The Association Between Retirement and Emotional Well-being: Does Prior Work–Family Conflict Matter?

Abstract

Objective.

This study investigates whether the association between retirement and emotional well-being depends on prior experience of work–family conflict.

Methods.

We use data from the 1993 and 2004 waves of the Wisconsin Longitudinal Study to estimate linear regression models of 2 dimensions of emotional well-being—depressive symptoms and positive psychological functioning. We also use fixed effects models to investigate whether key findings persist after controlling for stable, but unobserved, characteristics of individuals.

Results.

Retirement is associated with relatively fewer depressive symptoms among individuals who reported high levels of work stress interfering with family life in late midlife. We find suggestive evidence of a similar association with respect to positive psychological functioning after accounting for unobserved characteristics of individuals. Among individuals reporting high levels of family stress spillover into work life at late midlife, our results suggest that retirement tends to be associated with better emotional well-being among men than among women.

Discussion.

Retirement may come more as a relief than as a stressor for individuals previously experiencing high levels of work demands interfering with family life. However, particularly among women, retirement may not relieve the burdens of family life stressors.

CURRENT demographic trends in the United States such as the aging of the baby boom cohorts and growth in average life expectancy highlight the importance of understanding how retirement affects well-being in later life. A large body of previous research considers the association between retirement and emotional well-being, with most studies reporting no association, or only a weak association, between retirement and various indicators of mental health (e.g., Alpass, Neville, & Flett, 2000; George, Fillenbaum, & Palmore, 1984; Herzog, House, & Morgan, 1991; Warr, Butcher, Robertson, & Callinan, 2004). This work tends to focus on average effects of retirement despite substantial reason to believe that the consequences of retirement may vary depending on the contextual circumstances surrounding this major life transition. Although considerable research documents the importance of the interface between work and family life for the emotional well-being of individuals during early adulthood and midlife (e.g., Allen, Herst, Bruck, & Sutton, 2000; Kossek & Ozeki, 1998), relatively little is known about how prior exposure to work–family conflict may alter the experience of retirement. Using data from the Wisconsin Longitudinal Study (WLS), the current research investigates the association between retirement and two dimensions of emotional well-being: depressive symptoms and positive psychological functioning. In particular, we ask whether the association between retirement and emotional well-being is moderated by prior exposure to work–family conflict in late midlife. We further investigate whether the nature of these associations differs by sex and consider the robustness of conventional regression results to controls for unobserved characteristics of individuals that are fixed over time.

THEORY AND PREVIOUS RESEARCH

In his contextual model of role transitions, Wheaton (1990) argued that “potentially harmful mental health consequences of transition events will be moderated, if not entirely eliminated or reversed, by the presence of prior chronically stressful role problems” (p. 210). Rather than focusing on life events as universal stressors, Wheaton’s model suggests that an interaction may exist between a given life transition and prior levels of relevant role stress. With respect to retirement, this model implies that the emotional consequences of leaving the labor force may vary depending on prior stress experienced in the role of paid worker, even after adjusting for potential postretirement stressors, such as declines in health and caregiving responsibilities. There are a number of reasons to expect prior exposure to work–family conflict to influence the nature of the association between retirement and emotional well-being. Work–family conflict is associated with preferences for retirement (Raymo & Sweeney, 2006) and with well-being outcomes, such as anxiety, depression, irritability, feelings of low self-worth, fatigue, and alcohol use (Allen et al., 2000). Work–family conflict may be a particularly salient source of prior role stress for retirees, given well-documented interdependencies between work and family life and the reciprocal influence of close family members in shaping the outcomes of important life course transitions (Elder, 1994). Although a number of studies indicate that characteristics of family and employment environments are each associated with postretirement well-being (e.g., Alpass et al., 2000; Calasanti, 1996; Mein, Martikainen, Hemingway, Stansfeld, & Marmot, 2003; Reitzes, Mutran, & Fernandez, 1996; Szinovacz & Davey, 2004; Wheaton), little is known about whether and how prior stress resulting from the nature of the “interface” between work and family roles may alter the experience of retirement.

Work–Family Conflict

Work–family conflict arises when the demands of work and family roles interfere with one another and are perceived as being incompatible in some respect (Greenhaus & Beutell, 1985). Common correlates of work–family conflict originating from employment demands (work-to-family conflict) include nonstandard or inflexible work schedules, job stress, lack of autonomy on the job, and time pressure at work. Correlates of work–family conflict originating from demands of family roles (family-to-work conflict) include long hours spent on housework, childcare, or caregiving, and low marital quality (see Byron, 2005, for a review, and Michel, Mitchelson, Kotrba, LeBreton, & Baltes, 2009, for a meta-analysis). The extent to which a given set of work and family demands are perceived as being incompatible may vary across individuals and is likely related to factors, such as personality characteristics, time management skills, and coping styles (Blanch & Aluja, 2009; MacDermid, 2005; Mitchelson, 2009), which are often unmeasured in survey data. Key sources of work–family conflict may also tend to change over the life course. For example, moving from young adulthood to middle age may tend to bring a reduction in responsibilities associated with parenting young children but an increase in demands associated with caring for aging parents.

