Unemployment and the Transition From Separation to Divorce

Abstract

Informal marital separation often quickly leads to divorce, but can become long-lasting, especially among disadvantaged populations. In this study, we focus on the timing of divorce after separating and examine how unemployment before or during separation affects this pivotal moment in the divorce process. Using data from the National Longitudinal Survey of Youth, 1979 Cohort (N = 2,219), we track unemployment before and during separation and show that men’s unemployment during separation, rather than women’s, reduces the likelihood of divorce, independent of preseparation unemployment and other characteristics. For men, unemployment during a marital separation prolongs the divorce process, creating an extended period of uncertainty in marital relationships on the brink of dissolution. We discuss the gendered relationship observed between employment status during an informal separation and an estranged couple’s decision to complete the divorce process.

The Great Recession has renewed interest in unemployment as a predictor of divorce in the United States (Amato & Beattie, 2011; Cohen, 2014). During marriage, a spouse’s unemployment can trigger the beginning of the divorce process, due to both the psychological damage of job loss and the economic strain of losing a share of the household income (Amato & Beattie, 2011). This effect of unemployment on divorce initiation differs by gender (Eliason, 2012; Sayer, England, Allison, & Kangas, 2011) and is stronger among married men because it conflicts with the norm of the male breadwinner (Sayer et al., 2011). The role conflict experienced by married men who become unemployed exacerbates the marital strain that follows from the financial difficulties of unemployment and often ends in the breakup of the marriage. Yet, although unemployment during marriage is known to spur the beginning of the divorce process, it is unknown how unemployment after the initial separation affects couples’ decision to complete this process.

A separation does not always imply that a formal divorce is subsequently attained (Schaller, 2013; Tumin, Han, & Qian, 2015). After the initial separation, couples’ continuing or new unemployment may pose a barrier to divorce. Unemployment during separation may limit material resources available to complete the divorce and sustain separate households (Amato & Beattie, 2011). For example, estranged couples may continue to share health insurance, assets, or even a home, until the unemployed spouse’s financial situation improves (Roberts, 2009; Sohn, 2015). Furthermore, if there is a dispute about aspects of the divorce, such as division of assets and custody arrangements, the unemployed spouse may not be able to afford legal help to resolve issues and finalize the divorce. Unemployment can also discourage divorce—after the initial separation, the unemployed spouse may have little incentive to divorce if he or she perceives difficulties finding a new partner after divorce (Ono, 1995).

The theorized role of unemployment as a barrier to divorce after separation stands in contrast to the role of unemployment in the initial disruption of the marriage. Clearly, the effect of unemployment on the divorce process varies according to the timing of unemployment. In this study, we explore how unemployment during separation influences the transition from separation to divorce and whether this influence varies by gender. We build on prior research by distinguishing between informal separation and formal divorce as different stages of the divorce process and emphasizing that the timing of unemployment matters in the relationship between unemployment and divorce. Specifically, our study presents the first analysis of unemployment during separation as a potential barrier to the transition from separation to divorce and helps understand why disadvantaged populations are disproportionately likely to get stuck in the state of informal separation. We consider gender differences in the role of short- and long-term unemployment after separation. Our study contributes to the literature on unemployment and marital instability by demonstrating the importance of timing and length of unemployment for shaping the transition from separation to divorce.

Unemployment and Marital Disruption

Job loss creates economic hardship and psychological stress, which can jeopardize the stability of a marriage (Fox & Chancey, 1998; Ström, 2003). Previous work on employment and marriage suggested that marriage rewarded men for specializing in wage labor and women for specializing in housework and that the risk of divorce increased when married men were unsuccessful in fulfilling the specialized role of breadwinner (Becker, 1991). The entry of women into the labor force in recent decades means that a typical married American family includes two wage earners (Casper & Bianchi, 2002; Qian, 2013). Both men’s and women’s jobs could therefore contribute to marital stability (Sayer et al., 2011). Conversely, job loss by either the husband or wife may strain the marriage because it forces the couple to make do with fewer resources, upsets the “bargain” over what each spouse contributes to the relationship, and makes the unemployed spouse less attractive as a partner. Indeed, both husbands’ and wives’ unemployment contributes to higher odds of marital disruption (Hansen, 2005; Jalovaara, 2003).

Notwithstanding the increasing importance of wives’ labor force participation, the effect of unemployment on marital disruption remains gendered. In the United States, husbands’ unemployment motivates divorce initiation by both men and women (Sayer et al., 2011). Some studies using Western European data report similar results, with husbands’ unemployment being a stronger predictor of divorce risk than wives’ unemployment (Eliason, 2012; Jalovaara, 2013; Jensen & Smith, 1990). One study argues that when men are fired from their jobs (as opposed to losing a job due to the elimination of a worker’s position), this signals their low status and increases the chances of their marriage ending (Doiron & Mendolia, 2012). This suggests that men’s unemployment increases the risk of divorce initiation because unemployed men (but not unemployed women) are evaluated against gendered ideals of male breadwinning (Forret, Sullivan, & Mainiero, 2010; Sherman, 2009), and not simply because men’s unemployment is a blow to a couple’s finances.

