Lost jobs, broken marriages

Abstract

This paper examines the impact of both husbands’ and wives’ job displacement on the risk that the marriage ends in divorce. Using Swedish-linked employee–employer data, all married couples in which one of the spouses lost his or her job because of an establishment closure in 1987 or 1988 and a comparison sample were identified. Over a 12-year period, the excess risk of divorce among couples’ in which the husband was displaced was 13% and statistically significant. The estimated impact of wives’ job displacements was of almost the same size, but not statistically significant.

1 Introduction

The aim of this paper is to study the impact of job displacement on the risk of divorce. Although economists’ interests in the consequences of job displacement and unemployment have generated an extensive body of research, only a few studies have addressed the impact on marital stability or the risk of divorce. However, without also considering the non-economic consequences of job displacement, such as marital instability and ill-health, one cannot obtain a complete picture of the overall welfare loss caused by such events. These aspects of job displacement cannot be regarded as of secondary importance. In fact, divorce has been ranked as the most stressful life event and personal illness as the fifth most stressful life event after the death of a family member (Miller and Rahe 1997).

By linking administrative individual data to establishment data, all married couples in which at least one of the spouses lost a job because of an establishment closure in Sweden in 1987 or 1988 were identified, as well as a control group containing a large sample of couples who were not affected by these closures. To some extent, a closure can be viewed as a quasi-experiment since all employees are laid off regardless of their personal characteristics and behaviour. Using plant closures or other mass layoffs as a strategy to handle the selection issues that are otherwise associated with job loss is common in the displaced worker literature. Following the seminal study by Jacobson et al. (1993), a number of studies have examined the earnings impact of job displacements due to mass-layoffs (e.g., Eliason and Storrie 2006; Hijzen et al. 2010; Couch and Placzek 2010; Huttunen et al. 2011).¹ Most of these studies have found that job displacements are followed by long-lasting earnings losses, although the European and especially the Nordic studies generally have found smaller losses than the US studies over both the short term and the long term.

Recently, economists have also become increasingly interested in using mass-layoffs, especially business closings, to study the impact of job displacement on non-economic outcomes, such as mortality (Eliason and Storrie 2009a; Sullivan and von Wachter 2009; Browning and Heinesen 2011), morbidity (Browning et al. 2006; Eliason and Storrie 2009b, 2010; Browning and Heinesen 2011), disability pension (Rege et al. 2009a, b), fertility (Huttunen and Kellokumpu 2010; Del Bono et al. 2011), children’s school performance (Oreopoulos et al. 2008; Coelli 2011; Rege et al. 2011), and criminality (Rege et al. 2009a, b).

I will follow this methodological strategy to estimate the causal effect of job displacement on the risk of divorce while avoiding contamination by either reverse causality (i.e., marital problems caused the job loss) or selection bias (i.e., people who have lost their jobs have certain traits that render them less likely both to keep a job and to keep a marriage intact).^{2 , 3} Most previous studies in this area have focused on the impact of unemployment on the risk of divorce (e.g., Jensen and Smith 1990; Starkey 1996; Kraft 2001; Hansen 2005) and found evidence for at least an immediate increase in the risk of marital dissolution. However, the design of these studies does not clearly answer the question of whether job loss or unemployment causes marital instability or divorce but only of whether an association exists. A recent exception is Rege et al. (2007), who examined the impact of Norwegian husbands’ job displacement due to plant closure on marital dissolution. They found that the married men who lost their jobs because of plant closures during 1995–2000 were 11% more likely to be divorced by 2003 than the married men working in stable plants. Their results also suggest that the destabilising impact on marriages could not be explained by unexpected reductions in earnings.

Another exception is Charles and Stephens (2004) who examined the impact of job loss on divorce by cause of job loss. They found that job losses due to plant closures had no effects on the probability of divorce, whereas other types of job losses had a positive impact on the likelihood of divorce.⁴ They speculated that a job loss due to a plant closure does not affect the divorce decision because such an event does not reveal any information about non-economic traits that are relevant to the spouse, since all employees are laid off regardless of their individual characteristics in these situations.

Nevertheless, a job displacement may affect marital stability and increase the risk of divorce through several other channels. From an economic perspective, job loss may unexpectedly affect a partner’s earnings capacity and cause the other spouse to reconsider his or her initial choice of a marriage partner.⁵ However, most marriages and divorces in contemporary industrialised societies do not seem to be the result of utility-maximising behaviour in the conventional economic sense (Frey and Eichenberger 1996; Weiss 1997).

