Does living together or apart matter? A longitudinal analysis of relationship formation in later life in association with health,well-being,and life satisfaction

Abstract

Despite a growing body of research on later-life relationship formation, little is known about the health predictors and outcomes of later-life Living-Apart-Together (LAT) relationships. A LAT living arrangement is understood to be a possible way for older adults with age-related limits to partner and balance the consequences of being single. Using both selection and resource models to capture the links between health and relationship status, we analyzed unpartnered people 50+ from the longitudinal Survey of Health, Ageing and Retirement in Europe. We examined who entered LAT or cohabiting relationships between Time 1 and 2, who remained unpartnered, and what effect the transition into one of these statuses had on the well-being of those who remained in that status at Time 3. The respondents in LAT relationships did not differ from their cohabiting counterparts in health indicators before the relationship formation, nor was there a difference in life satisfaction and well-being 2 years after partnering. Compared to unpartnered persons, LAT respondents reported better self-perceived health before relationship formation and slightly higher life satisfaction 2 years later. Health status does not influence the choice for a relationship form, but a LAT relationship may constitute a resourceful living arrangement in later life, which provides some support for both the selection and resource models.

Keywords

Aging health Living-Apart-Together relationships longitudinal survey relationship formation SHARE

Introduction

Relationships in later life have rightfully received a growing amount of attention among scholars because a non-negligible proportion of older people have undergone various union dissolutions that have exposed them to health risks related to the absence of a relationship partner (Beekman et al., 1995, 2000; Federal Statistical Office of Germany, 2016). Although partnering generally reduces health risks, establishing a new relationship in later life is challenging because markers of health, health deterioration, and caregiving issues become more salient (Beekman et al., 1995, 2000; Umberson et al., 2013). In this respect, later-life Living-Apart-Together (LAT) relationships may represent an alternative to cohabitation partnerships because they provide the flexibility to negotiate caring duties and dealing with health deterioration (Connidis et al., 2017; Koren et al., 2016). However, our knowledge of the associations of health and relationship formation in later life remains limited.

Theoretical framework for studying associations between relationships and health

Prior research provides evidence that family status is closely linked to health status and this association may remain pertinent even after controlling for other relevant factors (Umberson et al., 2013). This conclusion is predominately drawn from studies on marriage—a partnership status that has consistently been found to be a beneficial factor for health for both women and men (Schoenborn, 2004; Umberson et al., 2013), even when compared to a cohabitation living arrangement during a later-life period (Brown et al., 2005). The positive effect of marriage on health is often explained by two competing models: the selection model and the resource model. The selection model states that changes in marital status are influenced by health status, specifically by individual characteristics that may lead to both poorer health and lower chances for marriage. In other words, people with poorer mental and physical health have a lower likelihood of marrying or maintaining a stable marriage. Whereas the resource model suggests an effect in the opposite direction, marital roles influence the health of individuals (Kim & McKenry, 2002). The resource model explains this relationship with a marital resource model, where health is improved by marriage because it provides social, psychological, and economic resources (Ross et al., 1990; Umberson, 1992). However, these explanatory models have been developed without consideration of health changes in later life and without the specifics of the aging population, which include an increased number of divorced older people and the possible cessation of marriage due to a partner’s death (Wu & Schimmele, 2005).

Relationship formation among older persons

In older age, the chance for establishing a new relationship generally decreases with age. This is especially true for women and older persons with a poor financial situation (Brown et al., 2016; de Jong Gierveld, 2004; Schimmele & Wu, 2016; Vespa, 2012; Wu et al., 2015).

Apart from sociodemographic characteristics, health also plays an important role in later-life relationship formation. While chronic illness decreases the chance of remarriage among older men (Schimmele & Wu, 2016), good self-rated health increases the likelihood of remarriage in older women (Brown et al., 2016). This indicates that good health status increases the chances for later-life remarriage. This may be due to the fact that the better health of a couple helps to attenuate the worries related to future caregiving—a commitment that is likely to be expected in marriage and cohabitation relationships (Koren et al., 2016). Nonetheless, existing studies on later-life relationship formation have mainly focused on remarriage and later-life cohabitation despite the fact that some persons opt for a more flexible relationship in order to not remain unpartnered (Duncan & Phillips, 2010; Koren, 2014; Levin, 2004). One of the relationship options is a Living-Apart-Together (LAT) union, with estimated rates that range from 2.1% to 5.1% (based upon studies from the Netherlands and Canada; de Jong Gierveld, 2015; Turcotte, 2013).

Specifics of later-life LAT relationships, health, and their associations from the perspectives of the selection and resource models

The LAT relationship is defined as a family form in which partners are considered to be a couple despite keeping separate households (Levin, 2004). With respect to older people, later-life LAT relationships are characteristic for their stability because couples tend not to transform the LAT relationship into cohabitation (Duncan et al., 2013; Kobayashi et al., 2017; Régnier-Loilier et al., 2009; Turcotte, 2013).

The main function of this relationship form in later life is to have someone with whom older people may share hobbies, spend leisure time, and have fun (Benson & Coleman, 2016; Bildtgård & Öberg, 2015; Koren, 2014, 2015), while maintaining a balance between intimacy and autonomy, including family commitments (Benson & Coleman, 2016; de Jong Gierveld & Merz, 2013; Duncan et al., 2013; Funk & Kobayashi, 2016; Koren, 2015; Malta & Farquharson, 2012).

In support of the selection model, the literature shows that health and health deterioration represent issues that may shape relationship formation and the preference for a LAT relationship over other family forms for older persons who desire to (re)partner (Benson & Coleman, 2016; Bildtgård & Öberg, 2015; Funk & Kobayashi, 2016; Koren, 2015). These studies commonly point out that the LAT preference functions as protection against the risks associated with later-life (re)partnering, the presence of health problems (e.g., chronic health diseases, a history of alcohol abuse), and questions of caregiving (Duncan et al., 2013; Koren et al., 2016). Older persons take into consideration issues associated with the deterioration of physical health. Specifically, those in LAT relationships balk at the expectation to provide physical care to their partner or to be the recipient of their partner’s care (Bildtgård & Öberg, 2015; Duncan et al., 2013; Koren et al., 2016). In this view, a later-life LAT relationship, as a partnership type, is considered by older people who desire to partner and whose health status could be perceived as a possible source of worry with respect to future caregiving demands. We hypothesize that those who remain unpartnered may differ in physical health before their relationship formation such that (1) unpartnered persons have more health problems than those who partner in LAT relationship forms and (2) those who opt for a LAT living arrangement have poorer health than those who establish a cohabitation relationship or get married.

