Socio-economic status and the rise of divorce in Sweden: The case of the 1880–1954 marriage cohorts in Västerbotten

Abstract

An established negative association between socio-economic status (SES) and divorce has applied to most Western nations since 1960. We expected a positive association between SES and divorce for low-divorce contexts historically because only individuals in higher social strata had the resources to overcome barriers to divorce. According to Goode’s socio-economic growth theory, this relationship was reversed as industrialization and modernization began removing the economic and normative barriers. Making use of longitudinal data from parish registers, we investigated SES and other micro-level determinants of divorce among men and women in northern Sweden who married between 1880 and 1954. Results indicated a positive association between SES and divorce among those who married 1880–1919, with the middle class, not the elite, featuring the highest divorce risks. This association changed for couples who married in the 1920s, for whom divorce became more common and the working class faced similar divorce risks to the higher social strata.

Keywords:

• divorce
• Sweden
• socio-economic status (SES)
• economic independence
• event history analysis

Introduction

Divorce increased dramatically across the Western world during the twentieth century. Divorce law reform was a prerequisite for this increase (Stone 1990; Sandström and Garðarsdóttir 2018). Explanations also include economic and cultural factors. One strand of this literature, based on William Goode’s socio-economic growth theory (Goode 1951, 1963), emphasizes that industrialization together with modernization prompted socio-economic and cultural change through altered modes of production (e.g. the growth of wage work done first by men, then by women), which had implications for living standards, gender relations, and family patterns. A negative association between men’s socio-economic status (SES) and divorce has been established for many Western countries since 1960, while findings regarding the association between women’s SES and divorce vary (see Lyngstad and Jalovaara 2010; Härkönen 2014). However, we know little about this relationship in the more distant past when divorce was rare.

We expected a positive association between SES and divorce for historical contexts. Our assumption was not that higher social groups in the past experienced worse marriages than others but that they were better equipped both financially and socially to overcome existing barriers to divorce. In low-divorce contexts in the past, such barriers were legal, as well as economic and normative, making divorce costly in many ways. The higher classes could afford to cover monetary costs related to separation and divorce, such as legal advice, lodging, and transportation, and were probably better able to navigate the legal system and make informed choices that minimized the social costs involved. The high costs of divorce early on effectively constrained and thus excluded those with fewer resources (Stone 1990, pp. 386–87). According to Goode’s theory, the positive relationship between SES and divorce was reversed in parallel with industrialization and modernization, whereby costs were reduced, constraints on the average married couple’s efforts to address their relationship problems were eased, and broader layers of the population could divorce.

In this paper, the main question is whether Goode’s theory really does hold up and correspond with the socio-economic patterns observed during the transition from low to high divorce rates in Sweden. Until now there have been few empirical tests of Goode’s theory, and there is limited evidence as to whether there was a positive SES gradient in divorce initially, whether it was reversed with industrialization, and when, if so, this occurred. We contribute to this literature by looking at Sweden in answering these questions. We tested Goode’s socio-economic growth theory and the hypothesis that the SES gradient of divorce was reversed from positive to negative during the country’s modernization. We considered both husband’s and wife’s SES in the analysis, thus adding a gender dimension to socio-economic growth. We also tested the economic independence hypothesis: that women with a recorded occupation and labour market experience were more likely to divorce than other women. The analyses exploited longitudinal data based on parish registers drawn from the POPLINK database covering Västerbotten County; this database include SES and other micro-level determinants of divorce for men and women in northern Sweden who married in 1880–1954 and were included in our observation up to 1960.

This paper contributes to the literature on SES and divorce by means of a detailed yet comprehensive analysis of historical divorce patterns. Previous research has focused primarily on the period after 1960 and missed the initial rise in divorce that occurred during the first half of the twentieth century in many Western countries. This is the period referred to in Goode’s predictions but not fully investigated in a country context until now. Moreover, the hypothesis that increased economic independence among women is important for the increase in divorce has not been tested for this period either. Our results indicated that both the negative association between SES and divorce and the positive association between women’s economic independence and divorce, as documented in modern contexts, emerged in the first half of the twentieth century, at least in societies (such as Sweden) that emphasized both class equality and gender equality early on.

Background

Figure 1 serves as the backdrop of this study. It shows the transition from a low-divorce to a high-divorce society by the trend in the crude divorce rate in Sweden in 1911–2019 and a three-year moving average of the crude divorce rate in Västerbotten. The lower-level trend in Västerbotten closely reflected developments at national level. The fact that northern Sweden (including Västerbotten) lagged behind the national trend in nineteenth-century demographic processes such as fertility decline (Sundbärg 1910) yet followed the general national trend in demographic shifts during the twentieth century has been shown elsewhere (Junkka 2018; Sandström and Marklund 2019). That the trends in these settings were very similar would suggest that the determinants of divorce were the same and that we can generalize the findings from Västerbotten. This long-term trend was characterized by two periods of marked increase during which most (80 per cent) of the growth in divorce occurred. The first took place during the 1940s and 1950s, when the divorce rate more than doubled, while the second started in the mid-1960s. There was a spurt in the divorce rate in the mid-1970s with the introduction of unilateral no-fault divorce in 1974. Figure 1 shows that the period of high divorce fell outside the scope of our study, which ended in 1960. The results from this study should be assessed in the light of a slow yet consistent increase in the divorce rate in the early twentieth century until the more distinct increase during the 1940s and early 1950s was seen nationwide.

Figure 1 (a) Crude divorce rate in Sweden and (b) three-year moving average of crude divorce rate in Västerbotten County, 1911–2019

Source: Statistics Sweden (1911, 1963, 1967, 2022).

No-fault divorce came into law in Sweden in 1915. This meant that spouses could apply for divorce if they agreed on the ‘irretrievable breakdown’ of their marriage. Fault-based reasons, such as adultery, allowed one of the spouses to apply for divorce unilaterally, but this became uncommon: as early as the 1930s, four out of five divorces were ‘no fault’ (Sandström 2011). The 1915 divorce law was an integral part of social reformist ambitions regarding family relationships and welfare, and thus influenced the reform of marriage legislation that resulted in the Marriage Code of 1920. Of note, the way it regulated the legal relationship between spouses was progressive by contemporary standards. It emphasized equality between spouses, although this concept had a different meaning from today’s (Melby et al. 2006).