An important caveat to most studies on unemployment and divorce is that they do not distinguish between divorce initiation (i.e., the beginning of an informal separation) and the end stage of the divorce process (the transition from separation to formal divorce). Therefore, these studies do not account for the ambivalence many separated couples express about completing the divorce process (Doherty, Willoughby, & Peterson, 2011; Wineberg, 1996; Wineberg & McCarthy, 1994). Because of this ambivalence, couples may move into and out of informal separation multiple times as they attempt to work through marital difficulties or may remain informally separated for many years (Binstock & Thornton, 2003; Bramlett & Mosher, 2002). The phenomenon of separation without divorce (e.g., 9% of separated women not divorcing within 3 years of separation; Bramlett & Mosher, 2002) means that the decision to complete the divorce process after the couple has initially separated is theoretically significant in its own right (Schaller, 2013). Indeed, predictors of divorce initiation, such as low educational attainment, may not lead to divorce attainment after separation and may even be barriers to formal divorce after separation (Ono, 1995; Tumin et al., 2015).

Although unemployment is strongly associated with the risk of marital disruption, continuous unemployment or new unemployment during separation may discourage divorce from happening. The reasons are multiple. First, the “cost of divorce” perspective suggests that unemployment limits a person’s resources to divorce (Amato & Beattie, 2011; Hellerstein & Morrill, 2011). For example, the loss of income due to unemployment may limit a person’s ability to enlist legal help for settling the terms of a divorce or may compel separated couples to continue sharing economic resources (Roberts, 2009). Among people who are unemployed during a marital separation, dependence on the estranged spouse for economic assistance may lead to deferring a formal divorce or eventually reconciling (Wineberg, 1996).

Second, the “modified remarriage” perspective suggests that the decision to formally divorce after separating is motivated by the prospects of finding a new partner on the marriage market (Ono, 1995). Unemployment signals low socioeconomic status (i.e., lack of or unstable economic resources), which lowers a person’s prospects of remarriage (Shafer & James, 2013). Furthermore, unemployment may socially isolate a person and prevent them from meeting potential new partners. Therefore, unemployed separated people may have less incentive to divorce because their economic instability makes them less attractive to potential mates. Consequently, unemployment during separation may further delay divorce. This pattern may be more relevant in the case of men’s unemployment, due to gendered norms of labor force participation that place higher value on men’s employment, as opposed to women’s employment.

Unemployment, especially among men, increases the likelihood of marital disruption but may slow the divorce process after separation. Of course, unemployment can happen at any time during an individual’s life course. Unemployment during marriage and during separation may lead to different divorce outcomes. Whereas prior studies have examined time-invariant effects of socioeconomic disadvantage on the outcome of marital separation (Bramlett & Mosher, 2002; Morgan, 1988), employment history changes many times over adults’ life course (U.S. Bureau of Labor Statistics, 2012b), and may change during the course of the separation. To date, no study has examined how job loss while a person is separated from their spouse affects the outcome of the separation. From the cost of divorce perspective, the economic challenges posed by unemployment during separation are more immediate than those posed by a past history of unemployment during the marriage and therefore should be a greater obstacle to divorce. From the modified remarriage perspective, current unemployment during the separation is a more relevant attribute for a separated person’s marriageability than their past unemployment history. Therefore, both perspectives suggest that unemployment during separation should discourage divorce independently of—and to a greater extent than—unemployment during the marriage.

Current Study

In this study, we examine the contribution of unemployment during separation to the odds of divorce after separating. Our first hypothesis is that unemployment during separation will decrease the odds of divorce independently of unemployment prior to separation. Our second hypothesis, mirroring the research on unemployment and the odds of the initial marital disruption, is that the effect of unemployment during separation will be stronger on men’s odds of divorce than on women’s, because the cultural salience of men’s breadwinner role amplifies the penalty to men’s unemployment in the marriage market and reinforces the “modified remarriage” mechanism discouraging formal divorce among unemployed men.

