The changing educational and social class gradients in union dissolution: Evidence from a latecomer of the Second Demographic Transition

Most studies on the changing socioeconomic gradient of divorce have operationalized individuals ’ socioeconomic status (SES) through education, often neglecting social class differences. Education may proxy cultural and cognitive skills, whereas social class could more accurately capture economic means. Additionally, existing research has predominantly focused on women and marital dissolutions. This study addresses these oversights by analyzing the educational and social class gradients of both marriage and cohabitation dissolutions among men and women in Italy — a latecomer to the Second Demographic Transition. We used non-proportional hazard models to estimate survival curves and union dissolution probabilities stratified by education, social class, and cohort. Our findings reveal a vanishing socioeconomic gradient of marital dissolution among women and a reversal from positive to negative among men across cohorts. These results challenge the conventional view that men ’ s higher SES always stabilizes unions and support Goode ’ s hypothesis on the reversal of the socioeconomic gradient of divorce for both genders. No clear SES gradient was found for cohabiting unions. Overall, the study demonstrates the significant predictive power of social class for marital dissolutions, even when controlling for education, emphasizing the need to consider both measures of SES to comprehensively account for different underlying mechanisms.


Introduction
Over the last century, union dissolution has become an increasingly common phenomenon in Western societies, with relevant social and economic consequences for individuals.Union dissolution can be a disruptive event, and separated men and women face higher risks of economic deprivation compared to their partnered counterparts, especially if they have children (Andreß et al., 2006;Leopold & Kalmijn, 2016).Accordingly, the socioeconomic gradient of union dissolution, or the differential dissolution risk between different socioeconomic groups, is increasingly attracting scholarly attention.
The most prominent theoretical perspective on the socioeconomic gradient of union dissolution was proposed in the seminal work of William Goode (1962Goode ( , 1993)).According to his view, early adopters of divorce correspond to the "social vanguard."When the legal, social, and economic barriers to divorce are high, only high socioeconomic status (SES) partners have the necessary cultural and economic resources to break such barriers.As these barriers diminish, divorce becomes more accessible to less privileged couples, thereby spreading through the population.When divorce becomes commonplace, the unions of high-SES partners may result in being more stable than those of lower SES partners.Individuals from higher social strata tend to form more successful matchings and are less exposed to stressful life events (e.g., unemployment, health issues) that affect relationship quality (Lyngstad & Jalovaara, 2010).Moreover, upper-class partners share more financial assets and long-term investments (e.g., home ownership), which raises the financial costs of divorce (Boertien & Härkönen, 2018).Goode's narrative has been widely supported by within-and between-country empirical evidence, revealing a generalized reversal-from positive to negative-of the educational gradient of divorce over time, with differences based on institutional and cultural contexts (Harkonen & Dronkers, 2006;Matysiak et al., 2014).Nonetheless, the literature on the changing socioeconomic gradient of divorce contains several shortcomings.
First, while Goode's original thesis referred to "class differentials" in divorce, virtually all studies have operationalized individuals' SES through education.The importance of social class for demographic behaviors, as a well-defined concept distinct from education or income, has been increasingly recognized in the literature (Baizan, 2020;Kreyenfeld et al., 2023).Partners' education and social class may both influence union dissolutions through different underlying mechanisms.Education should more accurately proxy partners' cultural resources to overcome the legal and social barriers to divorce or the cognitive skills required to form more stable unions.Instead, social class may more precisely capture individuals' economic means to cope with the direct and indirect costs of union dissolution.Moreover, social class is a more direct indicator of economic hardship, which may increase the risk of union dissolution, or of the partners' financial assets and long-term investments, which may deter them from dissolving the union.We thus ask: Does social class influence union dissolution over and above education?Does the effect of social class, net of education, change across cohorts?
Second, most studies have only included married couples in the analysis, despite the rising popularity of unmarried cohabitation as a living arrangement, conceived both as a pathway or alternative to marriage (Manning, 2020;Perelli-Harris & Sánchez Gassen, 2012).Marriage and cohabitation differ in terms of partners' socioeconomic characteristics and union stability, with the latter often considered a more flexible and easier-to-terminate living arrangement (Perelli-Harris & Sánchez Gassen, 2012).Married and cohabiting couples may thus encounter different types of barriers to union dissolution (Cherlin, 2017).We ask: Is there a socioeconomic (educational and social class) gradient in the dissolution of cohabitation?Does it change across cohorts?
Third, the majority of existing research has focused on women only.Goode (1962Goode ( , 1993)), however, generally referred to the couple's social class, often using the husband's social class as a proxy because, at the time, married women's employment was limited.If we consider the change in the socioeconomic gradient of divorce partly the result of a cultural diffusion process, it should operate in the same fashion for men and women.Nevertheless, it is well-known that SES can influence partnership choices differently for women and men, potentially affecting their union stability in distinct ways (Sayer et al., 2011).We thus ask: Does the socioeconomic gradient in the dissolution of marriage and cohabitation differ between women and men?Does the pattern change across cohorts?
We answer these research questions by analyzing the educational and social class gradients in the dissolution of marriage and cohabitation for both men and women.Studies in this field have generally focused on countries with relatively high separation rates.We complement previous work with analyses for Italy, a country that-despite being all-to-often caricatured by the international literature as a "traditional" country in terms of family demographics-over the last three decades has witnessed a strong increase in total divorce and separation rates, and a rapid diffusion of cohabitations (Aassve, Mencarini, Pirani, & Vignoli, 2024).
The present paper moves beyond existing research in three ways: (1) We conceptualize and test the potential independent role of the educational and social class gradients in union dissolution; (2) we explore whether the (changing) socioeconomic gradient in union dissolution differs between marriage and cohabitation; and (3) between men and women.While previous research has hitherto neglected latecomers of the SDT, we offer these three contributions with a focus on the Italian case, hence also elucidating on how the socioeconomic gradient of union dissolution develops across cohorts in a country with a postponed, yet somewhat-accelerated SDT.
We combined the two latest and largest statistically representative family surveys conducted by the Italian Institute of Statistics (Istat) in 2009 and 2016.From a methodological perspective, we adopted an event-history analysis approach relying on stratified Cox models.With this analytic strategy, we sought to relax the proportionality assumption and were able to estimate survival curves and union dissolution probabilities stratified by education and cohort, and social class and cohort (separately for married and cohabiting women and men) while also adjusting our estimates for other covariates.
Two considerations need to be addressed upfront.First, individuals may self-select into marriage or cohabitation based on their education and social class, among other characteristics.However, our goal is not to establish causation but to illustrate the evolving composition of those who separate, and selection is part of the story (Sigle-Rushton et al., 2014).Second, we were compelled to analyze women and men separately due to the lack of collected information on former partners in the retrospective data available, thus impeding a comprehensive understanding of within-couple dynamics.Nonetheless, we were able to illuminate gender disparities and generational shifts, which are essential for understanding patterns of social stratification in union dissolution.

