Existing Research

Existing research has, with little exception, found turnout to be higher among married individuals (Cutts and Fieldhouse Reference Cutts and Fieldhouse2009; Leighley and Nagler Reference Leighley and Nagler2013; Plutzer Reference Plutzer2002; Plutzer and Wiefek Reference Plutzer and Wiefek2006; Strate et al. Reference Strate, Parrish, Elder and Ford1989; Timpone Reference Timpone1998; Wolfinger and Rosenstone Reference Wolfinger and Rosenstone1980; Wolfinger and Wolfinger Reference Wolfinger and Wolfinger2008) and that marital disruptions, caused by divorce or the death of a spouse, have a negative effect on the probability of voting (Bhatti, Fieldhouse, and Hansen Reference Bhatti, Fieldhouse and Hansen2020; Hobbs, Christakis, and Fowler Reference Hobbs, Christakis and Fowler2014; Kern Reference Kern2010; Stoker and Jennings Reference Stoker and Jennings1995; Wolfinger and Wolfinger Reference Wolfinger and Wolfinger2008). Although these studies generally find a divorce effect (see appendix A.2.12 for an overview), the possibility of a gendered effect of individuals’ marital status on voter turnout has rarely been raised.

On the one hand, there exist theoretical reasons to expect that marital status will be more consequential for women’s turnout than for men’s. In the words of Wolfinger and Wolfinger (Reference Wolfinger and Wolfinger2008, 1515), “[d]ivorced and, to a lesser extent, widowed people fare worse than their married and never-married peers on a variety of social indicators … that are related to voter turnout.” And although both women and men suffer significant socioeconomic consequences from divorce, women’s economic standard of living declines more in the wake of divorce (Bianchi, Subaiya, and Kahn Reference Bianchi, Subaiya and Kahn1999; Smock, Manning, and Gupta Reference Smock, Manning and Gupta1999; Weitzman Reference Weitzman1985). In addition, women face greater responsibilities as single parents (Leopold Reference Leopold2018) and are more likely to move after a divorce (Mulder and Wagner Reference Mulder and Wagner2010). In view of this, and given the prominence of resource-based explanations for political participation (Rosenstone Reference Rosenstone1982; Verba, Schlozman, and Brady Reference Verba, Schlozman and Brady1995), one possible expectation would be that women’s turnout will be more adversely affected by marital disruption than that of men (cf. Kern Reference Kern2010; Shore Reference Shore2020; Voorpostel and Coffé Reference Voorpostel and Coffé2012).

On the other hand, the idea of spousal mobilization suggests that marital status will instead be more consequential for men’s turnout. On this account, married individuals are more likely to vote than their unmarried counterparts because of interpersonal influence between partners (Bhatti, Fieldhouse, and Hansen Reference Bhatti, Fieldhouse and Hansen2020; Cutts and Fieldhouse Reference Cutts and Fieldhouse2009; Glaser Reference Glaser1959; Leighley and Nagler Reference Leighley and Nagler2013; Stoker and Jennings Reference Stoker and Jennings1995; Timpone Reference Timpone1998; Wolfinger and Wolfinger Reference Wolfinger and Wolfinger2008).Footnote ⁶ This spillover effect may exist because partners remind each other to vote, or they go to the polls together (Bhatti, Fieldhouse, and Hansen Reference Bhatti, Fieldhouse and Hansen2020), thus reducing the information costs of voting and also increasing the satisfaction derived from the act itself. The spillover effect may also exist because partners exert normative pressure on each other (Blais, Galais, and Coulombe Reference Blais, Galais and Coulombe2019). In the words of Abrams, Iversen, and Soskice (Reference Abrams, Iversen and Soskice2011, 234), “if politics is seen as important during an election period in one’s network of friends and family, then voting gains social approval and not voting leads to social disapproval.”