The extent to which work–family conflict moderates the association between retirement and emotional well-being may vary depending on the origin of stressful role demands. Because work-based sources of stress should be ameliorated after leaving the labor force, we expect retirement to be associated with relatively more positive emotional outcomes for individuals previously experiencing high levels of work-to-family conflict. Retirement for this group should be perceived as more of a relief than as a stressor. The implications of family-to-work conflict for postretirement well-being, however, are more ambiguous. Retirement should relieve the active struggle to balance work and family life, allowing individuals with a high level of family stress to spend more time at home and possibly improve these conditions. In this way, retirement may be associated with some improvement in emotional well-being. However, some individuals may have immersed themselves in paid work as a haven from a stressful home environment (Hochschild, 1997) and although individuals nearing retirement age are less likely than younger adults to struggle with the “time bind” associated with raising young children, the demands associated with housework, caregiving, or marital problems may persist after leaving the labor force. Thus, retirement may reduce emotional well-being among those previously exposed to high levels of family-to-work conflict by eliminating an outlet for coping with stress and increasing exposure to family-based stressors. This may particularly apply to women, who tend to feel a relatively greater sense of responsibility for home-based demands and are more likely to have friends (and thus social support) at work (Hochschild; Morrison, 2009).

Using data from a large longitudinal study of women and men, the current research investigates whether the association between retirement and emotional well-being is moderated by prior exposure to work–family conflict. We separately consider the potentially moderating influence of work stress interfering with family life (work-to-family conflict) and family stress interfering with work life (family-to-work conflict). Both for women and for men, we expect that retirement will be associated with relatively better emotional well-being outcomes for individuals previously experiencing high levels of work-to-family conflict. We do not, however, expect fewer depressive symptoms or higher positive functioning to be associated with retirement among those previously exposed to high levels of family-to-work conflict. Particularly among women in this group, we expect that potentially negative effects of retirement associated with reduced employment-based social support and increased exposure to family-based stressors may tend to offset benefits associated with relief from the day-to-day struggle to manage work and family role responsibilities. Finally, we use fixed effects models to investigate the robustness of conventional regression results to unmeasured factors, such as aspects of personality or coping style, which may be correlated with emotional well-being.

METHODS

Sample

The WLS is a long-term cohort study of 10,317 randomly selected Wisconsin high school graduates from the class of 1957. The graduates were interviewed in 1957, 1975, 1993, and most recently in 2004. Our analysis relies on data from the 1993 and 2004 waves of the study, both of which included a phone interview and a mail questionnaire. In 1993, the response rates were 80% for the phone interview and 70% for the mail questionnaire, and in 2004, these rates were 81% and 78%, respectively. The WLS data are particularly well suited for our purposes because respondents were between 64 and 65 years old in 2004, and the data include a rich array of measures of psychological well-being, retirement status, work–family conflict, and relevant control variables assessed at multiple points in time. The longitudinal design of the WLS permits an investigation of the association between retirement transitions and subsequent psychological well-being while controlling for preretirement levels of emotional well-being and assessments of work–family conflict. It is important to note, however, that all respondents are high school graduates and most are White and reside in or near the state of Wisconsin. Results thus may not be generalizable to the entire population of similarly aged Americans.

We impose several restrictions on our analytic sample. First, we limit the sample to those respondents who completed both the telephone and the mail components of the 1993 and 2004 waves of the WLS (5,528 cases), who had valid responses for both 1993 and 2004 retirement items (5,281 cases), and who had not yet retired as of the 1993 interview (4,642 cases). Because the meaning of work–family conflict may differ for those who did not have a spouse during the preretirement period, we further limit the sample to those who were married in both 1993 and 2004 (3,505 cases). It is worth noting that of the respondents who were married in 1993, only 7% became widowed and 3% divorced or separated by 2004 (authors’ tabulations). Finally, we limit the sample to cases with no missing data on any of our analytic variables. Additional analyses (not shown) indicate that our substantive conclusions are not sensitive to alternate approaches to handling missing data, such as applying multiple imputation techniques. Differing patterns of missing data for our two outcomes leaves a total of 2,666 individuals for the analysis of depressive symptoms and 2,855 individuals for the analysis of positive psychological functioning.

Measures

Outcomes.—

As emotional well-being is multifaceted and reflects more than the absence of distress, and retirement may have different implications for well-being depending on which outcome is considered, we rely on two distinct indicators of this concept. Our first measure is the Center for Epidemiologic Studies Depression (CES-D) scale, which is shown to be suitable for use with the general population and has high reliability and validity (Radloff, 1977). The scale consists of 20 items indicating how many days in the past week respondents experienced a particular depressive symptom. For comparability with prior work, the scale was constructed by grouping the number of times during the past week that a particular symptom was experienced into categories of “0,” “1–2,” “3–4,” and “5–7 days.” Responses were then summed over the 20 items, with a relatively higher score indicating more frequent symptoms of depression. Respondents tended to report more symptoms of depression in 1993 than in 2004, with sample averages of 9.1 and 7.3, respectively (see Table 1).

Table 1.

Descriptive Statistics and Variable Descriptions: 1993 and 2004 Wisconsin Longitudinal Study (WLS)