Another set of mechanisms that may contribute to an adverse impact of job loss on marital stability and the risk of divorce is related to the stress associated with job loss. In family stress theory, a stressor event, such as a job loss, will depending upon the family’s coping resources and the family’s perception of the event produce a crises or a resolution (Hill 1949).⁶ In the case of a “crisis”, the level of adaptation will also depend upon the family’s coping resources, perception of the crisis, coping strategies, and additional stressors (McCubbin and Patterson 1983). A job loss may operate as a stressor in various ways. It may produce financial strain, the extent of which would depend on, among other things, the re-employment possibilities, the eligibility for unemployment insurance, the other spouse’s income, and the duration of unemployment. Even without financial strain, a job loss may produce psychological distress because a job, aside from income, also provides such elements as social networks, time structure, and identity.⁷ Psychological problems are not necessarily limited to the job loser either but may also be transmitted to the spouse.⁸ Certain maladaptive individual coping strategies, such as alcohol abuse and violent behaviour, may also operate as additional stressors.⁹

The next section gives a brief overview of the macroeconomic environment during the study period and of the nature of marriage and divorce in the Swedish context. Section 3 presents the data, the empirical method, and some descriptive statistics. Section 4 presents the results by starting with the estimated effects of job displacement on earnings, non-employment, and unemployment for this particular sample. The section continues with the main analyses of the impact of job displacement on the risk of divorce, followed by a subgroup analysis and a supplementary analysis of how the impact of job displacement varies with time. Finally, Section 5 summarises and concludes the study.

2 Background: the Swedish context

2.1 The labour market

Table 1 summarises the Swedish labour market from 1985 to 1999. In the mid- to late 1980s, when the job displacements examined in this study occurred, Sweden experienced remarkably low unemployment rates. The unemployment rate had been falling since 1983 and was down to 1.5% in 1989, whereas the employment rate during the same period rose continually from 79.0% to 82.9%. During the years that followed, Sweden experienced the most severe recession since the 1930s. The unemployment rate rose to 8.2% in 1993 and stayed at approximately this level until 1997, whereas the employment rate fell by ten percentage points to 72.6%. Thus, the displaced married men and women faced a very buoyant labour market at the time of the job loss and had good opportunities to find new jobs before they had to face the impending recession.

Table 1 The Swedish employment, unemployment, and labour force participation rate during 1985–1999

Sweden has been internationally recognised for its high proportion of women in the labour force. During these years, the labour force participation was almost as high for women as it was for men and peaked at 82.3% in 1990 after rising for decades.¹⁰ The increase was then halted by the economic crisis and subsequently fell to 73.9% in 1998. Similarly, the female employment rate fell from 81.0% in 1990 to 69.4% in 1998.

2.2 Marriage and divorce

Sweden has been internationally recognised not only for the high participation of women in the labour force but also for its high divorce rate and low marriage rate. Figure 1 depicts the crude marriage and divorce rates from 1985 to 1999. Although the marriage rate decreased dramatically during the 1970s and early 1980s, the marriage rate decreased at a much slower pace during the period of this study (see Andersson and Guiping 2001). Except for the spike in 1989, which was due to changes in the eligibility of widow(er)s’ pensions, the rate varied between 5.2 marriages per 1,000 population in 1988 to 3.6 per 1,000 population ten years later.

Fig. 1

The crude marriage and divorce rates (i.e., the number per 1,000 population) between 1985 and 1999

The crude divorce rate has been relatively stable during the same time period and has varied only between 2.1 and 2.5 divorces per 1,000 population. However, it is important to note that the crude measure, which accounts for the divorce rate by population instead of by marriages at risk, conceals a somewhat increasing trend in the divorce rate.

The relatively high rate of marriages ending in divorce can probably be largely attributed to a liberal divorce legalisation. From 1973 to 1974, when the liberalising changes accepting unilateral and no-fault divorces came into effect, the divorce risk doubled. The court must still approve a divorce, but it cannot refuse a divorce application. If both spouses agree to divorce and jointly request it and if neither of them has custody of a child younger than 16 years or if the couple has lived apart for at least 2 years, then the court can grant the divorce immediately. Otherwise, a reconsideration period of at least six months must elapse before a new application can be submitted and before the divorce can be subsequently granted. Upon the dissolution of the marriage, all economic ties between the spouses are terminated. The court grants alimony only under exceptional circumstances, although disagreements about ancillary questions (e.g. maintenance issues or custody of children) can prolong the divorce process. On request, however, the court can grant the divorce immediately and make decisions regarding the ancillary questions later.¹¹

3 Data

3.1 The samples

The data used in this study is a linked employee–employer data set containing all married couples in which at least one of the spouses was displaced because of an establishment closure in 1987 or 1988 and a random sample of otherwise comparable couples. To create the data set, four registers (i.e., the Register Based Labour Market Statistics; the Longitudinal Database for Education, Income and Occupation; the Income and Wealth Register; and the Hospital Discharge Register) were merged to obtain information on the couples during three pre-displacement years and 12 post-displacement years.¹² One can link various registers to one another because each resident and each establishment in Sweden has a unique identity number (i.e., a civic registration number or an organisation number). Moreover, since the obligatory income statements, which are filed with the taxation authorities by the employer, contain both the employee’s civic registration number and the establishment’s organisation number, one can also link all employees to the establishments that employed them. This feature of the data enabled the identification of both the closing establishments and the employees who lost their jobs.