The loss of one’s more politically active partner may thus cause one to become more passive.Footnote ⁷ Although there is a considerable degree of cross-country variation, recent research suggests that in the case of voting, the more active partner is often the woman. Historically, turnout in many democratic countries was lower among women than men, perhaps due to the lingering effects of female disenfranchisement (Franklin Reference Franklin2004). More recently, however, this pattern has been reversed in many countries (Carreras Reference Carreras2018; Cascio and Shenhav Reference Cascio and Shenhav2020; Coffé and Bolzendahl Reference Coffé and Bolzendahl2010; Leighley and Nagler Reference Leighley and Nagler2013). One potential explanation for this can be found in psychological research on personality traits, which finds that women consistently exhibit higher degrees of conscientiousness (Schmitt et al. Reference Schmitt, Realo, Voracek and Allik2008). This trait is in turn related to greater adherence to norms, such as the one that states that it is a civic duty to vote (Gallego and Oberski Reference Gallego and Oberski2012). Consequently, Carreras (Reference Carreras2018, 40) argues that “these personality differences are critical to explain why women are more likely to go to the polls on Election Day.” All this suggests that spousal influence will be asymmetrical. When women vote more than men, we should expect men’s turnout to be more adversely affected by divorce.

Given the above, it is surprising that there are only a couple of studies that examine the gendered effect of divorce and separation on political participation. Kern (Reference Kern2010) uses the British Household Panel Survey to examine the influence of divorce on, among other outcomes, turnout. The study presents separate estimates for women and men, and although they are rather imprecise—in all likelihood due to data limitations—the magnitudes of these effects do not appear to differ substantially. Voorpostel and Coffé (Reference Voorpostel and Coffé2012) instead use the Swiss Household Panel and find that women’s self-reported participation in referenda falls in the wake of divorce or separation, whereas no effect is observed for men. In both studies, the authors had expected a more negative effect of divorce and separation for women because women fare worse on a number of sociodemographic indicators related to turnout.

Although we should be cautious in overinterpreting the results from a few studies that rely on smaller datasets than the one used here, the results, if taken together, are more in line with the theoretical accounts emphasizing spousal mobilization than those emphasizing sociodemographics. Specifically, although women’s turnout in many democratic countries has surpassed that of men during recent decades, this is not true of Great Britain and Switzerland. In Great Britain, voter turnout is very similar across genders, so we would not expect gendered spousal mobilization. In Switzerland, where turnout is still higher among men (perhaps due to the late enfranchisement of women), spousal mobilization implies a more negative effect of divorce on women’s voter participation. In a new study of voters in an Italian city, finally, Bellettini et al. (Reference Bellettini, Ceroni, Cantoni, Monfardini and Schafer2020) find that marriage has a greater positive influence on voter turnout among men, although there is little evidence of a gendered divorce effect.

As we shall see in the next section, Sweden, which is the focus of the following empirical analysis, exemplifies the broader trend toward higher turnout among women than among men. If spousal mobilization is the main driver, men’s turnout in Sweden should fall more in the wake of divorce than that of their female partners. If, on the other hand, sociodemographic changes are the main explanation for the divorce effect, the turnout of women should be more adversely affected than that of men.

Institutional Setting

This study uses populationwide data from Sweden to uncover the influence of divorcing on the turnout of women and men. In order to divorce individuals must, however, first get married. Despite ongoing secularization, it is still fairly common to get married in Sweden. The marriage rate in Sweden lies slightly above the EU average and has been increasing over the last couple of decades. According to available statistics, about 70% of all persons living together in a relationship in Sweden do so as married (Eurostat 2016).

As in many other Western countries, divorces in Sweden became increasingly common during the course of the last century. The right to no-fault divorce was introduced in Sweden in 1915. Divorce rates increased gradually during the first half of the twentieth century. However, in the mid-1960s, partly as a result of progress in women’s social and economic emancipation, divorce rates began to rise rapidly. Between 1965 and 1975, Swedish divorce rates more than doubled, going from about 5 divorces per 1,000 married women to more than 12 divorces per 1,000. In recent decades divorce rates in Sweden have been fairly stable, currently standing at about 13.8 divorces per 1,000 married women.