Variable	M	SD	Description
Psychological well-being
Depressive symptoms (2004)	7.3	6.8	Score on CES-D scale. Based on self-reported frequency of 20 symptoms of depression (e.g., felt sad, felt lonely, slept restlessly) experienced during the past week. Sample range = 0 (fewest depressive symptoms) to 49 (most depressive symptoms), α = .86.
Depressive symptoms (1993)	9.1	7.4	Score on CES-D scale. Based on self-reported frequency of 20 symptoms of depression (e.g., felt sad, felt lonely, slept restlessly) experienced during the past week. Sample range = 0 (fewest depressive symptoms) to 55 (most depressive symptoms), α = .86.
Positive psychological functioning (2004)	95.1	12.3	Score on abridged version of Ryff’s (1989) positive well-being scale. Based on 20 questions assessing self-acceptance, relations with others, autonomy, environmental mastery, purpose in life, and personal growth. Range = 46 (lowest positive well-being) to 120 (highest positive well-being), α = .81.
Positive psychological functioning (1993)	99.5	12.3	Score on abridged version of Ryff’s positive well-being scale. Based on 20 questions assessing self-acceptance, relations with others, autonomy, environmental mastery, purpose in life, and personal growth. Range = 34 (lowest positive well-being) to 120 (highest positive well-being), α = .89.
Retirement status (2004)
Not at all retired	0.25		Based on responses to the following question: “At this time do you consider yourself partly retired, completely retired, or not retired at all?”
Partially retired	0.24
Fully retired	0.51
Work–family conflict (1993)
Work interfering with family	0.01	0.98	Scale constructed by summing responses to 3 items asking respondents the extent to which they agree or disagree with the following: “My job reduces the amount of time I can spend with my family”; “Problems at work make me irritable at home”; and “My job takes so much energy I don’t feel up to doing things that need attention at home.” Variable standardized with a M equal to 0 and a SD equal to 1 based on full WLS sample, α = .58.
Family interfering with work	−0.02	0.97	Scale constructed by summing responses to 3 items asking respondents the extent to which they agree or disagree with the following: “Family matters reduce the time I can devote to my job”; “Family worries or problems distract me from my work”; and “Family activities stop me from getting the amount of sleep I need to do my job well.” Variable standardized with a M equal to 0 and a SD equal to 1 based on full WLS sample, α = .62.
Other family characteristics
Shares “very similar” life outlook with spouse (1993)	0.56		Based on responses to the following question: “In terms of your outlook on life, would you say you and your (husband/wife) share very similar views, somewhat similar views, not very similar views, or not at all similar views?” Coded 1 if the respondent reports “very similar” views; 0 otherwise.
“Very close” to spouse (1993)	0.82		Based on responses to the following question: “How close would you say you are to your (husband/wife)? Would you say you are very close, somewhat close, not very close, or not at all close?” Coded 1 if the respondent reports being “very close”; 0 otherwise.
Ever divorced or widowed (2004)	0.16		Coded 1 if the respondent was ever divorced or widowed; 0 otherwise.
Duration of current marriage (2004)	27.4	8.2	Total number of years respondent married to his or her current spouse.
Spouse’s labor force status (1993–2004)			Following questions in 1993 and 2004: “What is the labor force activity of your current spouse?” Valid responses were “currently working” and “not currently working.”
Employed in 1993 and 2004	0.39
NOT employed in 1993 or 2004	0.16
Employed in 1993 but NOT in 2004	0.45
Spouse in poor health (1993)	0.08		Coded 1 if respondent reported his or her spouse’s health was very poor, poor, or fair; 0 if good or excellent.
Spouse's health declined from 1993 to 2004	0.33		Coded 1 if respondent reported his or her spouse’s health was poorer in 2004 than in 1993; 0 otherwise.
Respondent provided care in past 12 months (1993)	0.11		Coded 1 if respondent provided care to a friend or family member; 0 otherwise.
Respondent provided care in past 12 months (2004)	0.05		Coded 1 if respondent provided care to a friend or family member; 0 otherwise.
Child(ren) were living with the respondent (1993)	0.40		Coded 1 if respondent’s child(ren) were living with the respondent in 1993; 0 otherwise.
Child(ren) were living with the respondent (2004)	0.09		Coded 1 if respondent’s child(ren) were living with the respondent in 2004; 0 otherwise.
Characteristics of current or last job (1993)
Worked 50+ hours per week	0.28		Coded 1 if respondent worked 50 or more hours per week at his or her primary job in 1993; 0 otherwise.
Class of worker
Government worker	0.24		Respondent’s class of worker at his or her primary job in 1993.
Self-employed	0.04
Other	0.72
Job required intense concentration	0.42		Coded 1 if respondent’s primary job always required intense concentration or attention in 1993; 0 otherwise.
Exposed to dangerous conditions at job	0.32		Coded 1 if respondent’s primary job required exposure to dangerous conditions in 1993; 0 otherwise.
Job required work under time pressure	0.30		Coded 1 if respondent’s primary job always required work under time pressure in 1993; 0 otherwise.
Very satisfied or fairly satisfied with job	0.94		Coded 1 if respondent felt “very satisfied” or “fairly satisfied” with his or her primary job in 1993; 0 if the respondent felt “somewhat dissatisfied” or “very dissatisfied.”
Log hourly wage	2.5	0.86	Log hourly wages in 1993 U.S. dollars. Those reporting $0 were assigned a value of $1.
Employer offered a pension plan	0.71		Coded 1 if the respondent’s 1993 employer offered a pension plan; 0 otherwise.
Employer offered health insurance	0.78		Coded 1 if the respondent’s 1993 employer offered a health insurance plan; 0 otherwise.
Background characteristics
Respondent is female	0.45		Coded 1 if the respondent is female; 0 if the respondent is male.
Log total net worth (1993)	11.9	2.3	Respondent’s total assets in log 1993 U.S. dollar. Those reporting less than $1 were recoded to $1.
Poor self-rated health (1993)	0.08		Coded 1 if the respondent’s self-rated health was “very poor,” “poor,” or “fair”; 0 if “good” or “excellent.”
Health declined from 1993 to 2004	0.23		Coded 1 if the respondent reported his/her health was poorer in 2004 than in 1993; 0 otherwise.
Education (2004)
High school graduate	0.45		Education categories based on respondent’s most recent degree.
Some college	0.19
College degree or more	0.36

Notes: Descriptive statistics presented for the 2,518 cases with no missing data on the complete set of independent variables and two outcome measures.

