Living Arrangement Preferences of Elderly People in Taiwan as Affected by Family Resources and Social Participation

Abstract

This article investigates living arrangement preferences of elderly people in Taiwan, including coresidence with family (their children or spouse), coresidence with spouse only, living alone, and living in an institution. The authors consider the effects of three factors: the elderly persons’ health situation, their family resources, and their social participation, such as community workshops or political activities. Accordingly, the authors propose empirical logit models based on the well-developed discrete choice theory. Empirical results reveal that (1) elderly people with higher socioeconomic status, prefer either independent living arrangements or coresidence with their children, (2) elderly people with more family resources, such as large family size, prefer to coreside with their children, and (3) elderly people with adequate social support and/or contact networks prefer independent living arrangements.

Keywords

elderly people living arrangement family resources social participation discrete choice theory logit model

Introduction

People sixty-five years and older accounted for 10.7 percent of Taiwan’s total population in 2010 due to its low birth rate and increased life expectancy. Taiwan is projected to become an aged society by 2017, with elderly people accounting for 14 percent of the total population, according to the 2010 report by the Council for Economic Planning and Development of Taiwan. This increased elderly population and the progressive decrease in birth rate imply that there will be less chance for coresidence with adult children. In this study, we investigate living arrangement preferences of elderly people from the perspectives of their value for privacy and the level of their needs/resources. In addition, we consider the social participation of the elderly as affected by their improved socioeconomic status.

This study consists of the following sections. The second section presents the theoretical background of this study. The third section proposes an empirical model. The fourth section focuses on model estimation and discussion. The last section provides conclusions.

Literature Review and Situation Summary for Taiwan

Living arrangements of elderly people can be separated into several subcategories. In terms of family structure, living locations¹ and the relationship with coresident/coresidents, there are (1) living alone, (2) coresidence with spouse only, (3) coresidence with children, (4) coresidence with multiple generations,² and (5) coresidence with friends or siblings. The present study focuses on the relationship with coresident/coresidents and regards each type of living arrangement as a different alternative.

Demographic Factors and Living Arrangements

As an indispensable variable in research on the living arrangements of elderly people, demographic factors include age, gender, and health conditions with the later being the major determinant.^3,4 The literature subdivided the elderly people into three categories: “the younger elderly” (defined by age of sixty-five to seventy-four), “the middle elderly” (seventy-five to eighty-four), and “the older elderly” (eighty-five and over). It was found that “the older elderly” group is more likely to coreside with children than the “the younger elderly,”⁵ indicating that the elderly people depend on children for care because of deteriorating health or mobility as they grow older. The study indicated that living with children seems to be the least desired option for the elderly in poor health, implying that in Korea elderly parents want to avoid burdening their children.⁶ On the contrary, Indonesian elderly who have physical disability are more likely to live with a child.⁷ This indicates that living arrangement preferences are influenced by values of privacy and the level needs in the elderly. One the other hand, “Family resources” is a major determinant for living arrangement preferences in addition to the demographic factors.

Family Resources and Living Arrangements

“Low birth rate” and “high labor force participation of women” are more direct causes.⁸ In particular, the number of children available to participate in caregiving is related positively to the elderly people’s likelihood of coresiding with an adult child. The literature defined coresidence with children is an “opportunity” for elderly,⁹ but in this study, we defined it as “family resources.” However, the likelihood of elderly people coresiding with their children might also depend on the children’s willingness.¹⁰

Adult children provide instrumental and emotional support to elderly people.^11,12 Studies in Taiwan, however, demonstrate that adult children offer primarily financial support.¹³ In other words, adult children in Taiwan prefer to supply financial support to their elderly parents instead of coresiding with them. Other studies explain a high possibility of coresidence based on “social exchange theory.”^14,15

Married elderly people who have better financial condition prefer coresidence with their spouse than their children.^7,16 In Malaysia, the married elderly and children live together to reduce living costs or for assistance with household activities.⁹ Women have a longer life expectancy and elderly widows, who have lost the physical and emotional support from their spouse, prefer coresidence with children.¹⁷ But due to low birth rates and the increase of unmarried adults, the tradition of coresidence with children is gradually fading. Traditionally, the elderly have depended on children for emotional support, though they are now more open to receiving emotional support from a social network support instead of from their children. As a result, the elderly are more likely to live with spouse instead of their children. Furthermore, the elderly in Taiwan are now more open to the option of living in an institution instead of with their children. Therefore, living in an institution also is considered in this study.

