Abstract
Political gender differences have moved to the forefront of debates about politics and democracy. Scholars have documented that women are becoming more likely to vote, while men are more supportive of populist/radical right parties and positions, including elite distrust in mainstream politics. Past research has only been partly successful in explaining these differences, with most work concentrating on labour market-related economic explanations or value-based cultural factors. This paper shifts the attention to the family domain. Specifically, it investigates perceived risk of divorce/separation as a source of distrust in political institutions and actors. Three-wave panel data from Norway suggest this effect may not be strongly instantaneous, following short-term changes in risk. Rather, the effect develops over time and mainly among partnered fathers with persistently high levels of perceived divorce risk. This supports a ‘gendered post-divorce realities’ interpretation, under which male risks concerning post-divorce parental involvement matter politically among partnered fathers.
Introduction
Gender differences in political orientations are increasingly debated and researched. In some respects, and places, these differences may have grown over time with generational shifts and with the growing salience of ‘second dimension’ political conflict (also known as the ‘libertarian-authoritarian’, ‘socio-cultural’ or ‘populist’ dimension). Men are consistently more supportive of populist radical right parties and positions (Amengay and Stockemer, 2019; Harteveld et al., 2015), including greater elite distrust and resentment towards mainstream politics. By its own admission, however, extant research has only been partly successful in explaining political gender gaps. Many studies examine labour market economic explanations or ‘cultural backlash’, that is, the impact of values and attitudes. While these explanatory themes are important for understanding populism and distrust in general, there are typically remaining gender gaps also controlling for economic and cultural variables. As Coffé (2018: 201, 209) concludes in an overview of gender and support for the populist/radical right, ‘a significant part of the gender gap remains unexplained in most research […] even when various socioeconomic characteristics and attitudes were controlled for’.
This paper shifts the focus to the family domain, which is more rarely invoked to explain growing gender gaps in general, and male disaffection in particular. Specifically, the paper studies perceived risk of divorce/separation (shorthand: divorce) among partnered individuals as a source of political distrust. I build on recent work using administrative register data on divorce and electoral turnout (Bellettini et al., 2023; Dehdari et al., 2022). On the one hand, this new research confirms the handed-down insight that stable family relationships and childrearing foster political integration. At the same time, it shows more clearly than before the gendered nature of family effects. Childrearing may have especially positive effects (Bellettini et al., 2023), and divorce especially negative effects, on male turnout (Dehdari et al., 2022). Overall, men seem to rely more on family inclusion for political participation and democratic integration.
I use survey panel data, rather than register data, to study the impact of perceived divorce risk on political trust. This allows several contributions to the emerging literature. First, I study how subjective worries about divorce matter among currently partnered people. Subjective perceptions take us closer to how individuals think about their situation and the political consequences of this thinking. A further advantage comes from an observation often made by family sociologists: a separation is not a discrete event that occurs, as it were, at the flick of the switch (Mortelmans, 2020; Zartler, 2021). It is better described as a long and uncertain process where perceived risks and their psychological consequences arise long before any subsequent legal or physical divorce event. By implication, subjective divorce risk perceptions can exist and exercise political impact also in couples that ultimately withstand the turbulence.
A second contribution concerns the dependent variable. Recent research on gendered family effects concentrates on electoral turnout, partly because this is the one political indicator typically offered by register data. By contrast, the survey panel used here measures multiple aspects of trust in political actors and institutions. This is useful as turnout, while a crucial democratic feature, is still a remote indicator of the ‘second’ conflict dimension that now concerns analysts of democracy. Political-institutional distrust, by contrast, lies at the heart of this conflict and is part of the very conceptualization of populism (Müller, 2016; Norris and Inglehart, 2019). However, we know little about how couple instability affect political orientations that are more immediately tied to this conflict. In addressing this knowledge gap, I also contribute to political trust research more generally. Recent work reports that political trust is relatively stable in adulthood (Devine and Valgarðsson, 2024), but can change, at least in the short term, after major life changes, crises and political events (Valgarðsson et al., 2024; Zmerli and van der Meer, 2017). Studying consequences of couple instability contributes to our general understanding of the nature and development of political trust in adulthood.
A first theory section will discuss general reasons why perceived divorce risk may shape political trust. Two subsequent sections present hypotheses about why such effects may be stronger among men. Specifically, a section on ‘gendered spousal socialization’ draws on the notion that men in a turbulent heterosexual union lose communication with, and political influence from, a female partner who is now likely to be the more politically engaged and trusting socializing agent in the relationship. A subsequent section on ‘gendered post-divorce realities’ rather uses the fact that while both women and men are adversely affected by divorce, the precise life consequences are strongly gendered. Women suffer more economically whereas men are often hurt in terms of social exclusion, particularly in terms of parental involvement. This distinction, together with the empirical regularity that women are less politically reactive to economic grievances, generates the prediction that male trust is more strongly affected by divorce risk, but mainly if children are involved.