The changing socioeconomic gradient of divorce
Whether relationship dissatisfaction converts into an actual separation depends partly on the monetary and social costs of separation.Generally speaking, new social behaviors and trends first emerge in specific social groups-who are defined as "trendsetters," or "prior adopters" (Rogers, 1962)-and only later gradually spread to others.According to the influential work of Goode (1962Goode ( , 1993)), prior adopters of divorce correspond to high-SES couples with the cultural and economic means to afford such a separation.In contexts where divorce is uncommon, its economic and social costs are high: Divorce is considered a severe breach of social norms and is thus strongly stigmatized; it is expensive and time-consuming in terms of legal proceedings; and it has important economic consequences.High-SES individuals, due to their greater levels of autonomy, a higher degree of rejection of traditional institutions and religion, and, more broadly, embracement of "post-modern" values, are generally more tolerant of divorce than the lower social strata (Lesthaeghe, 2014(Lesthaeghe, , 2020)).Moreover, high-SES couples tend to be better prepared to weather the economic costs of divorce, e.g., legal expenses, the costs of moving into a new home, and bearing living expenses alone.Thus, in low-divorce contexts, the socioeconomic gradient of union dissolution is usually positive.
Only later, as divorce spreads, it gradually becomes affordable to less privileged social groups.As separation becomes more common, it begins to be seen as an eventuality of the life course, free from stigma, and its economic cost decreases.Goode (1963) predicted that, in this second phase, the class gradient of divorce will be negative due to several factors that imply a lower propensity to dissolve the union for the upper than for the lower social strata.First, a higher SES may be an indicator of marital attraction (Boertien & Härkönen, 2018) and provide non-economic benefits that enhance the quality of the marriage.It may also correspond to more advanced cognitive and communication skills, and problem-solving ability (Becker et al., 1977;Conger et al., 2010;Holley et al., 2006).In addition, high-SES couples generally show a higher level of gender egalitarianism in the domestic sphere, which should lead to greater relationship satisfaction and stability (Cooke, 2006;Hochschild, 1989;Oláh & Gahler, 2014).Finally, high-SES couples have more financial assets and goods than their low-SES counterparts.Since separation leads to a decrease in net worth, high-SES couples encounter more significant economic barriers to divorce due to having more to lose (Boertien & Härkönen, 2018).By contrast, individuals with lower SES are more likely to be exposed to stressful life events and behavioral issues, such as unemployment, health problems, alcohol or drug abuse, and economic hardship ( de Graaf & Kalmijn, 2006;Whelan, 1994), which may decrease relationship satisfaction and increase their risk of union dissolution (Conger et al., 1990;Howe et al., 2004).
E. Bastianelli et al. 2.2.The socioeconomic gradient of divorce: education or social class?Despite Goode's work originally referring to "class position" and "class differentials" (Goode, 1962), virtually all empirical research has operationalized individuals' SES through education alone (e.g.Chen, 2012;Cheng, 2016;Harkonen & Dronkers, 2006;Kalmijn & Leopold, 2021;Musick & Michelmore, 2018).Education is a more accurate indicator of the cultural resources (e.g., rejection of traditional institutions and religion, post-modern values, gender equality) necessary to overcome the social barriers to divorce (Lesthaeghe, 2014), and of the cognitive skills that lead to more stable unions (e.g., communication and problem-solving skills) (Conger et al., 2010).While some of the mechanisms underlying his hypothesis of a reversal, from positive to negative, of the socioeconomic gradient of divorce are thus directly linked to education, others are more strictly economic, and may be more precisely captured by social class.
In contemporary capitalist societies, social class (i.e., the position in the occupational division of labor (Weber, 1978)), shapes individual life chances, behavioral patterns, and inequalities (Breen & Rottman, 1995;Chan & Goldthorpe, 2007).Our definition of social class is derived from the well-known and recognized European Socio-economic Classification (ESeC), an evolution of the Erikson-Goldthorpe-Portocarero (EGP) class schema (Erikson et al., 1979;Erikson & Goldthorpe, 1992), which refers to categories of individuals sharing similar positions in term of employment relations and occupations (Rose & Harrison, 2007).Within this classification, the salariat class includes large employers, professionals, high administrative and managerial occupations, higher grade technicians, and supervisory occupations.The middle class is constituted of such intermediate occupations as higher grade white collar ("non-manual") workers, small employers and self-employed in nonprofessional occupations, and lower supervisory or technician occupations.Finally, the routine class refers to lower grade white collar workers, lower technical occupations, and semi-and unskilled workers (Harrison & Rose, 2006;Rose & Harrison, 2007).Social class is not interchangeable with education or other individual attributes, such as employment status or income.Rather, it emphasizes individuals' positions in society and is strictly linked to different risks of job loss and being trapped in temporary employment, as well as to different levels and continuity of earnings (Baizan, 2020;Breen & Rottman, 1995;Chan & Goldthorpe, 2007).Job loss, unstable employment, and low permanent income can generate stress within couples and are associated with higher risks of union dissolution (Bastianelli & Vignoli, 2022;Kalmijn et al., 2007;Ono, 1998).
As predicted by Goode, many countries have documented a clear weakening in the positive educational gradient of divorce over time, which has gradually become more common in the least educated fraction of the population (Chen, 2012;Cheng, 2016;Harkonen & Dronkers, 2006;Kalmijn & Leopold, 2021;Matysiak et al., 2014;Musick & Michelmore, 2018;Perelli-Harris & Lyons-Amos, 2016).In a comparative study of 17 countries, Harkonen and Dronkers (2006) found that the de-institutionalization of marriage and the spread of unconventional family practices were linked to an increasingly negative educational gradient of divorce.This finding aligns with Goode's hypothesis that as marriage becomes less institutionalized, higher education levels are associated with lower divorce risks.Interestingly, Dronkers (2002) also revealed a cohort-specific trend in the relationship between intelligence and divorce in the Netherlands.For those born in 1940, when divorce laws were strict, and divorce was an "elite phenomenon", higher intelligence correlated with increased divorce risk.Conversely, among individuals born in 1958, higher intelligence was associated with lower divorce risk as divorce laws became more lenient.Furthermore, in a meta-analysis of European research findings, Matysiak et al. (2014) found that increases in divorce rates and in women's participation in the labor force were the main factors driving the reversal of the educational gradient.This confirms the notion that the change in the educational gradient can be linked to a decrease in both the economic and social costs of divorce.
To the best of our knowledge, only a handful of studies have considered social class differentials in union dissolution (Gibson, 1974;Haskey, 1984;Kalmijn et al., 2011).Existing research has predominantly employed historical administrative data including information on divorce and occupation (Gibson, 1974;Haskey, 1984;Kalmijn et al., 2011).Gibson (1974) and Haskey (1984) analyzed the case of England and Wales-the former using data from divorce petitions filed in 1961 and the latter using 1979 census data-and found mixed evidence.Kalmijn et al. (2011), using historical data stemming from marriage records, found that occupational class was positively associated with divorce in 19th-century Netherlands, in line with theories identifying upper-class individuals as trendsetters of new social behaviors.Less is known, however, on how these trends have evolved in more recent times.

