Linked Lives: Does Disability and Marital Quality Influence Risk of Marital Dissolution among Older Couples?

Using fourteen waves of data from the Health and Retirement Study (HRS), a longitudinal panel survey with respondents in the United States, this research explores whether marital quality—as measured by reports of enjoyment of time together—influences risk of divorce or separation when either spouse acquires basic care disability. Discrete-time event history models with multiple competing events were estimated using multinomial logistic regression. Respondents were followed until they experienced the focal event (i.e., divorce or separation) or right-hand censoring (i.e., a competing event or were still married at the end of observation). Disability among wives was predictive of divorce/separation in the main effects model. Low levels of marital quality (i.e., enjoy time together) were associated with marital dissolution. An interaction between marital quality and disability yielded a significant association among couples where at least one spouse acquired basic care disability. For couples who acquired disability, those who reported low enjoyment were more likely to divorce/separate than those with high enjoyment; however, the group with the highest predicted probability were couples with low enjoyment, but no acquired disability.


Linked Lives: Does Disability and Marital Quality Influence Risk of Marital Dissolution among Older Couples?
Recent research has spotlighted the "graying of divorce" in the United States (Brown and Lin 2012). Although limited, prior empirical research has noted an increased risk of divorce following a serious health shock such as disability or illness onset among midlife and older adults (Blekesaune and Barrett 2005;Karraker and Latham 2015;Singleton 2012); however, other research has failed to find significant associations between poor health and divorce in older ages (Lin et al. 2018). These equivocal results may stem from differences in samples and measures of health. Alternatively, risk of divorce following a serious health shock may be conditioned on marital characteristics. To illustrate, Lin et al. (2018) examined antecedents of gray divorce and underscored the role of typical factors such as marital quality as the primary drivers of divorce among older adults. It is possible that disability or illness may increase the risk of divorce, but only among couples with certain marital characteristics; however, empirical evidence to date is quite limited.
The ability for a couple to adapt to major life stressors may be shaped by marital characteristics including marital quality (Bulanda et al. 2016;Umberson et al. 2005). The purpose of this research is to examine the role of marital quality in the risk of marital dissolution following disability (for either spouse) among different-sex older couples. We focus on basic care disability as measured by Activities of Daily Living (ADLs), which captures difficulty with bathing, eating, dressing, walking short distances, getting in and out of bed, and toileting. This type of impairment is closely linked with caregiving and is more common with advancing ages (Verbrugge and Jette 1994). We apply a life course A major tenet of the life course perspective is the concept of linked lives, which states, " [l]ives are lived interdependently and socio-historical influences are expressed through this network of shared relationships" (Elder et al. 2003, p. 13). The concept of linked lives recognizes that each individual experiences transitions and changes in the context of other relationships, and suggests that relationships with significant others are an essential consideration when studying life experiences (Elder et al. 2003). Indeed, Settersten (2015, p. 217) suggests that linked lives is a key theoretical perspective that needs to be more thoroughly incorporated into research; although linked lives encourages recognition that "individuals affect and are affected by others", much research still examines individuals and struggles to incorporate relational perspectives.
Moreover, marriages and families are key sites for gender performance and socialization. Traditional gender-role attitudes emphasize the supremacy of the husband breadwinner and wife homemaker family model and undergird differential power relations in families (Rogers and Amato 2000). Scholarship on marriage has long suggested "his" and "her" experiences of marriage (Bernard 1972) in which men and women have different appraisals of their relationship, with women tending to report lower levels of marital satisfaction (Boerner et al. 2014). These different marriage experiences are also linked with unequal health outcomes among older married men and women. For example, Bulanda et al. (2016) found that marital status was an important predictor of mortality among men with married men experiencing a clear health benefit, yet marital status did not matter for women. Instead, low marital quality was associated with higher risk of mortality for women.
According to Flood et al. (2018), married couples are primary groups whose relationships offer protective benefits relating to well-being and health. Using a linked lives perspective, they explore how time spent together is associated with marital quality among older adults. They conclude that couples spend more passive time together (time spent in physical proximity but not engaged in face-to-face activities) when the marriage feels more strained or less supportive. Conversely, couples experiencing low strain and high support spent more time in face-to-face activities; face-to-face time seems more driven by wives, as this type of togetherness increases particularly when the wife feels more support from the husband (Flood et al. 2018). Both husbands and wives reported higher levels of happiness when they engaged in face-to-face activities, as opposed to passive time together (Flood et al. 2018). This research underscores how time spent together is an important dimension of marital quality and provides insight into how the shared context of a marriage is best understood through a linked lives perspective.