We aim to test these hypotheses while controlling for variables that may confound the relationship between unemployment and divorce. It is likely that some of the association between unemployment and divorce risk arises from prior socioeconomic disadvantages that contribute to both unstable employment and marital strain (Jensen & Smith, 1990). Specifically, we aim to isolate the contribution of unemployment from social and economic characteristics for which unemployment may be a proxy, both in early life (e.g., mother’s educational attainment) and during the first marriage (e.g., a history of poverty and whether the person had children while married). We also aim to control for confounders emerging during the separation itself, including health problems that may cause both unemployment and marital disruption (Singleton, 2012); the birth of new children during separation (Suchindran, Koo, & Griffith, 1985); and, among women, pregnancy during separation. Finally, we aim to isolate the effect of a person’s own unemployment from the effects of between-spouse differences in human capital, recognizing that the decision to reconcile, remain separated, or divorce is based not only on a separated person’s own economic characteristics but also on their economic attractiveness to their estranged spouse or their need for the estranged spouse’s financial support during separation.

Data and Method

Data

We use data from the National Longitudinal Survey of Youth, 1979 Cohort (NLSY79; U.S. Bureau of Labor Statistics, 2012a) to analyze separations from the first spouse. The NLSY79 is a nationally representative survey of Americans who were aged 14 to 21 years in 1979. Respondents were surveyed annually from 1979 to 1994, and biennially since then, reaching the 24th wave of the survey in 2010. The NLSY79 originally included 12,686 cases. We exclude 2,923 cases in the military and economically disadvantaged non-Hispanic White oversamples, as these oversamples were discontinued in 1985 and 1990, respectively. As we are interested in tracking the marital trajectories of NLSY79 respondents over time, we exclude 2,226 cases where the respondent either never reported a marriage or reported being never married as of 2010.

We focus on the disruption of a first marriage because the process of marital disruption may be different for remarried adults, who are more likely to respond to marital problems by seeking a divorce, as compared with people in their first marriage (Whitton, Stanley, Markman, & Johnson, 2013). Therefore, we exclude 96 respondents already divorced at the 1979 baseline interview. We also exclude a small number of respondents (169) whose first marriage was disrupted by widowhood. Of the remaining 7,272 ever-married respondents, 2,396 reported a separation from their first spouse. We wish to examine how long respondents remained separated before completing the transition to divorce. Therefore, we exclude 88 respondents with unknown separation or marriage dates; 23 respondents whose separation ended in the death of their ex-spouse; and 41 respondents whose separation ended in a divorce but on an unknown date. After excluding 25 respondents with no employment data (discussed below) collected during or prior to their separation, we arrive at a sample of 2,219 respondents of whom 1,537 have complete data on all variables in the analysis and 682 are missing data on time-invariant control variables. We use multiple imputation by chained equations (Royston, 2004) to fill in missing data on covariates, creating five imputed data sets. Each imputed data set was reshaped into long (person-month) format for further analysis.

Dependent Variable

We identify marital separations from the first spouse using data on baseline marital status (collected in 1979) and subsequent marital transitions (collected since the last interview and including marriage, separation, reunification, divorce, remarriage, and widowhood; Haurin, 1994). Separations not ending in divorce or reunification were considered ongoing and were censored at the time of the respondent’s most recent interview. Due to the small number of reconciled couples, reconciliation was treated as a censoring event. We create a person-month file for respondents who separated from their first spouse, including person-months from the beginning of a separation until the separation ends or is censored. Therefore, the outcome is a time-varying dichotomous measure of ongoing separation (coded 0) as compared with divorce (coded 1) in a given month. The longitudinal design of the NLSY79 means that some respondents exit the study after separating but before reporting the end of their separation. This may lead to classifying some divorced respondents as remaining separated. Yet only 3% of respondents separated from their first spouse and exited the study before 2010 without contributing data on a divorce or reconciliation. Therefore, despite attrition, the outcome of separations from the first spouse is known for the majority (97%) of such separations in the data.

Employment Status

Data on labor force participation were compiled in a weekly array classifying respondents as employed, unemployed, or out of the labor force in a given week. We use these data to calculate weeks spent unemployed during the first marriage. Similarly, we create a time-variant variable counting weeks spent unemployed during the separation. The duration of unemployment (before or after separation) may have a nonlinear association with the odds of divorce, so we examine quadratic transformations of both of these unemployment variables. We also explore how well the relationship between unemployment and divorce is represented with categorical measures of unemployment. We code dichotomous measures of long-term unemployment, defined as unemployment for 27 consecutive weeks (Ilg, 2010), during the marriage and separation. The contrast between these dichotomous variables allows comparing among four categories of employment history: (a) no long-term unemployment before or during separation, (b) no long-term unemployment before separation but long-term unemployment during separation, (c) long-term unemployment before separation but no long-term unemployment during separation, and (d) long-term unemployment before and during separation. As few respondents fall into the last category of unemployment before and during separation, we focus our interpretation on the first three categories of this variable. Twenty-seven respondents were excluded from the models using categorical measures because they did not contribute employment data for 27 consecutive weeks.

