Gender differences in years of remaining life by living arrangement among older Singaporeans

Abstract

Living arrangements of older adults have often been studied as a measure of the support available to them. Given the rapidly ageing and low fertility context of Singapore where the prevalence of older adults living alone and without children is expected to increase, we construct multistate life tables to estimate the number of years that older persons can expect to live in different living arrangements at a population level (population-based) as well as based on their initial living arrangement (status-based). We focus particularly on whether there are gender differences in the expected years of life in different living arrangement states. We use the Panel on Health and Ageing of Singaporean Elderly, a 2009 nationally representative survey of 4990 Singaporeans aged 60 years and older, with follow-up surveys in 2011 and 2015. In calculating the probabilities of transition between different states, we control for number of children, housing type, and time-varying ADL limitations. We find that at age 60, women can expect to spend more than twice the proportion (18%) of their remaining lives living alone compared to men (7%). Status-based estimates indicate that the proportion of remaining years living with a child is higher for women initially living alone, with a spouse only or already with a child, compared to males. Our results indicate that while older women are more likely to live alone compared to their male counterparts, older women living alone are also more likely to transition to living with children. Our research sheds light on the importance of expanding research on life expectancy beyond health, to consider analysis using other forms of social stratification, particularly gender differences in states of living arrangement.

Supplementary Information

The online version of this article contains supplementary material available at 10.1007/s10433-020-00594-3.

Introduction

In a number of Asian societies including Singapore, the widely prevalent norm is for older adults to co-reside with children. This is rooted in the concept of filial obligations and reciprocity and expected to enable the provision of direct support by family members to older adults (Thang 2010; Verbrugge and Ang 2018). In recent years, however, there has been an increase in older adults living only with a spouse in two-member households, a result of adult children either having migrated for education or employment, or having set up their own independent households, as well as an increasing desire among older adults to maintain an independent household when their finances and health allow (Jamieson and Simpson 2013; Thang 2010). Considerably less common than co-residence with a spouse or a child is living alone (Podhisita and Xenos 2015), although household surveys across Asia have shown that the proportion of older adults living alone, while still low, has increased in the past two decades (Teerawichitchainan et al. 2015; Visaria et al. 2019). There are also gender differences in living arrangements, with studies finding that compared to older men, older women in various countries across South, Central, and Southeast Asia were more likely to live alone and less likely to live with a spouse (Bongaarts and Zimmer 2002), a finding corroborated by recent research in Singapore (Chan et al. 2018; Gubhaju et al. 2017).

Living arrangements of older adults have often been studied as a measure of the level of care and family support available to them (Chan 2005). Co-residence, with children in particular, has been found to contribute to improved physical and mental health through the receipt of financial and material support, emotional support from kin, assistance with activities of daily living, greater access to health information and services, as well as cognitive stimulus received through grandparenting (Arpino and Bordone 2014; Chen and Liu 2011; Chen and Short 2008; Merz and Huxhold 2010; Tang and Hooyman 2018). There is considerable research on the association of living arrangements, particularly living alone, with the health and wellbeing of older adults. A number of studies indicate that living alone is associated with adverse outcomes, such as greater depressive symptoms (Chan et al. 2011; Chou et al. 2006), short-term morbidities (Samanta et al. 2015), unplanned and higher frequency of hospitalization (Pimouguet et al. 2016), and a higher risk of falls (Elliott et al. 2009). Research suggests that older adults who live alone have fewer sources of social support, decreased interaction with social network members, reduced life satisfaction and greater loneliness (Djundeva et al. 2018; OʼSúilleabháin et al. 2019; Russell and Taylor 2009). At the same time, some studies suggest that the relationship between living arrangements and health and wellbeing outcomes varies by gender, for example that living along is associated with reduced life satisfaction (Gaymu and Springer 2010) or greater dependence in instrumental activities of daily living (Gubhaju et al. 2017) among older females only and not older males.

As a rapidly ageing society contained within a city-state, Singapore offers a unique opportunity to study living arrangements of older adults and specifically to estimate the remaining years of life that older adults can expect to live in different living arrangements. Total life expectancy at birth increased in Singapore from about 65 years in 1960 to 78 years in 2000, to currently being one of the highest in the world at about 85 years (Singapore Ministry of Health and Institute for Health Metrics and Evaluation 2019). Increasing longevity combined with several decades of total fertility well below replacement have meant that family sizes in Singapore have been declining over the years, from nearly five persons in 1980 to fewer than three in 2019 (Singapore Department of Statistics 2020b). A recent comparison of two cohorts of older adults in Singapore showed that the proportion of older adults aged 60 years and older living alone in one-person households increased by about 19% between 2009 and 2016 (Visaria et al. 2019). At the same time, social policy in Singapore strongly emphasizes family caregiving of older adults and makes available a number of financial incentives to encourage older adults and children to co-reside or live in close proximity (Chan 2008). These include a housing subsidy under the Proximity Housing Grant provided to adult Singapore citizens for purchasing property within 4 km of where their parents or children live, and priority in the allotment of new public housing for joint applications between married children and parents, or flexible lease 2-room apartments (Housing Development Board 2020).

