Separations of romantic relationships are experienced differently by initiators and noninitiators

Significance

In economic and sociological theory of divorce, the link between divorce consequences and the decision to divorce is central: A couple divorces if at least one spouse expects to improve their life by initiating divorce. The present study provides empirical evidence in support of this theoretical link: Separation initiators become better off in terms of subjective well-being after a separation, whereas noninitiators become worse off, before they eventually experience a full recovery. Because separations are predominantly initiated by only one partner, this finding suggests that one partner typically benefits from the separation (the initiator), while the other is disadvantaged (the noninitiator). Accordingly, analyses of average divorce trajectories convey only limited information about the causal effects of divorce on individuals’ well-being.

Abstract

Divorces are predominantly initiated by one spouse alone. This might suggest that one spouse typically benefits from divorce (the initiator), while the other is disadvantaged (the noninitiator). At the same time, empirical research on the consequences of divorce commonly focuses on the average effect for both partners. In contrast, I estimate separation trajectories individually for initiators and noninitiators of formerly cohabitating or married couples. The analysis covers a wide range of outcomes and a long period of time surrounding the separation. I employ an event-study design based on individual fixed effects, thereby accounting for time-invariant individual heterogeneity that could be linked to initiator status and the outcomes. The results reveal substantial differences in separation trajectories between initiators and noninitiators. Initiators indeed improve their subjective well-being after a separation and also see gains in other life domains, with the exception of the economic domain. Noninitiators experience significant short-term losses in subjective well-being, from which they recover in the long run. Noninitiators’ trajectories in other life domains vary.

In economic and sociological theory of divorce, the link between divorce consequences and the decision to divorce is central: A couple divorces if at least one spouse believes that a better life would be lived without their current partner. Sociological work often assumes that persons who expect to improve their lives through divorce will be the ones to initiate divorce. Conversely, persons who believe that they would worsen their lives will not initiate divorce. Accordingly, initiators and noninitiators should experience opposing overall effects from divorce.* While this line of reasoning predicts the direction of the overall effect of divorce on initiators and noninitiators, it does not predict the life domains in which they gain or lose; neither does it predict when potential effects may occur or whether the timing is the same for initiators and noninitiators.

Empirical evidence on these points is very scarce. A small number of studies have analyzed retrospective self-evaluations of the divorce based on initiator status, e.g., questions such as “Has your divorce improved or worsened your peace of mind?” (1, 2). A drawback to this study design is that respondents may have biased recalls, as this type of question directly links the divorce decision to subsequent outcomes and people tend to attribute positive features to options they have chosen in the past (3). Only one published study uses contemporaneous longitudinal data to overcome this limitation (4). However, the study focuses exclusively on mental and physical health outcomes. It finds that noninitiators’ health tends to worsen after divorce. In contrast, the consequences for initiators’ health are more ambiguous: Depending on the outcome variable, their physical health either improves or remains unchanged after divorce whereas their mental health worsens or remains unchanged. While some additional studies analyze initiator status and postdivorce outcomes, these studies cannot inform us on any of the theoretical predictions made above, as they focus exclusively on a comparison of initiators and noninitiators in the period after the divorce, often at only one point in time (5–12). Results from these studies either indicate that initiators do better than noninitiators after separation or indicate that there is no difference.

This lack of evidence on the link between the decision to separate and its consequences stands in stark contrast to the interest these topics have generated individually and to how central this link is in theory. An enormous literature explores the consequences of divorce, spanning various academic fields including demography, economics, epidemiology, psychology, and sociology.^† However, this literature does not disentangle the potentially opposing trajectories of initiators and noninitiators. This likely conceals considerable heterogeneity in divorce trajectories, as prior research shows that 70 to 90% of divorces are predominantly initiated by only one partner (37–41).

As with interest in divorce consequences, the divorce decision has also received considerable attention from researchers. This work often focuses on identifying the determinants of divorce initiation (37, 38, 40–44). The assumption in these studies is that factors leading to divorce initiation reflect how well someone would be doing in the marriage versus after a divorce. This logic implies that divorce initiation is associated with better postdivorce outcomes. As laid out above, there is little evidence on whether this is actually the case.

Therefore, I aim to provide more evidence on the link between the separation decision and subsequent outcomes. For this purpose, I estimate separation trajectories for initiators and noninitiators of formerly cohabitating or married couples from 4.5 y before to 4.5 y after the separation. I analyze 12 outcomes covering subjective well-being and three life domains: partnership, social life, and the economic domain. By analyzing a wide range of outcomes in the same sample, I provide insight into the domains in which initiators and noninitiators gain or lose. This represents a significant extension of the range of outcomes compared to prior literature. While the analysis of well-being represents some overlap with Hewitt and Turrell (4), the analyzed indicators of well-being differ. The outcomes analyzed by Hewitt and Turrell (4) stem from the Short Form 36 Health Survey (45, 46). In contrast, this study relies on a standard life satisfaction question and a depression indicator based on the state-trait-depression scales (47). Finally, I explore heterogeneity in the separation trajectories of initiators and noninitiators by gender and by the speed with which they repartner. The analysis of heterogeneity by gender represents an important robustness check because most initiators of separation are women. On a methodological level, I employ an event-study design based on individual fixed-effects regressions, which accounts for the fact that unobserved time-invariant individual heterogeneity may be correlated with initiator status and outcomes. This is a difference compared to Hewitt and Turrell (4), who rely on a random-effects model.

