A New Look at the Living Arrangements of Older Americans Using Multistate Life Tables

Abstract

Objectives

We extend existing research on the living arrangements of older Americans by focusing on geographic proximity to children, examining transitions in living arrangements across older ages, and describing differences by both race/ethnicity and educational attainment.

Method

We use data from the Health and Retirement Study (HRS) over a period of 10 years (2000–2010) to construct multistate life tables. These analyses allow us to describe the lives of older Americans between ages 65 and 90 in terms of the number of expected years of life in different living arrangements, reflecting both mortality and living arrangement transitions.

Results

Americans spend a substantial proportion of later life living near, but not with, adult children. There is a good deal of change in living arrangements at older ages and living arrangement-specific life expectancy differs markedly by race/ethnicity and educational attainment. However, overall life expectancy is not strongly related to living arrangements at age 65.

Discussion

Multistate life tables, constructed separately by sex, race/ethnicity, and educational attainment, provide a comprehensive description of sociodemographic differences in living arrangements across older ages in the United States. We discuss the potential implications of these differences for access to support and the exacerbation or mitigation of inequalities at older ages.

Research on living arrangements at older ages has a long history in social gerontology and demography. This work has paid particular attention to levels and correlates of intergenerational coresidence, reflecting the importance of support provided by coresiding family members for economic, emotional, and physical well-being at older ages. In the context of increased longevity, growing disparities in both health and employment circumstances across the life course, and increased individual responsibility for later-life well-being, living arrangements are likely to play a critical role in shaping processes of inequality and stratification at older ages. For these reasons, the United Nations has identified living arrangements of older persons and possible government responses as one of the most pressing issues of population aging (United Nations, 2001).

Although the body of research on living arrangements at older ages is large, it is limited in two fundamental ways. First, conventional conceptualization of living arrangements has emphasized the distinction between intergenerational coresidence with children and other arrangements, typically ignoring residential proximity to adult children. This is problematic in that proximate residence is a key feature of contemporary family living arrangements that may provide many of the posited benefits of coresidence without some of its posited disadvantages (e.g., lack of privacy; Zhang, Engelman, & Agree, 2013). Studies that do consider proximity to adult children are limited by a tendency to focus on distance from children (with coresidence being zero distance), an approach that precludes an understanding of close proximity as a qualitatively different living arrangement with distinct correlates (Compton & Pollak, 2015).

Second, few studies have examined transitions in living arrangements over an extended period of time or summarized these transitions in the form of living arrangement-specific life expectancies. Past work has typically focused on correlates of living arrangements in the cross-section or has examined change in living arrangements over short periods of time. The only other analysis of life expectancy in different living arrangements of which we are aware (Wilmoth, 1998) was based on older data, used a relatively small sample of older individuals, covered a short period of time, did not focus on proximate residence, and did not consider differences across racial/ethnic groups or socioeconomic status. We thus know little about how the duration of different types of living arrangements across older ages is related to key sociodemographic characteristics or important dimensions of well-being (e.g., economic need, physical health, emotional health).

We begin to address these limitations by using data from the Health and Retirement Study (HRS) over a period of 10 years (2000–2010) to construct multistate life tables of living arrangements between ages 65 and 90. These analyses allow us to describe the lives of older Americans in terms of the number of expected years of life in different living arrangements—reflecting both mortality and transitions in living arrangements across this 25-year period of later life. We limit our focus to ages 65–90 for two reasons. The first is small cell sizes and the associated difficulty of estimating transition probabilities beyond age 90. (According to 2004 U.S. life tables, the proportion of life after age 65 that is lived beyond age 90 is .08 for women and .05 for men [https://www.cdc.gov/nchs/data/nvsr/nvsr56/nvsr56_09.pdf].) The second is that age 65 is a conventional marker of “old age” beyond which living arrangements are unlikely to reflect the presence of young adult children still enrolled in higher education or in the process of transitioning to independence. Conducting analyses separately by sex, race/ethnicity, and educational attainment allows us to provide one of the first descriptive portraits of sociodemographic differences in living arrangement transitions and living arrangement-specific life expectancy at older ages in the United States.

Background

Living Arrangements and the Provision of Care

Where adult children live in relation to their elderly parents is often viewed as an indicator of the amount of potential support available to older parents (Litwak & Kulis, 1987; Rossi & Rossi, 1990; Wolf, 1994). This is particularly true for types of support that require physical interaction, such as caregiving and companionship, and is less true for monetary support. Coresidential, proximate, and distant living can thus be seen as gradations of accessibility from the “face-to-face access” inherent in coresidence to the “less intense sustained access” of proximate living (Wolf, 1994) and the irregular access of distant residence. Proximate living is sometimes portrayed as an ideal arrangement—adult children and their older parents can interact frequently while both are able to retain a degree of autonomy and privacy (Shanas, 1979). Rosenmayr and Köckeis (1963) famously described this living arrangement as “intimacy at a distance.”

There is a good deal of research on the geographic proximity of children and parent–child contact and interaction (e.g., DeWitt & Frankel, 1988; Greenwell & Bengtson, 1997; Lawton, Silverstein, & Bengtson, 1994; Roan & Raley, 1996). In contrast, empirical evidence of the prevalence of proximate living arrangements and especially of the duration and correlates of intergenerational proximity is quite limited. In part, this reflects the tendency for living arrangements to be constructed from survey household rosters, which typically indicate whether adult children live with parents, but not their distance from parents.