The samples of married couples were constructed in four steps. First, all closing establishments with at least 10 employees were identified by the disappearance of their identity numbers from the administrative registers. The problem of ‘false firm deaths’ (i.e., cases in which the disappearance of an identity number was due to, for example, a change of owner) emphasised in Kuhn (2002) was eliminated or at least greatly reduced by Statistics Sweden’s extensive examinations and corrections of the establishment numbers (Statistics Sweden 2005).

In the second step, the employees at these establishments were identified. In administrative data, one can observe separations between employees and employers, but usually no distinction can be made between quits and layoffs. Thus, it is necessary to define displacements as separations in connection to the closures. However, one can question whether selection bias truly does not exist in plant closure studies, since there is reason to assume that those employees with better outside options will be more likely to quit before the actual shutdown. On the other hand, the firm will be more likely to lay off its less valuable employees first during a preceding downsizing period. Thus, it is essential to identify not only the employees who are laid off at the time of the actual shutdown but also the employees who depart earlier because of the impending closure. The case study evidence in Pfann and Hamermesh (2008) indicates that this issue is important.

Therefore, if a closing process was deemed to last longer than one year based on the size of the establishment and the employee flows during the 3 years prior to the closure, then not only the employees who separated from this establishment in the same year as the shutdown but also the employees who separated in the preceding year were included, and in a few cases those who separated in the year before that as well (i.e., a 3-year-window). However, most of the closing processes were considered to last no more than a calendar year.¹³ As a sensitivity test, all estimations were performed also using a sample containing only those employees who separated within the calendar year preceding the closure (i.e., a 1-year window).

In the third step, two comparison groups were constructed that comprised random samples of married men and women who were employed in November of 1986 at non-closing establishments with at least 10 employees. However, these employees could have been displaced in any subsequent year. Correspondingly, the displaced employees could have experienced multiple displacements.

In the fourth step, all of the married employees were linked to their spouses. This was possible because married couples are taxed jointly in Sweden and because Statistics Sweden collects the administrative records from the National Tax Board.¹⁴ Therefore, the same information could be obtained for both spouses in each couple. This procedure resulted in two samples: 1) a sample of married couples in which all of the husbands were employed at baseline, but some of them were displaced; and 2) a sample of married couples in which all of the wives were employed at baseline, but some of them were displaced. Henceforth, I will refer to the two samples as the “male sample” and the “female sample”.

A few sample restrictions were applied before arriving at the final samples used in the empirical analyses. To fully appreciate the basis of these restrictions, it should be noted that many conditioning baseline variables contain information that was measured 2 to 3 years prior to the job displacement. The first restriction excluded the couples who had missing information in any of the three baseline years.¹⁵ The second restriction excluded those who were not married in all three baseline years. This latter restriction ensures that the baseline variables correspond to information about the two spouses as a married couple. It also implies that the establishment closures were unlikely to be expected at the time that the couples became married. Finally, the samples were restricted to the couples in which both spouses were 20 to 64 years old as of December 31 in the selection year.

These restrictions reduced the number of couples in which the husband was displaced from 4,227 to 3,692 and the number of couples in which the wife was displaced from 3,318 to 2,786. The corresponding comparison groups were reduced from 51,899 to 46,990 and from 49,548 to 43,393.

3.2 The divorces

Divorces could not be directly observed in that data but only the marital status of each spouse in each year. Therefore, a divorce was defined as having occurred if at least one of the spouses was registered as divorced. This definition would underestimate the number of actual divorces in the unlikely event that there were divorced couples in which both the former spouses remarried within the same calendar year of their divorce.

Table 2 shows the divorce rate for 12 post-displacement years in terms of the percentage and cumulative percentage for each of the populations defined in the previous section. From these raw figures, one can see that, on average, somewhat more than 1% of the marriages ended in divorce each year and that the risk of divorce decreased over time. The figures also indicate that those couples who were experiencing job displacements were more likely to divorce. An increased risk of divorce seems to be present regardless of whether the husband or the wife lost the job.

Table 2 The divorce rate in percentages and cumulative percentages

3.3 The baseline variables

The baseline variables were drawn from three registers: the Register Based Labour Market Statistics, the Income and Wealth Register, and the Hospital Discharge Register. The choice of variables included in the analyses was based on the economic theory of marriage and divorce and on family stress theory. They can be categorised as socio-demographic/economic variables and as variables corresponding to marital investments, match quality or homogamy, economic dependency/independency, coping resources, the establishment, and the local labour and marriage market. All of the variables will be discussed briefly below.

Demographic and socioeconomic variables   These variables include the spouses’ ages (9 categories each), immigrant statuses, and education levels (4 categories each) as well as the couple’s county of residence (21 categories).