Theoretical Background and Predictions

Theoretical insights into divorce decisions and trajectories have originated from several academic disciplines, including economics, sociology, and psychology. This section presents these insights along with the hypothesis central to this study: Initiators will benefit from a separation, while noninitiators will be disadvantaged.

Basic Economic and Sociological Modeling of Divorce.

Economic theory models marriage and divorce as rational choices within an expected utility framework (48, 49). Two individuals marry if both of their individual expected utilities are higher being married to each other than being single or being in a relationship with someone else. In other words, marriage occurs if the expected marital gain is positive for both partners. Once married, it is often assumed that utility within the couple is (perfectly) transferable. This leads to bargaining between spouses about the distribution of the marital surplus (the sum of both partners’ marital gains).

According to this framework, divorce occurs if the expected utility of divorcing (and potentially remarrying) is higher than the expected utility of remaining married for at least one partner. In other words, divorce occurs if the marital gain is negative for at least one partner (after potential transfers of utility between partners).^‡ It is generally assumed that divorce only occurs due to unexpected shocks to at least one partner’s utility (50, 51).

In sociology and social psychology, social exchange theory provides a very similar analysis of the decision to divorce (52). Levinger (52) describes three forces that influence a couple’s cohesiveness. On the one hand, the attractiveness of staying in the couple and the barriers to separation (such as social norms that ostracize divorce) make divorce less likely. On the other hand, “alternative attractions,” i.e., potential benefits from leaving the partner, make divorce more likely.

Separation Initiation and Separation Payoffs.

It seems reasonable to think of divorce initiation as resulting from an individual’s comparison of the two alternatives of staying in the couple versus leaving: The partner for whom the marital gain turns negative is the one who will initiate the divorce. It is also possible that the net gains from marriage become negative for both partners at the same time, causing a consensual divorce.^§

This view of separation initiation is frequently taken in empirical sociology, usually when interpreting the fact that a certain factor is linked to divorce initiation.^¶ For example, the fact that women are much more likely to initiate divorce than men can be interpreted as evidence that women benefit less from marriage, at least among couples that divorce (37). Moreover, “initiating” and “wanting” the divorce are often used interchangeably (38, 44). This is supported by evidence that there is reasonable to very good overlap between survey answers on initiating the divorce and wanting it or being in favor of it (5, 41, 56, 57).

Predictions.

For the purpose of this study, one property of the theory presented so far is central: People who initiate a separation expect to become better off through the separation than through remaining in the marriage. Conversely, the noninitiating partner expects to become worse off through a separation, compared to staying together (otherwise, the partner would have initiated a separation). From the economic perspective, “better off” and “worse off” refer to changes in discounted expected lifetime utility.

Perhaps just as interesting as the above prediction are two aspects on which the theory remains silent. First, the theory does not predict how initiators and noninitiators fare in individual life domains. It predicts only that initiators expect to do better overall and noninitiators expect to do worse overall through the separation. It is quite possible that initiators experience deterioration in some life domains but gain disproportionately in others, since only the overall impact of a separation is expected to be positive. Second, the theory does not make predictions about the timing of the (expected) effects. While gains for initiators may be immediate, they may also suffer short-term losses for which they are compensated only in the long run. The timing of gains and losses may also differ across life domains. Thus, the distribution of gains and losses across life domains and over time remains an empirical question.

Refinements of the Basic Framework.

Further insights, drawn primarily from psychology, can enrich this basic framework. Individual experience may differ from the theory presented above for several reasons. For example, there is evidence that people do not always accurately predict the effects of major life events on their well-being (58, 59). Accordingly, if their expectations of the consequences of separation are inaccurate, their actual experiences may differ substantially from expectations. Even if there were a bias in expectations, it is not clear whether this bias would be different for initiators and noninitiators. If initiators and noninitiators were biased in the same way, the effects for them might not have the expected direction. Nonetheless, a difference between initiators and noninitiators should be apparent.

Moreover, regarding subjective well-being (SWB), set-point theory enriches predictions about the timing of potential effects. Set-point theory posits that major life events like a divorce can influence SWB in the short term but not in the long term, because individuals revert to a baseline level of well-being in the long run (60, 61). Consequently, any effect of separation on SWB should be transient.

Other arguments suggest that there will be larger differences between initiators and noninitiators. Earlier research has highlighted that initiators are more likely to anticipate the separation (7, 62). Accordingly, they may prepare better for it, which could be reflected in better adjustment before and after the separation. Moreover, SWB could be affected because noninitiators may feel that they have less control over the divorce situation (7), which can lead to a higher likelihood of experiencing stress, as well as feelings of helplessness and hopelessness (63–65). Similarly, Weiss (66) indicates that initiators are more likely to experience feelings of guilt and remorse, whereas noninitiators feel hurt and rejected.