In this context, we use longitudinal data to assess the extent and type of transitions in living arrangements that are made at older ages in Singapore. In particular, we apply the methodology of constructing multistate life tables, to estimate the number of years of remaining life that older adults in Singapore can expect to live in different living arrangements. Multistate life tables have been widely used to calculate life expectancy by health status (Saito et al. 2014), but only a handful of prior studies have used multistate life table methods to examine life expectancy by living arrangements (Gu et al. 2009; Hermalin et al. 2005; Ogawa et al. 2010; Raymo et al. 2019). Hermalin et al. (2005) found that older Taiwanese spent equal amounts of time coresiding and living alone during their later lives. They found minimal differences by gender in coresident life expectancy. Gu et al. (2009) used the Chinese Longitudinal Healthy Longevity Survey (CLHS) and found that Chinese oldest-old can expect to spend most of their remaining lived years co-residing in multigenerational households. A higher proportion of remaining life can be expected to be spent living in multigenerational households and living alone for women compared to men, while men can expect to spend a higher proportion of their lives living with a spouse only and living in skipped-generation households. Ogawa et al. (2010) found that in Japan, men could expect to spend nearly twice the proportion of total life expectancy at the age of 65 as women living with a spouse and less than half the proportion as women living alone. A recent study of living arrangements in the USA constructed multistate life tables using a methodology similar to this study and adopted a novel approach of defining living arrangements in terms of institutional living, co-residence with a child, and proximity of distance to a child (Raymo et al. 2019). The authors found a high level of transitions across living arrangements at older ages, with the largest proportion of remaining years spent living in close proximity but not in co-residence with a child for men and women both.

The overall aim of our study is to estimate the average number of years of remaining life that older males and females in Singapore can expect to spend living alone, with a spouse, with a child, and with others, conditional as well as unconditional on their initial living arrangement.

A recent study in Singapore found that the proportions of those living alone and living with a spouse only were substantially higher among Chinese older adults compared to their Malay or Indian counterparts (Chan et al. 2018). On the other hand, the proportions of those living with a child but not a spouse, and living with both a spouse and child were the highest among the Malays. The same study found that older adults with tertiary education had the highest proportions of living alone or living with a spouse only. Older adults with no formal education had the highest proportion of living with a child only and the lowest proportion of living either with a spouse only or with a spouse and child. A previous study of active and inactive life expectancy at older ages in Singapore, measured in terms of years of life at older ages expected to be spent with and without health-related difficulties in independently performing activities of daily living (ADLs) and instrumental ADLs, found no gender differences in active life expectancy, but found higher active life expectancy among older adults with higher education compared to those with no formal or only primary education, as well as among the Chinese compared to the non-Chinese (Chan et al. 2016). Therefore, in this study of gender differences in estimated years of life in different living arrangements, we also calculate if any gender differences are present across levels of educational attainment and ethnic groups.

Additionally, we study status-based estimates, i.e., to what extent the remaining years of life in different living arrangements at different ages are the same or different across the initial living arrangement of older adults at those ages. The estimates provided by this study will enable policy makers to better understand the extent and type of transitions that occur in living arrangements at older ages, and their implications for the availability of family-based support, and future plans for the housing, health, and care needs of older adults.

Methods

Data

We used data from the Panel on Health and Aging of Singaporean Elderly (PHASE), a nationally representative longitudinal survey of community-dwelling older Singapore citizens and permanent residents aged 60 years and older. PHASE had 3 waves, in 2009 (n = 4990, inclusive of 453 proxy respondents), 2011, and 2015. Further details about PHASE are available elsewhere (Chan et al. 2019). Of the total 4990 respondents who were interviewed in wave 1, those who remained uncontactable at the time of waves 2 and 3 and for whom mortality status could not be ascertained were not included in the analysis (n = 1538), since they would have contributed data for only one time point and their subsequent living arrangement would have been unknown. This includes respondents who may have been alive in different living arrangements, but moved to a different location within Singapore or overseas, or transitioned to institutional settings such as nursing homes. The survey was unable to contact them and ascertain the extent or type of their living arrangement transitions. The impact of this on our study is likely to be low; based on available data on the capacity of residential institutional care—namely, nursing homes and inpatient hospice care—in Singapore (Singapore Department of Statistics 2020a), and assuming both full occupancy and that all residents of these facilities were older adults aged 60 years and over, residents in such institutional living arrangements would have represented a small fraction of the total older adult population in Singapore, about 1.9% in 2009, and 1.8% in 2011 and in 2015 (Singapore Department of Statistics 2020c).

Our analytical sample is based on the 3452 respondents of wave 1 in 2009 who were re-interviewed during wave 2 in 2011 (n = 3103) or who were reported as deceased at the time of fieldwork in 2011 (n = 349). Of those interviewed during wave 2 in 2011, 1572 respondents were re-interviewed and 295 respondents were reported as deceased at the time of wave 3 in 2015. The analysis of de-identified data from PHASE wave 1, data collection for PHASE waves 2 and 3, and linkage of the survey data with mortality databases were approved by the institutional review board at the National University of Singapore.