In support of the resource model, prior research shows that being in a relationship is perceived to be more rewarding or bring additional benefits than remaining unpartnered in later life (Koren, 2015). However, older persons experience tension when balancing autonomy and intimacy, which results in the fact that some LAT persons have their needs for emotional closeness insufficiently fulfilled since a relationship with two separate households requires continually negotiating the presence of the partner (Koren, 2014). This may be challenging in later life and it may lead to lower satisfaction with their life situations. Other research has shown that older people in non-residential unions reported a lower amount of perceived instrumental support than those in cohabitation relationships (Strohm et al., 2009). Moreover, Lewin (2017) used a large-scale, cross-sectional research project to conclude that LAT relationships are slightly less likely to be happy than other cohabitation relationship forms, pointing to possible differences in the extent to which these partnership types are rewarding. In this vein, we propose to test the hypothesis that those who remain unpartnered and those who opt for LAT and cohabitation relationships may differ in well-being and life-satisfaction such that (1) unpartnered persons have poorer well-being and life-satisfaction than those who partner in LAT relationship forms and (2) those who opt for a LAT living arrangement have poorer well-being and life satisfaction than those who establish a cohabitation relationship or get married.

Research aim

Based up on previous studies, which are still limited in number, we propose to view the establishment of a later-life LAT relationship as a possible way for older adults to partner because worries related to becoming responsible for a partner’s care may be attenuated (i.e., the perspective of the selection model). We also propose to re-examine whether a later-life LAT relationship is less rewarding compared to more committed partnership types, such as cohabitation and marriage in older age (i.e., the perspective of the resource model). The present study aims to explore (1) the health predictors of establishing a later-life LAT relationship, and (2) the associations of establishing a later-life LAT living arrangement with subsequent changes in the quality of well-being after the formation of the relationship. These associations are studied with respect to the following research question: to what extent do older people who entered into a LAT relationship differ in both (1) the health status before relationship formation and (2) in their well-being after partnering, from those who established a cohabiting relationship/got married, and those who remained unpartnered. We assume that, in the former case, unpartnered persons have more health problems than those who partner in LAT relationship forms. And we assume that those who opt for a LAT living arrangement have poorer health than those who establish a cohabitation relationship or get married. In the latter case, we hypothesize that unpartnered persons have poorer well-being and life-satisfaction than both those who entered into a LAT relationship and those who opted for a LAT living arrangement have poorer well-being and life satisfaction than those who established a cohabitation relationship or got married.

Methods

Data description

We used data from the Survey of Health, Ageing and Retirement in Europe (SHARE; see www.share-project.org). SHARE is a longitudinal cohort study of people aged 50 and over who regularly live at the time of data collection in the respective participating country. SHARE is based on probability sampling and it is conducted every 2 years in 17 European countries and Israel. The first wave of data collection was initiated in 2004–2005 and seven waves have already been conducted. Each wave has a sample that is refilled with new respondents in order to address the issue of drop-outs and each wave continuously monitors the birth cohorts that had since reached age 50. Participants were excluded from the SHARE sample if (a) they were unreachable at home during the entire survey period, (b) they were unable to speak one of the national languages, or (c) they were less than 50 years old at the time of data collection. Recruitment was done by telephone or in person at the participants’ home addresses. Computer-assisted personal interviews were used for data collection. The SHARE study has been reviewed and approved by the Ethics Committee of the University of Mannheim (Waves 1 to 4) and the Ethics Council of the Max Planck Society (Waves 4 to 7). The prospective study was based on data from Waves 4, 6, and 7, which all included the key variables on health status and well-being. Wave 5 was omitted because the key variable for identification in a LAT living arrangement could not be identified in the data due to the routing rules set up in the computer-aided interviews. These waves of data collection were conducted in 2011–2012, 2015, and 2017, respectively (Börsch-Supan, 2018, 2019, 2020). Waves 4 and 6 were used to explore (1) the health predictors of establishing a later-life LAT relationship; and data from Waves 6 and 7 were analyzed in order to study (2) the effects of establishing a later-life LAT living arrangement on the quality of well-being after the formation of the relationship. The former study focus examined survivors of Waves 4 and 6, while the latter study focus was based on survivors of all the used waves. The total number of cases varied per wave (total N_{wave 4} of 58,129; total N_{wave 6} of 68,188; total N_{wave 7} of 76,520). We refer to the waves of data collection as T1 (Wave 4), T2 (Wave 6), and T3 (Wave 7). Only 5,329 respondents who reported no partner in T1 and simultaneously participated in both T1 and T2 waves of data collection were followed for the first study focus, and 3,831 respondents who reported no partner in T1 and simultaneously participated in all the waves of data collection (i.e., T1-T3) were followed in the second study focus (see the flowchart in Figure 1).

Figure 1.

Flowchart of the sample selection process.

The sample and the identification of respondents with no partner in T1

For this study we selected only those who were unpartnered at T1. To identify these respondents, we had the following variables administrated at T1 (as well as T2 and T3) at our disposal: (1) marital status, with response options “married,” “partnered,” “married, but living separated,” “divorced,” and “widowed”; (2) having a partner in their household, with response options “Yes” or “No”; and (3) having a partner outside of their household, with response options “Yes” or “No.” Upon the combination of the key information, the study sample included only those who participated in the T1 and T2 waves and who simultaneously indicated being unpartnered at Time 1 (n = 5,329 participants; age_mean = 68.80; SD_age = 10.41; minimum = 50; maximum = 100; 77.27% of women).