Having discussed these theoretical variants, the paper considers the context, data, and measures. I use a Norwegian panel survey from 2014 to 2017. Multilevel models of the three-wave data suggest that divorce risk generates distrust among partnered men with children. Comparisons of ‘random effects within-between’ and ‘growth curve models’ suggest this effect may not be strongly instantaneous following short-term changes in risk. Rather, the effect develops over time among partnered fathers with persistently high levels of perceived divorce risk.
Why would perceived divorce risk generate political distrust?
Political behaviour research has become increasingly interested in subjective risks and their political effects. A first wave of interest arose with ‘the great recession’ and concentrated on perceived risk of job loss and poverty (Margalit, 2019; Rehm et al., 2012). More recently, perceived employment threat from technological change has been a prominent theme (Busemeyer and Tober, 2022; Gallego and Kurer, 2022). This research generally suggests that political orientations are not only affected by the actual realization of problematic life events. Subjective worries about those events also have meaningful political effects. That said, we know little about political ramifications of perceived divorce risk.
Why would such worries affect general trust in political actors and institutions? The personal consequences of divorce are inherently complex, contingent and individual. However, the average short- to medium-term ramifications tend to be negative in key respects of relevance for political trust. Examples include multiple dimensions of life satisfaction and well-being (Köppen et al., 2020), as well as health and work disability (Brüggmann, 2020). These personal life aspects are relevant for political trust. Studies suggest that subjective well-being predicts, and even causes, decline in political distrust indicators such as satisfaction with democracy (Esaiasson et al., 2020). Similarly, an overview of the political consequences of health (Gidengil and Wass, 2023) reports a ‘consistent connection’ between poor health and political distrust in Europe as well as the United States.
Such trust ramifications can be understood using key assumptions about electoral accountability and assignment of responsibility in rich western welfare states. In these contexts, where up to one-third of the GDP is spent on social policy and services, citizens see a non-negligible political responsibility for their own well-being and health. Equally important, however, the precise locus of responsibility is unclear. People often lack information for assigning responsibility to specific governments, parties and political-administrative levels. This is partly due to complicated institutional power sharing features producing poor ‘clarity of responsibility’ in western democracies (Anderson, 2007; Powell, 2000; Powell and Whitten, 1993). It is also due to actors engaging in ‘blame avoidance’ strategies that deliberately aim to confuse citizens (Hood, 2011; Weaver, 1986). As a result, citizens who worry about diffuse outcomes such as well-being and health often react with general political distrust and disaffection, rather than precise electoral accountability. We formulate the expectation that:
The discussion above envisages a relatively direct, albeit diffuse, political responsibility attribution. Additionally, there may be more indirect processes driven by social interpersonal trust. Empirically, there is a well-known association between social and political trust, although its strength and causal status remain debated (Inglehart, 1990; Kaase, 1999; Sønderskov and Dinesen, 2016). Theoretically, however, it is possible to imagine that couple instability, especially when involving severe conflict, may damage one's generalized faith in others and ability to trust. This could spill over onto political trust. Moreover, such processes could be gendered as studies suggest gender conditions social trust formation in complex ways (c.f. Qiu et al., 2022). Therefore, the analysis will address the possibility of an indirect impact by examining if divorce risk effects on political trust remain when controlled for social trust.
Gendered spousal socialization?
Recent work using longitudinal register data linking marriage, divorce and turnout imply that the effect described in H1 differs between men and women. Bellettini et al. (2023: 1037) use Italian register data and find that ‘marriage increases men's turnout to the higher pre-marital levels of their wives’. Similarly, using register from Sweden, Dehdari et al. (2022: 1306) report that both men and women become less likely to vote due to divorce but ‘the effect is much more pronounced for men’. Interestingly, and consistent with the notion of separation as a longer process, the effects can be measured already 2–3 years before an actual separation event (and are still statistically significant at least 7 years after).
Several mechanisms could produce gendered political divorce effects. Both Bellettini et al. (2023) and Dehdari et al. (2022) favour ‘spousal mobilization’ as the key explanation. Spouses affect each other politically but the democratically more active and integrated partner will have a larger (positive) impact on the less active one. Of course, men were more likely to vote some decades ago and still are in the oldest generations alive today. However, this has changed in many western democracies, such that turnout is now often higher among women. Wherever this is the case women will tend to be the more influential socializing agent, lifting their spouses up to their level of political engagement and democratic integration.
Consistent with spousal mobilization, Dehdari et al. (2022) find that divorce actually has a stronger negative effect among women whenever the male spouse is more politically active. Again however, this is now the more unusual situation. More often than not, men ‘marry up’ in terms of spousal electoral participation, such that divorce hurts male turnout more in this by now more common situation. Moreover, the spousal mobilization argument seems to account for studies not finding gendered divorce effects (Kern, 2010), or even reporting a stronger impact among women (Voorpostel and Coffé, 2012). On closer inspection, such studies tend to analyse the increasingly unusual western contexts with male overrepresentation or simply no gender differences in turnout (see Dehdari et al., 2022, for a discussion).
