Culture Matters? Family Norms, Living Arrangements, and Loneliness

Abstract

The increasing prevalence of loneliness has become a new global concern of public health. Previous research devoted substantial efforts to exploring micro-level risk factors of loneliness but generally ignored the potential role of macro-level forces, such as culture. The theory predicts that an individual's loneliness results from unmet expectations for desired social relations or living conditions, a process deeply influenced by societal-level cultural norms. This study empirically examines the theoretical hypothesis by investigating how family culture affects people's preferences for living arrangements and how this affects loneliness in older people. Using nationally representative survey data of older Chinese and historical records of family genealogies, the study shows that prefecture-level genealogy density-a measure of family culture-significantly shapes older adults' perceptions of ideal living arrangements. Meanwhile, it finds that while older Chinese living alone exhibit higher levels of loneliness than those living with adult children, the loneliness disparity is more pronounced in prefectures with stronger family norms. The findings are consistent with the theoretical hypotheses and underscore the essential role of cultural norms in shaping people's preferences for living arrangements and loneliness.

Full text

Culture Matters?
Family Norms, Living Arrangements, and Loneliness*
Shu Cai† Wei Li‡ Jia Wang§
Abstract
The increasing prevalence of loneliness has become a new global concern of public health.
Previous research devoted substantial efforts to exploring micro-level risk factors of loneliness
but generally ignored the potential role of macro-level forces, such as culture. The theory
predicts that an individual’s loneliness results from unmet expectations for desired social
relations or living conditions, a process deeply influenced by societal-level cultural norms. This
study empirically examines the theoretical hypothesis by investigating how family culture
affects people’s preferences for living arrangements and how this affects loneliness in older
people. Using nationally representative survey data of older Chinese and historical records of
family genealogies, the study shows that prefecture-level genealogy density—a measure of
family culture—significantly shapes older adults’ perceptions of ideal living arrangements.
Meanwhile, it finds that while older Chinese living alone exhibit higher levels of loneliness
than those living with adult children, the loneliness disparity is more pronounced in prefectures
with stronger family norms. The findings are consistent with the theoretical hypotheses and
underscore the essential role of cultural norms in shaping people’s preferences for living
arrangements and loneliness.
Keywords: loneliness, living arrangements, oldest old, cultural norms
JEL Classification: I14, I15, J14
* Shu Cai acknowledges financial support from the National Natural Science Foundation of China (Project
No.: 72173056) and the Fundamental Research Funds for the Central Universities (Project No.:
23JNQMX29). Jia Wang acknowledges financial support from the Start-up Fund of the Hong Kong
Polytechnic University (Project No.: P0045993). All errors are our own. The authors contribute equally to
the paper.
† Institute for Economic and Social Research, Jinan University. Email: shucai.ccer@gmail.com.
‡ Department of Economics, the Hong Kong University of Science and Technology. Email:
wlick@connect.ust.hk.
§ Department of Applied Social Sciences, the Hong Kong Polytechnic University. Email: jia-
apss.wang@polyu.edu.hk.

1
1. Introduction
The increasing prevalence of loneliness has become a new public health concern across many
countries (Cacioppo & Cacioppo, 2018; Surkalim et al., 2022), posing significant challenges
for individuals, particularly older adults, to maintain good mental and physical health during
this global “loneliness pandemic.”1,2 Researchers and policymakers have devoted substantial
efforts to exploring potential risk factors of loneliness among the general population and the
unequal exposures to loneliness across socioeconomic groups. So far, the current research has
primarily concentrated on micro-level factors of loneliness, including individual and household
attributes such as gender, age, marital status, ethnicity, education, family income, living
arrangement, and early-life experiences (e.g., De Jong Gierveld & Van Tilburg, 1999; Furuya
& Wang, 2023; Hawkley et al., 2008; Luo et al., 2012; Raymo & Wang, 2022; Russell, 2009).
However, we know relatively little about whether and how the macro-level factors affect
individuals’ experiences of loneliness. This oversight is unfortunate, as loneliness is not merely
a subjective feeling associated with individual and household characteristics but also has
societal-level roots, such as cultural norms, that may shape people’s perceptions of ideal life
circumstances and thus their feelings of loneliness (De Jong Gierveld & Tesch-Römer, 2012).
Culture is a shared set of beliefs, expectations, and collective practices of a society, which
can affect individuals’ preferences and attitudes (Bachrach, 2014; Giuliano, 2007). In particular,
as a salient societal-level factor, family culture norms significantly influence individuals’
perceptions of ideal living arrangements. Individuals may feel lonely when their social
relationships or living conditions do not match their desired situations (De Jong Gierveld &
Tesch-Römer, 2012). Since the prevalent cultural scheme sets a socially valued or endorsed
living standard or condition, it is reasonable to view the prevalent cultural norms as a collective
reference against which individuals evaluate their actual social relationships. This implies that
individuals whose actual living conditions align with the culturally prescribed standards
1 “Increased loneliness has become a global public health issue.” Open Access Government, February 10,
2022. https://www.openaccessgovernment.org/loneliness-health-countries/129381/
2 These challenges are magnified by the fundamental life changes accompanied by the COVID-19 pandemic,
which has led to elevated social isolation and loneliness (Mckeown et al., 2021; Rebechi et al., 2024).

2
experience “cultural concordance” and thus report lower loneliness. On the other hand, those
whose living conditions deviate from the culturally valued ones experience “cultural
discordance” and are consequently exposed to higher loneliness. Moreover, the loneliness
disparity between adults in “cultural discordance” versus “cultural concordance” situations is
more salient in the presence of more powerful cultural norms. These scenarios are particularly
relevant for societies with rapid socioeconomic and demographic transformations, where the
cultural norms remain relatively stable. Unfortunately, these theoretical possibilities of how
societal-level cultural norms shape loneliness have not been thoroughly examined in empirical
research.
This study closes the gap by investigating how family culture shapes people’s preferences
for living arrangements and how this affects loneliness in older people. Specifically, we use
prefecture-level genealogy density to measure family culture and show that it significantly
shapes older adults’ perceptions of ideal living arrangements. We then investigate whether and
how loneliness disparities by living arrangements would change across prefectures with
varying intensity of family culture. The Chinese context suits our research purposes on several
fronts. First, China has the largest aging population and the highest accelerated aging rate in
the world (Zhang et al., 2012), and recent studies reported an increasing prevalence of
loneliness among older Chinese since the early 1990s (Yan et al., 2014). Second, China has
undergone dramatic family change and household structure transformation over the past several
decades (Cai et al., 2023; Li et al., 2020), which to some extent contributes to the increasing
proportions of living alone or living with a spouse only among older Chinese (Lei et al., 2015).
Third, despite dramatic socioeconomic and demographic changes, the traditional family culture
that values filial piety and intergenerational coresidence with adult children persists, and many
older Chinese strongly value living with adult children (Chen, 2019; Chen & Short, 2008). The
collision between rapid and dramatic social transformations and persistent culturally endorsed
family norms exposes an increasing number of older Chinese to a situation of “cultural
discordance” in family living arrangements. This may lead to an elevated prevalence of
loneliness among older people, which is likely to be more pronounced in regions with stronger

3
family culture.
The rest of the paper is organized as follows. Section 2 describes a conceptual framework
and the theoretical hypotheses. Section 3 introduces the background of the study and reviews
previous research. Section 4 presents the data, measurement, and analytical strategy. Section 5
reports the empirical results. Section 6 discusses our main findings and policy implications.
2. Conceptual Framework
2.1. Why culture matters for loneliness?
Individuals’ preferences for desired social relations and living conditions are deeply
influenced by societal culture. As Bachrach (2014) noted, the fundamental cognitive dimension
of culture forms social norms that characterize socially (un)acceptable and (un)desirable ways
of living and behaving. These social norms consist of powerful reference systems dictating
individuals’ preferences and standards for certain behaviors throughout their lives (Carr & Utz,
2020; Lin & Yi, 2011; Park et al., 1999).
As a subjective evaluation of one’s social relationships and interactions, loneliness
accompanies perceived deficiencies in the quantity and quality of social relationships (Hawkley
et al., 2008). This indicates a cognitive component inherent in loneliness when people’s actual
social conditions fall behind their desired standards. In particular, the cognitive discrepancy
perspective views loneliness as the result of individuals’ unmet social expectations (De Jong
Gierveld & Tesch-Römer, 2012; Peplau & Perlman, 1982). People feel lonely primarily
because they perceive and experience discrepancies between their existing and expected social
relationships. The latter, however, is profoundly shaped by local cultural norms. Since the
majority of social members tend to internalize or abide by the prevalent cultural norm to
calibrate their behaviors, it is reasonable to view the prevalent local culture as a collective
reference system against which individuals evaluate their social relationships.
In particular, the prevalent family culture in a given society sets a collectively valued
living standard or condition. People whose actual living circumstances do not align with the