Measurements

Relationship status and changes over time were measured using three variables administrated at T2: (1) marital status, with response options “married,” “partnered,” “married, but living separated,” “divorced,” and “widowed”; (2) having a partner in a household, with response options “Yes” or “No” along with the cohabiting partner’s identification document (ID), and (3) having a partner outside of the household, with response options “Yes or No.” All those whose relationship status remained unchanged were treated as unpartnered (n = 5,001). All those who indicated that they had a partner outside of their household—and no partner in their household or were not married or in registered partnership—were treated as having a Living-Apart-Together relationship status (n = 215). Due to the lower number of partnered respondents, the remaining participants who had a partner in their household—and not outside of the household or who were married or entered into a registered partnership (n = 5)—were assigned a cohabitation relationship/married status (n = 113).

The same variables for partner status were also administered at T3 so we could monitor changes in relationship status between T2 and T3. Cohabiting respondents were identified based on their marital status and the cohabiting partner’s ID. In addition, the SHARE data set for Wave 7 (i.e., T3) included items on the end of non-cohabiting. This variable represents the year in which a certain partnership ended. We used this variable for the sole purpose of verifying that respondents in the LAT cohabitation/relationship stayed with the same partner between T2 and T3. We found no respondents who established a new LAT relationship within T2 and T3.

Only those whose relationship status remained unchanged (i.e., those who stayed in the LAT, cohabitation, and unpartnered categories and engaged in data collection at T3) were included in the testing resource models, with data measured at T2 and T3 (n_unpartnered = 3,678; n_LAT = 80; n_{cohabiters/married} = 73; n = 6 in registered partnership). These persons were survivors at T3 within the original categories (n_unpartnered = 5,001; n_LAT = 215; n_{cohabiters/married} = 113). The decrease of respondents in the studied categories was primarily due to dropout and relationship dissolution in both the partnered categories between T2 and T3. Additionally, within each analysis, a small portion of respondents was excluded due to missing data. For the final sample sizes for each analysis, see Tables 2, 3, and 4.

Variables measured at T1, T2, and T3

Sex referred to a binary category with the following options: “Male” or “Female.”

Age is the age of the respondent in years.

Years of education refers to the reported number of years the respondent had spent being educated.

Able to make ends meet was used to estimate participants’ socioeconomic situation and it was measured with the question “Is your household able to make ends meet?” with response options that ranged from 1 (with great difficulty) to 4 (easily).

Number of children refers to the number of living children connected to the participant (including adoptive children, stepchildren, and those who were in foster care).

Variables measured at T1 and included in the analyses testing selection models

The SHARE data included several indicators for health that monitor self-reported general health; illnesses or disabilities; eyesight and hearing; and difficulties with a range of (instrumental) activities normal for daily living. We selected three indicators for health difficulties (two objective and one subjective) that were conceptualized as potential barriers for partnering in later life: a number of chronic disease; the Global Activity Limitation Indicator (GALI); and self-perceived health. We chose the GALI scale and subjective self-perceived health to control for everyday life impairment—difficulties that are not directly attributable to disease, but which can negatively affect relationship formation through the adverse effects on spending leisure time together and sharing hobbies (Benson & Coleman, 2016; Bildtgård & Öberg, 2015; Koren, 2014, 2015).

Number of chronic diseases—The number of chronic diseases at the time of interview. This objective indicator was measured by asking respondents whether they received a doctor’s diagnosis on a list of major chronic diseases. The card that was shown included 21 chronic conditions (e.g., high blood pressure or hypertension, Diabetes or high blood sugar, Parkinson disease).

The Global Activity Limitation Indicator (GALI)—This is a global, single-item instrument that used the question “For the past 6 months or more, have you been limited in activities people usually do because of a health problem?” to measure long-standing activity limitations (e.g., related work, household chores, leisure, personal care) and referred to general health problems and activities in which people usually participate, with “0” referring to no limitations and “1” referring to being limited in activities (van Oyen et al., 2006).

Self-perceived health (US Version)—A single-item, self-perceived health scale was measured with the question “Would you say your health is…” Response options ranged from 1 (poor) to 5 (excellent); higher scores indicate better self-perceived health.

Variables measured at T2 and T3 included in the analyses testing resource models

The SHARE data includes two of the following indicators of well-being that were administered in Waves 6 and 7:

Quality of life and well-being (CASP)—Respondents were presented with a 12-item scale adapted from the original questionnaire, CASP-19 (Hyde et al., 2003). Seven items were negatively formulated (e.g., “How often do you think your age prevents you from doing the things you would like to do?”), with response options that ranged from 1 (often) to 4 (never). Higher scores mean a higher quality of life. The total score was computed as the sum of the items. The scale showed good internal consistency: Cronbach´s α _T2 = .812, α _T3 = .820.

Life satisfaction—This was measured with an 11-point scale with the question “How satisfied are you with your life?” where 0 meant completely dissatisfied and 10 meant completely satisfied.

Descriptive statistics of these variables are presented in Table 1. Note that Table 1 describes only the participants with known relationship statuses.

Table 1.

Descriptives per participant Category at Time 1, Time 2, Time 3.

	Participant category
Variable	Repartnered LAT	Repartnered Cohabitation/Married	Unpartnered
Survivors of Time 1 and Time 2	215	113	5001
T1
Age	62.46 (8.94)	62.28 (9.20)	69.22 (10.36)
Female	112 (52.6%)	64 (56.6%)	3942 (78.8%)
Years of education	11.29 (4.63)	11.01 (4.11)	10.10 (4.27)
Makes ends meet	2.75 (1.03)	2.62 (0.92)	2.52 (0.98)
No. of children	1.72 (1.21)	1.68 (1.31)	1.79 (1.41)
No. of chronic diseases	1.55 (1.47)	1.42 (1.35)	2.12 (1.67)
Global Activity Limitation—limited in activities	38.1%	43.4%	55.9%
Self-perceived health	3.09 (1.09)	3.00 (1.15)	2.52 (1.04)
Survivors of Time 2 and Time 3	80	73	3678
Time 2
Age	65.92 (8.57)	66.71 (9.16)	72.86 (9.94)
Female	37 (46.3%)	43 (58.9%)	2955 (80.3%)
Makes ends meet	3.17 (0.95)	3.07 (0.91)	2.63 (1.01)
No. of children	1.66 (1.23)	2.15 (1.57)	1.76 (1.38)
CASP^a	39.38 (5.6)	39.43 (5.17)	35.83 (6.46)
Life satisfaction	7.89 (1.89)	7.96 (1.54)	7.18 (2.00)
Time 3
CASP^a	38.98 (6.45)	38.05 (5.17)	35.43 (6.75)
Life satisfaction	8.09 (1.60)	7.90 (1.28)	7.14 (2.02)

Note: Means or counts are shown with standard deviations or percentages in parentheses.