The idea of asymmetrical spousal mobilization may be relevant also for psychological orientations and attitudes, such as political trust. To denote such processes, however, we may speak about spousal ‘socialization’ rather than mobilization. Terminology aside, the mechanisms are similar: marital turbulence comes with a decline or breakdown in communication and hence less cross-gender political socialization. Moreover, in many European democracies women tend to be more generally trusting in political institutions and actors, even controlling for multiple socioeconomic and political factors (de Blok et al., 2020). In the Norwegian context studied here these gender differences in trust have arisen over time, even when averaged across age groups (Segaard et al., 2020). The upshot is that men in a turbulent union often lose influence from a more politically trusting partner, one who is typically also more likely to vote. Again, there is no reason to think that these processes commence only once a separation is legally or physically finalized. Rather, the loss of spousal socialization likely starts earlier as communication between spouses deteriorates and perceived risk of divorce increases. We formulate the following hypothesis:
Gendered post-divorce realities?
A second set of gendered mechanisms concern post-divorce life trajectories. Both men and women experience significant negative average life consequences, but their precise nature is gendered. The stylized fact in family sociology is that women are to a greater extent hurt economically by divorce (Andreß et al., 2006; Mortelmans, 2020). This is in large part because women's pre-divorce labour participation rates and incomes are still lower, generating a larger post-divorce drop in adjusted household income.
By contrast, men tend to be hurt more in terms of social exclusion and ‘immaterial domains of life satisfaction’ (Köppen et al., 2020: 239). There are two main mechanisms here. First, men tend to rely more on their partners for emotional support (Brüggmann, 2020) and less on other relationships (Reeves, 2022). Second, post-divorce male social exclusion is driven by reduced contact with children, as these typically continue to reside, and spend more time with, mothers than fathers (NOU, 2024: 8; Wiik, 2022). This can be a severe grievance for fathers in and of itself, but it also has knock-on effects on relationships with other parents, teachers and other adults, in settings and activities where children participate. Overall, these gendered post-divorce trajectories, which are mirrored in subjective measures of post-divorce well-being, have led scholars to speak about ‘gendered realities after separation and divorce’ (Köppen et al., 2020: 235), in which ‘men, and especially fathers, tend to experience larger declines in immaterial domains of life satisfaction than women’.
Theoretically, gendered post-divorce realities could produce equally strong negative trust effects among men and women, albeit through different socioeconomic mechanisms. However, this scenario is not the most plausible one given how economic and immaterial grievances affect men and women politically. Women are affected more economically by divorce, but also seem less reactive to economic grievances in terms of political orientations (Coffé, 2013; Kurer and Van Staalduinen, 2022), subjective well-being (Heyne and Voßemer, 2023), as well as health and mortality (Heggebø, 2022). The standard explanation for these differences in economic reactivity highlights gender and work norms: men still more often self-identify as economic ‘breadwinners’ and are viewed as such by society. Thus, even though female divorcees lose economically this may not translate into stronger effects on political trust among women. Meanwhile, and as discussed above, men are affected by divorce in terms of social exclusion in general, and parental involvement in particular. We can thus formulate the following ‘gendered realities’ hypothesis:
The context
Like in most Western societies, divorce became markedly more common in Norway during the postwar decades and plateaued at a high level in the 1990s and early 2000s. Since then, Norway is one of several European countries where divorces and separations have become somewhat less likely again, compared to the peak years (Kreyenfeld and Trappe, 2020). Based on current trends, Statistics Norway now predicts that about 37% of legal marriages will end in divorce (SSB, 2024).
The Norwegian context conforms broadly to the key assumptions underlying our hypotheses. For instance, this is true for the ‘spousal socialization’ idea underlying H2, that men are now less politically trusting on average (Segaard et al., 2020:, see also empirical analysis section). As for the ‘gendered post-divorce realities’ notion underlying H3, studies of custody and time-sharing arrangements show that children of divorced parents spend considerably more time with mothers on average (NOU, 2024: 8). Relevant data are generally hard to come by but a 2020 survey reported by Wiik (2022) suggests 43% of previously married/cohabiting parents in Norway practice shared residency. In this situation, the child resides legally with both parents also after divorce. Parents tend to live close to each other and practice near-equal time-sharing, such that each parent typically spends on average close to 50% of the days during a normal month with the child. However, this is still not the most common arrangement. Rather, in close to 50% of cases the child legally resides with the mother only after divorce. The father transitions to a new role as ‘non-residential’ parent, spending on average 7–8 days per month with the child (albeit with much variation upwards and downwards). Finally, excluded from these averages is a large category comprising 30% of all non-residential parents (mostly male) who do not see the child during a normal month. Key predictors of this situation are parental conflict and geographical distance (NOU, 2024: 8).
These arrangements structure divorced fathers’ subjective evaluations (Wiik, 2022). Shared-residence fathers are generally satisfied with residential arrangements (78%), whereas satisfaction among non-residential fathers is markedly lower (37%). Similarly, whereas only 17% of shared-residence fathers prefer more time with the child (25% among mothers), such a preference for more involvement is found among 63% of non-residential fathers. Overall, these numbers fit the notion of male ‘non-material’ divorce risks (Kreyenfeld and Trappe, 2020) as these tend to arise among divorced fathers and shape subjective evaluations of custody arrangements. By implication, these matters are a realistic source of worry among fathers who are currently married/partnered but perceive a divorce risk. Judging from the large variation in situations and evaluations described, such worries could range from very considerable fears of dramatic deterioration in child contact to less severe, but still subjectively unsatisfactory, situations where the father must transition to a less involved role in terms of residence and time-sharing.