While the theory presented so far is gender neutral, sociologists have posited some arguments that support the notion that male initiators differ from female initiators in their separation trajectories (as male noninitiators differ from female noninitiators). Based on gender role theory, Symoens et al. (11) predict that male noninitiators fare worse emotionally than female noninitiators and that their mental health is affected more negatively. They argue that men suffer more from feelings of powerlessness and a loss of control during a separation because they cherish power, control and autonomy more than women (67, 68). Conversely, it has been suggested that male initiators might be less prone to experience feelings of guilt and remorse than female initiators (69), implying that male initiators may fare better than female initiators in terms of mental health.

Empirical Strategy

Pairfam Data.

The empirical analysis relies on the German Family Panel (pairfam) (70). Pairfam is a panel survey focused on relationships and family dynamics that surveys three birth cohorts (1971 to 1973, 1981 to 1983, 1991 to 1993) over 11 yearly waves from 2008 to 2019. Pairfam offers a representative sample of 13,891 individuals in the three cohorts, with the exception of oversampling East German individuals (in the DemoDiff subsample with n = 1,489). Individuals in the DemoDiff subsample participate only from wave 2 onward. As long as a couple stays together, both partners are interviewed. Once a couple separates, only the primary respondent (anchor) is tracked. A detailed description of the dataset is given in Huinink et al. (71).

Separation Initiation in Pairfam.

Pairfam provides information on separation initiation in the following way: Individuals are asked about separation initiation if (at the time of the current interview) they are no longer in a romantic relationship with their partner from the previous interview and if this is not due to the death of their partner.^# The question on separation initiation is phrased “How strongly did you take the initiative to end the partnership?”. Respondents can answer on a scale of 1 (not strongly at all) to 5 (very strongly).^ǁ Alternatively, they can also answer “I don’t know” or “I don’t want to answer this.” Based on their answers to this question, I classify respondents as either initiators or noninitiators to analyze the separation trajectories of these two groups. I classify respondents as initiators if they answer with a 4 or 5 to the initiation question or as noninitiators if they answer between 1 and 3. Individuals who answer in the middle category (with a 3) are categorized as noninitiators because earlier research indicates that people attribute slightly higher levels of initiation to themselves than their former partners do (38, 55). SI Appendix, section 4.C provides some further discussion of this methodological choice and shows that it has only a very modest influence on the results.

Analyzed Outcomes.

Since a separation potentially influences many facets of a person’s life, this study presents evidence on a wide array of outcomes. The analyzed outcomes cover SWB and three life domains: partnership, social life, and the economic domain. SWB is assessed using two indicators: life satisfaction (measured on a 0 to 10 scale) and a binary indicator for being depressed. The two outcomes measure two components of SWB: cognitive well-being (CWB) (in the life satisfaction scale) and affective well-being (AWB) (in the depression scale). CWB is conceptualized as a cognitive overall evaluation of one’s life, while AWB refers to the presence and frequency of positive and negative affect (72, 73). Including measures for both CWB and AWB is important because prior research has shown that the relationship between external factors and CWB and AWB is not necessarily the same (74, 75). In particular, a metastudy reports that the effects of divorce for CWB and AWB can differ (20).

The indicator for being depressed is based on the German version of the trait part of the state-trait-depression scales (T-STDS) (47). The T-STDS comprises 10 items, of which half assess negative mood and half assess positive mood. Respondents are presented with a statement assessing mood and are asked to choose the answer that best describes how they feel in general on a scale ranging from 1 (almost never) to 4 (almost always). Examples include the statements “I am happy,” “I enjoy life,” “I am sad,” and “My mood is melancholy.” After inverting the positive mood scale, I added the answers and assigned the depression dummy a value of 1 if the sum is greater than or equal to 25 out of a maximum of 40 points. I selected this cutoff based on an evaluation of various cutoff points by Lehr et al. (76).** The dummy is created only if individuals have answered all 10 items of the T-STDS.

Outcomes depicting partnership and family outcomes include having a romantic partner and satisfaction with multiple spheres, including satisfaction with one’s sex life, satisfaction with the relationship, and satisfaction with family. Satisfaction with the relationship is surveyed only if the individual reports having a partner. The next set of outcomes portrays a person’s social life. These outcomes include satisfaction with leisure activities, hobbies, and interests, as well as satisfaction with friends and social contacts. Another group of outcomes refers to the economic domain. This group contains net household income according to the equivalence scale of the Organisation for Economic Co-operation and Development (OECD), employment status, and hours worked per week, as well as satisfaction with school, education, and career.^†† All satisfaction measures are reported on a 0 to 10 scale. Except for depression symptoms, all variables were surveyed in every wave. Depression symptoms on the STDS scale were included only from wave 2 onward in the main sample and from wave 4 onward in the DemoDiff subsample.

Event-Study Approach.

This paper analyzes whether separation trajectories show a positive development for initiators and a negative development for noninitiators, as the theory laid out before suggests. For this purpose, I adopt an event-study design based on individual fixed effects that is similar to the design used in other studies on divorce trajectories (13, 19). The regressions include a set of treatment dummies that measure the time to separation and a set of control variables including individual fixed effects, as well as age- and sex-specific time-fixed effects. For the main results, I run separate regressions for initiators and noninitiators. The regressions include a control group that consists of cohabitating and married couples that did not experience a separation.