Marital investments   This category includes the number of children within two age ranges¹⁶ and an indicator of whether the couple owned a house. Marriage-specific capital should theoretically reduce the risk of divorce because its value is higher within marriage. Prior empirical work has also found that children, especially younger ones, reduce the risk of divorce (Andersson 1997). However, the presence and number of children could have a reverse effect on a couple’s perception of job loss because more children mean more dependants to support. A positive correlation between marital investment and marital stability could also be spurious. That is, since marital investments decrease in value if the marriage dissolves, the possibility of divorce might lead less stable couples to engage in more cautious investment behaviour (Becker et al. 1977). Svarer and Verner (2008) found this case to be true in their analysis of Danish couples and also found that children actually destabilised marriages when they corrected for this selection bias.

Match quality or homogamy   Three dimensions of homagamy were measured: age, education, and ethnicity. Age homogamy is represented by a categorical variable with seven categories related to the difference in age between the spouses. Similarly, educational homogamy is represented by a categorical variable with four categories related to the spouses’ differences in terms of education levels. Ethnic homogamy was proxied with an indicator of whether the husband and wife were born in the same region of the world.

Prior studies have empirically established that positive assortative mating occurs among couples (Weiss and Willis 1997; Mare 1991), and Becker (1973) theoretically shows that positive assortative mating over a large range of personal traits is optimal.

Economic dependency/independency   One categorical variable with five categories representing the degree of one spouse’s economic dependency on the other spouse was included, and measured as the wife’s earnings relative to the couple’s total earnings. Several studies have shown that higher relative earnings from wives destabilise marriages (e.g., Jalovaara 2003; Kalmijn et al. 2007), although some studies suggest that this impact is confined to bad marriages (e.g., Sayer and Bianchi 2000).

Coping resources   How a couple will perceive a job loss or any other adverse life event and how they will adapt to it depend on their coping resources. The couple’s financial situation (measured as their disposable family income, taxable wealth, and earnings) is likely to influence their perception of the severity of the job loss. Several studies have also shown an inverse relationship between the measures of socioeconomic status and the divorce risk (e.g. Jalovaara 2001, 2003).

A number of variables representing previous hardship in various dimensions were also included. First, three variables measuring both spouses’ previous experience of unemployment and whether the couple had received means-tested social benefits. However, the expected effect of previous hardship is unclear. On the one hand, previous hardship could have an accumulated adverse impact, but on the other hand, it could also decrease a couple’s perception of the current displacement as stressful.

Financial resources alone do not constitute the full set of coping resources. The spouses’ mental and physical health could clearly affect their ability to cope with subsequent adverse life events. However, little attention has been devoted to how spouses’ health statuses affect marital stability and the risk of divorce, and the direction of causality is ambiguous. Some previous studies have shown that husbands’ ill-health increases the risk of divorce (Jensen and Smith 1990; Svarer 2002), whereas other studies have found no effect of disability on the divorce risk, regardless of which spouse was disabled (Charles and Stephens 2004). The spouses’ health statuses were measured both as incidence of hospital inpatient treatment with a discharge diagnosis of alcohol-related diseases/conditions, psychiatric conditions, or violence and as incidence of hospital inpatient treatment regardless of diagnosis.¹⁷

Regional characteristics   In addition to 21 county indicators, four measures of the characteristics of the local labour and marriage market were included. These measures were population size, unemployment rate, and the shares of both divorced and married persons. The latter two variables could be seen as measures of the availability of spousal alternatives. A low share of married men/women would increase the opportunity to find a new potential spouse and, thus, decrease the search costs, which could threaten the stability of marriage (see South and Lloyd 1995).¹⁸

Establishment characteristics   Although it can be argued that an establishment closure is a quasi-experiment, it is far from a true natural experiment because closures do not, for example, occur randomly over regions or sectors. Thus, in addition to the regional variables discussed previously, six variables measuring establishment characteristics were included: economic sector (10 categories); number of employees; the proportions of employees who had completed compulsory education, secondary education, and university education; and the fraction of female employees.

3.4 Empirical method

The exclusive focus on job displacements due to plant closures as a strategy to address the selection issues that are otherwise associated with job losses should not, which was also noted above, be interpreted as if having a natural experiment in hand. Although, any selection of which employees at a particular establishment are laid off is eliminated or at least greatly reduced, one cannot ignore that there is a non-random selection of which establishments are going out of business. Hence, it is still necessary to control for any baseline differences between the displaced and non-displaced couples.

A propensity score weighted estimator that is similar to those proposed in Hirano and Imbens (2001) and Robins et al. (2000) was adopted. By propensity score weighting one will ideally obtain a pseudo-sample in which the distribution of observed characteristics is the same for the samples of exposed (i.e., displaced) couples and non-exposed (i.e., non-displaced) couples. The propensity score (p) is the probability of exposure (Rosenbaum and Rubin 1983), which was estimated by a logit model: $p_i =\Pr \left[ {D_i =1\vert X_i } \right]=\left\{ {1+\exp \left( {-\alpha -\beta X_i } \right)} \right\}^{-1}$, where D _i is an indicator taking a value of one if the husband/wife in couple i was displaced at baseline and zero otherwise, and X _i is a vector of baseline covariates.