Social Participation and Living Arrangement Preferences of Elderly People in Taiwan

There has been a lack of research on the social participation of elderly people in Taiwan. This current study focuses on the issue of how living arrangements are affected by social participation, which has not been previously considered. We hypothesize that social participation enhances the social support¹⁸ for elderly people and therefore lowers the likelihood of the elderly to reside with their children. Some elderly people, because of family structure change, retiring from work, or losing their spouse, have more time to participate in family, social, or leisure activities, while others have more limited time. Still others maintain current social participation because their role transition from adulthood to old age has not yet occurred. The elderly engage in varying degrees of social interaction with others and the environment.¹⁹ This results in diverging levels of emotional dependence and perceived social support, which influence living arrangement preferences. According to the empirical research, the lack of social support is common among the elderly community who live alone.²⁰ Besides for the middle elderly, and the older elderly, the kin network was found to offer the greatest degree of support, and the family intensive type, the least. The network-type construct proved to be a significant predictor of social support even when taking into account respondents’ living arrangements, physical disability level, and personal coping resources.²¹ Similar research in Toulouse pointed out that elderly women rarely lived on their own, and when they could not rely on their children for support, they found care in more distant kin and friends.²² Therefore, we hypothesize that social participation would influence living arrangement preferences.

Overall, the choice of living arrangements also depends on the resources and needs of the family and the elderly individual.²³ This study focuses on (1) the influence of individual needs and family resources on living arrangement preferences of the elderly and (2) how social participation is associated with living arrangement preferences of the elderly.

Empirical Models

Data and Methodology

This study uses data from the “2005 Report on the Status of Elderly People in Taiwan and Fukien Areas,” which was based on stratified random sampling of people aged fifty. But only questionnaires answered by people aged sixty-five²⁴ and over were selected. Questionnaires were eliminated if they had items in “living arrangement preferences” that were unanswered. As a result, a final size of 1,430 samples was used.

Living arrangements in this study are grouped into four alternatives: “coresidence with children” (denoted by G), “coresidence with spouse only” (S), “living alone” (A), and “living in an institution” (I). The choice structure is shown in Figure 1. “Coresidence with children” refers to elderly people who prefer to live with their children in a household, either with or without their spouse, “coresidence with spouse only” to households composed elderly people and their spouses, “living alone” to elderly people living by themselves, and “living in an institution” to the elderly living in a care facility. According to samples distributions, most of the subjects (57.0 percent) prefer coresidence with children. Of the rest, 43.0 percent, prefer to live with a spouse only (25.4 percent), to live alone (15.4 percent), or to live in an institution (2.3 percent).

Figure 1.

Multinomial choice structure of living arrangements.

The key dependent variable in this study is the preferred coresident/coresidents of the elderly people. The study employs a discrete choice model to predict the probability of choice for each living arrangement because the variable is a categorical one, given the assumption of rational choice behavior. The study explains living arrangement preferences of elderly people based on expected utility theory.^7,25,26

Assume that the probability (denoted by $P_{j}^{i}$ ) of each living arrangement alternative ( $j$ ) of the elderly ( $i$ ) depends on the utility consumed from each arrangement alternative ( $U_{j}^{i}$ ). Thus, higher consumption utility indicates stronger preference for the given alternative.