The setup can be described more formally as follows:^‡‡ Let E_i denote the survey wave after a separation event for an individual and let t denote the survey wave. Then, the relative time to separation for the individual (measured in survey waves) is $t-E_i$. Fig. 1 shows an example of a separation between waves 5 and 6. In that case, the relative time to separation $t-E_i$ in wave 6 is 0. For the regression, I create treatment dummies that are equal to one if the relative time to separation lies within a certain range. For example, $1\{t-E_i\in [1,2]\}$ denotes a dummy variable equal to one if the time to separation is either 1 or 2. I split relative time periods symmetrically around the separation into six disjoint sets covering the time period from five waves before to five waves after the separation: One set is the observation immediately after (before) the separation ($t-E_i=0$ and $t-E_i=-1$, respectively). Observations 2 to 3 after (before) the separation also form a set. Finally, observations 4 to 5 after (before) a separation form a set. Thus, the superset G of treatment dummies included in the regression contains $\{[-5,-4],[-3,-2],\{0\},[1,2],[3,4]\}$ as the observation before the separation ($E_i-t=-1$) is chosen as the base category and is consequently excluded from the regression. Observations for relative time periods outside the analyzed interval are also excluded from the estimation. The following equation describes the regression specification:

$$y_{it}={\displaystyle \sum _{g\in G}{\gamma }_g}·\mathbb{1}\{t-E_i\in g\}+{\alpha }_i+{\lambda }_{\mathit{cst}}+u_{it}.$$ [1]

Fig. 1.

Timeline for a separation between survey waves 5 and 6.

Here, y_it is the outcome for individual i at survey wave t; α_i denotes the individual fixed effect; and λ_cst a fixed effect resulting from an interaction of pairfam age cohort, sex, and wave. λ_cst is supposed to flexibly capture age- and sex-specific time trends that could influence the outcomes. ${\displaystyle {\sum }_{g\in G}}\mathbb{1}\{t-E_i\in g\}$ denotes the treatment dummies discussed above. The coefficients γ_g on these dummies measure the deviation of the outcome variable from the baseline period immediately before the separation, after adjusting for the other control variables. That is, they measure the deviation from the baseline after accounting for individual time-invariant heterogeneity (via the individual fixed effects) and the age- and sex-specific time trends. In this setup, the purpose of the control group is to contribute to the estimation of the time trends.^¶¶

Estimation Sample.

Separations must fulfill the following criteria to enter the estimation sample. First, the study includes only separations of couples who were together at one or more interviews and who were cohabitating or married for at least 1 mo in the 12 mo prior to the separation.^## This coincides with the separations for which there is information on initiation (except for the aforementioned peculiarity of wave 3 and the filter error in the survey). The inclusion of formerly cohabitating couples is motivated by the fact that cohabitation is an important form of partnership in Germany, with 32% of children in Germany born into nonmarried relationships (79). As a second criterion, only the first such event is used for each individual. Thus, an individual does not enter with multiple separations into the estimation. As a third criterion, respondents’ answer to the separation question must lie between 1 and 5. I thus drop cases in which respondents chose the answer options “I don’t know” or “I don’t want to answer that.” Applying these three criteria, 1,150 separation events are obtained, of which 670 are for women and 480 for men.^###

The estimation sample also includes a control group. Individuals enter the control group from the moment that they are in a cohabitating or married relationship at an interview date if they remain in that relationship for their remaining time in the survey. In other words, they contribute to the control group with all available survey waves from the moment that the cohabitation/marriage begins. Further criteria are that individuals were excluded from the control group if they had experienced a separation from a prior cohabitating or married relationship during their survey participation or if they were concerned by the filter error in the question on separation initiation. Additionally, individuals also must have had at least two observations that fulfill these criteria. Otherwise, they would not contribute to the estimation due to the individual fixed effects. Applying these criteria adds 4,937 individuals with 32,084 observations to the estimation.

Descriptive Statistics.

Fig. 2 shows the distribution of the answers to the question on separation initiation by sex. The striking feature of the data is that nearly 60% of all women answer that they initiated the separation very strongly, i.e., with a 5 on the 1 to 5 scale. In contrast, men’s answers are more evenly distributed, with a tendency to say that their partner initiated the separation. Because the pairfam survey tracks only the main respondent after a separation, men’s and women’s answers do not need to be mirror images to be consistent. Instead responses could differ by sex because male and female responses refer to different relationships. SI Appendix, Table S8 provides further information on the underlying relationships and indeed shows that there are some differences between the (former) relationships of the men and the women in the sample. Furthermore, studies that are able to observe both partners’ answers on separation initiation find that their responses coincide most of the time (38, 55). Taking this into account, Fig. 2 indicates that women are more likely to initiate separations. This finding aligns with prior research from other contexts (37–39, 41, 43, 80–82). Given that the majority of initiators are women, I analyze initiators and noninitiators separately by gender in the section on heterogeneity.

Fig. 2.

Histogram of separation initiation by sex in the estimation sample.