A longitudinal and comparative perspective is useful as there have been considerable increases in post-divorce male parental involvement. Shared residency was uncommon in Norway (roughly 10%) as late as the early 2000s (Lyngstad et al., 2014). Although trends are similar in many Western societies (Steinbach et al., 2021), the current level (40+%) suggests Norway is among the countries with the most equal post-divorce arrangements. On the one hand, this could mean that Norway is a difficult context for divorce risks to trigger male worries and political trust consequences. On the other hand, significant inequalities remain, with mothers assuming more parental post-divorce responsibility. Additionally, analysts note sweeping changes in societal expectations on male post-divorce involvement during the same period. Higher expectations are likely to increase salience of these matters and amplify the effects studied here such that remaining inequalities in post-divorce involvement are now likely to be experienced more negatively by fathers than the same inequalities would have some decades ago.
Data and measures
The data come from the ‘Support for the affluent welfare state (SuppA) survey’, collected in Norway in 2014, 2015 and 2017, administered by TNS Gallup (Kumlin et al., 2020). The sample was drawn from Gallup's standing web panel with a response rate of 50.1% and is broadly representative of the Norwegian population (aged 18–75). Specifically, the data include 5420 first wave respondents, and 5008 second wave respondents (the latter included 2847 who answered wave 1 and 2161 new recruits). The third wave consisted of respondents who participated in one of the previous waves (N = 1560). The mean number of repeated observations per individual is 1.6.
The central independent variable comes from a question battery with the head question: ‘Please tell me how likely you think it is that during the next 12 months you will …’. One of the response items was ‘become divorced or separated from your partner’. Respondents used a four-point continuum running from ‘not at all likely’ to ‘very likely’. Among men and women living with a partner, 28% chose one of the three alternatives that express some risk. Most (26%) expressed the mildest form of risk where divorce was still ‘not very likely’, whereas only about 2% were more certain of divorce. In the main analyses we shall retain the four-point scale, while remembering that most variation occurs between not perceiving any risk and perceiving some risk, with very few being more certain of divorce (‘likely’ or ‘very likely’).
There are no significant average gender differences in perceived divorce risk among partnered individuals. However, among those living with children both overall levels, as well gender differences, in perceived divorce risk are greater (women 31%; men 39%). Thus, about two in five partnered men living with children perceive some divorce risk.
The dependent variable is a general political-institutional trust index. It taps the average response to four survey items (α = .85) from a question battery with the head question: ‘Various organizations and institutions are listed below. How much trust do you have in each of these?’ A four-point scale was used, ranging from ‘great trust’ to ‘no trust’. The resulting index also varies from 1 to 4 (M = 2.5; SD = .60) but is reversed so that higher values mean greater trust. Don’t know responses were coded as missing for individual items. However, since the index averages over valid responses only those not answering any of the four items are missing. Specifically, the index taps trust in ‘political parties’, ‘parliament’, ‘the government’ and ‘the city council of your municipality’. This index, then, captures trust in a wide range of political institutions and actors at the national and local level.
The analyses also contain control variables. These include age, education, income, general satisfaction with public services, political interest, perceived risk of poverty, perceived risk of unemployment, and perceived risk of poor health (see A1 and A2 for more information about these variables).
Statistical models
The data consist of repeated observations of respondents at up to three timepoints. Such observations are not independent of each other. I therefore estimate multilevel models that take this dependency into account, in that responses at different time points (Level 1) are explicitly modelled as nested within individuals (Level 2). Specifically, this multilevel framework is used to estimate three types of statistical models. Each provides a distinct test of the hypotheses, with different advantages and drawbacks.
Random effects (RE) models estimate if there is an association between the perceived divorce risk, and the trust level, reported by an individual at a given timepoint. Furthermore, the strength of this association can be allowed to vary between men and women by means of cross-level gender (Level 2) × divorce risk (Level 1) interactions. Such interactions, which are well-powered in our random models, can be further compared across groups of theoretical interest (i.e., those with and without children).
Random effects models are useful for understanding broad patterns of association in the data. However, they do not offer very strong causal tests, partly because they rely on both variation ‘between’ and ‘within’ individuals to arrive at one single estimate of, in our case, the effect of currently perceived divorce risk. This assumes that stable variation between individuals in divorce risk, as well as dynamic changes over time within an individual, exercise effects that are so similar that it makes sense to express them in one single effect estimate. This typically tenuous assumption can be addressed more directly using random effects within-between models (REWB; sometimes called ‘hybrid’ models). REWB models have become a standard tool for analysing survey panel data (see Andreß et al., 2013). As shown by Bell and Jones (2015), it is a more flexible, and often preferable, tool compared to popular alternatives like fixed effects or lagged dependent variable models.