The utility function of living arrangement alternative of elderly people can be specified as shown in equation (1). This function indicates that the consumption utility of the elderly derives from observed utility $(V_{j}^{i})$ and unobserved utility $(ϵ_{j}^{i})$ , where $V_{j}^{i}$ represents the observable components of the utility for each alternative. The study separates utility into three categories. The first category refers to the demographic factors of the elderly people ( $X_{j}^{i}$ ), the second category is variables related to the family resources of the elderly people ( $Z_{j}^{i}$ ), and the third category is variables of the social participation of the elderly people ( $S_{j}^{i}$ ). Let observed utility be the linear combination of all attributes as given in equation (2).²⁷

U_{j}^{i} = V_{j}^{i} + ϵ_{j}^{i},

V_{j}^{i} = f (X_{j}^{i}, Z_{j}^{i}, S_{j}^{i}) = \sum_{k = 1}^{K} β_{k} X_{j k}^{i} + \sum_{l = 1}^{L} α_{l} Z_{j l}^{i} + \sum_{m = 1}^{M} φ_{m} S_{j m}^{i} .

By assuming the term of unobserved utility (

ε_{j}^{i}

) to be an independent and identical Gumbel distribution, a logit-type model of probability estimation (

P_{j}^{i}

) for living arrangement of the elderly could be expressed in the following equation (3).²⁸ In addition, a multinomial logit model is referred to as the original multinomial logit model (OMNL)²⁹ or the conditional multinomial logit model (CMNL, or the McFadden logit model).²⁵ The conditional multinomial logit model is adopted in this study.

P_{j}^{i} = exp (V_{j}^{i}) \times {[\sum_{j} exp (V_{j}^{i})]}^{- 1} .

Definition of Variables

Dependent variables in the study include coresidence with children, coresidence with spouse only, living alone, and living in an institution. Table 1 presents the definitions of explanatory variables. Tables 2 and 3 provide descriptive statistics.

Table 1.

Definition of Explanatory Variable

Variable	Symbol in model	Definition/value domain
Age	age	Questionnaire items include ranges 65–69, 70–74, 75–79, and 80+. The values are 67, 72, 77, and 82, which are the median ages
Education background	edu	Education is the sum of years of schooling
Self-reported health conditions	hl	The scores include 1 (extremely poor health), 2 (poor health), 3 (normal), 4 (good health), and 5 (extremely good health)
Gender	sex	Categorical variable, 1 for male and 0 for otherwise
Disposable income	d_inc	Estimated based on subjects’ responses (in NTD)
Financial support	sup	Categorical variable, including self-support (sup_se), support by spouse (sup_sp), support by children (sup_ch), and support by others (sup_oth)
Marital status	smar	Categorical variable, including spouse (smar_1), spouse deceased (smar_2), and unmarried/divorced (smar_3)
Number of children	N_ch	Based on the number of male and female children
Care for grandchildren	D_care	Categorical variable, 1 means elderly need to care for their grandchildren and 0 for otherwise
Primary caretaker when sick	depend	Categorical variable, including spouse (depend_sp), children (depend_ch), and others (depend_oth)
Social participation (1)	social_sum	Summation of frequency of participation in six categories of activity. Values are between 0 and 18
Social participation (2)	social	Religious activity (socia1_re), Voluntary services (social_vo), Learning activity (social_lea), Health-promoting activity (social_hea), Leisure activity (social_le), and Political activity (social_po). The scores are: 0 for never, 2 for sometimes, and 3 for regular
Social participation × Health conditions	hsocial	Categorical variable, including poor health with regular social participation (hsoical_bad), normal health with regular social participation (hsoical_me), and good health with regular social participation (hsoical_good)

Table 2.

Descriptive Statistics of Selected Variables

Variable	Alternative	Average	SD
age	j = G	72.57	5.23
	j = S	72.64	5.30
	j = A	74.89	5.62
	j = I	76.77	5.40
	Total	73.04	5.40
edu(year)	j = G	5.27	5.59
	j = S	6.86	5.01
	j = A	5.58	4.90
	j = I	8.10	5.30
	Total	5.78	4.82
N_ch_ male	j = G	2.19	1.18
	j = S	1.90	1.25
	j = A	1.84	1.38
	j = I	1.12	1.32
	Total	2.04	1.25
N_ch_ female	j = G	2.12	1.46
	j = S	1.81	1.39
	j = A	1.91	1.42
	j = I	1.23	1.27
	Total	1.99	1.44
d_inc(in NTD)	j = G	11,100	10,700
	j = S	14,000	11,000
	j = A	11,700	10,800
	j = I	9,700	9,900
	Total	11,900	10,800

Table 3.

