Introduction
Along with the (re)negotiation of ageing and what a good life in older age could be like in the light of the second demographic change (van Dyk, Reference van Dyk2015), relationships and sexuality in older age, including the topic of repartnering, have become increasingly relevant research topics (Bamler, Reference Bamler, Buchen and Maier2008; Gildemeister, Reference Gildemeister, Buchen and Maier2008; Koren, Reference Koren2015). Against the backdrop of demographic and normative changes, this article assumes that repartnering in older age is and will continue to gain relevance (Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004). By theoretically framing repartnering in older age as a lifecourse transition which is embedded within structural and processual dimensions of gendered ageing (Krüger and Levy, Reference Krüger and Levy2001), this article poses the following question: How do later-life relationships form in relation to gender- and ageing-specific structural and processual dimensions?
On the one hand, and due to the rise in life expectancy, relationships and especially marriages are lasting longer and often end with the death of one partner, in most cases the male partner – when considering different-sex relationships (Bamler, Reference Bamler, Buchen and Maier2008). On the other hand, a pluralisation of family arrangements took place in Germany and other so-called Western societies (Beck-Gernsheim, Reference Beck-Gernsheim2002; Brückner and Mayer, Reference Brückner and Mayer2005; Fasang, Reference Fasang, Blanchard, Bühlmann and Gauthier2014). This pluralisation expresses itself in changing relationship practices like serial relationship biographies entailing higher divorce rates at a constantly rising average age in middle life (destatis, 2019a) or non-institutionalised respectively non-normative relationships in older age, like living apart together (Beck-Gernsheim, Reference Beck-Gernsheim2002; Mauritz and Wagner, Reference Mauritz and Wagner2021). Based on these developments, sociologists proclaim that marriage and the nuclear family are gradually losing their traditional value as a union for the entire lifecourse (Beck-Gernsheim, Reference Beck-Gernsheim2002; Lenz, Reference Lenz2009, Reference Lenz, Schöer, Stauber, Walther, Böhnisch and Lenz2013; Kuchler and Beher, Reference Kuchler, Beher, Kuchler and Beher2014). This norm shift becomes especially relevant when considering older age, since the generation that was socialised after the late 1960s in the context of these new relationship norms and practices is now becoming older (Krüger and Levy, Reference Krüger and Levy2001; Beck-Gernsheim, Reference Beck-Gernsheim2002; Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004; Bamler, Reference Bamler, Buchen and Maier2008).
In the following, I present a gender- and ageing-sensitive perspective on lifecourse transitions as the theoretical framework underlying my analysis. I hypothesise that repartnering is more likely for men than for women, that for the older (one is getting) repartnering is less likely and that these effects are becoming more intense with higher age. Additionally, I hypothesise that more positive attitudes towards ageing and being divorced compared to being widowed promotes repartnering. After discussing previous key findings, I present the method and data – hybrid panel regressions based on the German Ageing Survey (1996–2017) – and the findings. They confirm the intersectional effect of gender and ageing, but ageing does only influence repartnering processually. While the hypothesis regarding attitudes towards ageing can be accepted, the results show that in the short term the widowed are more likely to repartner than the divorced, while in the long term the divorced are more likely to repartner than the widowed. In the final section, I discuss the results and draw a conclusion.
Theoretical framework and hypotheses: repartnering as a lifecourse transition in older age
Sociologically speaking, the lifecourse is defined as a socially institutionalised set of rules structuring individual participation ‘movements through social space’ (Krüger and Levy, Reference Krüger and Levy2001: 158; see also Mayer, Reference Mayer, Schäfers and Zapf1998) in a certain society at a certain point in historical time (Kohli, Reference Kohli2007). This conceptualisation refers to a sequential – not necessarily linear (Abbott, Reference Abbott1992) – alternation of ‘participational changes’ (Krüger and Levy, Reference Krüger and Levy2001: 158) or transitions. They vary systematically and are thereby embedded in social and institutional structures as well as temporal norms while also (re)producing them (Mayer, Reference Mayer, Schäfers and Zapf1998; Krüger and Levy, Reference Krüger and Levy2001; Becker, Reference Becker, Walther, Stauber, Rieger-Ladich and Wanka2020). Therefore, the lifecourse perspective allows for a focus on structural and processual dimensions of lifecourse transitions. By ‘structural’, this paper is referring to differences or similarities in the social position between individuals at a certain point in time, whereas ‘processual’ relates not only to the individual social position but also to the temporal dynamic of social positions one individual takes in over time.
Since the lifecourse emerges in different, yet overlapping, social spaces, this perspective has been developed in an inter-disciplinary context and used intensively as a heuristic in the sociology of the family (Krüger and Levy, Reference Krüger and Levy2001; Levy, Reference Levy, Levy and Widmer2013; Aeby and Gauthier, Reference Aeby, Gauthier, Castrén, Cesnuityte, Crespi, Gauthier, Gouveia, Martin, Mínguez and Suwada2021). From this perspective, the formation of a relationship can – as any other participational change in relationship status – be defined as a familial lifecourse transition (Aeby and Gauthier, Reference Aeby, Gauthier, Castrén, Cesnuityte, Crespi, Gauthier, Gouveia, Martin, Mínguez and Suwada2021), as it describes the status change from participating in singlehood to living in a relationship.
How such transitions are embedded structurally and processually in later life can be conceptualised using the cumulative advantage/disadvantage model (Dannefer, Reference Dannefer2003) linking the somewhat stable, structural (re-)production of social inequalities to the dynamic, processual temporality of cumulation and ageing. This concept describes the cumulation of both resources and capacities based on differences in (institutional) opportunity structures – over the lifecourse in an inequality-generating and -stabilising manner. Thereby social positions change and possibly intensify over the lifecourse.