Covariates

We control for respondents’ early-life social and economic background, including race/ethnicity (non-Hispanic White, non-Hispanic Black, or Hispanic), mother’s years of schooling (less than 12 years, 12 years, or more than 12 years), and whether the respondent was living with both biological parents at age 14. We also control for socioeconomic status during marriage, including years of education completed as of the separation date (less than 12 years, 12 years, 13–15 years, or 16 or more years), poverty history during the first marriage, and welfare receipt during the first marriage, including Aid to Families with Dependent Children (after 1996, Temporary Assistance to Needy Families); the Food Stamp program (after 2008, Supplemental Nutrition Assistance Program); and Supplemental Security Income; but excluding unemployment benefits. Respondents who had never been in poverty during their marriage were compared with respondents who had been in poverty during some, but less than half of the years in which they were married, and to respondents who had been in poverty during half or more of the years in which they were married. Welfare receipt was coded the same way.

We also control for relationship and fertility factors, including age at first marriage, duration of first marriage, whether the respondent cohabited before the first marriage, and whether the respondent had any children at the time of separation. We add a time-variant binary measure for having a new child during separation (1 = new child born since separation, 0 = no new children since separation). The other parent of a new child born during separation was not identified in the data. Among female respondents, we control for a time-variant measure of pregnancy during separation. To account for the confounding effect of health status, we control for a binary variable indicating any new health problem that limits the kind or amount of work a person could do (1 = yes, 0 = no; Teachman, 2010). Finally, we control for spouse characteristics, and, especially, the relative differences between spouses in income and education during the first marriage. Spouse covariates include age, spouse earnings equaling or exceeding respondent earnings prior to separation (1 = yes, 0 = no), and spouse educational attainment equaling or exceeding respondent educational attainment prior to separation (1 = yes, 0 = no).

Plan of Analysis

We use discrete-time event history analysis to model how time-invariant and time-varying characteristics affect the odds of divorce in a given month after separation (Allison, 1982; Box-Steffensmeier & Jones, 2004). We add a respondent-level random effect to the discrete-time logistic regression model to account for between-person differences on unobserved variables. In our first model, we include a time-invariant measure of weeks unemployed during marriage and a time-varying measure of weeks unemployed during separation to examine how unemployment during marriage and separation influences the odds of divorce. We then fit a second model including quadratic transformations of the continuous unemployment measures to examine nonlinearity in the association between weeks spent unemployed and the odds of divorce, Next, we examine models in which unemployment is treated categorically. The third model includes dichotomous measures of long-term unemployment before and during separation. The fourth model includes a single categorical measure interacting long-term unemployment before and during separation as described above. We present all regression results separately by gender and test the significance of gender differences by fitting pooled models where each covariate is interacted with gender (full results from pooled models are available on request). Our regression analysis does not use survey weights, following the recommendation to fit unweighted regression models when the sampling weights are not a function of the dependent variable (Winship & Radbill, 1994).

Results

We describe the characteristics of our sample in Table 1. This table summarizes background characteristics, characteristics at the beginning of the first separation, and characteristics at the end of the first separation for 917 women and 620 men who separated from their first spouse. In this sample, 742 women (81%) and 503 men (81%) divorced after separating, whereas the remainder either remained separated or reconciled. Median durations of separations ending in divorce were 20 months among men and 21 months among women. These figures exceed estimates of separation duration calculated from retrospective marital histories such as those included in the Survey of Income and Program Participation (Kreider & Ellis, 2011) or the National Survey of Family Growth (Copen, Daniels, Vespa, & Mosher, 2012). The discrepancy may be explained by differences in methodology between longitudinal and retrospective surveys collecting data on marital separation, such that very short separations are underrepresented in the NLSY79, leading to higher estimates of median separation length (Tumin et al., 2015).

Table 1.

Characteristics of Men and Women Who Separate From Their First Spouse.