The socioeconomic gradient in the dissolution of cohabitation
Most theoretical reflections and empirical studies described in the previous sections have only considered the dissolution of marriages.Nonetheless, in Western societies, from the 1970s onwards, non-marital cohabitation has become an increasingly popular living arrangement.Married and cohabiting couples share fundamentally similar features.Members of both types of union share a household, usually resulting in economies of scale, and present themselves socially as a couple (Smock, 2000).It follows that many of the implications of a couple's breakup are virtually identical regardless of union type, and the union dissolution has been found to affect cohabiting partners' economic well-being, emotional health, and parental responsibilities (Avellar & Smock, 2005;Manning, 2020;Tavares & Aassve, 2013).It is thus important to uncover the socioeconomic gradient of the dissolution of cohabiting unions, which has been neglected in the literature, with few exceptions (e.g.Jalovaara, 2013;Mäenpää & Jalovaara, 2014), often due to data availability.
Despite the similarities in the implications of union dissolution for married and cohabiting couples, the two types of unions exhibit notable differences.In contexts where cohabitation is less common, individuals may choose cohabitation over marriage to avoid cultural expectations concerning marriage.Cohabiting couples often self-select as individuals seeking to challenge traditional views of marriage, including its gendered division of housework and childcare (Manning, 2020;Perelli-Harris et al., 2014).Accordingly, they typically represent the most educated and the "social vanguard" of a society.Especially among older Italian cohorts, marriage and cohabitation considerably differed in terms of socioeconomic composition and union stability, with cohabitation showing both higher SES and dissolution rates (Guetto et al., 2016;Rosina & Fraboni, 2004).
The theoretical discourse around similarities and differences in the dissolution of marriage and cohabitation is rooted in the fact that, across many wealthy countries, marriage and cohabitation continue to have distinct meanings.Marriage predominantly signifies a stronger level of commitment than cohabitation (Perelli-Harris et al., 2014;Perelli-Harris & Sánchez Gassen, 2012).As such, cohabitors have been shown to experience far higher rates of dissolution, even if the partners have children in common (Kelly Raley & Wildsmith, 2004;Musick & Michelmore, 2018).Dissolving a cohabiting union is easier and less costly than divorcing, as it does not require legal procedures and usually involves fewer long-term economic investments.While it may be more accessible for the lower social strata, it also implies fewer deterrents (such economic barriers as home ownership) for the higher social strata.Thus, education and social class may be less significant factors for the dissolution of cohabitants compared to spouses.
As a matter of fact, despite a trend toward a negative educational gradient of divorce having been found in many countries, the educational gradient in the dissolution of cohabiting unions is more varied (Cherlin, 2017).A negative educational gradient in the dissolution of E. Bastianelli et al. cohabiting unions was found in two studies on Finland (Jalovaara, 2013;Mäenpää & Jalovaara, 2014), and a recent study by Kalmijn and Leopold (2021) including eight European countries.However, they found that the negative gradient was stronger in married than in cohabiting unions (Kalmijn & Leopold, 2021).
Ultimately, evidence on the socioeconomic gradient of union dissolution for cohabitors is limited.Given the rising importance of cohabitation in contemporary family life courses, this paper seeks to address this oversight.