4
culturally prescribed one—those who experience “cultural discordance”—are vulnerable to
heightened levels of loneliness. This is particularly true in areas where the local cultural norm
is more powerful. In contrast, people whose actual social relationships match the culturally
prescribed standard—those who experience “cultural concordance”—tend to exhibit lower
loneliness. Meanwhile, their feelings of loneliness are less likely to vary depending on the
strength of local family culture.
The above understanding of individuals’ loneliness as a cognitive process is presumably
more relevant in societies where the sociodemographic landscape is transforming rapidly, but
the societal-level cultural traditions remain relatively stable. The contradiction between
dramatic sociodemographic change and cultural continuity creates a breeding ground for the
perceived discrepancy and unmet needs between individuals’ actual and socially prescribed
living conditions, a precondition prone to the rise of loneliness.
2.2. Theoretical hypotheses
In societies valuing family life and intergenerational coresidence with adult children, the
cognitive process of loneliness described above implies that older adults living with adult
children realize “cultural concordance” in terms of family living arrangements and thus exhibit
lower levels of loneliness than their counterparts in the situation of “cultural discordance,” i.e.,
living in any other arrangements without adult children. Moreover, the difference tends to
enlarge as local family culture strengthens.
For illustration, Figure 1 depicts that loneliness disparities between older adults
experiencing “cultural discordance” (the upper solid line) versus “cultural concordance” (the
lower solid line) in family living arrangements increase monotonically across regions with
relatively weak-, middle-, and strong-level of family culture.3 We expect that older adults who
do not live with adult children (e.g, living alone, living with spouses without adult children;
living with others without family members) are increasingly prone to feeling lonely in areas
with stronger family norm that collectively values intergenerational coresidence (the positive
3 For simplicity, we assume that the increase in the loneliness disparity between the two groups of older
adults is a linear function of the strength of family culture.

5
slope). On the other hand, loneliness among older adults living with adult children remains
relatively stable across regions with various levels of family culture as they already meet the
culturally prescribed living arrangement (the flat line). Absent the role of family culture, the
levels of loneliness of older adults not living with adult children should not change across areas
with varying intensity of family culture (the counterfactual, dash flat line).
Accordingly, we have the following hypotheses regarding the relationships between living
arrangements and loneliness and how their associations would vary by the strength of family
culture across regions:
Hypothesis 1: All else being equal, older adults living without adult children have higher
levels of loneliness than those who live together with adult children.
Hypothesis 2: The difference in loneliness of older adults living with versus without adult
children is larger in regions with stronger family cultures.
Hypothesis 2a: Older adults living with adult children experience similar levels of
loneliness despite the varying strengths of local family culture. On the contrary, older
adults living without adult children are exposed to increasing levels of loneliness in
regions with stronger family cultures.
Prior studies have mainly focused on the examination of Hypothesis 1. However, scant
research attention has been paid to test Hypotheses 2 and 2a. In the empirical analyses below,
we examine these hypotheses throughout.
3. Background and Literature Review
3.1. The rise of loneliness and the Chinese context
There has been a rising trend of reported loneliness in many countries (Banerjee et al.,
2023; Surkalim et al., 2022). Loneliness has also been increasingly viewed as a public health
crisis, given its close associations with a range of unfavorable outcomes, including poor health,

6
mortality, functional limitations, cognitive impairment, and depression (Banerjee et al., 2023;
Cacioppo et al., 2006; Holt-Lunstad et al., 2015; Luo et al., 2012; Wilson et al., 2007). It is thus
pivotal to understand why and under what conditions people feel lonely, based on which
targeted policy efforts could be more effectively implemented to reduce the occurrence of
loneliness and its associated risks for the population.
Around the globe, China has the largest aging population and the highest accelerated aging
rate (Zhang et al., 2012). Data from the National Bureau of Statistics show that the number and
proportion of older people aged 65 and above in China have increased continuously since 1978,
with the annual growth rate accelerated after 2000 (Lu & Liu, 2019). Against the backdrop of
rapid population aging, there has been a growing prevalence of loneliness among older Chinese
since the early 1990s (Yan et al., 2014). This increasing trend of loneliness is also accompanied
by the growing incidence of poor mental health, such as distress and hopelessness (Niu et al.,
2020), which pose significant challenges for the government, society, and families to provide
adequate care and support for older adults.
Over the past few decades, China has witnessed dramatic family and household structure
transformation since the onset of the economic reform, which brought substantial changes in
the living arrangements of older Chinese. In particular, scholars have demonstrated that the
proportion of the traditional large linear family with two or more generations living under the
same roof has dramatically shrunk from about 40% in 1982 to less than 5% after the 2000s (Li
et al., 2020). Meanwhile, studies also show that older Chinese have become increasingly
independent in terms of living arrangements, where the percentages of living alone (or solo-
living) or living with only a spouse but without any adult children appear to be rising
substantially among older Chinese over time (Cheung & Yeung, 2015; Hu & Peng, 2015; Lei
et al., 2015).
The rapid shift toward more diverse and independent living arrangements that deviate
from the traditional family culture could increase exposure to feeling lonely among older adults.
Chinese societies have long been featured by strong family ties and intergenerational solidarity.
Ideal family life often involves multiple generations living in the same household, such as “a

7
five-generation family” or “five generations co-residing” (wu shi tong tang 五世同堂), in
which older people are believed to enjoy family love and joy (Baker, 1979). Although the
prevalence of the ideal family life of the “five-generation family” is quite rare in reality (Baker,
1979), multigenerational households where older people co-reside with adult children are still
widely perceived to be ideal living arrangements by traditional family values (Logan & Bian,
1999; Sun et al., 2019). Therefore, the increasingly diverse family living arrangements other
than living with adult children may expose more older people to the situation of “cultural
discordance” that leads to elevated loneliness relative to those who co-reside with adult
children. In addition, the loneliness disparity between older adults experiencing “cultural
discordance” versus “cultural concordance” is presumably larger in regions with more
powerful family norms.
3.2. Prior research and limitations
Prior studies have primarily focused on individual traits and household characteristics as
potential correlates of loneliness (Luhmann et al., 2022). For example, research has reported
that gender, age, ethnicity, marital status, and household structure are significant demographic
factors associated with one’s loneliness (Hawkley et al., 2008; Russell, 2009). Socioeconomic
status indicators, such as income and education, also strongly predict loneliness, where
disadvantaged groups with lower income and education report feeling lonely more frequently
relative to their more affluent counterparts (Raymo & Wang, 2022).
However, we know relatively little about whether and under what circumstances the
macro-level social forces and contexts would shape individuals’ loneliness. Theoretically,
whether people feel lonely and how they react to loneliness can be shaped by the larger social
milieu, such as culture (De Jong Gierveld & Tesch-Römer, 2012; Luhmann et al., 2022).
Specifically, cultural norms offer a collectively valued standard against which people may
evaluate their actual living conditions. The process of cognitive comparison between one’s
actual and culturally desired living standard would generate “cultural discordance” (or “cultural
concordance”) that may (or may not) make a person vulnerable to living alone or living with a

8
spouse only.
The theoretical possibility of loneliness arising from a contrast between individuals’ own
living circumstances and the collectively valued standard has not been rigorously evaluated in
the empirical literature, although family living arrangements have long been recognized as a
significant predictor of loneliness. A handful of studies have examined how the relationships
between living arrangements and loneliness vary across different cultural settings (De Jong
Gierveld & Van Tilburg, 1999; De Jong Gierveld et al., 2012; Jylhä & Jokela, 1990). However,
they lacked a unified measurement of culture and thus did not formally examine the role of
family culture in shaping individuals’ preference for living arrangements and their feelings of
loneliness.
Specifically, in an earlier study on six European countries, Jylhä & Jokela (1990)
investigated the association between living alone and loneliness among older people and found
that older adults living alone experienced loneliness more frequently than people living with
others, a pattern particularly true in countries where older people are expected to live together
with their families. De Jong Gierveld & Van Tilburg (1999) compared associations between
living arrangements and loneliness in the Netherlands and Italy. They found that coresidence
with adult children (without partners) resulted in more loneliness among older adults in the
Netherlands but less loneliness among older Italians. The authors attributed observed
disparities in the coresidence-loneliness associations to the more traditional family orientation
of Italians and the more individualized family orientation of the Dutch people.
In a more recent study, De Jong Gierveld et al. (2012) re-examined the associations
between living arrangements and older adults’ loneliness in five Eastern and Western European
countries. The authors argued that the normative ideas about family responsibilities and shared
households shape living-arrangement differences in loneliness and expected that living alone
would make older adults more likely to feel lonely in Eastern than in Western Europe. However,
their empirical findings do not support the hypothesized regional disparities in loneliness
associated with living arrangements and suggest that the types and patterns of family
relationships might not be directly comparable across countries (Dykstra, 2009).