	Proportion or mean (standard error)
	Women (n = 917)		Men (n = 620)
	Divorced after separation (n = 742)	Remained separated or reconciled (n = 175)	Divorced after separation (n = 503)	Remained separated or reconciled (n = 117)
	Background characteristics
Race/ethnicity
Non-Hispanic White	0.53	0.36	0.58	0.36
Non-Hispanic Black	0.25	0.34	0.23	0.50
Hispanic	0.22	0.30	0.19	0.24
Mother’s years of education
Less than 12	0.45	0.60	0.40	0.52
12	0.42	0.34	0.45	0.32
More than 12	0.13	0.06	0.15	0.16
Lived with biological parents at age 14	0.68	0.57	0.70	0.56
Characteristics at end of marriage
Age at first marriage	21.48 (0.15)	22.69 (0.44)	23.13 (0.17)	25.34 (0.58)
Length of first marriage (years)	7.26 (0.21)	9.54 (0.57)	6.97 (0.24)	8.21 (0.60)
Ever cohabited	0.19	0.23	0.26	0.31
Ever had children	0.76	0.89	0.75	0.85
Poverty during first marriage
Never below poverty line	0.70	0.58	0.78	0.62
Below poverty line in less than half of years married	0.21	0.33	0.15	0.27
Below poverty line in half or most of years married	0.09	0.10	0.07	0.11
Welfare receipt during first marriage
Never received welfare	0.74	0.63	0.80	0.74
Received welfare in less than half of years married	0.17	0.19	0.12	0.16
Received welfare in half or most of years married	0.09	0.17	0.08	0.10
Years of education
Less than 12	0.07	0.11	0.13	0.16
12	0.44	0.50	0.50	0.57
13–15	0.31	0.24	0.22	0.17
16 or more	0.19	0.15	0.15	0.10
Unemployment during first marriage
Weeks unemployed	16.58 (0.98)	23.22 (3.22)	19.54 (1.76)	24.21 (3.93)
Long-term unemployment (27+ week spell)	0.12	0.16	0.13	0.18
Spouse characteristics at end of marriage
Spouse age	31.34 (0.28)	34.09 (0.76)	28.48 (0.31)	31.29 (0.77)
Spouse education ≥ respondent education	0.57	0.61	0.68	0.74
Spouse income ≥ respondent income	0.79	0.79	0.28	0.22
Characteristics at the end of separation
Length of separation (months)	35.47 (1.60)	58.04 (5.86)	32.28 (1.68)	51.20 (5.63)
Unemployment during separation
Weeks unemployed	7.44 (0.80)	11.46 (2.49)	9.13 (1.42)	18.85 (4.42)
Long-term unemployment (27+ week spell)	0.06	0.10	0.07	0.14
New children while separated	0.10	0.09	0.11	0.08
Health limitation while separated	0.08	0.11	0.05	0.09

Men and women who did not divorce after separating were more likely to be Black or Hispanic and more likely to have been raised by a mother who did not complete 12 years of education and did not live with the respondents’ biological father. During the first marriage, respondents who did not divorce after separating were more likely to have had children, were more likely to have spent part of the marriage living in poverty, and, among women, were more likely to have received welfare. Although the proportion of respondents experiencing one or more spells of long-term unemployment during marriage was somewhat higher among men and women who did not divorce after separating (18% and 16%, respectively, compared with 13% and 12% in the divorced group), these differences were not statistically significant. Yet, during separation, men who eventually divorced were significantly less likely to have been long-term unemployed (7% compared with 14%, p < .05).

In Table 2, we use the person-month data to model the odds of divorce among men and women in a given month as a function of continuous measures of unemployment before and during the separation. Among women, the odds of divorce in a given month decrease by 0.4% with each month spent in separation (p < .001), whereas among men, the odds of divorce are not related to time spent separated. In Model 1, continuous unemployment measures are modeled as linear terms, with weeks spent unemployed during the first marriage unrelated to divorce among both men and women, and weeks spent unemployed during the separation significantly associated with lower odds of divorce in a given month among men. For men, each 10 weeks spent unemployed during separation decreases the odds of divorce in a given month by 5.2% (p < .05), and this association is significantly different from the estimate in the women’s model at the 95% confidence level.

Table 2.

Odds Ratios From Random Effects Logistic Regression of Divorce After Separation on Continuous Measures of Unemployment in the Person-Month File.

	Model 1		Model 2
	Women	Men	Women	Men
Time in separation (months)a	0.996***	0.999	0.996***	0.999
Cumulative unemployment experience
Weeks unemployed during first marriageb	0.980	1.011	0.988	1.014
Weeks unemployed during first marriage, squaredc	—	—	0.999	1.000
Weeks unemployed during separationa, b	0.993	0.948***	1.035	0.910***
Weeks unemployed during separation, squareda, c	—	—	0.998	1.002*
Pregnanta	0.945	—	0.943	—
New children while separateda	0.737**	0.911	0.731*	0.941
Health limitation while separateda	1.131	0.731	1.114	0.731
Race/ethnicityd
Non-Hispanic Black	0.377***	0.426***	0.375***	0.429***
Hispanic	0.560***	0.545***	0.563***	0.551***
Mother’s years of educatione
Less than 12	0.841*	0.893	0.835*	0.889
More than 12	0.970	0.876	0.974	0.874
Lived with biological parents at age 14	1.183*	0.998	1.183*	0.993
Age at first marriage	0.979	0.950***	0.980	0.949***
Length of first marriage (years)	1.008	1.009	1.009	1.007
Ever cohabited	0.874	1.042	0.871	1.038
Ever had children	0.737***	0.882	0.731***	0.884
Poverty during first marriagef
Below poverty line in less than half of years married	0.854	0.705**	0.857	0.704**
Below poverty line in half or most of years married	0.850	0.681*	0.846	0.675*
Welfare receipt during first marriageg
Received welfare in less than half of years married	0.879	1.008	0.879	1.018
Received welfare in half or most of years married	0.965	0.818	0.975	0.829
Years of educatione
Less than 12	0.627***	0.970	0.627***	0.970
13–15	1.327**	1.525***	1.325**	1.519***
16 or more	1.364**	1.416*	1.370**	1.417*
Spouse age	0.988	0.993	0.988	0.995
Spouse education ≥ Respondent education	1.072	1.212	1.071	1.218
Spouse income ≥ Respondent income	0.998	1.167	0.994	1.160
ρ	0.000	0.073	0.000	0.065
Sample size	1,289	930	1,289	930
Person-month observations	61,999	38,572	61,999	38,572