We do know, however, that proximate residence is a common living arrangement. Cross-national data from 10 European countries showed that 43% of people aged 50 years and older live less than 25 km away from their nearest living child, ranging from 31% in Italy to 63% in the Netherlands (Hank, 2007). In the United States, 45% of men and women aged 65 and older live within 10 miles of an adult child (authors’ tabulation of data from the Health and Retirement Survey) and the average American adult lives only 18 miles from his or her mother (Bui & Miller, 2015). We also know that residential moves that increase intergenerational proximity at older ages—so-called “second moves” in Litwak and Longino’s (1987) life course model of migration—are associated with older age, widowhood, and health decline (e.g., Zhang et al., 2013).

It is clear that the prevalence of intergenerational coresidence has declined and the proportion of elderly adults living alone has risen in many developed countries, suggesting that proximate living and preferences for living near, but not with, children may have increased over time (Tomassini, Glaser, Wolf, van Groenou Broese, & Grundy, 2004). Long-term decline in intergenerational coresidence has been attributed to increased affluence of elderly parents (McGarry & Schoeni, 2000; Michael, Fuchs, & Scott, 1980) and increased economic opportunities among the younger generation (Ruggles, 2007), whereas the more recent growth in proximate residence in the United States likely reflects declining internal migration, especially long-distance migration (Cooke, 2018; Molloy, Smith, & Wozniak, 2011).

Much of the research on older adults’ living arrangements focuses on support from adult children, but it is clear that support provision often goes in the other direction, or in both directions. Though some elderly parents need support from their adult children, parents are also important resources to their adult children, providing instrumental support (e.g., childcare and shopping) as well as financial and emotional support (Compton & Pollak, 2015; Ingersoll-Dayton, Neal, & Hammer, 2001, Knipscheer, Dykstra, de Jong Gierveld, & van Tilburg, 1995). Indeed, a recent study suggested that intergenerational coresidence in the United States is increasingly driven by the economic needs of adult children rather than the needs of older parents (Kahn, Goldscheider, & García-Manglano, 2013). Research on intergenerational exchanges of support has focused on motivations (e.g., reciprocity, altruism, strategic investment, insurance) and on the implications of support receipt and provision for multiple dimensions of well-being (e.g., Ingersoll-Dayton et al., 2001; Silverstein, Conroy, Wang, Giarrusso, & Bengtson, 2002). To the extent that exchanges of support are both beneficial and associated with intergenerational geographic proximity, cross-group differences in patterns of living arrangements may play an important role in exacerbating or mitigating inequality across the life course.

In addition to living arrangements defined by the geographic proximity of adult children, older adults may live in institutional settings (i.e., nursing homes). Because this is far less common than other arrangements, it is not often a central focus in analyses of population data. Nevertheless, the proportion of older Americans who experience nursing home admission is not trivial—in 2013, there were 15,700 nursing homes with 1.4 million residents in the United States (Harris-Kojetin, Sengupta, Park-Lee, & Valverde, 2013). Nursing home residents comprise 2.8% of the 65+ population and 10.2% of the 85+ population. (These figures are from https://www.cms.gov/medicare/provider-enrollment-and-certification/certificationandcomplianc/downloads/nursinghomedatacompendium_508.pdf.) These figures highlight the value of better understanding the duration of nursing home residence, the likelihood of returning to the community relative to dying, and how these relationships vary across sociodemographic groups.

A large body of research on the correlates of nursing home admission demonstrates that the likelihood of institutionalization increases with age, especially following health decline, and is higher among women, whites, rural residents, and those with lower income (e.g., Choi, Schoeni, Langa, & Heisler, 2015; Cohen & Bulanda, 2016; Gaugler, Duval, Anderson, & Kane, 2007). Because the risk of nursing home residence is negatively associated with access to family-provided support (Choi et al. 2015), our focus on transitions between nursing home residence and three other distinct living arrangement states (and death) as well as our estimation of nursing home life expectancy by sex, race/ethnicity, and educational attainment are important extensions of previous research.

Sociodemographic Correlates of Living Arrangements

Studies of the demographic correlates of living arrangements have found a curvilinear relationship between age and distance from adult children, increasing among the “young–old” and then decreasing among the “old–old” as the need for care grows (Clark & Wolf, 1992). Living arrangements are clearly associated with correlates of need for care. For example, Compton and Pollak (2015) showed that coresidence is most likely when mothers are older, in poor health, and unmarried, but found that proximity does not depend on these care-related characteristics of mothers. Other studies have documented a rural–urban differential, with older rural residents more likely to live farther from their children than those living in urban areas (Lee, Dwyer, & Coward, 1990; Lin & Rogerson, 1995). Increased availability means that parents with a larger number of children are more likely to live close to a child and higher life expectancy means that older women (especially widows) are more likely than older men to live with or near their children (Clark & Wolf, 1992; Lee et al., 1990). The correlates of living arrangements also differ by the gender and marital status of children—for example, both married and unmarried daughters are more likely to live with older mothers (75+) while married sons are less likely to live near older mothers (Compton & Pollak, 2015).

Other research has focused on the socioeconomic correlates of living arrangements. In the United States, socioeconomic status is positively associated with distance between parents and adult children—for example, adult children and parents with higher levels of education are more likely to live far from each other relative to those with lower levels of education (Clark & Wolf, 1992; Compton & Pollak, 2015). There are several theoretical explanations for these socioeconomic differences in living arrangements. For example, middle-class children are able to pursue specialized education and jobs that are not available near their parents while working-class children more often enter less-specialized jobs closer to home. Similarly, the fact that deterioration of health and economic well-being begins at earlier ages for working-class parents makes them more likely to need the support of proximate children relative to their middle-class counterparts (Greenwell & Bengston, 1997).