Socio-economic status and divorce

The way that individual SES relates to risk of divorce has been extensively investigated for Western contexts after 1970. Various indicators of men’s SES, such as occupation, education, employment, and income, have been found to be negatively associated with divorce in Scandinavia, the UK, the United States (US), and Japan (e.g. Ross and Sawhill 1975; Becker et al. 1977; Cherlin 1979; Haskey 1984; Murphy 1985; Bracher et al. 1993; Jalovaara 2001, 2003, 2013; Lyngstad 2004; Raymo et al. 2004; De Graaf and Kalmijn 2006; Ono 2009; Kreager et al. 2013; Boertien and Härkönen 2018). In contrast, results for women’s education, labour market attachment, and income have been inconclusive (see Spitze 1988; Sayer and Bianchi 2000; Lyngstad and Jalovaara 2010 for reviews). Cross-national comparisons indicate that SES gradients in divorce vary across countries—depending on how common divorce is (Härkönen and Dronkers 2006)—through a positive SES gradient, particularly for women in societies characterized by low rates of divorce. Differences have, however, become less marked in recent decades (Matysiak et al. 2014).

Scholars have argued that these patterns may be understood by Goode’s predictions of change in the determinants of divorce during the transition from a low- to a high-divorce regime. An element of diffusion is implicit in this reasoning. In societies where divorce is uncommon due to legal, economic, and normative barriers, it may be a cultural innovation that spreads over time through communication and interaction among members of a social system (Rogers 1962). Usually the educated, financially comfortable upper classes are early adopters of such innovations, including divorce, partly because they can act on their motivation and marshal enough resources to overcome the extant barriers (Goode 1962, 1993). Thus, according to Goode’s (1951, 1963) socio-economic growth theory, there should have been a positive relationship between SES and divorce in the early industrialization period, which was characterized by low divorce rates. As the different barriers to divorce were overcome through change in legislation and social norms, improved living standards, and women’s economic independence, the adoption of this innovation should have extended to broader groups in the population. According to Goode, divorce among the lower classes was initially restricted by their lack of resources, although they experienced greater economic hardship and social problems. Goode argued that the association between SES and divorce would have reversed over time and turned negative as constraints to divorce were relaxed, with divorce gaining momentum as a social convention. While the lower classes could not initially afford to divorce, they could do so in increasing numbers at later stages of the divorce transition. This was especially true for women who were married to unskilled men or men with no occupation; such women were burdened by the dual responsibilities of unpaid domestic work and paid work to make a living (Goode 1963, p. 13).

Few studies have established a positive association between SES and divorce in low-divorce contexts before the divorce transition. Extant studies (e.g. Goode 1951, 1962, p. 517; Phillips 1988, p. 608) have drawn inferences primarily from aggregate data in published divorce statistics rather than analysing individual-level data and adjusting for compositional changes in the population. Studies using individual-level historical data from the Netherlands and Flanders have shown a positive association between social class and divorce during the very early stages of the divorce transition (i.e. the late nineteenth century through to the 1920s, see Van Poppel 1997; Matthijs et al. 2008; Kalmijn et al. 2011). For later stages, more research exists on the determinants of divorce, and it supports a negative association between SES and divorce. In the Netherlands, the impact of education among those who married in 1942–99 became more negative across cohorts (De Graaf and Kalmijn 2006), although Teachman (2002) found a stable association between education and risk of divorce for the 1950–84 cohorts in the US. South (2001) found that wife’s education had a stable impact for those who married between 1969 and 1993. Using Swedish register data and comparing marriage cohorts in Sweden that corresponded to those studied by South, Hoem (1997) found few differences with respect to education and divorce risk before 1974, after which reform triggered a trend towards increased divorce for all, although more so for lower educational categories.

Because tests of Goode’s socio-economic growth theory for historical contexts are few and limited to the Netherlands, and because most studies of SES and divorce have focused primarily on the period after 1960, when divorce was already on the increase, extant research misses the initial rise in divorce that occurred during the first half of the twentieth century (mainly 1920–60) in most Western countries. We believe it is important to extend the historical scope to capture the span of industrialization and modernization, including the period at the core of Goode’s conjectures. In the case of Sweden, this means extending research back to the 1870–80s, when industrialization began (Schön 2012). In doing so, we must remember that we cannot fully distinguish between different trends, such as those in industrialization and modes of production, in modernization (including egalitarianism, women’s economic independence, new family roles, and propensity to marry), and in divorce, these being parallel processes. Our long-term perspective (covering the period 1880–1960) allowed us to cover the transition from a traditional agricultural society to a modern industrial economy. Of note, Goode did not elaborate on this aspect of industrialization and was admittedly vague about the links between industrialization and family change (Goode 1963, pp. 8, 10–27). While he asserted that the role of the family and patterns of marriage, childbearing, and divorce change along with the economic system’s base, his emphasis was on the features specific to industrial society. Goode was quite reticent about the agricultural sector and the farmers staying in it, although we might expect increasing differences in family patterns between those staying (and gradually forming a more select group over time) and the rest of the population (Goode 1963, pp. 13–15, 19).

Economic independence and divorce

Another influential strand of the divorce literature emphasizes the role of women’s economic independence. Historical trends support a link between married women’s increased labour force participation and the rising divorce rate across contexts during the past century.

The hypothesis that married women’s employment destabilizes marriage is found in both economics and sociology. Proponents of theories of the family have argued that specialized and differentiated marital roles make a marriage both efficient and stable (Parsons 1949; Becker 1974, 1981); thus, the tendency among married women to spend more time on paid work and less time on childrearing and housework reduces the gains from marriage and increases the propensity for couples to divorce. Married women’s paid work outside the home has increased their bargaining power within marriage and also improved their economic resources in terms of income, enabling them to leave a dysfunctional marriage, especially where public support exists in terms of transfers and services for single mothers (England 2003). The total impact of a married woman’s economic resources on divorce is known as the ‘independence effect’ (Ross and Sawhill 1975).