For our purposes, a key advantage is that REWB models allow analysis of Level 1 individual dynamic variation (within effects), as well as Level 2 associations across individuals (between effects). In terms of our research problem, we can analyse if persistent Level 2 variation in divorce risk between individuals affect trust in the same way as short-term Level 1 changes within individuals. Both estimates are interesting given our hypotheses, but the latter are generally regarded as a more demanding causal test. This is because estimates based on within-individual change eliminate heterogeneity bias arising from the spurious impact of stable third variable confounders. That said, within-individual estimates are still subject to bias arising from non-stable third variable confounders.
A further limitation of REWB models is that they only capture short-term dynamics. In our context, they reveal how the fact that an individual, in a particular wave, deviates from his/her own average divorce risk affects trust in that wave. While this is a valuable test of causality, divorce risk might exercise long-term effects. We therefore also employ growth curve models (GCM). These allow examination of longer-term trust trajectories over the entire period covered by the panel (in our case three years). More exactly, GCMs assess how a persistent between-individual difference in an independent variable affects within-individual development in the dependent variable. In principle, GCMs can gauge linear as well as non-linear change trajectories following various functional forms (exponential, logarithmic, etc.). However, with three panel waves it is mainly possible to explain variation in linear change trajectories (Singer and Willett, 2003). The empirical question becomes if persistent cross-individual differences in perceived divorce risk generate different individual trust trajectories over time. Beyond their longer-term perspective, GCM models offer a valuable causal test also as they provide a certain temporal separation between independent variable (average divorce risk variation across individuals that is stable across waves) and dependent variable (linear dynamic development in political trust across the waves).
In sum, then, RE, REWB and GCM models offer distinct advantages and drawbacks when it comes to testing our causal hypotheses. That said, one needs to bear in mind that the tests themselves are inherently observational and may be affected by specification decisions. A robustness section will probe the sensitivity of results further.
Empirical analysis
A premise for the ‘spousal socialization’ perspective is that there are gender differences in political trust in the context under study. Table A3 in the appendix contains a cross-sectional regression analysis of participants’ first-time responses, showing how political trust varies with age and gender. Model 1 confirms that males are on average less trusting, controlling for age. As an aside, we also see age differences between less trusting younger respondents and the more trusting older ones. Model 2 adds interaction terms showing how gender differences in trust vary across age groups. Gender differences are the largest among the very young (−.20), but not significantly smaller in other young to middle-aged groups. The exception is the 36–45 group, where trust differences are more than halved compared to the youngest. Still, further analysis (not shown) reveals that the remaining gender gap is statistically significant also in this group. Among older respondents (mid 50s and higher) gender differences in trust are small or absent. Overall, these results support the premise that Norwegian men and women differ in political trust, especially in the earlier life phases when couples are formed and later dissolved.
Table 1 reports three random effects models of political trust where the central independent variable is perceived divorce risk. As discussed, this is a multilevel analysis of repeated panel responses nested in individuals. All three are estimated among respondents currently living with a spouse/partner. Further, each model contains a multiplicative perceived risk × male interaction term. By example, this means that the main perceived risk coefficients show the impact among women, and where the interaction term reveals if the effect is different among men.
Random effects models of political trust.
Standard errors in parentheses.
* p < 0.10, ** p < 0.05, *** p < 0.01.
The first model applies to all observations of currently partnered respondents. It suggests no significant association between female divorce risk and trust, at the same time as this effect is more negative among men (−.04; p < .10). The other male × risk interactions show that this gendered pattern is particular to divorce risk. Thus, the effects of perceived risk of poverty, unemployment, and sickness are all statistically indistinguishable comparing men and women.
The second model uses a more restricted sample where respondents are cohabiting with a partner but do not have children. Here, there is no significant divorce risk effect, neither among women nor among men. The third model involves only observations of currently partnered individuals living with children. There is no significant impact on female trust, but a significantly more negative effect on male trust (−.08; p < .05).
Overall, comparing the models suggests that the more negative impact on male trust that could be discerned already in model 1 is driven by the subset of men living with children. In terms of our hypotheses, the results do not support H1 and H2, suggesting a more widespread negative impact among all cohabiting men and women (H1), or all cohabiting men (H2). Rather, the results support H3. Drawing on the ‘gendered post-divorce realities’ perspective, this hypothesis assumes that men with children are especially prone to fear, and react politically to, perceived divorce risk.
Random effects models (RE) use both between- and within-individual variation to estimate a single estimate. This assumes that these types of variations exercise a very similar impact. To relax this restrictive assumption, Table 2 reports REWB models. They are identical to those in Table 1, but with the difference that perceived divorce risk effects are now captured by two variables, corresponding to between- and within-variation respectively.
Random effects within-between models of political trust.
Standard errors in parentheses. Models contain the same controls and random effects as Table 1.
* p < 0.10, ** p < 0.05, *** p < 0.01.
The first model is again based on all cohabiting individuals. It confirms that there is no significant impact of divorce risk among women, as shown by main effects of between- and within-individual variation. The model also confirms that the between-individual divorce-trust association is significantly more negative among men (b = −.08; p < 0.05). However, there is no within-individual effect of divorce risk in this larger male group.