One-Way ANOVA of Social Participation and Living Arrangement Preferences

Variable		j= G	j = S	j = A	j = I	Total
social_sum	Average	2.31	2.57	2.13	2.84	2.36
social_sum	SD	2.87	3.02	2.47	3.85	2.88
socia1_re	Average	0.74	0.75	0.74	0.61	0.74
socia1_re	SD	1.08	1.13	1.12	1.10	1.10
social_vo	Average	0.30	0.33	0.19	0.20	0.29
social_vo	SD	1.27	1.47	1.39	1.99	1.48
social_lea	Average	0.17	0.18	0.08	0.08	0.16
social_lea	SD	0.66	0.67	0.45	0.46	0.63
social_hea	Average	0.42	0.49	0.49	0.73	0.46
social_hea	SD	0.98	1.07	1.09	1.24	1.03
social_le	Average	0.55	0.64	0.47	0.92	0.57
social_le	SD	1.00	1.05	0.98	1.33	1.02
social_po	Average	0.13	0.17	0.16	0.28	0.15
social_po	SD	0.53	0.57	0.57	0.70	0.55

Scores for social participation range from 3 “regular participation”, 2 “sporadic participation,” to 0 “never.” Social participation could be further categorized into voluntary services, religious, learning, health promotion, leisure, and political activities. Zero indicates no participation while seventeen indicates regular participation in five of the six categories of activity. Those who prefer to live in an institution have more frequent social participation, followed by those who prefer coresidence with spouse (p= .0001).

The elderly who prefer to live in an institution have less frequent participation in religious activity, and those who prefer to live alone participate in voluntary services less often. Elderly people across all four living arrangements seldom undertake learning activity, and this is especially true for those who prefer to live alone or in an institution. Elderly people who prefer to live in an institution have more frequent participation in health promotion, leisure, and political activity (p= .0001). Statistical results overall show that the elderly people who participate in different activities are similarly distinguished by their living arrangement preferences.

Model Estimation and Discussion

Parameter Analysis

Table 4 presents the results of this study. Models 1–4 analyzed how demographic factors and family resources are associated with elderly people’s living arrangement preferences. Many of variables of the elderly attributes in this study were converted into alternative specific variables and alternative specific dummy variables. This was done because the socioeconomic attributes of elderly people did not vary with alternatives of living arrangements when we employed CMNL to calibrate the variables. As shown in equation (3), the variables of personal attributes in the numerator and the denominator were canceled out. Then a denoted choice has a value, while the undenoted one is zero, indicating that a variable is correlated with a choice of alternative. For example, the number of children is associated with coresidence with children, and therefore the number of children is denoted as the variable-influencing choice of coresidence with children.

Table 4.