Measures

Living states: living arrangement

We defined living arrangement for each respondent at the time of the survey wave in terms of four mutually exclusive states: (a) living alone, i.e., in a one-person household, (b) living with a spouse only, i.e., as a couple, in a two-person household, (c) living with a child or grandchild in a two-or-more-persons household, i.e., with a child or grandchild and with or without any others, and (d) living with others, i.e., within a household but not in one of the three living arrangement states noted above. Our measure for living with a child is intentionally an expansive definition to include the respondent’s own biological or adopted children as well as grandchildren, spouses of children, and spouses of grandchildren. As noted above, transition to institutional living is not considered in this analysis.

Absorbing state: death

The date of death of deceased respondents (until 31 December 2015) was obtained from the national Registry of Births and Deaths databases and supplemented by information gathered in survey waves 2 and 3.

As shown in Fig. 1, respondents could remain within or transition between any of the four living states between waves, or transition between one of the four living states and death, the final absorbing state. In order to show the schematic simply, Fig. 1 does not show respondents lost to follow-up between waves.

Fig. 1.

Diagrammatic representation of the transitions represented in the multistate life tables

Control variables

In the calculation of transition probabilities, we chose to include three control variables that capture elements of the family, socioeconomic status and health status that can influence the living arrangement options available to older adults as well as their preferences. The first is the total number of surviving biological and adopted children for all respondents at the time of the first wave. Data on the total number of surviving children were not collected in all three waves of the survey, preventing us from using this as a time-varying variable. Socioeconomic status of older adults was measured using the type of housing that the respondent was living in at wave 1. We dichotomized the type of housing into two categories: public apartments (i.e., constructed by the Housing Development Board) with 3 or fewer rooms, and public apartments with 4 or more rooms or private housing, with the larger apartments and private housing indicative of a higher socioeconomic status of the older adult (Malhotra et al. 2013). Finally, the health status of an older adult can strongly determine living arrangement options, with changes brought on by declining ability to function independently, for example, transitioning from living alone or with a spouse only to living with a child or with others such as a domestic worker. We also noted earlier that a number of studies have found an association between living alone in particular and adverse health outcomes. Establishing the causal direction of the relationship is not the objective or within the scope of this study, but in order to estimate transition probabilities that take into account the time-varying nature of physical health status at older ages, we control for health using a time-varying indicator of activity of daily living (ADL) limitations, operationalized as any health-related difficulty at each wave with performing any of six ADLs, namely bathing, dressing, eating, standing up from a bed or a chair or sitting down on a chair, walking around the home, and toileting.

Analysis

We constructed multistate life tables to estimate the remaining years of life that individuals at ages 60, 70, and 80 can expect to live within each of the four living arrangement states, as well as total life expectancy and status-based life expectancy. We use the Stochastic Population Analysis for Complex Events (SPACE) program in SAS version 9.4. SPACE is a composite program that includes fitting multinomial logistic regression models to the data to estimate the log odds of different transitions over time, using these to calculate the probability of transitioning between states in terms of 1-year age intervals, and then constructing a multistate life table based on the transition probabilities (Cai et al. 2010; Chiu 2018). For all individuals for whom the living arrangement changed between two survey waves, the exact time of the transition is unknown. Therefore, the timing of the transition was randomly assigned in SPACE to an individual’s age between the first (2009) and second (2011) waves of data, i.e., age in either 2010 or 2011, and between the second (2011) and third (2015) waves of data to the age at one among 2012, 2013, 2014, or 2015.

Multinomial logistic regression models were used to estimate the log odds of transition in living arrangement states across waves: within and between the four living arrangement states, and between the four living states and the absorbing state of being deceased, as shown in Fig. 1. Given the longitudinal nature of the data, separate equations were specified for each of the four living states at wave 1. The first set of models regressed living arrangement states on sex as the key explanatory variable. In order to ascertain whether there were gender differences across educational attainment and ethnic groups, two additional sets of models were run with living arrangement states regressed on educational attainment (coded dichotomously as no formal education or primary education only, and higher), and ethnic group (Chinese and non-Chinese), with sex also included as a covariate in both. Each of the three sets of models included age, coded as a continuous variable, and the three controls described earlier, i.e., number of children, housing type, and a time-varying variable for ADL limitations. All analysis was weighted by wave 1 survey weights. The log odds from the multinomial logistic regression models were converted into age- and sex-, age- and ethnicity-, and age- and educational attainment-specific probabilities of the transition within and between each living arrangement state. These age-specific transition probabilities were used to assign a living arrangement state at each age beginning at age 60 to a synthetic cohort of 100,000 individuals, yielding a multistate life table with (1) ‘population-based’ point estimates for remaining years of life at a specific age expected to be lived in each of the living arrangements, unconditional on the origin living arrangement, and (2) ‘status-based’ estimates for remaining years of life at a specific age expected to be lived in each of the living arrangements a, conditional on initially being in a specific living arrangement state (Palloni 2000; Saito et al. 2014). The remaining years of life at a certain age expected to be lived in a particular state are based on a summation of the years in this particular state until death and do not necessarily represent a continuous period. Bootstrapping with 200 resampled datasets were used to estimate the 95% confidence intervals from the 2.5th and 97.5th percentile of the resulting distribution of point estimates. We also used the 95% confidence intervals around the absolute difference in the life expectancy estimates for males and females from the distribution across all bootstrap resamples to determine whether the male–female differences in point estimates were statistically significant. If the 95% confidence interval of the difference between males and females included zero, we deemed the male–female difference to not be statistically significant.