Turning to the dependent variable of our study, elections to the Swedish Parliament are held in September every fourth year, and all Swedish citizens aged 18 and older are eligible to vote. By international standards, Swedish voter turnout is fairly high. In recent decades, overall turnout has varied between 80% and 90%. From the 1980s, average turnout for women has been slightly higher than that for men (Statistics Sweden 2012). However, the gender differences in voter turnout become more marked if we zoom in on turnout in certain age groups, as is done in Figure 1.

Figure 1.

Turnout by Gender and Age

The graph displays average turnout by gender and age in the three general elections that will be studied here—that is, those held in 1994, 2010, and 2018. As can be seen, women’s turnout lies clearly above that of men for all but the very oldest age groups. The reversal of the relationship for older individuals is in all likelihood due to a combination of a cohort effect—the individuals in the oldest age groups were socialized into voting at a time when men voted more than women—and gender differences in mortality patterns. However, from the perspective of this study, the primar insight from Figure 1 is that the average turnout is higher for women than for men in the age span during which divorces usually occur.

Data and Method

Turning to the data at our disposal, they mostly derive from various public registers maintained by Statistics Sweden. One of these registers contains yearly information on the civil status of all Swedish inhabitants together with the date when the individual first obtained this civil status code. From these data we can construct start and end dates of the universe of marriages recorded in Sweden during the period 1972–2019 (an individual is coded as divorced in the registers until he or she remarries). We then match the sample of divorced individuals to another database, which contains detailed yearly sociodemographic data for all Swedes above age 16 for the period 1990–2018. The latter data also include household identifiers that make it possible to link the divorced individuals to their previous spouses.Footnote ⁸

In the next step, we add turnout data from the three general elections held in 1994, 2010, and 2018. For these three elections, individual-level data on voter turnout have been made available through digitization of the publicly available election rolls. The turnout data cover more than 90% of the electorate for the elections in questions and have been shown to be highly reliable (Lindgren, Oskarsson, and Persson Reference Lindgren, Oskarsson and Persson2019). In addition to this, we have access to digitized turnout data for the two general elections of 1970 and 1982, which we will use for some of our analyses.Footnote ⁹ The reason for not including these two earlier elections in our main analyses is that annual data on the socioeconomic indicators are only available from 1990.

From these data, we select all individuals who divorced during the period 1985–2019. We choose 1985 as the start period because 1994 is the first year in which we observe voter turnout. By using 1985 as the start date we get a sufficiently long period before the first election in 1994 while at the same time making sure that a sufficiently large share of the individuals divorcing early in the period are still alive during the two later elections.

In total, there are about 1.4 million unique individuals with a divorce date between 1985–2019 in our data. If the turnout of these individuals could be observed in each of the three elections under study, we would thus obtain a sample of approximately 4.2 million observations. In reality this is not possible, as some of the individuals divorcing late in the period were too young to vote in 1994 and some of those divorcing early in the period died or emigrated before the later elections. There is also a small amount of missing data in the turnout variable, which further reduces the maximum sample size that can be used for the study. Dropping the individuals who divorced between 1985–2019, yet for whom we lack complete turnout data, leaves us with a sample of about 1.2 million unique individuals and 3.3 million individual-year observations.

We will report some results for this maximum sized sample, but in an effort to strengthen identification and examine potential mechanisms we conduct most of our analyses using a more restricted sample. We provide a detailed description of the various sample restrictions that we invoke and their effects on the sample size in the appendix (see Table A.1). Some of these restrictions are more technical in nature and state that we must be able to link the individuals to their previous spouses and that we must have at least one data point for both spouses. However, two of the restrictions are of more substantive nature. First we exclude all individuals who divorced more than once during the period 1972–2019. The reason for this is that individuals with multiple divorces can be simultaneously “treated” by two or more divorces, which blurs both identification and interpretation. Second, we exclude information from all elections occurring before the divorcees were initially married. By invoking this restriction we make sure that we use married individuals, rather than a mixture of married and later to be married individuals, as the comparison group when estimating the divorce effect. Once we have imposed these additional sample restrictions, we are left with a final sample of 1,684,967 observations for 739,664 individuals in 369,832 unique couples.

