Peeking Under the Hood of Job Stress: How Men and Women’s Stress Levels Vary by Typologies of Job Quality and Family Composition

Abstract

Changes to work and family norms and polices over the last several decades have reshaped both the job quality and the nature of job and family formation in the United States. Neoliberal policies have generated a slew of flexible-but-precarious working conditions; labor force participation is now the modal path for all genders, regardless of parental or marital status. Leveraging data on 3,419 working men and women from the National Longitudinal Study of Adolescent to Adult Health (Add Health), I use granular measures of job quality to identify distinct job quality–family typologies among both men and women in early-mid adulthood to examine differential implications for psychological and physiological stress. I find four types among men and three among women. Family formation and job prestige appear to differentiate stressful from non-stressful jobs for men; stress outcomes for women are more complex, with job characteristics such as flexibility playing a greater role.

Introduction

The United States employment landscape has transformed over the last few decades, particularly with regards to 1) who works and 2) job quality. With regards to “who”, most U.S. women now participate in the labor force, increasing from 43.3% in 1970 to 57.4% 2019 (Bureau of Labor Statistics, 2022). Further, combining work and parenthood is now nearly as modal a path for women as it is for men: in 2019, 72.3% of all mothers with children under the age of 18 participated in the labor force, compared to 93.3% of all fathers (Bureau of Labor Statistics, 2022). At the same time, job quality has changed. The service sector has boomed as blue collar, manual labor has dwindled, resulting in a preponderance of contingent and non-standard jobs (Kalleberg, Reskin, & Hudson, 2000). Wages have been stagnant since the 1970s (Desilver, 2018) and labor union membership has declined (Desilver, 2018). Finally, the 2007–2009 recession reshaped both employment opportunities and expectations for young adults entering the workforce, with many struggling to find sustainable employment (Aronson, Callahan, & Davis, 2015; Johnson, Sage, & Mortimer, 2012; von Wachter, 2020).

These labor force changes have had important implications for population health and well-being. Jobs matter for health throughout the life course: both genders experience slower declines in health when employed full-time (Ross & Mirowsky, 1995). However, not all jobs are created equal, and jobs characterized by inflexibility (Chandola, Booker, Kumari, & Benzeval, 2019; Karasek, 1979), unpredictable scheduling (Schneider & Harknett, 2019) or negative work-home spillover (Moen, Fan, & Kelly, 2013; Schieman, Milkie, & Glavin, 2009; Schieman, Whitestone, & Van Gundy, 2006) can lead to chronic stress and subsequent health deterioration. Similarly, the gendered nature of the standard employment relationship (SER)—characterized by standardized working times, continuous employment, and a traditional employer-employee contract (Vosko, 2009)--continues to create challenges for working women, particularly mothers. Although full-time, continuous employment has documented health benefits, (Frech & Damaske, 2012; Ice et al., 2020) the nature of traditional employment relationships can make it difficult for women to enter into and remain in these positions. Further, women continue to shoulder the “second-shift” of childcare and domestic labor, making them more vulnerable to work-family conflict and stress (Byron, 2005; Gilbert-Ouimet, Brisson, & Vezina, 2019).

While the literatures regarding job quality, health and gender clearly highlights that 1) precarious, often non-standard employment circumstances are worse for a slew of health outcomes, but 2) traditional employment structures still create conflict for working women (particularly mothers), most research explores the effect of job characteristics in isolation, such as decision latitude (Moen, et al., 2016), hours worked (Lallukka, et al., 2008), job status or fringe benefits (Hill, 2013); or conversely, explores workplace differences by gender, or among women with/without children who are/are not working (Damaske & Frech, 2016; Nomaguchi & Fettro, 2019). However, jobs are rarely defined by a single characteristic, but are in fact determined by bundles of characteristics which tend to cluster together (eg, low wages and non-standard scheduling). Similarly, prior research has demonstrated that family arrangements and employment trajectories inform and shape one another—and that these combinations have consequences for health and well-being, particularly for women (Lippert & Damaske, 2019). Although there is some prior research that takes a multidimensional or typological approach to work and family, to-date most research has focused on either a) work or employment quality (Peckham et al., 2019; Peckham et al., 2022) typologies among older workers and their associations with health outcomes(Eisenberg-Guyot, et al., 2020), or b) typologies of employment status and parental status among women (Lippert & Damaske, 2019) and their implications for stress. Whether and how typologies of job quality and family formation influence health—and whether these typologies might vary by gender—is less clear.

A note on terminology: since the meanings of “work”, “employment” and “job” are not consistent across literatures, I will be leveraging the definitions from the labor literature. In this case, “work” refers to the tasks being performed; “employment” to the contract under which a person carries out said work; and “job” to the combination of work and employment. Within this context, this study uses latent class analyses to identify typologies of job quality and family composition among early-midlife working U.S. men and women. I then explore whether these typologies—stratified by gender—are differentially associated with two important measures of health and well-being: self-reported stress and physiological stress. I ask the following: 1) what patterns of job quality and family formation can be found among contemporary men and women in later young adulthood; 2) how do these typologies associate with men and women’s perceived and physiological stress levels, and 3) in what ways do stress-promoting typologies vary from stress-reducing typologies?

Background

The Multidimensional Relationship Between Work and Stress

Work-stress is an increasingly common phenomena and has serious implications for individual and population health. Work-stress has been linked to numerous negative outcomes, including increased risk of coronary heart disease (Kivimaki, et al., 2006), worker absenteeism (Borritz et al., 2006), and burn-out (Shoman, et al., 2021). Despite the documented health—and economic—risks of work-stress, work-stress continues to be pervasive in the U.S. labor force. The World Health Organization (WHO) recently included psychological burn-out in the 11^th revision of the International Classification of Diseases (ICD-11), specifically identifying it as an “occupational phenomenon” (W.H.O., 2021). A 2022 report from Gallup found that 55% of U.S. workers reported experiencing high levels of stress, making them some of the most stressed workers in the world (Gallup, 2022). And qualitative research has highlighted how even workers employed in objectively “good” jobs—well-paying jobs with stable schedules, social prestige, benefits, etc—are not immune, with work hour- and responsibility-creep resulting in work overload (Kelly & Moen, 2020). In short, work-stress is quickly becoming normative in the U.S.