This accumulation of resources and capacities, in turn, becomes relevant for the emergence of opportunities to participate in the partner market, ergo to both meet and mate with new partners. The social (e.g. the family) and institutional (e.g. the workplace) settings in which individuals participate are socially organised according to social positions. The opportunity structures of the partner market emerge within these settings and are therefore also structured by one's social position (Kalmijn and Flap, Reference Kalmijn and Flap2001; Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004). At the same time, the numerical distribution of certain structural characteristics, e.g. gender, education, age or ethnicity, is essential for the availability and thereby the attractiveness of potential partners within social and institutional settings (Klein and Stauder, Reference Klein, Stauder, Niephaus, Kreyenfeld and Sackmann2016; Corti and Scherer, Reference Corti and Scherer2021). This numerical availability might in turn lead to changes in individual preferences, which are relevant for the mating processes, e.g. in changing norms of who is considered to be an appropriate or attractive partner (Corti and Scherer, Reference Corti and Scherer2021). Accordingly, the likelihood to repartner is interrelated with one's social position.
Hypotheses
In the sense of linked lives (Elder et al., Reference Elder, Johnson, Crosnoe, Shanahan, Mortimer and Johnson2003), the cumulation of advantages and disadvantages is tied to other individuals, especially to the family or relationship form (Dannefer, Reference Dannefer2003; Aeby and Gauthier, Reference Aeby, Gauthier, Castrén, Cesnuityte, Crespi, Gauthier, Gouveia, Martin, Mínguez and Suwada2021). Within families, a ‘gendering of life courses’ (Levy Reference Levy, Levy and Widmer2013) emerges due to interrelated mechanisms like gender-specific decisions in the occupational field and the unequal division of labour within couples. While men tend to focus on their participation in the occupational field, women's participation in this field interferes with commitments in other fields, e.g. child care, care for older family members or domestic work in the familial field. This leads to systematic gender differences in various fields of the lifecourse, resulting in gender-typical lifecourses and reproducing gender as a socially constructed category (Krüger and Levy, Reference Krüger and Levy2001; Levy, Reference Levy, Levy and Widmer2013). Linking the impact of gendered lifecourses with the distribution of advantages and disadvantages, it can be assumed that men especially accumulate more advantages concerning economic and educational resources. Women, however, accumulate more advantages concerning social capacities over their respective lifecourses (Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004). This unequal cumulation of advantages and disadvantages thus becomes relevant to the partner market in older age: men are not only more admired partners but are also less capable of maintaining social relationships without a partner (Davidson, Reference Davidson2002; Barrett, Reference Barrett2005; Koren, Reference Koren2016). These considerations lead to the following hypothesis:
• Hypothesis 1: Men are more likely to form new relationships than women.
When regarding repartnering in later life, it is pivotal to theorise and analyse mating in older age with an ageing-sensitive perspective. In this article, ageing is conceptualised in its multi-dimensionality or complex ‘Doppelcharakter’ (van Dyk, Reference van Dyk2015: 6). It highlights both the inherent processual character of becoming older and the structural character which marks someone as old in contrast to being young. This understanding allows me to include ageing in a more complex way and to link it systematically to the processual and structural dimensions of lifecourse transitions. Considering the partner market, we can assume that repartnering is less likely for older than for younger individuals (structural) and less likely the older an individual is getting (processual). As individuals become older, both the embeddedness in favourable opportunity structures like the workplace and the likeliness to become a potential partner weaken due to the ageing-specific accumulation of disadvantages (Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004; Klein and Stauder, Reference Klein, Stauder, Niephaus, Kreyenfeld and Sackmann2016). The concrete mechanisms behind this assumption will be further elaborated, but based on it, the following hypothesis emerges:
• Hypothesis 2: The older a person is (getting), the less likely repartnering becomes.
In order to better understand the interrelated structural and processual dimensions of repartnering in later life, the influences of ageing and gender on repartnering are linked in an intersectional manner. Silke van Dyk proposes to link gender and ageing theoretically to also focus on interdependencies by describing ‘gendered ageing and respectively age-inherent gender relations’ (van Dyk, Reference van Dyk, Denninger and Schütze2017: 44, own translation). This intersection is pivotal for repartnering in older age, because one of the consequences of gendered lifecourses is an unbalanced partner market in older age: there are proportionally more women than men, even though the distribution is increasingly adjusting (Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004; Nowossadeck and Engstler, Reference Nowossadeck and Engstler2013; destatis, 2019b). The mechanisms behind this are firstly that women tend to live longer than men (Hoffmann et al., Reference Hoffmann, Romeu Gordo, Nowossadeck, Simonson and Tesch-Römer2017; destatis, 2019c) and secondly that there is an average age difference in different-sex relationships, with the man being notably older than the woman (Backes and Wolfinger, Reference Backes, Wolfinger, Blättel-Mink and Kramer2009; Brown and Shinohara, Reference Brown and Shinohara2013). This leads to a higher likelihood of becoming a widow than becoming a widower (Bamler, Reference Bamler, Buchen and Maier2008). Accordingly, there is a bigger proportion of single women than single men in older age, which in turn raises men's and restricts women's chances to repartner (Klein and Stauder, Reference Klein, Stauder, Niephaus, Kreyenfeld and Sackmann2016). Thirdly, men are presumed to be more attractive in older age than women (Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004; van Dyk, Reference van Dyk, Denninger and Schütze2017). The ‘double standard of ageing’ (Sontag, Reference Sontag, Allman and Jaffe1972: 286) describes this unequal assessment of physical attractiveness. Thus, normative expectations consolidate these three mechanisms: it is more common that men date younger women than that women date younger men (Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004). These intersectional considerations result in the following hypothesis:
• Hypothesis 3: With rising age, men are increasingly more likely than women to repartner.
Since cultural norms are considered to be a pivotal factor in the explanation of meeting and mating, it is important to not only consider ageing but also the normative context within which different cohorts were socialised (Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004). In Germany, as in other Western cultures, later-born cohorts grew up in a context of more liberal relationship norms – like the decreasing importance of marriage – and practices – like serial relationship biographies and higher divorce rates (Beck-Gernsheim, Reference Beck-Gernsheim2002; Lenz, Reference Lenz2009, Reference Lenz, Schöer, Stauber, Walther, Böhnisch and Lenz2013; Kuchler and Beher, Reference Kuchler, Beher, Kuchler and Beher2014). Accordingly, later-born cohorts had, in comparison with earlier-born cohorts, the possibility to accumulate relationship capacities outside a traditional or even restrictive context (Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004). Therefore, I hypothesise the following:
• Hypothesis 4: Later-born cohorts are more likely to repartner compared to earlier-born cohorts.