Disability, Divorce, and Marital Quality among Older Couples
Although the divorce rate in the United States has been relatively stable overall, the divorce rate among older adults doubled between 1990 and 2010 (Brown and Lin 2012). In many ways, those who are older follow similar patterns as those who divorce when younger (Lin et al. 2018). Those who are in remarriages and have lower levels of education are more likely to divorce, and divorce patterns still reflect racial disparities in which those who are non-white are more likely to divorce (Lin et al. 2018). Furthermore, Lin et al. (2018) find that many of the factors associated with divorce at younger ages, such as marital duration and marital quality, are also associated with gray divorces. Thus, it is important to keep in mind that divorce in later life is a complex life transition that likely has numerous contributing factors. Divorce in later life may carry particular public health and policy implications, due to increased financial difficulties and adverse health experiences 3 of 15 (Brown and Lin 2012), and it is important to identify possible contributing factors specific to later life.
Many factors influence marital quality and decision-making regarding divorce. According to social exchange theory, decisions to divorce include appraisals relating to the costs and benefits of the relationship, barriers to exiting the relationship, and potential alternatives (Amato and Hohmann-Marriott 2007). For example, earlier in the life course, the presence of children has a "braking" or delaying effect on divorce (Waite and Lillard 1991;White et al. 1986). Biological grandchildren can exert a similar braking effect on gray divorce (Brown et al. 2021). In interviews of older adults who experienced divorce after age 50, Crowley (2019) documented several reasons for gray divorce including growing apart, spouse's mental health problems, infidelity, and conflicts over finances and children. People often anticipate that marriage will provide a buffer when they encounter health or financial shocks, but because of changing norms regarding marriage and divorce, the experience of those shocks may also increase the likelihood of divorce (Singleton 2012).
Research examining the influence of disability and illness on risk of divorce is limited and mixed. There is some empirical support that onset of life threatening conditions (e.g., cancer and heart disease) or other health problems is associated with increased risk of divorce among midlife and older adults (Joung et al. 1998;Karraker and Latham 2015;Teachman 2010;Wilson and Waddoups 2002); however, this association may be gendered where health declines from specific illnesses (e.g., heart problems or cancer) among wives, but not husbands, increased risk of divorce (Glantz et al. 2009;Karraker and Latham 2015;Kirchhoff et al. 2012). Yet, others have found no significant association between physical illness and divorce (Lin et al. 2018).
Research examining the risk of marital dissolution following disability, specifically, is also limited. Singleton (2012) demonstrated an elevated risk of divorce following work disability among a sample of working age midlife men (i.e., ages 30-54). Similarly, using data from the 1979 National Longitudinal Study of Youth (NLSY-79), Teachman (2010) found an increased risk of divorce following onset of work-related limitations among husbands, but not wives. However, Charles and Stephens (2004) found no support for an elevated risk of divorce following work disability.
We speculate that these mixed findings may reflect differences in sample ages and type of health condition or impairment, and, furthermore, risk of marital dissolution following a health shock may be contingent on marital characteristics. In other words, we posit that couples with high levels of marital quality, across multiple dimensions, are more likely to stay married as these marital characteristics indicate shared resources and contexts. High levels of marital quality may act as a buffer against relationship strain that can accompany major health shocks or the taking on of caregiving responsibilities within couples (Bookwala 2012;Carr et al. 2014).
Although we conceive marital quality broadly, this research focuses on enjoying time together. Previous research demonstrates that enjoying time together is a strong and robust predictor of divorce among older couples (Lin et al. 2018) and that those who spend more active time together report better well-being (i.e., happiness and less stress) (Flood et al. 2018;Genadek et al. 2019). According to Flood et al. (2018), time spent together is an indicator of couple-level adaption strategies, where couples who experience high levels of strain are more likely to spend less time together. As major work and family transitions occur in later life and leisure time increases, some older couples may grow apart or find that they do not enjoy time together (Brown and Wright 2017). In these instances, their risk of divorce is substantially increased (Lin et al. 2018). Yet, less is known about how measures of marital quality, such as enjoying time together, may interact with other later-life transitions like age-related acquired disability to either increase or decrease risk of divorce or separation.