Despite the wealth of research demonstrating the relationship between work, stress and health, the mechanisms by which work is hypothesized to produce stress vary. For example, higher decision-making latitude—or the control over what one does at work and how one does it—is associated with lower stress over time (Lippert & Damaske, 2019). Long working hours are another mechanism, with longer work hours associated with a higher risk of psychological and physical health problems (Lallukka, et al., 2008; Wong, Chan, & Ngan, 2019); so, too, are unpredictable or precarious schedules (Schneider & Harknett, 2019). Some theoretical frameworks attempt to bring these ideas together, such as the job demands-control (JD-C) model, which theorizes that all jobs can be defined by two domains: high/low job demands, and high/low job control. Combinations of high and low job demands/control result in better or worse health, with high demands and low control placing workers at risk of the poorest health outcomes (Karasek, 1979). However, job demands/control can be conceptualized in a number of ways, and operationalization of the JD-C model frequently focuses on a single variable, leaving us with only part of the picture. Given the myriad mechanisms by which jobs can generate stress—not to mention different forms of stress, e.g. psychological versus physiological—some work-health scholars have begun to take a typological approach to job quality within the context of health. For example, the “Good Jobs, Bad Jobs” framework (Kalleberg A. L., 2011) posits that a “good” job must include 1) relatively high earnings, 2) adequate fringe benefits, 3) worker autonomy or control over their work, 4) flexibility over scheduling, and 5) control over termination of one’s job. The Employment Quality framework (EQ) conceptualizes job quality similarly, determined by seven different domains: 1) employment stability; 2) material rewards; 3) worker’s rights and social protection; 4) working time arrangements; 5) employability opportunities; 6) collective organization; and 7) interpersonal power relations (Julià, 2017; Peckham et al., 2019; Van Aerdan et al., 2014).

While the “Good Jobs/Bad Jobs” and EQ framework advance a more nuanced and dynamic approach to jobs and health, to-date much of the research using a typological approach has either been conducted in the European setting (Julià, 2017; Van Aerdan et al., 2014; Van Aerden et al., 2016), or among older U.S. workers (Andrea et al., 2021; Eisenberg-Guyot, et al., 2020; Peckham et al., 2019); less clear is how these typologies might form for workers who entered the labor force during the Great Recession, or how gender and family formation might be instrumental in shaping these typologies.

Gender, Family Formation and Work

Despite advances in gender equity in- and out- of the home, structural hurdles continue to make it difficult for women to enjoy career trajectories comparable to those of their male counterparts. As many feminist scholars have pointed out, the way work is organized, conceptualized and institutionalized is inherently gendered: “formal work” is siloed from other responsibilities, such as caregiving at home, and it is taken for granted that the worker will prioritize formal work (Acker, 1990). Further, even in dual-earner households, women continue to shoulder much of the responsibility of domestic work, particularly when it comes to managing childrearing (Clawson & Gerstel, 2014; Gerson, 2010; Hochschild & Machung, 2012). This relationship is bidirectional, with mothers also falling under greater scrutiny at work: a robust literature has demonstrated that women with children are likely to experience a “Motherhood Penalty” in the labor market when compared to both their male- and their childless female-counterparts (Fuegen et al., 2004; Yu & Kuo, 2017), and that women in high-skill high-wage occupations may receive the greatest motherhood penalty (England et al., 2016). In contrast, men appear to earn a “Fatherhood Premium”, whereby their wages actually increase as a function of having children—although this effect may be limited to married, biological fathers (Killewald, 2013; Zhang & Soderberg, 2022; Yu & Hara, 2021).

One would imagine that this structural disadvantage would result in greater work stress and poorer health among women, particularly mothers; research suggests a more complex relationship. On average, working mothers experience better long-term health than mothers who opt out of the workforce or work part-time (Damaske & Frech, 2016), and that consistent work in earlier life is associated with greater longevity (Caputo, Pavalko, & Hardy, 2020). Similarly, women report greater happiness and experience lower levels of cortisol when at work than when at home (Damaske, Smyth, & Zawadzki, 2014), regardless of whether or not they have children. However, this relationship is likely moderated by work characteristics: flexible work hours are associated with lower allostatic load for both male and female workers but appear to be particularly beneficial for women with greater caregiving duties. Conversely, jobs characterized by high demands and low control are associated with a higher likelihood of major depression in male workers, but not female workers (Wang et al., 2008).

Along with parental status, marital (or partner) status is also likely to differentially shape health outcomes across gender and job characteristics. A life course framework—specifically with a focus on individual agency and “constrained choices” (Bird & Rieker, 2008)—posit that people are embedded in meaningful social networks which shape both their life trajectories and their wellbeing, and that individuals’ agency is constrained by the historical, social and geographic places in which they find themselves (Elder Jr, Kirkpatrick Johnson, & Crosnoe, 2003; Bird & Rieker, 2008). Through this lens—and given that most contemporary family and career tempos are still shaped by a breadwinner/homemaker dynamic (Acker, 1990)—one might expect that marriage might benefit men’s (or constrain women’s) job prospects. Indeed, married (and to an extent, cohabitating) men experience higher wages and even work-related favoritism than their un-partnered counterparts: the so-called “marriage premium” (McDonald, 2020; Cohen P. N., 2002). Additionally, work-and-marital contexts combine to shape health behaviors (and subsequent outcomes). Research suggests that, particularly for men working long hours, the work-schedule flexibility of their wives only somewhat shaped their odds of engaging in health-promoting behaviors; in contrast, women’s odds of engaging in health-promoting behaviors was heavily dependent on the flexibility of their husband’s schedule (Fan, et al., 2015). Inequalities occur outside the confines of marriage as well: marital dissolution appears to place women at greater risk of economic precarity (Leopold T., 2018; Chanda, 2022), particularly so if the divorced couple has children (Leopold & Kalmijn, 2016).

This variation suggests that, much like the work-stress literature, this field might benefit from a typological approach, in which gender, marital status, parental status and employment characteristics are jointly taken into account. Some steps have already been taken in this regard: using Wave IV of the Add Health data, Lippert & Damaske (2019) identified a work-family typology among young adult women consisting of seven different latent classes. Further, they found significant differences in both self-reported and physiological stress outcomes as a function of class membership: women who were without full-time work, a partner(s) or child(ren) reported the highest levels of perceived stress; the lowest levels of perceived stress were reported by professional workers engaged in full-time work, regardless of whether they had children or not. However, this study focused primarily on higher-level job quality-family characteristics—such as occupational classification or employment status—and restricted its sample to young women.