To estimate the effect on the displaced couples a weight defined as $w_i =D_i +\left( {1-D_i } \right)\cdot p_i /\left( {1-p_i } \right)$ was assigned to each couple i (see Hirano and Imbens 2001). Hence, all of the couples in which at least one of the spouses was displaced were assigned a weight equal to one, whereas each comparison couple j was assigned a weight equal to p _j /(1-p _j ). After the weights were normalised, as suggested in Hirano and Imbens (2001), they were used to estimate a weighted discrete-time logit model: $h_i \left( t \right)=\left[ {1-\exp \left\{ {-\lambda \left( t \right)-\gamma Z_i -\delta D_i } \right\}} \right]^{-1}$, where h(t) is the hazard rate or the conditional probability of divorce, λ(t) is the baseline hazard function, Z _i is a vector of baseline covariates, and δ is the estimated effect of job displacement at baseline. The latter was assumed to be constant over time. No time-varying covariates were included since they may mediate the effect of displacement on the divorce risk and, therefore, bias the estimated effect of job displacement.

The choices of covariates included in X _i and Z _i can produce four different estimators. Including covariates in neither the estimation of the propensity scores nor the following discrete-time logit model will result in an unadjusted estimator. Including covariates only in the estimation of the propensity scores will result in a propensity score weighted estimator (PSW). If covariates, instead, are included only in the discrete-time logit model, this will result in a standard unweighted discrete-time logit model (DTL). Finally, including covariates in both estimations produces a propensity score weighted discrete-time logit model (PSW + DTL). As a check of robustness to the choice of estimator the estimates from all four estimators will be presented.

3.5 Estimation of the propensity scores

The estimation of the propensity scores was performed separately for the male and female samples, the two time windows, and for each of the subgroups investigated in Section 4.3. The covariates included in the full model were all those discussed in Section 3.3.¹⁹ For the sake of brevity, no estimation results are reported other than summary statistics for the propensity scores and the corresponding weights for the male and female samples using the preferred 3-year window.

Based on an assessment of these summary statistics in Table 3, it is evident that the samples of displaced and non-displaced couples are fairly similar with regard to the estimated propensity scores, although the non-displaced couples have a more positively skewed propensity score distribution. Moreover, since the propensity score is not near one for any observation, there are no corresponding weights that are unduly large for any comparison couples.

Table 3 Summary statistics of the estimated propensity scores and the corresponding weights

3.6 Descriptive statistics and balancing assessment

Table 4 presents the pre- and post-weighting means of the baseline variables for the main populations. To further assess whether the weighted samples are comparable with respect to these variables, the standardised differences in means (SDM; i.e. the difference in covariate means between the displaced and the non-displaced couples in percentage of the pooled standard deviation of that covariate before the weighting) are presented for the two samples before and after the propensity score weighting.

Table 4 Sample characteristics for the displaced (D = 1) and non-displaced (D = 0) couples

The largest pre-weighting differences in baseline characteristics between the displaced and non-displaced couples correspond to differences in establishment characteristics and spouses’ attained educational level. Both the displaced men and women were, on average, employed at much smaller establishments with less educated workforces. As expected, the economic sector in which the establishments operated also differed greatly between the displaced and non-displaced employees (see Table 4). The displaced employees (both married men and married women) also had more previous experiences of unemployment and lower earnings.

These differences related mainly to the establishments’ characteristics repeat the previously claimed need to adjust for baseline differences. Even though the problems associated with a non-random selection of who is laid off within a particular establishment are eliminated or greatly reduced in the case of a closure, there is a non-random selection of which establishments that goes out of business, which may affect the distribution of such worker characteristics that are correlated with future divorce risk.

However, the weighting of the samples reduced the mean of the absolute values of the SDMs from 7.8 to 0.5 in the male sample and from 9.4 to 0.7 in the female sample.²⁰ Hence, the weighting process generated pseudo-samples of non-displaced couples who were, on average, similar to the samples of displaced couples.

4 Results

4.1 The impact of job displacement on earnings, non-employment, and unemployment

A natural point of departure would be to first make a convincing case that the job displacements produced a shock to earnings and changed the future employment situations of the couples. In these estimations the second step discrete-time logit model was replaced with a fixed effect regression: $y_{i,t} =\alpha_i +\gamma_t +\sum\nolimits_{k=-2}^{11} {\delta ^kD_{i,t}^k } +\varepsilon_{i,t}$, where y _i,t is the particular outcome of interest for spouse i in year t, and the $D_{i,t}^k$ ‘s are a set of indicators for the number of years relative to the displacement that allow the temporal impact of job displacement to be estimated (i.e., δ ^k is the estimated impact k years after the displacement).²¹ The parameter α _i is the individual-specific fixed effect, γ _t is a time-specific effect, and ε _i,t is the error term.