^a CASP is the quality of life and well-being.

Data analysis

Data analyses were conducted in R (R Core Team, 2019b). We used the “foreign” package (R Core Team, 2019a) to load the data; “DescTools” (Signorell, 2019), “psych” (Revelle, 2018), and “ggplot2” (Wickham, 2016) packages to explore and describe the data; and “nnet” (Venables & Ripley, 2002), “lme4” (Bates et al., 2015), and “lmerTest” (Kuznetsova et al., 2017) packages to conduct statistical analyses. We used a multinomial regression model to predict the establishment of a LAT relationship, a cohabitation relationship, or remaining unpartnered at T2, based on the health indicators measured at T1. This model was built with respect to the so-called selection model in which we entered Able to make ends meet as an ordinal variable in our models (due to its very short range) and all other scale variables as cardinal variables. To assess whether the relationship status had an effect on the quality of life, well-being, and life satisfaction measured at T3, we used two linear mixed models with quality of life, well-being, and life satisfaction as dependent variables. The linear mixed models were chosen over multiple linear regression models in order to better control for within-subject variance (i.e., change in time, correlation among respondent’s scores across T2 and T3). Given the fact that relationship status was treated as a time-invariant category for T2 and T3, autoregressive models could not be performed. In all of the analyses, we included age, sex, education, socioeconomic situation, and the number of children as control variables because these variables might be associated with health and well-being (as pointed out in the introduction). Only in the case of the original education variable—on which approximately 400 participants had a missing value—we used its imputed variable, which was available in the SHARE data set from Wave 4 (i.e., Time 1). Additionally, we controlled for the time change between T2 and T3 of the dependent variables in both of the linear mixed models. We set the LAT group as a reference category for the unpartnered and cohabitation groups in the models. Due to unbalanced sample sizes across the relationship status groups, we conducted a power analysis for this variable in both linear mixed models using a “simr” package (α = .05, Nsim = 1,000, Green & MacLeod, 2016). For the CASP variable, we set the minimum effect of interest to be two points (i.e., the sum of 12 items with a 4-point scale, a possible range of 12–48), and, for the life satisfaction, we set the minimum effect to be three-quarters of a point (i.e., one 11-point item, a possible range of 0–10). We expected to see a negative effect for the unpartnered group and a positive effect for the cohabiting group, as compared to the LAT reference group, in both models. The power analysis results suggest that we achieved very high power for the unpartnered group (i.e., 90.7% for CASP and 98.6% for life satisfaction). Although the estimated power was adequate for the cohabitation group in the life satisfaction model (82.5%), the CASP model was slightly underpowered (64.3%) in this regard. This means that our analysis had a low chance to find a statistically significant effect of cohabitation on CASP.

Results

Table 1 shows a greater prevalence of women in the unpartnered group (78% in T1 and 80% in T2) compared to a roughly balanced sex ratio in the LAT and cohabitation groups. Also, unpartnered people seemed to have consistently lower well-being and life satisfaction, as well as worse ability to make ends meet, compared to the other two groups.

With respect to the research question about the extent to which older people who entered into a LAT relationship differed in the health status before relationship formation (i.e., the selection model), we conducted the multinomial regression model. The results showed that older age (OR = 1.06), being female (OR = 2.79), and having fewer children (OR = 0.90) raised (p < .05) the odds of staying unpartnered rather than entering into a LAT relationship at T2 (see Table 2). In addition, having better self-perceived health (OR = 0.38) lowered the odds of staying unpartnered (p < .001) (see Table 2). Regarding the LAT versus cohabitation comparison, no variable discriminated between the groups, pointing to the lack of differences in the health indicators before relationship formation. The results only provide support for the assumption that those who were unpartnered had more health problems than those who entered in a LAT relationship.

Table 2.

Multinomial regression model.

Group^a		B	SE	Wald Z	p	Exp(B)
Unpartnered^b	(Intercept)	−0.981	0.582	−1.686	.092	0.375
	Education	−0.018	0.017	−1.079	.281	0.982
	Gender (female)	1.026	0.149	6.903	<.001	2.791
	Age	0.058	0.009	6.735	<.001	1.060
	Able make ends meet	−0.268	0.177	−1.516	.130	0.765
	No. of children	−0.109	0.053	−2.068	.039	0.897
	No. of chronic diseases	−0.027	0.057	−0.478	.633	0.973
	Global Activity Limitation	0.077	0.180	0.429	.668	1.080
	Self-perceived health	−0.967	0.284	−3.399	<.001	0.380
In cohabitation/married^c	(Intercept)	−0.386	0.961	−0.401	.688	0.680
	Education	−0.009	0.028	−0.337	.736	0.991
	Gender (female)	0.148	0.245	0.602	.547	1.159
	Age	−0.001	0.014	−0.072	.943	0.999
	Able make ends meet	−0.095	0.312	−0.306	.760	0.909
	No. of children	−0.050	0.088	−0.573	.567	0.951
	No. of chronic diseases	−0.135	0.099	−1.365	.172	0.874
	Global Activity Limitation	0.181	0.293	0.617	.537	1.198
	Self-perceived health	−0.158	0.457	−0.345	.730	0.854

Note: ^a Reference group—Living-Apart-Together (n = 212).

^b (n = 4912).

^c (n = 110).