A gender perspective
Most empirical studies on the changing socioeconomic gradient in union dissolution-often focusing on the educational gradient in divorce-have exclusively examined women.This focus likely stems from the common belief in the literature, aligned with Becker's theory of the family (Becker, 1991), that men's higher SES should always stabilize marriage.However, Goode's (1962) theory does not explicitly refer to women.Instead, Goode primarily considered the husband's social class as a proxy for the couple's SES, reflecting the limited employment of married women at the time.His theoretical reflections, especially those concerning sociocultural barriers to divorce, generally addressed the couple as a unit of analysis and thus should apply to both men and women.Therefore, it is reasonable to expect a positive socioeconomic gradient in union dissolution for both genders when societal barriers to divorce are high, challenging the conventional view on the role of men's SES.
However, individuals' SES may have a different relevance for men and women's partnership patterns (Bastianelli & Vignoli, 2022;Killewald, 2016;Sayer et al., 2011).Goode's theory predicts that when divorce is commonplace, low SES should be linked to more unstable unions.Nevertheless, while this may clearly be the case for men, whether a negative association between SES and union dissolution emerges also for women may partly depend on society's division of gender roles (Gonalons-Pons & Gangl, 2021;Killewald, 2016).In contexts with a prevalent male-breadwinner family model, where women only have (and are only expected to have) marginal or complementary roles in the labor market, low-SES women are generally more likely to be financially dependent on the partner and thus less likely to separate (Killewald, 2016;Sayer et al., 2011;Vignoli et al., 2018).Becker's theory posits that as education enhances women's prospects in the labour market and their economic autonomy, it diminishes the economic benefits of marriage.Thus, women with higher levels of education or higher social class should be more inclined to divorce (Becker et al., 1977).As the prevalence of the dual-earners family model increases, expectations toward women's employment change and women's contribution to family income becomes more substantial and valued.Under these different circumstances, low-SES women may become comparably less desirable partners than high-SES women and form less stable unions as predicted by Goode.Thus, the reversal from positive to negative of the socioeconomic gradient in union dissolution for women is only likely to occur in contexts where women's employment is both established and widespread.Indeed, the diffusion of women's employment has been found to be negatively associated with changes in women's educational gradient in divorce (Matysiak et al., 2014).
In conclusion, a proper empirical investigation of the socioeconomic gradient in union dissolution should ideally test the role of both education and social class, and adopt a gender perspective, so as to disentangle potentially different (and gendered) cultural and economic mechanisms.

Italy, a latecomer of the SDT
Until the last couple of decades, Italian marital stability seemed to be an exception in the European landscape.Divorce was only introduced in 1970, and divorce rates have always been low compared to most Western societies (Sobotka & Toulemon, Chap. 4, 2008).Moreover, unlike in Northern and Western European countries, marriage in Italy has consistently maintained its centrality in family formation (Rosina & Fraboni, 2004).Divergences with other Western countries have been attributed to the lower level of secularization and the strong role of the Catholic Church, as well as to strong parent-child ties and the importance of parental approval rooted in Italian society (Reher, 1998;Rosina & Fraboni, 2004;Vignoli & Salvini, 2014).
Nevertheless, separation and divorce rates in Italy have considerably grown in the last three decades, while marriage rates have decreased to such an extent that, in 2019, the number of divorces reached almost 50 % of the number of marriages celebrated in the same year (Guarneri et al., 2021).Cohabitations almost quadrupled between 2000 and 2020, when roughly 16 % of all partnered individuals aged 25-54 were in a cohabiting union (Tomassini & Vignoli, 2023), particularly in the country's northern and central regions and in urban areas (Castiglioni & Dalla Zuanna, 2009).Non-marital cohabitations are increasingly accepted even for childbearing, to the extent that, in 2020, 35.8 % of children were born from unmarried parents (Istat, 2022).
These new family patterns began to spread across more secularized individuals, those with the highest socioeconomic profiles, and predominantly among those living in the north of the country (Guetto, Mancosu, Scherer, & Torricelli, 2016;Pirani & Vignoli, 2016).Research on the socioeconomic gradient of union dissolution in Italy is, however, limited and dates back to a couple of decades.Regarding social class, the empirical evidence is virtually non-existent.To the best of our knowledge, the only trace of a reversal, from positive to negative, in the educational gradient of divorce in Italy was found by Salvini and Vignoli (2011).Although they found an overall positive educational gradient in marital dissolution for both women and men, highly educated women were characterized by a strong increase in separation risk during the early 1990s, followed by a stabilization and decline in the early 2000s, thus revealing a potential emergence of a negative gradient in union dissolution.However, empirical evidence for younger Italian cohorts is lacking.

Research hypotheses
The present paper offers fresh empirical evidence on the (changing) educational and social class gradients in the dissolution of marriages and cohabitations for men and women.In so doing, it addresses several research gaps in the literature by focusing on a rather unexplored case study, Italy, where the diffusion of union dissolution was delayed relative to many European countries but dramatically accelerated in the last three decades.In the following, in line with the presented theoretical arguments, we test four analytical research hypotheses.
Given that education and social class may capture different mechanisms related to union dissolution, and in light of the recent remarkable changes in the Italian family demographics: HP1.: We expect a reversal, from positive to negative, in the educational gradient of union dissolution across cohorts.

HP2. :
We hypothesize that, net of education, social class is also significant for union dissolution.We thus anticipate a reversal, from positive to negative, even in the social class gradient of union dissolution across cohorts.
Due to the lower barriers to union dissolution for non-marital cohabitations and in light of the existing empirical evidence for other countries, HP3.: We hypothesize that education and social class are less determinant for the dissolution of non-marital cohabitations than for marriages.
Finally, given the still limited diffusion of women's employment and the persistence of the male-breadwinner model in the Italian society, HP4.: We hypothesize that the change in the socioeconomic gradient of union dissolution across cohorts may be more evident for men than for E. Bastianelli et al. women.