9
The previous studies offer valuable insights into cultural variation in the linkages between
living arrangements and loneliness. However, they lack direct measures of culture. As a result,
their conclusions were based on comparative analyses across countries. Such analyses cannot
exclude country-level factors other than cultural norms that may drive the observed variation
in relationships between living arrangements and loneliness across countries. Additionally,
although existing studies have viewed loneliness to be rising from a process of cognitive
discrepancy between one’s preferences and reality (De Jong Gierveld & Tesch-Rӧmer 2012),
the empirical challenge is that the personal or individual expectations or preferences are often
endogenous and using self-reported preferences may introduce bias that blurs the genuine
process of cognitive discrepancy.
To overcome these challenges, this study measures the family culture valuing family life
and intergenerational coresidence with the China’s regional-level lineage genealogy (jia pu 家
谱) density. In Chinese society, genealogy books describe family history and relationships
among lineage members within and across generations. The genealogy carries traditional
family culture for two main reasons. First, family genealogy records show people’s interest in
family origins, lineage continuity, intra-lineage member connections, and intergenerational
solidarity (Shiue, 2016; Xie & Yuan, 2022). Second, family genealogy involves experiences
living in an extended family, corresponding to the culturally ideal family life that emphasizes
adult children’s (often sons’) filial piety and care responsibilities for supporting older parents
by living in the same household. In this sense, family genealogy stresses the instrumental
function of the extended family in providing services, insurance, and other resources to family
members, which is viewed as a form of strong family culture (Alesina & Giuliano, 2011; Xie
& Yuan, 2022).
Family genealogy has been shown to be an appropriate measure of family culture. For
instance, using data from the China Family Panel Studies (CFPS), earlier research demonstrates
that residents of regions with a higher genealogy density are more likely to agree that family is
important, to have close family relationships, and to interact more frequently with other family
members (Xie & Yuan, 2022). To further strengthen the validity of family genealogy in

10
representing family culture, we draw on multiple survey datasets and investigate the
associations between the prefecture-level family genealogy density and older adults’
preferences for living arrangements (see Appendix A1 for details). First, using the Chinese
Longitudinal Healthy Longevity Survey (CLHLS) data, we found a significant positive
relationship between the family genealogy density (the logged number of genealogies per
million people) in a prefecture and local older people’s preferences for living with adult
children (Column (1) in Table A1). Second, relying on data from the China Health and
Retirement Longitudinal Study (CHARLS), we documented that the higher the genealogy
density in a prefecture is, the more likely people (aged 45 and above) would agree that living
with adult children is the best living arrangement for an older person, regardless of the presence
of a spouse (Columns (2) and (3) in Table A1). Besides, the prefecture-level family genealogy
density positively relates to an older respondent’s willingness to rely on adult children for late-
life financial support (Column (4) in Table A1). Taken together, it is reasonable to view regional
genealogy density as a valid measure of the traditional family culture that deems coresidence
with adult children as an ideal living arrangement for older Chinese.
It is worth noting that there is substantial variation in the strength of the local family
culture measured by the family genealogy density. Figure 2 depicts the geographic variation in
the family genealogy density across all the prefectures in the CLHLS data, where darker color
indicates higher density (stronger family norm). On average, Southeast China tends to have a
higher density of family genealogy, i.e., stronger family culture, than Northeast and Southwest
China. Leveraging the geographic variation in the intensity of family norms within China, we
examine whether and how the loneliness difference between older Chinese living with versus
without adult children would change across prefectures with varying levels of local family
culture.4
4 Theoretically, one can use people’s stated preferences of living arrangements and their actual living
arrangements to construct a measure of the perceived discrepancy in living arrangements and examine how
that affects their reported loneliness. However, people may prefer to live with adult children if they feel
lonely. To avoid the endogeneity problem, we use the regional-level genealogy density to measure family
culture.

11
4. Empirical Strategy
4.1. Data
This paper uses data from several sources. The primary one is the Chinese Longitudinal
Healthy Longevity Survey (CLHLS) conducted by the Center for Healthy Aging and
Development Studies at Peking University. The CLHLS launched its baseline survey in 1998
and executed follow-up surveys in 2000, 2002, 2005, 2008, 2011, 2014, and 2018, respectively.
The survey was conducted in 22 of the 31 provinces in mainland China, where 85% of the
population resides. The initial target of the CLHLS is older people aged 80 and above. Since
2002, the survey has extended the sample to those aged 65 and above.
The CLHLS provides comprehensive information on older people’s demographic and
socioeconomic characteristics. Additionally, for CLHLS 2008 – 2014, the Center for Healthy
Aging and Development Studies at Peking University also collected county- and prefecture-
level demographic and economic conditions from various statistical yearbooks in the
community data based on which we identified the residential prefectures of respondents. Since
the location information is crucial for merging individual data with the prefecture-level
measure of traditional family cultures (elaborated below), we only used the CLHLS 2008 –
2014 for our main analyses.5
We restricted the sample to those aged 65 and above. As recording family genealogy is a
prevalent cultural practice only among the Han Chinese, we excluded all ethnic minorities
other than Han in the main analyses and obtained a total sample of 28,704 observations. Among
them, 3,792 (13%) have missing values on loneliness, and another 359 (1%) have missing
values on explanatory variables. We finally obtained an analytical sample of 24,553
observations contributed by 14,226 older people.
We measured traditional family culture using historical records of family genealogies
5 By tracing the respondents, we can impute respondents’ residential prefectures and extend the sample back
to the wave 2002. However, such an exercise can only be conducted among older people who were
successfully followed over time in the survey, resulting in potential selection bias. We use the extended
sample from CLHLS 2002 – 2014 as a robustness check and report the results in Section 5.4.

12
from The General Catalog of Chinese Genealogy (GCCG) edited by the Shanghai Library. As
the most comprehensively digitized data source of family genealogies in mainland China, the
GCCG documents each genealogy’s surname, location, and compilation date. Based on the
location information, we merged the genealogy indicators with the CLHLS data by prefecture
names. Among the 186 prefectures in the CLHLS sample, the majority (166 prefectures, 89%)
were successfully matched with genealogy records. The unmatched prefectures only contain
5.7% of all respondents from the CLHLS survey. In the empirical analyses, we assumed the
unmatched prefectures did not have genealogy records in history and assigned zero values to
them. Excluding the unmatched samples from the analysis does not change our main
conclusions.6
In addition to the two primary datasets, supplementary analyses drew on several other data
sources, including the CHARLS survey data, the China City Statistical Yearbooks, and other
historical data. Appendix A2 provides a detailed description of these data sets.
4.2. Measurements
Loneliness. The CLHLS asked respondents, “Do you feel lonely?” The respondents rated
their subjective feelings on a Likert scale ranging from one to five, indicating “always,” “often,”
“sometimes,” “seldom,” and “never,” respectively. We constructed two measures of loneliness.
First, we created a continuous, 1-5 level of loneliness by reverse-coding options so that higher
values indicate more loneliness. Second, we constructed a dummy indicator of loneliness where
options of “always,” “often,” and “sometimes” were collapsed as 1 – “feel lonely,” whereas
“seldom” and “never” were categorized as 0 – “not feeling lonely.”7
Living arrangements. In the family roster, for each person living with the focal older
respondent in the same household, the CLHLS collected information on his or her relationship
with the respondent and other basic sociodemographic characteristics. Based on the
information on family relationships, we created four mutually exclusive categories of living
6 See Section 5.4 for more details about the robustness check.
7 Findings remain qualitatively the same based on an alternative coding scheme where we coded “always,”
“often,” “sometimes,” and “seldom” as 1 and “never” as 0. Results are available upon request.