Additionally, there is a close tie between the formation of new relationships and physical attractiveness – which is normatively equated with youth. In combination with the normative devaluation of ageing, this results in a taboo surrounding sexuality and relationships in later life (Bamler, Reference Bamler, Buchen and Maier2008; Buchen and Maier, Reference Buchen, Maier, Buchen and Maier2008; van Dyk, Reference van Dyk, Denninger and Schütze2017). Thus repartnering appears as a non-normative lifecourse transition in older age (Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004). Since older people might internalise such devaluative or even ageist attitudes towards ageing (Bytheway, Reference Bytheway1995), they might also have dismissive attitudes towards sexuality and relationships in later life. These attitudes could prevent them from developing the capacity to actively undermine the social norm of staying single in older age (Gildemeister, Reference Gildemeister, Buchen and Maier2008; Koren, Reference Koren2015). However, more positive attitudes towards ageing could increase the likelihood of resisting such norms. Therefore, the fifth hypothesis is:
• Hypothesis 5: Positive attitudes towards ageing promote repartnering.
Based on the assumption that lifecourse transitions are linked to each other (Aeby and Gauthier, Reference Aeby, Gauthier, Castrén, Cesnuityte, Crespi, Gauthier, Gouveia, Martin, Mínguez and Suwada2021; Bischoff et al., Reference Bischoff, Franke and Wanka2021), an additional aspect can influence repartnering in later life: the transition into singlehood (Poortman, Reference Poortman2007). Thus, a differentiation between various origins of singlehood is pivotal (Brown et al., Reference Brown, Lin, Hammersmith and Wright2018).
The initiation of a divorce or separation is usually a deliberate act by at least one of the partners, while the other partner might not have decided whether to continue or end the relationship (Sweeney, Reference Sweeney2002; Lenz, Reference Lenz, Schöer, Stauber, Walther, Böhnisch and Lenz2013). Due to this difference, emotional reactions can include mourning, anger and feelings of guilt or shame, but a separation or divorce can also serve as a liberating momentum (Lin et al., Reference Lin, Brown, Wright and Hammersmith2019).
The form of singlehood which can be understood to be opposite to divorce or separation is the death of a partner, as it does not happen deliberately or intentionally (Lenz, Reference Lenz, Schöer, Stauber, Walther, Böhnisch and Lenz2013). On the contrary, both partners wanted to maintain their relationship and spend their futures together (Wolf, Reference Wolf, Schöer, Stauber, Walther, Böhnisch and Lenz2013). As a result, the relationship is often carried on in spirit and the partner is remembered in an idealised way (Hockey et al., Reference Hockey, Penhale and Sibley2001; Lopata, Reference Lopata2002). Commonly, and contrary to divorce or separation, a process of grief, socially accompanied by explicit mourning rituals like the funeral, follows the death of a partner (Wolf, Reference Wolf, Schöer, Stauber, Walther, Böhnisch and Lenz2013).
As there are basal differences between various forms of singlehood and based on the significance the former partner has after transitioning into singlehood, the last hypothesis states the following:
• Hypothesis 6: The death of a partner deteriorates the likelihood of repartnering – as opposed to divorce or separation.
Previous studies on repartnering in older age
Repartnering in older age has not been a research interest for a long time, even though there still is a longing for love, closeness, intimacy and sexuality in older age (Bamler, Reference Bamler, Buchen and Maier2008; Watson and Stelle, Reference Watson and Stelle2011; Fileborn et al., Reference Fileborn, Thorpe and Hawkes2015). However, in recent years there has been an increasing number of studies concerning repartnering in older age, especially in so-called Western countries (Koren, Reference Koren2015). These studies can partly be transferred to the German context, due to the similar societal circumstances.
It is a general observation that only a small share of the older population is actively searching for a partner or repartnering, even though their share is rising (Bulcroft and Bulcroft, Reference Bulcroft and Bulcroft1991; Brown and Shinohara, Reference Brown and Shinohara2013; Wu et al., Reference Wu, Schimmele and Ouellet2015). Many studies not only investigate repartnering itself, but also attitudes towards repartnering.
The gender-specific findings shed light on the (mostly) female narrative of neglecting the idea of repartnering (Talbott, Reference Talbott1998; Davidson, Reference Davidson2001) and the practice of remaining single (Poortman and Hewitt, Reference Poortman and Hewitt2015; Schimmele and Wu, Reference Schimmele and Wu2016; Brown et al., Reference Brown, Lin, Hammersmith and Wright2018). The empirical explanation often lies in the wish to maintain autonomy and self-confidence, and a fear of losing (re)gained independence due to having to do care- or housework again (Cooney and Dunne, Reference Cooney and Dunne2001; Davidson, Reference Davidson2002; De Jong Gierveld, Reference De Jong Gierveld, Davidson and Fennel2002; Moorman et al., Reference Moorman, Booth and Fingerman2006; Watson and Stelle, Reference Watson and Stelle2011; Koren, Reference Koren2016). Likewise, studies reveal a (mostly) male narrative and practice: men wish to and tend to repartner due to frustration (divorcees), grief (widowers), and feelings of social isolation and loneliness (Talbott, Reference Talbott1998; Davidson, Reference Davidson2001; Carr, Reference Carr2004; Bennett et al., Reference Bennett, Arnott and Soulsby2013). In traditional relationships, men were not responsible for socialising and have therefore fewer close relationships. By repartnering, they can elude their loneliness and experience emotional closeness and intimacy (Cooney and Dunne, Reference Cooney and Dunne2001; Davidson, Reference Davidson2001, Reference Davidson2002; Koren, Reference Koren2015, Reference Koren2016). These gender-specific preferences – that older single men tend to wish and actively search for a new partner more than older single women – are recurring patterns in previous studies.