Data and methods
We used data from the two Italian surveys on Families and Social Subjects (FSS) conducted by the Italian Institute of Statistics (Istat) in 2009 and 2016.These are the most complete and reliable retrospective, nationally-representative surveys on Italian individuals and their families.Both had an overall response rate of approximately 80 %.These data include detailed retrospective information (recorded on a monthly basis) on men's and women's partnership histories, which allowed us to follow an event-history approach.
The event studied corresponded to the date of de-facto separation provided by the survey.We considered the respondents' first marriage or cohabitation.Higher-order unions may in fact suffer from selection effects, and approximately 90 % of the individuals in our data only had one union (married or unmarried).The sample of marriages included direct marriages, as well as those preceded by pre-marital cohabitations.Thus, if a respondent first cohabited and then married the same partner, said respondent would appear in both samples (cohabitators and married).The time was measured in months since the date of marriage or beginning of cohabitation, to its end.Episodes were right censored if the partner died, if the union had not ended, and (for cohabitors only) if unmarried cohabitations became marriages.
For marriages, we are able to observe three birth cohorts: Those born before 1960, who mostly grew up when divorce was not yet allowed in Italy; those born between 1960 and 1969, who were raised in the years when divorce was publicly debated and eventually introduced; and finally, those born after 1970, who were born when divorce was already established.As cohabitations spread later in Italy, and the sample size was much smaller (especially for the oldest cohorts) we combined the two oldest cohorts for the analysis of cohabitations, and observed differences between those born before and after 1970. 1 Prior research analyzing changes in the educational gradient of divorce has tended to focus on union cohorts, rather than birth cohorts.We estimated our models with both union and birth cohorts and found that, despite the similarity of the results, birth cohorts better captured the changes in the socioeconomic gradient, highlighting the importance of the generational dimension.
Respondents' education at the time of union formation was measured through the ISCED scale, recoded as "low" for ISCED 0-2, "mid" for ISCED 3-4, and "high" for ISCED 5-6.Those who were still in education (less than 2 % of the sample) were excluded from the analysis.Individuals' social class at the time of union formation was measured according to the ESeC in the three-class version, i.e., routine jobs (lower class), middle class, and salariat (upper class), and an additional category for those not employed for whom the social class could not be detected (Harrison & Rose, 2006;Rose & Harrison, 2007).Among married women, the latter group constitutes approximately the 49 %, with about 55 % in the oldest cohort and 39 % in the youngest cohort.Among cohabiting women, the proportion of non-employed is notably lower, at around 29 % in the oldest cohort and 20 % in the youngest cohort, averaging around 24 %.Among men, the non-employed represent approximately 8-9 % of the sample across all cohorts, regardless of marital status (see Table A2 in the Appendix).About 15 % of individuals who were initially not employed at the beginning of the union later obtained employment.To address this, we performed supplementary analyses using the social class associated with the available employment spell.The findings, accessible in the online supplementary material, remained unchanged.Moreover, social class may have improved over the course of the relationship; however, this only occurred to a limited number of respondents (roughly 6 %).For a robustness check, we ran our models excluding those cases: Our main results proved robust to this test (available in the online supplementary material).
In principle, social class could be assessed at either the individual or household level (Rose & Harrison, 2007), each with its own advantages and limitations.Assessing social class from a household perspective allows for consideration of broader factors such as life opportunities and consumption standards.Ideally, both approaches should be tested.Nevertheless, the FSS survey employed for this study provides detailed retrospective data on socioeconomic characteristics and separations at the individual level, while household-level information is only available at the time of the interview.Consequently, we had to adopt an individual-level approach in defining social class.
All models controlled for region, parents' separation, and parents' education.Region was coded into three categories, measuring whether the respondent resided in the north, center, or south (and Islands) of the country.Parents' separation was coded as a dummy variable indicating whether the respondents' parents were separated or not when the respondents were aged 18.Finally, parents' education was coded as a dummy variable measuring whether at least one between mother and father is higher educated (ISCED 5-6).
We also controlled for other well-known predictors of union dissolution, such as age at union formation, pre-marital cohabitation for marriages, and number and age of children.Despite the results remaining essentially unchanged, we opted for the simpler models (only controlling for region, parents' separation, and parents' education) to observe the total effects of education and social class.Age at union formation, pre-marital cohabitation, and number and age of children may also be dependent on education and social class, and thus partly mediate their effects2 .
We applied stratified Cox models (Kleinbaum & Klein, 2012), stratified by education and cohort, and by social class and cohort.With this approach, the proportionality assumption was relaxed, and the baseline hazard was allowed to vary across education, social classes, and cohorts.Due to there being different baseline hazard functions, the fitted stratified Cox model yielded different estimated survival curves (and survival probabilities) for each combination of education and cohort, and social class and cohort.With this analytical strategy, we were thus able to estimate the survival functions predicted by our models for different population subgroups using a minimum of assumptions while adjusting for covariates.Furthermore, displaying our results as survival probabilities, instead of hazard ratios, allowed for a clearer and more accurate perception of the actual magnitude of the associations.
Within the retrospective section of the survey, couple-level information was unavailable as information of ex-partners was not collected.Hence, we computed our analyses separately for women and men.Moreover, we segmented the analysis by type of union to assess differences and similarities in the relationship between respondents' SES and union dissolution in marriages and cohabiting unions.Our sample consisted of 23,641 married women, of whom 2175 experienced a union dissolution; 19,621 married men with 1790 dissolutions; 3256 cohabiting women with 782 union dissolutions; and 3446 cohabiting men with 1027 separations.For each subgroup, we analyzed to what extent the association between respondents' SES and union dissolution evolved across cohorts.
Unions of individuals born in the youngest cohorts could only be observed for a relatively short time-span, especially compared with the oldest cohorts.In order to maintain a consistent observational window across birth cohorts, we displayed predicted survival curves and probabilities of union dissolution for the first 10 years of marriage and the first 5 years of cohabitation, as cohabitations have (on average) a considerably shorter duration.
It is noteworthy that marriage and cohabitation differ in term of socioeconomic composition.Tables A1 and A2 in the appendix display the distributions of education and employment status and social class by relationship status, gender, and cohort.Notably, differences among women are particularly pronounced.Cohabiting women show significantly higher levels of education compared to married women, with 24 % having tertiary education compared to only 11 % among married women.However, there has been a noticeable increase in the education of married women across cohorts, so that, in the 1970-1990 cohort, married and cohabiting women show similar levels of education.Married women are overrepresented in the "non-employed" category, whereas higher proportions of cohabiting women are in the upper class, approximately 16 % compared to less than 9 % of married women.However, when comparing married and cohabiting women in the youngest cohort , the primary differences lie in the proportions of non-employed individuals and those employed in routine jobs.Among married women, there is a higher share of non-employed, whereas among cohabitants, there is a higher share of women employed in routine jobs.In summary, in Italy, cohabiting women used to represent a selected group of higher educated and higher social class women.However, differences between married and cohabiting women gradually reduce across cohorts.For men, while the distribution across social classes is relatively similar among married and cohabitants, marked differences emerge in educational levels, with cohabiting men exhibiting much higher levels of education, even when comparing married and cohabiting men in younger cohorts.
Furthermore, Table A3 in the appendix displays differences in employment status and social class in marriages and cohabitations by education.Roughly 19 % of highly educated married women were not employed at the beginning of the union, whereas only 48 % were in the salariat class.Among highly educated married men, merely 6 % were not employed, while 63 % were in the salariat class.This difference underscores the existence of a strong gender gap in the occupational returns to education, emphasizing the importance of considering both education and social class for accurately assessing socioeconomic status.Comparing marriages and cohabitations, the main difference among women is that low-educated cohabitors are more represented in the labor market, mainly in routine jobs, whereas low-educated married women are predominantly non-employed.Among men, instead, there are no considerable differences between marriages and cohabitations.