By splitting individuals into initiators (answer between 4 and 5) and noninitiators (answer between 1 and 3), we end up with 662 initiators and 488 noninitiators. Hence, there are more initiators than noninitiators in the sample. This is congruent with the fact that there are more women in the sample and that women are more likely to initiate separations.

Results: Separation Trajectories of Initiators and Noninitiators

This section graphically presents the estimated coefficients and 95% confidence intervals in Figs. 3–6. Each figure contains one category of outcomes. Fig. 3 shows SWB surrounding a separation, Fig. 4 partnership outcomes, Fig. 5 social life, and Fig. 6 economic outcomes. Initiators’ trajectories are always depicted in black; noninitiators’ trajectories are in gray. The x axis in Figs. 3–6 displays the average time to separation, in years. For a more fluid reading experience, the time frame 0.5 y after the separation is referred to as the short run, 1.5 to 2.5 y after the separation is the medium run, and 3.5 to 4.5 y after the separation is the long run. In addition to the graphical presentation in this section, SI Appendix, section 4 contains the exact regression results and tests for differences in the coefficients between initiators and noninitiators.

Fig. 3.

(A and B) SWB outcomes for initiators (black lines) and noninitiators (gray lines).

Fig. 4.

(A–D) Partnership outcomes for initiators (black lines) and noninitiators (gray lines).

Fig. 5.

(A and B) Satisfaction with social life for initiators (black lines) and noninitiators (gray lines).

Fig. 6.

(A–D) Economic outcomes for initiators (black lines) and noninitiators (gray lines).

Subjective Well-Being.

Fig. 3 displays the estimated trajectories for the SWB outcomes. Fig. 3A shows the trajectories of initiators and noninitiators regarding life satisfaction. For initiators, life satisfaction significantly decreases before the separation, by 0.25 points on the 0 to 10 scale. After the separation, it remains lower in the short run but then increases by even more than the initial loss. For noninitiators, there is no significant pretrend, although there is a very drastic drop of 0.73 points immediately after the separation. Noninitiators still experience lower life satisfaction 1.5 to 2.5 y after the separation than before the separation. However, in the long run, they recover completely.

The results look very similar for depression (Fig. 3B). Initiators experience an increase in depression levels of 5 percentage points before the separation, but then rebound quickly to initial levels after the separation. Unlike the results on life satisfaction, this rebound begins immediately after the separation. For noninitiators, there is (again) no significant pretrend, although there is a large and sudden increase in depression levels of 9.1 percentage points immediately after the separation, followed by a return to the baseline. Considering that the initial depression levels of noninitiators immediately before the separation are at 12.8%, this 9.1 percentage point increase corresponds to a 71% increase in depression levels.

These results are consistent with set-point theory, in that both initiators and noninitiators return to their initial levels of SWB after a separation. For initiators, affective well-being as measured by the depression indicator returns to baseline somewhat faster than cognitive well-being (as measured by life satisfaction). However, one could argue that the return to baseline materializes only because the chosen definition of initiator status does not show the most extreme trajectories. To address this potential limitation, SI Appendix, section 6 repeats the analysis using a slightly modified definition of initiators and noninitiators. In SI Appendix, section 6, individuals are classified as initiators if they answered “very strongly” to the question on separation initiation (a 5 on the 1 to 5 scale) and as noninitiators if they answered “not at all” (a 1 on the same scale). The results remain consistent with set-point theory—both initiators and noninitiators return to baseline. The primary difference is that noninitiators now do worse immediately after a separation. Noninitiators now experience a short-run drop in life satisfaction by a full point and an increase in depression levels by 16 percentage points. Nevertheless, they still recover completely over the long run. In comparison, initiators’ trajectories are less affected. They now experience slightly larger declines in both life satisfaction and depression before the separation and slightly weaker recoveries afterward.

Partnership Outcomes.

Fig. 4 shows trajectories for outcomes that describe individuals’ partnership and family lives. In contrast to the methodology used for the other outcomes, the results for having a romantic partner (Fig. 4A) stem from a simple regression of the outcome on the time-to-separation dummies. Therefore, the coefficients measure the average deviation from the baseline level immediately before the separation. Since, by construction, everyone has a partner immediately before the separation, the coefficients are equal to 1 minus the share of individuals having a partner. (For the outcome “having a partner,” it is not purposeful to include the individual and year fixed effects. The reason for including these fixed effects in Eq. 1 was to compare divorcees’ trajectories to a scenario in which they had not separated. For the outcome “having a romantic partner,” this is achieved by the simplified regression, since it compares divorcees’ trajectories to the case in which everyone has a partner, i.e., a flat time trend.) The results indicate that trajectories prior to the separation are nearly identical for initiators and noninitiators. The coefficients of approximately –0.25 (3.5 to 4.5 y before the separation) imply that 75% of initiators and noninitiators already had a partner at that point in time. After the separation, the paths of initiators and noninitiators diverge sharply. Initiators are much faster to enter into a relationship again, with 30% already having a partner 0.5 y after the separation. In the long run, this value rises to 75% for initiators. In contrast, noninitiators are slower to enter a relationship again and do not catch up with initiators in this regard even in the long run. At the observation right after the separation, there is a gap of 15 percentage points in the likelihood of having a partner between initiators and noninitiators. In the long run, this gap still amounts to 8.5 percentage points.

