Background
There are several potential detrimental effects of divorce, such as lowered well-being, financial problems, long-term impaired functioning [
1], and an elevated risk of mortality [
2].
Mental distress and/or mental illness are important factors associated with divorce [
3-
10]. Explanations for this association are commonly classified within two different frameworks; social selection and social causation [
11]. In social selection it is assumed that mental distress leads to divorce and that mentally distressed people are thus “selected” out of marriage. Social causation posits that divorce leads to mental distress due to the stresses of the transition into divorce and the new status as divorcee.
Although there is broad support in the literature for the social causation hypothesis, selection effects may occur simultaneously [
1]. In fact, several studies have found support for both social causation and social selection [
12-
14]. For example, in a longitudinal study based on three-wave panel data from 930 respondents, Mastekaasa [
15] reported evidence for a short-term selection effect as well as a long-term social causation effect.
The present study focuses on social selection. There are a number of different pathways from mental distress and mental illness to divorce. Mental disorders may negatively affect people’s ability to maintain marital relationships [
5], which may in turn lead to divorce. For example, depressed patients and even patients with depressive symptoms tend to have impaired functioning both physically and socially, experience bodily pain and spend as many days in bed as people with chronic medical conditions [
16]. Mental illness is also related to low levels of social capital [
17]. Consequently, this may reduce mentally ill people’s capacity to participate in joint activities and provide their spouse with emotional support, which is important for companionship and individual well-being in spouses [
18]. Another pathway from mental distress to divorce may lead through socioeconomic status. Mental illness is associated with low education, unemployment [
19] and low family income [
20]. Since level of education, receiving welfare and level of income are also related to marital quality and (in) stability [
21], socioeconomic status may play a role in divorce. A third approach concerns similarities between partners. Research has consistently shown that partners resemble each other on mental and physical health as well as health behaviours [
22]. Explanations for this have been categorized into two main categories: non-random mating on the one hand, which entails similarity in partners even before they meet, and, on the other hand, increased similarity due to shared experiences and mutual influence after partnering, such as emotional contagion [
23]. Thus, people that are vulnerable and/or predisposed to develop mental distress may tend to select each other as partners, and may be exposed to the same negative life events or to mutual influence. This may in turn increase the risk of divorce, since couples with two mentally distressed partners are at an especially high risk of getting divorced [
7,
24].
An important limitation in the literature is the lack of research based on data from both spouses [
25]. Couple data are necessary in order to identify to what extent the association between mental distress and divorce exists at the individual level or at the couple level. Data on mental distress status of only one spouse is sufficient to examine whether people with mental distress get divorced more frequently, which is important in itself. However, data on mental distress status of both spouses give a unique possibility to take one step further than what has been possible in most previous studies. The risk in couples with two mentally distressed spouses could be exactly as calculated from combining the individual risk in each of the spouses (multiplying the excess risk in distressed wives with the excess risk in distressed husbands). However, there could also be an additional risk in doubly burdened couples not accounted for by the individual risks alone. For instance the distressed partners might be unable to care for the other. Or, on the contrary, the risk in doubly burdened couples could be lower than expected from combining the individual risks, perhaps because sharing the burden may give some comfort. Detecting such effects, in which the presence or absence of distress in one spouse moderates the risk of divorce associated with distress in the other spouse, requires data from both spouses in very large samples, like ours.
An early study based on couple data from a small, clinical sample reported that couples in which one partner was depressed did not have higher divorce rates than the general population, whereas couples in which both partners were depressed had a divorce rate 8 times higher, suggesting a couple-level effect in which the two partners did not have sufficient capacity to support and help each other [
24]. Butterworth and Rodgers [
7] recently tested the generalizability of this finding utilizing representative data, investigating whether the mental health problems of both spouses in 3,230 couples measured at baseline could predict marital dissolution during the subsequent 36 months. The results showed that couples with one or two partners with mental health problems were significantly more likely to separate/divorce than couples without mental health problems, but there were no significant interaction effect between mental health problems in husbands and wives. Although these results supported Merikangas’ [
24] finding that couples with two depressed partners have high divorce rates, they did not support the notion of a couple-level phenomenon. Butterworth and Rodgers [
7] thus concluded that their results seemed to reflect an additive effect of individual mental health problems rather than a couple-level mental health effect. Finally, the authors noted that although their results suggested a selection effect, the time frame of three years did not allow for an exclusion of social causation.
In sum, most previous research is based on information from one spouse only, and where information from both spouses were available, statistical power may have been insufficient despite relatively large samples. Hence, little is still known about couple-level selection effects.