The reasons for invoking these additional sample restrictions become even clearer once we consider the statistical model that we will use to study the divorce effect. To identify the effect of a divorce on the turnout of women and men, we will estimate a dynamic linear probability model of the following form:

(1) $$ {\displaystyle \begin{array}{rcl}{y}_{i,c,t}& =& \sum \limits_{k=-7}^7{\beta}_k^g{D}_{c,t}^k+{\alpha}_1^g{D}_{c,t}^{k\le -8}+{\alpha}_2^g{D}_{c,t}^{k\ge 8}+{\theta}^g{\boldsymbol{X}}_{i,c,t}+{\eta}_c\\ {}& & +\gamma {\overline{y}}_{g,c,t}+{\varepsilon}_{i,c,t},\qquad \end{array}} $$

where y _i,c,t is turnout of individual i, in couple c, in election t, and $ {D}_{c,t}^k $ refers to a set of dummy variables set equal to one if an individual divorced k years before an election (negative values of k indicate that the divorce is to take place k years after the election). The superscript g attached to the coefficients of these indicators indicates that the divorce effect is allowed to differ by the gender of the individual. For reasons of statistical precision, we estimate a joint effect for divorces taking place more than seven years before ( $ {\alpha}_1^g $ ) or after ( $ {\alpha}_2^g $ ) the election, respectively. In estimating this model, we will use k = –3 as the reference category—that is, all estimates will be presented in relation to the turnout three years before a divorce. The reason for choosing –3 as the base year is that, as we will see later, the effect of the marriage troubles are often starting to show some years prior to the actual divorce.

For our main analyses we will thus be reporting the effect of divorcing up to seven years before or after an election, although the coefficients ( $ {\alpha}_1^g $ ) and ( $ {\alpha}_2^g $ ) will capture the average effect of divorces occurring outside this seven-year window. Admittedly, it would be interesting to extend the observation window beyond seven years to further analyze the persistence of the divorce effects. A drawback of increasing the observation window, however, is that the composition of divorces and elections will change with the length of the follow-up period. For instance, it is only for individuals who divorced fairly early in the period that we can study the development of turnout 15–20 years after a divorce. Therefore, we consider the observation window of seven years as a reasonable compromise, but we will be reporting results for a longer window (±15 years) as a robustness check.

Along with the variables that capture the effect of divorce timing, our model includes some additional controls. The vector X _i,c,t comprises a small set of predetermined individual characteristics, such as sex, years of education, and immigrant background, whereas $ {\eta}_c $ denotes a full set of couple fixed effects. The inclusion of the couple fixed effects in the model means that we will only use the within-couple variation in turnout to identify the divorce effects for women and men, which greatly reduces the risk of omitted variable bias.

In studying voter turnout, it is fairly standard to control for age and election year fixed effects. In the presence of the couple fixed effects, however, the election year variable will be perfectly collinear with the divorce timing variable. If we instead replace the couple specific effects with individual fixed effects, as will be done as a robustness check, the age effects too will become perfectly collinear with divorce timing. To circumvent this problem, we instead add a control for expected turnout given an individual’s birth cohort, sex, and the year of election (ӯ _g,c,t ). More precisely, we do this by calculating the mean turnout in the entire electorate for all unique combinations of birth year, sex, and election year and add this average as a control to our model. This procedure is akin to residualizing the dependent variable with respect to the three variables used to construct the groups. By adding this control to the model, we can adjust for the effect of age and election year without causing collinearity. Our approach rests on the assumption that the age and election year fixed effects are the same for the population of divorcees as for the electorate as a whole.

In our main analyses we will thus apply the estimation procedure described by Equation 1 to our final estimation sample, which includes about 1.7 million observations for the group of individuals who got divorced between 1985 and 2019. In Table 1, we present some basic descriptive statistics for this sample.

Table 1.

Descriptive Statistics

Voter turnout in general elections is high in Sweden, and in our sample the average turnout is 87% for men and 89% for women. The average number of years of schooling is slightly below 12 years for men and slightly above 12 years for women, and about a quarter of our sample has immigrant background, which here means that either the individuals themselves or at least one of their parents were born abroad. The average age when experiencing a divorce is 44 years for men and 42 years for women. As explained above, the exact number of observations will differ across individuals and couples, but we have the maximum number of observations (three for individuals and six for couples) for about about 40% of all individuals and one-third of all couples (see Table A.2 in the appendix).