Basic Care Disability, Caregiving, and Marital Dynamics among Older Different-Sex Couples
Prior empirical research has documented that divorced/separated older adults are more likely to have a physical disability relative to their married peers (Connolly et al. 2017;Picavet and Hoeymans 2002). Although this line of research does not examine the direction of the relationship (i.e., disability causing divorce versus divorce causing disability), it is plausible that this relationship exists, in part, due to increased risk of divorce/separation following disability onset. Moreover, acquired basic care disability may elevate risk of divorce because this type of impairment is often associated with caregiving, which can have a destabilizing effect on relationships depending on the context of the relationship (Penning and Wu 2019).
Research has indicated that men and women may have different patterns in carework, including aspects such as overall time spent in carework and the types of tasks included (Allen et al. 1999). Research into caregiving and union dissolution has documented an elevated risk of divorce when older wives take on the primary caregiver role for their spouses (Penning and Wu 2019). Yet, other research highlights the potential for caregiving to strengthen relationships. Yorgason et al. (2008) found increases in marital quality among older husbands whose wives acquired disability. The complex and dynamic processes related to gender, acquired disability, caregiving, and relationship quality among older adults were captured by Thomeer and Clark (2021) through interviews with older couples; they identify three patterns of caregiving. In the first model, gender inequities are present in that women provide support when husbands experienced health issues, but men provided less support when their wives experienced health issues. The second pattern was one in which caregiving responsibilities had more mutuality, and the third was one of independence in which neither spouse provided much support. Using a life course perspective, Thomeer and Clark (2021) note that patterns of caregiving are established earlier in relationships, but continue when health issues arise in later life. Thus, long-term marital dynamics may shape caregiving patterns and decisions relating to divorce/separation in later life. Keating et al. (2019) explore theoretical connections between family caregiving and life course perspectives. In their review, they draw a distinction between care as doing and care as being in a relationship. In addition to a focus on caregiving as engaging in specific behaviors, Keating et al. (2019) suggest that more attention needs to be given to the shifts that might occur in relationships as caregiving responsibilities increase. Over the course of care, the dyad relationship between the carer and the care-recipient likely undergoes significant shifts. Although caregiving might enhance some relationships through deepening bonds, it can also increase marital strain (Keating et al. 2019).
Likewise, Noël-Miller's (2011) exploration of caregiving among cohabiting and married older couples highlights possible gender effects in caregiving patterns. Both specialization of tasks and gender role socialization theories contribute to the development of gender roles in which men tend to focus more on outside employment and earnings, whereas women focus more on nurturing and caring for the family and home. Because of this foundation and other patterns associated with aging and disability, older women are more likely to care for a frail spouse (Noël-Miller 2011). Additionally, Noël-Miller's (2011) findings suggest that strength of relationship may affect caregiving behaviors; in this study, cohabiting partners with a disability were less likely to receive partner care but, when cohabiting partners did provide care, they provided approximately the same number of hours as married partners. Thus, factors such as relationship quality, emotional closeness, sense of relationship permanence, and interpersonal commitment could influence decisions related to caregiving behaviors and/or to remain in the relationship. Therefore, we examine whether risk of divorce/separation increases following acquired basic care disability among wives and husbands and whether this association is moderated by marital quality (i.e., enjoying time together).

Summary of Hypotheses
The purpose of this research is to explore whether acquired disability (for either spouse) increases risk of divorce/separation among a sample of older different-sex couples. Based on previous research, we have developed the following hypotheses: Hypothesis 1. Disability among wives will be associated with increased risk of divorce/separation. Because previous research has noted that wives' health status is a more important predictor of marital dissolution, we anticipate that wives' disability will increase risk of divorce.
Hypothesis 2. Couples with high levels of marital quality (i.e., enjoy time together) will have a decreased risk of divorce/separation. Hypothesis 3. The effect of acquired disability on divorce/separation risk will be moderated by marital quality. We posit that marital quality will buffer relationship strain that may accompany acquired basic care disability.

Data
Data for this research come from Waves 1 through 14 (1992-2018) of the Health and Retirement Study (HRS 2021). The HRS is an ongoing nationally representative longitudinal survey in the United States that uses a multi-stage, clustered probability frame. It is primarily focused on the health, finances, and employment history of a late midlife cohort (b. 1931-1941). Data collection occurs every two years. Because this research is interested in risk of marital dissolution among late midlife couples (i.e., 51-61 years), the sample was restricted to married different-sex couples in Wave 1 (1992). Spouses were interviewed regardless of age-meaning that each couple had at least one member aged 51-61 years in Wave 1. Additionally, we further restricted the sample to couples without ADL disability in Wave 1 and with valid marital history and biography information yielding a sample of 3965 couples.