Ageing is defined as chronological age in most studies, which demonstrate consistently that the older are less likely to repartner than the younger (Mahay and Lewin, Reference Mahay and Lewin2007; Meggiolaro and Ongaro, Reference Meggiolaro and Ongaro2008; Schimmele and Wu, Reference Schimmele and Wu2016; Rapp, Reference Rapp2018). The processual effect of becoming older is only rarely analysed, since most studies do not use longitudinal data (e.g. Schimmele and Wu, Reference Schimmele and Wu2016) or use methods which do not include ageing as a process, but rather as a structural comparison category (e.g. Jaschinski, Reference Jaschinski2011). Thus, they cannot account for the multi-dimensional complexity of ageing. Sporadic studies take into account the influence of the relationship status on attitudes towards ageing, but not vice versa (Mauritz and Bischoff, Reference Mauritz and Bischoff2023), and often without considering processual dimensions (Jung et al., Reference Jung, Cham, Siedlecki and Jopp2021).
Previous findings concerning the influence of the relationship biography on repartnering show how the form of singlehood influences the decision to repartner. For widowhood, it is often argued that due to the idealisation of the late partner (Bennett et al., Reference Bennett, Arnott and Soulsby2013), a narrative arises that the deceased person cannot be replaced by a new partner (Davidson, Reference Davidson2001, Reference Davidson2002; Stevens, Reference Stevens, Davidson and Fennel2004). At the same time, studies show that repartnered widow(er)s do not feel like they betrayed their former partner (Davidson, Reference Davidson2002; Stevens, Reference Stevens, Davidson and Fennel2004), despite their ongoing bond to the deceased (Dekel et al., Reference Dekel, Shorer and Nuttman-Shwartz2022). Divorced or separated individuals often do not want to repartner because of the fear of experiencing another rejection or disappointment (Crowley, Reference Crowley2019). However, comparisons illustrate consistently that divorced or separated individuals are more likely to repartner than widowed individuals (Treas and Van Hilst, Reference Treas and Van Hilst1976; Schimmele and Wu, Reference Schimmele and Wu2016; Brown et al., Reference Brown, Lin, Hammersmith and Wright2018; Rapp, Reference Rapp2018).
To my knowledge, there are no previous findings which describe and analyse repartnering in older age in Germany by longitudinally comparing divorce and separation with the death of a partner for men and women. This article begins to close this research gap and focuses on gender and ageing specificities, since previous findings indicate that men and women express different motivations and behave in gender-specific patterns when they repartner in later life.
Methodology
A prospective, longitudinal panel data structure is especially useful when ageing is of certain interest. It is additionally eligible when analysing lifecourse transitions because they can be interrogated while they occur and because it is possible to compare transitional and non-transitional individuals structurally, as well as to observe processual changes within transitional individuals.
Data
The German Ageing Survey (DEAS) is a sociological and psychological panel survey covering the living conditions of the German population in the second half of life. The survey offers seven waves (1996, 2002, 2008, 2011, 2014, 2017, 2020/21) – of which the first six waves are analysed in this article – and is conducted via face-to-face interviews and a complementary fill-in questionnaire. With a cohort-sequential design, a new random sample (of respondents aged 40–85) is pulled every six years (1996, 2002, 2008, 2014, not in 2020/21) and included within the longitudinal design. The participation rate in the panel rose continuously with each new wave (from 1996 to 2002: 32%, from 2008 to 2011: 46%, from 2014 to 2017: 52%). Panel mortality is selective with increasing age, and men are more likely to drop out when controlling for education. The DEAS covers topics like occupation, retirement, housing, leisure activities, familial and other social relationships, social participation, health and wellbeing, and is therefore suitable for this analysis (Klaus et al., Reference Klaus, Engstler, Mahne, Wolff, Simonson, Wurm and Tesch-Römer2017).
Sample
The six waves of the DEAS used for the analysis consist of 39,446 observations from 20,129 respondents. To focus on the third and fourth ages rather than middle age, all respondents who were only interviewed when they were under 50 (2,254) and who participated only once (8,041) were excluded. Additionally, the sample does not enclose respondents possessing an inconsistent relationship status (1,182), living in a relationship over the whole observation period (3,631), being never married and/or never partnered (149) or becoming single in their last participation (244). After deleting individuals with missing values on the variables included in the analysis (975), the analysis sample consists of 3,653 respondents and 11,628 observations, within which all respondents were single in at least one wave and 179 new relationships occurred. The analytical sample is on average 65 years old when being interviewed, which is the same average age as in the whole dataset after removing those aged under 50. Most participants belong to the cohort born between 1940 and 1954, whereas the cohort born between 1910 and 1924 is the smallest cohort in the sample. The gender ratio shows a slightly higher share of female respondents (54%) compared to the complete sample, within which the gender ratio is nearly equal. Therefore, women are overrepresented within the analytical sample, which already illustrates the gender-specific structure of the partner market (see Table 1).
Table 1. Specification of the distribution over the explanatory variables (descriptive results)
Notes: SPA: self-perception of ageing. 1. Mean over waves. 2. Mean.
Source: German Ageing Survey (1996–2017).
Measures
The dependent variable (couple) indicates whether a respondent is living in a relationship. It is dichotomous and time-varying (0 = no, 1 = yes) and summarises civil unions as well as marriages living in co-habitation or apart together.
The first explanatory variable (gender) is being interrogated binarily and is thereby a dichotomous and time-invariant variable (0 = male, 1 = female). There are no gender transitions within either the original sample or the analytical sample. Following an analytical-pragmatic strategy, this article includes both different- and same-gender couples, based on the assumption that ‘couples share more similarities than differences in their social dynamic’ (Kuchler and Beher, Reference Kuchler, Beher, Kuchler and Beher2014: 10, own translation).
