average characteristics are statistically significant (using joint significance tests for categorical variables) for all variables except preference for employment at age 35, beliefs about the ideal number of children, and religious attendance.

Table 3 shows the results of our multinomial logit model using the Post-1979 Sample. The first four columns describe how changes in the predictors are associated with changes in the log odds of falling into a given cluster versus the Full-Time cluster (the omitted category); positive coefficients indicate a positive association with odds of membership in the designated cluster versus the Full-Time cluster. Other comparisons can be obtained by differencing coefficients across columns; the fifth through tenth columns indicate which of these other pairwise comparisons are statistically significant. Given the number of predictors and pairwise comparisons among clusters, some statistically significant differences are likely due to chance. Furthermore, by treating clusters as observed rather than estimated outcomes, our standard errors are understated. Our discussion focuses on what we believe to be the most robust patterns distinguishing clusters. In general, we focus on cluster predictors that statistically significantly change the odds of membership in that cluster relative to at least two other clusters. We recognize that this is somewhat a matter of judgment, so we present the full results in the tables and additional analyses in the online appendix.

Table 4 shows differences in predicted probabilities of cluster membership with the covariate set to the given value versus the reference value. For example, the value of –0.11 in the first cell indicates that, on average, having no high school diploma rather than a high school diploma (and no further education) reduces the predicted probability of membership in the Full-Time cluster by 11 percentage points.

Human Capital

All else equal, having no high school diploma versus a high school diploma is associated with heightened odds of membership in the Nonemployed cluster, increasing the predicted probability of membership in this cluster by 13 percentage points, on average. Holding other covariates constant, having a bachelor’s degree or more rather than a high school diploma is associated with diminished odds of membership in the Early Return cluster, decreasing the predicted probability of membership in this cluster by 6 percentage points, on average.

Higher employment intensity during the two years prior to a first birth is associated with increased odds of membership in the Full-Time group. In the year prior to the birth, compared with not working, working full-time is also associated with increased odds of membership in the Part-Time group relative to the Late Return and Nonemployed groups. All else equal, compared with not working either year, working full-time both years is associated with average increases of 45 and 7 percentage points in the predicted probabilities of membership in the Full-Time and Part-Time groups, respectively, but average declines of 14 and 27 percentage points in the predicted probabilities of membership in the Late Return and Nonemployed groups, respectively.

Attitudes and Cultural Models

Women’s beliefs about the ideal number of children for a family and their preferences for employment at age 35 do not strongly distinguish among the clusters. More traditional gender role attitudes are associated with increased odds of membership in the Nonemployed cluster; having gender role attitudes at the 75th (more traditional) rather than the 25th (less traditional) percentile of the distribution is associated with, on average, a 3 percentage point increase in the predicted probability of membership in the Nonemployed cluster.

Frequent attendance of religious services is associated with diminished odds of membership in the Part-Time group. On average, attending at least weekly or at least monthly but less than weekly are associated with declines of 6 and 5 percentage points, respectively, in the predicted probability of membership in the Part-Time cluster, relative to attending less than monthly.

Mothers’ work patterns are also associated with their own mothers’ characteristics. Having a mother who was employed is associated with heightened odds of membership in the Full-Time group: an average increase of 8 percentage points in the predicted probability. Compared with having a mother with a high school diploma, having a mother without a high school diploma is associated with lowered odds of membership in the Part-Time group: an average decline of 5 percentage points in the predicted probability. Having a mother with a college degree rather than a high school diploma is associated with increased odds of membership in the Nonemployed cluster: an average increase of 14 percentage points in the predicted probability.¹²

Family Experiences

As discussed previously, to describe variation by age across clusters, we rely on the results from the Extended Sample, which includes a smaller set of covariates but does not disproportionately exclude young mothers as our Post-1979 sample does. We find that holding constant race/ethnicity, education, marital status, and the education and employment of the respondent’s mother, being a teen mother—rather than having a first birth between ages 20 and 25—is associated with significantly lowered odds of membership in the Part-Time cluster. Compared with becoming a mother between ages 20 and 25, having a first birth at age 26 or older is associated with heightened odds of membership in the Full-Time cluster and diminished odds of membership in the Early Return cluster (see the online appendix, Table S14).

Returning to the results based on the Post-1979 Sample, we find that being married just prior to the first birth is associated with higher odds of membership in the Part-Time and Early Return clusters relative to the Full-Time and Nonemployed clusters; being married is associated with average increases of 4 and 5 percentage points in the predicted probabilities of membership in the Part-Time and Early Return clusters, respectively, but declines of 7 and 4 percentage points in the predicted probabilities of membership in the Full-Time and Nonemployed clusters.¹³

Race/Ethnicity

Net of other characteristics, being African American as opposed to white is associated with higher odds of membership in the Full-Time group and lower odds of membership in the Part-Time group: an average increase in the predicted probability of membership in the Full-Time cluster of 16 percentage points and an average decrease in the predicted probability of membership in the Part-Time group of 8 percentage points. Being Hispanic rather than white is also associated with diminished odds of membership in the Part-Time group: an average decline of 8 percentage points in the predicted probability.