Current Study

Both the work-stress literature and the gender-family-work literature highlight the importance of work quality for health and well-being. However, while the work-stress literature excels by including rich and multidimensional measures of jobs and employment quality, it often overlooks variation by gender and family composition. Similarly, studies of gender, family and work highlight the issues of gendered institutions and role conflict, but are often limited in their operationalization of job quality. I aim to bring the nuance of these two literatures together, to identify more granular job quality-family typologies among men and women in a similar life stage. Building on the work of Lippert and Damaske (2019), I ask what job quality-family typologies might be found among working men and women in their mid-30s and early 40s, a time when both family and career paths have likely solidified. I examine these typologies in a sample of workers who entered the labor force around the time of the Great Recession, a period of time in which jobs and work expectations were entirely reshaped. Finally, I aim to identify whether these typologies are differentially associated with worker stress, and whether distinct patterns emerge as a function of gender.

Data and Methods

Data

Data for this study are from Wave V of the National Longitudinal Study of Adolescent to Adult Health (Add Health). Add Health is a nationally representative survey begun in 1994 of adolescents then in grades 7–12, who have been followed through their transition to adulthood. By Wave V, 12,300 respondents have reached the early years of mid-adulthood and are between the ages of 33 and 43 years: old enough to be engaged with family and career formation, but young enough to have entered the labor market during the Great Recession. Respondents in Wave V were excluded from the sample if they did had not participated in biomarker data collection (n=6,923); were not working (n=926); did not have data for the hs-CRP measure (n=644); were missing data on their occupational code (n=72), or were coded as “unclassified” (n=15); were in the military (n=4); and who were missing data on covariates, including personal income (n=15), health insurance (n=15), retirement benefits (n=9), paid time off (n=5), job decision latitude (n=5), race (n=7), gender (n=9), educational attainment (n=5), parental status (n=3), weekly work hours (n=3), number of jobs held (n=2), job tenure (n=16), work sector (n=23) and partner status (n=3). Additionally, respondents who were missing data for the stress scale were dropped (n=38) as were pregnant persons (n=139) since pregnancy may bias both forms of stress. My final sample is comprised of N=3,419 persons, 1,934 women and 1,485 men. Although Add Health includes some measures of job quality, they are somewhat spare—as such, data from the Occupational Information Network (O*Net) are cross-walked with the occupational codes used in Add Health to provide a more robust measure of job quality.

Outcomes of Interest: Self-reported and Physiological Stress.

I measure a single dimension of health—stress—via both a perceived and physiological lens, specifically 1) self-rated stress, and 2) high-sensitivity C-Reactive Protein (hs-CRP). Prior studies regarding stress, work and family have shown that perceived stress and physiological measurements of stress are not always aligned, and that stress-experiences for workers may be moderated by 1) parental status 2) marital status and 3) income (Damaske et al., 2014; Robinson, Magee, & Caputi, 2016). Self-rated stress is measured using an abbreviated version of Cohen’s Perceived Stress Scale (PSS) (Cohen, Kamarck, & Mermelstein, 1983). Respondents are asked to rate, on a scale from never (=0) to very often (=4) how often in the prior 30 days they felt 1) they were unable to control the important things in their life; 2) confident in their ability to handle their personal problems; 3) that things were going their way; 4) that difficulties were piling up so high that they could not overcome them. Responses were coded such that a higher number indicated higher levels of stress. The stress scale is discrete and continuous, ranging from 0 (the lowest possible score) to 15 (the highest possible score). While the stress scale has acceptable kurtosis, it is left-skewed. However, I believe it remains appropriate to use OLS regression despite the non-normal distribution of the stress variable. Post-estimation tests (kernel density plot, inter quartile range test) revealed that the errors were only modestly deviated from a normal distribution; that there were no severe outliers; and that the distribution is relatively symmetric. I conducted these tests for both genders. Physiological stress is measured using high-sensitivity C-reactive Protein (hs-CRP). hs-CRP is a measure of low-grade, systemic inflammation, and has previously been linked to psychosocial stress (Woojae, et al., 2016). Add Health collected venous blood from a consenting subset of wave V participants, which was then analyzed by the Laboratory for Clinical Biochemistry Research (LCBR) at the University of Vermont. Hs-CRP is measured dichotomously, with respondents classified as either low or average (up to 3 mg/L) versus high (>3 mg/L) (Whitsel, et al., 2002).

Independent Measure: Job Quality Typologies.

My independent measure combines a vector of job and family characteristics to form a typology of job quality and family. I use latent class analysis (LCA) to identify the distinct types present in my typologies, stratified by gender. LCA identifies response patterns across a specified and theoretically associated set of questions with the intention of identifying latent “classes” in the data. Variables included in my independent measure are a function of prior research exploring the implications of 1) job quality, 2) family composition and 3) gender for health and wellbeing within the context of work. My choice of job characteristics is driven by the Employment Quality framework and the “Good Jobs/Bad Jobs” framework. As previously mentioned, “job” in this context is an umbrella term for both the employment contract and work characteristics. Given that EQ is the slightly more detailed framework, I use this as my primary guide for variable selection, guided by its seven domains: employment stability, material rewards, workers’ rights and social protection, working time arrangements, employability opportunities, collective organization, and interpersonal power relations (Peckham et al., 2019; Van Aerdan et al., 2014). In keeping with prior work, I represent these domains via proxy variables. A complete breakdown of the dimensions, their proxies and the data sources (Add Health vs. O*Net) is presented in Figure 1. I use ten proxies to represent six of the seven dimensions of EQ to identify the job quality facet of my typology. Proxies for eight of these dimensions were identified in the Add Health data; two were merged in from the O*NET. An in-depth description of the O*NET-to-Add Health crosswalk process is available upon request. It should be noted that the domain of collective organization is excluded from the analysis due to the fact that the proxy—reception of healthcare insurance from a union—was too sparely represented in the final sample. While this is a limitation, I would argue that it does not hinder the analysis too much, as collective organizing has declined in the U.S. (Bureau of Labor Statistics, 2022). Additionally, the O*Net variable for schedule stability should be understood as macro-level (rather than daily) schedule stability. For example, in the O*Net, construction workers reported a schedule that was seasonal 13% of the time, and irregular/contract based/needs based 53% of the time; as such, this particular job is coded as having an unstable schedule. In comparison, waiters and waitresses reported a schedule that was irregular 23% of the time, and regular, with an established schedule and routine, 63% of the time. As a result, job is coded as having a stable schedule. There are both strengths and weaknesses to this approach, which I discuss in the limitations section. Finally, as a result of this limitation, I round out my dimension of employment stability with two additional measures: multiple-job holding and employment sector. Prior research has shown that multiple-job holders experience greater physical and psychological stress; there is also precedence for using this measure to capture precarity (Campion, Caza, & Moss, 2020; Marucci-Wellman, Willetts, Lin, Brennan, & Verma, 2014; Koryani, Jonsson, Ronnblad, L, & T., 2018). Employment sector can also serve as an indirect indicator of employment stability: private sector jobs are on average more precarious than public sector jobs (Kopelman & Rosen, 2016), and public sector workers are far likelier to be represented by a union (37.6% compared to 7% in the private sector) (Shierholz, Poydock, Schmoitt, & McNicholas, 2023). Additionally, self-employment can either signal precarity or greater stability, depending on the context in which it is embedded, such as wages, autonomy, etc. (Smeaton, 2003). Finally, “full time” hours are informed by the Internal Revenue Service (Internal Revenue Service, 2022); “over-time” hours are informed by the Department of Labor (U.S. Department of Labor, 2023); and “health hazard hours” are informed by findings from the World Health Organization and the International Labor Organization that 55+ hours increases risk of cardiovascular disease (World Health Organization, 2021).