Education
Fig. 1 displays the survival curves predicted by the stratified Cox model for married men and women, stratified by education and cohort, and adjusted for region, parents' separation, and parents' education.Table 1 reports the corresponding predicted cumulative probabilities of marital dissolution for men and women after 10 years of marriage.
For men born before 1960, despite the probabilities being low for all educational groups, the cumulative probability of separation for the highly educated (6 %) was double that for those with low education (3 %), thus denoting a positive educational gradient.In the 1960-1969 birth cohort, there was a drastic increase in the probability of marital dissolution for all educational groups, reaching 9 % for low-and mideducated, and 12 % for the highly educated.Finally, in the 1970-1990 cohort, we observed a clear reversal in the educational gradient of marital dissolution.Overall, while the probability of marital disruption for the highly educated increased from 6 % to 8 % from the oldest to the youngest cohort, for those with lower education it increased from 3 % in the oldest cohort to 13 % in the youngest.Statistically significant differences across cohorts were observed for both educational groups.Thus, in line with Goode's hypothesis, we found a reversal from positive to negative in the educational gradient of marital dissolution for men, confirming HP1.
We observed a similar pattern for married women.For those born before 1960, despite the rarity of marital dissolutions, the marriages of highly-educated women had lower survival rates than those of the lowand mid-educated.After 10 years of marriage, only 2 % of low-educated women belonging to this cohort experienced marital dissolution, while the corresponding figures were 5 % and 7 % for mid-and highlyeducated women, respectively.While differences were small in absolute terms, highly educated women were more than three times more likely to experience union dissolution than the low-educated.Also, the estimation of these differences is statistically precise.Among those born between 1960-1969, mid-and highly-educated women continued to show higher probabilities of marital dissolution than their low-educated counterparts.Finally, in the 1970-1990 cohort, the cumulative probability of experiencing marital dissolution after the first 10 years of marriage was neither substantially nor statistically different across educational groups.
Thus, for married women, HP1 was only partially confirmed as we observed a vanishing (and not a reversal) of the positive educational gradient of marital dissolution across cohorts.This result, however, aligned with HP4: The change in the socioeconomic gradient in union dissolution is less marked when considering women instead of men.

Social class
Fig. 2 and Table 2 display the survival curves and probabilities of marital dissolution for men and women, this time by social class and cohort, net of education.
Among men born before 1960, the probabilities of marital dissolution after 10 years for those in the middle and salariat social classes were higher than for those in routine jobs, and those non-employed.In the 1960-1970 birth cohort, the probabilities of marital dissolution increased for all social classes, with still slightly higher probabilities for the middle and salariat classes (10 %) relative to the routine class (8 %).Finally, in the 1970-1990 cohort, we noted a drastic reversal in the social class gradient of marital dissolution.In particular, relative to the previous cohorts, there has been a substantial increase in the probability of marital dissolution for those in the lowest social class (from 3 % to 11 %), as well as for the non-employed (from 3 % in the oldest cohort to 18 % in the youngest cohort).Therefore, also when considering married men's social class gradient in marital dissolution, Goode's hypothesis was confirmed.In line with HP2, net of education, social class appears to play an independent and crucial role in the prediction of marital dissolution, and the social class gradient in marital dissolution turned from positive to negative across cohorts.
For women, the analysis of the social class gradient in marriage dissolutions (net of education) led to similar conclusions.In the two oldest cohorts, the social class gradient appeared positive.Among women born before the 1960s, although the probabilities of marital dissolution were relatively low for all social classes, marriages in the salariat and middle classes had lower survival probabilities.In the 1960-1969 cohort, we noted a drastic increase in marital dissolution probabilities for all social classes, and group-specific differences narrowed.Moving to the youngest cohort, we found that the probability of marital dissolution in the salariat social class stabilized compared to the 1960-1969 cohort (with 11 % of women experiencing marital dissolution), while there was a substantial increase in the probabilities for those in the lower social classes (up to 14 % for women in the routine class), and for non-employed women, which overtook the probability for salariat women.
We thus identified a distinct role of education and social class even among married women.Considering social class instead of education, we found a mild trace of a reversal of the socioeconomic gradient, in line with HP2, although it must be noted that the differences between social classes were not statistically precise.These findings again support HP4, in that the change in the socioeconomic gradient of union dissolution across cohorts is more evident for men than for women.