The Swedish context we studied (as well as the contexts studied by Goode) may be labelled a male breadwinner society. Although married women’s labour force participation was low, men and women worked for the same reason: to sustain themselves and their families. In cases where the husband could not provide for his family, the wife had to work for pay (alongside the responsibility for housework and childcare) because no real public support existed until the 1960s, when the Swedish welfare state became more generous. For low-SES groups, wage levels were low. Goode (1963) was largely silent about sex differences, although he pointed out that one reason for the higher divorce rates among the lower classes was that low-SES women had relatively less to lose economically from divorce because they were more likely to work for pay than married middle- and upper-class women, and their wages were more like their husbands’ than higher-SES women’s wages were (Goode 1963, pp. 13, 15–17). Of importance, Goode’s argument implies that middle-class and elite women had higher stakes both economically and socially in the stability of their marriage compared with low-SES women, thus having more to lose from a divorce (Goode 1963, pp. 81–6; Stone 1990, p. 403). Goode was also vague on the implications of women’s changing economic roles for both women’s employment and the economy as industrialization progressed and services became more important, although he acknowledged that if women could support themselves (albeit only just) through their own efforts, divorce rates would increase. This reasoning is in line with Becker’s argument regarding specialization and gains to marriage, whereby women’s economic independence is positively associated with divorce (Becker 1974). Both Becker and Goode addressed the way that individual-level barriers to divorce are overcome through economic restructuring, allowing more equal access to divorce, although Goode focused more on social class than sex differences.

In Sweden during our period of investigation, gender roles were traditional and complementary, and married women’s economic dependence on their husbands was typically high. The labour force participation rate among married women was less than 1 per cent in 1880 and increased slowly over the decades. Only in the 1940s did married women’s gainful employment exceed 10 per cent, and by 1960 it reached 23 per cent. We acknowledge that these figures undercount married women’s paid work, especially that of farmers and the working class. Although the wage gap between men and women narrowed, it was still 30 per cent in 1960 and applied across sectors. These trends are very similar to those documented for other countries, including the US (Stanfors and Goldscheider 2017).

In the context of our study, we have reason to believe that women’s economic independence mattered for divorce (if only to a limited extent) because married women became more able to overcome the economic barriers in their way. Access to own income—either directly through employment (rare among married women) or indirectly through work experience (common)—improved women’s opportunities for opting out of a dysfunctional marriage. Over time, the increasing acceptance of married women in paid work, as well as that of divorce, eased the normative consequences of leaving an unhappy marriage. From this perspective, divorce became more ‘affordable’ (Goode 1963, 1993; Phillips 1991). At the same time, women were the primary caregivers for children and other dependants and, in the event of divorce, were both main breadwinner and caregiver. Unlike in the US, alimony was uncommon in Sweden. As early as 1920, the Swedish Marriage Code stated that spouses were equally responsible for the family’s provision (the man through income from paid work and the woman through unpaid housework and caregiving), and if separated, the woman was expected to provide for herself as, in practice, post-divorce allowances were provided only for children (Melby et al. 2006, chapters 1 and 6).

Data and methods

We explored data on men and women who married in Västerbotten County, northern Sweden, in 1880–1954. The POPLINK data are based on information from parish registers and cover approximately 350,000 individuals. The data consist of longitudinal demographic information for individuals and households plus records of vital events including divorce. Administrators of the parish records were required to keep track of occupation for all individuals and update this information at the time of vital events, such as moves, marriages, and births/baptisms, and this provides us with individual occupational histories. We used data from ten parishes, including the urban areas of Skellefteå and Umeå and their rural surroundings (see Figure 2 for location of these parishes). For a detailed discussion of POPLINK, see Westberg et al. (2016).

Figure 2 Study area in Västerbotten County

Source: Demographic Data Base, Umeå University.

The requisite information was available for 47,460 couples in their first marriage, of which 691 experienced divorce during the period of observation. We decided to exclude higher-order marriages from the analysis due to selection bias, because the mechanisms leading to higher-order divorces could be different from those for first marriages and divorces. We grouped individuals into three broad marriage cohorts: 1880–1919, 1920–1929, and 1930–1954. The set-up of marriage cohorts partly related to data binning and partly reflected institutional change before, during, and after the notable increase in divorce during the 1940s. Descriptive characteristics of couples included in the analytical sample are displayed in Table 1. From the distributions, we see proof of first marriage at quite an early age (typical for this period), because in about half the couples the woman was under 25. We see evidence of industrialization when looking at the occupational distributions of both men and women. Across marriage cohorts, an increasingly smaller share of the study population were labelled farmers, whereas increasingly larger shares are categorized as unskilled or skilled workers, particularly among men who were more involved in industrial work than women. The middle class also grew over time. Among women, the share with a recorded occupation was around 60 per cent. This is far higher than estimates of married women’s economic activity according to census data, which indicate that most women left the labour force on marriage or when they had their first child. Concomitant with industrialization was a geographical displacement, whereby an increasingly smaller share of the couples studied lived in rural parishes while a growing proportion lived in urban parishes.

Table 1 Descriptive statistics of couples in analytical sample (numbers and percentages): men and women who married in 1880–1954 in Västerbotten County, Sweden