In terms of race and ethnicity, prior research shows that Hispanics are much more likely to live with (but not near) their mothers than whites, and blacks are both more likely to live with and live near their mothers than whites (Compton & Pollak, 2015). These differences partially reflect the fact that several well-established correlates of intergenerational coresidence including poor health, constrained economic circumstances, and absence of a spouse are more prevalent among racial/ethnic minorities (Mutchler & Burr, 2008). Racial/ethnic differences have also been linked to culture and preferences, with emphases on independence and individualism contributing to greater distance between older whites and their children and valuation of extended family ties contributing to higher levels of coresidence among blacks and Hispanics (Burr & Mutchler, 1999; Mutran, 1985). Support from older parents is also relevant for understanding intergenerational proximity among blacks, with coresidence often initiated in response to the younger generation’s childbirth, union dissolution, unemployment, and other life events (e.g., Seltzer & Bianchi, 2013).

Compton and Pollak’s (2015) examination of the prevalence and correlates of proximate residence as a distinct living arrangement category provides a valuable theoretical and empirical basis for our study. We move beyond their analyses in several important ways. Although they used data from Wave 2 (1992–1994) of the National Survey of Family and Households (NSFH), we use more recent data from the much larger sample of older Americans in the HRS. Further, while Compton and Pollak (2015) focused exclusively on the living arrangements of mothers and their adult children, we examine the living arrangements of both older men and women in relation to their adult children. Importantly, we move beyond their cross-sectional analysis by using multiple observations across 10 years of the HRS to characterize life expectancy beyond age 65 in terms of living arrangements, including nursing home residence (an arrangement that they did not consider in their analyses). This allows us to examine how theoretically important characteristics (e.g., sex, race/ethnicity, educational attainment) are related not only to the relative likelihood of being in a given living arrangement at one point in time, but also the number of years that various groups live, on average, in different arrangements.

Data

We use data from the Waves 5–10 of the HRS (2000–2010) provided by RAND (version O of the main file and version C of the family file). Wave 5 (2000) is the first wave of the HRS to both include a representative sample of Americans aged 65 and older and provide weights for nursing home residents. (Weighted HRS data are representative of the U.S. population. We use sampling weights [RWTRESP] in all analyses to reflect oversampling and patterns of differential nonresponse that may be related to living arrangements [e.g., age, health, race/ethnicity, region of residence]. Weights for nursing home residents [RWTR_NH] are available from Wave 5.) In our analyses, we focus on older adults with at least one living child in consecutive survey waves. A total of 3,868 observations (5.7% of the total sample) are excluded by virtue of having no living children at a given wave and results should be interpreted with this sample restriction in mind.

Living arrangements can be categorized in several ways including household headship, relationship to the head of the household, and whether the elderly live with children or others, cross-classified by marital status. In this article, we define respondents’ living arrangement at the time of each interview using the following categories: (a) coresidence with children (regardless of whether other coresident adults were present), (b) proximate residence (living independently, including both living alone and with spouse only, but within close proximity (10 miles) to at least one living child), (c) distant residence (living independently, including both living alone and with spouse only, and more than 10 miles from the nearest living child), and (d) institutional living (i.e., nursing home). [Ten miles is an arbitrary threshold used to define close proximity in the HRS (and other surveys including the NSFH). Rationale for this threshold is that regular contact and provision of instrumental and emotional support become difficult when travel between residences requires more than 30 min (Zhang, Engelman, & Agree, 2013). Previous research using the HRS shows that this measure of proximate residence is significantly associated with transfers of support to older parents (e.g., Leopold, Raab, & Engelhardt, 2014).] Transitions to death before the subsequent wave are treated as an additional, absorbing state. Figure 1 provides a diagrammatic representation of the transitions summarized in the multistate life tables. In this figure, the black bidirectional arrows represent transition probabilities both from state x to state y and from state y to state x whereas the gray unidirectional arrows represent transition probabilities from state x to death. For the sake of simplicity, this diagram does not include panel attrition, a censoring event.

Figure 1.

Diagrammatic representation of the transitions represented in the multistate life tables.

Recognizing that the different categories of living arrangements are not mutually exclusive, we apply the hierarchy of categories listed above when defining a person’s living arrangement category. For example, an HRS respondent will be coded as coresiding if s/he was living with a child, regardless of whether s/he lived within 10 miles of another living child. We recognize that some respondents may have both co-resident and proximate children or may have multiple proximate children (indeed, 57% of HRS respondents coresiding with a child had another child living within 10 miles). (All respondents living in a nursing home are included in that category, even if an adult child lives within 10 miles of the nursing home.) However, given the analytical strategy summarized in Figure 1, we prefer this hierarchical classification to a more complex classification scheme that involves additional categories and smaller cell sizes.