Measures
For Waves 2-14, basic care disability was measured using six common Activities of Daily Living (ADL) tasks: (1) difficulty bathing; (2) difficulty dressing; (3) difficulty eating; (4) difficulty getting in/out of bed; (5) difficulty walking across the room; (6) difficulty toileting. Respondents indicating any difficulty, including "can't do" or "don't do," with at least one ADL task were considered to have basic care disability. For Wave 1, respondents were only asked about five ADL tasks (i.e., bathing, dressing, eating, getting in/out of bed, and walking). Similar to Waves 2-14, a dichotomized measure, where "any difficulty" = 1 and "no difficulty" = 0 was used. Toileting was not asked in Wave 1, so the inclusion criterion that couples did not have disability at baseline was based on the other five ADL tasks. Difficulty toileting was the least common ADL impairment with only two husbands reporting difficulty toileting in Wave 2 and no wives reporting any difficulty.
Disability was measured as a time-varying variable that was lagged one survey interval. Although there are many valid criticisms of ADLs as a measure of disability including an overemphasis on clinical aspects of bodily impairment (see Verbrugge and Jette 1994), ADL disability represents difficulty completing tasks essential for daily living. Compared with other indicators of functional health, basic care disability is often viewed as a loss of independence by individuals, which has important implications for both the individual with disability as well as the spouse. Basic care disability is closely tied to caregiving need and related to personal, often intimate, activities; therefore, it is a valuable measure for these analyses. We created a separate disability measure for each spouse based on gender (i.e., wives' acquired disability and husbands' acquired disability) as well as a couple-level measure, where at least one spouse acquired disability. We created the couple-level measure to combat concerns about statistical power given that divorce was a rare outcome.
The independent variable of interest was marital quality, which was measured using a variable assessing how much respondents enjoy spending time together. In Wave 1, respondents and spouses were asked "Generally speaking, would you say that the time you spend together with your husband/wife/partner is extremely enjoyable, very enjoyable, somewhat enjoyable, or not too enjoyable?" We reverse coded the measure, so that higher values reflected more enjoyment. Next, we averaged the husband and wife's reports of marital quality. To minimize missing data, when only one respondent had valid marital quality information, we used the information provided by the one spouse. Fewer than 10% of couples did not have valid information for both spouses-this was mostly due to the use of proxy interviews in Wave 1. Because most couples reported high levels of enjoyment, a binary indicator of "low enjoyment" was created, where couples whose enjoyment was one standard deviation below the mean (i.e., 2.56 or lower) were classified as having low enjoyment.
We adjust for several key covariates that could account for an observed relationship among marital quality, disability and divorce. Spousal homogamy measures were timefixed and only measured at baseline. A continuous measure (in years) of husbands' age (range: 50-83) and spouses' age difference (range: −30-36) was included in the analyses. A three category variable was created to capture the ethnoracial couple-level characteristics, where both spouses identified as white (reference), both spouses identified as nonwhite (including Black/African American, Hispanic/Latinx, or other race), and one spouse identified as nonwhite and one identified as white. A continuous measure of husbands' education (range: 0-17 years) was employed, and a measure of spouses' education difference was created by subtracting husbands' education from the wives' education (range: −13-16). Total household wealth was measured using four categories with low wealth (assets totaling fewer than 50,000 United States dollar (USD)), low-middle (50,000-99,999 USD), upper-middle (100,000-249,999 USD), and upper (250,000 USD or more). Wealth was measured at baseline and the dollar amounts reflect USD in 1992. Marital characteristics were measured at baseline (i.e., time-fixed). A dichotomous indicator of remarriage (=1) was created. Additionally, marriage duration was measured in years and ranged from 0.10 to 53.2.
Among the time-varying covariates, a measure of logged total household income included. Additionally, similar to Lin et al. (2018), we included two measures of life course transitions related to family and work: (1) empty nest; and (2) husband currently works for pay. Both measures were dichotomized, where living alone with your spouse was classified as having an "empty nest" and any amount of paid labor was classified as working for pay among husbands. We include a measure of "empty nest" to account for any potential braking effect that may accompany having children in the home. Finally, we included an additional health indicator to help disentangle the potential effect of basic care disability on risk of divorce, above and beyond general health concerns. We included a measure of self-rated health (SRH) for both husbands and wives. Self-rated health was reverse coded, so that higher scores reflected better health (range: 1 = poor to 5 = excellent). Finally, a dichotomous measure was created to capture proxy interviews, where proxy interview was used for husband = 1.