13
arrangements: living with adult children (with or without the spouse), living alone, living with
the spouse only, and living with others (people other than a spouse or adult children).8 The
first category is our reference group, including older adults who meet the culturally valued
living arrangement, i.e., “cultural concordance.” In contrast, older adults in the other three
living arrangements were experiencing “cultural discordance” as they were not co-residing
with adult children.9 For simplicity, we refer to the group living with adult children (with or
without the spouse) as “living with adult children” hereafter.
Family culture. We used family genealogy density in a given prefecture to measure its
local family culture. To account for differences in population size, we normalized the number
of genealogies in a particular prefecture by its total population (in millions) in the year 2000.10
We took the logarithm form of the density of genealogy per million people to reduce the
influence of extreme values. We then subtracted the mean value of the logged family genealogy
density among all the sampled prefectures for easier interpretation of the regression results.
Other control variables. We controlled for a set of individual-level and regional-level
characteristics in the regressions. The individual-level variables include older people’s gender
(0=female and 1=male), age (in years), age squared, marital status (0=unmarried and
1=married),11 educational attainment (illiterate, primary school, middle school or above), main
occupation categories before age 60 (farmer, industrial worker, manager, and others), and
residential type (0=rural area, 1=urban area). To account for regionally geographic and
economic factors that may correlate with both loneliness and living arrangements (or family
culture), we also control for the regional-level covariates, including the share of the plain on
the county level and the logarithm of one-year-lagged GDP per capita on the prefecture level.
8 People in the categories of “living with the spouse only” and “living with adult children (with or without
the spouse)” may also live with some other people. Older people living in nursing homes or institutions were
classified as “living with others.”
9 For the older people who were childless (only 3.7% of the CLHLS sample), we also considered them to
be experiencing “cultural discordance.”
10 Statistics of the total population sizes of all prefectures were from the 2000 Population Census. Following
the practice of prior studies (Chen et al., 2020; Zhang, 2020), we only utilized the family genealogies
compiled in or before 1950 to avoid reverse causality.
11 Unmarried respondents included those who were widowed, divorced, and never married. The vast
majority (98%) of them were widowed in our analysis sample.

14
4.3. Empirical specifications
We follow prior studies and estimate the following equation to examine the relationship
between living arrangements and loneliness:
 =   +  ℎ ℎ  
+ ℎ ℎ
+ + + +  , (1)
where  denotes the individual,  denotes the resident prefecture,  denotes the resident
province, and  denotes the survey wave. The outcome of interest, , is the loneliness of
the respondent measured by either a dummy indicator (1=feeling lonely) or a continuous
variable (1-5) with larger scores indicating higher levels of loneliness. The main independent
variables are the categories of living arrangements, including “Living alone,” “Living with the
spouse only,” and “Living with others,” where “Living with adult children” is the reference
group and omitted from the regression.
 is a set of control variables including older people’s gender, age, age squared,
marital status (the indicator of married), education categories (including indicators of primary
school and middle school or above), main occupation before age 60 (including indicators of
industrial worker, manager, and others), and residential type (the indicator of urban area). 
also contains the county-level share of plain and the prefecture-level logarithm of one-year-
lagged GDP per capita obtained from the community data of CLHLS.  and  are the
survey wave fixed effect and the residential province fixed effect, respectively. Standard errors
are clustered at the individual level. All estimations are adjusted by the sampling weights.
According to Hypothesis 1 (H1), we expect , , and  to be positive, that is, older
adults living without adult children exhibit higher levels of loneliness than those living with
adult children in the same household.
Furthermore, to investigate how the disparities in older adults’ loneliness across their
living arrangements are affected by family culture, we extend Equation (1) and estimate the
following equation:

15
 =   +  ℎ ℎ  
+  ℎ ℎ
+  × 
+ ℎ ℎ   × 
+ ℎ ℎ × + 
+ × +  + + + , (2)
where  is the logarithm of the family genealogy density (mean-centered) in
prefecture  and province . To account for the heterogeneous relationship between marital
status and loneliness along family culture that may confound the relationships we are interested
in, we also include the interaction term between the indicator of being married and family
culture as an additional control variable in Equation (2). The other variables are defined in the
same way as in Equation (1).
We are interested in the coefficients of interaction terms, , , and , and expect them
to be positive according to Hypothesis 2 (H2). That is, differences in loneliness of older people
not living with adult children versus those living with adult children tend to be larger in
prefectures with stronger family cultures. The coefficient of family culture, , represents the
influence of family culture on loneliness among older people living with adult children (the
reference group). According to Hypothesis 2a (H2a), we expect it to be not significantly
different from zero. That is, the average level of loneliness of older people living with adult
children is the same across prefectures with varying strengths of family culture.
5. Results
5.1. Summary statistics
Table 1 reports the summary statistics of the main variables we examined in the analytical
sample. The mean of the dummy variable “ ” indicates that 32% of observations
report feeling lonely at least for some time. The average score of the five-level measure of

16
loneliness is 2.05. As for the living arrangements, 19% of observations live alone, another 24%
live with a spouse only, over half of the observations (56%) live with adult children (with or
without the spouse), and only 1% live with people other than spouses and children (including
older adults in nursing houses). The summary statistics also indicate substantial geographic
variation in family culture measured by genealogy density across the analytical sample.
Regarding other demographic and socioeconomic characteristics, 46% of the observations
are male, the average age is 85, and 38% are married. As expected, the respondents’ educational
attainment is rather low. Over half of the observations are illiterate (58%), 31% have primary
school education, and only 11% have a middle school degree or above. Most observations were
farmers before age 60 (65%), and less than half reside in urban areas (47%).
5.2. Main results
Table 2 presents the ordinary least squares (OLS) estimates of Equations (1) and (2)
specified in Section 4.3. Columns (1) and (2) use the dummy indicator of feeling lonely as the
dependent variable, whereas Columns (3) and (4) use the five-level loneliness measure as the
dependent variable.
As shown in Column (1), compared to older adults living with adult children, those living
alone and living with others are significantly more likely to feel lonely. Other factors being
equal, the probability of feeling lonely is 15.0% higher for those living alone and 15.2% for
those living with others relative to those living with adult children. For older people living with
the spouse only, the probability of feeling lonely is not significantly different from those living
with adult children. After adding the genealogy density measure and its interactions with living
arrangements and marital status in Column (2), the main coefficients of living arrangements
now indicate that among older adults in prefectures with an average level of the strength of
family culture (mean of the logged genealogy density), those who live alone or live with others
are more likely to feel lonely (14.6% and 15.4% higher in probabilities of reporting loneliness,
respectively) than those who living with adult children. The main coefficient of genealogy
density indicates little association between family norms and feelings of loneliness among

17
respondents living with adult children (= -0.001, =0.13), the group who achieved cultural
concordance under China’s traditional family norms.
Nevertheless, traditional family norms may interact with living arrangements to influence
one’s perception of loneliness. The coefficients of the interaction terms between the genealogy
density and indicators of living arrangements are all positive: The interaction term coefficient
between genealogy density and living alone is 0.018 and statistically significant at the 10%
level; The interaction term coefficient between genealogy density and living with the spouse
only is 0.021 and statistically significant at the 1% level; The interaction term coefficient
between genealogy density and living with others is large in magnitude (0.035), although it is
statistically insignificant due to a much larger standard error.
Overall, these results suggest that older people living without adult children—groups
failing to achieve cultural concordance—are more likely to feel lonely than those living with
adult children. In particular, these relationships tend to be stronger for respondents who resided
in regions with more powerful family norms (higher family genealogy density). The
magnitudes of the coefficients of the interaction terms are large. For example, one standard
deviation increase in the family genealogy density (i.e., 1.48; see Table 1) is associated with a
3.1 percentage points (i.e., 1.48 times 0.021) larger gap in the likelihood of feeling lonely
between those living with the spouse only and those living with adult children. This is about
half of the difference in the likelihood of feeling lonely between older people with primary
school education and those without formal education. Similarly, one standard deviation
increase in the family genealogy density is associated with a 2.6 percentage points (i.e., 1.48
times 0.018) larger gap in the likelihood of feeling lonely between those living alone and those
living with adult children.
Columns (3) and (4) in Table 2 report the OLS regression results using the five-level
loneliness measure as the dependent variable. The directions and significance levels of the main
coefficients are comparable to those in Columns (1) and (2), except that the estimates of the
interaction terms are less precise. In terms of magnitudes, the estimates in Column (3) indicate
that, other factors being equal, the average level of loneliness among the older adults living