Is the Divorce Effect Gendered?

In analyzing how divorce affects voter turnout among men and women, it is illuminating to begin by looking at the bivariate relationship between turnout and divorce timing. To this end, Figure 2 displays average voter turnout by month since divorce at election time. For instance, turnout at month 12 is the average turnout for those that divorced 12 months before the election, and turnout in month –12 is the average turnout for those experiencing a divorce 12 months after the election. This figure is based on the largest sample possible—that is, we only require that the divorce should have taken place within seven years (84 months) and that we have information on turnout for the individuals being divorced. The results presented in Figure 2 are thus not affected by the additional sample restrictions that we use in our later regression analyses, and which, among other things, require that an individual is only divorced once and is married in the predivorce period.

Figure 2.

Turnout and Time since Divorce

As can be seen, married women and men have fairly similar turnout rates before their divorces. For instance, in elections occurring three years prior to a divorce, the average turnout is around 87% among women as well as men. Turnout then starts to decline for both sexes. For women and men divorcing right around the time of an election, the turnout rates are only about 80% and 78%, respectively. Once the divorce has taken place, turnout rates start to rise again, but as can be seen they still remain below their predivorce levels seven years after the divorce.

The pattern displayed in Figure 2 suggests three important findings. First, a divorce has a substantial negative effect on voter turnout for both women and men, and this effect starts to materialize already 2–3 years prior to the actual divorce, although it is most marked right around the time of the divorce. Second, the reduction in turnout seems to be fairly long-lasting as it is clearly visible still seven years after the divorce. Third, whereas a divorce is associated with lower turnout rates among both sexes, men seem to experience a larger drop in turnout when going through a divorce.

To examine whether these findings hold up to closer scrutiny, we next estimate the regression model described in Equation 1 using yearly data. In doing so, we also impose the larger set of sample restrictions described above, meaning that the following analyses will be based on a subset of the data used for Figure 2. One important advantage of the regression framework is that it allows us to account for the influence of age and election-year effects. Equally important, the inclusion of couple-fixed effects means that the divorce effect is estimated based only on within-couple variation in turnout. The couple-fixed effects will effectively control for all factors that remain constant over time for the couples going through a divorce. All models include controls for sex, immigrant background, and years of schooling.

The upper panel of Figure 3 shows estimated divorce effects for women (solid line) and men (dashed line) based on our regression model. Because Figure 2 indicated that turnout rates start to decline already a couple of years before the divorce, we use the year –3 as our reference category. Thus, all effects are measured in comparison with turnout in elections taking place three years prior to the divorce.

Figure 3.

Divorce Effect on Turnout

Note: Complete coefficient estimates together with standard errors clustered on the couple identifier are reported in Table A.3 of the appendix.

Overall, the regression results mimic those of the bivariate analysis. We observe stable turnout levels for both sexes until three years before the divorce, then turnout starts to decline. If we compare turnout in the year of the divorce to turnout three years prior to the divorce, the drop in turnout is about 7.5 percentage points for women and slightly above 8 percentage points for men. As time passes, the turnout rates gradually start to increase again, but the recovery is faster for women than for men. Seven years after the divorce, women are about 3 percentage points less likely to vote than they were as married, whereas the corresponding figure for men is about 5.5 percentage points.

The regression results thus corroborate the finding that men’s turnout is more adversely affected by a divorce. To further illustrate this, the lower panel of Figure 3 shows the differences in the divorce effects for women and men together with 95% confidence intervals. As can be seen, the gender gap in the divorce effect begins to materialize a couple of years before the divorce and then the gap gradually grows until three years after the divorce when it stabilizes around 2.5 percentage points. The gender differences in the divorce effect are statistically significant at the 5% level from year 0 and onward (the regression coefficients together with clustered standard errors are reported in Table A.3 of the appendix).

Why is the Divorce Effect Gendered?