Subsamples

As shown in Table 5, in the Married Sample, we find no statistically significant associations between spouse’s work hours and cluster membership.¹⁴ However, as expected, higher spouse earnings are associated with lower odds of membership in the Full-Time cluster, all else equal.¹⁵ Changing a spouse’s earnings from the 25th to the 75th percentile of the distribution is associated with an average reduction of 13 percentage points in the predicted probability of membership in the Full-Time cluster.

As shown in Table 6, in the Working Sample, higher prematernity wages are associated with lower odds of membership in the Early Return group. Having a prematernity wage at the 75th rather than the 25th percentile of the distribution is associated with a 3 percentage point decline, on average, in the predicted probability of membership in the Early Return cluster. Prematernity job tenure is negatively associated with the odds of membership in the Nonemployed cluster; having job tenure at the 75th rather than the 25th percentile of the distribution is associated with a 5 percentage point decline in the predicted probability of membership in the Nonemployed group.

Conclusion

We identify five common maternal employment patterns. The representative sequences of these patterns are: permanent disaffiliation from the labor market (Nonemployed); consistent part-time employment (Part-Time); consistent full-time employment (Full-Time); and a multiyear spell of nonemployment lasting either about 6 (Early Return) or about 11 (Late Return) years, followed by an eventual transition to full-time employment. Our sequence analysis shows the variability in how American women currently at midlife balanced paid employment and motherhood.

Our multivariable analyses draw out key prematernity features distinguishing each cluster. The cluster predictors we consider do not strongly distinguish members of the Late Return cluster from those of other clusters, so we focus on the other four clusters. The Full-Time cluster shows economic disadvantage in some respects. All else equal, being African American, unmarried, and (if married) having lower spouse earnings are positively associated with membership in this cluster. At the same time, high prematernity work intensity, being older at the time of the first birth, and having a mother who was employed are also positively associated with membership in this cluster. The mothers in this cluster may need to work to support themselves and their families financially, and they are successful in securing long-term full-time employment and have personal and family histories of employment.

The Early Return cluster is characterized by modest prematernity human capital; low prematernity wages and lacking a college degree are associated with membership in this cluster, holding constant other characteristics. Being married at the time of the first birth is also associated with membership in this cluster, as is having a first birth at younger ages. Rather than investing in human capital while childless and continuing employment after maternity, mothers in this cluster appear to have relatively early family formation and delayed careers, (re)entering the workplace postpartum and then experiencing steady full-time employment.

All else equal, being married prior to the birth, white, and not a teen mother are associated with membership in the Part-Time cluster, as are having a mother who received a high school diploma and working full-time in the year before the birth. These mothers are not particularly traditional: they stand out for their low levels of religious attendance. Mothers in this cluster maintain consistent employment but at reduced intensity. The social advantage of this group is notable because in 2012, a plurality of Americans reported viewing mothers’ part-time employment as “ideal” for young children, and the same is true when mothers of minor children are asked what their ideal work situation would be (Parker and Wang 2013). Whether this reflects that white, married women who become mothers after their teen years are most likely to achieve the most desired work-family balance, or whether the behaviors of this group of mothers shape what is considered ideal by the public at large, the association results in the concentration of advantaged women in the most approved-of and desired employment pattern.

Mothers in the Nonemployed cluster are characterized by weaker attachment to employment prematernity: shorter job tenure and not working in the year before the birth are both associated with membership in this group. This group also shows signs of being internally heterogeneous. Some mothers in this group appear to be disadvantaged. All else equal, lacking a high school diploma is positively associated with membership in this cluster, as is being unmarried. On the other hand, net of other covariates, having a mother with a college degree and holding more traditional gender role attitudes are also positively associated with membership in the Nonemployed cluster. Thus, some mothers in the Nonemployed cluster may prefer not to work, whereas others are unable to find and keep a job.

This study makes two main contributions to the literature on maternal employment: (1) extending the observation window beyond the first few years after a birth to include 18 years of maternity, and (2) considering maternal employment as a connected set of states forming a sequence rather than a single transition or a status at a given moment in time. Previous research studying only a few years postpartum has overlooked distinctions among the Early Return, Late Return, and Nonemployed groups. These distinctions are important to uncover both because of their implications for mothers’ midlife labor market experience and wages and because of the differences in the socioeconomic and family characteristics of mothers in each group. Thus, we urge future research on maternal employment to consider the experiences of mothers beyond the immediate postpartum years. This recommendation can be adopted by scholars using event-history models as well as those using sequence analysis or trajectory-based methods.