	Broad marriage cohort			Total
	1880–1919	1920–29	1930–54
Marriage cohort
1880–89	2,367 (14.4)	–	–	2,367 (5.0)
1890–99	3,508 (21.4)	–	–	3,508 (7.4)
1900–09	4,728 (28.8)	–	–	4,728 (10.0)
1910–19	5,793 (35.3)	–	–	5,793 (12.2)
1920–29		7,135 (100.0)	–	7,135 (15.0)
1930–39	–	–	9,914 (41.4)	9,914 (20.9)
1940–54	–	–	14,015 (58.6)	14,015 (29.5)
Age at marriage
<20	955 (5.8)	344 (4.8)	1,333 (5.6)	2,632 (5.5)
20–24	6,640 (40.5)	2,913 (40.8)	11,018 (46.0)	20,571 (43.3)
25–29	5,464 (33.3)	2,398 (33.6)	7,325 (30.6)	15,187 (32.0)
30–34	2,062 (12.6)	923 (12.9)	2,512 (10.5)	5,497 (11.6)
35–39	774 (4.7)	335 (4.7)	965 (4.0)	2,074 (4.4)
40+	501 (3.1)	222 (3.1)	776 (3.2)	1,499 (3.2)
Number of children
Childless	1,723 (10.5)	1,069 (15.0)	5,079 (21.2)	7,871 (16.6)
One child	1,717 (10.5)	1,391 (19.5)	6,884 (28.8)	9,992 (21.1)
Two children	2,070 (12.6)	1,635 (22.9)	6,249 (26.1)	9,954 (21.0)
Three or more children	10,886 (66.4)	3,040 (42.6)	5,717 (23.9)	19,643 (41.4)
Socio-economic position man
Unskilled workers	4,760 (29.6)	2,425 (34.5)	9,657 (41.1)	16,842 (36.2)
Skilled workers	1,771 (11.0)	1,015 (14.4)	4,104 (17.5)	6,890 (14.8)
Farmers	6,558 (40.8)	1,780 (25.3)	3,095 (13.2)	11,433 (24.5)
Middle class	1,900 (11.8)	1,179 (16.8)	4,130 (17.6)	7,209 (15.5)
Elite	1,088 (6.8)	627 (8.9)	2,492 (10.6)	4,207 (9.0)
Socio-economic position woman
No occupation	5,848 (35.7)	3,128 (43.8)	10,405 (43.5)	19,381 (40.8)
Unskilled workers	5,160 (31.5)	1,634 (22.9)	7,191 (30.1)	13,985 (29.5)
Skilled workers	341 (2.1)	329 (4.6)	1,657 (6.9)	2,327 (4.9)
Farmers	4,472 (27.3)	1,625 (22.8)	2,627 (11.0)	8,724 (18.4)
Middle class and Elite	575 (3.5)	419 (5.9)	2,049 (8.6)	3,043 (6.4)
Parish
Skellefteå rural	10,038 (61.2)	3,599 (50.4)	11,428 (47.8)	25,065 (52.8)
Skellefteå urban	704 (4.3)	787 (11.0)	3,171 (13.3)	4,662 (9.8)
Umeå rural	3,677 (22.4)	1,397 (19.6)	4,904 (20.5)	9,978 (21.0)
Umeå urban	1,977 (12.1)	1,352 (18.9)	4,426 (18.5)	7,755 (16.3)
N	16,396	7,135	23,929	47,460
Events	142	176	373	691

Note: Figures in parentheses are percentages. Woman’s occupation (if any) is at the time of marriage but is updated in the analysis when new information is provided in the parish register at the time of vital events, such as changes of address, births of children etc.

Source: POPLINK database, Demographic Data Base, Umeå University.

The main event of interest in the analysis was formal divorce. No information was available on informal separation, although this was known as ‘poor man’s divorce’. We analysed first marriages entered between 1880 and 1954 to estimate exposure-time-specific risk of divorce during the first 25 years of marriage. Individuals entered the risk set either when they married during the period of observation or when they moved into the study area as a married couple. Individuals were followed until divorce, death (of either partner), outmigration, or the year 1960, whichever occurred first. We limited follow-up to 25 years to reduce differences in exposure time between the 1930–54 marriage cohort and preceding cohorts. As a robustness check, we estimated additional models that excluded couples married after 1945 to ensure that individuals were followed for at least 15 years. Results were robust and did not change the conclusions (see Table A1, Appendix).

To estimate the impact of SES on divorce as a function of time at risk, we used occupational status as a time-varying covariate. Occupational titles in the database are coded to follow the Historical International Standard Classification of Occupations (HISCO) 2002 standard and guidelines from Van Leeuwen et al. (2002). The classification scheme is a hierarchical five-digit code. Each digit describes a new level of detail, and codes sharing the same first three digits show similar occupational groups. The SES measure was coded into socio-economic strata using the HISCO classification system for occupations and the Social Power scheme (social stratification system) developed by Van de Putte and Miles (2005).

We used non-parametric Kaplan–Meier survival analysis to estimate how risk of divorce varied with time at risk for different social strata across the three broad marriage cohorts. The 1880–1919 cohort represented primarily those couples who married before the introduction of no-fault divorce in 1915. The second cohort consisted of couples who married in 1920–29 under the reformed Marriage Code of 1920. The third cohort comprised those who married in 1930–54, just before Sweden truly changed from a traditional agricultural society to a modern industrial economy. We used this non-parametric analysis to estimate divorce patterns for men and women separately by cohort and socio-economic position without adjusting for other factors. We then estimated semi-parametric Cox proportional hazard regression models, focusing on the net impact on divorce of the man’s or woman’s SES, respectively, while holding their spouse’s SES constant. To the best of our knowledge, no previous study covering this long historical period of industrialization and modernization has been able to explore the impacts of both the husband’s and wife’s SES (thereby testing for SES and for women’s economic independence) at that time.

We excluded men with no information on occupation, as they formed a very select group (less than 2 per cent of the men in our sample). We retained women with no occupation for well-known reasons of how this information was recorded after marriage (Humphries and Sarasúa 2012; Stanfors and Goldscheider 2017). Many women in our sample changed on marriage from employed, typically in agriculture or domestic service, to not employed and/or homemakers, as was common at that time. Although most women left the labour force on family formation, some worked intermittently. After the Second World War, it became increasingly common for women to re-enter the labour market once their children were of older school ages. We estimated that approximately 10 per cent of the women in our sample resumed paid work at age 40 or older. Given the few women classed as elite, we chose to combine the elite and middle class categories for women.

For both men and women, we used a time-varying definition of SES based on individual occupational status. This allowed us to capture the net impacts of men’s and women’s SES independently of each other and to account for changes in occupational status over time. Of relevance for our study, it also allowed us to capture an increasing number of married women with changing occupational status as they left the labour force and re-entered after a period as homemaker. We focused on SES at the individual level and did not account for different combinations of SES within the couple. We controlled for age at marriage, number of children (at time t), and parish-level context (essentially differentiating between individuals living in urban and rural parishes, with the latter typically being later adopters of new social mores such as divorce).