Multinomial Logit Model for Living Arrangement Preferences of the Elderly

Variable		Model 1	Model 2	Model 3	Model 4
Constant_G		3.46^*(471.33)	1.64^*(123.99)	1.32^*(96.85)	1.27^*(93.21)
Constant_S		1.27^*(35.69)	2.22^*(60.70)	2.77^*(72.32)	2.71^*(70.71)
Constant_A		1.92^*(287.81)	1.92^*(287.81)	1.92^*(287.80)	1.92^*(287.82)
Constant_I (reference group)		—	—	—	—
age [ASV_S]		0.03^*(53.21)	0.03^*(61.35)	0.03^*(49.56)	0.03^*(50.68)
edu [ASV_G]		−0.03^*(−61.25)	−0.04^*(−88.91)	−0.03^*(−57.98)	−0.03^*(−56.54)
hl	extremely poor health [ASDV_G]	−0.44^*(−46.82)	−0.39^*(−43.17)	−0.38^*(−41.11)	−0.44^*(−46.72)
	poor health [ASDV_G]	−0.08^*(−14.88)	−0.01(−1.34)	−0.02^*(−2.96)	−0.03^*(−4.93)
	extremely good health [ASDV_G]	0.33^*(48.72)	0.21^*(32.68)	0.21^*(31.41)	0.21^*(31.17)
	good health [ASDV_G]	0.28^*(49.90)	0.20^*(36.99)	0.19^*(34.35)	0.20^*(34.96)
	normal (reference group)	—	—	—	—
sex	male [ASDV_S]	−0.19^*(−34.51)	−0.27^*(−49.11)	−0.39^*(−63.24)	−0.40^*(−64.14)
sex	female (reference group)	—	—	—	—
d_inc [ASDV_S]		0.02^*(78.66)	0.01^*(60.22)	0.01^*(59.84)	0.01^*(59.83)
sup	sup_se [ASDV_S]	—	—	−0.22^*(−33.75)	−0.21^*(−32.99)
	sup_sp [ASDV_S]	—	—	0.16^*(12.98)	0.18^*(14.70)
	sup_ch [ASDV_G]	—	—	0.79^*(156.81)	0.78^*(153.56)
	sup_oth (Reference group)	—	—	—	—
smar	smar_1[ASDV_G]	—	2.83^*(236.82)	2.79^*(227.96)	2.82^*(229.38)
	smar_2 [ASDV_G]	—	0.73^*(67.87)	0.61^* (55.01)	0.63^*(56.31)
	smar_3 (reference group)	—	—	—	—
N_ch	N_ch_male [ASV_G]	—	0.14^*(81.30)	0.11^*(62.29)	0.11^*(62.11)
N_ch	N_ch_ female [ASV_G]	—	0.09^*(60.77)	0.07^*(46.82)	0.07^*(45.40)
D_care [ASDV_G]		—	—	1.43^*(165.33)	1.44^*(165.73)
depend	depend_sp[ASDV_S]	—	—	—	0.15^*(15.28)
	depend_ch [ASDV_G]	—	—	—	0.34^*(43.19)
	depend_oth (reference group)	—	—	—	—
Statistic	−2LL	968174.35	899884.70	863850.78	862822.43
Statistic	$ρ^{2}$	0.3299	0.3771	0.4014	0.4021

Note: ^* p < .001.

[ASV_S] = alternative specific variable_S; [ASV_G] = alternative specific variable_G; [ASDV_G] = alternative specific dummy variable_G; [ASDV_S] = alternative specific dummy variable_S.

Models 1–4 shows that older elderly prefer coresidence with their spouse only, and those with more years of schooling also prefer not to coreside with children. Models 1, 3, and 4 indicate that elderly people with self-reported poor health do not prefer to coreside with their children and those with self-reported better health go opposite. This implies that elderly people with self-reported poor health tend not to coreside with children in order not to be a burden. This finding is similar to a Korean study.⁶ Male elderly people are less willing to coreside with a spouse only. The elderly with more disposable income prefer coresidence with spouse only. Previous research presumes that the aforementioned phenomenon means that the elderly devote higher share of their income to “purchase privacy.”⁹ Therefore, elderly people who are financially independent are less likely to rely on children for living arrangements.

Models 2–4 include family resources as a variable. The elderly who provide their own income prefer not to live with a spouse only; otherwise, they tend to live with those who provide financial support. This implies that the financial support provider determines living arrangement preferences. Elderly people who provide their own financial support have more choices of living arrangements and are significantly less likely to live with a spouse only. Elderly people who are married or whose spouse is deceased prefer to coreside with their children. Elderly people who have more (male or female) children prefer to coreside with them. Past research has pointed out that the elderly with more children prefer coresidence with them, especially with a married son.⁷ The elderly who have only daughters also tend to live with adult daughters because of the fact that women now have became more financially independent, late marriage, and remaining single. Elderly people who are cared for by their children or their spouse prefer to live with children only or spouse only. It implies that caretaker provides companionship and emotional support to the elderly.