The current study applies a longitudinal design attempting to replicate the findings of Mastekaasa [
15] and Butterworth and Rodgers [
7]. Our study expands on previous research in important ways. Whereas the abovementioned studies were based on 930 and 3,230 couples, respectively, our sample includes more than 20,000 couples, implying high statistical power. Unlike the relatively short three-year period in Butterworth and Rodgers’ study [
7], our 16 years follow-up period makes it possible to draw firmer conclusions with respect to selection effects not confounded by causation effects. As opposed to the study by Mastekaasa [
15], our data include information from both spouses, which allow us to test for combined main effects and/or interaction effects. To the best of our knowledge, this is the first prospective study based on a large, representative sample including data from both spouses to examine long-term selection effects. The aims of the study are as follows:
1.
Examine the association between mental distress and divorce over time, testing for long-term selection effects.
2.
Investigate the extent to which such effects reflect an individual-level or a couple-level phenomenon.
Discussion
The first aim of our study was to examine the association between mental distress and divorce over time, testing for long-term selection effects. In general, the results from the present study show that there is a significant association between mental distress and divorce. The current results expand on previous research by testing this association on longitudinal couple data from a large, representative sample. Couples with a mentally distressed husband or wife had more than a twofold risk of divorce compared to couples in which neither spouse suffered from mental distress, even after controlling for demographic variables and other covariates. The results show a peak in the effect of mental distress on hazard of divorce in the years immediately preceding the event and, further, that mental distress predicts divorce for as long as 8 years or more into the future. A social causation explanation for the peak in the effect of mental distress during the years around the divorce is plausible. However, the results showing a risk for divorce many years after the observation of mental distress provide evidence of a strong selection effect. This contradicts Mastekaasa’s [
15] finding, but supports other longitudinal research [
11,
32,
33].
Because marital problems have been found to predict divorce [
34], one might argue that our results may reflect chronic problems within the marriage leading to mental distress and subsequent divorce, rather than mental distress being a direct cause. In fact, a recent review of couple- and family-based treatments in depression stated that there seems to be a reciprocal relationship between marital quality and depressive symptoms [
35]. On the other hand, one study based on couple data found that although both partners’ degree of psychopathology was related to both partners’ degree of marital satisfaction, the more important factor for marital satisfaction was one’s own degree of psychopathology [
36]. In other words, the poorer one’s mental health, the more dissatisfied one may be with one’s marriage. Thus, marital problems may also be a result of mental distress. Unfortunately, our data did not include information on marital satisfaction.
Like mental distress, alcohol use has been shown in the same data material to predict divorce [
37], and alcohol use could well be suspected to mediate as well as confound the effect of mental distress. However, entering the demographic factors, social support, and life style, including alcohol use, as covariates did not change the estimates very much. These results imply that neither alcohol use nor the other covariates strongly confound or mediate the effects of mental distress on divorce.
The second aim of our study was to investigate whether the observed selection effects reflect an individual-level or couple-level phenomenon. To the extent that couples in our study concordant on mental distress have an especially high risk of divorce, our results support the findings of both Merikangas [
24] and Butterworth and Rodgers [
7]. However, our results show a significant interaction effect between husbands’ mental distress and wives’ mental distress in the first and second analyses, indicating that the elevated divorce risk among these couples is also related to mental distress at the couple-level. This contradicts the results from Butterworth and Rodgers’ study [
7] which seemed to reflect only an additive effect of individual mental health problems, and no interaction effect between mental health in each of the spouses. The interaction effect in our study was no longer significant when excluding couples who divorced within 8 years after baseline, but this is likely to result from loss of statistical power. The number of couples in which both partners were mentally distressed and who also experienced divorce was reduced from 72 couples in the first analysis, to 20 couples in the second analysis, and to 10 couples in the final analysis, still the estimates in the final model were very similar to those in the previous model. Likewise, lack of power may explain why Butterworth and Rodgers [
7] did not find evidence for an interaction effect.
Our finding of the interaction effect indicates that there may be a certain protective effect of being married to a person with a level of mental distress similar to ones own level, even in couples with two mentally distressed partners. This is supportive of the health mismatch hypothesis [
38] which posits that couples with concordant health status are at a lower risk of getting divorced than couples with discordant health status. Similar findings have been reported in other research. In a recent study on alcohol use, concordant heavy drinking predicted divorce to a lesser extent than what was expected from the combined main effects, possibly due to perceived compatibility or a judgement that it may be difficult to deal with the problems alone and to find a new partner [
37]. It is not difficult to imagine that this scenario might hold also for people with mental distress. Another explanation may be related to assortative mating, referring to the tendency for individuals to choose life partners with similar characteristics as themselves, which has been reported for psychiatric disorders [
39]. It may be that people with similar mental health understand each other better, and are thus better able to cope with challenges related to couple mental distress. In conclusion, couples with one or two mentally distressed partners in our study have a persistently higher risk of divorce than couples in which neither partner suffers from mental distress. The divorce rate for couples with two mentally distressed partners was lower than expected, but still high. Thus, our results suggest that mentally distressed individuals are indeed selected out of marriage.