To check robustness, we estimated models with SES at time of marriage instead of using a time-varying specification. These models (not shown) rendered similar yet somewhat weaker results (based on Akaike and Bayesian information criteria). This finding applied both to men’s and women’s SES combined and to the relationship of each to divorce risk. We tested alternative measures of the wife’s SES, for example, using her father’s SES, which had a weak impact on divorce risk compared with her own SES. We also estimated models controlling for period instead of marriage cohort by splitting analysis time, although this did not impact the results. We believe that using cohorts captured differences in the gender regime and human capital between groups of women, making the cohort approach more theoretically justifiable than the period approach. Further, we introduced a shared frailty at parish level to control for time-invariant unobserved heterogeneity across parishes, but this made no meaningful change to the results, did not affect the conclusions drawn, and was discarded because we preferred a parsimonious approach. Of note, the final models referred to in the text showed no signs of misspecification. The only sign of non-proportionality as indicated by Schoenfeld residuals related to the impact of number of children at time t. Estimating a more complex model that relaxed the assumption of proportionality for this variable by stratification or through an interaction with analysis time did not affect the estimates of the theoretically relevant SES variables. Therefore, we chose to report results from the more parsimonious specification (without stratification or a time interaction for the impact of number of children). All estimations were performed using Stata 17.0.

Results

Figure 3 shows risk of divorce as a function of time since marriage for the analytical sample in Västerbotten, using Kaplan–Meier survival estimates. The pattern shows a gradual and steady increase in the risk of divorce over time, with each marriage cohort experiencing higher divorce intensities than the previous one. There is a marked shift towards higher divorce intensity for those who married from 1920 onwards. Although this indicates a large relative increase in the risk of divorce, we should remember that divorce was a very rare phenomenon in Sweden at that time. Even later, among those who married in or after 1940, only one marriage in 20 resulted in divorce after 20 years’ exposure. By contrast, one-third of couples who married in 1975 had divorced by 1995, illustrating how the period of our analysis predates the sharp increase in divorce that took place in the late 1960s and 1970s (Statistics Sweden 2008, p. 381).

Figure 3 Kaplan–Meier marriage survival estimates by marriage cohort as a function of time since marriage for couples married 1880–1954 in Västerbotten County, Sweden

Source: POPLINK database (Demographic Data Base, Umeå University).

Figure 4 complements the picture by displaying non-parametric Kaplan–Meier estimates of risk of divorce as a function of time since marriage for men who married in 1880–1954 (and were followed until 1960). We distinguished men who married in 1880–1919, 1920–29, and 1930–54 and focused on the differentiation according to SES across marriage cohorts. Among men who married in 1880–1919 (panel (a)), those categorized as middle class (e.g. white-collar professionals) were most likely to divorce. We did not find that the highest-SES men (belonging to the elite) exhibited high rates of divorce compared with working-class men in this cohort. These results are similar to those found by Van Poppel (1997) and Kalmijn et al. (2011) for the Netherlands.

Figure 4 Kaplan–Meier marriage survival estimates by socio-economic status for men married 1880–1954 in Västerbotten County, Sweden

Source: As for Figure 3.

For men who married in the 1920s (panel (b)), the pattern differs from that of the previous cohort in that all SES groups except farmers caught up with the middle class. This is noticeable among both skilled and unskilled workers, who together constituted approximately 50 per cent of the married male population. The coordinated change in behaviour of this large segment of the male population was thus an important factor contributing to the increase in divorce in the 1940s and early 1950s. Moreover, men marrying in or after 1930 also contributed to the increase during this period: panel (c) confirms that in this cohort all SES groups except farmers experienced similar divorce risks to the middle class. The graph also shows higher divorce risks among working-class men, particularly after 15–20 years of marriage, although these differences are not statistically significant. Thus, there are indications of a shift from a positive association between SES and divorce among men in the earliest marriage cohort to a negative association among men in the latest cohort; this aligns with the clearly negative gradient found for men in studies on divorce in Sweden after 1970.

Turning to the results for women, much the same pattern of change as described for men emerges in Figure 5, which shows Kaplan–Meier estimates of divorce intensities as a function of time at risk for women by SES. For women in the earliest marriage cohort, there was a positive SES gradient in divorce. Women in the middle class and elite strata were more likely to divorce than those in other SES groups. The results for women also confirm that farmers were least likely to divorce. In later cohorts, women who were skilled workers were considerably more likely to divorce than other groups. Patterns among women who married in 1930–1954 resembled those of the previous cohort, apart from a greater propensity to divorce among unskilled workers. For women, as for men, there was a shift in the SES gradient in risk of divorce, with the largest increases occurring among the working class. This suggests that socio-economic growth among women as well as men was an integral part of increasing divorce rates during this period. It is also worth noting that the results for women suggest that having a recorded occupation, as an indication of economic activity (i.e. independence), was significantly related to increased risk of divorce in comparison to non-employment, and this held for all SES groups except farmers, who were distinctly different throughout much of the analysis.

Figure 5 Kaplan–Meier marriage survival estimates by socio-economic status for women married 1880–1954 in Västerbotten County, Sweden

Source: As for Figure 3.

Turning to the multivariate results, Table 2 shows the results from Cox regressions by marriage cohort. Stepwise modelling provides evidence of how different variables modulate the association between SES and divorce. Model 1 is a baseline model for the man’s SES and divorce. Model 2 includes covariates of relevance for the association between SES and divorce (wife’s age at marriage, number of children at time t, and geographical context in terms of parish, with a distinction made between urban and rural (also at time t)). In Model 3, we estimated the impact of the husband’s and wife’s SES net of each other, controlling for potentially confounding factors. Of note, the association between SES and divorce was robust, irrespective of model specification.