The second model is estimated among cohabiting individuals without children. Here, there is a larger but still insignificant tendency for between-individual variation to affect trust more negatively among men (interaction b = −.07; p = .13). However, there is no corresponding tendency within individuals. Still, these results illustrate the value of the REWB technique in that an overall insignificant RE association (see Table 1) can mask differences across between- and within associations.
The third model is based on cohabiting individuals living with children. The results reveal why the earlier RE analysis showed a stronger impact among men with kids. This is because both between- and the within-individual interactions are sizeable, and very similar to the random effects estimates. Overall, the result in this group is satisfactory in the sense that the Male × Divorce risk estimates do not vary in size across modeling techniques, or across between- and within coefficients. By contrast, among men without kids there was no RE association in the first place, and only a statistically insignificant between effect, coupled with no within effect. Overall, then, the evidence for a short-term impact of divorce risk emerges as somewhat stronger for men with kids.
Less satisfactory, however, is the fact that the two Male × Divorce risk estimates in this group are only near-significant (p values = .13 and .12). Thus, the conclusion must be cautious. On the one hand, and consistent with H3, there is some evidence of a stronger short-term association between divorce risk and distrust among partnered men with kids. But we cannot be entirely sure, at conventional levels of statistical significance, that this result is driven by both cross-sectional between-individual variation and within-individual change in divorce risk.
Longer-term effects of divorce risk?
The analyses so far assessed short-term effects. Divorce risk – whether measured within or between individuals–was assumed to make an instantaneously detectable imprint on trust. By contrast, the growth curve models (GCMs) that we now turn to examine longer-term trust trajectories over the three-year period covered by the panel waves. More exactly, GCMs model how between-individual differences in selected independent variables affect within-individual trajectories in the dependent variable (Singer and Willett, 2003).
A first question concerns the existence of individual variation in linear change. The ‘empty’ models reported in Table A4 show that this is the case here. These models have two noteworthy components. One is the average trust effect of a linear time variable varying within individuals at level 1 (wave year, coded 2014 = 0; 2015 = 1; 2017 = 3). The estimates show no tendency for trust to up or down on average over time among partnered respondents. The second noteworthy component is a level 2 error term, with an associated standard deviation, showing the amount of between-individual variation in the linear time effect. We see that while there is significant variation in linear time trajectories among partnered men and women, the standard deviation is considerably larger among men. This is a first clue suggesting that perceived divorced risk may not affect trust trajectories much among women as there is not a lot of variation in such trajectories in the first place. By contrast, there is considerably more variation in individual trust trajectories among partnered men. The question we now ask is: can perceived divorce risk explain this variation among partnered men, such that greater risk generates more negative trust trajectories? This question is addressed through a series of growth curve models reported in Table 3.
Growth curve models (GCMs) of political trust.
Standard errors in parentheses. The models contain the same controls as Table 1.
* p < 0.10, ** p < 0.05, *** p < 0.01.
Importantly, the models contain a multiplicative interaction term between mean divorce risk and year of the wave. Consistent with H3, this interaction term is significant only among partnered men with kids. This suggests that their trust trajectories become more negative than they otherwise would have been as an individual's mean perceived divorce risk increases (b = −.07; p < .01). Inconsistent with H1 and H2, this growth curve effect is absent in the other groups.
The models contain the same control variables as previous RE and REWB models. Additionally, they control for a rival growth tendency, whereby high education makes overtime trust trajectories more positive. This control was selected based on analyses showing that only two of the previously controlled factors (education and perceived risk of unemployment) interacted with the year effect. Further analysis shows that the growth curve effects of these correlated variables are in fact driven by education. Unemployment risk was therefore dropped from the final analysis, but the results are robust to its inclusion (see Table A5).
Figure 1 uses the estimates for partnered men with kids. It graphs the impact of another year on trust at different levels of perceived divorce risk. For comparison, this is done separately for education levels. This comparison illustrates that high education has a similarly sized impact as that of moving one step along the divorce risk scale.
Effect of another year on political trust (among partnered men with children).
As discussed earlier, no divorce risk (1) is the most common perception (about 60% of partnered men with kids). At this level, the predicted impact of time is positive among the higher educated (b = .04; p = .054) and zero among those without. Said differently, men in stable relationships with kids and higher education develop greater political trust with the passing of each year. Men in stable relationships with kids, but with low education, display stable political trust. Look now at the mildest level of perceived divorce risk (2). This is a large category containing almost 40% of partnered men with kids. Here, the previously positive time effect among the higher educated disappears. Meanwhile, those with lower education are now negatively affected by the passing of another year (b = −.08; p = .001).
It is also interesting to look at even higher levels of perceived divorce risk (3 and 4). Here, the passing of another year hurts political trust quite strongly, also among the higher educated. The 1-year effect predictions in these groups range from −.09 to .−22. These are considerable 1-year changes as the overall standard deviation in political trust is .60. Thus, the predicted 3-year changes in these groups range from roughly one half to a full standard deviation. We must remember, however, that these more severe divorce perceptions are only held by about 3% of respondents. In conclusion, therefore, the principal effect of divorce risk is to offset an otherwise positive trust development among higher educated fathers, and to generate a negative development among lower educated fathers.