Figure 1.

Dimensions, Variables, and Operationalizations of Employment Quality.

Note: Add Health = National Longitudinal Study of Adolescent to Adult Health; O*Net = Occupational Information Network.

Gender and family composition are also central to my typologies. Gender was coded dichotomously as male (=0) or female (=1). While it is important to recognize that there are other gender identities, the data in Wave V of Add Health does not capture an array of gender identities. I operationalize family formation as partnership status and parental status. As discussed in the background section, partner status (including marital dissolution) can shape both health outcomes and job opportunities differently for men and women. As such, examine partnership at a fine-grain level, specifically married, never divorced (=0); divorced, re-partnered (includes a new spouse or cohabitating partner) (=1); divorced, living alone (=2); never married, cohabitating (=3); never married, living alone (=4). Parental status is a binary, coded as having any children (=1) or not (=0). While prior research has examined parity (Lippert & Damaske, 2019), given the focus on the literature on the “motherhood penalty”/“fatherhood premium”, I ultimately opted to focus on the role of parenting, rather than the number of children.

Other Covariates.

I include a number of relevant controls in my analysis. As college education is associated with both better employment opportunities and better health (Zajacova & Lawrence, 2018), I include a binary indicator of those with/without a college education. Whiteness in the United States is associated with a number of economic and social structural privileges which, in turn, shape employment and health (Weisshaar & Cabello-Hitt, 2020; Wilson, 2003), so I include measures of respondent racial and ethnic identity, specifically dummy variables for White, Black, Asian/Pacific Islander, Native American/other. I also include a control for Hispanic ethnicity. Age may shape childbearing, so I control for age. I control for job tenure, as it may inform to what extent men and women experience a “motherhood penalty” or “fatherhood premium” (Yu & Hara, 2021). In order to disentangle current health from childhood health, I include a measure of self-reported health collected at Wave I of Add Health, coded as fair or poor (=0), good (=1), very good or excellent (=2), or not reported (=3). Finally, since hs-CRP is sensitive to infection or the use of anti-inflammatory medications, binary variables for whether the respondent was currently taking an anti-inflammatory medication and whether the respondent reported having an infection or inflammatory disease within the past four weeks were included.

Analytic Strategy

Stage 1: Latent Class Analysis.

I first identify the gender-stratified job quality-family typologies present in my data. Latent class analysis—or LCA—is a finite mixture method for detecting unobserved and discrete constructs (“latent classes”) that the researcher believes to be represented in the data (Lanza et al., 2007). LCA uses observed, categorical indicators believed to share a common conceptual identity to classify a sample into mutually exclusive and exhaustive qualitative categories (Lanza, Bray, & Collins, 2013). LCA requires the researcher to have as strong theoretical justification for their presumed classes; I anchor my assumptions in both empirical precedence and existing theoretical frameworks. Empirically, there is precedence for exploring typologies of job quality, family composition and gender: LCA has previously been used to identify typologies of employment quality (Peckham et al., 2019; Peckham, et al., 2022), typologies of work and family orientations among adolescents as a function of gender (Hayford & Halliday, 2021), and work-and-family composition among women (Lippert & Damaske, 2019). Theoretically, I rely on two separate literatures. With regards to the job quality component of my typology, I turn to the literature regarding 1) Employment Quality and 2) “Good” versus “Bad” jobs. Both of these literatures argue that jobs, broadly defined, should be conceptualized as a bundle of characteristics which work in tandem to inform quality. Central themes are the precarity (stability) of the job (e.g. hours worked, schedule stability) and the material benefits provided by the job (e.g. benefits, income). The characteristics included in my typology draw from both of these literatures to address these broader themes. Finally, as discussed in the background section, theories of role-strain and the stress-process inform my choice to include gender and family composition as meaningful dimensions in this typology.

I use Stata’s gsem LCA package to identify my typologies. First, I stratified my sample by gender, in order to run separate analyses for men and women. I stratify my classes by gender because prior literature has demonstrated that family formation and work experiences vary along these lines: e.g., the fact that women still manage the “second shift” (Hochschild & Machung, 2012) of domestic housework, or that men tend to work longer hours (Moen & Yu, 2000). However, I do run a supplementary analysis on a combined sample of men and women to specifically explore how (if at all) gender interacts with job quality-family types to inform stress—results of these analyses can be found in the supplementary material.