18
alone (or those living with others) is 0.339 (or 0.264) points larger than those living with adult
children, whereas the average level of loneliness is not significantly different between older
adults living with the spouse only and those living with adult children. Column (4) shows that,
as the logged family genealogy density increases by one standard division, the difference in the
level of loneliness between older adults living with adult children and those living alone is
0.046 (i.e., 1.48 times 0.031) points larger (t=1.48), and the difference between older adults
living with adult children and those living with the spouse only is 0.044 (i.e., 1.48 times 0.030)
points larger (t=1.88).
To better visualize the results, we plot predicted levels of loneliness of older adults by their
living arrangement across different prefectures with varying strength of family culture. We do
not plot the relationship between living with others and loneliness, as only 1% of the
respondents lived with others in our sample (see Table 1). Empirical patterns illustrated in
Figure 3 are largely consistent with the theoretical pattern shown in the conceptual Figure 1.
Loneliness exhibited by older Chinese living under “cultural concordance” (coresidence with
adult children) is basically a flat line that is immune to the varying strength of family norms
across prefectures. Older Chinese who experience a “cultural discordance,” such as those who
live alone or with a spouse only (the two dash lines), are exposed to elevated loneliness as the
family culture in a given region strengthens.
Taken together, the results in Table 2 are consistent with H1. All else being equal, older
adults living without adult children (particularly those living alone and those living with others)
exhibit significantly higher levels of loneliness than those living with adult children. Moreover,
the positive relationships between loneliness and living arrangements without adult children
are more pronounced in prefectures with a higher family genealogy density. In other words, the
differences in the reported levels of loneliness between older Chinese living without adult
children and those living with adult children are higher in prefectures with stronger traditional
family cultures (support H2). Lastly, the levels of loneliness of older people coresiding with
adult children are relatively stable despite the regional variation in family culture (consistent
with H2a).

19
5.3. Heterogeneity analyses
In this section, we investigate the heterogeneous patterns of the interactions between
loneliness, living arrangements, and family norms by gender, age, and residence type. The
results are presented in Table 3. Columns (1) – (2), (3) – (4), and (5) – (6) report the subsample
analysis by gender (female vs. male), age (age 80 or below vs. above age 80), and residence
type (rural vs. urban), respectively. For each subsample analysis, the first column presents
results using the dummy indicator of loneliness as the outcome, and the second column presents
results based on the five-level measure.
Across different subsamples, the main coefficients of living arrangements consistently
show robust patterns that older people living alone or living with others feel lonely more
frequently than those living with adult children. The differences in loneliness between older
people living with a spouse only and those living with adult children are small and statistically
insignificant. The only exception is the subgroup of urban residents, among whom older people
living with a spouse only also feel lonely more frequently than those living with adult children.
As for the role of culture, the interaction terms between genealogy density and living alone
are significant among females and older adults aged 80 or below; the interaction terms between
genealogy density and living with spouse only are significant for both gender groups, those
who are aged 80 or below, and rural residents. On the other hand, there is not much variation
in the linkages between loneliness and living alone (or living with a spouse only) across
prefectures with varying family cultures among respondents above 80 years old and living in
urban regions. We also observe positive and significant interaction terms between genealogy
density and living with others among men, people aged above 80, and rural residents. These
results together suggest that the heterogeneous associations between living arrangements and
loneliness along family culture are mainly driven by women, older people aged 80 or below,
and those living in rural areas. We discuss these results in more detail in Section 6 below.
5.4. Robustness checks
This section reports a set of robustness checks to bolster our baseline results. First, as

20
compiling family genealogies is a cultural practice among the Han Chinese, the genealogy
measure should not shape the relationship between living arrangements and loneliness among
ethnic minorities. Table 4 reports findings of a battery of placebo tests: (a) Columns (1) and (2)
present regression results among respondents who are ethnic minorities (i.e., non-Han); (b)
Columns (3) and (4) present regression results restricted to those from prefectures or counties
that are ethnic autonomous regions; (c) Columns (5) and (6) report results further restricting
the sample to respondents who are ethnic minorities and living in ethnic autonomous regions.
For each set of analyses, the first column presents results using the dummy loneliness indicator
as the dependent variable, and the second column shows results using the five-level loneliness
measure. As expected, the majority of interaction terms between living arrangements and
family genealogy density are statistically insignificant across all the placebo tests.
Next, we conduct a series of tests on the potential bias related to sample construction. First,
we re-estimated the regressions using an extended CLHLS 2002 – 2014 sample. By tracing
respondents who were successfully followed in the survey and did not move, we can impute
their residential prefectures information back to the CLHLS 2002 wave. However, these
samples may suffer from selection problems since only the respondents in waves 2002 and
2005 who were re-interviewed in later waves are included in the analyses. Column (2) in Table
A2 reports the estimates. Utilizing the longer period of observations does not change the sign
and the statistical significance of the main coefficients compared to benchmark estimates in
Column (1).
Second, in the baseline analyses, we assume the CLHLS prefectures that fail to match
with the genealogy data do not have any genealogy records. In Column (3) in Table A2, we
examine the robustness of the results by excluding respondents living in these unmatched
prefectures. As shown, the results are similar to the benchmark estimates.
Another concern is that our estimates may suffer from selection bias due to different
mortality rates across age groups, which might correlate with one’s feeling of loneliness. To
account for potential survival bias, we additionally control for the wave by age-group fixed
effects. Specifically, we define an age group every five years starting from age 65, with the last

21
group as “Aged 95 or above.” The wave by age-group fixed effects are constructed by
interacting indicators of survey waves and age-group dummies. Column (4) in Table A2 shows
that results from such an exercise are very similar to the benchmark estimates.
We also conduct a series of robustness checks regarding omitted variable bias in Table A3.
First, as the measure of family culture is historical genealogy records, our estimation may be
vulnerable to omitted variable bias due to other historical factors that simultaneously affect
clan culture and contemporary people’s feelings of loneliness or living arrangements. To
address such a concern, we additionally control for a set of historical variables, including the
logarithm form of the distance to the nearest commercial center, silk center, and tea center in
the Ming-Qing dynasties and the distance to the nearest large city in 1920 (see Appendix A2
for more details), as well as their interactions with various living arrangements. As shown in
Column (2) of Table A3, after accounting for these historical factors, the results are robust
compared to the benchmark estimates presented in Column (1).
Another set of confounding factors is the support that older people receive from other
sources beyond immediate family support. To account for these sources of support, we further
control for the indicators of obtaining help from the extended family when respondents get sick,
talking with people from the extended family, and receiving support from the community, along
with their interactions with the family genealogy density. Column (3) of Table A3 shows that
the results remain robust compared to the benchmark estimates.
Similarly, contemporary macro-level factors, such as regional economic development
other than GDP per capita, may confound the previously observed relationships. To mitigate
the concern, we replace the original province fixed effect and the survey wave fixed effect with
the province-by-wave fixed effect and add city-specific time trends to the regressions. Column
(4) of Table A3 shows that the main results are robust to the alternative specifications.
Finally, we check the robustness of the functional form specification of the regression
models. Specifically, we use a probit model for the dummy loneliness indicator and an ordered-
probit model for the five-level loneliness measure. As shown in Table A4, we obtain consistent
findings using these alternative estimation methods.