Construction of Multistate Life Tables

We use multistate life tables to summarize information on the transitions between different living arrangements between ages 65 and 90 depicted in Figure 1. Multistate life tables allow us to characterize old age in terms of the number of years that a hypothetical cohort of 65-year-olds (or x-year-olds, more generally) would expect to live in different living arrangement states (coresiding with children, living proximate to children, living far from children, in institutional care) over the next 25 years if they were to experience the age-specific living arrangement and mortality transition probabilities observed in the data (see Palloni, 2001 and Schoen, 1988 for descriptions of these models). They also allow for comparison of time spent in different living arrangements across groups of interest, for example, men versus women, racial and ethnic groups, different levels of educational attainment. The results presented later are based on data pooled across waves and cohorts in the HRS. This hybrid period-cohort life table allows us to more precisely estimate transitions across multiple living arrangements (and death) for men and women aged 65–90. We construct 12 separate life tables—for three racial/ethnic groups (white, black, Hispanic) separately by sex and for three educational groups (less than high school, high school, and some college or more) separately by sex.

To construct these life tables, we first estimated multinomial logistic regression models to generate state-specific transition probabilities based on person-year data constructed from the HRS. Because the HRS is conducted at 2-year intervals, we generated random numbers to assign interwave transitions to one of the 2 years (i.e., we assign the event to age x if the random number is 0 and to age x + 1 if the random number is 1). In these models (estimated separately by race/ethnicity, educational attainment, and sex), living arrangement transitions are a function of age and age squared, a specification that adequately captures the age pattern of mortality beyond age 65 and flexibly and parsimoniously represents the age pattern of other transitions. (An exception to this specification is transitions from nursing home residence. The small number of observations in this origin state necessitated constraining the coefficient for age squared to equal zero.) We transformed estimated probabilities from these regression models into rates using conventional conversion methods (Preston, Heuveline, & Guillot, 2001) to construct origin state-specific multistate life tables using a SAS macro available at the Centers for Disease Control and Prevention website. (The SAS macro is titled “Mark Hayward’s SAS macro to compute MSLT function via the deterministic approach.” It implements the linear method described in Schoen (1988) and is available at http://www.cdc.gov/nchs/data_access/space.htm.) Finally, we combine these state-specific life tables into a single life table for each combination of sex and race/ethnicity or sex and educational attainment by using the observed living arrangement distribution at age 65 as weights. That is, we divide the radix (l₆₅) into four different origin states $l_{65}^i$ that are equal in size to the observed distribution of living arrangement categories (i) at age 65 for each of the racial/ethnic and educational groups of interest.

Results

Sex- and age-specific distributions of living arrangements (from data pooled across waves) summarized in Figures 2 and 3 show that the prevalence of different living arrangements is relatively stable across older ages, with 15%–20% coresiding with children, 45%–50% living near children, and 30%–35% living far from children. The prevalence of nursing home residence is low, reaching 5%–10% only after age 80.

Figure 2.

Distribution of women’s living arrangements, by age. Note: Authors’ tabulation of HRS data for female respondents with at least one living child, n = 33,489.

Figure 3.

Distribution of men's living arrangements, by age. Note: Authors’ tabulation of HRS data for male respondents with at least one living child, n = 25,296.

However, cross-wave living arrangement transitions presented in Table 1 clearly show that a sizeable proportion of HRS respondents 65 years and older change living arrangements. Between 20%–30% of those in coresidential, proximate, or distant living arrangements at a given age x are in a different state (including death) approximately 2 years later. (The percent in a different state 2 years later is 100 minus the percent in the same state 2 years later.) Transitions out of institutional living are much more common, with roughly half of both men and women transitioning to a different state, typically death, 2 years later. Patterns of transitions are very similar for men and women, with men somewhat more likely to transition to death. The contrast between aggregate stability in living arrangement distributions across older ages and the large proportion of individuals experiencing change in living arrangements is consistent with Hermalin, Ofstedal, Baker, and Chuang’s (2005) earlier findings from Taiwan. Table 1 also presents unweighted numbers and mean age for each cell. The former show that our estimates of transitions from nursing home residence are based on small numbers and the latter show that, not surprisingly, transitions to institutional living and to death tend to be at older ages.

Table 1.

Living Arrangement Transitions, by Initial State

Group	Coresident	Proximate	Distant	Institutional	Dead	Total
Women
Coresident (%)	75.6	8.3	3.9	2.3	9.8	100.0
(n)	5,659	628	271	166	738	7,462
(mean age)	74.0	72.0	71.7	79.7	78.5	74.3
Proximate (%)	4.4	81.5	4.8	2.6	6.7	100.0
(n)	705	12,112	757	359	989	14,922
(mean age)	74.5	74.4	74.1	80.8	78.7	74.8
Distant (%)	3.4	7.7	90.1	2.6	6.2	100.0
(n)	337	772	7,485	222	571	9,387
(mean age)	74.9	74.6	74.1	81.9	79.0	74.6
Institutional (%)	1.2	2.7	1.1	49.1	45.9	100.0
(n)	11	31	12	601	536	1,191
(mean age)	77.8	78.8	82.5	81.1	81.8	81.4
Total	18.8	41.8	26.4	4.3	8.7	100.0
(n)	4,642	11,755	8,792	610	3,167	32,962
(mean age)	74.1	74.3	74.0	81.0	79.3	74.9
Men
Coresident (%)	72.4	9.6	5.8	1.4	10.8	100.0
(n)	3,308	470	255	69	546	4,648
(mean age)	72.7	71.9	71.5	76.9	77.1	73.1
Proximate (%)	3.5	80.1	5.2	1.4	9.4	100.0
(n)	447	9,027	622	153	1,085	11,334
(mean age)	73.6	73.6	73.1	80.3	77.8	74.1
Distant (%)	2.6	6.7	80.0	1.5	9.2	100.0
(n)	240	626	6,698	117	789	8,470
(mean age)	73.4	73.7	73.2	78.9	77.6	73.7
Institutional (%)	0.1	3.5	2.6	42.4	50.5	100.0
(n)	6	15	11	203	254	489
(mean age)	72.1	77.1	79.6	78.0	80.9	79.4
Total	15.8	40.1	31.1	2.2	10.4	100.0
(n)	7,746	15,702	9,893	1,563	3,255	29,941
(mean age)	74.0	74.2	74.0	80.8	78.7	74.6

Note: Rows represent initial state (at HRS Wave t) and columns represent follow-up state (at HRS Wave t + 1).