Table 2 Cox regressions: relative hazard of divorce for couples married in 1880–1954 in Västerbotten County, Sweden

	Marriage cohort
	1880–1919			1920–29			1930–54
Variables	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3
Socio-economic position: man
Unskilled worker	ref	ref	ref	ref	ref	ref	ref	ref	ref
Skilled worker	1.37	1.19	1.08	1.05	0.93	0.87	1.29	1.13	1.10
Farmer	0.21***	0.24***	0.25***	0.28***	0.37***	0.40***	0.52***	0.68*	0.74
Middle class	3.24***	2.54***	2.17***	0.87	0.80	0.77	1.04	0.91	0.90
Elite	0.91	0.64	0.58	1.00	0.84	0.85	0.89	0.76	0.78
Socio-economic position: woman
No occupation	–	–	0.92	–	–	1.06	–	–	0.88
Unskilled worker	–	–	ref	–	–	ref	–	–	ref
Skilled worker	–	–	2.57*	–	–	3.52***	–	–	2.11***
Farmer	–	–	0.75	–	–	0.75	–	–	0.47***
Middle class and Elite	–	–	3.46***	–	–	1.77	–	–	1.19
Age at marriage
<20	–	ref	ref	–	ref	ref	–	ref	ref
20–24	–	0.57	0.54	–	0.39***	0.40***	–	0.47***	0.45***
25–29	–	0.47*	0.43*	–	0.26***	0.26***	–	0.26***	0.25***
30–34	–	0.44*	0.35**	–	0.13***	0.12***	–	0.22***	0.19***
35–39	–	0.26*	0.22**	–	0.28**	0.22***	–	0.15***	0.13***
40+	–	0.19*	0.15**	–	0.11**	0.07***	–	0.17***	0.13***
Number of children
Childless	–	ref	ref	–	ref	ref	–	ref	ref
One child	–	0.87	0.86	–	0.84	0.81	–	0.78	0.77
Two children	–	0.49*	0.49*	–	0.53*	0.52*	–	0.31***	0.31***
Three or more children	–	0.26***	0.26***	–	0.50**	0.52**	–	0.28***	0.29***
Parish
Skellefteå rural	–	ref	ref	–	ref	ref	–	ref	ref
Skellefteå urban	–	1.13	1.18	–	2.11**	2.12**	–	1.81***	1.74***
Umeå rural	–	2.47***	2.52***	–	1.59*	1.56*	–	1.66***	1.60***
Umeå urban	–	1.91**	1.84*	–	2.20***	2.02***	–	2.81***	2.64***
Akaike information criterion	2,478.8	2,431.1	2,410.9	2,855.7	2,813.8	2,793.3	6,724.3	6,548.5	6,515.1

*p < 0.05; **p < 0.01; ***p < 0.001.

Note: ref indicates the reference category.

Source: As for Table 1.

The estimates from Model 1 indicate a positive association between (the man’s) SES and divorce among those who married in 1880–1919, with the middle class significantly more likely to divorce than other groups, particularly farmers. For the later marriage cohorts (1920–54) there was no such positive association. Farmers always display a significantly lower divorce risk compared with unskilled workers (and with other SES groups, as the result is robust irrespective of reference category). In Model 2 the addition of covariates, such as woman’s age at marriage, number of children, and parish of residence, adjusts the coefficients downwards but does not affect the results in any meaningful way. All covariates render results in line with expectations. Age at marriage is negatively associated with divorce, as is number of children. Individuals living in the most rural parish (Skellefteå rural) showed significantly lower divorce risks than those living in other parishes.

Model 3’s coefficients show that the woman’s SES also mattered for divorce; their magnitudes indicate the importance of women’s economic independence, even in a context where married women’s employment was still limited and far from universal. The addition of this variable to the model further adjusts the coefficients on the association between the husband’s SES and divorce. The negative coefficient for farmers belonging to the 1930–54 cohort becomes insignificant, but otherwise the results do not change. Woman’s SES is interesting in itself. The coefficients confirm what can be inferred from Figure 5: namely, a positive association between SES and divorce for the 1880–1919 cohort which then becomes insignificant among the later cohorts (1920–54). There is also evidence that skilled workers were more likely to divorce than women with no occupation recorded. The results for woman’s SES support not only the claim that a process of socio-economic growth was taking place but also the economic independence hypothesis. For women who married in 1880–1954, having an occupation and work experience in the growing industrial sector had a bearing on their economic independence and decision-making in family transitions, including divorce. Among the early cohorts, women with the highest SES experienced the highest risk of divorce, but this changed with time as (skilled) working-class women became those most likely to divorce.

Although we did not analyse the impact of different combinations of men’s and women’s SES in this study, the similar associations between men’s and women’s SES and divorce might be seen as indications of homogamy based on SES. This was not, however, the case. When we assessed the degree of homogamy among couples for which a recorded occupation at marriage for the woman was available (approximately 60 per cent of the sample), we found only a modest correlation between spouses’ SES (about 0.2), irrespective of marriage cohort. Although women belonging to the highest socio-economic strata based on own occupation were few, we found that about half of the women in the middle class and elite categories were heterogamously married to men with lower SES. We therefore interpret the associations between SES and divorce for men and women, respectively, in a similar manner.

In sum, the regression results confirmed the results of our non-parametric analysis. Additional analysis, where we estimated the full model (Model 3) for all marriage cohorts pooled and included an interaction between SES and cohort, confirmed the temporal dimension of the shifting association between SES and divorce across cohorts (see Table 3). When we set the first cohort (1880–1919) and middle class as the reference categories for men and women, we found strong, positive interactions for all SES categories among those who married after 1920. This illustrates a process of socio-economic growth and diffusion of divorce impacting on broader layers of the population, a process that coincided with an increase in divorce risks at the aggregate level.