With respect to the research question about the association between the relationship status and well-being and life satisfaction after relationship formation (i.e., the resource model), we conducted mixed linear model analyses. In the first mixed linear model analysis, we used relationship status to predict the quality of life and well-being (CASP) (see Table 3). In total, we defined three models: Model 0, without predictors; Model 1, with education, sex, age, ability to make ends meet, number of children, and time as control variables; and Model 2, with additional relationship status dummy variables. Model 2 was fitted using the Restricted Maximum Likelihood (REML) method for more accurate results. The LAT group was set as a reference group. After controlling for socioeconomic variables, we found no substantial differences in CASP between the LAT group and the unpartnered (B = 1.17, SE = 0.60, p = .052) or cohabitation groups (B = −0.27, SE = 0.87, p = .758). The inefficiency of the relationship status as a predictor of CASP in Model 2 is further illustrated by the practically unchanged random effect variances (see Table 3) and the lack of model-fit improvement (AIC_M1 = 44131, AIC_M2 = 44129, χ ²(2) = 5.53, p = .063). The analysis showed no support for differences in CASP according to the relationship status.

Table 3.

Mixed linear model for the quality of life and well-being (CASP) in Time 3.

Fixed effects	Model 0 (No predictors)					Model 1 (Control variables)					Model 2 (Final model)
Fixed effects	Estimate	SE	t	df	p	Estimate	SE	T	df	p	Estimate	SE	t	df	p
(Intercept)	35.802	0.100	356.7	3648	<.001	44.245	0.761	58.131	3692	<.001	45.160	0.924	48.861	3656	<.001
Years of education						0.084	0.022	3.865	3662	<.001	0.083	0.022	3.850	3652	<.001
Gender (female)						0.033	0.221	0.148	3650	.882	0.092	0.222	0.412	3641	.680
Age						−0.137	0.009	−14.617	3673	<.001	−0.135	0.009	−14.235	3662	<.001
Able make ends meet						5.995	0.201	29.794	3670	<.001	5.952	0.202	29.441	3659	<.001
No. of children						0.184	0.065	2.840	3672	.005	0.177	0.065	2.721	3663	.007
Change ^CASPT2→T3						−0.389	0.086	−4.542	3561	<.001	−0.389	0.086	−4.543	3560	<.001
Unpartnered^ab											−1.174	0.603	−1.946	3612	.052
Cohabitation^ac											−0.268	0.870	−0.309	3622	.758
Random effects^d	Variance			SD		Variance			SD		Variance			SD
Residual	13.13			3.624		13.04			3.610		13.04			3.611
ID	30.20			5.495		21.03			4.586		21.07			4.590

Note: ^a Reference group—Living-Apart-Together (n = 80).

^b (n = 3524).

^c (n = 68).

^d This section shows the amount of variance attributable to within-subject factors (labeled ID) and variance unexplained by the model (Residual).

In the second mixed linear model analysis, we used the relationship status to predict life satisfaction (see Table 4). We defined three models in the same manner as in the previous mixed linear model analysis to predict CASP. The final model was again estimated using REML. The LAT group members reported slightly higher life satisfaction compared to the unpartnered group (B = −0.47, SE = 0.18, p = .007), but there was no significant difference in life satisfaction between the LAT and cohabitation groups (B = −0.03, SE = 0.26, p = .899). However, the lack of change in the random effect variances (see Table 4) of Model 2 suggests a small effect of the relationship group on life satisfaction. Also, Model 2 showed only a slight, albeit statistically significant, improvement over Model 1 (AIC_M1 = 28596, AIC_M2 = 28588, χ ²(2) = 12.077, p = .002). The findings provide support only for the hypothesis that at T3 unpartnered persons had slightly poorer life satisfaction than those who entered into LAT relationship.

Table 4.

Mixed linear model for life satisfaction in Time 3.

Fixed effects	Model 0 (No predictors)					Model 1 (Control variables)					Model 2 (Final model)
Fixed effects	Estimate	SE	t	df	p	Estimate	SE	t	df	p	Estimate	SE	t	df	p
(Intercept)	7.190	0.029	248.6	3606	<.001	7.216	0.230	31.419	3659	<.001	7.588	0.278	27.257	3613	<.001
Years of education						−0.005	0.007	−0.729	3613	.469	−0.005	0.007	−0.747	3603	.455
Gender (female)						0.142	0.067	2.128	3597	.033	0.167	0.067	2.500	3588	.012
Age						−0.005	0.003	−1.836	3627	.066	−0.004	0.003	−1.418	3616	.156
Able make ends meet						1.659	0.061	27.349	3624	<.001	1.640	0.061	26.933	3612	<.001
No. of children						0.093	0.020	4.733	3626	<.001	0.089	0.020	4.548	3617	<.001
Change Life satisfaction^T2→T3						−0.014	0.033	−0.414	3536	.679	−0.014	0.033	−0.417	3535	.676
Unpartnered^ab											−0.486	0.181	−2.680	3550	.007
Cohabitation^ac											−0.033	0.262	−0.127	3564	.899
Random effects^d	Variance			SD		Variance			SD		Variance			SD
Residual	1.928			1.389		1.930			1.389		1.930			1.389
ID	2.076			1.441		1.527			1.236		1.525			1.235

Note: ^a Reference group—Living-Apart-Together (n = 80).

^b (n = 3524).

^c (n = 68).^d This section shows the amount of variance attributable to within-subject factors (labelled ID) and variance unexplained by the model (Residual).

Discussion

The aim of this study was to investigate health and well-being in association with Living-Apart-Together (LAT) relationships that are established at older age, while taking into consideration other relationship statuses, specifically cohabitation/marriage and remaining unpartnered. To outline the possible directions of the association, the study was guided by testing the selection and resource models. Specifically, we expected differences in health, well-being, and life satisfaction indicators among the people who entered into a LAT partnership, people who decided to cohabit with or marry a new partner, and people who remained unpartnered. The findings revealed few differences in the health and well-being markers among the studied groups, predominately between the LAT persons and the unpartnered people.