In short, the elderly with longer schooling or more disposable income prefer coresidence with spouse only, indicating that those with better socioeconomic status prefer independent living arrangements. Both married and widowed elderly prefer coresidence with children, perhaps because coresidence with children provides companionship and emotional support. Specifically, previous empirical research suggests that widowed-elderly people tend to provide child care or perform household services in exchange for emotional support and financial assistance from children. In Taiwan, this phenomenon is significant, because elderly men have tended to have higher levels of income and more institutional income sources than women. Furthermore, the study indicated that reduced work and spouse’s death were identified as the two events most likely to cause income change for the elderly³⁰ so that widowed-elderly people tend to coresidence with children. “Financial support” and “primary caretaker when sick” demonstrate that living arrangement preferences are determined by the resource provider.

Table 5 presents the correlation between social participation and living arrangement preferences. The major difference among the three models is in how social participation indicators are represented. Model 5 includes the frequency of participation in six categories of activity to explain how each category of activity is associated with living arrangement preferences. For ease of comparison, all variables were tested with “co-residence with their children.” Results also indicate that the elderly people who participate in “religious activity,” “health-promoting activity,” “leisure activity,” or “political activity” prefer not to live with children. Model 6 presents total scores of each category to explain how social participation is correlated with living arrangement preferences. Results demonstrate that elderly people with more frequent social participation prefer not to coreside with children.

Table 5.

Multinomial Logit Model for Living Arrangement Preferences of the Elderly When Social Participation is Included

Variable		Model 5	Model 6	Model 7
Constant_G		1.32^**(95.71)	1.29^**(93.87)	1.29^**(95.54)
Constant_S		2.64^**(68.36)	2.68^**(69.90)	2.49^**(65.35)
Constant_A		1.92^**(287.81)	1.92^**(287.82)	1.92^**(287.81)
Constant_I (reference group)		—	—	—
age [ASV_S]		0.03^**(52.67)	0.03^**(51.32)	0.03^**(57.51)
edu [ASV_G]		−0.03^**(−52.03)	−0.03^**(−53.75)	−0.02^**(−34.30)
hl	extremely poor health[ASDV_G]	−0.46^**(−48.38)	−0.45^**(−47.64)	—
	poor health [ASDV_G]	−0.04^**(−7.74)	−0.04^**(−6.74)	—
	extremely good health [ASDV_G]	0.19^**(28.28)	0.22^**(31.85)	—
	good health [ASDV_G]	0.20^**(35.13)	0.20^**(35.82)	—
	normal (reference group)	—	—	—
sex	male [ASDV_S]	−0.41^**(−65.34)	−0.39^**(−63.32)	−0.42^**(−67.89)
sex	female (reference group)	—	—	—
d_inc [ASDV_S]		0.02^**(55.52)	0.02^**(60.12)	0.02^**(62.32)
sup	sup_se [ASDV_S]	−0.23^**(−34.82)	−0.22^**(−33.59)	−0.24^**(−36.79)
	sup_sp [ASDV_S]	0.16^**(12.70)	0.18^**(14.72)	0.15^**(12.72)
	sup_ch [ASDV_G]	0.78^**(153.28)	0.78^**(154.13)	0.77^**(153.40)
	sup_oth (reference group)	—	—	—
smar	smar_1 [ASDV_G]	2.82^**(227.74)	2.83^**(229.78)	2.85^**(232.60)
	smar_2 [ASDV_G]	0.63^**(55.82)	0.63^**(56.79)	0.64^**(57.16)
	smar_3 (reference group)	—	—	—
N_ch	N_ch_ male [ASV_G]	0.11^**(61.45)	0.11^**(62.46)	0.11^**(63.79)
N_ch	N_ch_ female [ASV_G]	0.07^**(44.32)	0.07^**(45.52)	0.06^**(42.105)
D_care [ASDV_G]		1.45^**(166.37)	1.43^**(165.02)	1.46^**(168.69)
depend	depend_sp [ASDV_S]	0.17^**(16.97)	0.14^**(14.89)	0.24^**(24.75)
	depend_ch [ASDV_G]	0.34^**(42.68)	0.34^**(42.87)	0.23^**(29.80)
	depend_oth (reference group)	—	—	—
social_sum [alternative specific variable_G]		—	−0.01^**(−14.18)	—
social	socia1_re [ASV_G]	−0.02^**(−9.66)	—	—
	social_vo[ASV _G]	0.11^**(41.06)	—	—
	social_lea [ASV _G]	0.06^**(18.48)	—	—
	social_hea [ASV _G]	−0.03^**(−16.19)	—	—
	social_le [ASV _G]	−0.08^**(−34.78)
	social_po [ASV _G]	−0.07^**(−18.16)	—	—
hsocial	hsoical_bad [ASDV_G]	—	—	−0.01 (−1.35)
	hsoical_me [ASDV _G]	—	—	−0.20^*(−28.45)
	hsoical_good [ASDV _G]	—	—	−0.24^**(−37.04)
	hsocial_oth (reference group)	—	—	—
statistic	−2LL	860541.11	862284.28	863857.28
statistic	$ρ^{2}$	0.4044	0.4031	0.4021