The previous findings provide clear evidence that going through a divorce suppresses voter turnout, and that this effect is particularly pronounced for men. However, so far our analyses tell us little about why this is the case. In this section, we attempt to remedy this by investigating various potential causal mechanisms.

In the section on existing research, we discussed some different reasons why a divorce should affect voter turnout and why that effect may differ between women and men. A first possibility is that the divorce effect is driven by changes in socioeconomic status and residential mobility. Going through a divorce is typically associated with a decline in the socioeconomic position of an individual and a higher probability of residential mobility, two factors that previous research suggests are associated with lower voter turnout (e.g., Rosenstone Reference Rosenstone1982; Squire, Wolfinger, and Glass Reference Squire, Wolfinger and Glass1987). To the extent that divorces affect the socioeconomic status and residential mobility of women and men differently, this could potentially also help explain why going through a divorce seems to affect the turnout of women and men to different degrees.

In addition, women are more likely to continue to live with their children after a divorce. This have led some scholars to suggest that a divorce should have more detrimental consequences for women’s turnout because they will have “less time to participate in political and social life” (Voorpostel and Coffé Reference Voorpostel and Coffé2012, 30). On the other hand, because women are more likely to continue to live together with their children and take responsibility for their daily social activities they are also less likely to become socially isolated after going through a divorce. Because social isolation is known to be strongly related to political inactivity (e.g., Putnam Reference Putnam1995; Reilly Reference Reilly2017), this mechanism posits that women’s role as the primary caregiver will mitigate, rather than enhance, the negative divorce effect on turnout.

In Figure 4 we report how residential mobility, socioeconomic status, and household composition develop for men and women after a divorce. The analyses are based on the same data and regression setup as that used for our main analyses.

Figure 4.

Effect of Divorce on SES, Mobility, and Household Composition

Note: Complete coefficient estimates together with standard errors clustered on the couple identifier are reported in Table A.4 of the appendix.

The first two subgraphs show that the the probability of moving both within and between municipalities increases substantially for both sexes around the time of the divorce. The next three subgraphs show how family disposable income (expressed in 1,000 SEK and 2010 constant prices) and the probability of receiving positive amounts of means-tested social and housing allowances develops with divorce timing. As can be seen, a divorce has large negative economic consequences for both sexes, but in particular for women. For women, the average family disposable income drops by 40%–50%, and their probability of receiving social or housing benefits more than doubles when going through a divorce. The last subgraph shows that women are more likely to continue to live with their children after a divorce. For women, the number of children under the age of 18 in the household drops by about 0.5 children when divorcing, whereas the corresponding reduction for men is about one child per household.Footnote ¹⁰

The key question here, however, is to what extent the effects reported in Figure 4 can help account for the divorce effect on voter turnout. One way to examine this is by performing a mediation analysis in which we add controls for socioeconomic status and residential mobility to our baseline regression model and observe how much the divorce effect changes in response. Figure 5 displays the results from such an analysis. For reasons of comparison, the left part of the figure reproduces the results for the model without controls for potential mechanisms, whereas the rightmost part shows the divorce effects once the variables studied in Figure 4 have been added as controls.

Figure 5.

Mediation Analysis

Note: Complete coefficient estimates together with standard errors clustered on the couple identifier are reported in Table A.3 of the appendix.

By comparing the two graphs in Figure 5, we can see that the drop in voter turnout associated with the divorce becomes somewhat less pronounced once we control for the potential mechanisms. In the baseline model, divorce reduces turnout by about 7.5 percentage points for women and 8 percentage points for men, whereas the corresponding figure in the model with mechanism controls is about 6 percentage points for both sexes. To judge from these results, the intermediary variables can help account for about one-quarter of the overall effect of divorce on turnout.

Turning instead to the difference in the divorce effect for women and men, the gap decreases by about one percentage point from 2.5 to 1.5 percentage points when adding the additional controls to the model. Closer inspection reveals that this drop is entirely due to the variable measuring the number of children living in the household. If we remove this variable from the analysis but retain the controls for SES and residential mobility, the gender difference in the divorce effect remains at 2.5 percentage points. The results thus suggest that changes in SES and residential mobility are not very helpful in explaining why turnout develops differently for women and men following a divorce. However, the fact the gender gap in the divorce effect diminishes by about 40% when controlling for the number of children in the household could be taken to indicate that the reason for the gendered divorce effect is social rather than economic in nature. This observation thus points us in the direction of the spousal political mobilization hypothesis, which was the second potential mechanism highlighted in the theoretical section.