Table 3 Cox regressions: relative hazard of divorce for men and women married in 1880–1954 in Västerbotten County, Sweden, including interaction between marriage cohort and SES

Variables	Men	Women
Marriage cohort
1880–1919	ref	ref
1920–29	0.69	0.62
1930–54	0.74	0.60
Socio-economic position: man
Unskilled worker	0.40***	–
Skilled worker	0.46**	–
Farmer	0.11***	–
Middle class	ref	–
Elite	0.25**	–
Socio-economic position: woman
No occupation	–	0.23***
Unskilled worker	–	0.21***
Skilled worker	–	0.58
Farmer	–	0.13***
Middle class and Elite	–	ref
Marriage cohort × Socio-economic position: man
1920–29 × Unskilled worker	3.54***	–
1920–29 × Skilled worker	2.61**	–
1920–29 × Farmer	5.08***	–
1920–29 × Elite	4.25**	–
1930–54 × Unskilled worker	2.86***	–
1930–54 × Skilled worker	2.79***	–
1930–54 × Farmer	6.81***	–
1930–54 × Elite	3.51*	–
Marriage cohort × Socio-economic position: woman
1920–29 × No occupation	–	3.09**
1920–29 × Unskilled worker	–	3.29**
1920–29 × Skilled worker	–	3.79*
1920–29 × Farmer	–	4.03**
1930–54 × No occupation	–	3.27***
1930–54 × Unskilled worker	–	4.24***
1930–54 × Skilled worker	–	3.15*
1930–54 × Farmer	–	3.22**
Akaike information criterion	12,968.2	12,988.6

*p < 0.05; **p < 0.01; ***p < 0.001.

Note: ref indicates the reference category. Models include controls for woman’s age at marriage, number of children, and parish.

Source: As for Table 1.

Concluding discussion

In this study we set out to test Goode’s theory of socio-economic growth. This states that the association between SES and divorce was positive in low-divorce contexts in the past, but that it was reversed with industrialization and modernization as the economic and normative barriers to divorce were overcome for many people. Although many studies have documented a negative association between SES and divorce after 1970, few tests have been made of Goode’s theory for historical contexts and longer time spans. Evidence is thus limited as to whether there was a positive SES gradient in divorce to begin with, whether it was reversed over the course of industrialization, and, if so, when this shift occurred. To fill this gap, we investigated the association between SES and divorce using individual-level longitudinal data for cohorts of men and women who married in Västerbotten County, northern Sweden, between 1880 and 1954. We analysed a long historical period during which Sweden underwent industrialization and considerable change in terms of socio-economic modernization. Sweden was still best characterized as a low-divorce country; it was not yet the forerunner in terms of family dynamics and gender equality as it is now known.

We found indicative support for Goode’s predictions of changing socio-economic patterns of divorce from positive to negative as divorce rates increased over the course of industrialization and modernization. The results imply the diffusion of a cultural innovation and are in line with those found for the Netherlands and Flanders (Kalmijn et al. 2011). Our belief, which is of relevance for divorce (and other behavioural innovations), is that SES encompasses more than simply economic resources. We found that, among men, it was not the economic elite but rather the educated middle class who led the way for divorce among those who married before 1920. These early adopters did not belong to the privileged upper class and were neither landed gentry nor capitalists in the sense of being large business owners. They were teachers, lawyers, bank tellers, clerks, and editors or worked in another high-status occupation (for that time) requiring a secondary or tertiary education. This indicates that factors specific to the middle class, other than economic resources, played an important role in determining divorce. It is possible that for this group, unlike the elite, the cultural and social capital acquired through education and profession was more important than economic capital. A potential mechanism behind the higher divorce rates among the middle class was that these men, who based their status primarily on human and cultural capital rather than on economic capital in terms of assets, held more individualistic or modern values than did men belonging to the elite. In comparison, the social context experienced by wealthy business proprietors and other members of the elite was likely to be more conservative and less accepting of divorce, and thus the potential scandal of a divorce as well as the costs related to the division of household assets served as a disincentive for them. This is in line with Rogers’ theory of diffusion of innovations (Rogers 1962), which emphasizes the role of human capital in the (early) adoption of innovations, be they social or technological. Our findings, as well as those presented by Kalmijn et al. (2011), indicate that some modification of Goode’s theory might be warranted, in that cultural rather than economic capital (or a combination of the two) was the important resource needed for overcoming the barriers to divorce in low-divorce settings in the past.

We found a change, although not an altogether significant reversal, in the association between SES and divorce among those who married in the 1920s and thereafter, which represented a substantial growth in divorce across socio-economic groups (i.e. what Goode would call ‘socio-economic growth’ in divorce). In the latest marriage cohort, the highest estimated divorce risk was faced by working-class men, showing a burgeoning tendency towards a negative SES gradient, while for women, high risks were faced by skilled workers across cohorts although risks were highest among the middle class and elite among women married prior to the 1920s. The sharp increase in the divorce risk for working-class men and women coincided with industry and services crowding out agriculture as the dominant sector of the Swedish economy.

Farmers consistently faced the lowest divorce risks. This may be explained not only by the group’s identity but also by the changing function of the family farm over time. Some would argue that this group was more traditional, but the division of labour along gendered lines was probably more pertinent because it made farming highly dependent on the joint efforts of an intact couple well into the 1930s. Thereafter, Swedish agriculture was highly regulated through national production goals, alongside goals for increasing unit size, mechanization, and productivity. Agricultural change and industrialization combined meant that farmers formed an increasingly select group that tended to turn their farms into large business enterprises, which were generally difficult to divide between partners in the case of divorce (Goode 1963, p. 19; Flygare 2008; Martiin 2012).

Agriculture represented more than 50 per cent of the labour force until 1910, after which it declined to 16 per cent by 1960. At the same time, industry expanded from 32 to 40 per cent of the labour force (both sexes), while services increased from 19 to 44 per cent between 1910 and 1960 (Stanfors and Goldscheider 2017, p. 190). With industrialization, wage work in factories involved a growing segment of the working class, drawing peasants out of agriculture and into industry. This prompted a diffusion of innovations regarding family life, including divorce. Family size diminished in association with industrialization and urbanization. Moreover, the role of marriage had changed, although marriage itself was still widespread (Cherlin 2012). Parental control over partner choice had diminished and the role of the family was no longer that of an economic production unit but rather one of consumption, for fulfilling individual rather than collective needs. For most young men and women—many of whom were wage workers in manufacturing and service occupations—marriage was based primarily on love and companionship. It is within this context that we find support for Goode’s conjectures regarding the socio-economic growth of divorce.