Extensions and robustness
Additional analyses were performed to test robustness and improve our understanding of the findings. First, all results were controlled for social (interpersonal) trust, using a classic dichotomy (‘yes, people can generally be trusted’ or ‘no, you can’t be too careful’). They confirm that Norway is a high-trust context with 74% of first-time respondents expressing trust (with 5% answering don’t know). Moreover, previous results are entirely robust to controlling for this social trust dichotomy, with relevant divorce risk estimates remaining the same or even growing slightly (Tables A6–8). This suggests that the previously reported results represent a direct impact of divorce risk on political trust, one that is not an indirect function of divorce risk affecting generalized trust, with knock-on implications for political trust.
Second, I analysed the sensitivity of the significant results among men with kids to dropping controls. Specifically, this involved either dropping all background features tapped by income, education, political interest and public service experiences or dropping all the non-divorce related risk perceptions. The RE and REWB estimates showed very small fluctuations even after such large specification shifts, with estimates varying by less than .01 and remaining either significant or insignificant as before at p < .10. Similarly, the GCM estimates were entirely robust to dropping these sets of controls, as well as to dropping the growth effect of education (Table A9).
Third, several alternative dependent variables were analysed. One was a political interest/awareness index that combined a political interest and a political news consumption item (1–4; M = 2.8; SD = .66; α = .82). This is a relevant extension as political interest (and conceptually adjacent variables) are often at lower levels among women (Fraile and Gomez, 2017; Fraile and Sánchez-Vítores, 2020; Grasso and Smith, 2022; Pfanzelt and Spies, 2019). Spousal socialization assumptions could therefore be taken to imply that divorce risk effects are more negative among women for these outcomes. But while the data confirm that males score higher on interest/awareness (b = .23; p < .001) there is no indication that divorce effects are more negative among females. If anything, RE models suggest it is male interest/awareness that responds negatively to divorce risk, although REWB models showed this to be driven by between-individual variation (Table A10). Additional GCM analysis revealed no significant long-term impact of persistent divorce risk on interest/awareness in any male or female group. Overall, effects of divorce risk on interest/awareness are generally weak, occasionally stronger among men, but not consistently gendered one way or the other.
Fourth, I analysed effects on a less inclusive political trust index that dropped local trust and retained the three national/general items (parliament, government, parties). The direction and size of estimates were similar suggesting that both local and national items contribute to effects in the same direction. An interesting nuance can be noted, however. The short-term estimates in RE and REWB models became slightly weaker using the narrower trust index (typically dropping from −.08 to −.06, while remaining significant or insignificant like before at p < .10. By contrast, the longer-term processes captured by GCM estimates were almost exactly the same with a narrower index (Table A11). These nuances underline our overall conclusion that it is the longer-term effects of perceived divorce risk among men with kids that are the most significant and robust.
Conclusions and future research
Research on ‘second-dimension’ political conflict, including populist support, gravitates towards labour market ‘economic’, and value-based ‘cultural’, explanations. This is true also for studies of gender gaps along this conflict dimension (Coffé, 2018). By contrast, this paper emphasizes the family as a site for gendered formation of political orientations. The specific research question was how perceived divorce risk shapes trust in political actors and institutions. The findings suggest such an impact, but there is no support for H1 suggesting a broad-based effect among partnered men and women alike. Neither did the data show much support for a general impact among all partnered men (H2). Instead, results are more in line with H3, predicting an effect among men with children. In this group, ‘within-between’ models showed substantively meaningful, but not quite statistically significant, evidence of an instantaneous impact of short-term changes in perceived divorce risk. By contrast, ‘growth curve models’ showed strong and robust evidence that the effect develops over time among fathers with persistently high perceived divorce risk. Still, the more modest within-individual evidence offers useful clues. The statistically weaker, but substantively meaningful, short-term associations among men with kids are suggestive of small and inconsistent instantaneous reactions that accumulate over time into a more robust long-term link between persistent divorce risk and longer-term political trust development.
These survey findings join recent work on register data (Bellettini et al., 2023; Dehdari et al., 2022) showing more clearly than before that political-democratic integration is especially dependent on family life among men. At the same time, the results make novel contributions to this research. Starting with the independent variable, the analysis of risk perceptions shows that political effects of separation can be broader than implied by measures of divorce as an actual legal/physical transition. Behind such ‘events’ may lie a larger pool of perceived risks, many of which are never realized, but nonetheless matter politically also among fathers who remain partnered. As for the dependent variable, our findings contribute by showing that political effects of divorce are not limited to electoral participation (the one political variable typically available in register data). Instead, divorce risk effects can extend to a key psychological orientation – political trust – that lies at the heart of the second-dimension conflict currently dominating political analysis and debate. Interestingly, this reaction does not seem mediated by social trust. That is, the explanation for the political trust development among men with kids worrying about divorce is not that they grow distrusting in everything and everyone, with a spillover on politics. Rather, results fit better with the notion of a direct political, albeit diffuse, reaction in terms of trust in democratic actors and institutions, a reaction that is independent of generalized social trust.