22
To sum up, all the robustness checks corroborate our benchmark results, underscoring the
significance of family norms in shaping the associations between living arrangements and
loneliness, particularly the feelings of loneliness among older adults living under arrangements
that deviate from the culturally prescribed type.
6. Discussion
Capitalizing on nationally representative survey data of older Chinese and a dataset of
historical records of family genealogies, this study examines how family culture (proxied by
genealogy density) on the prefecture level shapes the disparities in older adults’ loneliness by
their family living arrangements. In particular, we hypothesize that a “cultural concordance”
or a “cultural discordance” scenario between one’s actual living arrangement and the
collectively valued situation may expose individuals to different levels of loneliness, and the
loneliness gaps by living arrangements will enlarge as the family norm strengthens.
Our main findings show that, in contemporary China where the strength of traditional
family culture valuing family cohesion and intergenerational coresidence is still strong in most
areas, older Chinese who do not live with adult children (cultural discordance groups), such as
those living alone or living with others, report significantly higher levels of loneliness relative
to those living together with adult children (cultural concordance group). Moreover, the
positive relationship between living alone and loneliness is more pronounced in prefectures
with stronger family culture (measured by a higher density of family genealogy). The
interaction term between living with a spouse only and genealogy density is also positive in
shaping loneliness. On the other hand, loneliness reported by older Chinese co-residing with
adult children does not change much as the family culture varies. These findings are consistent
with our theoretical hypotheses and remain robust after several sensitivity checks on possible
bias due to sample selection, omitted variables, or functional form specification.
Our main findings shed light on the important but relatively neglected cultural factor on
the aggregate level in shaping individuals’ experiences of loneliness. More specifically, the
demonstrated moderating role of family culture reflects a cognitive discrepancy process in the

23
rise of loneliness: When the culturally prescribed living arrangement is coresidence with adult
children, older people living with children in reality likely perceive their actual living
arrangement as meeting the culturally desired modal, for whom the culture has little correlation
with feelings of loneliness. Conversely, older people whose living arrangements do not align
with cultural modals, such as those living alone and living with a spouse only, are vulnerable
to heightened levels of loneliness, and the feelings of loneliness are more pronounced in
prefectures with a more traditional family culture. These patterns suggest that the perceived
deviation from the cultural ideal of living arrangements is crucial in shaping loneliness,
especially in regions with entrenched family norms.
We also detect interesting patterns of heterogeneity by respondents’ gender, age group, and
residence area in the interrelationships between family culture, living arrangements, and
loneliness. First, both older men and women are sensitive to the local family culture regarding
how their feelings of loneliness relate to living arrangements. Nevertheless, it appears that older
women’s loneliness is vulnerable to family culture when they live alone or live with a spouse
only, whereas older men’s loneliness is sensitive to family culture when they live with a spouse
only and live with others. Unlike prior research on other East Asian societies (Lee & Yeung,
2021), we do not find evidence supporting a centered role of family circumstances in shaping
older women’s well-being only; instead, unmet expectations regarding desired living
arrangements in influencing loneliness are important for both gender groups.
Second, the observed patterns regarding the role of family norms are primarily a
phenomenon within the subsample of respondents aged 80 or below. In contrast, almost no
significant interactions were found among the much older subsample aged above 80, despite
their reported level of loneliness being significantly higher than that among aged 80 or below.12
Our finding is consistent with earlier studies that found the prevalence of loneliness increases
at advanced ages like 80 years and above (Dykstra, 2009), but reveals a reversed age pattern in
how older adults’ feelings of loneliness might change as a function of their living arrangements
12 In our analytical sample, the average level of loneliness among people older than 80 is 0.36, 0.12 points
larger than that of older adults aged 80 and below (a difference significant at 1% level).

24
and the societal family culture.
Lastly, we find evidence supporting the moderating role of family culture mainly among
rural residents but not among urban subsamples. Consistent with earlier studies on China
reporting rural-urban differences in older adults’ social networks related to culturally preferred
family structure and living arrangements (Zhao et al., 2022), our study demonstrates that the
role of traditional family norms is stronger in less modernized rural villages than in urban cities.
This study has several limitations that future studies could better address with appropriate
data and measurement. First, our loneliness measure is a simple global one that cannot capture
multi-dimensional variation in feelings of loneliness and is also subject to recall bias. If many
older adults are vulnerable to declining cognitive function and fading memory, the recall bias
in reporting loneliness might bias the observed patterns in our study, particularly among the
oldest old (Abedini et al., 2020; Cai, 2022). Second, we do not have detailed information on
other characteristics of older people’s living environment, such as the geographic proximity
and living distance between the respondent and their adult children and the amount and
frequency of support the respondent receives from non-coresident children. These omitted
variables could affect respondents’ living arrangement decisions and subjective feelings of
loneliness, thus biasing our findings. Third, we cannot draw a causal relationship between
living arrangements and loneliness in our study. One concern is the reverse causality if older
adults choose to live with adult children when they feel lonely. This possibility indicates a
positive association between living with adult children and loneliness, which could be more
salient (i.e., more positive) in regions with a stronger valuation of cultural obligations to
support older parents. However, this scenario is unlikely to be true in our study. Supplemental
analyses in Table A5 examine the relationship between one-wave lagged loneliness and living
arrangements. The results show no significant relationships between the lagged loneliness and
older adults’ living arrangements, which help mitigate the concern of reverse causality.
We acknowledge the existence of other omitted factors that may still confound the
relationship between living arrangements and loneliness, although we have conducted a wide
range of robustness checks. For instance, Sun et al. (2019) show that childcare for

25
grandchildren and economic gains from labor division are important driving forces of
intergenerational coresidence. Unfortunately, we cannot account for these factors in our
analyses due to data limitations. However, offering childcare is more common among the
middle-aged and the youngest-old Chinese adults (e.g., 65 years or younger), whereas a large
proportion of the CLHLS respondents are well beyond the age range. Thus, the potential
confounding from grandchild care and labor division should be less of a concern for the older
Chinese examined in this study.
Despite the limitations mentioned above, this study makes several contributions. First, our
research adds fresh evidence to the literature focusing on society-level cultural factors as
deeper-root determinants of people’s perceptions of their life circumstances, health status, and
feelings of loneliness (Beller & Wagner, 2020; De Jong Gierveld & Tesch-Römer, 2012; De
Jong Gierveld et al., 2012; Holt-Lunstad, 2018; Jylhä & Jokela 1990; Takagi et al., 2007).
Second, our findings highlight one crucial mechanism via which cultural norms would matter
for loneliness: the cognitive discrepancy process that emphasizes the difference between the
actual and collectively desired living conditions, which is shown to be more salient in regions
with more powerful cultural schemes. Third, our study is situated in contemporary China
featured by profound transformations in its sociodemographic landscape and the continuity of
traditional family culture, offering more analytical leverage in empirically testing the role of
cultural norms in influencing individuals’ loneliness (and presumably many other aspects of
psychological well-being) compared to other societies. Our analyses provide one helpful way
to understand the increasing prevalence of loneliness among older adults from a cultural
perspective regarding the relationship between living arrangements and loneliness. The
findings from China are likely to be generalized to other societies with similar cultural contexts
featuring strong family ties and intergenerational relationships and also experience rapid
sociodemographic transformations.

26
References
Abedini, Nauzley C., HwaJung Choi, Melissa Y. Wei, Kenneth M. Langa, and Vineet
Chopra. 2020. “The Relationship of Loneliness to End-of-Life Experience in Older
Americans: A Cohort Study.” Journal of the American Geriatrics Society, 68(5), 1064-
1071.
Alesina, Alberto, and Paola Giuliano. 2011. “Family Ties and Political Participation.”
Journal of the European Economic Association, 9(5), 817-839.
Bachrach, Christine A. 2014. “Culture and Demography: From Reluctant Bedfellows To
Committed Partners.” Demography, 51(1), 3-25.
Baker, Hugh D.R. 1979. Chinese Family and Kinship. Columbia University Press.
Banerjee, Abhijit, Esther Duflo, Erin Grela, Madeline McKelway, Frank Schilbach,
Garima Sharma, and Girija Vaidyanathan. 2023. “Depression and Loneliness among
the Elderly in Low-and Middle-Income Countries.” Journal of Economic Perspectives,
37(2), 179-202.
Beller, Johannes, and Adina Wagner. 2020. “Loneliness and Health: The Moderating Effect
of Cross-Cultural Individualism/Collectivism.” Journal of Aging and Health, 32(10),
1516-1527.
Cacioppo, John T., and Stephanie Cacioppo. 2018. “The Growing Problem of Loneliness.”
The Lancet, 391(10119): 426.
Cacioppo, John T., Mary Elizabeth Hughes, Linda J. Waite, Louise C. Hawkley, and
Ronald A. Thisted. 2006. “Loneliness as a Specific Risk Factor for Depressive
Symptoms: Cross-Sectional and Longitudinal Analyses.” Psychology and Aging, 21(1),
140-151.
Cai, Shu. 2022. “Does Social Participation Improve Cognitive Abilities of the Elderly?”
Journal of Population Economics, 35, 591-619.
Cai, Shu, Yuyang Jiang, Lingwei Wu. 2023. “Risk Behaviors of Only-child Parents:
Evidence from the One-child Policy in China.” SSRN working paper.