There are, however, limitations to keep in mind regarding our study. First, we should not interpret results as causal impacts on divorce. Although reverse causality can be ruled out, there may be unobservable, individual-level factors associated with men and women who divorce for which we cannot account. Second, we note that although occupational status was meant to capture SES and women’s economic independence and to proxy their roles in the divorce decision-making process, these concepts should preferably be validated by other measures. Such measures are, however, hard to come by for historical contexts. An important next step in this line of research is to model women’s economic independence explicitly in the divorce process, for example by wife’s income. Further, we did not model relationship quality and we explored only formal marriages and divorces. We also acknowledge that our data and study design introduced stayer bias, although our data and variables had several advantages compared with materials used in previous studies.

We believe it was important to extend the historical scope further back—to the 1880s—thereby capturing the entire period of industrialization and modernization in Sweden including the early phase of industrialization, which was at the core of Goode’s conjectures. Furthermore, recognizing the diffusion of divorce behaviour from the higher social strata to the rest of the population is important for understanding the divorce transition more generally. The higher classes, if not the true elite, were innovators in the divorce transition, although the working class adopted new standards quite rapidly once divorce became accessible and more acceptable. In Sweden, the spread of divorce throughout the population contributed to the growth in divorce from the mid-1940s through to the early 1950s. This was the first marked increase in an otherwise slowly upward-trending divorce rate, but, unlike the second and major increase that followed in the late 1960s and early 1970s, it was unrelated to legislative reform covering divorce and comprehensive state support for families and working parents. We interpret this as evidence of divorce becoming less costly, more acceptable, and thus a more ready option than before for those wanting to end their marriage.

Worth noting is that the first marked increase in the divorce rate in Sweden took place in a context where married women’s labour force participation was low (e.g. lower than in the US). As early as the first decades of the twentieth century, most young single women worked outside the family household but left the labour force when they married and had children. Estimates show that almost all women in the early 1900s and more than half of all working women in 1960 left the labour force on marriage. After the Second World War it became increasingly common for mothers to re-enter the labour force once their children required less care (Stanfors 2014).

We found evidence supporting the economic independence hypothesis. There were significant impacts of wife’s SES as indicated by class, based on own occupation, with positive parameters for all economically active women apart from those categorized as farmers. For women who married after 1920, in contrast to those who married earlier, it was not the middle class and elite who exhibited the highest risks of divorce but rather the skilled working class whose divorce risks were by far the highest compared with those of women with no recorded occupation. Being economically active strongly increased the risk of divorce among women, particularly at the end of the study period. This points to the importance of the economic independence hypothesis in the context of a traditional division of labour and role specialization among spouses. It also suggests that the slowly changing economic roles of women during the first half of the twentieth century mattered for divorce. During the early stages of the divorce transition, when divorce became an option for broader groups, women able to support themselves through market work could act more independently. The most likely mechanism for this was that working-class women in the labour market, or with previous work experience, faced fewer economic and normative constraints in seeking divorce.

Furthermore, the reversal of the association between SES and divorce, driven by divorce extending to the working class, coincided with an increased emphasis on individualism and egalitarianism along class and gender lines, through law reform regarding marriage and divorce. It also occurred alongside the early development of the Swedish welfare state, which offered improved economic security for individuals engaged in wage labour, thereby reducing the individual’s dependence on the family for risk-pooling and income security in the event of health problems, unemployment, and other shocks to individual economic security. This and the fact that women’s economic independence increased divorce risk, other things being equal, should be assessed in context: one which, although emphasizing as of 1920 the partners’ equal contributions to the household, did not in practice extend income support to women in the case of divorce (dependent children did, however, receive an allowance, and primary education was free of charge). As part of the change during the 1920s, women gained suffrage in Sweden’s general election in 1921. The role of the welfare state as we know it today was limited during this period of study. Public expenditure was growing, more markedly during the post-war years than before. There was, however, an early emphasis on women’s economic independence, in that single mothers were expected to be able to support themselves and their children through market work rather than through alimony or public support. Over time, the welfare state became increasingly important for women in general, and single mothers in particular, and was an important determinant of divorce in Sweden after 1960 (Stanfors et al. 2020). Our study captured the importance of socio-economic growth and the increase in economic and social opportunities for divorce for many before the modern welfare state came into being.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Glenn Sandström is based in the Department of Historical, Philosophical and Religious Studies and the Centre for Demographic and Ageing Research (CEDAR), Umeå University, Sweden. Maria Stanfors is based in the Department of Economic History and the Centre for Economic Demography, Lund University, Sweden.

2 Please address all correspondence to Maria Stanfors, Department of Economic History and the Centre for Economic Demography, Lund University, PO Box 7080, 220 07 Lund, Sweden; or by Email: [email protected].

3 Previous versions of this paper were presented at the European Network for the Sociological and Demographic Study of Divorce in Florence, October 2019, and at the Social Science History Association’s Annual Meeting in Chicago, November 2019.

4 This work was supported by Handelsbankens forskningsstiftelser under grant number P17:0184 and by the Marcus and Amalia Wallenberg Foundation, grant number MAW 2019.0029.

Appendix

Table A1 Cox regressions: relative hazard of divorce for couples married in 1930–54 compared with those married in 1930–45 (i.e. under observation for a minimum of 15 years)

	Marriage cohort
	1930–54	1930–45
Socio-economic position: man
Unskilled worker	ref	ref
Skilled worker	1.10	1.13
Farmer	0.74	0.63*
Middle class	0.90	0.87
Elite	0.78	0.89
Socio-economic position: woman
No occupation	ref	ref
Unskilled worker	1.14	1.18
Skilled worker	2.40***	2.45***
Farmer	0.54**	0.60*
Middle class and Elite	1.35	1.42
Age at marriage (woman)
<20	ref	ref
20–24	0.45***	0.46***
25–29	0.25***	0.22***
30–34	0.19***	0.18***
35–39	0.13***	0.12***
40+	0.13***	0.09***
Number of children
Childless	ref	ref
One	0.77	0.70*
Two	0.31***	0.29***
Three or more	0.29***	0.27***
Parish
Skellefteå rural	ref	ref
Skellefteå urban	1.74***	1.73***
Umeå rural	1.60***	1.62***
Umeå urban	2.64***	2.70***
Akaike information criterion	6,515.1	5,661.8

*p < 0.05; **p < 0.01; ***p < 0.001.

Source: POPLINK database, Demographic Data Base, Umeå University.