Analytic Strategy
Discrete-time event history models with multiple competing events were estimated using multinomial logistic regression (see Allison 1982Allison , 2014. Because this research examines marital dissolution (i.e., divorce or separation) in later life, it is vital to address selection bias due to death and attrition. The competing risks modeling approach allows for an examination of the risk of divorce and separation conditional on both spouses remaining alive and in the study. We employed a closed-cohort design. The risk group at baseline (Wave 1) were married couples and possible outcomes included: (1) continuously married (reference); (2) divorce or separated (or remarriage/repartnering between waves); (3) respondent died or spouse died (widowed); and (4) attrited. Respondents were followed over the observation window until they experienced the focal event (i.e., divorce or separation) or right-hand censoring (i.e., a competing event or were still married at the end of observation) (Brown and Warner 2008). Similar to Karraker and Latham (2015), we created a marriage-period dataset, where each married couple contributed to the marriage period until they experienced the focal event or right-hand censoring. The best fitting parameterization of period (i.e., length of observation) was linear. We present multinomial coefficients for divorce/separation with continuously married as the referent group. In Supplemental Table S1, we present the coefficients for the competing event of widowhood/death. Analyses were conducted using SAS 9.4. To adjust for clustering due to repeated measures, we used robust standard errors via the PROC CLUSTER statement. Using PROC SURVEYLOGISTIC, analyses were weighted using the Wave 1 HRS household-level weights to adjust for oversampling and complex survey design. Table 1 presents marital status by wave among couples without basic care disability at Wave 1. At baseline, 3965 couples were currently married and had valid marital biography and quality information. From Wave 1 to Wave 2, 56 couples divorced, 84 were widowed, and 270 attrited (i.e., lost to follow up). Across the 14 waves, divorce and attrition became less frequent, while widowhood become more frequent. By the end of the observation period, 18.1 percent of the couples were still alive and married, 5.3 percent had experienced divorce or separation, 43.5 percent had been widowed, and 33.1 percent had attrited.

Sample Characteristics of Analytic Sample
The analytic sample represents the marriage-period dataset. Table 2 presents the sample characteristics for the 29,487 marriage periods. Of the potential observations, 285 couples were dropped due to nonpositive weights. For the outcome measure, divorce/separation represented only 1 percent of the events, whereas about 89 percent of couples were continuously married. Another 6 percent of couples experienced widowhood, and 4 percent attrited. Prior research estimates that 10 in 1000 older adults (or approximately 1%) experience gray divorce (Brown and Lin 2012), which is reflected in our analytic sample.
Among the time-invariant measures, about 19% of the couples reported low enjoyment. In general, marital quality was quite high with an average level of enjoyment of time together of 3.16 out of 4. When examining wives and husbands' ratings of enjoyment separately, wives had slightly lower ratings. For the spousal homogamy measures, the average age of husbands was 56.81 years and couples had an average age difference of 3.81 years. About 87 percent of couples were white. Husbands' average educational attainment was nearly 13 years, and there were, on average, small education differences within couples. About one-third of the couples were low or low-middle wealth. More than a quarter of husbands were remarried, and the average length of marriage at baseline was nearly 28 years. Husband currently works (yes = 1) 0.51 0-1 Empty nest (yes = 1) 0.72 0-1 Self-Rated Health: Husband's self-rated health 3.40 1.08 1-5 Wife's self-rated health 3.54 1.04 1-5 Proxy interview (Husband) 0.10 0-1 Note: a USD = United States dollars; Means and proportions weighted using Wave 1 household-level weights.
For the time-varying covariates, seven percent of husbands acquired basic care disability, while 8 percent of wives acquired disability. At the couple-level, 14 percent of couples had at least one spouse acquire disability. The average logged household income was 10.82. A little over half of husbands currently worked for pay and the majority-nearly three-quarters-of couples were empty nesters. The average self-rated health for husbands was 3.40 with wives reporting slightly higher self-ratings of health with an average of 3.54. About 10 percent of interviews for husbands were completed by proxy interviewees. Table 3 presents the estimated log odds coefficients and standard errors of divorce/ separation from multinomial logistic regression. Three models are presented for genderspecific acquired disability: (1) main effects; (2) low enjoyment*wife acquired disability;