The completed, chronological age when being interviewed is scaled metrically and time varying. The functional form of the ageing-term is linear, not squared, because interactional terms for ageing are included (Spieß, Reference Spieß, Wolf and Best2010).
Belonging to a cohort is scaled ordinally, time-invariant and based on the year of birth (0 = 1910–1924, 1 = 1925–1939, 2 = 1940–1954, 3 = 1955–1969). Due to the ‘identifiability problem’ (Rutherford et al., Reference Rutherford, Lambert and Thompson2010: 606) of age–period–cohort effects and the ‘overparameterization’ (Rutherford et al., Reference Rutherford, Lambert and Thompson2010: 606) when all three effects are included, this analysis accounts for ageing and cohort effects. Using longitudinal data and separating within- and between-effects allows one to separate these effects and to address possible confounding (Palmore, Reference Palmore1978; Holford, Reference Holford1992; Rutherford et al., Reference Rutherford, Lambert and Thompson2010).
The self-perceptions of ageing (Diehl et al., Reference Diehl, Wettstein, Spuling and Wurm2021; Jung et al., Reference Jung, Cham, Siedlecki and Jopp2021) are the basis for the attitudes towards ageing. These scales measure different dimensions of attitudes towards ageing (For most people, getting older means…), namely physical decline (…that you aren't able to withstand as much as you used to, …that you are less able to compensate for physical limitations, …that you are less energetic and fit, …that your health gets worse), social loss (…that you aren't really needed anymore, …that you get bored more and more often, …that people treat you with less respect, …that you feel lonely more often) and continuous growth (…that you keep making plans, …that you're still able to learn new things, …that you can still put your ideas into practice, …that you can expand your skills and abilities). They are metrically scaled mean-indices, time-varying and based on the items, which use a four-point scale (1 = strongly agree, 2 = agree, 3 = disagree, 4 = strongly disagree; the higher the index, the more positive the respondent's attitudes towards ageing for physical decline and social loss, vice versa for continuous growth; Engstler et al., Reference Engstler, Lozano Alcántara, Luitjens and Klaus2019).
The different forms of singlehood are time-varying and scaled dichotomously and separately for widowhood (0 = not widowed or partner died, 1 = widowed or partner died) and for divorce or separation (0 = not divorced or separated, 1 = divorced or separated).
The full model controls for functional health (number of physical illnesses), self-rated health (0 = much better, 1 = somewhat better, 2 = the same, 3 = somewhat worse, 4 = much worse), education (0 = low (International Standard Classification of Education (ISCED) 0–2), 1 = medium (ISCED 3–4), 2 = high (ISCED 5–6)) and income (monthly equivalence income – oriented on per capita demand of every household member and based on the modified Organisation for Economic Co-operation and Development equivalence scale, rounded to fit a 100-point scale; Engstler et al., Reference Engstler, Lozano Alcántara, Luitjens and Klaus2019). The variables for income and education are additionally included as interactional terms with gender, because – as argued in the theoretical framework – resources are being accumulated gender-specifically over the lifecourse (Krüger and Levy, Reference Krüger and Levy2001).
Analytical strategy
For the analysis of repartnering transitions, hybrid panel regressions are modelled. Hybrid models combine random-effects (RE, or between) with fixed-effects (FE, or within) estimators (Brüderl, Reference Brüderl2005; Rabe-Hesketh and Skrondal, Reference Rabe-Hesketh and Skrondal2008; Wooldridge, Reference Wooldridge2009). An FE model is not being used, since it does not allow the inclusion of time-invariant explanatory variables like gender or education, which are of major interest for the analysis, and it includes only transitioning individuals, which are very few in the analytical sample (Brüderl, Reference Brüderl2005).
RE models do allow accounting for inter-individual variation by including time-invariant factors and all available research units. However, they are based on the assumption that the unobserved effect and the idiosyncratic error term are in constant variance, identically distributed and uncorrelated with all explanatory variables (Rabe-Hesketh and Skrondal, Reference Rabe-Hesketh and Skrondal2008; Wooldridge, Reference Wooldridge2009). Since in social sciences the latter assumption often cannot hold, time-constant unobserved heterogeneity is problematic, contrary to an FE model and even though it can be attenuated by including time-invariant, explanatory variables. Time-varying, unobserved heterogeneity or the idiosyncratic error term can bias the analysis in both FE and RE models (Rabe-Hesketh and Skrondal, Reference Rabe-Hesketh and Skrondal2008).
In this article, hybrid models are conducted even though they cannot overcome the described drawbacks of RE models. To account for the drawbacks, an FE model is created to check the robustness of the hybrid model. The choice of a hybrid regression model was made due to the possibility of combining the advantages of the FE and RE estimators. Accordingly, the problems of the usage of the RE estimator can be minimised since time-varying variables can be included within-transformed and between-transformed in a hybrid model (Rabe-Hesketh and Skrondal, Reference Rabe-Hesketh and Skrondal2008; Wooldridge, Reference Wooldridge2009). This inclusion of time-varying variables for both within- and between-effects and of time-invariant variables for between-effects is the main advantage of hybrid models, which allows us to simultaneously analyse variations within individual processes and comparisons between individual positions. Linking this analytical approach to the theoretical considerations, I propose to understand the within-effects as processual – developing social positions of one individual over time – and between-effects as structural – differences in social positions between individuals at one point in time – dimensions of repartnering.
Due to the dependent variable being dichotomous, a logistic regression is modelled. The first model contains gender as an explanatory variable, the second model contains ageing and the third model adds cohort. The fourth model contains gender and ageing, the fifth model additionally contains cohort, the sixth model excludes cohort and includes an interactional term of gender and ageing, the seventh model adds cohort again, and the eight model adds an additional interactional term for gender and cohort. The interactional terms are added in order to account for the interdependency of the variables. Since these measures represent the theoretical focal points, they are included within all following models, which are computed separately for the other explanatory variables (Model 9 tests Hypothesis 5; Model 10 tests Hypothesis 6). The eleventh and full model includes all explanatory and control variables in order to compare the impact of each explanatory variable and to identify the main influencing factors.