Note: ^* p < .01. ^** p < .001.

[ASV_S] = alternative specific variable_S; [ASV_G] = alternative specific variable_G; [ASDV_G] = alternative specific dummy variable_G; [ASDV_S] = alternative specific dummy variable_S.

Interest in and frequency of social participation, as well as self-reported health conditions are included in Model 7. We combined the variables of self-reported health conditions and the variable of social participation into a composite variable in order to examine the cross effect of these two conditions in influencing the living arguments of elderly people. Results indicate that both healthy and unhealthy elder people with regular social participation prefer not to coreside with children.

Results of the three models demonstrate that the elderly people with regular social participation prefer independent living arrangements because of the social and affective support they obtain thereby. This indicates that for the elderly, the emotional support from social networks can help replace that from their children.

Odds Ratio and Direct Elasticity

We further discuss the marginal effects of the living arrangement preferences and computed the so-called odds ratio to analyze some of the alternative specific dummy variables. Table 6 presents the odds ratios based on Models 6 and 7. Elderly people who have self-reported extremely poor health and poor health do not prefer coresidence with children because the odds ratio only is 0.64 and 0.96, indicating that coresidence with children is not a preferred living arrangement for elderly people with poor health conditions. The odds ratio of elderly people with self-reported good health and those who prefer coresidence with children are 1.22 and 1.24, respectively, suggesting that the elderly with good health conditions prefer to coreside with children. The odds ratio of male elderly people who prefer coresidence with a spouse is only 0.67 times of that of their female counterparts, showing that coresidence with their wife is not a preferred living arrangement for elderly males.

Table 6.

Odds Ratios of Choice of Living Arrangement

Alternative	Variable	Model 6	Model 7
$j = G$	extremely poor health	0.64	—
	poor health	0.96	—
	Extremely good health	1.22	—
	good health	1.24	—
$j = S$	sex_male	0.67	0.66
$j = S$	sup_se	0.80	0.79
$j = S$	sup_sp	1.20	0.86
$j = G$	sup_ch	2.18	2.17
$j = G$	smar_1	16.86	17.31
$j = G$	smar_2	1.89	0.53
$j = G$	D_care	4.19	4.32
$j = S$	depend_sp	1.16	1.27
$j = G$	depend_ch	1.40	1.26
$j = G$	hsoical_bad	—	0.87
	hsoical_me	—	0.82
	hsoical_good	—	0.78

Married and widowed-elderly people prefer coresidence with children. Especially, the odds ratio of married elderly people is 16.86, indicating that married elderly people prefer coresidence with children. The odds ratios of compound variables (social participation and health conditions) including poor health with regular social participation, normal health with regular social participation, and good health with regular social participation are 0.87, 0.82, and 0.78, respectively, indicating that both healthy and unhealthy elderly people with regular social participation prefer not to coreside with their children.