Although gender differences are not a focus of our study, we note that the effect of mental distress on divorce was stronger for women than for men, contrary to the finding in Butterworth and Rodgers’ study [
7]. However, the sizes of the differences are well within what could be due to random fluctuations.
Limitations
Unfortunately, our data did not include information on marital satisfaction. This is important, as the effect of women’s mental health problems on marital disruption disappeared when controlling for women’s relationship dissatisfaction in the study by Butterworth and Rodgers [
7]. Likewise, Breslau and colleagues [
5] noted that the observed relation between mental disorder and divorce across 12 countries in their study may have partly been a result of preceding marital distress.
Despite our large sample, lack of power is probably the reason why an interaction effect was not detected in our final analysis, since the estimates of the interaction effect is highly similar for all three sets of analyses.
The design of our study did not permit us to investigate both long-term social causation and social selection effects.
We do not know how well our results generalize to other societies. For instance the risk associated with mental illness in both spouses could be higher in a society in which mental health services are less available than in ours, and in which mentally ill spouses to a larger extent are left to take care of each other.
Another limitation pertains to the lack of information on whether some of the couples in our study were remarried, since people who have previously divorced are more likely to get divorced again. Furthermore, our sample included married couples only and not cohabiting couples. It is, however, unlikely that the inclusion of cohabiting couples would have represented a substantial change of the results. A study by Ask and colleagues [
23] based on couple data from HUNT 1 estimated that about 1.2% of all the participating couples were cohabiting whereas the rest were married.
We chose to dichotomize our principal explanatory variables, mental distress in husband and wife. While this may be considered a limitation, because it implies losing some information, it makes the results more easily interpretable. Also, very skewed distributions of the MD variables make treating them as continuous predictor variables a little problematic.
Finally, despite our efforts to control for a wide range of covariates, we do not have information about circumstances that may occur in the period from baseline to year of divorce, such as the birth of (more) children, the death of relatives, changes in social support, changes in socioeconomic status, fluctuations in mental health and so on. Thus, our results should be interpreted with caution. For example, a major negative life event such as losing one’s job may negatively affect a family’s socioeconomic status and also the climate in a couple’s relationship and contribute to an eventual divorce.
However, our study has several strengths. It is based on couple data from a large, population based sample followed for many years. We were able to control for a range of relevant variables, including years of marriage, which is important because of higher dissolution rates in the earlier years of marriage [
40].
Future studies should be based on data from both spouses and, ideally, follow people for some years before they marry and then for many years after, in order to be able to examine social causation as well as social selection in the same sample. It is not surprising that divorce may lead to mental distress, but the question of whether mentally distressed people are selected out of marriage may be less straightforward. In our study, mental distress apparently seems to lead to divorce, but this association may also be due to unknown third factors such as marital dissatisfaction, economic hardship or shared negative life events. Such factors should be studied in more detail. The dynamics of the shared climate of two mentally distressed spouses is also a subject that deserves more attention. Why is it that such couples in our study did not divorce as frequently as would have been expected from the double risk? Is it because both spouses lack the resources to implement the process of divorce, or have some of these couples developed certain strategies that help them understand each other and lead a relatively well functioning life together despite it all? Or maybe shared exposure to a major negative life event caused mental distress in both spouses but tied them closer together rather than result in marital conflicts. Answers to these questions may aid mental health professionals in identifying couples at risk for divorce, and in helping such couples to understand and deal with challenges related to mental distress both as individuals and couples.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
MI participated in the conception and design of the study, performed the statistical analyses, interpreted the results and drafted the manuscript. FAT provided support in performing the statistical analyses, participated in the interpretation of the results and revised the manuscript critically. IB participated in the interpretation of the results and revised the manuscript critically. KR participated in the interpretation of the results and revised the manuscript critically. ER participated in the interpretation of the results and revised the manuscript critically. KT participated in the conception and design of the study, provided support in performing the statistical analyses, participated in the interpretation of the results and revised the manuscript critically. All authors read and approved the final manuscript.