The models are interpreted using average marginal effects (AME) because – contrary to log odds or odds ratios – the intensity of AMEs can be interpreted and they can be compared across models and between groups (Mood, Reference Mood2010).
All models were estimated with the xtlogit commands in Stata (version 16.1) and a significance level of 5 per cent (p < 0.05) was set.
Findings
As elaborated in the sample section, 9.7 per cent of the singles repartnered within the observation period, meaning that 179 new relationships were formed. Contrary to the formulated expectation of the relevance of this topic, the proportion of singles forming a new relationship does not increase over time. Although the number of new relationships is somewhat growing, so is the proportion of singles, and therefore the proportion of singles forming new relationships is decreasing (see Table 1).
The results from the logistic hybrid regression models show that, when gender is the only independent variable, the expected effect cannot be confirmed. Rather, it shows that women – not men – are significantly more likely to form a new relationship. With stepwise adding ageing and cohort to the model, gender stays significant (on a 0.1% level), even though the effect is being lessened. The gender effect changes its direction, as the interactional term for gender and ageing is added to the model, indicating that men are more likely to repartner than women. When the variables for form of singlehood are added to the models, the significance of the gender term drops on a 1% level. Still, Hypothesis 1 can be confirmed (see Table 2, Models 1, 4–11).
Table 2. The influence of structural and processual dimensions on repartnering in older age (average marginal effects from hybrid logistic models)
Note: SPA: self-perception of ageing.
Source: German Ageing Survey (1996–2017).
Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001.
On the structural level and contrary to Hypothesis 2, the results of the univariate model indicate that older individuals are more likely to repartner than younger individuals. The effect remains similar when cohort is added to the model but becomes insignificant as soon as form of singlehood is included in the model. Contrary to the gender effect, the ageing effect does not change its direction with the inclusion of the interactional term between gender and ageing. The processual effect of ageing is negative in all models and significant (on a 1% level) in the full model. Hypothesis 2 can therefore be confirmed on the processual level: the older an individual gets, the less likely repartnering becomes, but it has to be rejected on the structural level, since the effect is not significant in the full model (see Table 2, Models 2–11).
Hypothesis 3 – with rising age, men are increasingly more likely than women to repartner – must be rejected since the results are contrary to the hypothesis. On the structural level, the findings show that the likelihood to repartner rises to a greater extent for older women than for older men (significant on a 1% level). On the processual level, ageing has a stronger effect on the likelihood to repartner for women than for men. Similar to gender and ageing, the effect of the interactional term lessens but remains significant, when form of singlehood is included in the model (see Table 2, Models 6 and 11).
The effects of cohort and the interactional term of cohort with gender are not significant. Thus, Hypothesis 4 has to be refused. When gender is included in the model additionally to ageing, their direction indicates that later-born cohorts are more likely to repartner than earlier-born cohorts. Accordingly, rather gender and ageing influence repartnering, which allows one to understand the ageing effects as actual ageing and not cohort-mediated effects (see Table 2, Models 3, 5 and 7–11).
Two of the three dimensions of self-perceptions of ageing – physical decline and continuous growth – are insignificant in the final model and the former shows an effect that is as hypothesised, whereas the latter does not. The dimension of social loss is significant (on a 0.1% level) and influences repartnering in the opposite direction than hypothesised. The structural finding shows that individuals with negative attitudes towards ageing are more likely to repartner than individuals with more positive attitudes. The processual effect indicates that when the attitudes towards ageing are worsening, repartnering is getting more likely. In sum, Hypothesis 5 has to be rejected on both levels (see Table 2, Models 9 and 11).
The coefficients indicating the influence of form of singlehood are both positive, high and significant (on a 0.1% level). Hypothesis 6 can be confirmed on the structural level, because the divorced or separated are more likely to repartner, compared to the individuals whose partner died. On the processual level, however, Hypothesis 6 must be rejected, because the findings illustrate how, when the partner dies, repartnering becomes more likely than when a divorce or a separation occurs within the observation period (see Table 2, Models 10 and 11).
The control variables affect repartnering only minimally. On the structural level, higher education lessens the likelihood of repartnering significantly (on a 5% level). Additionally, in interaction with gender, the educational effect is significantly stronger for women than for men (on a 1% level). Changes in functional health do not influence the likelihood to repartner, whereas individuals with a better functional health are less likely to repartner compared to those with a worse functional health (on a 5% level). Individuals who rate their health lower are significantly (on a 1% level) more likely to repartner than individuals rating their health higher. On the processual level, self-rated health does not affect the likelihood to repartner. Income and the interactional term of income and gender do not influence repartnering (see Table 2, Model 11).
As a robustness check, linear hybrid regressions and logistic FE regressions for the time-varying variables were modelled (with xtreg, re and xtogit, fe commands; see Table 3 for the FE models). Even though the findings are not completely congruent, they indicate the robustness of the hybrid regression models.
Table 3. Robustness check (average marginal effects from logistic fixed-effects models)
Note: SPA: self-perception of ageing.
Source: German Ageing Survey (1996–2017).
Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001.
Discussion
To sum up, the results on the structural level illustrate that women and individuals with more positive attitudes towards ageing are less likely than men and individuals with more negative attitudes towards ageing and the widowed are less likely than the divorced or separated to repartner. On the processual level the results indicate that the likelihood of repartnering decreases when getting older, when the attitudes towards ageing are getting worse and when experiencing divorce or separation compared to experiencing widowhood. For gendered ageing, the results show that the ageing effect is stronger for women than for men. This means that women's likelihood to repartner increases more with age compared to the rise in men's likelihood to form a new relationship.