Education
Fig. 3 displays the predicted survival curves for cohabiting men and women, and Table 3 the corresponding cumulative probabilities of union dissolution after the first 5 years of cohabitation.
For men, dissolution rates from non-marital cohabitations were far higher than from marital dissolutions: After 5 years, roughly 30 % of men experienced dissolution.However, our results did not reveal a clear educational gradient.In the older cohort, mid-educated men had a higher probability of separating than their low-and high-educated counterparts.In the youngest cohort, however, the educational gradient was positive, as low-educated men were less likely to dissolve their unions than those with middle and high education, while there were no differences between these latter groups.Thus, we found no

Table 1
Cumulative probability of marital dissolution after 10 years of marriage for men and women, by education and birth cohort.Note: adjusted for region of residence, parent's separation, and parent's education.Control variables are set to their modal category: region=north, parents not separated, and non-tertiary educated parents.
pattern of changes across cohorts consistent with Goode's hypothesis.This result aligned with our expectation that socioeconomic differences are less relevant for the dissolution of cohabiting unions (HP3).Also considering cohabiting women, we found them to have a far higher probability of union dissolution than married women.Similarly to men, in the cohort of women born before 1970, those with an intermediate level of education had the highest probability of separating, whereas we found no differences between high-and low-educated women.In the 1970-1990 cohort, probabilities of union dissolution after 5 years of union increased for all women, but especially for the low

Table 2
Cumulative probability of marital dissolution after 10 years of marriage for men and women, by employment status and social class, and birth cohort.Note: adjusted for region of residence, parents' separation, parents' education, and respondent's education.Control variables are set to their modal category: region=north, parents not separated, non-tertiary educated parents, education=mid.
educated, meaning that we observed no substantially or statistically significant differences in the cumulative probability of cohabitation dissolution across educational groups.Our HP1 on a reversal, from positive to negative, of the educational gradient in union dissolution was thus unconfirmed also for cohabiting women.

Social class
The social class gradient in the dissolution of cohabitating unions for men and women is displayed in Fig. 4 and Table 4.In line with the results concerning education, no clear pattern of social class gradient (net of education) was evident among men born in both cohorts.In fact, notwithstanding the high estimation uncertainty, a negative class gradient seemed to emerge after 5 years of union in the oldest cohort, contrary to theoretical expectations (HP2).However, the results suggest that being out of employment is particularly detrimental to men's union stability.The differential in the probability of union dissolution between the non-employed and the other groups considerably increased across cohorts.Among men born before the 1970s, the probability of union dissolution for those outside of the labor market was 33 %, and rose to 44 % for the 1970-1990 cohort.Therefore, what appears most significant for cohabiting men is being out of employment, rather than education or social class.
Regarding social class differentials for women, we again detected no clear pattern.Among those born before the 1970s, women with routine jobs had the lowest risk of dissolution.However, throughout our observational window, we detected hardly any differences between women in the other social classes or the non-employed.In the 1970-1990 cohort, the probability of dissolution after 5 years of cohabitation increased for all women, and differences among social classes were virtually null.
Thus, the results aligned with HP3 in that education and social class appeared less relevant for the dissolution of cohabitations, to the point that we observed no clear pattern of change across cohorts in the socioeconomic gradient, neither for men nor for women.

Conclusion and discussion
The present study adds to the existing literature on the changing socioeconomic gradient in union dissolution by addressing three main shortcomings.First, unlike most studies, we considered the social class gradient in union dissolution, in addition to the educational gradient.Second, we included dissolutions from both marriages and cohabiting unions.Third, while the majority of the existing research on the topic only focused on women, the present study analyzed the socioeconomic gradient in union dissolution of both women and men.We focused on Italy, a country long considered to be an exception in the diffusion of SDT-related family behaviors, and for which evidence on the socioeconomic gradient in union dissolution is scant and outdated.
It is well-known in the literature that the socioeconomic gradient in union dissolution-usually operationalized in terms of educational differences-tends to turn from positive to negative over time (Harkonen & Dronkers, 2006;Matysiak et al., 2014).Following the pioneering work of Goode (1962Goode ( , 1993)), the positive educational gradient in divorce among older cohorts can be traced back to the ability of high-educated