The direct elasticity analyzes the percentage change in each living arrangement alternative due to the change of economic situation. Table 7 presents the probability of each living arrangement alternative preferred by the elderly based on Model 7. Age and disposable income were tested with coresidence with spouse only. Direct elasticity represents the percentage increase in the probability of living arrangement alternative preferred by elderly people relative to increase in either age or disposable income. Educational background (negative), number of children, and social participation (negative) were tested with coresidence with children. Direct elasticity shows a percentage decline in the probability of coresidence with children relative to the percentage increase in educational background and social participation. Significant differences exist between male and female children, indicating that the number of male children is significantly correlated with preference for coresidence with children.

Table 7.

Elasticity Analysis of Choice of Living Arrangement

Variable		Percentage	Variable		Percentage
age( $j = S$ )	67 years	0.69	d_inc( $j = S$ )	3000 NTD	17.87
	72 years	0.75		6000 NTD	35.73
	77 years	0.80		10000 NTD	59.56
	82 years	0.85		15000 NTD	89.34
social_sum ( $j = G$ )	3 scores	−0.02	N_ch ( $j = G$ )	1 person	0.07 (male)
	12 scores	−0.08		1 person	0.05 (female)
	16 scores	−0.11		2 person	0.15 (male)
				2 person	0.09 (female)
edu ( $j = G$ )	6 years	−0.11
	12 years	−0.21		4 person	0.29 (male)
	16 years	−0.28		4 person	0.18 (female)

Policy Implications

For the elderly, the ideal living arrangement is coresiding with and being attended by their children since it reduces social costs and increases family support for elderly people. Taiwan’s low birth rate, however, makes this arrangement increasingly difficult there. In addition to increasing the birth rate, the government can support increased economic security for the elderly. Strengthening corresponding measures is crucial which may include setting up a stable pension system to maintain financial independence after retirement and encouraging social participation to increase emotional support. Results of this study demonstrate that elderly people with regular social participation are less dependent on coresidence with children and are more receptive to independent living arrangements. Therefore, promoting care in communities through which elderly people may assist each other by means of community networks is a possible means to mitigate the impact of aging population and low birth rate. In the long run, providing adult children economic motivation in the form of tax incentives may encourage them to live with their parents.

Conclusions

In this study, we use the discrete choice theory to depict the living arrangement preferences of people aged sixty-five and over. Results of this study illustrate that elderly people of the “older elderly” group and those with higher disposable income prefer coresidence with spouse only. Elderly people with higher education have lower preference for coresidence with their children, implying that the elderly with better socioeconomic conditions prefer independent living arrangements. Living arrangement preference are affected by the source of financial support, that is, the elderly prefer to coreside with their children when their financial support comes from the children, and the elderly with poor health conditions prefer not to coreside with children in order not to be a burden.

The elderly who have more frequent social participation (especially in religious, health promotion, leisure, or political activities) coreside with their children less frequently, indicating that elderly people with more frequent social participation prefer independent living arrangements. This may be because they have more sources for social support or not want to disturb children’s life. In brief, the elderly who have more need for family resources, including financial and emotional assistance, are more likely to coreside with their children. In addition, the elderly who are more independent, such as those with higher income or with more social participation/networks are more likely to prefer independent living arrangements.

Considering potential and effective demand, this study found that elderly persons with better socioeconomic status prefer either independent living arrangements or coresidence with their children. Elderly people with more family resources prefer to coreside with their children. The elderly with more frequent social participation also prefer independent living arrangements.

It is also necessary to include children’s willingness and ability to coreside with their elderly parents in order to fully understand expected living arrangements. In addition, to fully explain living arrangement preferences of the elderly, it is crucial to include the harmony of relationships among family members. These two factors, however, were not measured in the “Report on the Status of Elderly People in Taiwan and Fukien Areas” and so we have not included their influence in this study.

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This article is part of results of the research project supported by National Science Council in Taiwan (NSC 95-2415-H-006-008).