Note. Multiple imputation used to fill in missing data on covariates. In each model, underlined pairs of coefficients indicate statistically significant gender difference (p < .05 of interaction with gender in model fitted to pooled sample).

^aThis variable is time-varying in the person-month file. See text for details.

^bScaled by 10 (1 unit = weeks unemployed/10).

^cScaled by 100 (1 unit = weeks unemployed squared/100).

^dThe reference category is non-Hispanic White.

^eThe reference category is 12 years of education.

^fThe reference category is never living in poverty during first marriage.

^gThe reference category is never receiving welfare during first marriage.

^*p < .05.

^**p < .01.

^***p < .001. (Two-tailed tests)

Model 2 elaborates on the relationship between unemployment and divorce risk by introducing quadratic terms for unemployment during marriage and unemployment during separation. As in Model 1, only unemployment during separation is significantly associated with the outcome of the separation, and only among men. The odds ratio (OR) for the square of weeks spent unemployed during separation is statistically significant in the men’s model (OR = 1.002, p < .05), indicating that men’s unemployment during separation is nonlinearly related to the odds of divorce in a given month. The shape of this nonlinear relationship is illustrated in Figure 1, which shows predicted probabilities of divorce in a given month by length of time spent unemployed during separation, as based on Model 2.

Figure 1.

Predicted probability of divorce in a given month among separated men as a quadratic function of weeks spent unemployed during separation.

Figure 1 shows the predicted probability of divorce in a given month for men who were non-Hispanic White; whose mother had a high school education; who lived with their biological parents at age 14; who had never cohabited, had children, lived in poverty, or received welfare; who completed high school; who earned more and were more highly educated than their spouses; and whose age at first marriage, length of first marriage, and spouse age were at the sample means. Given these baseline characteristics, men who were never unemployed during separation had about a 20% chance of divorcing in a given month while they were separated, whereas men who were unemployed for half a year (27 weeks) had about a 15% chance of divorcing in a given month. At longer durations of unemployment, divorce probability continued decreasing for each additional week spent unemployed but at a weaker rate. Therefore, men with very long unemployment spells during separation did not have much lower chances of divorce in a given month than men with moderately long unemployment spells during separation.

ORs of covariates in Table 2 suggest that, in addition to men’s unemployment, other indicators of socioeconomic disadvantage delay divorce among both men and women. Black and Hispanic men and women are significantly less likely to divorce in a given month, and men and women with more than 12 years of education are significantly more likely to divorce than men and women who completed 12 or fewer years of schooling. Furthermore, women whose mothers had less than 12 years of education and men who lived in poverty during some or most of their marriage were less likely to divorce in a given month. Additionally, women who had children before or during the separation were less likely to divorce. This result may represent the effects of financial interdependence between single mothers and their estranged husbands.

Net of these control variables, results for unemployment shown in Table 2 support both of our hypotheses. Consistent with the first hypothesis, men’s unemployment during separation decreases the probability of divorce after separating, whereas men’s unemployment prior to separation is not associated with this outcome. Consistent with the second hypothesis, the association found between unemployment and divorce is significantly stronger among men than among women, where the latter’s odds of divorce in a given month of separation are not affected by unemployment either before or during separation. In Table 3, we show that these findings are robust to a different specification of the unemployment variables. Model 3 uses binary measures of long-term unemployment, and estimates that men who are long-term unemployed during the separation have 44.1% lower odds of divorce in a given month (p < .001). Model 4 includes a combined categorical measure of long-term unemployment before and during the separation. This model clarifies that this reduction in the probability of divorce only applies when men who were never long-term unemployed during the marriage become long-term unemployed during the separation (OR = 0.547, p < .001). This result refines our conclusions about unemployment during separation by showing that men’s entry into long-term unemployment after separating, rather than a continuation of unemployment experienced during the marriage, is a factor delaying couples’ transition to formal divorce.

Table 3.

Odds Ratios From Random Effects Logistic Regression of Divorce After Separation on Categorical Measures of Unemployment in the Person-Month File.