This study provides support for the selection model only for the instance where persons who establish a LAT relationship were compared to those who remained unpartnered. Even after controlling for age, LAT participants reported better self-perceived health when compared to unpartnered persons. This suggests that the way a person evaluates their physical health plays an important role in later-life relationship formation. An interpretation of this finding could be that better health may allow older people to be more active and engage in various social activities where they can make new acquaintances. It is also possible that older people may be motivated to seek potential partners who look and act in an invigorating way and who may fulfill the main function of the LAT relationship, which consists of sharing hobbies, spending leisure time together, and having fun (Benson & Coleman, 2016; Bildtgård & Öberg, 2015; Koren, 2014, 2015).

Interestingly, the other health markers, such as limitations in activities or chronic diseases, showed no role in later-life relationship formation. Although these markers are objective, the findings suggest that, while considering a new relationship, it may depend more on how persons interpret their physical conditions. As mentioned above, self-perceived health may influence a personal level of self-efficacy, thanks to which a person could be open to engaging in various activities and embark on establishing a new relationship.

The present study provided no support for the assumption that older people with health difficulties are more likely to opt for LAT relationships than cohabitation relationships or marriage. Those who entered into either a LAT living arrangement or cohabitation/marriage were similar in health markers before partnering. Since health indicators were measured at least 4 years before the identification of changes in relationship status, it is possible that health deterioration might not have been detected. The lack of the effects of health indicators in both partnered (i.e., LAT and cohabitation/marriage) groups could imply that later-life relationship formation is associated with the prospects that partnered persons will still be partners toward each other, meaning that the partners are for sharing activities and hobbies and not for providing care. This interpretation of the findings may correspond to the broader concept of the third age, which denotes older people’s tendency for putting an emphasis on autonomy (i.e., that may manifest by being independent of the care of others), staying active, and taking measures for alleviating infirmity (Gilleard & Higgs, 2007). In this respect, forming any new relationship in later life could be apparent in persons who are active and relatively healthy. It is also possible that LAT relationships could have become a more legitimate relationship form that started to be treated as equal to cohabitation. This interpretation should be taken with caution since more research is needed to understand older people´s attitudes toward LAT living arrangement.

This study provides some support for the resource model only in the case where persons who establish a LAT relationship were compared with those who remained unpartnered. Staying in LAT relationships predicted slightly higher scores on the life satisfaction scale when compared to those with a “single” status, but the effect was small after controlling for other covariates. Nonetheless, this finding corroborates prior qualitative studies that concluded that being in a LAT relationship is perceived to be more rewarding than remaining unpartnered in later life (Koren, 2015).

This longitudinal study provides no support for later-life LAT arrangements to be less resourceful than other cohabitation relationships/marriages, which indicates that these two relationship types were comparable in well-being and life satisfaction and differences in living arrangement, per se, do not seem to be associated with different effects on well-being and life satisfaction constructs. Other factors, such as socioeconomic situation, might be responsible for the studied outcome variables in partnered people. These findings are not in line with primarily cross-sectional research, according to which later-life LAT arrangements has been communicated as a less resourceful relationship type than other cohabitation relationships or marriage due to its association with a lower amount of perceived support or happiness (Lewin, 2017; Strohm et al., 2009). Nonetheless, it should be noted that the present study differed from prior work in the studied outcome variables. Furthermore, despite the large sample, relatively few participants partnered across the three study waves, which caused our estimates for these groups to be less accurate. Due to a lower number of partnered respondents, we were forced to jointly analyze cohabitating people and married persons, although the latter group is consistently the most likely to profit from the marriage type of partnership (Lewin, 2017; Schoenborn, 2004; Strohm et al., 2009; Umberson et al., 2013). This might represent another source of discrepancy between these findings and prior work. For cumulating research evidence, future research with a longitudinal study design should focus on other outcome variables, such as loneliness, in order to disentangle the extent to which a relationship with separate households may challenge the overall well-being of LAT people when compared to older persons in more traditional partnership forms.

Lastly, the present study also showed that later-life relationship formation (i.e., entering into a LAT relationship) was predicted by male sex and lower age. These findings corroborate prior research, according to which the chance of relationship formation decreases with age; this is especially true for older women who outnumber their male counterparts in the dating market and thus have a lower chance to (re)partner (Brown et al., 2016; de Jong Gierveld, 2004; Schimmele & Wu, 2016; Wu et al., 2015). This study used SHARE data that supports previous findings on the gender gap in the dating market because about 76% of unpartnered persons at the initial level were women.

Our study had several limitations. Despite the large-scale SHARE dataset, the data included relatively few people before and after partnering and in the follow-up analyses. This prevented the incorporation of a gender perspective into the data analyses (i.e., conducting the analyses separately for women and men). Therefore, we are not able to reliably estimate partnering trends for both female and male; due to the over-representation of women in the total sample, our results may be more applicable for females than males, if the partnering trends do considerably differ based on sex. Moreover, according to the power analysis, our study had a lowered chance to find a statistically significant effect for cohabitation/marriage on CASP. However, given that a theoretically well-powered unpartnered group effect in the CASP model was not significant either, we assume the observed lack of effect is not a Type II error. Furthermore, the LAT group was identified by a single item about having a partner outside the household without probing the level of relationship commitment because such a measure was not part of the SHARE dataset. Nonetheless, studies on later-life LAT relationships have shown that people find it inappropriate to use the term “partner” when in a dating relationship and not a serious LAT relationship (Benson & Coleman, 2016; Kobayashi et al., 2017). In addition, those LAT persons who remained in Time 3 (i.e., years later) were those who did not indicate ending the relationship and starting a new one between T2 and T3. Another study limit is that the dichotomization of longstanding limitations in activities did not allow us to probe the severity of this health problem. Lastly, this study focused on the relationship forms and the relationship status, and their effects on the well-being indicators. It did not take into consideration relationship quality, which might have varied within the studied living arrangements (Koren, 2014).