Given the speed at which initiators find new partners, it is reasonable to suspect that some of these new relationships had already started before the previous relationship ended. The pairfam data yield insights into this by asking participants about infidelity in the year before the separation. The answers indicate that infidelity can indeed partially explain why initiators find a partner more rapidly after a separation: 22.5% of initiators report having been unfaithful versus only 7.4% of noninitiators. (Histograms of the answers to the question on infidelity can be found in SI Appendix, section 7.) Taking these answers at face value and assuming that affairs likely lead to new relationships, this would imply that the likelihoods that faithful initiators and noninitiators find a new partner are more alike than they appear from the results in Fig. 4A.

There are also large differences in relationship satisfaction between initiators and noninitiators (Fig. 4B). This question is asked only if the respondent has a partner. Before the separation, both initiators and noninitiators experience significant declines in relationship satisfaction, with the decline being larger for initiators. Initiators lose a full point on the 0 to 10 scale from 4.5/3.5 y before the separation until immediately before the separation. After the separation, initiators’ and noninitiators’ trajectories both move upward, but this time their differences are even more pronounced. Initiators experience a massive 2.1 point increase in relationship satisfaction in the short run, while noninitiators gain by nearly a full point over the preseparation level. These differences become smaller in the long run but still persist. The fact that individuals are asked about relationship satisfaction only if they have a romantic partner can explain why even noninitiators experience an increase immediately after the separation. It also implies that the gap between initiators and noninitiators cannot be explained by the differential likelihood of finding a partner. Instead, the results indicate that initiators are more likely not only to be in new relationships but also to find better ones.

Regarding satisfaction with one’s sex life (Fig. 4C), initiators experience a drop of 0.24 points before the separation that is significant at the 10% level. Immediately after the separation, they gain 0.45 points. This gain more than doubles in the long run. In comparison, noninitiators do not experience any significant pretrends but do experience a strong decrease of –0.86 points immediately after the separation. However, in the medium run, they have already recovered completely. In the long run, their satisfaction with their sex life has even increased by 0.54 points. This is remarkable, as even noninitiators seem to gain from the separation, at least in the long run.

In terms of satisfaction with family, initiators experience a drop right before the separation but then stabilize (Fig. 4D). The coefficients in the medium and the long run after the separation are slightly positive but not statistically significant. In contrast, noninitiators do not exhibit any significant pretrends, although they experience a large drop of –0.75 points immediately after the separation. Even in the long run, they recover only partially from this drop.

Social Life.

Fig. 5 illustrates trajectories for satisfaction with leisure activities and for satisfaction with friends and social contacts. For both outcomes, there are no significant pretrends for either initiators or noninitiators. Immediately after the separation, initiators gain in both domains (at least at the 10% significance level). However, they cannot sustain these gains: While coefficients remain slightly positive in the long run, they are no longer significant. Noninitiators’ postseparation trajectories are nearly identical to those of initiators in terms of satisfaction with friends and social contacts. In contrast, noninitiators do not experience any gains in satisfaction with leisure activities after the separation. These findings suggest that both initiators and noninitiators can use time previously spent with the ex-partner on other activities after a separation. However, these gains vanish in the long run. This may be linked to both initiators and noninitiators becoming more likely to enter a relationship again.

Economic Outcomes.

Fig. 6 depicts the trajectories for variables pertaining to the economic domain. Interestingly, both initiators and noninitiators experience a drop in monthly OECD equivalized household income after the separation (Fig. 6A). For noninitiators, this decrease already begins before the separation. However, both groups rebound from this shock in the long run. Heterogeneity by Gender reveals that these negative short-run coefficients are largely driven by the women in the sample. Patterns are less clear for the subjective evaluation of their educational and career situation (Fig. 6B). Initiators’ trajectories resemble a flat line, even though the coefficient for the period 1.5 to 2.5 y before the separation is significant at the 10% level. Compared to the situation 3.5 to 4.5 y before the separation, noninitiators experience an increase of 0.29 points above their level 0.5 y before the separation. In the short run after the separation, they remain at this higher level; in the medium run, their satisfaction decreases before rebounding in the long run.

Both initiators and noninitiators see small increases in the likelihood of employment of ∼2 percentage points in the period before the separation (Fig. 6C). However, these increases are not statistically significant. After the separation, initiators begin a downward trend and are 6.3 percentage points less likely to be employed in the long run. As with household income, these negative consequences for initiators are driven by female initiators. While noninitiators mirror this postseparation trajectory over the short and medium runs, they return to their baseline level of employment in the long run, even though the coefficient is not very precisely estimated. This yields the surprising conclusion that, in terms of employment, noninitiators do better than initiators in the long run (at least at the 10% significance level). With respect to weekly working hours, initiators see an increase of 2 h worked before the separation but then no significant further changes (Fig. 6D). While noninitiators experience a similar increase before separation, it is not statistically significant.