Risk of Divorce/Separation among Late Midlife Couples
(3) low enjoyment*husband acquired disability. Two additional models are presented (i.e., Models 4 and 5) with a couple-level acquired disability main effects model (Model 4) and interactional effects model (Model 5) testing low enjoyment*couple-level disability. In Model 1, low enjoyment was strongly associated with risk of divorce/separation. Among couples with low ratings of enjoying time together, their odds of divorce/separation were four times that of couples with high ratings of enjoyment. If wives acquired disability, there was an increased risk of divorce/separation at an alpha-level of 0.10. Among the covariates, husband's age and marital duration were negatively associated with risk of divorce, while age difference and race or ethnicity differences among couples were positively associated with divorce/separation.   In Model 2, we introduced the interaction for low enjoyment*wife acquired disability. Using an alpha-level of 0.10, the interaction term was negatively associated with divorce/separation risk. In Model 3, we see a similar magnitude for the interaction term between low enjoyment*husband acquired disability. Models 2 and 3 provide suggestive evidence that risk of divorce/separation following wives or husbands acquiring basic care disability varies by marital quality. However, given the large standard errors due to divorce being a rare event, we elected to focus our interpretation of the interactions on the couplelevel disability measure, which were more robust. In Model 4, low enjoyment continued to be a strong predictor of divorce/separation risk. Similarly, couple-level acquired disability was associated with higher risk of divorce/separation using an alpha-level of 0.10. In Model 5, the interaction between low enjoyment and couple-level acquired disability was significantly associated with marital dissolution.
To aid in the interpretation of the interaction, Figure 1 displays the predicted probabilities of divorce/separation-with all covariates set to the mean value-by marital quality and couple-level acquired disability status. Although divorce/separation risk was low across all groups, among couples without acquired disability, those reporting low enjoyment were five times more likely to divorce/separate than those reporting high enjoyment. Among couples who had at least one spouse acquire basic care disability, the predicted probabilities were more similar among the low and high enjoyment groups. Surprisingly, the group with the highest predicted probably were couples with low enjoyment, but no acquired disability.
In Model 2, we introduced the interaction for low enjoyment*wife acquired disability. Using an alpha-level of 0.10, the interaction term was negatively associated with divorce/separation risk. In Model 3, we see a similar magnitude for the interaction term between low enjoyment*husband acquired disability. Models 2 and 3 provide suggestive evidence that risk of divorce/separation following wives or husbands acquiring basic care disability varies by marital quality. However, given the large standard errors due to divorce being a rare event, we elected to focus our interpretation of the interactions on the couple-level disability measure, which were more robust. In Model 4, low enjoyment continued to be a strong predictor of divorce/separation risk. Similarly, couple-level acquired disability was associated with higher risk of divorce/separation using an alpha-level of 0.10. In Model 5, the interaction between low enjoyment and couple-level acquired disability was significantly associated with marital dissolution.
To aid in the interpretation of the interaction, Figure 1 displays the predicted probabilities of divorce/separation-with all covariates set to the mean value-by marital quality and couple-level acquired disability status. Although divorce/separation risk was low across all groups, among couples without acquired disability, those reporting low enjoyment were five times more likely to divorce/separate than those reporting high enjoyment. Among couples who had at least one spouse acquire basic care disability, the predicted probabilities were more similar among the low and high enjoyment groups. Surprisingly, the group with the highest predicted probably were couples with low enjoyment, but no acquired disability.

Sensitivity Analyses
As an additional robustness check, we completed parallel analyses on all couples regardless of disability status at baseline. In Supplemental Table S2, the same five models are presented. With the added power from including couples who had basic care disability at baseline, the results show similar coefficients, but smaller standard errors-lending more confidence to the findings. In these analyses, wives' disability status was associated

Sensitivity Analyses
As an additional robustness check, we completed parallel analyses on all couples regardless of disability status at baseline. In Supplemental Table S2, the same five models are presented. With the added power from including couples who had basic care disability at baseline, the results show similar coefficients, but smaller standard errorslending more confidence to the findings. In these analyses, wives' disability status was associated (p = 0.02) with divorce/separation risk at a more conventional alpha-level of 0.05, while husbands' disability status was not. While we still interpret this finding with caution, we view this as suggestive evidence that wives' disability is more predictive of divorce/separation risk than husbands' disability among older couples. Other supplemental analyses (not shown) included additional controls including marital satisfaction. The substantive findings were the same. In Supplemental Table S3, number of chronic conditions (ever diagnosed) was added. Wives' disability status continued to be significantly associated with risk of marital dissolution. Finally, all models were run using a continuous measure of enjoying time together. Again, the substantive findings were identical.