Results

Table 1 shows the distribution of background characteristics and the different living arrangements across the three waves. At the outset at wave 1, more than two-thirds of the older adults (71%) live with a child. About 16% live with only a spouse, and about 6% of older adults live by themselves. The proportions change among those surveyed at wave 2 and wave 3, although in absolute terms the proportion of those living with a child remains the highest at each wave, and those living alone the lowest. At the same time, we see that the proportion of those living with a child declines from wave 1 to wave 2 and wave 3, whereas the proportion of those living with a spouse only or living alone increases.

Table 1.

Distribution of basic characteristics and living arrangement states among respondents by wave

	Wave 1		Wave 2		Wave 3
	Mean	%/SD	Mean	%/SD	Mean	%/SD
Age	72.8	8.1	74.6	7.8	78.1	7.6
Female	53.8	%	54.5	%	58.6	%
Educational attainment
No formal/primary	72.6	%	70.8	%	70.7	%
Secondary or higher	27.4	%	29.2	%	29.3	%
Ethnicity
Chinese	71	%	71.5	%	72.9	%
Non-Chinese	29	%	28.5	%	27.1	%
ADL limitations	9.0	%	12.2	%	17.6	%
Number of surviving children	3.6	2.2	–		–
Housing type
1–3 room public housing	35.7	%	–		–
4 + room public/private housing	64.3	%	–		–
Living arrangements
Living alone	6.0	%	7.2	%	8.5	%
Living with spouse only	16.0	%	16.6	%	18.4	%
Living with a child	70.8	%	68.0	%	64.1	%
Living with others	7.2	%	8.2	%	8.9	%
Observations	3452		3103		1572

In Table 2, we present the transitions in living arrangements as observed for males and females between waves. Including transitions within and between living arrangements and between living arrangements and death, there were a total of 1857 transitions for females and 1595 transitions for males between wave 1 and wave 2, and 1080 transitions for females and 787 transitions for males between wave 2 and wave 3. We note that for both females and males, the majority of respondents do not change their living arrangement between wave 1 and wave 2, and that relatively more transitions in living arrangements are made from wave 2 to wave 3, i.e., when respondents are older, between ages 62 + and 65 +. More females compared to males remained living alone, living with a child, or living with others from wave 1 to wave 2. With the exception of living with others, more females remained within each of their living arrangements from wave 2 to wave 3 compared to males. We also note that a higher proportion of men compared to women transitioned to death across nearly all living arrangements and during both transition periods.

Table 2.

Transitions in living arrangement among respondents by wave

From wave 1	To wave 2	n	%	From wave 2	To wave 3	n	%
Females
Alone	Alone	98	69.0	Alone	Alone	60	75.0
	Spouse	1	0.7		Spouse	0	0.0
	Child	19	13.4		Child	8	10.0
	Others	18	12.7		Others	8	10.0
	Deceased	6	4.2		Deceased	4	5.0
Spouse	Alone	10	4.6	Spouse	Alone	10	7.3
	Spouse	160	73.1		Spouse	91	66.4
	Child	32	14.6		Child	20	14.6
	Others	11	5.0		Others	12	8.8
	Deceased	6	2.7		Deceased	4	2.9
Child	Alone	30	2.2	Child	Alone	16	2.0
	Spouse	37	2.7		Spouse	38	4.9
	Child	1130	82.8		Child	572	73.1
	Others	31	2.3		Others	13	1.7
	Deceased	136	10.0		Deceased	143	18.3
Others	Alone	11	8.3	Others	Alone	15	18.5
	Spouse	4	3.0		Spouse	3	3.7
	Child	16	12.1		Child	9	11.1
	Others	83	62.9		Others	46	56.8
	Deceased	18	13.6		Deceased	8	9.9
Transitions	1857				1080
Total, females	2937
Males
Alone	Alone	47	72.3	Alone	Alone	21	70.0
	Spouse	0	0.0		Spouse	0	0.0
	Child	5	7.7		Child	2	6.7
	Others	8	12.3		Others	4	13.3
	Deceased	5	7.7		Deceased	3	10.0
Spouse	Alone	7	2.1	Spouse	Alone	4	2.5
	Spouse	244	73.5		Spouse	96	58.9
	Child	39	11.7		Child	26	16.0
	Others	17	5.1		Others	15	9.2
	Deceased	25	7.5		Deceased	22	13.5
Child	Alone	8	0.7	Child	Alone	6	1.1
	Spouse	59	5.5		Spouse	57	10.5
	Child	861	79.7		Child	367	67.8
	Others	14	1.3		Others	9	1.7
	Deceased	138	12.8		Deceased	102	18.9
Others	Alone	13	11.0	Others	Alone	2	3.8
	Spouse	11	9.3		Spouse	5	9.4
	Child	7	5.9		Child	4	7.5
	Others	72	61.0		Others	33	62.3
	Deceased	15	12.7		Deceased	9	17.0
Transitions	1595				787
Total, males	2382