Despite these limits, this study was based on longitudinal data, which allowed the disentangling of possible causal links between health and relationship formation, as well as the effects of relationship status on the markers of well-being and life satisfaction. Upon these findings, several conclusions may be drawn. Older persons are no more likely to be in less committed LAT relationships due to their poorer health status. Better self-perceived health is positively associated with partnering in later life and staying in a LAT relationship could be beneficial to older people; however, its effect is rather limited. Hereby, LAT living arrangements may constitute a resourceful relationship form in later life in that the level of well-being and life satisfaction remain preserved when compared to cohabiting and married counterparts, indicating that a LAT relationship is not inferior to cohabitation or marriage. Nonetheless, future research should test this assumption with respect to other psychological factors, such as loneliness. This could be a relevant step in order to build cumulative evidence about the quality of later-life LAT relationships.

Footnotes

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by research grant no. GA20-25752S provided by the Czech Science Foundation. The SHARE data collection was primarily funded by the European Commission through FP5 (QLK6-CT-2001-00360), FP6 (SHARE-I3: RII-CT-2006-062193, COMPARE: CIT5-CT-2005-028857, SHARELIFE: CIT4-CT-2006-028812), and FP7 (SHARE-PREP: N°211909, SHARE-LEAP: N°227822, SHARE M4: N°261982). Additional funding came from the German Ministry of Education and Research, the Max Planck Society for the Advancement of Science, the U.S. National Institute on Aging (U01_AG09740-13S2, P01_AG005842, P01_AG08291, P30_AG12815, R21_AG025169, Y1-AG-4553-01, IAG_BSR06-11, OGHA_04-064, HHSN271201300071C), and various national funding sources, which are gratefully acknowledged (see ).

ORCID iD

Anna Ševčíková

Open research statement

As part of IARR’s encouragement of open research practices, the authors have provided the following information: This research was not pre-registered. The data used in the research are available. The data can be obtained at: . The materials used in the research are available. The materials can be obtained by emailing: asevciko@fss.muni.cz.

References

Bates

Mächler

Bolker

Walker

(2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67, 1–48. https://doi.org/10.18637/jss.v067.i01

Beekman

A. T.

de Beurs

van Balkom

A. J.

Deeg

D. J.

van Dyck

van Tilburg

(2000). Anxiety and depression in later life: Co-occurrence and communality of risk factors. American Journal of Psychiatry, 157, 89–95. https://doi.org/10.1176/ajp.157.1.89

Beekman

A. T.

Deeg

D. J.

van Tilburg

Smit

J. H.

Hooijer

van Tilburg

(1995). Major and minor depression in later life: A study of prevalence and risk factors. Journal of Affective Disorders, 36, 65–75. https://doi.org/10.1016/0165-0327(95)00061-5

Benson

J. J.

Coleman

(2016). Older adults developing a preference for living apart together. Journal of Marriage and Family, 78, 797–812. https://doi.org/10.1111/jomf.12292

Bildtgård

Öberg

(2015). Time as a structuring condition behind new intimate relationships in later life. Ageing & Society, 35, 1505–1528. https://doi.org/10.1017/S0144686X14000452

Börsch-Supan

(2018). Data from: Survey of Health, Ageing and Retirement in Europe (SHARE) Wave 6. Release version: 6.1.0. SHARE-ERIC. Data set. https://doi.org/10.6103/SHARE.w6.610

Börsch-Supan

(2019). Survey of Health, Ageing and Retirement in Europe (SHARE) Wave 4. Release version: 7.1.0. SHARE-ERIC. Data set. https://doi.org/10.6103/SHARE.w4.710

Börsch-Supan

(2020). Survey of Health, Ageing and Retirement in Europe (SHARE) Wave 7. Release version: 7.1.1. SHARE-ERIC. Data set. https://doi.org/10.6103/SHARE.w7.711

Brown

S. L.

Bulanda

J. R.

Lee

G. R.

(2005). The significance of nonmarital cohabitation: Marital status and mental health benefits among middle-aged and older adults. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 60, S21–S29. https://doi.org/10.1093/geronb/60.1.S21

10.

Brown

S. L.

Lin

I. F.

Hammersmith

A. M.

Wright

M. R.

(2016). Later life marital dissolution and repartnership status: A national portrait. The Journals of Gerontology Series, Series B, gbw051. https://doi.org/10.1093/geronb/gbw051

11.

Connidis

I. A.

Borell

Karlsson

S. G.

(2017). Ambivalence and living apart together in later life: A critical research proposal. Journal of Marriage and Family, 79(5), 1404–1418. https://doi.org/10.1111/jomf.12417

12.

de Jong Gierveld

(2004). Remarriage, unmarried cohabitation, living apart together: Partner relationships following bereavement or divorce. Journal of Marriage and Family, 66, 236–243. https://doi.org/10.1111/j.0022-2445.2004.00015.x

13.

de Jong Gierveld

(2015). Intracouple caregiving of older adults living apart together: Commitment and independence. Canadian Journal on Aging, 34, 356–265. https://doi.org/10.1017/S0714980815000264

14.

de Jong Gierveld

Merz

E. M.

(2013). Parents’ partnership decision making after divorce or widowhood: The role of (step) children. Journal of Marriage and Family, 75, 1098–1113. https://doi.org/10.1111/jomf.12061

15.

Duncan

Carter

Phillips

Roseneil

Stoilova

(2013). Why do people live apart together? Families, Relationships and Societies, 2, 323–338. https://doi.org/10.1332/204674313X673419

16.

Duncan

Phillips

(2010). People who live apart together (LATs)—How different are they? The Sociological Review, 58, 112–134. https://doi.org/10.1111/j.1467-954X.2009.01874.x

17.

Federal Statistical Office of Germany (2016). Older people in Germany and the EU 2016. https://www.destatis.de/EN/Publications/Specialized/Population/BrochureOlderPeopleEU0010021169004.pdf?__blob=publicationFile (accessed 25 March 2017).

18.

Funk

L. M.

Kobayashi

K. M.

(2016). From motivations to accounts: An interpretive analysis of “living apart together” relationships in mid- to later-life couples. Journal of Family Issues, 37, 1101–1122. https://doi.org/10.1177/0192513X14529432

19.