To display the estimated impact on earnings, non-employment, and unemployment, the coefficients δ ^k are plotted in Figs. 2, 3 and 4. First, Fig. 2 presents the estimated effects of the job displacements of married men and women on annual earnings in thousands of Swedish kronor (SEK). All amounts were deflated to the 1999 values by using the consumer price index. It is clear that the married men’s displacements generated earnings losses that in monetary terms were much larger than those of women. In the year of the displacement, the relative drop in earnings corresponded to SEK 22,000 (8.8%) for the married men and approximately SEK 11,000 (8.1%) for the married women.²² For the married men earnings dropped further to a gap of SEK 34,000 (14.1%) in the following year. Thereafter, the earnings gap slowly diminished in both the male and female sample, although for the men, the gap was still as large as SEK 9,000 (4.9%) at the end of the 12-year follow-up period. For the women, the gap disappeared 8 years after the displacement.

Fig. 2

The impact of job displacement on earnings in thousands of SEK with 95% confidence intervals for married men (left) and married women (right)

Fig. 3

The impact of job displacement on non-employment (i.e., zero annual earnings) with 95% confidence intervals for men (left) and married women (right)

Fig. 4

The impact of job displacement on annually received unemployment insurance in thousands of SEK with 95% confidence intervals for men (left) and married women (right)

Based on Fig. 3, it is apparent that these earnings losses can be explained to a large extent by the displaced employees leaving paid employment. The difference between the shares of displaced and non-displaced married men who had zero annual earnings increased to 7.9 percentage points in the year following the displacement, whereas the corresponding difference for the married women was 6.5 percentage points. For both the male and female sample, the difference between the displaced and non-displaced employees diminished only slowly and at the end of the follow-up period there was a remaining statistically significant difference of 3.0 and 1.3 percentage points, respectively.

Figure 4 presents similar estimates of the impact on annually received unemployment insurance in thousands of SEK. There are obvious spikes in the year of displacement and in the following year for both the men and women. These spikes corresponded to SEK 10,000–12,000 (580%–670%) and SEK 8,000–9,000 (330%–350%), respectively. However, there was a rapid return to smaller differences in the following 2 years (i.e., approximately SEK 3,000). Thereafter, there was no more than minor additional recovery for either the displaced men or the displaced women.

4.2 The impact of job displacement on the risk of divorce

After having established that job displacement inflicted both immediate and rather persistent earnings losses (though modest in an international context) as well as increased non-employment and unemployment among displaced married men and women, we can proceed to the main part of this paper, which examines whether job displacement also increased the risk of a subsequent divorce.²³ As a robustness check, estimates from all four estimators will be showed in the main analysis. In all of the analyses, the estimates using both applications of the time window procedure that defines the displaced employees will be displayed.

The main estimates are presented in Table 5. The unadjusted estimator yields a statistically significant increase in the risk of divorce from displacement in all of the samples. For the preferred 3-year window, the excess risks of divorce among couples in which the husband or wife was displaced were estimated to be 30% (HR, 1.30; 95% CI, 1.18–1.42) and 27% (HR, 1.27; 95% CI, 1.15–1.41), respectively.

Table 5 Estimated impact of husbands’ and wives’ job displacement on the risk of divorce

Regarding the adjusted estimates, it is clear from the table that the choice of estimator has a minimal impact on the estimates as long as one somehow adjusts for the baseline differences. Therefore, I will refer only to the propensity score weighted (PSW) estimator in the text. It is also clear that some of the unadjusted excess divorce risk is attributable to differences in the baseline sample characteristics since all of the adjusted estimates are considerably smaller. For the male sample, both time windows yield a statistically significant excess divorce risk from job displacement. Using the 3-year window, the estimated excess divorce risk is estimated to be 14% (HR, 1.14; 95% CI, 1.03–1.26), whereas the corresponding estimate using the 1-year window is ten percentage points higher (HR, 1.24; 95% CI, 1.11–1.39).²⁴ For the female sample, the 3-year window yields a statistically insignificant estimate of the impact of job displacement (HR, 1.10; 95% CI, 0.98–1.23). However, the 1-year window yields a statistically significant estimate also of the impact of wives’ job displacement corresponding to a 16% excess divorce risk (HR, 1.16; 95% CI, 1.02–1.32).

4.3 A subgroup analysis

In the previous section, it was established that the risk of divorce following job displacement was increased during the 12-year follow-up period. The impact was somewhat larger if the husband was the job loser. In this section, by performing a subgroup analysis, the aim is to learn about how the impact of job displacement depended upon the couples’ marital investments, economic dependency/independency, and coping resources.

In multiple subgroup analyses, one would expect both more false negative results (i.e., a failure to reject the null hypothesis given that the alternative hypothesis is actually true) because of the smaller sample sizes and more false positive results (i.e., a rejection of the null hypothesis given that it is actually true) because of multiple significance testing. Thus, the findings should be interpreted with caution. Because of the reduced number of divorces in each subgroup only the propensity score weighted estimator without additional covariate adjustments was used.²⁵ However, the estimates presented in Table 6 contain those obtained by applying both the 1- and 3-year windows.