Population-based estimates for total life expectancy (TLE) at age 60, 70, and 80 as well as remaining years of life in different living arrangements for males and females are presented in Table 3. There were substantial differences between females and males, with females at age 60 expected to spend 5.0 years [95% CI 3.0–7.1] or 18% of remaining life living alone compared to 1.6 years [0.5–2.7] or 7.2% for males. For females, the expected years of life living alone were thus more than three times compared to males, and the proportion of TLE spent living alone twice that of men, at age 60, as well as ages 70 and 80. The remaining years of life living with only a spouse was higher for males at age 60 (6.2 [5.0–7.3]) compared to females (4.9 [3.6–6.2]) and the proportion of TLE substantially higher for males (27.6%) compared to females (17.1%); the male–female difference, however, was not statistically significant. Male–female differences for living with others were also not statistically significant. A little over half of remaining lives are expected to be spent living with a child for both males and females. Females at age 60 can expect to spend a higher number of years living with a child (15.9 [14.6–17.2]) compared to males (12.4 [11.1–13.6]), and while the male–female difference is statistically significant, the number of years represents a similar proportion (55%) of TLE for both females and males.

Table 3.

Population-based estimates for total life expectancy, and years of life spent in different living arrangements at ages 60, 70, and 80, by sex

		Total life expectancy (TLE)		Living alone			Living with a spouse only			Living with a child			Living with others
Age	Sex	No. of years (95% CI)	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff
60	F	28.6 (26.3–31.0)	*	5.0 (3.0–7.1)	17.6	*	4.9 (3.6–6.2)	17.1	NS	15.9 (14.6–17.2)	55.5	*	2.8 (1.9–3.8)	9.9	NS
	M	22.3 (20.7–24.0)		1.6 (0.5–2.7)	7.2		6.2 (5.0–7.3)	27.6		12.4 (11.1–13.6)	55.3		2.2 (1.4–3.1)	9.9
70	F	20.6 (11.4–16.7)	*	4.3 (2.1–6.4)	20.7	*	2.7 (1.9–3.6)	13.2	NS	11.4 (10.4–12.4)	55.3	*	2.2 (1.3–3.2)	10.8	NS
	M	15.3 (13.7–17.0)		1.4 (0.2–2.6)	9.4		3.6 (2.8–4.4)	23.6		8.4 (7.6–9.3)	55.0		1.9 (1.0–2.7)	12.1
80	F	14.1 (11.4–16.7)	*	3.5 (1.3–5.8)	25.2	*	1.5 (0.9–2.1)	10.6	NS	7.4 (6.4–8.4)	52.6	*	1.6 (0.7–2.6)	11.5	NS
	M	9.8 (8.0–11.7)		1.2 (0.0–2.6)	12.4		2.0 (1.4–2.6)	20.3		5.1 (4.5–5.8)	52.4		1.5 (0.6–2.4)	14.9

TLE total life expectancy, F females, M males, CI confidence intervals, NS not significant, Sign. of diff. the significance of difference in the female-male point estimates at p < 0.5

In Table 4, we present population-based estimates for TLE overall and life expectancy in each of the four living arrangements by educational attainment and ethnic group. In terms of educational attainment, the results show that the gender differences are significant at both lower and higher educational levels. At age 60, higher educated females can expect to spend a little over a decade living alone (10.3 years [3.5–17.1]) representing 27% of TLE. The comparable figures for higher educated males were 5.2 years [1.1–9.3], or about 17% of TLE, similar to those for lower educated females. The remaining years of life expected to be spent with a child were higher for females with no formal or primary education (16.2 [14.7–17.6]) compared to their male counterparts (12.7 [11.2–14.2]), with similar differences for higher educated females and males, and both representing similar proportions of TLE across females and males. Male–female differences by educational attainment in years of life spent living with a spouse only or living with others were not statistically significant. Among the gender differences by ethnic group presented in Table 4, the proportion of TLE expected to be spent living alone at age 60 was more than twice for Chinese females (18.4%) compared to Chinese males (7.6%), with a similar difference seen for Others, i.e., those not of Chinese origin. Chinese females can expect to spend a higher number of years living with a child (15.8 [14.4–17.2]) compared to Chinese males (12.3 [10.9–13.7)], with a similar male–female difference among Others, although the proportion of TLE expected to be spent living with a child is similar for females and males within the two ethnic groups.

Table 4.

Population-based estimates for total life expectancy, and years of life spent in different living arrangements at age 60 by educational attainment and ethnic group

		Total life expectancy (TLE)		Living alone			Living with a spouse only			Living with a child			Living with others
Edu	Sex	No. of years (95% CI)	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff
Low	F	27.0 (24.8–29.3)	*	4.5 (2.7–6.4)	16.7	*	4.0 (2.8–5.2)	14.8	NS	16.2 (14.7–17.6)	59.9	*	2.3 (1.5–3.2)	8.6	NS
	M	20.0 (18.3–21.7)		1.2 (0.3–2.2)	6.2		4.5 (3.4–5.6)	22.5		12.7 (11.2–14.2)	63.3		1.6 (0.9–2.3)	7.9
High	F	38.3 (32.5–44.1)	*	10.3 (3.5–17.1)	26.9	*	7.2 (4.7–9.6)	18.7	NS	16.6 (13.8–19.5)	43.4	*	4.3 (1.6–7.0)	11.2	NS
	M	30.0 (25.3–34.7)		5.2 (1.1–9.3)	17.4		8.3 (6.1–10.4)	27.6		12.8 (10.8–14.8)	42.7		3.7 (1.4–6.1)	12.5