27
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4417673
Carr, Deborah, and Rebecca L. Utz. 2020. “Families in Later Life: A Decade in
Review.” Journal of Marriage and Family, 82(1), 346-363.
Chen, Taichang. 2019. “Living Arrangement Preferences and Realities for Elderly Chinese:
Implications for Subjective Well-being.” Ageing & Society, 39(8), 1557-1581.
Chen, Feinian, and Susan E. Short. 2008. “Household Context and Subjective Well-Being
Among The Oldest Old in China.” Journal of Family Issues, 29(10), 1379-1403.
Chen, Ting, James Kai-sing Kung, and Chicheng Ma. 2020. “Long Live Keju! The
Persistent Effects of China’s Civil Examination System.” The Economic Journal,
130(631), 2030-2064.
Cheung, Adam Ka-Lok, and Wei-Jun Jean Yeung. 2015. “Temporal-Spatial Patterns of
One-Person Households in China, 1982–2005.” Demographic Research, 32, 1209-1238.
de Jong Gierveld, Jenny, Pearl A. Dykstra, and Niels Schenk. 2012. “Living Arrangements,
Intergenerational Support Types and Older Adult Loneliness in Eastern and Western
Europe.” Demographic Research, 27, 167-200.
de Jong Gierveld, Jenny, and Clemens Tesch-Römer. 2012. “Loneliness in Old Age in
Eastern and Western European Societies: Theoretical Perspectives.” European Journal of
Ageing, 9, 285-295.
de Jong Gierveld, Jenny, and Theo Van Tilburg. 1999. “Living Arrangements of Older
Adults in the Netherlands and Italy: Coresidence Values and Behaviour and Their
Consequences for Loneliness.” Journal of Cross-Cultural Gerontology, 14, 1-24.
Dykstra, Pearl A. 2009. “Older Adult Loneliness: Myths and Realities.” European Journal of
Ageing, 6, 91-100.
Furuya, Shiro, and Jia Wang. 2023. “The Long Shadow: Early-Life Adversity and Later-Life
Loneliness in the United States.” The Journals of Gerontology Series B: Psychological
Sciences and Social Sciences, 78(2), 370-382.
Gao, Yufei. 2014. Lineage: Clan Genealogy and Family Culture (Xueyuan Maixi: Jiazu Zupu
Yu Jiating Wenhua) [in Chinese]. Modern Press Co., Ltd.

28
Giuliano, Paola. 2007. “Living Arrangements in Western Europe: Does Cultural Origin
Matter?” Journal of the European Economic Association, 5(5), 927-952.
Greenfield, Emily A., and David Russell. 2011. “Identifying Living Arrangements That
Heighten Risk for Loneliness in Later Life: Evidence from the US National Social Life,
Health, and Aging Project.” Journal of Applied Gerontology, 30(4), 524-534.
Hawkley, Louise C., Mary Elizabeth Hughes, Linda J. Waite, Christopher M. Masi,
Ronald A. Thisted, and John T. Cacioppo. 2008. “From Social Structural Factors to
Perceptions of Relationship Quality and Loneliness: The Chicago Health, Aging, and
Social Relations Study.” The Journals of Gerontology Series B: Psychological Sciences
and Social Sciences, 63(6), S375-S384.
Henning-Smith, Carrie. 2016. “Quality of Life and Psychological Distress among Older
Adults: The Role of Living Arrangements.” Journal of Applied Gerontology, 35(1), 39-
61.
Holt-Lunstad, Julianne. 2018. “Why Social Relationships Are Important for Physical Health:
A Systems Approach to Understanding and Modifying Risk and Protection.” Annual
Review of Psychology, 69, 437-458.
Holt-Lunstad, Julianne, Timothy B. Smith, Mark Baker, Tyler Harris, and David
Stephenson. 2015. “Loneliness and Social Isolation as Risk Factors for Mortality: A
Meta-Analytic Review.” Perspectives on Psychological Science, 10(2), 227-237.
Hu, Zhan, and Xizhe Peng. 2015. “Household Changes in Contemporary China: An Analysis
Based on the Four Recent Censuses.” The Journal of Chinese Sociology, 2, 1-20.
Hughes, Mary Elizabeth, and Linda J. Waite. 2002. “Health in Household Context: Living
Arrangements and Health in Late Middle Age.” Journal of Health and Social Behavior,
43(1), 1-21.
Jylhä, Marja, and Jukka Jokela. 1990. “Individual Experiences as Cultural – A Cross-
cultural Study on Loneliness among the Elderly.” Ageing & Society, 10(3), 295-315.
Lai, Daniel W.L. 2005. “Cultural Factors and Preferred Living Arrangement of Aging Chinese
Canadians.” Journal of Housing for the Elderly, 19(2), 71-86.

29
Lee, Yeonjin, and Wei-Jun Jean Yeung. 2021. “The Country that Never Retires: The
Gendered Pathways to Retirement in South Korea.” The Journals of Gerontology Series
B: Psychological Sciences and Social Sciences, 76(3), 642-655.
Lei, Xiaoyan, John Strauss, Meng Tian, and Yaohui Zhao. 2015. “Living Arrangements of
the Elderly in China: Evidence from the CHARLS National Baseline.” China Economic
Journal, 8(3), 191-214.
Li, Lydia W., Jiaan Zhang, and Jersey Liang. 2009. “Health among the Oldest-Old in China:
Which Living Arrangements Make a Difference?” Social Science & Medicine, 68(2), 220-
227.
Li, Ting, Wenting Fan, and Jian Song. 2020. “The Household Structure Transition in China:
1982–2015.” Demography, 57(4), 1369-1391.
Li, Ting, Yang Claire Yang, and Yanlong Zhang. 2018. “Culture, Economic Development,
Social-Network Type, and Mortality: Evidence from Chinese Older Adults.” Social
Science & Medicine, 204, 23-30.
Lin, Ju‐Ping, and Chin Chun Yi.‐ 2011. “Filial Norms and Intergenerational Support to
Aging Parents in China and Taiwan.” International Journal of Social Welfare 20: S109-
S120.
Lin, Xiaoping, Christina Bryant, Jennifer Boldero, and Briony Dow. 2015. “Older Chinese
Immigrants’ Relationships with Their Children: A Literature Review from a Solidarity–
Conflict Perspective.” The Gerontologist, 55(6), 990-1005.
Logan, John R., and Fuqin Bian. 1999. “Family Values and Coresidence with Married
Children in Urban China.” Social Forces, 77(4), 1253-1282.
Lu, Jiehua, and Qin Liu. 2019. “Four Decades of Studies on Population Aging in China.”
China Population and Development Studies, 3, 24-36.
Luhmann, Maike, Susanne Buecker, and Marilena Rüsberg. 2023. “Loneliness across
Time and Space.” Nature Reviews Psychology, 2(1), 9-23.
Luo, Ye, Louise C. Hawkley, Linda J. Waite, and John T. Cacioppo. 2012. “Loneliness,
Health, and Mortality in Old Age: A National Longitudinal Study.” Social Science &

30
Medicine, 74(6), 907-914.
Lykes, Valerie A, and Markus Kemmelmeier. 2014. “What Predicts Loneliness? Cultural
Difference between Individualistic and Collectivistic Societies in Europe.” Journal of
Cross-Cultural Psychology, 45(3), 468-490.
Mckeown, Brontë, Giulia L. Poerio, Will H. Strawson, Léa M. Martinon, Leigh M. Riby,
Elizabeth Jefferies, Cade McCall, and Jonathan Smallwood. 2021. “The Impact of
Social Isolation and Changes in Work Patterns on Ongoing Thought During the First
COVID-19 Lockdown in the United Kingdom.” Proceedings of the National Academy of
Sciences, 118(40), e2102565118.
Newmyer, Lauren, Ashton M. Verdery, Haowei Wang, and Rachel Margolis. 2022.
“Population Aging, Demographic Metabolism, and the Rising Tide of Late Middle Age to
Older Adult Loneliness Around the World.” Population and Development Review, 48(3),
829-862.
Niu, Lu, Cunxian Jia, Zhenyu Ma, Guojun Wang, Bin Sun, Dexing Zhang, and Liang
Zhou. 2020. “Loneliness, Hopelessness and Suicide in Later Life: A Case–Control
Psychological Autopsy Study in Rural China.” Epidemiology and Psychiatric Sciences,
29, e119.
Park, Denise C., Richard Nisbett, and Trey Hedden. 1999. “Aging, Culture, and
Cognition.” The Journals of Gerontology Series B: Psychological Sciences and Social
Sciences, 54(2), 75-84.
Peplau, Letitia A., and Daniel Perlman. 1982. “Perspectives on Loneliness.” Loneliness: A
Sourcebook of Current Theory, Research and Therapy, 1-18.
Raymo, James M., and Jia Wang. 2022. “Loneliness at Older Ages in the United States:
Lonely Life Expectancy and the Role of Loneliness in Health Disparities.” Demography,
59(3), 921-947.
Rebechi, Alessio, Anthony Lepintuer, Andrew E. Clark, Nicholas Rohde, Claus Vögele,
and Conchita D’ambrosio. 2024. “Loneliness during the COVID-19 Pandemic:
Evidence from Five European Countries.” Economics & Human Biology, 55: 101427.