Our second measure is a modified version of Ryff’s (1989) scale of psychological well-being, an instrument widely used to assess positive psychological functioning. The questions composing the scale ask individuals to indicate the extent to which they agree or disagree with statements pertaining to levels of self-acceptance, positive relations with others, autonomy, environmental mastery, purpose in life, and personal growth. Because disagreement exists as to whether the full scale is composed of six substantially independent factor (Ryff & Singer, 2006; Springer & Hauser, 2006), we combine all items into a global index of positive psychological functioning as suggested by Springer, Hauser, and Freese (2006). We construct the scale by summing together total scores for the 20 items that appeared on the WLS mail instrument in both 1993 and 2004, with a higher score indicating relatively better positive psychological functioning. The sample means of positive psychological functioning were 99.5 and 95.1 in 1993 and 2004, respectively (see Table 1).

Work–family conflict and retirement.—

We construct two measures of work–family conflict for the current analysis that are shown to have high validity and reliability in prior research (Grzywacz & Marks, 2000). The first measure focuses on the extent to which work demands interfere with family life (work-to-family conflict) and the other focuses on the extent to which family demands interfere with work life (family-to-work conflict). These indices are assessed in 1993 and are each based on three items that are summed together and standardized to a M of 0 and a SD of 1 (see Table 1). Individuals were considered missing for each index if they were missing on any of the individual items. The correlation between these two measures of work–family conflict in our analytic sample is .36.

Retirement is difficult to measure. An individual may reduce his or her hours of work, leave a main job, or reduce the level of effort put into a job before leaving the labor force altogether (Gustman & Steinmeier, 2000). Previous research suggests that although employment is the main basis for subjective self-definitions of retirement status, other life circumstances influence this definition as well, especially for women (Szinovacz & DeViney, 1999). For example, two individuals may report different retirement states even when working the same number of hours. In addition, partial retirement is conceptually distinct from and associated with different correlates than full retirement (Mutchler, Burr, Pienta, & Massagli, 1997; Quinn & Burkhauser, 1994). In light of these issues, and because we argue that an individual’s “perceived” retirement status is most meaningful for understanding emotional well-being, we use a subjective measure of retirement status in 2004 based on self-reported full and partial retirement status (for exact question wording, see Table 1). WLS respondents report wide variation in retirement status, with 25% of respondents reporting being not at all retired, 24% being partially retired, and 51% being fully retired. Preliminary analyses revealed no significant differences across retirement groups in mean levels of emotional well-being or work–family conflict in 1993 (results available upon request).

Control variables.—

Because we are interested in how subjective perceptions of work–family conflict moderate the relationship between retirement and emotional well-being, we hold constant a series of factors that may be related to the experience of retirement or emotional well-being, such as other potentially stressful or influential aspects of the home environment. Potentially stressful family circumstances might be related to retirement, postretirement emotional well-being, or work–family conflict. For example, men appear to have lower emotional well-being after retirement when their wives are still employed (Moen, Kim, & Hofmeister, 2001) and poor marital quality is positively related to work–family conflict (Frone, Russell, & Cooper, 1992). Controls for marital context include two binary measures of marital closeness and shared outlook on life between spouses. We also construct a series of other family context variables, including whether the respondent was ever divorced or widowed prior to 1993, duration of the current marriage, spousal employment status, spousal health status, caregiving responsibilities, and whether children were living in the household. Measures of family context are drawn mainly from the 1993 interview (see Table 1). We further control for a variety of 1993 employment characteristics, including whether the respondent generally worked long hours; the respondent’s class of worker (government, self-employed, or other); whether the respondent’s primary job required intense concentration, exposed the respondent to dangerous conditions, or required work under time pressure; respondent’s general job satisfaction; and availability of pension plans or health insurance through one’s employer. Finally, we construct a series of control variables known to be associated with retirement or postretirement adjustment, including gender, wages, assets, physical health, and educational attainment (e.g., Dwyer & Mitchell, 1999; Gower, 1997; Hayward, Grady, Hardy, & Sommers, 1989; Herzog, Franks, Markus, & Holmberg, 1998; Kim & Moen, 2002). Detailed descriptions and descriptive statistics for our full array of variables are presented in Table 1.

Methods

Ordinary least squares models.—

In the first stage of the analysis, we investigate the association between retirement and emotional well-being by estimating a set of linear regression models. For each of our two outcome measures, we regress well-being in 2004 on a parallel measure of well-being in 1993, retirement status, work–family conflict, and the control variables just described. Because our primary interest lies in the potentially moderating effect of work–family conflict on the association between retirement and well-being, we also include an interaction between retirement status and work–family conflict. As our work–family conflict measures are standardized to have a M of 0, the main coefficients for partial and full retirement reflect estimated effects of retirement for individuals who experienced average levels of work–family conflict. Finally, to investigate whether moderating effects of work–family conflict differ by sex, we estimate a set of supplementary models that include the full set of two-way interactions of sex with retirement status and work–family conflict, as well as the three-way interaction among sex, retirement status, and work–family conflict.