		Total life expectancy (TLE)		Living alone			Living with a spouse only			Living with a child			Living with others
Ethnic group	Sex	No. of years (95% CI)	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff
Chinese	F	29.0 (26.5–31.5)	*	5.3 (3.2–7.5)	18.4	*	5.2 (3.7–6.6)	17.8	NS	15.8 (14.4–17.2)	54.5	*	2.7 (1.8–3.6)	9.3	NS
	M	22.8 (20.9–24.7)		1.7 (0.6–3.0)	7.6		6.7 (5.3–8.1)	29.4		12.3 (10.9–13.7)	54.0		2.1 (1.2–2.9)	9.0
Other	F	26.5 (23.4–29.7)	*	3.5 (1.0–6.0)	13.1	*	3.4 (2.1–4.7)	12.9	NS	16.0 (14.0–18.1)	60.4	*	3.6 (1.7–5.6)	13.6	NS
	M	20.8 (18.7–23.0)		1.2 (0.1–2.3)	5.6		4.3 (2.9–5.7)	20.6		12.3 (10.7–14.0)	59.3		3.0 (1.4–4.7)	14.5

TLE total life expectancy, F females, M males, NS not significant. Educational attainment: low = no formal education or primary education only, high = secondary education and higher. CI = confidence intervals. Sign. of Diff. refers to the significance of difference in the female-male point estimates at p < 0.5

Status-based estimates for males and females at ages 60, 70, and 80 are shown in Table 5. We note overall that for females and males in each initial living arrangement, the highest proportion of TLE is expected to be spent within the same living arrangement, representing what we saw in Table 2 on the transitions between waves. Across the status-based estimates, the highest TLE for both females and males is when the initial state is living with a spouse. For status-based estimates with the initial state of living alone, male–female differences were not statistically significant for remaining years living alone, living with a spouse, or for living with others. On the other hand, we note that females initially living alone at all ages can expect to live a higher number of years living with a child compared to males (at age 60, 7.7 [5.7–9.7] years compared to 3.5 [1.5–5.6] among males), representing a higher proportion of their TLE (27.2%) compared to males (16%). When the initial state is living with a spouse, we see that a higher proportion of remaining years are expected to be lived alone for females compared to males (at age 60, 5.0 [2.7–7.4] years representing 16.7% of TLE) compared to males (1.3 [0.1–2.5], or 5.9% of TLE). The number of years living with a child is higher for females initially living with a spouse, (10.2 [8.3–12.0]) at age 60 compared to males (7.5 [5.5–9.4]) although the proportion of TLE is similar. We find similar differences with an initial state of living with a child, with the exception that the proportion of TLE spent by males living with a child when their initial state is already living with a child is higher than females. Supplementary Tables S1 and S2 show status-based estimates at age 60 by educational attainment and ethnicity which largely mirror these results.

Table 5.

Status-based estimates for total life expectancy, and years of life spent in different living arrangements at ages 60, 70, and 80, by sex