For both men and women, I took an iterative approach to class construction. I began with a single-class model, adding additional classes until the best-fitting model was identified. Among men, one-, two- three- and four- class models were fit: the four-class model fit the data the best, with the smallest AIC and BIC. Among women, a three-class model fit the data best. Likelihood-ratio tests for both samples confirmed these models fit the data at least as well as the saturated model. Finally, I calculated posterior probabilities of membership in order to assign each individual in the data to a specific class. Relative entropy (E_K), an index of classification fit for the data, was calculated for posterior probabilities for the respective samples: among men, E_K= .96, and among women, E_k= .98. E_k is bounded by zero and one, with a value of one indicating that there is perfect posterior classification for all sample observations (Masyn, 2013); as such, I concluded that the classes in both samples were well-differentiated and defined.

Stage 2: Linear and Logistic Regressions.

The second stage of my analytic process involved regressing self-reported stress and hs-CRP on class membership. I regressed self-reported stress on just the work quality-family types, separately for men (1a) and women (1b). Covariates were then added and the full models were run (2a and 2b). Due to the dichotomous nature of the outcome, logistic regression was used to regress hs-CRP level on just the types (3a and 3b), and then the full-model (4a and 4b). Cross-sectional survey weights for Wave V were applied to all models, including the latent class analyses.

Results

Descriptive Results

A full set of descriptives are presented in Table 1. Married, never divorced was the modal status (60%); 17% of the sample reported being never married, living alone; 10% were divorced and re-partnered; 7% never married and living alone; and 6% were divorced and living alone. The majority of respondents were parents (70%) and women were slightly more likely to be parents than men (75% vs. 65%). Men were more likely to be employed in a job which allowed them to utilize their strengths (43% vs. 36%); provided retirement pensions (79% vs. 75%), paid time off (84% vs. 81%) or health insurance (83% vs 78%); and had irregular scheduling (16% vs. 7%). Men were more likely to report decision making latitude at work most-to-all of the time than were women (71% vs. 61%) and were more likely to report making 75k or more per year (34% vs. 19%), but also more likely to work over 55 hours a week (43% vs. 36%). Proportions of multiple job-holding and employment sector were relatively similar by gender. Finally, women were more likely to have an above-average hs-CRP value (43% vs. 25%) and a higher self-reported stress score (5.13 vs. 4.42).

Table 1.

Descriptive Statistics (Weighted), N = 3,419.

	All (N = 3,419)	Men (n = 1,485)	Women (n = 1,934)	Chi-Square test
High hs-CrP class	.34	.25	.43	χ2(1) = 120.46, ρ =.00
Stress score	4.79	4.42	5.13	χ2(15) = 80.42, ρ = .00
Working hours
Part-time: 0–29/week	.08	.03	.13	χ2(3) = 233.38,
Full-time: 30–40/week	.47	.42	.52	ρ =.00
Overtime: 41–54/week	.31	.36	.26
Health hazard: 55+/week	.13	.19	.08
Job promotes skill use	.39	.43	.36	χ2(1) = 19.99, ρ =.00
Personal income
< $25,000	.18	.11	.24	χ2(3) = 155.10,
$25,000–$49,999	.32	.29	.33	ρ =.00
$50,000–$74,999	.24	.25	.22
$75,000+	.27	.34	.19
retirement benefits	.77	.79	.75	χ2(1) = 8.51, ρ =.04
Paid time off	.83	.84	.81	χ2(1) = 5.48, ρ =.11
Health insurance	.80	.83	.78	χ2(1) = 16.29, ρ =.01
High decision latitude	.66	.71	.61	χ2(1) = 35.83, ρ =.00
Multiple job holder	.14	.14	.14	χ2(1) = .22, ρ =.74
Employment sector
Private	.73	.73	.74	χ2(2) = 2.58,
Public	.18	.17	.18	ρ =.53
Self-employed	.09	.10	.08
regular/predictable work schedule	.89	.84	.93	χ2(1) = 74.55, ρ =.00
Partnership status Married: never divorced	.60	.60	.59	χ2(4) = 5.11,
Divorced, widowed, separated: repartnered	.10	.09	.11	ρ =.69
Divorced, widowed, separated: not cohabitating	.06	.07	.06
Never married: cohabitating	.07	.07	.07
Never married: not cohabitating	.17	.17	.17
respondent is a parent	.70	.65	.75	χ2(1) = 40.49, ρ =.00
racial identity	.23	.23	.23	—
White	.77	.77	.77
Black	.14	.14	.15
Asian/Pacific Islander	.04	.04	.04
Native American/other	.03	.05	.01
Ethnicity: Hispanic	.10	.11	.09	—
Mean age	37.63	37.85	37.43	—
Job tenure	6.48	6.22	5.76	—
College education or more	.47	.43	.51	—
Childhood health
Poor/fair	.06	.06	.06	—
Good	.23	.20	.26
Very good/excellent	.42	.42	.43
Childhood health not reported	.28	.31	.26

Note: Data from National Longitudinal Study of Adolescent to Adult Health. hs-CrP = high sensitivity C-reactive protein.

Latent Classes

Tables 2 and 3 present the findings from my latent class analyses, stratified by sex. I identified a total of seven latent classes of job quality-family composition: four for men, and three for women.

Table 2.

Item response Probabilities for Job Quality/Family Classes, Men, N = 1,485.

	High Pressure, Flexible: Married	High Security, Stability:	Low Control, Precarious: Childless	High Prestige, Security: Married Dad
	Dad (.10)	Unmarried Man (.25)	Bachelor (.11)	(.54)
Working hours
0–29 hours/week	.13	.01	.12	.00
30–40 hours/week	.30	.48	.64	.37
41–54 hours/week	.16	.45	.18	.44
55+/week	.41	.16	.06	.19
Job allows worker to use skills
Yes	.64	.50	.42	.62
No	.36	.50	.56	.38
Personal income
< $25,000	.17	.10	.51	.02
$25,000–$49,999	.29	.39	.49	.21
$50,000–$74,999	.14	.31	.00	.30
$75,000+	.40	.20	.00	.46
Multiple job holder	.26	.14	.18	.10
Work: retirement benefits
Yes	.11	.98	.14	.96
No	.89	.02	.86	.04
Work: paid time off
Yes	.20	.96	.47	.99
No	.80	.04	.53	.01
Work: health insurance
Yes	.07	.99	.39	.98
No	.93	.01	.71	.02
Employment sector
Private	.30	.77	.88	.76
Public	.00	.20	.03	.22
Self-employed	.70	.02	.08	.02
Work-decision latitude
None to some of the time	.04	.37	.60	.23
Most to all of the time	.96	.63	.40	.77
regular/predictable work schedule
Yes	.62	.86	.78	.88
No	.38	.14	.22	.12
Marital status
Married: never divorced	.65	.05	.30	.91
Divorced, widowed, separated: repartnered	.06	.24	.05	.04
Divorced, widowed, separated: living alone	.10	.08	.08	.05
Never married: cohabitating	.04	.19	.14	.01
Never married: living alone	.14	.44	.42	.00
Parent
Yes	.68	.21	.40	.89
No	.32	.79	.60	.11

Note: Data from National Longitudinal Study of Adolescent to Adult Health.