Heterogeneity

This section investigates two factors that could aid the understanding and interpretation of the findings presented in the section before. In a first step, I analyze the extent to which the results depend on finding a new romantic partner after a separation. In a second step, I check whether trajectories of initiators and noninitiators differ by gender. Additionally, SI Appendix, section 10 compares the separation trajectories of cohabitating and married couples by initiator status.

The Importance of Repartnering.

This section examines how much results depend on finding a new romantic partner after a separation. (Technically, a few individuals in the sample repartner with their ex-partner so that speaking of a “new” partner is not always entirely correct.) This analysis is motivated by a combination of two factors. First, initiators are more likely to be in a relationship after a breakup and this gap persists even in the long run. Second, having a partner is likely linked to the analyzed outcomes. This is obvious for the partnership outcomes, such as satisfaction with sex life, but also holds true for the other outcome categories. For instance, there is empirical evidence that having a romantic partner positively affects SWB (83, 84). Moreover, social lives could also be affected because there is likely a trade-off between spending time with a romantic partner and time with other social contacts or on leisure activities. Furthermore, economic theory predicts that relationships facilitate household specialization and thus affect economic outcomes, at least in cohabitating and married partnerships. OECD equivalent household income also depends in a mechanic way on whether a person lives with a romantic partner.

To investigate whether having a partner indeed influences the results, I reran the analysis but split initiators and noninitiators into two groups depending on whether or not they had a new partner in the survey wave after the separation. (A total of 40 individuals skipped the survey wave immediately after their separation. These individuals are excluded from this analysis.) The results are summarized below. Figures comparable to those of the prior section and exact estimation results can be found in SI Appendix, section 8.

The overall picture that emerges from this analysis is that finding a new romantic partner is important for some outcomes but not for others. SWB outcomes and satisfaction with one’s sex life are highly affected by finding a new partner, while family and social life are not. In terms of economic outcomes, household income and satisfaction with education/career are unaffected, while employment and hours worked are affected, at least for initiators.

Regarding SWB (SI Appendix, Fig. S8), initiators and noninitiators who immediately find a new partner now have remarkably similar separation trajectories. They both see immediate increases in life satisfaction after the separation of 0.52 and 0.56 points, respectively. In contrast, initiators and noninitiators who do not find a new partner experience losses in the short run. However, their trajectories still differ in this case, as noninitiators’ losses are much larger. The results for depression are similarly affected. Initiators and noninitiators who find a new partner have very similar depression trajectories, mostly because noninitiators do much better if they find a new partner than if they do not.

With respect to partnership outcomes (SI Appendix, Fig. S9), the number of outcomes to be analyzed is halved, as having a partner is used to split initiators and noninitiators into groups and the question on relationship satisfaction is posed only to respondents who have a partner. These two outcomes are thus excluded from the heterogeneity analysis. In terms of satisfaction with one’s sex life, initiators and noninitiators who immediately find a new partner fare very similarly, with increases of 2.4 and 1.96 points in the short run, respectively. In contrast, if they do not find a new partner, initiators experience significant increases in satisfaction with their sex life only in the medium and the long run, but not in the short run. Noninitiators see a large decline of –1.4 points immediately after the separation, from which they recover in the medium and the long run. Thus, there is again large heterogeneity from finding a new partner, especially for noninitiators. For satisfaction with family, finding a new partner is not associated with significant heterogeneity for initiators, although it is for noninitiators. The analysis reveals that the negative effect for noninitiators in terms of satisfaction with family found in the previous section that does not take repartnering into account is mostly driven by those who do not find a new partner.

Regarding the two outcomes representing satisfaction with one’s social life (SI Appendix, Fig. S10), the results are only minimally affected by the speed at which individuals find a new partner. Comparing initiators who immediately find a new partner to those who do not, the former are somewhat less satisfied with leisure activities before the separation. Moreover, they are less satisfied with friends and social contacts in the medium run after the separation.

With economic outcomes (SI Appendix, Fig. S11), I do not detect any significant heterogeneity for household income and satisfaction with education/career, which may partially result from the relatively large SEs. In contrast, for employment and hours worked, an interesting pattern emerges among initiators. Initiators who immediately find a partner after separation experience a large increase in employment probability of 11 percentage points from 3.5 to 4.5 y before the separation to immediately before the separation. After the separation, their employment probability begins to decrease until reaching a decline of 15 percentage points after the separation. Trajectories of initiators who do not find a new partner do not exhibit such a pattern and instead resemble a flat line. The results for hours worked reflect these changes in employment probability.

Heterogeneity by Gender.

This section investigates whether there is heterogeneity in the trajectories of initiators and noninitiators by gender. Several reasons suggest that such an analysis is important. Because women account for roughly three-quarters of initiators in the sample but only 40% of noninitiators, we might otherwise confound differences driven by initiator status with differences driven by gender. This concern is substantiated by prior literature that reports differences between genders in adaptation to divorce, especially in the short run (13). Moreover, some sociological work predicts differences in the magnitude of effects between female and male initiators, as well as female and male noninitiators. Symoens et al. (11) predict that male noninitiators will be worse off emotionally than female noninitiators because being a noninitiator entails a feeling of powerlessness, which affects men more than women. Conversely, male initiators may be better off than female initiators because they experience less guilt due to the breakdown of the relationship (69). This could be reflected in the SWB outcomes, particularly in the depression dummy that measures AWB. To analyze separation trajectories by gender, I ran regressions as in Eq. 1 separately by gender. The results are summarized below. Tables of estimated coefficients can be found in SI Appendix, section 4.A and the figures for these coefficients are found in SI Appendix, section 9. [SI Appendix, section 4.A also contains estimates of separation trajectories by gender but irrespective of initiator status, which are similar to those in the analysis by Leopold (13).]