Discussion
Using a life course perspective and employing the concept of linked lives, the main purpose of this research was to examine whether acquired basic care disability and marital quality (i.e., enjoy time together) influenced risk of divorce/separation among older different-sex couples. Furthermore, we investigated whether marital quality buffered the risk of divorce/separation following acquired disability. As documented in previous research, divorce/separation was rare among our sample. About 5.3% of older couples experienced a divorce by the end of our observation period.
In relation to our first hypothesis, we found strong support that marital quality, as measured by enjoying time together, predicted risk of marital dissolution among older couples. Similar to Lin et al. (2018) study of the antecedents of gray divorce, we echo that gray divorce is similar to more youthful divorces, where marital quality is a strong and robust predictor. Despite being measured only at baseline, we found compelling evidence that couple-level reports of enjoying time together predicted risk of divorce/separation. For couples who reported having low enjoyment, the odds of divorce/separation were quadrupled. Like previous work among older couples, we find that enjoying spending time together signals a high-quality marriage that is less likely to end in divorce or separation.
For our second hypothesis, in the gender-specific acquired disability models, we observed an increased risk of divorce among couples whose wives had basic care disability. This finding is in no way definitive, but given the rarity of divorce and the large standard errors, we believe that this suggestive evidence adds to the current literature that has examined the gendered nature of health status and divorce risk. Although evidence is mixed, previous research has suggested that wives' health status is a larger driver of marital dissolution compared with husbands' health status (see Glantz et al. 2009;Karraker and Latham 2015;Kirchhoff et al. 2012). This may be true with regard to disability status among older married different-sex couples.
Interactional analyses provided support for the moderating effect of marital quality on disability and risk of marital dissolution. However, these findings provided only partial support for our original hypothesis. We posited that high levels of marital quality would mitigate divorce/separation risk among couples who acquired basic care disability. Among the couples who had at least one spouse acquire basic care disability, the predicted probability of divorce was slightly lower among those with high levels of enjoyment. Yet, among those with low enjoyment, it was the couples without basic care disability that had the highest risk of divorce. It is possible, due to left-hand censoring, that the couples who had acquired basic care disability had already experienced health-related strain and subsequent divorce or separation. In sensitivity analyses, we control for baseline chronic conditions as an attempt to address this issue. The substantive findings were the same (see Supplemental  Table S3). An alternative explanation is that couples without basic care disability may have higher expectations regarding spending time together. If these expectations are not met, then it is possible that spouses will seek to end the marriage or separate. Another potential explanation may come from the added closeness that can arise from supportive caregiving arrangements among spouses (Yorgason et al. 2008). Regardless of these explanations, our findings suggest that risk of marital dissolution following acquired basic care disability is conditional on marital quality-underscoring that it is not just experiencing changes to health or impairment that elevates risk of divorce/separation, but the context in which these changes happen within couples.
Previous qualitative research reiterates this claim. For example, Racher (2002, pp. 35-36) raises interesting questions about the functionality of the dyad unit and linked lives.
Although not addressing gender and ability directly, the study included both men and women who were primary caregivers and focused on the synergism among frail older couples. Some couples were able to develop "role reciprocity and interdependence," whereas others were characterized by "friction, strain, and frustration." Racher (2002) suggests that, among the former, the strength and functionality of the dyad was enhanced such that the couple could navigate challenges beyond what the individual partner could handle on their own. In some cases, this was manifested in a division of emotional labor and physical contributions. In this line of research, the focus is the resource that the dyadic unit becomes for those individuals within the unit. Allen et al. (1999) also highlight the significance of relationship dynamics, suggesting that an underlying assumption of spousal availability for caregiving in later life is pervasive. However, this fails to account for diversity in relationship dynamics, as not all relationships are supportive and not all spouses would automatically be appropriate caregivers.
This research contributes to the extant literature in several ways. First, we explored the importance of marital quality and acquired disability among older different-sex couples. The research examining the role of acquired disability is limited. Moreover, research exploring how marital characteristics shape risk of divorce/separation following the onset of disability or health-event is even more limited. The current literature has generated mixed results as to whether health factors shape risk of marital dissolution in later life; however, few studies have looked at disability, specifically, rather than chronic conditions. In supplemental analyses (see Supplemental Table S3), when chronic conditions are entered into the model, wives' disability was associated with marital dissolution. For number of ever diagnosed chronic conditions, which included heart disease, high blood pressure, diabetes, cancer, lung disease, stroke, psychiatric problems, and arthritis, wives' number of conditions was associated with marital dissolution at an alpha-level of 0.10. This is suggestive evidence that wives' disability status and chronic conditions may be independent predictors of marital dissolution among older couples.
For older, married different-sex couples where the woman acquires disability, there may be an increased risk of marital dissolution. Compared with other health indicators, basic care disability may uniquely contribute to risk of marital dissolution because of its association with caregiving. This research provides renewed, albeit tempered, support that wives' health is more important for predicting divorce or separation among older couples. Yet, this may not be the full story. We also observed that marital quality shaped risk of divorce/separation if couples had at least one spouse acquire basic care disability. This finding moves the literature forward by examining the potential effect of disability on marital dissolution among older adults in the context of marital characteristics. Although we are not able to capture change in marital quality over time, we believe that this research speaks to the potential for marital characteristics that are often developed early in a relationship to have a lasting influence on shared resources and understanding, which, in turn, shapes the graying of divorce.