Table 2 presents the average number of years that 65-year-old men and women are expected to live in each of the four living arrangement states, by race/ethnicity. The proportion of remaining life in each state is presented in parentheses. The “total” column shows that, for all race/ethnicity groups, women are expected to live 2–3 years more than men between ages 65 and 90. (Supplementary Figure 1 shows that the HRS data produce overall life expectancies that are very similar to those in the life tables based on vital statistics for whites and blacks and slightly understate life expectancy for Hispanics [https://www.cdc.gov/nchs/products/life_tables.htm].) The “proximate” column indicates that white women and men spend slightly less than half of their life between ages 65 and 90 living within 10 miles of an adult child. Roughly one-third of this period of life is spent living distant from adult children (“Distant” column) and about 15% is spent coresiding with an adult child (“Coresident” column). On average, white men and women spend less than 1 year of life (between ages 65 and 90) in a nursing home (“Institutional” column).

Table 2.

Living Arrangement-Specific Life Expectancy Between Ages 65 and 90, by Sex and Race/Ethnicity

Group	Coresident		Proximate		Distant		Institutional		Total
Women
White	3.06	(0.17)	8.26	(0.47)	5.73	(0.32)	0.65	(0.04)	17.70
Black	6.87	(0.42)	6.07	(0.37)	3.01	(0.18)	0.53	(0.03)	16.48
Hispanic	8.32	(0.46)	5.60	(0.31)	3.60	(0.20)	0.57	(0.03)	18.10
Men
White	2.42	(0.15)	7.22	(0.45)	6.08	(0.38)	0.34	(0.02)	16.06
Black	4.30	(0.32)	5.55	(0.41)	3.42	(0.25)	0.34	(0.03)	13.62
Hispanic	6.47	(0.41)	5.02	(0.32)	3.86	(0.25)	0.25	(0.02)	15.60

Note: Proportion of life between ages 65 and 90 is presented in parentheses.

Blacks and Hispanics differ markedly from whites, typically spending more than twice as many years coresiding with adult children. Black and Hispanic women also live 42% and 46%, respectively, of their life between ages 65 and 90 living with an adult child compared with only 17% for white women. The facts that duration of coresidence is shorter and “intimacy at a distance” is more common for whites suggest that attention to geographic proximity may be particularly important for understanding racial/ethnic differences in access to support from children at older ages (and the associated implications for variation in health and well-being). Of particular importance is whether the difference between whites and other groups in terms of years living with or near children serves to limit racial/ethnic disparities in health and financial well-being. In contrast to the pronounced differences in residential proximity, racial/ethnic differences in expected years of life in institutional living are small.

Table 3 presents corresponding figures for men and women of different educational backgrounds. The most striking differences in living arrangements are the relatively long duration of coresidence among the least educated (especially women) and the strong positive educational gradient in distant residence. Women who did not complete high school spend 35% of their life between ages 65 and 90 living with an adult child compared with only 16% for college-educated women. Like whites in Table 2, men and women with some college differ markedly from their less-educated counterparts in spending more time (in both years and proportion of life) living more than 10 miles from their nearest child. These pronounced differences highlight the importance of better understanding linkages between living arrangements and well-documented educational differences in health and other dimensions of well-being at older ages. Also similar to Table 2, educational attainment is not strongly associated with time spent in institutional living.

Table 3.

Living Arrangement-Specific Life Expectancy Between Ages 65 and 90, by Sex and Educational Attainment

Group	Coresident		Proximate		Distant		Institutional		Total
Women
Less than HS	5.47	(0.35)	6.33	(0.41)	3.05	(0.20)	0.60	(0.04)	15.45
High school	3.53	(0.20)	8.60	(0.48)	4.93	(0.28)	0.72	(0.04)	17.79
College	3.10	(0.16)	8.23	(0.43)	7.10	(0.37)	0.60	(0.03)	19.02
Men
Less than HS	3.88	(0.27)	6.81	(0.48)	3.15	(0.22)	0.32	(0.02)	14.16
High school	2.80	(0.18)	7.79	(0.49)	4.93	(0.31)	0.38	(0.02)	15.91
College	2.48	(0.15)	6.54	(0.39)	7.49	(0.45)	0.31	(0.02)	16.82

Note: Proportion of life between ages 65 and 90 is presented in parentheses.

Life Expectancy by Living Arrangements at Age 65

Because the life expectancies in Tables 2 and 3 reflect trajectories of experience averaged across initial living arrangement states at age 65, it is also informative to examine living-arrangement-specific life expectancy conditional on initial living arrangements. These estimates are presented in Tables 4 and 5. One unsurprising pattern in these tables is that, for all groups, life expectancy is longest in the initial living arrangement state at age 65. However, in many cases, duration in the origin state is not much more than half of remaining life expectancy, highlighting the fluid nature of living arrangements at older ages demonstrated in Table 1.

Table 4.