To recap the argument, the spousal mobilization theory posits that because voting is often a joint household activity, more politically active individuals should mobilize their less politically active partners. In a context such as the Swedish one, where voter turnout is higher among women than men, this could potentially help explain why men’s turnout is more negatively affected by a divorce. It is equally important to note that this line of reasoning also implies that men’s voter turnout should increase more when a couple is formed.

Unfortunately, we cannot observe how the turnout levels of women and men are affected by initial family formation because the household indicator at our disposal only identifies couples who are either married or cohabiting with joint children, whereas cohabiting couples without children are treated as separate households. So as a second best, we decided to examine how turnout varies with the time from marriage for the individuals in our sample. To do this, we use additional data on turnout from the general elections in 1970 and 1982. The results from this analysis are reported in Figure 6. In the leftmost panel, we plot the average turnout, together with 95% confidence intervals, for women and men by years elapsed from marriageFootnote ¹¹ . The rightmost panel instead reports the results from a regression model of the same type as that used to study the effect of a divorce.

Figure 6.

Marriage Effect on Turnout

Note: Complete coefficient estimates together with standard errors clustered on the couple identifier are reported in Table A.3 of the appendix.

Beginning with the leftmost panel, we see that men have lower turnout than women when unmarried, but that this difference largely disappears once they get married. Although marriage seems to foster higher turnout among both women and men, the effect is thus particularly pronounced for men. This effect becomes even clearer when examining the regression results presented in the rightmost panel. Despite the fact that it is common for Swedish couples to cohabit for some time before getting married, we observe a clear marriage effect on turnout. Moreover, this effect is stronger for men. Turnout develops very similar for both sexes until 1–2 years before the marriage, after which men’s turnout suddenly starts to increase at a faster rate. If we compare post- and premarriage turnout, we see that this difference is about 1–1.5 percentage points larger for men than it is for women.

The pattern shown in Figure 6 is thus a mirror image of the effects of martial disruptions presented earlier. Just as men’s turnout is more negatively affected by a divorce, it is also more positively affected by marriage, which is in line with the spousal mobilization hypothesis. Thus, before getting married, turnout is usually somewhat lower for men than for women; however, once they marry, men’s turnout converges to that of women’s. One likely reason for this is that having a voting spouse tends to mobilize some individuals who would have otherwise abstained from voting, and because the baseline probability of voting is lower for men, husbands are more likely to become mobilized by their wives than vice versa. However, when a marriage is dissolved, this spousal mobilization effect disappears and the divorce becomes particularly detrimental to the turnout of men.

We have argued that Swedish women tend to possess higher underlying propensities to vote and that they are therefore more likely to act as mobilizers in a married couple. But not all women have a higher baseline probability to vote than their spouses. Therefore, an alternative means of examining the validity of the spousal mobilization mechanism is to study how the divorce effect varies with a proxy for the spouses’ underlying propensities to vote: premarriage turnout. To do this, we first classify the premarriage turnout of husbands and wifes as either high—if they voted in all premarriage elections for which we have data—or low—if they abstained from voting in at least one election when unmarried. We then estimate our baseline model for couples where the premarriage turnout of the two spouses differ. If the gendered divorce effect is, at least partly, driven by the spousal mobilization mechanism the direction of the divorce effect should differ depending on whether it is the husband or the wife who is the most politically active partner.

The main thing to note is that the gender gap in the divorce effect is reversed as we go from the leftmost to the rightmost graph in Figure 7. Thus, in couples where the wife was more likely to vote as unmarried (left graph), the negative effect of divorce is more pronounced for men. However, in couples where it was the husband who had the higher premarriage turnout (the right graph), we instead observe a larger effect of divorce on the wife. It is important to note that the gendered effect of divorce that we have documented exists because the scenario depicted in the left graph is more common.Footnote ¹²

Figure 7.