	Model 3		Model 4
	Women	Men	Women	Men
Time in separation (months)a	0.996***	0.997	0.996***	0.998
Experience of long-term (27-week) unemployment
Unemployed during first marriage	0.775*	1.213	—	—
Unemployed during separationa	0.923	0.559***	—	—
Unemployed during but not prior to separationa,b	—	—	0.934	0.547**
Unemployed prior to but not during separationa,b	—	—	0.781*	1.199
Unemployed both prior to and during separationa,b	—	—	0.690	0.715
Pregnanta	0.945	—	0.945	—
New children while separateda	0.737**	0.895	0.737**	0.893
Health limitation while separateda	1.135	0.724	1.135	0.724
Race/ethnicityc
Non-Hispanic Black	0.379***	0.428***	0.380***	0.427***
Hispanic	0.557***	0.548***	0.557***	0.548***
Mother’s years of educationd
Less than 12	0.839*	0.904	0.839*	0.903
More than 12	0.971	0.890	0.971	0.891
Lived with biological parents at age 14	1.185*	0.987	1.185*	0.988
Age at first marriage	0.977*	0.946***	0.977*	0.946***
Length of first marriage (years)	1.008	1.006	1.008	1.006
Ever cohabited	0.889	1.071	0.890	1.071
Ever had children	0.744***	0.900	0.744***	0.899
Poverty during first marriagee
Below poverty line in less than half of years married	0.860	0.708**	0.860	0.707**
Below poverty line in half or most of years married	0.847	0.675*	0.847	0.677*
Welfare receipt during first marriagef
Received welfare in less than half of years married	0.876	1.005	0.876	1.006
Received welfare in half or most of years married	0.979	0.831	0.978	0.830
Years of educationd
Less than 12	0.624***	0.934	0.625***	0.934
13–15	1.316**	1.537***	1.316**	1.534***
16 or more	1.335**	1.415*	1.336**	1.412*
Spouse age	0.988	0.996	0.988	0.996
Spouse education ≥ Respondent education	1.071	1.226*	1.070	1.225*
Spouse income ≥ Respondent income	1.004	1.135	1.004	1.136
ρ	0.000	0.027	0.000	0.028
Sample size	1,271	921	1,271	921
Person-month observations	61,871	38,481	61,871	38,481

Note. Multiple imputation used to fill in missing data on covariates. In each model, underlined pairs of coefficients indicate statistically significant gender difference (p < .05 of interaction with gender in model fitted to pooled sample).

^aThis variable is time-varying in the person-month file. See text for details.

^bThe reference category is never experiencing unemployment before or during separation.

^cThe reference category is non-Hispanic White.

^dThe reference category is 12 years of education.

^eThe reference category is never living in poverty during first marriage.

^fThe reference category is never receiving welfare during first marriage.

^*p < .05.

^**p < .01.

^***p < .001. (Two-tailed tests)

The results from analyses of long-term unemployment (Table 3) reveal two patterns not observed in the analysis of continuous measures of unemployment (Table 2). First, in Models 3 and 4, women’s long-term unemployment during the first marriage was associated with lower odds of divorce in a given month of the separation. Although this result is at odds with null findings from Models 1 and 2, it raises the possibility that women’s past unemployment may contribute to delayed divorce after separation, albeit not in the immediate way in which men’s unemployment during separation appears to delay divorce. Second, among men, the odds of divorce were greater when their estranged spouse had equal or greater educational attainment during the marriage. This is consistent with the association between greater educational attainment and accelerated divorce among women and suggests that well-educated, high-earning women may be especially likely to pursue a formal divorce when their husband had “underachieved” economically during the marriage.

Discussion

Unemployment strains a marriage, and couples in which a husband or wife is unemployed tend to experience marital disruption (Jalovaara, 2003; Jensen & Smith, 1990). Yet not all couples progress from separation to divorce (Bramlett & Mosher, 2002; Tumin et al., 2015; Vennum, Lindstrom, Monk, & Adams, 2014). Unemployment during marriage may lead a couple to initiate a separation, but unemployment during separation may delay the completion of the divorce process. Drawing on the cost of divorce and the modified remarriage perspectives, we have described how unemployment during separation may delay formal divorce after separation, especially if the husband becomes unemployed. The cost of divorce perspective suggests that people who become unemployed during separation may be dependent on their estranged spouse and may lack the resources to resolve disputes about the terms of a divorce. The modified remarriage perspective suggests that people who become unemployed may be reluctant to divorce because they perceive themselves to have poor prospects in the marriage market. We have hypothesized that these mechanisms will be more salient for men’s rather than women’s unemployment during separation, reflecting a gendered view of men’s unemployment as more stigmatizing than women’s unemployment. Our initial analysis shows that men’s unemployment, rather than women’s, significantly reduces the odds of divorce in a given month of the separation. This result is independent of the effects of unemployment and other characteristics prior to separation and robust to different specifications of the unemployment measures.