For SWB trajectories (SI Appendix, Fig. S12), I do not find large gender differences. Female noninitiators experience a downward trend in life satisfaction before separation while male noninitiators do not experience such a trend. Moreover, initiators’ medium-run gain in life satisfaction stems largely from the women in the sample. For depression, the short-run increase of noninitiators after a separation appears to be driven by women.

Men in the sample seem to be driving some of the results regarding partnership outcomes (SI Appendix, Fig. S13). Compared to their female counterparts, male initiators and noninitiators tend to experience larger increases in the likelihood of having a partner before the separation, while declines after a separation are more similar between men and women. Male initiators may have stronger short-term increases in relationship satisfaction and satisfaction with their sex life after a separation. Among male noninitiators, the large drop in satisfaction with family immediately after a separation is noteworthy.

Men and women also show some differences in satisfaction with their social life (SI Appendix, Fig. S14). Regarding satisfaction with leisure activities, the analysis indicates that women are driving the short- and medium-run positive results for initiators. Male initiators experience a significant drop in satisfaction with friends and social contacts before the separation while female initiators do not.

In the economic domain (SI Appendix, Fig. S15), the most noteworthy result is that, regardless of initiator status, women experience immediate drops in household income while men’s household income remains unaffected or even increases in the long run compared to the base level immediately before the separation. The levels of male initiators’ satisfaction with school and career 1.5 to 2.5 y before the separation and immediately after the separation are significantly higher than the baseline levels right before the separation at the 10% level. For the other economic outcomes, there do not seem to be important differences between genders, although any interpretation of the coefficients must be careful to consider the large SEs.

Conclusion

This paper analyzes the link between the separation decision and separation trajectories. Economic and sociological theory predicts that a separation affects initiators and noninitiators differently: Initiators improve their overall well-being while noninitiators become worse off. The present study supports this prediction. For initiators, the separation coincides with a downward trend in subjective well-being before the separation and a rebound to initial levels after the separation. For noninitiators, well-being is stable before the separation, declines sharply immediately after the separation, and then also rebounds. These patterns are also consistent with set-point theory as they indicate that a separation does not have a lasting effect on subjective well-being.

While theory makes very clear predictions about overall well-being, it was not clear a priori in which life domains initiators and noninitiators would gain or lose. This study illuminates this issue regarding three life domains: partnership, social life, and the economic domain. For partnership outcomes, the results reveal that noninitiators are significantly slower than initiators in finding a new romantic partner and also experience smaller gains in satisfaction with new relationships after the separation. Whereas noninitiators experience large drops in satisfaction with sex life and with family after a separation, initiators’ satisfaction with sex life increases and their satisfaction with family remains stable. Both initiators and noninitiators improve their social lives after a separation, at least in the short run. In the economic domain, initiators and noninitiators experience short-term losses in their household income after a separation. In contrast, their satisfaction with education/career, employment status, and hours worked remain largely unaffected by the separation. Moreover, many effects are transitory, as both initiators and noninitiators adjust to the separation in the long run.

The study also analyzes heterogeneity in trajectories by gender and by how fast individuals repartner. The results reveal that finding a new romantic partner has a strong influence on subjective well-being and the partnership domain but only a modest impact on social life. Since initiators repartner more quickly, this suggests that some differences in separation trajectories between initiators and noninitiators are driven by this circumstance. Compared to the importance of repartnering, gender had only a limited influence on the trajectories of initiators and noninitiators.

These findings have implications for theory and existing studies. From a theoretical perspective, they lend support to the notion that separation trajectories are strongly linked to the separation decision. More specifically, they support the ideas that initiatorship indicates how individuals perceive the costs and benefits of divorce and that individuals base their decision to divorce on such a cost–benefit analysis. Furthermore, the results suggest that (depending on the outcome) studies analyzing the average effects of divorce without considering initiator status underestimate short-term effects. In addition, gender differences in divorce trajectories found in other studies (13, 85) may be linked to the gender imbalance in the divorce decision, as women are much more likely to initiate divorce.

The results also provide suggestive evidence on the reasons why people seek a separation. For example, economic factors may not be central to the decision to end a relationship, considering that initiation is not very predictive of economic trajectories. In contrast, the availability of other romantic partners may be more important in the separation decision, since initiation is very predictive of the likelihood of finding a new partner, which in turn is strongly linked to other outcomes. Such an interpretation would also be consistent with recent evidence showing a decline in the importance of economic factors vis-à-vis nonmaterial needs when searching for a romantic partner (86).

Data Availability

The pairfam data is made available by the GESIS Leibniz Institute for the Social Sciences at https://doi.org/10.4232/pairfam.5678.11.0.0 (70). See also https://www.pairfam.de/en.