Fixed effects models.—

Individuals with certain types of background or personality characteristics may be more likely than others to retire early or experience work-family conflict, and these same characteristics may also be related to emotional well-being. For example, individuals with good organizational or planning skills may be less likely to experience both work–family conflict and poor well-being (Adams & Jex, 1999). If such characteristics are indeed relevant and are unmeasured, the conventional regression estimates described earlier may be biased. To address this possibility, we estimate a set of fixed effects models in the second stage of the analysis. Specifically, we use ordinary least squares (OLS) to regress change in emotional well-being between 1993 and 2004 on change in retirement status and change in key time-varying control measures, including whether the respondent and his or her spouse experienced a decline in health between 1993 and 2004, whether the spouse’s labor force status changed during this period, whether the respondent experienced a change in caregiving responsibilities or child coresidence status, and change in the respondent’s reported wealth between the two survey waves (in 1993 dollars). Consider the separate regressions for emotional well-being in 2004 and 1993, such that:

(1)

(2)

where Y_it represents a particular emotional well-being measure in year t (1993 or 2004), W_it represents a vector of measured variables that change between 1993 and 2004, Z_i represents a vector of measured variables that are fixed over time, α_i represents a vector of unmeasured variables that are fixed over time, and X_i is a measure of whether partial or full retirement occurred between 1993 and 2004 (here entered into the regression as a series of dummy variables coded “1” if the relevant transition occurred). Subtracting the first equation from the second gives the following:

(3)

Note that both measured and unmeasured variables that are fixed over time (such as preretirement family and job characteristics and stable unobserved characteristics of individuals) effectively drop out of the model and thus are not expected to bias estimates of covariate effects in Equation (3). Fixed effects models do permit interactions between such fixed characteristics and other variables in the model, and thus, we again include interactions between work–family conflict and retirement to explore potential moderating effects of work–family conflict on the association between retirement and emotional well-being. To examine potential gender differences in the modifying effects of work–family conflict on the association between retirement and depressive symptoms, we estimate the previously described fixed effects models separately for men and for women. For a more detailed discussion of fixed effects methods, see Allison (1994, 2005).

RESULTS

Stage 1: Linear Regression Models

Depressive symptoms.—

We begin by considering the association between retirement and depressive symptoms, as displayed in the first column of Table 2. Not surprisingly, depressive symptoms in 1993 are positively associated with those in 2004, net of retirement status, work–family conflict, and our set of control variables. Among those reporting average levels of work–family conflict in 1993, being fully retired rather than not retired at all is associated with a 0.8-point lower CES-D score in 2004, net of other variables in the model. Although partial retirement is also associated with a net 0.4-point lower CES-D score, this coefficient is not significantly different from 0, net of other variables in the model.

Table 2.

Coefficients From Ordinary Least Squares Regression of 2004 Depressive Symptoms and Positive Psychological Functioning on Retirement Status, Prior Work–Family Conflict, and Control Variables: 1993 and 2004 Wisconsin Longitudinal Study

	Dependent variable
	CES-D (2004)	Positive functioning (2004)
Independent variables	Coefficient (SE)	Coefficient (SE)
Baseline well-being (1993)a	0.48*** (0.02)	0.63*** (0.02)
Retirement status (in 2004)
Not at all retired	—	—
Partially retired	−0.40 (0.30)	0.65 (0.49)
Fully retired	−0.77** (0.28)	0.42 (0.45)
Work–family conflict (in 1993)
Work interfering with family	0.76** (0.24)	−0.39 (0.39)
Family interfering with work	0.29 (0.24)	0.05 (0.37)
Partially retired ×
Work interfering with family	−1.19*** (0.33)	0.87 (0.53)
Family interfering with work	0.34 (0.33)	−0.78 (0.53)
Fully retired ×
Work interfering with family	−0.91** (0.29)	−0.14 (0.46)
Family interfering with work	0.33 (0.29)	−0.02 (0.46)
Other family characteristics
Very similar outlook on life with spouse	−0.53* (0.24)	0.73 (0.38)
Feels very close to spouse	−0.53 (0.31)	1.17* (0.48)
Duration of current marriage	−0.01 (0.02)	0.03 (0.04)
Ever divorced or widowed	0.35 (0.49)	0.42 (0.80)
Spouse’s labor force status (1993–2004)
Employed in 2004 and 1993	—	—
NOT employed in 2004 or 1993	0.56 (0.33)	−0.37 (0.53)
Employed in 1993 and NOT in 2004	0.12 (0.25)	0.37 (0.40)
Spouse in poor health (in 1993)	0.20 (0.41)	−0.46 (0.66)
Spouse's health declined from 1993 to 2004	0.22 (0.23)	−0.55 (0.37)
Provided care in past 12 months (1993)	−0.24 (0.34)	0.70 (0.55)
Provided care in past 12 months (2004)	0.69 (0.47)	−0.18 (0.76)
Child(ren) living in household (1993)	−0.55* (0.23)	0.37 (0.36)
Child(ren) living in household (2004)	0.37 (0.39)	−0.40 (0.63)
Characteristics of current or last job (in 1993)
Worked 50+ hours per week	−0.32 (0.27)	1.22** (0.43)
Class of worker (vs. other)
Government worker	0.18 (0.27)	0.54 (0.43)
Self-employed	−0.38 (0.55)	2.98*** (0.85)
Job required intense concentration	−0.16 (0.23)	0.51 (0.37)
Worker exposed to dangerous conditions	0.50* (0.24)	−0.03 (0.37)
Always under time pressure at job	−0.15 (0.25)	0.42 (0.39)
Very or fairly satisfied with job	−0.16 (0.44)	−0.23 (0.69)
Log hourly wage	−0.01 (0.15)	0.51* (0.24)
Employer offered pension plan	−0.46 (0.32)	−0.54 (0.52)
Employer offered health insurance	0.13 (0.35)	0.11 (0.55)
Background characteristics
Female	0.64* (0.26)	1.31** (0.42)
Log net worth	−0.10* (0.05)	0.07 (0.08)
Poor self-rated physical health (in 1993)	1.98*** (0.41)	−2.75*** (0.62)
Health declined from 1993 to 2004	1.23*** (0.25)	−2.07*** (0.41)
Educational attainment (vs. high school)
Some college	−0.26 (0.29)	0.90 (0.47)
College degree or more	−0.31 (0.27)	2.09*** (0.43)
Intercept	5.32*** (1.08)	26.41*** (2.13)
R2	.37	.48
N	2,666	2,855

Note: ^aParallel well-being measure (CES-D score or positive psychological functioning) assessed in 1993.