At first glance, it is paradoxical for unemployment to hasten separation but delay divorce. A key contribution of our study is the distinction between unemployment during marriage and unemployment during separation. By making this distinction, we have shown that it is the change in men’s employment status (from employed during the marriage to unemployed during separation), rather than a past history of unemployment during the marriage, that draws out the divorce process. Therefore, unemployment spells which spur the beginning of the divorce process (i.e., unemployment spells during the first marriage) are not the same unemployment spells as those which delay divorce after separation (i.e., men’s unemployment spells during the separation itself).

The delay of divorce among separated couples where the man becomes unemployed represents a response to an event (job loss) that was most likely unforeseen when the couple had initially separated. The effect of unemployment during separation on the divorce process is similar to previous findings showing that relationship transitions are shaped by unforeseen changes in material circumstances. For example, cohabiting couples’ decision to marry is sensitive to recent experiences of unemployment (Lichter, Qian, & Mellott, 2006). The same may be true of separated couples’ decision to divorce, which may be more responsive to present material circumstances than to a past history of economic disadvantage. Indeed, many separated people resist divorce and attempt reconciliation (Dougherty et al., 2011; Wineberg, 1996). This ambivalence regarding divorce may mean that some separated couples may be willing to remain legally married while living apart (or even reconciling) as a workable response to a husband’s job loss.

We have observed that men’s, but not women’s, unemployment during separation delays divorce. The role of gender in our findings is similar to the gender difference in the effects of unemployment on divorce initiation (Sayer et al., 2011). While a couple is married, men’s unemployment appears to be more disruptive and destabilizing than women’s unemployment (Doiron & Mendolia, 2012; Sayer et al., 2011), owing to gendered perceptions of unemployment, whereby men’s unemployment and especially men’s financial dependence on their wives are viewed as violating masculine roles (Forret et al., 2010; Sherman, 2009). Future research should explore how these gendered norms of labor force participation might sway couples’ negotiation over the formal end of their marriage. Particularly, it is unclear if men who become unemployed during separation intentionally attempt to delay divorce because they perceive their present unemployment to be a handicap on the marriage market. Confirming this mechanism with future analysis would further substantiate the modified remarriage perspective on unemployment during separation as a barrier to divorce.

The cost of divorce and modified remarriage perspectives explain why men who become unemployed might push to delay divorce, independent of their wives’ characteristics. Would estranged wives of unemployed men acquiesce to delaying divorce after separation, or would they resist this delay? Our results suggest this may in fact depend on the relative differences in human capital between husbands and wives. As shown in Table 3, men whose wives had equal or greater educational attainment prior to separation were more likely to complete the divorce process. Although measures of spouse socioeconomic status were assessed during the marriage, not during the separation, we suspect that women who earn significantly more than their estranged husbands during a separation may react unfavorably to their husbands’ job loss, and may attempt to accelerate the divorce process. Examining whether wives’ reaction to men’s unemployment in separated couples is conditioned by the estranged spouses’ relative earning power is an important direction for future research on this topic.

Our results also suggest a new perspective on recent studies examining aggregate-level associations between unemployment and divorce. The negative relationship we find between unemployment and divorce among separated men echoes the negative association between aggregate unemployment rates and divorce rates (Schaller, 2013). In recent decades, the divorce rate has increased when unemployment was low, and decreased when unemployment was high, including during the Great Recession of 2008 to 2009 (Cherlin, Cumberworth, Morgan, & Wimer, 2013; Cohen, 2014; Schaller, 2013). One reason for this phenomenon may be the effect of unemployment on the process of divorcing. Particularly, separations may last longer and be less likely to end in divorce in periods of high unemployment. Our results are consistent with this explanation, as men’s unemployment during separation decreases the odds of a transition to divorce. This individual-level effect of unemployment on the outcome of separation may explain aggregate-level results on the relationship between unemployment and divorce.

The role of job loss in delaying divorce after marital separation emphasizes heterogeneity in the divorce process and the need to distinguish between decisions to separate and decisions to divorce. Whereas economic circumstances, such as unstable employment, may strain a marriage to the breaking point, prior research suggests that economically disadvantaged couples are less likely to transition from separation to divorce. Our results build on earlier investigations of unemployment and the divorce process by showing that men’s job loss during the separation can delay divorce independent of preseparation characteristics. The gendered nature of this effect raises questions about how estranged couples respond to violations of traditional gender roles. Although we have proposed individualistic speculations of why unemployed men would want to delay formal divorce, it remains to be seen how both wives’ and husbands’ intentions to divorce change over the course of a separation in response to husbands’ sudden job loss, or other unforeseen changes in material resources or marriage market prospects. Future research tackling these questions will add to our understanding of the transition from living apart to formal divorce in disrupted marriages.