Our results highlight that analyses modeling traits associated with more or less paid work by mothers, without considering the pattern this employment takes, will capture some associations well but will miss others. For example, net of other characteristics, race/ethnicity is associated with differences in work intensity among mothers who experience consistent employment; compared with being in the Full-Time group, being white increases odds of membership in the Part-Time group. By contrast, holding other factors constant, we find no strong race/ethnicity differences among the groups characterized by spells out of the labor force (Early Return, Late Return, and Nonemployed). As another example, holding constant other characteristics, having low prematernity wages is associated with heightened odds of membership in the Early Return group relative to groups characterized by either more (Full-Time) or less (Nonemployed) postpartum work experience. These differences cannot be uncovered with models that lack a long-term categorical approach.

Theories of women’s and mothers’ employment have emphasized the roles of preferences and attitudes, the opportunity cost of labor market withdrawal, and husbands’ earnings (or other income sources), and prior research has evaluated how these characteristics are associated with how much women work. Our results suggest that further research is needed to test whether different theories are especially important for explaining different aspects of maternal employment patterns, such as whether to exit the labor force following maternity, whether and when to return, and how much to work while employed. For scholars using event-history methods to study specific employment transitions, these transitions can be studied more holistically by incorporating measures of prior work patterns and allowing interactions between time since maternity and other covariates.

Our analyses, of course, have limitations. They are descriptive rather than causal, and we lack information on some relevant correlates of mothers’ employment patterns, including the price of childcare (Kimmel 1998) and workplace characteristics, such as leave length and ability to avoid mandatory overtime (Glass and Riley 1998). For characteristics we do observe, associations may arise through a variety of pathways. For example, prematernity work experience may be associated with postpartum employment patterns because prior experience builds human capital that makes remaining in the labor force postpartum more attractive, or because women’s tastes for paid work shape their employment decisions both before and after becoming mothers. Furthermore, unlike event-history models, our approach is not designed to identify how time-varying conditions, such as the break-up of a marriage or the birth of a subsequent child, are associated with mothers’ employment. We view sequence analysis and event-history models as useful complements in the study of maternal employment.

Our results are specific to women born between 1957 and 1964 and living in the United States in 1979. The narrow range of birth cohorts prohibits adjudicating between associations with age at first birth versus period of the birth. As later cohorts of mothers complete their child-rearing years, future research will be able to test whether similar employment patterns and similar social correlates of these patterns can be found in more recent cohorts.

In supplemental models, we found that the association between marriage and maternal employment varied by race/ethnicity and education (see the online appendix, Tables S18 and S19), and future research is needed to explore these patterns and other interactions in more detail. One possible extension of our research would be to consider groups of predictors jointly. For example, the Nonemployed cluster may be a heterogeneous group including both women with low levels of human capital who are unable to find or keep a job and more-advantaged women with traditional gender role attitudes who choose to withdraw from the labor force. Analyses designed to identify combinations of characteristics that predict maternal employment patterns could test this possibility. Future research might also test whether the patterning of employment captured by our clusters is associated with later-life wages, children’s well-being, or other outcomes, net of more common measures of employment history, such as work experience and number of work interruptions.

In this article, we provide what we believe is the first description of American mothers’ long-term employment sequences following the birth of their first child, demonstrating considerable heterogeneity in how women combine employment and motherhood. Research that considers only total employment hours, does not consider part-time work separately, or focuses only on the years immediately following a first birth is likely to provide an underdeveloped picture of mothers’ work-family balance throughout the child-rearing years. This incomplete picture will, in turn, limit understandings of how the traits of mothers and their families are associated with their work outcomes, especially when women’s characteristics distinguish more sharply between the patterns of employment than the levels. Future research on mothers’ employment patterns and their later-life consequences will benefit from continuing to consider a rich set of employment statuses, expanding the length of time postpartum that is investigated, and conceptualizing employment statuses as interlinked and embedded within a holistic trajectory.

Acknowledgments

This research was funded in part by an Early Career Research Award from the W.E. Upjohn Institute for Employment Research, and an early version of this manuscript was published as Upjohn Institute Working Paper 15-247 (https://doi.org/10.17848/wp15-247). We are grateful to Siwei Cheng, Margaret Gough, Ian Lundberg, and Demography reviewers and editors for comments on earlier versions of the manuscript. The NLSY79 survey is sponsored and directed by the U.S. Bureau of Labor Statistics and conducted by the Center for Human Resource Research at The Ohio State University. Interviews are conducted by the National Opinion Research Center at the University of Chicago.

Publisher’s Note

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

Notes

References