Table 3.

Item response Probabilities for Job Quality/Family Classes, Women, N = 1,934.

	Flexible, Unpredictable: Married Mother (.09)	Precarious, Low Pay: Partnered Mother (.17)	High Security, Stability: Partnered Mother (.74)
Working hours
0–29/week	.41	.42	.04
30–40/week	.22	.43	.57
41–54/week	.14	.11	.31
55+/week	.16	.04	.08
Job allows worker to use skills
Yes	.73	.56	.65
No	.27	.44	.35
Personal income
< $25,000	.47	.64	.12
$25,000–$49,999	.22	.24	.38
$50,000–$74,999	.14	.07	.27
$75,000+	.16	.04	.23
Multiple job holder	.29	.15	.12
Work: retirement benefits
Yes	.06	.26	.94
No	.94	.74	.06
Work: paid time off
Yes	.07	.45	.99
No	.93	.55	.01
Work: health insurance
Yes	.06	.21	.99
No	.94	.79	.01
Employment sector
Private	.27	.93	.75
Public	.00	.07	.23
Self-employed	.73	.00	.02
Work-decision latitude
None to some of the time	.07	.49	.40
Most to all of the time	.93	.51	.60
regular/predictable work schedule
Yes	.78	.93	.95
No	.22	.07	.05
Parental status	.81	.80	.72
Partnership status
Married: never divorced	.68	.54	.59
Divorced, widowed, separated: repartnered	.08	.09	.13
Divorced, widowed, separated: living alone	.05	.07	.06
Never married: cohabitating	.06	.13	.05
Never married: living alone	.13	.17	.17

The classes I identified in my male sample were the Low Control, Precarious: Childless Bachelor (.10); the High Security, Stability: Unmarried Man (.23); the High Prestige, Security: Married Dad (.58); and the High Pressure, Flexible: Married Dad (.09) (Table 2). The Low Control, Precarious: Childless Bachelor had a low probability of a job with benefits, the lowest probability of decision latitude and skill utilization at work, and the highest probability of making less than $25k/year and working in the private sector. This class was also typified by a high probability of being unmarred (either living alone or cohabitating) and a low probability of having children. In comparison, although The High Security, Stability: Unmarried Man class also had a low probability of children. However, this class was more likely to be never married and living alone or divorced, etc. and re-partnered. Further, this class had a high probability of fringe benefits, a moderate probability of decision latitude at work, and a higher probability of working in the public sector. This class also had the highest probabilities of falling in the middle of the income range. Meanwhile, the High Prestige, Security: Married Dad class had the highest probability of making $75,000 or more a year, with high probability of fringe benefits, decision-making latitude and skill utilization at work. This class also had the highest probability of being married, never divorced and having children. Finally, the High Pressure, Flexible: Married Dad class had the highest probability of being self-employed, with the lowest probability of reporting a stable, regular schedule, and the highest probability of working “health hazard” hours. However, this class also had the highest probability of decision latitude at work and skill utilization, and a high probability of reporting an income towards the top of the spectrum. This class had a moderate probability of being married with children.

Among women the classes I identified were the Flexible, Unpredictable: Married Mother (.09); the Precarious, Low-Pay: Partnered Mother (.17); and the High Security, Stability: Partnered Mother (.74) (Table 3). The Flexible, Unpredictable: Married Mother had the highest probabilities of no fringe benefits and self-employment. This class had the highest probability of high decision-latitude and skill utilization at work, but the most mixed probabilities of all the classes when it came to hours worked. This class also had the highest probability of being married with children. Similarly, the Precarious, Low-Pay: Partnered Mother had a high probability of having children, but mixed probabilities with regards to relationship status, with the highest probability of being never married, cohabitating among the classes. This class had the highest probability of working part-time in the private sector and making $25,000 a year or less; this class also had low probabilities of reporting fringe benefits, and the lowest probabilities of decision latitude or skill utilization at work. Finally, the High Security, Stability: Partnered Mother had the highest probabilities of working 30–40 hours a week at a job with fringe benefits and making a moderate salary. This class also had the highest probability of working in the public sector. While still high, of the classes this class had the lowest probability of having children, and the highest probability of being divorced, etc. but re-partnered.

Multivariate Analyses

For both men and women, I conducted a series of regression models to examine the relationship between the classes and two measures of stress—hs-CRP and self-reported stress. The first set of models report the results for men. With regards to self-reported stress, compared to the High Prestige, Stability: Married Dad class, the Low Control, Precarious: Childless Bachelor and the High Security, Stability: Unmarried Man classes were associated with significantly higher stress scores (Table 4). These relationships were attenuated somewhat with the addition of covariates, but remained significant: the Low Control, Precarious: Childless Bachelor class was associated with a stress score that was on average 1.53 points higher than the High Prestige, Stability: Married Dad class (SE: .40), while the High Security, Stability: Unmarried Man class was .62 points higher (SE: .27). There was no difference between the High Pressure, Flexible: Married Dad and the High Prestige, Stability: Married Dad classes (Figure 2). When the reference class was rotated, analyses revealed significant differences in stress scores between additional classes. Compared to the Low Control, Precarious: Childless Bachelor class, the High Pressure, Flexible: Married Dad class was associated with a stress score that was 1.92 points lower (SE: .49), and the High Security, Stability: Unmarried Man class had a .91 point decrease in stress score (SE: .44). Finally, when compared to the High Pressure, Flexible: Married Dad class, the High Security, Stability: Unmarried Man class was associated with a 1.01 point increase in stress score (SE: .39) (Figure 3). With regards to odds of physiological stress, the only significant class difference I found was between the Low Control, Precarious: Childless Bachelor class and the High Prestige, Stability: Married Dad class: namely, the former was associated with 1.82 higher odds of clinically high inflammation compared to the latter (SE: .50).