Conclusions
There were several limitations worth noting. First, divorce is a rare event among older adults. We attempted to address this issue by completing multiple sensitivity analyses and weighing the evidence holistically. However, we still encourage caution when interpreting these results. Another major limitation stems from the inability to measure marital quality over time. The HRS only asked couples one time about marital quality. We are only able to include a baseline measure of marital quality, yet marital quality is shaped by health and may change throughout the course of a relationship (Bulanda 2011). Finally, because the original HRS cohort was focused on late midlife Americans, many of the couples were already married for a long period prior to first observation-leading to long-term marriages being overrepresented in the analytic sample. We believe this research provides a foundation for future research to explore the influence of marital quality (along with other marital characteristics) on divorce/separation risk following disability, yet we recognize the inherent data limitations.
Another important limitation of this study is an inability to tap into the nuances of lived experiences. For instance, we do not have data regarding the divorce process, such as who initially filed for the divorce. Prior research has noted that wives are more likely to initiate divorce (Brinig and Allen 2000;Kalmijn and Poortman 2006); however, this body of research has focused on younger populations. If acquired disability among wives is indeed a predictor of divorce/separation as suggested by our findings, then it is possible that this reflects "his" divorce in older ages given the more favorable remarriage markets for men in older ages (England and McClintock 2009). Alternatively, it may be that wives are electing to leave their husbands because their disability requires care that is not being adequately provided by their husbands. This would be similar to prior research into different-sex couples, which underscores that chronically ill wives may be dissatisfied with the care that they receive form their husbands because of differences in care-expectations by gender (Umberson et al. 2016).
This research indicates several directions for continued research. First, future research should move beyond the dyadic unit when exploring the nuances of linked lives, relationship transitions, and disability onset in later life. Although we posit that acquired basic care disability is closely tied to caregiving among couples, we do not specifically examine care-providing and care-receiving behaviors and attitudes. Among older couples with disability, spousal caregiving is quite common and often a sole endeavor with spouses reporting completing more high intensity carework compared with other caregiving arrangements (AARP and National Alliance for Caregiving 2020). Yet, spousal caregiving, even solo caregiving, may look very different for couples with strong social support networks. We encourage future research that examines caregiving and marital dissolution to address whether supportive social networks enable older adults to adapt to changes in the relationship and avoid divorce/separation.
Additionally, we note that the consequences of gray divorce on other social relationships remains understudied; however, there are a few examples of empirical studies related to this area. For example, Brown et al. (2019) explore how economic factors, health, and social ties impact patterns in repartnering after gray divorce. In this study, social ties did not impact women's likelihood of repartnering, and men were slightly more likely to enter a cohabiting partnership if they had few relatives or close friends nearby. Similarly, Van Gasse and Mortelmans (2020) explore how social support influences transitions after divorce, highlighting the role of social networks during challenging life transitions. Examining the intersection of disability onset, relationship quality, and social networks is an important next stage of research.
Future research could look into these questions with qualitative work-investigating the general patterns highlighted here by listening to the narratives of older different-sex couples. We found that marital quality moderated the relationship between acquired disability and risk of marital dissolution; however, we are hindered by secondary data to investigate how and why it matters. We draw on previous theoretical and empirical research to help explain our findings, but future research must extend this line of research by giving voice to older couples' experiences to understand how acquired disability influences marital characteristics and vice versa-particularly in light of the graying of divorce in the United States. Data Availability Statement: All data products are available to registered users at https://hrs.isr. umich.edu/.

Conflicts of Interest:
The authors declare no conflict of interest.