		Total life expectancy (TLE)		Living alone			Living with a spouse only			Living with a child			Living with others
Age	Sex	No. of years (95% CI)	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff	No. of years (95% CI)	% of TLE	Sign. of diff
Initial state: living alone
60	F	28.3 (24.9–31.6)	NS	15.1 (11.2–19.0)	53.5	NS	1.3 (0.5–2.1)	4.5	NS	7.7 (5.7–9.7)	27.2	*	4.2 (2.4–5.9)	14.7	NS
	M	22.1 (17.2–27.0)		12.7 (7.2–18.3)	57.6		1.5 (0.3–2.8)	7.0		3.5 (1.5–5.6)	16.0		4.3 (1.6–7.0)	19.4
70	F	22.7 (18.2–27.2)	NS	13.8 (9.4–18.2)	60.9	NS	0.5 (0.2–0.8)	2.2	NS	5.3 (3.6–7.1)	23.5	*	3.1 (1.5–4.7)	13.5	NS
	M	16.2 (10.8–21.7)		10.6 (5.4–15.9)	65.5		0.4 (0.0–0.8)	2.4		1.8 (0.3–3.3)	11.0		3.4 (1.1–5.7)	21.1
80	F	16.4 (10.4–22.4)	NS	11.2 (5.4–17.0)	68.2	NS	0.2 (0.0–0.4)	1.1	NS	3.5 (1.2–5.8)	21.2	*	1.5 (0.0–3.2)	9.4	NS
	M	11.5 (4.7–18.3)		8.4 (2.2–14.7)	73.4		0.1 (0.0–0.3)	0.9		1.2 (0.0–2.5)	10.4		1.8 (0.0–3.8)	15.3
Initial state: living with a spouse only
60	F	30.1 (27.1–33.0)	*	5.0 (2.7–7.4)	16.7	*	11.8 (9.5–14.1)	39.3	NS	10.2 (8.3–12.0)	33.8	*	3.1 (1.9–4.3)	10.2	NS
	M	22.7 (20.6–24.9)		1.3 (0.1–2.5)	5.9		11.8 (9.6–14.0)	52.1		7.5 (5.5–9.4)	32.8		2.1 (1.1–3.1)	9.3
70	F	23.2 (19.4–27.0)	*	5.1 (1.9–8.4)	22.1	*	9.6 (7.5–11.7)	41.4	NS	5.9 (4.5–7.3)	25.5	*	2.6 (1.1–4.0)	11.0	NS
	M	16.6 (14.3–18.9)		1.4 (0.0–2.9)	8.5		9.5 (8.0–11.0)	57.2		3.7 (2.7–4.6)	22.2		2.0 (0.9–3.1)	12.1
80	F	17.7 (13.1–22.2)	*	5.1 (1.1–9.1)	28.9	*	7.4 (4.8–10.1)	42.0	NS	3.0 (1.6–4.4)	16.8	*	2.2 (0.4–4.0)	12.3	NS
	M	11.6 (8.3–14.9)		1.2 (0.0–3.0)	10.6		7.2 (5.5–9.0)	62.3		1.5 (0.8–2.2)	13.0		1.6 (0.1–3.1)	14.0
Initial state: living with a child
60	F	28.5 (26.2–30.8)	*	4.1 (2.2–6.0)	14.3	*	3.7 (2.6–4.8)	13.0	NS	18.5 (17.2–19.8)	64.9	*	2.2 (1.4–3.1)	7.8	NS
	M	22.4 (20.8–24.0)		1.1 (0.2–2.1)	5.1		4.6 (3.6–5.6)	20.5		15.0 (13.8–16.2)	67.0		1.7 (0.9–2.4)	7.4
70	F	19.9 (17.8–21.9)	*	2.7 (1.1–4.3)	13.4	*	1.6 (1.0–2.2)	8.2	NS	14.3 (13.3–15.3)	71.8	*	1.3 (0.6–2.0)	6.7	NS
	M	14.9 (13.5–16.2)		0.7 (0.0–1.5)	4.7		2.1 (1.4–2.7)	13.9		1.1 (10.3–12.0)	75.0		0.9 (0.4–1.5)	6.4
80	F	12.7 (11.1–14.3)	*	1.5 (0.4–2.5)	11.5	*	0.6 (0.2–1.0)	4.8	NS	10.0 (9.0–10.9)	78.4	*	0.7 (0.2–1.2)	5.3	NS
	M	8.9 (7.8–9.9)		0.3 (0.0–0.8)	3.7		0.7 (0.4–1.1)	8.1		7.4 (6.7–8.1)	83.2		0.4 (0.1–0.8)	5.0
Initial state: living with others
60	F	26.8 (23.0–30.6)	*	5.8 (3.1–8.5)	21.7	NS	3.3 (0.9–5.7)	12.4	NS	10.5 (7.5–13.4)	39.1	*	7.2 (4.9–9.6)	26.9	NS
	M	21.2 (16.5–25.9)		2.4 (0.0–4.9)	11.3		6.0 (3.1–8.8)	28.1		5.5 (2.5–8.5)	26.1		7.3 (4.0–10.6)	34.4
70	F	19.8 (15.8–23.8)	*	5.6 (2.3–8.8)	28.2	NS	1.3 (0.05–2.0)	6.4	NS	4.9 (3.4–6.3)	24.6	*	8.1 (6.1–10.0)	40.9	NS
	M	16.5 (11.4–21.6)		3.1 (0.0–7.4)	18.5		2.6 (1.1–4.0)	15.6		2.5 (1.2–3.8)	15.3		8.3 (5.2–11.4)	50.6
80	F	14.8 (9.7–19.9)	*	5.3 (1.3–9.3)	35.9	NS	0.3 (0.0–0.6)	1.9	NS	2.3 (0.9–3.7)	15.6	*	6.9 (4.2–9.6)	46.5	NS
	M	12.8 (5.2–20.4)		3.1 (0.0–9.9)	24.2		0.7 (0.0–1.7)	5.3		1.2 (0.0–2.5)	9.6		7.8 (2.6–13.0)	61.0

TLE total life expectancy, F females, M males, CI confidence intervals, NS not significant, Sign. of diff. the significance of difference in the female-male point estimates at p < 0.5

Discussion

In this study, we applied the multistate life tables method to the study of living arrangements among older adults in Singapore. We showed that a substantial proportion of remaining life at older ages can be expected to be spent by males and females living alone, although females can expect to spend nearly three times the number of years as males and more than twice the proportion of remaining lives living alone. These differences between males and females are present across low- and higher-educated groups, as well as the Chinese and non-Chinese ethnic groups. Our study also found that females can expect to spend a higher number of years of remaining life compared to males living with a child. Our status-based estimates show that among those already living alone, females are significantly more likely to transition to living with a child and spend a substantially higher proportion of remaining lives living with a child. This suggests that it may be easier for women at older ages to transition from other living arrangements to living with a child, in particular move in with their children when living alone is no longer feasible. Factors that preclude males making a similar transition is an area for future enquiry.