Table 4:

Self-reported Stress and hs-CRP Class: Men (N=1,485)

	Self-reported Stress Score: Coefficients and SE		High hs-CRP class: Odds Ratio and SE
	Model 1a	Model 2a	Model 3a	Model 4a
Latent Classes (Ref: “High Prestige, Security: Married Dad”)	-	-	-	-
Low Control, Precarious: Childless Bachelor	1.76 (.40)***	1.52 (.40)***	1.97 (.54)*	1.83 (.50)*
High Security, Stability: Unmarried Man	.80 (.28)**	.60 (.26)*	1.21 (.26)	1.10 (.26)
High Pressure, Flexible: Married Dad	−.35 (.33)	−.40 (.33)	1.02 (.30)	.97 (.28)
Racial identity: Black	-	.16 (.27)	-	1.22 (.30)
Racial identity: Asian/Pacific Islander	-	.98 (.63)	-	.71 (.57)
Racial Identity: Native American/Other	-	.66 (.51)	-	1.62 (.67)
Ethnicity: Hispanic	-	−.29 (.36)	-	.57 (.16) †
Age	-	−.07 (.05)	-	1.05 (.05)
Job Tenure	-	−.00 (.00)	-	1.00 (.00)
Has college degree or more	-	−.27 (.20)	-	.80 (.15)
Childhood health (ref: poor or fair)	-	-	-	-
Good	-	−.26 (.49)	-	.59 (.23)
Very Good or Excellent	-	−.16 (.46)	-	.39 (.14)*
Childhood health not reported	-	−.43 (.47)	-	.46 (.18)*
Recent inflammation medication use	-	-	-	1.43 (.28)†
Recent inflammatory infection/disease	-	-	-	1.19 (.23)

^†p<.10

^*p<.05

^**p<.01

^***p<.00

Figure 2.

Stress Score across Job Quality–Family Classes: Men.

Note: reference group = High Prestige, Security: Married Dad. N = 1,485.

Data from National Longitudinal Study of Adolescent to Adult Health.

Figure 3.

Odds of High Inflammation across Job Quality–Family Classes: Men.

Note: reference group = High Prestige, Security: Married Dad. N = 1,485.

Data from National Longitudinal Study of Adolescent to Adult Health.

Turning to women, when compared to the High Security, Stability: Partnered Mother class and the Flexible, Unpredictable: Married Mother, the Precarious, Low Pay: Partnered Mother class was associated with higher stress scores (Table 5). These relationships were slightly attenuated with the addition of covariates, but remained significant: the Precarious, Low Pay: Partnered Mother class was associated with a .69 point increase in stress score (SE: .36) compared to the Flexible, Unpredictable: Married Mother class, and a .57 point increase in stress score compared to the High Security, Stability: Partnered Mother class (Figure 4). With regards physiological stress, only the Flexible, Unpredictable: Married Mother class was associated with significantly lower odds of high inflammation. When compared to the Flexible, Unpredictable: Married Mother class, the High Security, Stability: Partnered Mother class was associated with 2.16 higher odds of clinically high inflammation (SE: .60); the Precarious, Low Pay: Partnered Mother class was associated with 2.54 higher odds of clinically high inflammation (SE: .61). There was no significant different when comparing the High Security, Stability: Partnered Mother class and the Precarious, Low Pay: Partnered Mother class (Figure 5).

Table 5:

Self-reported Stress and hs-CRP Class: Women (N=1,934)

	Self-reported Stress Score: Coefficients and SE		High hs-CRP class: Odds Ratio and SE
	Model 1b	Model 2b	Model 3b	Model 4b
Latent Classes (Ref: “High Security, Stability: Partnered Mother”)	-	-	-	-
Flexible, Unpredictable: Married Mother	−.08 (.27)	−.12 (.27)	.37(.09)***	.39(.09)***
Precarious, Low-Pay: Partnered Mother	.93 (.29)**	.60 (.29)*	1.00 (.17)	.83 (.15)
Racial identity: black	-	.46 (.25)*	-	1.61 (.27)**
Racial identity: Asian/Pacific Islander	-	−.05 (.43)	-	.93 (.31)
Racial Identity: Native American/Other	-	−.16 (.72)	-	.99 (.47)
Ethnicity: Hispanic	-	.26 (.33)	-	.93 (.21)
Age	-	.03 (.05)	-	.94 (.03) †
Job Tenure	-	−.00 (.05)	-	1.00 (.00)
Has college degree or more	-	−.98 (.00)***	-	.59(.08)***
Childhood health (ref: poor or fair)	-	-	-	-
Good	-	−.72 (.43)	-	.59 (.23)
Very Good or Excellent	-	−.71 (.43)†	-	.39 (.14)*
Childhood health not reported	-	−1.26 (.44)**	-	.46 (.18)*
Recent inflammation medication use	-	-	-	1.04 (.15)
Recent inflammatory infection/disease	-	-	-	1.56 (.22)**

^†p<.10

^*p<.05

^**p<.01

^***p<.00

Figure 4.

Stress Score across Job Quality–Family Classes: Women.

Note: reference group = High Security, Stability: Partnered Mother. N = 1,934.

Data from National Longitudinal Study of Adolescent to Adult Health.

Figure 5.

Odds of High Inflammation across Job Quality–Family Classes: Women.

Note: reference group = High Security, Stability: Partnered Mother. N = 1,934.

Data from National Longitudinal Study of Adolescent to Adult Health.

Discussion

Prior literature has demonstrated that certain job attributes—such as long hours (Lallukka, et al., 2008; Wong et al., 2019), schedule instability (Schneider & Harknett, 2019), or low decision-making latitude (Karasek, 1979; Lippert & Venechuk, 2020)—have deleterious effects on worker stress; however, these characteristics have often been examined separately, which overlooks the multidimensional nature of jobs. Further, both gender and family composition have implications for employment options (and vice versa), with varying implications for worker stress and wellbeing (Frech & Damaske, 2012; Fuegen, Biernat, Haines, & Deaux, 2004; Ice et al., 2020). Finally, norms around gender and the workforce continue to evolve, with a greater number of job-family configurations slowly becoming available to men and women. Similarly, employment configurations have diversified and grown more complex—particularly in the wake of the Great Recession (von Wachter, 2020)—as the service sector has grown, gig work has become more common, and technology has enabled white collar workers to always be semi-on call (Vallas, 2012). As such, examining the relationship between worker stress and the current job quality-family configurations of adults who entered the labor force during the Great Recession provides a more detailed and contemporary understanding of how jobs, family and gender roles continue to evolve.