Living Arrangement-Specific Life Expectancy Between Ages 65 and 90, by Living Arrangements at Age 65, Sex, and Race/Ethnicity

Group	Coresident		Proximate		Distant		Institutional		Total
Women
Whites
Coresident	8.31	(0.50)	4.94	(0.29)	2.87	(0.17)	0.63	(0.04)	16.75
Proximate	1.77	(0.10)	12.72	(0.72)	2.59	(0.15)	0.67	(0.04)	17.74
Distant	1.56	(0.09)	4.10	(0.22)	11.93	(0.65)	0.65	(0.04)	18.25
Institutional	0.42	(0.06)	1.12	(0.17)	0.33	(0.05)	4.93	(0.72)	6.81
Blacks
Coresident	10.76	(0.66)	3.49	(0.21)	1.48	(0.09)	0.49	(0.03)	16.22
Proximate	3.67	(0.22)	10.53	(0.63)	1.99	(0.12)	0.58	(0.03)	16.77
Distant	3.17	(0.19)	3.78	(0.23)	9.12	(0.55)	0.51	(0.03)	16.58
Institutional	1.18	(0.17)	0.48	(0.07)	0.40	(0.06)	4.86	(0.70)	6.92
Hispanics
Coresident	12.15	(0.68)	3.41	(0.19)	1.77	(0.10)	0.57	(0.03)	17.90
Proximate	4.60	(0.25)	10.19	(0.55)	3.12	(0.17)	0.59	(0.03)	18.50
Distant	3.65	(0.20)	4.08	(0.23)	9.70	(0.54)	0.55	(0.03)	17.98
Institutional	2.61	(0.21)	3.01	(0.24)	0.89	(0.07)	5.86	(0.47)	12.37
Men
Whites
Coresident	7.71	(0.49)	4.39	(0.28)	3.13	(0.20)	0.34	(0.02)	15.59
Proximate	1.21	(0.07)	11.96	(0.74)	2.65	(0.16)	0.32	(0.02)	16.14
Distant	1.01	(0.06)	3.26	(0.20)	11.59	(0.72)	0.35	(0.02)	16.21
Institutional	0.60	(0.07)	2.07	(0.25)	0.89	(0.11)	4.82	(0.57)	8.39
Blacks
Coresident	8.52	(0.63)	2.99	(0.22)	1.76	(0.13)	0.36	(0.03)	13.63
Proximate	2.25	(0.17)	8.75	(0.65)	2.20	(0.16)	0.35	(0.03)	13.54
Distant	1.94	(0.14)	2.90	(0.21)	8.65	(0.63)	0.31	(0.02)	13.79
Institutional	1.17	(0.15)	1.38	(0.18)	1.54	(0.20)	3.73	(0.48)	7.82
Hispanics
Coresident	10.18	(0.65)	3.26	(0.21)	1.88	(0.12)	0.26	(0.02)	15.58
Proximate	3.27	(0.20)	9.50	(0.60)	2.95	(0.18)	0.23	(0.01)	15.94
Distant	2.52	(0.16)	3.76	(0.25)	8.77	(0.57)	0.25	(0.02)	15.29
Institutional	1.95	(0.30)	0.56	(0.09)	0.33	(0.05)	3.60	(0.56)	6.43

Note: Rows represent living arrangement state age 65. The proportion of life between ages 65 and 90 is presented in parentheses.

Table 5.

Living Arrangement-Specific Life Expectancy Between Ages 65 and 90, by Living Arrangements at Age 65, Sex, and Educational Attainment

Group	Coresident		Proximate		Distant		Institutional		Total
Women
Less than HS
Coresident	9.44	(0.63)	3.61	(0.24)	1.36	(0.09)	0.60	(0.04)	15.01
Proximate	2.83	(0.18)	10.15	(0.66)	1.83	(0.12)	0.58	(0.04)	15.38
Distant	2.85	(0.17)	4.13	(0.25)	8.82	(0.54)	0.64	(0.04)	16.45
Institutional	1.15	(0.18)	0.99	(0.15)	0.33	(0.05)	4.03	(0.62)	6.51
High school
Coresident	9.07	(0.53)	4.98	(0.29)	2.47	(0.14)	0.66	(0.04)	17.17
Proximate	1.84	(0.10)	12.88	(0.72)	2.47	(0.14)	0.76	(0.04)	17.95
Distant	1.64	(0.09)	4.47	(0.25)	11.24	(0.62)	0.69	(0.04)	18.05
Institutional	0.33	(0.04)	0.91	(0.11)	0.33	(0.04)	6.39	(0.80)	7.96
College
Coresident	8.82	(0.48)	5.10	(0.28)	3.81	(0.21)	0.56	(0.03)	18.29
Proximate	1.84	(0.10)	13.61	(0.70)	3.25	(0.17)	0.62	(0.03)	19.32
Distant	1.52	(0.08)	3.80	(0.20)	13.18	(0.69)	0.58	(0.03)	19.08
Institutional	0.68	(0.09)	1.46	(0.20)	0.46	(0.06)	4.61	(0.64)	7.21
Men
Less than HS
Coresident	8.18	(0.58)	3.91	(0.28)	1.66	(0.12)	0.31	(0.02)	14.06
Proximate	1.87	(0.13)	10.30	(0.72)	1.87	(0.13)	0.31	(0.02)	14.35
Distant	1.67	(0.12)	3.51	(0.25)	8.36	(0.60)	0.34	(0.02)	13.88
Institutional	1.35	(0.19)	1.16	(0.17)	0.42	(0.06)	4.02	(0.58)	6.94
High school
Coresident	7.79	(0.49)	4.94	(0.31)	2.70	(0.17)	0.39	(0.02)	15.82
Proximate	1.39	(0.09)	12.00	(0.75)	2.35	(0.15)	0.34	(0.02)	16.09
Distant	1.14	(0.07)	3.79	(0.24)	10.35	(0.66)	0.42	(0.03)	15.71
Institutional	1.32	(0.13)	2.33	(0.23)	1.02	(0.10)	5.50	(0.54)	10.17
College
Coresident	8.37	(0.52)	3.56	(0.22)	3.77	(0.23)	0.324	(0.02)	16.04
Proximate	1.11	(0.07)	12.09	(0.71)	3.43	(0.20)	0.30	(0.02)	16.93
Distant	1.00	(0.06)	3.09	(0.18)	12.69	(0.74)	0.31	(0.02)	17.09
Institutional	0.52	(0.06)	2.36	(0.28)	1.58	(0.19)	3.88	(0.47)	8.33