Table 6 Estimated impact of husbands’ and wives’ job displacement on the risk of divorce, by subgroups

In the first subgroup analysis, the couples were divided into those with and those without marital investments. Owning a house may obviously be viewed as a marital investment, which was why house ownership was included as a control variable in the previous analyses. However, a division based on house ownership is not meaningful here since more than three-quarters of the couples owned a house. Hence, marital investments will be defined based only on whether the couples had children.

A priori, it is not clear whether or not children would be expected to have a protective effect. On the one hand, if children are seen purely as marriage-specific capital, the presence of children would require the shock to the marriage to be larger to result in a divorce. On the other hand, having children could increase the likelihood that the couple perceives the displacement as a stressor because more children mean more dependants.

However, the estimates indicate that the protective effect was larger, at least for the male sample. The estimated impact of the husbands’ displacement is statistically significant for those with no children, but not for those with children, and corresponds to an increased risk of divorce by 22% (HR, 1.22; 95% CI, 1.00–1.49) using the 3-year window. For the female sample, the impact of job displacement is of similar size in both subgroups, and none of the estimates are statistically significant.

Traditionally, it has been suggested that women’s increased economic independence within marriages over time, which is the result of increased labour force participation, is an important factor behind the increasing divorce rates. However, the empirical support is inconclusive (see Sayer and Bianchi 2000). Beckerian economic theory (Becker 1973, 1974) suggests that decreased specialisation within a marriage will decrease the gains from marriage and, therefore, increase the risk of divorce. Increased own earnings will also lower the barrier to divorce. However, two-earner families are the norm today. Both spouses’ incomes will also have a positive effect on their living standards and provide a buffer against economic uncertainty (e.g. uncertainty caused by job displacement, as in this case).

A spouse is defined to be economically dependent on his/her partner if the spouse’s own earnings are less than 40% of the couple’s total earnings. In equally dependent marriages, each spouse’s contribution will range from 40% to 60% of total earnings.

The estimates indicate that equal dependence within a marriage has a buffering effect. In all of the estimations, the impact of job displacement on the risk of divorce is minor and non-significant among the couples whose spouses are equally dependent. A supplementary analysis, which is not presented here, did not reveal any differential impact depending on whether the husband or the wife was in a dependent position.²⁶

Finally, the impact of job displacement on the risk of divorce for groups of couples with varying degrees of coping resources was investigated. First, the samples were divided based on whether the couples had suffered from previous hardships or difficulties. Couples’ experience of previous difficulties was narrowly defined as either spouse’s experience of either insured unemployment or hospitalisation. The other measure of coping resources was equally narrowly defined as financial resources. The couples defined as lacking financial resources had either received means-tested social benefits or belonged to the lowest income quartile and also had no taxable wealth. In the female sample, both financial resources and the lack of previous difficulties seem to have protected the couple against the adverse effects of job displacement, whereas job displacement significantly increased the risk of divorce both among the couples lacking financial resources and among those who previously had suffered from unemployment or ill-health. However, in the male sample, these coping resources seem to have been of negligible importance to how the couple handled the husband’s current job displacement.

4.4 The impact of job displacement on divorce over time

In the analyses in the previous sections, the impact of job displacement on the risk of divorce was assumed to be constant over time. This assumption was necessary to obtain enough precision to be able to empirically determine whether the impact on the risk of divorce was statistically significant. Nevertheless, it could be argued that such an assumption may not be valid. If job displacement is viewed as a discrete event, one may expect the impact on the risk of divorce to be immediate only. However, a job displacement is more likely to be a process that begins with the anticipation of displacement and is then followed by the actual separation, possibly a period of unemployment, reemployment, a period in which the person adapts to the new job, and so forth. In Section 4.1, it was also demonstrated that the displaced employees in the current sample experienced long-lasting earnings losses (as shown in a number of previous studies) and extended periods of non- and unemployment.

Furthermore, also a divorce is more accurately described as a process than as a discrete event. This process probably begins with marital dissatisfaction or conflicts before proceeding towards separation and finally divorce. Therefore, the length of the transition from the initial trigger or stressor (i.e., job displacement) until the couple becomes legally divorced is an empirical matter. Following Charles and Stephens (2004), I will separate the estimated effect of job displacement into short-run (i.e., up to 3 years following the displacement), intermediate (i.e., 4–5 years), and long-run (i.e., more than 5 years) effects. In terms of the specification of the second-step discrete-time logit model, the coefficient δ is replaced with a function $\Gamma \left( t \right)=\delta_1 I\left( {t\le 3} \right)+\delta_2 I\left( {4\le t\le 5} \right)+\delta_3 I\left( {t<5} \right)$, where I(·) is an indicator function, and t is the number of years since displacement. The second step was again estimated without any additional covariates, and the first-step estimation of the propensity scores included the full set of covariates discussed in Section 3.4. The resulting estimates are presented in Table 7 for both samples and both time windows.