These findings contribute in several ways to contemporary research on political orientations. Importantly, they illustrate the relevance of the family domain as an origin of gender gaps in second-dimension orientations such as political distrust. More specifically, it is interesting to note that also male disaffection can be significantly driven by grievances in the family domain. On a more general level, the findings inform political trust research broadly, a field currently debating the extent, durability and origins of trust change in adulthood (Devine and Valgarðsson, 2024; Valgarðsson et al., 2024). The findings reported here contribute to the conversation by finding that gendered reactions to couple instability help shape gender gaps in political trust developments over a 3-year period.
A theoretical contribution arises from the distinction between two frameworks for thinking about divorce effects. Both predict a stronger impact among men, but the mechanisms and scope conditions differ. The idea of ‘gendered spousal socialization’ is emphasized in recent work on divorce and turnout. It implies that men in a turbulent relationship lose communication with a female partner now likely to be more politically engaged and trusting across several key dimensions. Spousal socialization implies a generally stronger effect among men as these are on average less trusting than women in our context. By contrast, the ‘gendered post-divorce realities’ framework highlights that negative life consequences of divorce are gendered. Women are hurt more economically but are also less reactive to economic grievances. Men suffer more in terms of social contacts, in particularly with children who go on to spend more time with the mother. This implies that male trust is hurt more strongly by divorce risk, but mainly if children are involved (H3). By contrast, ‘gendered spousal socialization’ implies negative effects among men generally (H2), and does not present clues as to why effects would occur only among men with kids.
The findings come with several caveats and related suggestions for future research. To begin with, these non-experimental observational findings naturally raise questions about control variables. All statistical models contained an arguably ambitious list of controls, including risk assessments beyond divorce. Robustness checks suggested that results were insensitive to dropping control variables. Still, there may be relevant controls not available in the data. For example, personality factors like neuroticism could conceivably affect both divorce perceptions and (changes in) trust orientations.
Moreover, I relied on three types of longitudinal statistical models, each with distinct advantages and drawbacks. Comparisons across these models showed that while there are some signs that short-term changes in divorce risk may matter, the longer-term effect of persistent risk is clearly stronger and more robust. This conclusion should be tested in future research using other designs. Longer-term panels, with more waves, will be useful to gauge the durability of perceived divorce risk and its political impact. Such data would also ease the application of newer data-demanding methods for analysing reciprocal causation across multiple waves (Zyphur et al., 2020). Reciprocal causation is possible to the extent that a distrusting outlook on politics can generate scepticism about one's own marriage, net of other perceived risks. This possibility is not directly addressed in the models used here, although as noted GCM models build on a certain temporal separation between a stable independent variable and dynamic development over time in the dependent variable.
A final avenue for future research concerns cognitive mechanisms. What are the precise thoughts and fears driving male distrust in unstable families with children? Are fathers mainly worried about their own post-divorce well-being or (also) that of their kids? Is there a policy component to these reactions, perhaps an alleged failure of politicians and public services to secure equal parental involvement? Our theory and results beg such questions without being able to answer them at this stage with these data. Relatedly, scholars could consider scope conditions. At an individual level, the thought processes implied by the results could be more salient among those with small children, as divorce consequences are in some ways larger and more lasting. At a contextual level, one wonders if the Norwegian context is an easy, difficult, or perfectly normal environment for cognitive connections between divorce risk and political orientations. As discussed, Norway has comparatively gender equal post-divorce arrangements, suggesting that risks and worries among partnered men with kids are mild. However, scholars agree that also societal expectations on male care work and involvement have grown over time and are comparatively high. Such expectations may fuel effects of remaining inequalities.
In closing, it is worth noting that this paper analysed political gender differences mainly in terms of ‘moderation’, whereby gender gaps arise as one gender (men) react differently to an underlying factor (perceived divorce risk). However, as pointed out by several scholars (Coffé, 2018; Harteveld et al., 2015) gender gaps can also arise through ‘mediation’, meaning that also levels of an underlying factor differ between men and women. Unlike most past research on political gender gaps, we have focused on moderation. However, the descriptive analysis suggests a role also for mediation: not only is political trust especially affected by perceived divorce risk among men with children. It is also in this group that we find the highest perceived divorce risk in the first place.
Supplemental Material
sj-docx-1-asj-10.1177_00016993261429680 - Supplemental material for Divorce risk and political distrust: Gendered consequences of couple instability
Supplemental material, sj-docx-1-asj-10.1177_00016993261429680 for Divorce risk and political distrust: Gendered consequences of couple instability by Staffan Kumlin in Acta Sociologica
Footnotes
Acknowledgements
I thank the participants in the Weltrust group at the Department of Political Science, University of Oslo for fruitful discussions.
Ethical considerations and informed consent
The study was approved by Norwegian Social Science Data Service (on August 21, 2014, Project 39089). Written consent concerning participation and recontacting was obtained.
Funding
The author disclosed receipt of the following financial support for the research, authorship and/or publication of this article: This research was funded by the Norwegian Research Council (Grant No. 358243 and 301443).
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.
Data availability statement
Anonymized data are archived with SIKT: Norwegian Agency for Shared Services in Education and Research. https://doi.org/10.18712/NSD-NSD2458-V6
Supplemental material
Supplemental material for this article is available online.
Author biography
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