Note: Rows represent living arrangement state age 65. The proportion of life between ages 65 and 90 is presented in parentheses.

Looking first at racial and ethnic differences in Table 4, we see that, apart from the lower life expectancy of those in institutional living arrangements at age 65, differences in overall life expectancy by initial living arrangements are typically less than 1 year. This is true for both men and women and reflects a combination of relatively fluid living arrangements between ages 65 and 90 and the fact that age-specific transition probabilities do not differ by living arrangements at age 65 (by definition). Interestingly, we also see that the proportion of life lived in the initial state is relatively high for blacks and Hispanics who were coresiding at age 65 and for whites who were living proximate to or far from their nearest adult child at age 65. The former reflects relatively high rates of transition out of coresidence for whites and the latter reflects relatively high rates of transition into coresidence for blacks and Hispanics. Among those who were living less than 10 miles from their nearest adult child at age 65 (i.e., coresiding or proximate), there is little difference across racial/ethnic groups in the proportion of life that older men and women spend living near an adult child. However, the proportion of life spent coresiding with an adult child is longer for blacks and Hispanics relative to whites. Not surprisingly, those who were in nursing homes at age 65 spend most of their remaining (much shorter) lives in nursing homes. However, as shown earlier in Table 1, the proportion of respondents experiencing a transition out of nursing homes to any state other than death is roughly 5%, and those in nursing homes at age 65 do spend two or more years of life in other living arrangement states. One notable pattern that deserves more attention is the relatively high life expectancy of Hispanic women living in nursing homes at age 65.

Turning to differences by educational attainment in Table 5, we see that the stability of initial living arrangements for men and women with less than a high school education resembles blacks and Hispanics in Table 4. Those who were coresiding with adult children at age 65 spend a greater proportion of their remaining years in that same state relative to their counterparts with a high school education or more. Conversely, distant residence is more stable for those with high school or college educations. For example, the proportion of remaining life spent in this initial state is .66 for men with a high school education and .74 for those with some college education, but only .60 for those with less than a high school education. The pattern for women is similar. The proportion of life that men and women living near an adult child at age 65 spend in this initial state differs little by educational attainment. Given the scarcity of empirical evidence on proximate residence, these similarities are notable. Among those who were living in a nursing home at age 65, the proportion of their (much shorter) remaining life spent in this initial state ranges between .50 and .80.

Discussion

Our goal in this article was to describe the living arrangements in which Americans spend their later years and how they differ by gender across two key dimensions of social stratification—race/ethnicity and educational attainment. To the extent that proximity to adult children reflects potential access to financial, instrumental, and emotional support at older ages, living arrangements may be an increasingly important correlate of multiple dimensions of inequality in aging populations. Not only are the absolute and relative numbers of older Americans projected to increase, but rising social, economic, and health inequality, and shifting policy emphasis toward greater individual responsibility for economic security at older ages (e.g., Hacker, 2008) also highlight the potential for growing variation in the resources available to individuals in later life. Furthermore, the bidirectional nature of financial, instrumental, and emotional support exchange across generations makes living arrangements a potentially important dimension of social and economic stratification across the life course. For these reasons, there is significant value in building a more comprehensive understanding of living arrangements at older ages that explicitly recognizes proximity to adult children as a distinct arrangement, that moves beyond cross-sectional analysis of the correlates of different arrangements, and that effectively summarizes subpopulation differences in living arrangement transitions across older ages.

Our application of multistate life table analyses to 10 years of HRS data represents an important step toward this more comprehensive understanding. Most notably, we have shown pronounced differences by race/ethnicity and educational attainment in life expectancy in different living arrangements. Whites differ from blacks and Hispanics in spending far less time (both in terms of years and proportion of remaining life) living with an adult child(ren), somewhat more time living near a child(ren), and more time living farther than 10 miles from children. Similarly marked differences characterize the living arrangement-specific life expectancy of highly educated men and women relative to their less educated counterparts. Older parents who attended college spend the most time living more than 10 miles from their adult children whereas those with less than a high school education spend the longest time coresiding with children. Expected years living proximate to adult children varied less by educational attainment, with all groups spending 39%–49% of their remaining life near, but not with, an adult child. One possible reason for this similarity in proximate life expectancy is heterogeneity in the reasons for living near an adult child. For some, proximity may reflect limited opportunity (either for the older parent or adult child) to move elsewhere. For others, it may reflect the ability to move closer to adult children (or older parents).