Table 7 Estimated time-varying impact of husbands’ and wives’ job displacement on the risk of divorce

During the first 3 years following the husbands’ displacements, the hazard ratio was estimated to 1.21 (95% CI, 1.02–1.43) using the 3-year window. The wives’ job displacements show no such short-run effect. Instead, the largest estimated effect of wives’ displacements was found in the following 2-year period (HR, 1.14; 95% CI, 0.89–1.46). In the male sample, there was an opposite effect in these years, although it was not statistically significant. Coinciding with the deep recession there was an emerging long-term effect of job displacement also on the divorce risk. More than 5 years after the displacement, there was an excess divorce risk of 23% (HR, 1.23; 95% CI, 1.01–1.49) among the couples whose husbands were displaced. The wives’ job displacements had a long-term effect on the risk of divorce of similar size as the intermediate effect and also not statistically significant.

5 Summary and conclusions

A job loss is, through various means, likely to change the conditions of marriage, which could affect marital stability and increase the risk of divorce. From an economic point of view, a job loss may result in long-lasting earnings losses and financial strain, which may lead to marital conflicts and increased marital instability. However, to obtain a better understanding of the relationship between job loss and divorce, one also has to recognise that employment has meaning beyond a source of income. That is, having a job also affects lifestyle, social networks, and psychological well-being.

This paper examined both husbands’ and wives’ job displacements due to establishment closures and extended the existing literature by analysing the marital impact over a longer period (i.e., up to 12 years after the displacement). The exclusive focus on displacements due to establishment closures is important because it reduces the selection problem (i.e., that the job loss might have been caused by difficulties following a divorce and not vice versa, or that the person may possess characteristics that render both a job loss and a divorce more likely). Moreover, the focus on job displacements also excludes one potential explanation for why job losses in general and firings in particular may increase the risk of divorce: since all of the employees at the workplace are laid off, it is not reasonable to assume that the job displacement reveals any information about the spouse’s fitness as a mate. However, a person may very well be fired because of personal characteristics that are unfavourable also within a marriage.

The estimates show a positive and statistically significant impact of job displacement on the risk of divorce. During the 12-year follow-up period, the risk of divorce was elevated by 14% if the husband was displaced at baseline. This finding is in line with the estimates in Rege et al. (2007) of the effect of Norwegian husbands’ job displacement due to plant closure on the risk of divorce. It is notable that the estimates in both studies contradict the findings in the one study that previously has investigated the impact of job displacement due to plant closure. Charles and Stephens (2004) found that the divorce risk did not increase following plant closures. Rather, the risk only increased following other layoffs. They suggested that the information about a spouse’s non-economic fitness as a mate drives the relationship between job loss and divorce and that a job loss due to a plant closure does not reveal any such information because all of the employees are laid off regardless of their individual characteristics and behaviour. Therefore, they considered any effect of job losses due to plant closures as the effect that relates to purely economic considerations. However, other research has linked job displacements (due to plant closures) to, for example, emotional distress and excessive alcohol consumption. Hence, it seems obvious that job displacements may destabilise marriages through other channels.

There is conflicting previous evidence with regard to whether wives’ job losses and unemployment also affect marital stability and the risk of divorce. The findings in this study indicate that marriages are affected by both husbands’ and wives’ job displacements, although marriages are affected more so by husbands’ displacements. The similar findings by Hansen (2005) for Norway may suggest that the impact of wives’ job losses is specific to Scandinavian countries, where the labour force participation among married women is high, and wives are expected to be self-supporting, even within marriage. However, a subgroup analysis revealed no effect of wives’ job displacement on marriages in which the spouses were equally dependent.

A final finding is related to the timing of divorce. Husbands’ job displacements seem to have affected the divorce risk both in the short term and in the longer term but not in the intermediate term. It is possible that less stable marriages were dissolved shortly after the displacement. Based on previous research showing that displaced workers also suffer long-term earnings losses and increased risk of repeated job losses, it seems reasonable to claim that job displacements may affect marriages even over the longer term. However, the finding that job displacement had no impact in the intermediate term may suggest that the long-term effect depends on the general labour market conditions and that the subsequent deep recession caused more of the previously displaced employees to lose their jobs again. However, the wives’ job displacements showed a different time pattern. There was no excess risk of divorce in the short term, whereas the divorce risk was elevated in both the intermediate and long term although not statistically significant. Given the much smaller earnings losses following the wives’ displacements, the lack of an immediate adverse impact on marital stability is not surprising. A subgroup analysis also revealed that the increased divorce risk following wives’ displacements was limited to the couples who had experienced previous difficulties or who lacked financial resources. Thus, in the case of wives’ displacements, the impact on the divorce risk appears to be caused by an accumulation of difficulties that finally results in divorce.

In conclusion, job displacement is a multifaceted event that appears to be associated with severe losses for some in both monetary and non-monetary terms. However, for several reasons one must also recognise and understand the complexity of the many non-economic adversities that may follow job displacements. Many of the adverse effects of job displacements, such as family problems, divorce, and ill-health, not only are important determinants of personal and family well-being but may also play an important role in a chain of adversities that begins with job displacement. For some people, these adversities may serve as a partial explanation for their long-term difficulties in the labour market.