We acknowledge that this study has a number of limitations. First, since we use reported living arrangement at the time of each survey wave or mortality subsequent to a survey wave, the exact timing of transition in living arrangements states between survey waves is unknown. We therefore randomly assign the timing of the transition to one of the inter-survey years. There are potentially more than one transitions in living arrangements that may occur in the inter-survey years, but we assume only one transition between the initial and subsequent states. Additionally, the two inter-survey periods between waves 1 and 2, and waves 2 and 3 are not the same, with two possible years for assigning the timing of transition between waves 1 and 2, and four possible years between waves 2 and 3. This may potentially somewhat over- or under-estimate the duration spent by individuals in a given living arrangement. Second, although we use mutually exclusive categories of living arrangements, we do not have data to account for differences in the proximity or actual location of a non-cohabiting adult child. The presence of a child living elsewhere within Singapore compared to one living abroad can differentially affect the probability of initially living alone or with a spouse only but then transitioning to living with a child. This may be an important factor to consider in future studies where information on the location of children is also available. Third, we do not account for marital status in the estimation of transition probabilities and the construction of the multistate life tables. Living arrangement options and preferences are likely to be correlated with marital status at the individual level. However, the introduction of marital status in the study of living arrangements leads to the comparison of very small cell sizes and model convergence challenges, as noted in other studies as well (Raymo et al. 2019). Finally, our study does not consider transitions to institutional living for older adults. Although the proportion of older adults in institutional living is very low as reported earlier, and current social policy in Singapore continues to emphasize ‘ageing-in-place’ with a key role for family members in caregiving (Malhotra et al. 2019), transitions to institutional living arrangements may be an important issue to study in the future.

At the same time, this study has a number of strengths. First, we control for three key possible determinants of living arrangements: number of children, socioeconomic status, and ADL limitations. In particular by including a time-varying covariate for ADL limitations, our estimation of transition probabilities between living arrangements takes into account the changes over time in the physical health status of older adults. Second, we extend the multistate lifetable method hitherto used primarily to measure health expectancy, to studying life expectancy in different living arrangements. Our expectation is that these methods can be applied further to understanding life expectancy at older adults in terms of other living states also. Third, we use recent nationally representative longitudinal data that makes the findings generalizable to community-dwelling older adults in Singapore.

Our study findings have important policy implications. In a context where the proportion of older adults living alone has increased, our finding that a substantial proportion of life for older males and females is expected to be spent living alone, including when their initial living arrangement is already that of living alone, points to the need for careful planning of available social, health, and financial resources at older ages. In particular, the finding that older women compared to men are likely to spend more than twice their remaining years of life living alone, suggests the strong need for policy making and innovative programmatic interventions to meet the specific needs of this female population. Older women in Singapore are likely to have built up substantially lower financial resources over their lifetimes compared to men, given that they spend fewer total years in formal employment, are more likely than men to have retired early, and less likely to be actively looking for employment at older ages (Visaria 2018). Interventions with a focus on ensuring that women have the capacity and resources to live independently, particularly those that center on improving mobility within and outside the household, are important to consider. Another key area for intervention is the development of programmes that cater specifically to ensure that older women living alone are socially connected and engaged. At the same time, our study finds that more than half of remaining lives will also be spent by both older women and men living with their children, and therefore assessing and meeting the needs of older adults in these households and exploring opportunities for productive engagement with a wider social network also remain key issues for families to prioritize.

Our study also suggests that it is important to consider the ways in which older adults may be enabled to transition between living arrangements, to live independently if they so desire, and to have a balance between autonomy and dependence when co-residing with family members. Initiatives towards this can include a wide range of center-based programmes for older adults including those living with family, that enable older adults to be socially engaged and active regardless of their living arrangement, and for social networks to be sustained, built, or rebuilt across living arrangement transitions. In order to address some of the potential health risks associated specifically with living alone, initiatives that support routine health maintenance while ‘ageing-in-place’ should be considered, including community-based health workers who can conduct home-based case management, coordinate appointments with health facilities, facilitate access to health screenings and diagnostic tests, and support post-hospitalization home-based care. For older adults who desire living independently, Singapore has recently announced an assisted-living public housing project where residents can have access to an ‘on-site’ community manager monitoring their health status, providing linkages to different social services and care programmes, and organizing group-based social activities (Teo 2020). Process as well as health and wellbeing outcome evaluations of interventions such as these can yield valuable lessons for further refining social policy and programmes that cater to the needs of older adults as they transition across different living arrangements at older ages.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

AC contributed to conception and design, acquisition of data, editing of manuscript. AV contributed to conception and design, data analysis, interpretation of analysis, initial drafting and revision of manuscript. BG contributed to conception and design, preliminary analysis, contribution to manuscript. SM contributed to acquisition of data, comments on manuscript. YS contributed to conception and design, interpretation of analysis, editing and revision of manuscript.

Funding

Waves 1, 2 and 3 of the Panel on Health and Ageing among Singaporean Elderly (PHASE) were funded or supported by the following sources: Ministry of Social and Family Development, Singapore; Singapore Ministry of Health’s National Medical Research Council under its Singapore Translational Research Investigator Award “Establishing a Practical and Theoretical Foundation for Comprehensive and Integrated Community, Policy and Academic Efforts to Improve Dementia Care in Singapore” (NMRC-STAR-0005-2009), and its Clinician Scientist—Individual Research Grant—New Investigator Grant “Singapore Assessment for Frailty in Elderly-Building upon the Panel on Health and Aging of Singaporean Elderly” (NMRC-CNIG-1124-2014); and Duke-NUS Geriatric Research Fund.

Compliance with ethical standards

Conflict of interest

The authors declare that they have no conflict of interest.