I firstly address the finding that only 179 new relationships form over the 21 years. Especially with regard to the result that repartnering becomes less likely with increasing age, this observation could be explained with the constantly increasing age of the sample. It is at the same time not in accordance with previous findings indicating a rise in repartnering rates in older age (Bulcroft and Bulcroft, Reference Bulcroft and Bulcroft1991; Brown and Shinohara, Reference Brown and Shinohara2013; Wu et al., Reference Wu, Schimmele and Ouellet2015). With the present analysis, it cannot be said if this is the result of unfavourable opportunity structures, of decreasing interest in new relationships, of ageist discourses and norms regarding love and sexuality, of the wish to enjoy the newfound freedom or if there is a completely different explanation.
Secondly, the observations regarding gender and ageing as well as their intersection show that the unexpected gender effect lessens when adding ageing to the model. Once the interactional term for gender and ageing is additionally included, the gender-effect changes its direction as hypothesised, increases strongly and the ageing effect decreases. However, when accounting for form of singlehood, the structural ageing effect loses its significance, while the processual ageing effect and the effects of the interactional term do not change. The strong correlation of the variables indicating the form of singlehood with ageing can methodically explain this finding (see Table 4). It could, however, theoretically be argued that the influence of ageing on repartnering is actually explained by the form of singlehood: differences in what happened over the lifecourse seem to be more relevant than differences in chronological age, at least for the lifecourse transition of repartnering. However, the findings for both gender and the interactional term of gender and ageing demonstrate how mating patterns are gendered and both systematically and strongly related to ageing-specific patterns – and vice versa. They can also be understood as a confirmation of the concept of ‘the double standard of ageing’ (Sontag, Reference Sontag, Allman and Jaffe1972), since ageing has a stronger effect on women compared to men. Moreover, these findings underline the gendering of ageing since gender-specific resources, which are relevant for repartnering, decrease for women and increase for men with advancing age (Bamler, Reference Bamler, Buchen and Maier2008; Gildemeister, Reference Gildemeister, Buchen and Maier2008; Backes and Wolfinger, Reference Backes, Wolfinger, Blättel-Mink and Kramer2009). As lifecourses are similarly gendered in other Western countries and the demographic changes are developing in a comparable manner, these gender- and ageing-specific differences and the entanglement between them are likely also to explain repartnering in later life beyond Germany. This could be given further notice in future cross-country comparative research.
Table 4. Correlation matrix of the key variables (Pearson's R)
Note: SPA: self-perception of ageing.
Source: German Ageing Survey (1996–2017).
Significance levels: ** p < 0.01, *** p < 0.001.
Thirdly, the insignificance of the cohort effect shows how the ageing effects are in fact based on becoming older and its interaction with gender. It is a main finding that the ageing process is more relevant for repartnering than the birth cohort, even though this study does not systematically account for period effects. Especially considering the empirically based proclamation of a cultural change in relationship norms and practices (Beck-Gernsheim, Reference Beck-Gernsheim2002; Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004), this finding is rather surprising. However, an alternative operationalisation of the cohort variable could display cohort belonging as before and during the late 1960s (Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004), which marks an important normative turning point in many Western countries. Future analyses could include additional period effects by using age–period–cohort models (Rutherford et al., Reference Rutherford, Lambert and Thompson2010; Fosse and Winship, Reference Fosse and Winship2019). This would be especially relevant considering the historical context in Germany, which includes the economic changes resulting from the Wirtschaftswunder, the German division and the fall of the Berlin Wall (Leontowitsch, Reference Leontowitsch, Denninger and Schütze2017; Vogel et al., Reference Vogel, Wettstein, Tesch-Römer, Vogel, Wettstein and Tesch-Römer2019).
Fourthly, the importance of complementing the socio-structural category of ageing with subjective valuations in order to operationalise the complexity of ageing is underlined by the finding that an individual's negative attitudes towards ageing in the dimension of social loss make them more likely to repartner. However, their attitudes concerning physical decline and continuous growth are irrelevant to the formation of a new relationship. This finding can be cautiously generalised beyond the German context, as it is in line with previous research from other Western countries, which identifies loneliness and/or social exclusion as driving factors in repartnering in later life (Cooney and Dunne, Reference Cooney and Dunne2001; Davidson, Reference Davidson2001, Reference Davidson2002; Koren, Reference Koren2015).
Finally, the form of singlehood does not influence repartnering as hypothesised. Even though, on a structural level, divorced or separated individuals are more likely to repartner than widowed individuals, the processual effects show that repartnering is more likely when experiencing the transition into widowhood compared to experiencing a divorce or separation. These different effects indicate that the death of a partner leads to a short-term increase in the likelihood of repartnering and that repartnering after divorce or separation is more likely to occur in the long term. These findings seem to imply differences in the temporal structures of repartnering following these transitions. Still, it is worth considering that individuals who experience widowhood tend to be older than individuals going through a divorce or separation. This is reflected in the positive correlation between widowhood and ageing as well as the negative correlation coefficient between divorce or separation and ageing (see Table 4), and it aligns with the finding regarding the structural dimension but not the processual one. The finding on the processual level is also out of accordance with previous quantitative findings (Schimmele and Wu, Reference Schimmele and Wu2016; Brown et al., Reference Brown, Lin, Hammersmith and Wright2018; Rapp, Reference Rapp2018) but aligns with and thereby confirms the emerging theory of qualitative findings. These state that older widowed individuals wish to repartner if they enjoyed their former relationship and that divorcees or separated individuals do not want a new relationship due to their possibly traumatic transition into singlehood (Davidson, Reference Davidson2001; Carr, Reference Carr2004; Stevens, Reference Stevens, Davidson and Fennel2004; Bennett et al., Reference Bennett, Arnott and Soulsby2013; Crowley, Reference Crowley2019). The results for form of singlehood demonstrate how hybrid panel models can enable the revelation of systematic temporal structures.