*p < .05; **p < .01; ***p < .001 (two-tailed tests).

The interaction between work–family conflict and retirement tests whether the association between retirement and depression is modified by prior exposure to work–family conflict in 1993. In short, we see evidence of such a modifying effect with respect to prior levels of work stress interfering with family life (work-to-family conflict) and reject the null hypothesis that the two coefficients associated with the interaction between work-to-family conflict and retirement are jointly equal to 0, F(2, 2628) = 7.26, p < .001. As shown in Figure 1, we see a relatively greater difference in depressive symptoms between retirees and nonretirees among those previously exposed to high levels of work-to-family conflict compared with those exposed to average levels of work–family conflict. For those experiencing levels of work-to-family conflict 1 SD above the mean, partial retirement is associated with a 1.6-point lower CES-D score relative to being not at all retired, and full retirement is associated with a 1.7-point lower CES-D score (vs. 0.4 points and 0.8 points fewer, respectively, for partial and full retirement among those with average levels of work-to-family conflict). No similar moderating effect is observed with respect to prior levels of family-to-work conflict, F(2, 2628) = 0.77, p = .46.

Figure 1.

Predicted depressive symptoms (Center for Epidemiologic Studies Depression [CES-D] score) by retirement status and work–family conflict.

Note: “High” conflict is defined as work-to-family (or family-to-work) conflict 1 SD above the mean.

We next ask whether sex differences exist in the moderating effect of work–family conflict on the association between retirement and depressive symptoms. As previously described, we add to the models presented in Table 2 the set of two-way interactions of sex with retirement status and work–family conflict, as well as the three-way interaction among sex, retirement status, and work–family conflict (detailed results not shown). An F test indicates that adding the full set of sex interactions significantly improves the overall fit of our model, F(8, 2620) = 2.34, p = .02. These results are displayed in Figure 2. For both men and women, we see relatively larger differences between retirees and nonretirees in predicted depressive symptoms among those who previously experienced high levels of work-to-family conflict (defined as 1 SD above the mean) than among those experiencing average levels of work–family conflict. Results differ somewhat with respect to family-to-work conflict, however. Among men who previously reported high family-to-work conflict, we see relatively lower levels of depressive symptoms among those who retired than among those who had not. The reverse relationship holds for women, however, such that we instead see higher levels of depressive symptoms among women who were partially or fully retired than among those who are not retired. Consistent with our initial hypothesis, supplementary F tests (not shown) indicate that coefficients for sex interactions involving family-to-work conflict are statistically meaningful, F(3, 2620) = 4.72, p < .01, whereas those involving work-to-family conflict are not, F(3, 2620) = 0.89, p = .44. In short, among individuals who experienced high levels of family stress spilling over into work at midlife, retirement is associated with lower levels of depressive symptoms among men but not among women.

Figure 2.

Predicted depressive symptoms (Center for Epidemiologic Studies Depression [CES-D] score) by sex, retirement status, and work–family conflict.

Notes: See Figure 1. Predicted values based on a single model that includes the full set of interactions among variables for sex, retirement status, and work–family conflict.

Positive psychological functioning.—

We next turn to positive psychological functioning, with results displayed in the second column of Table 2. We first note a positive association between psychological functioning in 1993 and 2004, but do not find a significant association between partial or full retirement and positive functioning among individuals with average levels of work–family conflict. We next consider whether the nature of the association between retirement and positive psychological functioning in 2004 depends on prior exposure to work–family conflict. Unlike findings for depressive symptoms, we find no significant interactions between retirement and work-to-family conflict, F(2, 2817) = 2.59, p = .08, or family-to-work conflict, F(2, 2817) = 1.55, p = .21, in the case of positive psychological functioning. Finally, F tests indicate that the additional set of two-way and three-way interactions among sex, retirement status, and work–family conflict do not significantly improve the fit of the model in the case of positive psychological functioning, F(8, 2809) = 1.43, p = .18. In sum, results from conventional linear regression models do not point to a significant moderating effect of work–family conflict on the association between retirement and positive psychological functioning either for women or for men.

Stage 2: Fixed Effects Models

Results described thus far provide some evidence that the association between retirement and emotional well-being varies depending on one’s prior level of work-to-family conflict—at least with respect to depressive symptoms. We next estimate a set of fixed effects models to determine whether these findings persist after adjusting for the potential presence of unobserved time-invariant factors.

Depressive symptoms.—

We first consider fixed effects model results for depressive symptoms, as shown in the left panel of Table 3. These results, which adjust for time-invariant factors, generally confirm key conclusions from the first stage of the analysis. For example, we find transitioning to full retirement to be associated with a significantly lower CES-D score among individuals previously experiencing average levels of work–family conflict, holding constant observed time-varying control measures and background characteristics that remain fixed over time. We also confirm our previous finding that work-to-family conflict significantly moderates the association between retirement and depressive symptoms, F(2, 2653) = 22.45, p < .001. For example, transitioning to full retirement by 2004 is associated with an almost 2-point reduction in CES-D score among individuals who previously experienced levels of work-to-family conflict 1 SD above the mean compared with a roughly 1-point reduction in CES-D score among otherwise similar individuals who experienced average levels of work-to-family conflict. We see only weak evidence that family-to-work conflict moderates the association between retirement and emotional well-being in the case of depressive symptoms, F(2, 2653) = 2.3, p = .10. Again, this is largely consistent with our OLS results from the first stage of the analysis.

Table 3.