Gilleard

Higgs

(2007). The third age and the baby boomers. International Journal of Ageing and Later Life, 2(2), 13–30. https://doi.org/10.3384/ijal.1652-8670.072213

20.

Green

MacLeod

C. J.

(2016). SIMR: An R package for power analysis of generalized linear mixed models by simulation. Methods in Ecology and Evolution, 7(4), 493–498. https://doi.org/10.1111/2041-210X.12504

21.

Hyde

Wiggins

R. D.

Higgs

Blane

D. B.

(2003). A measure of quality of life in early old age: the theory, development and properties of a needs satisfaction model (CASP-19). Aging & Mental Health, 7(3), 186–194. https://doi.org/10.1080/1360786031000101157

22.

Kim

H. K.

McKenry

P. C.

(2002). The relationship between marriage and psychological well-being: A longitudinal analysis. Journal of Family Issues, 23, 885–911. https://doi.org/10.1177/019251302237296

23.

Kobayashi

K. M.

Funk

Khan

M. M.

(2017). Constructing a sense of commitment in “Living Apart Together” (LAT) relationships: Interpretive agency and individualization. Current Sociology, 65(7), 991–1009. https://doi.org/10.1177/0011392116653237

24.

Koren

(2014). Together and apart: A typology of re-partnering in old age. International Psychogeriatrics, 26, 1327–1350. https://doi.org/10.1017/S1041610214000738

25.

Koren

(2015). The intertwining of second couplehood and old age. Ageing & Society, 35, 1864–1888. https://doi.org/10.1017/S0144686X14000294

26.

Koren

Simhi

Lipman-Schiby

Fogel

(2016). The partner in late-life repartnering: Caregiving expectations from an intergenerational perspective. International Psychogeriatrics, 28, 1555–1565. https://doi.org/10.1017/S1041610216000181

27.

Kuznetsova

Brockhoff

P. B.

Christensen

R. H. B.

(2017). lmerTest package: Tests in linear mixed effects models. Journal of Statistical Software, 82(13), 1–26. https://doi.org/10.18637/jss.v082.i13

28.

Levin

(2004). Living apart together: A new family form. Current Sociology, 52(2), 223–240. https://doi.org/10.1177/0011392104041809

29.

Lewin

A. C.

(2017). Health and relationship quality later in life: A comparison of living apart together (LAT), first marriages, remarriages, and cohabitation. Journal of Family, 38, 1754–1774. https://doi.org/10.1177/0192513X16647982

30.

Malta

Farquharson

(2012). The initiation and progression of late-life romantic relationships. Journal of Sociology, 50, 237–251. https://doi.org/10.1177/1440783312442254

31.

R Core Team. (2019a). foreign: Read data stored by “Minitab,” “S,” “SAS,” “SPSS,” “Stata,” “Systat,” “Weka,” “dBase,”… https://CRAN.R-project.org/package=foreign

32.

R Core Team. (2019b). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/

33.

Régnier-Loilier

Beaujouan

É.

Villeneuve-Gohalp

(2009). Neither single, nor in a couple: A study of living apart together in France. Demographic Research, 21, 75–108. https://doi.org/10.4054/DemRes.2009.21.4

34.

Revelle

(2018). psych: Procedures for psychological, psychometric, and personality research. Northwestern University. https://CRAN.R-project.org/package=psych

35.

Ross

C. E.

Mirowsky

Goldsteen

(1990). The impact of family on health: The decade in review. Journal of Marriage and Family, 52, 1059–1078. https://doi.org/10.2307/353319

36.

Schimmele

C. M.

(2016). Repartnering after union dissolution in later life. Journal of Marriage and Family, 78, 1013–1031. https://doi.org/10.1111/jomf.12315

37.

Schoenborn

C. A.

(2004). Marital status and health: United States, 1999-2002. Advance data from vital and health statistics, 351. National center for health statistics. http://www.cdc.gov/nchs/data/ad/ad351.pdf

38.

Signorell

(2019). DescTools: Tools for descriptive statistics. https://cran.r-project.org/package=DescTools

39.

Strohm

Seltzer

Cochran

Mays

(2009). Living apart together: Relationships in the United States. Demographic Research, 21, 177–214. https://doi.org/10.4054/DemRes.2009.21.7

40.

Turcotte

(2013). Insights on Canadian society: Living apart together. Statistics Canada, 75-006-X. http://www.statcan.gc.ca/pub/75-006-x/2013001/article/11771-eng.htm

41.

Umberson

, (1992). Gender, marital status, and the social control of health behavior. Social Science & Medicine, 34, 907–917. https://doi.org/10.1016/0277-9536(92)90259-S

42.

Umberson

Thomeer

M. B.

Williams

(2013). Family status and mental health: Recent advances and future directions. In Aneshensel

C. S.

Phelan

J. C.

Bierman

(Eds.) Handbook of the sociology of mental health (pp. 405–431). Springer.

43.

van Oyen

Van der Heyden

Perenboom

Jagger

(2006). Monitoring population disability: Evaluation of a new Global Activity Limitation Indicator (GALI). Sozial-und Präventivmedizin, 51, 153–161. https://doi.org/10.1007/s00038-006-0035-y

44.

Venables

W. N.

Ripley

B. D.

(2002). Modern applied statistics with S (4th ed.). Springer. http://www.stats.ox.ac.uk/pub/MASS4

45.

Vespa

(2012). Union formation in later life: Economic determinants of cohabitation and remarriage among older adults. Demography, 49, 1103–1125. https://doi.org/10.1007/s13524-012-0102-3

46.

Wickham

(2016). ggplot2: Elegant graphics for data analysis. Springer-Verlag. https://ggplot2.tidyverse.org

47.

Schimmele

C. M.

(2005). Repartnering after first union disruption. Journal of Marriage and Family, 67(1), 27–36. https://doi.org/10.1111/j.0022-2445.2005.00003.x

48.

Schimmele

C. M.

Ouellet

(2015). Repartnering after widowhood. Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 70, 496–507. https://doi.org/10.1093/geronb/gbu060