Our findings contribute to existing research on living arrangements and well-being at older ages in several ways. Most important is our documentation of the prevalence and duration of geographic proximity to children and the extent to which it varies by race/ethnicity and educational attainment. Evidence that geographic proximity to adult children is a common arrangement in which older Americans spend much of their remaining lives (on average) underscores previous researchers’ call for a better understanding of the correlates of proximate residence. This basic empirical evidence has been missing from the large body of research on the motivations for proximate residence and the nature of intergenerational relations and support exchange among those living near adult children. The same is true of research on nursing home residence which has examined the correlates of nursing home admission, but has not systematically documented differences in nursing home life expectancy. Evidence that nursing home life expectancy is relatively short and that small proportions of nursing home residents transition to other living arrangements are valuable new empirical insights. Another important contribution is our documentation of variation in living arrangement-specific life expectancy across two key dimensions of social stratification. Although past research has repeatedly documented racial/ethnic and socioeconomic differences in living arrangements at older ages, evidence on the duration of exposure to different living arrangements has not been available. To the extent that the duration of remaining life in different living arrangements reflects life-long trajectories of well-being and/or predicts subsequent well-being, our findings provide a valuable empirical basis for subsequent work on the role of living arrangement-specific expectancy in processes of cumulative dis/advantage.

The analyses presented above are limited in several ways. The first is that we have not incorporated information on marital status. Because marital status is a strong predictor of living arrangements and spouses are primary sources of physical and emotional support (e.g., Liang, Brown, Krause, Ofstedal, & Bennett, 2005; Silverstein, 1995), our omission of this characteristic limits our ability to evaluate the potential implications of our findings for variation in access to private support. However, the incorporation of information on marital status is not straightforward. Cross-classifying living arrangements by marital status would result in problems of small cell size and inclusion of marital status as a time-varying covariate in the transition models is complicated by the lack of information on the exact timing of living arrangement transitions. One promising extension of these analyses would be to include a measure of marital status at age 65 as a time-constant covariate in the transition models upon which the life tables are based.

A second limitation is that we have ignored the living arrangements of childless older men and women. This is an important shortcoming in view of the implications of childlessness for old age support and the fact that 15% of Americans currently aged 65 and older are childless and 30% of those aged 70–85 are projected to be childless in 2030 (OECD, 2011). A third limitation is that we have focused only on adult children and paid no attention to coresidence with, or proximity to, other adults who may be important sources of support at older ages (e.g., siblings, friends). Although we recognize the potential importance of kin other than spouse and children or of “fictive kin,” the HRS does not include sufficient information to incorporate such individuals into our analyses. A fourth limitation is our use of the 10-mile threshold that defines proximate residence in the HRS survey. Detailed geographic information on residential locations of older respondents and their adult children would allow for a better understanding of the distance thresholds within which different types of support can be readily provided. A fifth limitation is our inability to observe the precise timing of changes in living arrangements. Because we observe changes in living arrangements by comparing responses across consecutive survey waves, we are forced to make assumptions about the timing of those transitions. We have randomly assigned changes to one of the two interwave years rather than assigning all transitions to the first or second year, which would result in systematic understatement or overstatement of duration in the initial state, respectively. In the absence of external information on the timing of transitions across 2-year intervals and on the average number of transitions within 2-year intervals, random assignment is a reasonable strategy.

We have also not examined correlates of living arrangement transitions (other than age, sex, race/ethnicity, and educational attainment) and have not considered the association between different living arrangement trajectories and indicators of well-being. These are important questions, but are beyond the scope of our objectives for this article. The use of multistate life table techniques allows us to summarize large amounts of information on age-specific transitions (between living arrangement states and to death) in a way that provides intuitive and informative descriptive summaries of how different groups of Americans spend their later years. Although widely used in research on healthy life expectancy and on employment at older ages, application of multistate models to living arrangements has been limited.

In addition to extending previous research in the ways described previously, our findings provide an empirical basis for subsequent examination of reasons for educational and race/ethnic differences in living arrangements (e.g., economic need, poor health, absence of a spouse) and the implications of these differences for multiple dimensions of inequality (e.g., financial well-being, physical health, mental health). One potentially valuable extension of our analyses would be to examine interactions between educational attainment and race/ethnicity to evaluate the relative importance of socioeconomic circumstances and other factors such as norms and preferences for understanding racial/ethnic differences in living arrangement-specific life expectancy.

Additionally, stratification researchers, in particular, should seek to better understand the extent to which these differences in living arrangement-specific life expectancy may serve to mitigate well-documented racial/ethnic and socioeconomic disparities in health and other dimensions of well-being at older ages. Finally, we have shown that there is a good deal of change in living arrangements at older ages. In most of the groups we examined, roughly half of the years lived between ages 65 and 90 is spent in a different living arrangement than the initial state at age 65. Important foci for subsequent research include evaluation of relationships between proximate residence and well-being (broadly defined) and examination of the correlates of later-life living arrangement transitions and the implications of those transitions for well-being.

Supplementary Material

Supplementary data are available at The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences online.

Funding

This work was supported by a grant from the National Institute on Aging (R01 AG049716) and was conducted at the Center for Demography of Health and Aging and the Center for Demography and Ecology at the University of Wisconsin–Madison (P30 AG17266 and P2C HD047873, respectively).

Conflict of Interest

The authors declare that they have no conflicting interests.

Author Contributions

J. M. Raymo was responsible for conceptualization, data analysis, and writing. I. Pike was responsible for data preparation, literature review, and writing. J. Liang was responsible for conceptualization and writing.