Coefficients From Fixed Effects Regression Models of Depressive Symptomatology and Positive Psychological Functioning by Sex: 1993 and 2004 Wisconsin Longitudinal Study

	Depressive symptoms			Positive functioning
	Total	Men	Women	Total	Men	Women
Independent variables	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)
Retirement
Not at all retired	—	—	—	—	—	—
Partially retired	−0.61 (0.35)	−0.86* (0.43)	−0.39 (0.60)	0.28 (0.53)	0.68 (0.70)	−0.27 (0.83)
Fully retired	−0.97** (0.31)	−1.53*** (0.39)	−0.16 (0.51)	0.40 (0.47)	0.25 (0.64)	0.64 (0.71)
Partially retired ×
Work interfering with family	−1.17*** (0.27)	−0.80* (0.34)	−1.60*** (0.44)	1.42*** (0.41)	0.90 (0.56)	1.92** (0.62)
Family interfering with work	−0.28 (0.28)	−0.81* (0.35)	0.42 (0.46)	0.13 (0.42)	−0.33 (0.56)	0.82 (0.64)
Fully retired ×
Work interfering with family	−0.98*** (0.19)	−0.95*** (0.25)	−0.98*** (0.30)	0.42 (0.28)	−0.01 (0.40)	0.89* (0.40)
Family interfering with work	−0.37 (0.19)	−0.67** (0.26)	−0.04 (0.30)	0.98*** (0.29)	1.16*** (0.41)	0.79 (0.41)
Log net worth	−0.05 (0.04)	−0.12* (0.06)	0.01 (0.07)	0.17** (0.06)	0.24* (0.09)	0.11 (0.09)
Spouse became retired	−0.07 (0.26)	0.27 (0.33)	−0.59 (0.43)	0.48 (0.39)	0.11 (0.54)	0.80 (0.59)
Spousal health declined	0.44 (0.27)	0.40 (0.34)	0.52 (0.43)	−0.97* (0.40)	−1.16* (0.55)	−0.77 (0.59)
Respondent health declined	1.67*** (0.29)	1.34*** (0.38)	2.03*** (0.47)	−2.52*** (0.44)	−2.89*** (0.61)	−2.17*** (0.65)
Provided care	0.73* (0.34)	0.08 (0.50)	1.06* (0.47)	−0.27 (0.50)	−0.36 (0.79)	−0.20 (0.65)
Children living in household	0.63* (0.25)	0.66* (0.31)	0.65 (0.43)	−0.62 (0.38)	−0.48 (0.49)	−0.83 (0.59)
Intercept	−1.37*** (0.29)	−1.03** (0.36)	−1.73*** (0.48)	−4.39*** (0.44)	−4.13*** (0.60)	−4.72*** (0.67)
N	2.666	1,471	1,195	2,855	1,570	1,285
R2	.05	.06	.05	.03	.03	.04

*p < .05; **p < .01; ***p < .001 (two-tailed tests).

To further explore potential gender differences in whether work–family conflict moderates the association between retirement and depressive symptoms, we next estimate the previously described fixed effects models separately for men and for women. These results provide suggestive evidence of intriguing differences between men and women. Among men exposed to average levels of work–family conflict at midlife, we notice a significant reduction in depressive symptoms associated with both partial and full retirement. As observed for the full sample, this estimated effect is larger among men previously experiencing high work-to-family conflict at midlife than among men with relatively less work stress spilling over into family, F(2, 1458) = 10.1, p < .001. Unlike results from our pooled models, however, the findings from the sex-specific models also point to a reduction in depressive symptoms associated with retirement among men previously experiencing high levels of family-to-work conflict, F(2, 1458) = 6.1, p < .01.

Findings differ somewhat among women. As with men, we find retirement to be associated with relatively lower levels of depressive symptoms among women previously exposed to high levels of work-to-family conflict (i.e., 1 SD above the mean) than among those experiencing average levels of work-to-family conflict, F(2, 1182) = 12.0, p < .001. We do not, however, find a similar reduction in depressive symptoms associated with retirement among women previously exposed to high family-to-work conflict (vs. women previously exposed to average levels of family-to-work conflict), F(2, 1182) = 0.43, p = .65. Again, as suggested by our conventional regression models, retirement does not appear to bring the same reduction in depressive symptoms for women as for men among individuals who previously experienced high levels of family stress spilling over into work.

Positive psychological functioning.—

We next consider fixed effects regression results for positive psychological functioning, as shown in the right side of Table 3. Consistent with the previously described OLS models, our fixed effects results suggest no meaningful association between retirement and positive functioning among individuals previously experiencing average levels of work–family conflict. Unlike our previous set of results, however, moderating influences of work-to-family conflict on the association between retirement and positive well-being emerge after adjusting for unobserved factors that are fixed over time. For example, among individuals experiencing average levels of work-to-family conflict in late midlife, our results point to a (statistically insignificant) 0.28-point increase in positive well-being score associated with a transition to partial retirement. Yet among individuals with levels of work-to-family conflict 1 SD above the mean, our results suggest that transitioning to partial retirement is associated with a 1.7-point reduction in positive well-being score. Our results indicate that we should reject the null hypothesis that the two coefficients composing the interaction between work-to-family conflict and retirement are jointly equal to 0, F(2, 2842) = 7.07, p < .001. The fixed effects results also point to a similar, although somewhat smaller, moderating effect of family-to-work conflict on the association between full retirement and positive functioning after controlling for unobserved background characteristics that are fixed over time, F(2, 2842) = 5.74, p < .01.