Table 3
Cumulative probability of cohabitation dissolution after 5 years of cohabitation for men and women, by education and cohort.Note: adjusted for region of residence, parents' separation, and parents' education.Control variables are set to their modal category: region=north, parents not separated, and non-tertiary educated parents.
individuals to overcome the legal, social, and economic barriers to divorce at a time where it was rare and still viewed as a culturally "disruptive" behavior.When such barriers decline, divorce spreads among low-educated couples, eventually producing a reversal of the socioeconomic gradient.We found evidence of a change in the educational gradient in divorce over time, even in Italy.Across birth cohorts, we noted a reversal in the educational gradient from positive to negative for married men, and a vanishing of the positive gradient for married women.Contemporary Italian marriages formed by highly-educated women are not necessarily less stable than those formed by less-educated women.Not only does this finding confirm prior evidence for Italy based on older cohorts (Salvini & Vignoli 2011), it also generally strengthens the findings for Southern Europe (Bernardi & Martínez-Pastor, 2011).While some of the mechanisms underlying the changing socioeconomic gradient in union dissolution relate to the cultural resources available to individuals, and are thus more directly linked to individuals' level of education, others are more strictly economic, and may be more accurately grasped by social class.For instance, high-SES couples are less exposed to economic strain and are more likely to share financial assets and long-term investments (e.g., home ownership), which consequently raise the financial costs of divorce.These withincouple economic mechanisms have been found to play a crucial role in the emergence of a negative educational gradient in union dissolution (Boertien & Härkönen, 2018).Accordingly, our results show that, net of education, social class has an important and distinct role in the prediction of marital dissolution.These results are in line with those found for the educational gradient.Across birth cohorts, we detected a reversal (especially notable among men) from a positive to a negative social class gradient in marital dissolution.
Our findings on the significant and independent role of social class are especially important in the context of educational expansion.As elucidated in a recent study by Zilincikova, Skopek, and Leopold (2023), the emergence of a negative educational gradient in union dissolution is now somewhat tempered by the declining proportion of lower-educated individuals due to educational expansion.However, while there may be a decrease in the absolute numbers of lower-educated individuals, our findings of a reversal in the social class gradient, even after accounting for education, underscores the enduring relevance of the broader socioeconomic gradient in divorce dynamics.Despite educational expansion, low social classes are unlikely to vanish, thus highlighting the potential of social class as a crucial and more precise indicator to unravel socioeconomic patterns in family dynamics.
Regarding the educational and social class gradients in the dissolution of cohabitations, the results did not accord with our theoretical expectations.Educational differentials in cohabitation dissolution did not seem to follow any recognizable pattern, neither for men nor for women.Similarly, social class, net of education, showed no clear-cut

Table 4
Cumulative probability of cohabitation dissolution after 5 years of cohabitation for men and women, by employment status and social class, and birth cohort.Note: adjusted for region of residence, parents' separation, parents' education, and respondent's education.Control variables are set to their modal category: region=north, parents not separated, non-tertiary educated parents, education=mid.
E. Bastianelli et al. gradient.Instead, we found it to be detrimental to men's cohabitation stability being out of employment-a negative effect that increased considerably across cohorts.The economic disadvantage derived from non-employment had similar implications for married and cohabiting men.This finding recalls the incapability of non-employed men to fulfill their traditional male breadwinner role, which appears to be a destabilizing factor for couples in countries with low female labor market participation (Bastianelli & Vignoli, 2022).Our results suggest that socioeconomic status is less important in the prediction of union dissolution for cohabitation than for marriage.This supports the notion that the reversal of the socioeconomic gradient of union dissolution is driven by more committed partners, namely married couples (Kalmijn & Leopold, 2021).Generally speaking, gender differences emerged in the socioeconomic gradient of marital dissolution in Italy.Such differences in the role of education and social class in the prediction of divorce could be explained by the still limited labor market participation and low earnings of Italian women-especially among the low-educated and those in low social classes (Cantalini & Ballarino, 2023;OECD, 2017).Indeed, increased women's labor force participation has been identified as one of the crucial factors associated with the change in the educational gradient of divorce (Matysiak et al., 2014).Despite women of all educational levels having seen a considerable increase in union dissolution probabilities across cohorts, many of them are often housewives and, if employed, play complementary economic roles within the couple.This may explain why we observed no negative educational gradient in marital dissolution: Low-educated Italian women may lack the material means to separate, and consequently, their risk of union dissolution is likely to remain relatively low, irrespective of divorce's level of diffusion.For the same reasons, our results show the first traces of a reversal of the social class gradient in marital dissolution also for women, since social class more accurately captures women's economic independence.Another noteworthy finding is the change in the socioeconomic gradient in union dissolution for men.While this aligns with Goode's theory, the shift from a positive to a negative socioeconomic gradient in men's divorce only in recent cohorts contradicts the prevailing view in the literature that men's higher SES always stabilizes unions.
In all, our study leads to novel findings on the changing socioeconomic gradient in union dissolution in marriages and cohabitations.Our results show that, even in Italy, a general democratization of union dissolutions has occurred, which have spread across all educational levels and social classes, and a reversal from a positive to a negative gradient in marital dissolutions is now visible.This finding is of crucial importance to understanding the future consequences of union dissolutions on the reproduction of social inequalities in Italy (Guetto & Panichella, 2019).If marital dissolutions are more widespread among lower-SES groups and are associated with higher risks of (further) socioeconomic deprivation for their children, we can expect "diverging paths" to emerge (Kalmijn & Leopold, 2021;Mclanahan, 2004): The educational and socioeconomic outcomes of children who were already disadvantaged will be more negatively affected by growing marital instability compared to those of children from more advantaged families.Addressing to what extent these diverging paths might be mitigated by the diffusion of cohabitation-that does not display any clear educational and social class gradients-would be an interesting avenue for further research.

Fig. 1 .
Fig. 1. : Survival curves for marital dissolution, by education, and cohort, Note: adjusted for region of residence, parents' separation, and parents' education.Control variables are set to their modal category: region=north, parents not separated, and non-tertiary educated parents.

Fig. 2 .
Fig. 2. : Survival curves for marital dissolution, by employment status and social class, and cohort, Note: adjusted for region of residence, parents' separation, parents' education, and respondent's education.Control variables are set to their modal category: region=north, parents not separated, non-tertiary educated parents, education=mid.

Fig. 3 .
Fig. 3. : Survival curves for cohabitation dissolution, by education and cohort, Note: adjusted for region of residence, parents' separation, and parents' education.Control variables are set to their modal category: region=north, parents not separated, and non-tertiary educated parents.

Fig. 4 .
Fig. 4. : Survival curves for cohabitation dissolution, by employment status and social class, and cohort, Note: adjusted for region of residence, parents' separation, parents' education, and respondent's education.Control variables are set to their modal category: region=north, parents not separated, non-tertiary educated parents, education=mid.

Table A2 :
Distribution of employment status and social class by relationship type, gender, and cohort.

Table A3 :
Distribution of employment status and social class by education, relationship type, and gender.Supplementary data associated with this article can be found in the online version at doi:10.1016/j.rssm.2024.100954.