This study extends the existing literature on job quality, family formation, gender and stress in two important ways. First, I bring together the job quality literature and the gender-work-family literature on a more granular level to identify how job quality-family configurations may vary by gender. Second, I add to the stress literature by highlighting within-gender, between-class variation in perceived vs. physiological stress. I build on the recent work of Lippert & Damaske (2019) by a) including a job quality-family typology for men, 2) expanding the “jobs” side of the typology by including more granular measures of job quality, and 3) exploring these typologies in early-mid adulthood, a period during which careers and families have likely taken root. Using latent class analysis, I identify four job quality-family configurations among early-middle age men and three job quality-family configurations among early-middle age women. Among men, my two most advantaged classes both have high probabilities of being married, never divorced with children. Although their job quality characteristics vary in critical ways—the High Pressure, Flexible: Married Dad is characterized by greater control over work, but also longer hours and less stability, whereas the High Prestige, Security: Married Dad is characterized by higher wages and greater job stability—I find no significant differences between these classes when it comes to either form of stress. In contrast, the other two male classes—both less likely to be married or have children—are significantly more stressed. These findings add yet another dimension to the debate around the “fatherhood” and “marriage” premiums for men.

Among women, my findings are more varied. Although my two most advantaged classes—the Flexible, Unpredictable: Married Mother and the High Security, Stability: Partnered Mother—were statistically insignificant from one another with regards to self-reported stress, the latter class was significantly associated with lower odds of high inflammation. These differences may speak to important variation in how stress is perceived vs. how it is embodied. Conversely, reflecting on the Stress Process model (Pearlin, 1989), women in the Flexible, Unpredictable: Married Mother class may have access to additional coping mechanisms which attenuate their risk of high inflammation. For example, it could be that the more flexible and unpredictable nature of their job allows them to engage in more anti-inflammatory activities, such as regular exercise (Ihalainen, Hackney, & Taiple, 2019; Plaisance & Grandjean, 2006). Future work might examine how job quality-family classes correlate with reported incidence of exercise, diet, etc. and how these relationships might either shape (or be shaped by) stress.

An additional finding is variation in job quality-family formation and stress outcomes as a function of gender. While stratified LCA makes it difficult to conduct formal tests of gender differences, there are some interesting descriptive variations. For example, in two of my four male classes, the probability of being a parent is quite low (.21 and .39). Conversely, probability of parenthood is high for all three classes of women: even in the class with the lowest probability of parenthood, the probability is still .72. Similarly, for both men and women I find a job quality-family class typified by a high probability of being self-employed. While there are some strong similarities between these two classes—e.g., they both have a high probability of decision-making freedom at work and skill utilization—the male version of this class has a high probability of making a high salary and working long hours. Meanwhile, the female version of this class has a high probability of working part-time and making a lower income. Finally, although the supplementary latent class analysis and subsequent distal regressions of combined men and women does not offer as much nuance with regards to within-gender class differences, the significant interaction effects for physiological stress outcomes suggest types of job quality-family formation do differentially impact stress for men and women (see supplementary material).

This study has a number of important limitations that should be addressed here and may be considered for future analyses. First, this study is cross-sectional in nature. I made this choice in order to capture how job-quality and family formation are occurring on a detailed level for workers who are currently on the cusp of middle-age, and how these processes are related to said workers’ experiences of stress. This period has been largely overlooked in prior work-family latent classes. However, whether and how these classes change over time—and whether their association with stress changes over time—is beyond the scope of this project. Future work should include exploring trajectories of family formation and job quality across young adulthood and into early middle adulthood Second, while this study adds to prior work by taking a typological approach, the Add Health data set is limited with regards to data on work. Additional work-related characteristics were cross-walked in from the O*NET; however, this raises the issue of some measures being subjective while others are not. For example, a key limitation of this study lies with my measure of schedule stability. As it is measured, it captures the schedule predictability inherent to a given occupation, rather that the experiences of the individual. While I would argue there is still some value to this measure, as it captures the instability endemic to some jobs (such as construction workers or farm laborers), it does not capture the day-to-day instability experienced by many service workers. Additionally, whether a respondent is covered by a labor union is not captured in this study. While this is a limitation, union coverage was not particularly high at the time of the survey (2016–2018): approximately 12% of all employed persons were covered by a union in 2016, 11.9% in 2017, and 11.7% in 2018 (U.S. Bureau of Labor Statistics, 2023). Third, the sample size is fairly modest, and studies with larger samples have identified a greater number of classes. As such, it is likely that some job quality-family-gender configurations are not represented in this study. Fourth, gender and family definitions continue to broaden in scope. Unfortunately, due to few available observations, this study is limited to two genders, and is unable to account for same-sex pairings or the experiences of transgendered or non-gender conforming persons. Finally, these classes do not take into account the challenges faced at the intersection of other identities, such as race and gender. While these are critically important intersections that inform where people are situated in our social hierarchy, I simply did not have enough power to identify job quality-family classes stratified by race, gender and family composition. However, a second paper examining these processes at the intersection of gender and race is forthcoming.

Another limitation of this study is the focus solely on those who are employed. However, while unemployment and nonemployment are important aspects of the labor force experience overall, I believe exploration of these facets is beyond the scope of this study. The purpose of this study is to examine stress outcomes across detailed work-family types, specifically within the context of employment: as such, it would be inappropriate to compare these classes to persons experiencing unemployment and nonemployment. Future studies could explore latent classes of family and unemployment experiences, or examine trajectories of family and work over time.

As the job quality-family-gender and the work-stress landscape continues to evolve in the U.S., so too must our conceptualizations of how these concepts work together. While clearly identifying a typology of employment quality or a “good” job is critical to improving labor force experiences, it should also be acknowledged that the characteristics that make a job “good” for men may be very different from those that make it good for women, as well as for those with and without children. As we continue to strive for equitable job opportunities and a healthy workforce, taking greater stock of these forces can help us re-imagine work as something that is accessible, enjoyable and health-enhancing for all.