Limitations
The following aspects limit the analysis of this paper. (a) The lifecourse and its transitions can be analysed more holistically with other longitudinal methods like event history analysis or sequence analysis, as those can account for factors (like timing or relationship transitions between waves) that are central to the lifecourse perspective (Abbott, Reference Abbott1992; Fasang, Reference Fasang, Blanchard, Bühlmann and Gauthier2014). Future research on relational lifecourse transitions in older age could apply these methods. (b) Using solely the DEAS, the analysis could account neither for individual mating preferences nor for numerical and meeting opportunities or restrictions of the partner market in older age (for an example of the latter for middle age, see Corti and Scherer, Reference Corti and Scherer2021). Furthermore, one's family background or relationship history – especially the quality of the last relationship and the duration since or age at the transition into singlehood – could further clarify why some people repartner and others do not (Poortman, Reference Poortman2007). Considering variables which depict participation in the partner market, preferences for partnering and relationship history are highly relevant when analysing repartnering and should be addressed in future research. (c) Without differentiating for sexual orientation, this article reproduces the heteronormative perspective of love studies yet again (Lenz, Reference Lenz, Schöer, Stauber, Walther, Böhnisch and Lenz2013) and thereby makes relationship diversity invisible. Including non-heteronormative relationships in older age within a quantitative longitudinal analysis is nearly impossible, since sexual orientation was not a regular question in surveys in the 1990s and the number of cases is statistically speaking too small (for an exception, see Ophir and Boertien, Reference Ophir and Boertien2023). Still, in further quantitative research it will be crucial to find a way to shed light on the variety of relationships. (d) Additionally, this study cannot shed light on how new relationships come to be or whether the individuals who stayed single do so intentionally. A complementary qualitative analysis or a more extensive quantitative dataset – including information such as partner preferences or the structure of the partner market (Kalmijn and Flap, Reference Kalmijn and Flap2001; Mahay and Laumann, Reference Mahay, Laumann, Laumann, Ellingson, Mahay, Paik and Youm2004) – could enable an understanding of decisions, norms, wishes and attitudes towards repartnering, allowing the repartnering process in older age to be understood more deeply and broadly.
Enriching the lifecourse perspective with gender- and ageing-specific perspectives for repartnering in later life
The novel contribution of this paper is that it is the first longitudinal study analysing repartnering in older age among divorced, separated and widowed female and male individuals in a German context.
Its central contribution and significance, however, lie in the gender- and ageing-sensitive theoretical framework, which is not only relevant for the German context, but can easily be transferred to repartnering and other relationship transitions in other Western contexts. In this article, I enriched the lifecourse perspective on transitions with gender- and ageing-specific and intersectional considerations and focused on the differentiation of structural and processual dimensions of lifecourse transitions. The results demonstrate that not only should the structural social categorisation of male or female or of age groups be considered, but so should subjective ageing processes and gendered ageing. Applying this perspective to relationship transitions can help with understanding these seemingly individual processes as highly interrelated with structural circumstances as well as discursively generated and institutionally consolidated norms (Aubert, Reference Aubert1965; Kalmijn and Flap, Reference Kalmijn and Flap2001; Kuchler and Beher, Reference Kuchler, Beher, Kuchler and Beher2014).
The application of hybrid panel models – with the understanding of intra-individual effects as processual dimensions and inter-individual effects as structural dimensions – demonstrates how this method is especially fit to analyse lifecourse transitions in their specific timely structure. Moreover, the inclusion of form of singlehood illustrates how repartnering is – in reference to linked lives – linked to other transitions emerging over the lifecourse (Fasang, Reference Fasang, Blanchard, Bühlmann and Gauthier2014; Bischoff et al., Reference Bischoff, Franke and Wanka2021).
Conclusion
Resuming this analysis, it could be illustrated how structural and processual dimensions influence the formation of relationships in later life. The longitudinal analysis and comparison of structural and processual dimensions allows one to grasp not only social embeddedness, but also the timely structure of lifecourse transitions and their diachronic linkage. The results concerning form of singlehood and the difference in the temporal structuring of the occurrence of repartnering provide especially good starting points to systematically include and relationally link different lifecourse transitions. We could further ask how transitions (e.g. into new occupational fields, or another country in middle age, into grandparenthood, or new living arrangements in older age) mediate relationship transitions and their temporal structure in older age.
At the same time, the inclusion of attitudes towards ageing and the interactional term of gender and ageing demonstrate how ageing can be conceptualised according to its inherent complexity. Additionally, the inclusion of gender-specific interactional terms shows the importance of accounting for the gendering of social participations over the lifecourse. For future research questions on how gender- and ageing-specific patterns of repartnering or staying single emerge over time – e.g. regarding the living arrangement of new relationships or the decision (not) to marry – should be considered more systematically. Sensitising the lifecourse perspective with gender- and ageing-specific concepts allows us to advance our understanding of the structural and processual interrelatedness of these two pivotal categories (Aeby and Gauthier, Reference Aeby, Gauthier, Castrén, Cesnuityte, Crespi, Gauthier, Gouveia, Martin, Mínguez and Suwada2021).
However, the analysis could not shed light on how repartnering and the partner market in older age are linked to possibly ageist normative premises, e.g. concerning sexuality in older age. For future research, it could be asked whether and how repartnering actually challenges these stereotypes. Additionally, in a couple-oriented society, which considers being in a relationship an essential part of a good life (DePaulo and Morris, Reference DePaulo and Morris2005), we could also examine whether and how an active rejection of repartnering can question this ideal. These resistant practices are especially relevant because their emphasis on autonomy means that the conception of ageing as a phase of fragility and dependence can be questioned, and both ageing itself and the concept of a good life can be renegotiated.
To conclude, finding love in older age is a highly relevant research topic because (normative) premises about gendered ageing culminate within it, and because it reveals the interrelated functioning of both structural and processual dimensions of familial lifecourse transitions in older age.
Acknowledgements
I would like to thank the reviewers for their elaborate feedback, the participants of the PhD Summer School from the European Sociological Association in Ljubljana for the discussion of a first draft. Additionally, a large thank you to Prof. Dr Daniela Grunow, Prof. Dr Birgit Becker and Prof. Dr Frank Oswald for their helpful and detailed feedback.
Competing interests
The author declares no competing interests.
Ethical standards
The paper uses secondary data, therefore ethical approval is not required.
Open access