Divorce Effect by Premarriage Turnout

Note: Complete coefficient estimates together with standard errors clustered on the couple identifier are reported in Table A.3 of the appendix.

Admittedly, these results are a bit more noisy than our previous results, as we lose almost 90% of the data when zooming in on this subset of couples with incongruent premarriage turnout.Footnote ¹³ However, the difference in the gendered response between the two groups is fairly striking, and it serves to further underscore the importance of the spousal mobilization mechanism for understanding the gendered effect of divorces on voter turnout. Judging by these results, being in a marriage will have a larger influence on the turnout of the spouse with the lower baseline probability of voting, who is more often the husband than the wife. However, in contexts where men vote to a higher extent than their spouses, we could expect the opposite pattern to appear, which it does.

We have also performed a number of sensitivity analyses designed to check the robustness of our main findings (reported in the appendix). For instance, we show that we obtain very similar results if we drop the expected turnout control from the specification and instead include different combinations of couple, age, and election year fixed effects in the model (see Figure A.1) or if we estimate a model with individual fixed effects instead of couple fixed effects (Figure A.2). Moreover, none of the findings change if we relax all nonnecessary sample restrictions (Figure A.6) or if we restrict attention to the subset of the sample for which we can observe the turnout behavior of both spouses in all three elections (Figure A.3). We also obtain virtually identical results if we use a fixed effects logit model to estimate the divorce effect (Figure A.4).

The appendix also contains results from some additional analyses of more substantive interest. To further examine the persistence of the divorce effect, we extend the observation window around the election and find that the effect of the divorce extends far beyond the period used for the main analysis. Fifteen years after a divorce women’s turnout is 1.5 percentage points below their predivorce level, whereas the corresponding figure for men is almost 4 percentage points (Figure A.7).

Another set of analyses seeks to establish the generalizability of our main findings and determine the extent to which they are applicable to groups and countries other than those used for our main analyses. In one of these analyses, we compare our sample of divorcees with another group of individuals who got married between 1975 and 2005 but who never divorced. We find that the turnout gap between the two groups is largest during the period when the individuals in the divorcee group are going through their divorces, whereas turnout is more similar in young and old ages (Figure A.9). In our view, this pattern suggests that we would have observed similar drops in turnout among the individuals in the nondivorcee sample had they been subject to a divorce. This interpretation is further supported by the fact that we observe similar marriage effects on turnout for the nondivorcee groups as we do in our main sample (Figure A.10).

Not all couples marry, however, so we have used the available household data to study the effect of separations. A drawback with these data is that unmarried couples who live together without common children are treated as separate households. With this caveat in mind, we observe very similar effects of separations as we do for divorces (Figure A.5). This suggests that our findings should also be applicable to individuals who live together under marriage-like circumstances without being married.

On a slightly different note, we have also examined whether our findings depend on the age of the children at the time of divorce. We find that the main pattern looks similar regardless of children’s age, but the effects are more marked in families with younger children (Figure A.8).

Finally, to assess whether our findings can be generalized to countries outside Sweden, we have performed additional analyses using data from the National Longitudinal Survey of Youth (NLSY) and the European Social Survey (ESS). First, the results from a cross-sectional bivariate analysis of the NLSY data are in line with the pattern displayed in Figure 2 above and suggest that men’s voter turnout is more negatively affected in the wake of a divorce (see Figure A.11). Second, pooling ESS data from different countries and periods, we find clear evidence that the difference in turnout between married and divorced individuals is smaller for women than men and that this difference is of similar magnitude in Sweden as in the ESS as a whole. We also use the fact that both the direction and the magnitude of the gender gap in voting varies between different ESS countries to examine whether the gender difference in the divorce effect changes with the relative political activity of women and men, as suggested by the spousal mobilization hypothesis. We find this to be the case (Table A.5). Although our analyses of the NLSY and the ESS data suffer from many methodological problems that we are able to circumvent when using longitudinal administrative data, these findings suggest that both the empirical results and the theoretical arguments of this study have bearing outside the Swedish context.