Finally, we consider sex-specific findings for positive psychological functioning. These results again point to intriguing gender differences in the association between retirement and well-being. Turning first to our results for men, we do not find a significant moderating effect of work-to-family conflict on the association between retirement and well-being, F(2, 1557) = 1.3, p = .27. We do, however, find some evidence of a moderating effect of “family-to-work” conflict on this association, F(2, 1557) = 4.12, p < .05. Specifically, men previously exposed to high levels of family-to-work conflict tend to experience higher levels of positive well-being upon retirement than do men exposed to average levels of family-to-work conflict. Although we do find evidence of relatively better positive psychological functioning following retirement among women previously exposed to high levels of work-to-family conflict than among those exposed to average levels of such conflict, F(2, 1272) = 7.27, p < .001, we find only weak evidence of a moderating effect associated with prior exposure to family-to-work conflict, F(2, 1182) = 12.0, p = .07. Again, among individuals experiencing high levels of family stress spilling over into work at midlife, our results point more strongly to improved positive psychological functioning outcomes associated with retirement among men than among women.

DISCUSSION

Much prior work on the relationships between retirement and well-being focuses on average associations (e.g., Alpass et al., 2000; Bosse, Aldwin, Levenson, & Ekerdt, 1987; George et al., 1984; Herzog et al., 1991) despite the fact that the broader life circumstances within which this major life transition occurs can vary tremendously. Drawing on Wheaton’s (1990) contextual model of role transitions, this research asks whether emotional well-being after retirement tends to vary depending on one’s prior exposure to work–family conflict during late midlife. In short, our results suggest that work–family conflict may indeed moderate the relationship between retirement and emotional well-being, although our specific findings vary somewhat across domains of work–family conflict and across outcomes. For example, we find that respondents who previously experienced high levels of work-to-family conflict tend to do relatively well with respect to postretirement depressive symptoms. We expect that this is due to the relief of work-based sources of stress experienced upon retirement. Yet our results also suggest that retirement may not similarly reduce stress originating from within the family, at least for women. Although retirement should relieve the active struggle to balance work and family responsibilities, stressful family demands may persist after retirement and retirement may weaken access to work-based sources of emotional support—factors that may tend to be more strongly felt by women than by men (e.g., Hochschild, 1997). Other research similarly suggests that continued employment for women may offer some protection against stress at home. For example, retired women are more likely than employed women to experience depressive symptoms associated with caring for grandchildren, whereas the same relationship does not hold for men (Szinovacz & Davey, 2006).

Our study of emotional health after retirement is one of the few to date to use fixed effects methods to investigate whether conclusions drawn from conventional regression models are robust to the presence of unobserved personality traits or other characteristics that remain stable over time (see Lindeboom, Portrait, & van den Berg, 2002, for an exception). The fixed effects models largely confirm results from the linear regressions in the case of depressive symptoms. Furthermore, once we adjust for unobserved time-invariant variables, suggestive evidence also emerges that work–family conflict may moderate the relationship between retirement and positive psychological well-being. These results may reflect important unobserved personality or coping style differences related to positive psychological functioning. For example, research shows that individuals who are good at time management or are goal directed perceive lower levels of work–family conflict (Adams & Jex, 1999), characteristics that may also be associated with domains of Ryff’s (1989) positive well-being scale, such as environmental mastery or purpose in life.

Although our study is the first to highlight the importance of work–family conflict for understanding the relationship between retirement and emotional well-being, it is also limited in a number of respects that should be addressed in future work. For example, our measures of preretirement life context are largely drawn from a single point in time (1993), thus limiting our ability to fully untangle the complex casual processes underlying relationships among emotional well-being, work–family conflict, and decisions regarding retirement. Future work on this topic would benefit from assessments of work–family conflict and emotional well-being at multiple time points in the years preceding retirement. Indeed, the circumstances of work and family life may have changed in the years between 1993 and the time of retirement, and our snapshot measure of work–family conflict does not fully capture Wheaton’s (1990) concept of chronic prior role stress. We further expect that considerable variability exists in the work hours and other employment circumstances of our “partially retired” respondents. Future work should explore the nature of this variability and investigate the potential implications of this and other aspects of the retirement experience for emotional well-being, such as whether retirement was voluntary or involuntary and whether specific aspects of personality or coping style may jointly contribute to well-being, perceptions of work–family conflict, and/or retirement timing. Finally, it is important to keep in mind that the current findings may not extend to groups not well reflected in the WLS sample of mostly White high school graduates. Individuals with less education, for example, may experience a more difficult transition to retirement or a greater level of financial strain after leaving the labor force.

In conclusion, although much prior work documents a “balancing act” between work and family responsibilities during the early and middle life course (see Perry-Jenkins, Repetti, & Crouter, 2000, for a review), our study points to the continued importance of these issues later in life. We identify important variability in emotional health after retirement based on prior exposure to work–family conflict in a large cohort in the midst of the retirement process. Our findings suggest that workers nearing retirement who previously experienced work-based stress spilling over into family life may view retirement as a relief, whereas those, especially women, previously experiencing family-based stress spilling over into work may not benefit in the same way from retirement. In many ways, however, WLS respondents represent only the leading edge of contemporary shifts in the work–family nexus, with employment and family roles that are less gender segregated than their parents but more so than their adult children. It will be important to understand how the work–family interface may differently influence postretirement emotional health for future retirees, particularly as the large baby boom cohorts retire against a backdrop of unique efforts to combine work and family responsibilities.

FUNDING

This research was supported by the National Science Foundation (Grant No. 0550752); the California Center for Population Research, which receives core support from the National Institute of Child Health and Human Development (R24-HD41022); the UCLA Interdisciplinary Relationship Science Program, which receives core funding from the National Science Foundation; the UCLA Graduate Division; and UCLA Academic Senate Council on Research.

CONFLICT OF INTEREST

Any opinions expressed herein are those of the authors.