31
Russell, David. 2009. “Living Arrangements, Social Integration, and Loneliness in Later Life:
The Case of Physical Disability.” Journal of Health and Social Behavior, 50(4), 460-475.
Shiue, Carol H. 2016. “A Culture of Kinship: Chinese Genealogies as a Source for Research
in Demographic Economics.” Journal of Demographic Economics, 82(4), 459-482.
Sun, Ang, Chuanchuan Zhang, and Xiangting Hu. 2019. “Boys, Girls, and Grandparents:
The Impact of the Sex of Preschool-aged Children on Family Living Arrangements and
Maternal Labor Supply.” Demography, 56(3), 813-833.
Surkalim, Daniel L., Mengyun Luo, Robert Eres, Klaus Gebel, Joseph van Buskirk,
Adrian Bauman, and Ding Ding. 2022. “The Prevalence of Loneliness across 113
Countries: Systematic Review and Meta-Analysis.” BMJ, 376, e067068.
Swader, Christopher S. 2019. “Loneliness in Europe: Personal and Societal Individualism-
Collectivism and Their Connection to Social Isolation.” Social Forces, 97(3), 1307-1336.
Takagi, Emiko, Merril Silverstein, and Eileen Crimmins. 2007. “Intergenerational
Coresidence of Older Adults in Japan: Conditions for Cultural Plasticity.” The Journals
of Gerontology Series B: Psychological Sciences and Social Sciences, 62(5): S330-S339.
Tang, Zongli. 2017. “Changes with Chinese Characteristics: Rural Clan Culture, Clan
Communities, and Kinship Relations During Urbanization.” China’s Urbanization and
Socioeconomic Impact, 161-198.
Victor, Christina, Sasha Scambler, John Bond, and Ann Bowling. 2000. “Being Alone in
Later Life: Loneliness, Social Isolation and Living Alone.” Reviews in Clinical
Gerontology, 10(4), 407-417.
Wilmoth, Janet M. 2001. “Living Arrangements among Older Immigrants in the United
States.” The Gerontologist, 41(2), 228-238.
Wilson, Robert S, Kristin R Krueger, Steven E Arnold, Julie A Schneider, Jeremiah F
Kelly, Lisa L Barnes, Yuxiao Tang, and David A Bennett. 2007. “Loneliness and Risk
of Alzheimer Disease.” Archives of General Psychiatry, 64(2), 234-240.
Xie, Jian, and Song Yuan. 2021. “The Cultural Origins of Family Firms.” Working Paper.
Yan, Zhimin, Xun Yang, Lei Wang, Yuhan Zhao, and Lin Yu. 2014. “Social Change and

32
Birth Cohort Increase in Loneliness among Chinese Older Adults: A Cross-Temporal
Meta-Analysis, 1995–2011.” International Psychogeriatrics, 26(11), 1773-1781.
Yoo, Jeong Ah, and Allison Zippay. 2012. “Social Networks among Lower Income Korean
Elderly Immigrants in the US.” Journal of Aging Studies, 26(3), 368-376.
Zhang, Chuanchuan. 2020. “Clans, Entrepreneurship, and Development of the Private Sector
in China.” Journal of Comparative Economics, 48(1), 100-123.
Zhang, Ning Jackie, Man Guo, and Xiaoying Zheng. 2012. “China: Awakening Giant
Developing Solutions to Population Aging.” The Gerontologist, 52(5), 589-596.
Zhao, Menghan, Fan Yang, and Youlang Zhang. 2022. “The Power of Culture: The
Gendered Impact of Family Structures and Living Arrangements on Social Networks of
Chinese Older Adults.” Ageing & Society, 42(3), 657-680.

33
Figures and Tables
Figure 1. Conceptual framework
Source: Own construction.
Notes: The figure depicts monotonically increasing loneliness disparities between older adults experiencing “cultural
discordance” (the upper solid line) versus “cultural concordance” (the lower solid ine) in family living arrangements across
regions with relatively weak-, middle-, and strong-level of family culture.
Strength of family culture
Level
of
loneliness
Living without
adult
children
(cultur
al
discordance)
Living with
adult
children
(cultur
al
concordance)
Weak
Difference 1
The loneliness difference
enlarges as fa mily culture
becomes stronger
Low
Strong
Middle
High
Difference
2
Difference 3

34
Figure 2. Geographic distribution of genealogy density in the CLHLS prefectures
Source: Own construction based on family genealogy data from the General Catalog of Chinese Genealogy, the Population
Census 2000, and CLHLS.
Notes: The figure illustrates the logarithm form of the number of genealogy per capita in the sampled CLHLS prefectures. The
prefectures with grey color are not in the CLHLS sample.

35
Figure 3. Predicted level of loneliness by living arrangements across prefectures with
different family culture
Source: Own construction based on regression results reported in Column (4) of Table 2.
Notes: The figure illustrates the predicted level of loneliness by living arrangements across prefectures with different family
culture which is measured by the logarithm form of the number of genealogy per capita in the prefecture substracting the mean
values among all the prefectures sampled in the CLHLS.

36
Table 1. Summary statistics
M ean S.D. M in. M ax.
(1) (2) (3) (4)
Feel lonely (dummy) 0.32 0.47 0.00 1.00
Feel lonely (5-level) 2.05 1.01 1.00 5.00
Living alone 0.19 0.39 0.00 1.00
Living with the spouse only 0.24 0.43 0.00 1.00
Living with adult children (with or without the sp ouse) 0.56 0.50 0.00 1.00
Living with others 0.01 0.12 0.00 1.00
Ln (number of genealogies per million p eople) 0.08 1.48 -2.23 4.33
M ale 0.46 0.50 0.00 1.00
Age 85.41 10.77 65.00 116.00
M arried 0.38 0.48 0.00 1.00
Illiterate 0.58 0.49 0.00 1.00
Primary school 0.31 0.46 0.00 1.00
M iddle school and above 0.11 0.32 0.00 1.00
Farmer 0.65 0.48 0.00 1.00
Worker 0.13 0.34 0.00 1.00
M anager 0.08 0.27 0.00 1.00
Others 0.13 0.34 0.00 1.00
Urban resident 0.47 0.50 0.00 1.00
Variable
Source: CLHLS 2008-2014.
Notes: The table presents the summary statistics for the sample of respondents with Han ethnicity. The number of observations
is 24,558.

37
Table 2. OLS regressions of loneliness on living arrangements and family culture
among older Chinese
(1) (2) (3) (4)
Living alone 0.150*** 0.146*** 0.339*** 0.333***
(0.018) (0.018) (0.039) (0.040)
Living with the spouse only 0.013 0.007 0.046 0.038
(0.016) (0.017) (0.035) (0.036)
Living with others 0.152* 0.154* 0.264** 0.268**
(0.091) (0.090) (0.132) (0.131)
Ln (number of genealogies per million peop le) -0.001 -0.002
(0.008) (0.017)
Living alone 0.018* 0.031
× Ln (number of genealogies per million people) (0.010) (0.021)
Living with the spouse only 0.021*** 0.030*
× Ln (number of genealogies per million people) (0.007) (0.016)
Living with others 0.035 0.062
× Ln (number of genealogies per million people) (0.041) (0.070)
M ale 0.011 0.013 0 .017 0.019
(0.015) (0.015) (0.032) (0.032)
Age 0.007 0.006 0.023 0.022
(0.007) (0.007) (0.015) (0.015)
Age squared/100 -0.004 -0.004 -0.014 -0.013
(0.004) (0.004) (0.009) (0.009)
M arried -0.179*** -0.175*** -0.397*** -0.390***
(0.019) (0.019) (0.040) (0.041)
M arried -0.025*** -0.041**
× Ln (number of genealogies per million people) (0.008) (0.017)
Primary school -0.060*** -0.060*** -0.132*** -0.132***
(0.017) (0.017) (0.035) (0.035)
M iddle school and above -0.074*** -0.077*** -0.151** -0.156***
(0.025) (0.025) (0.060) (0.060)
Worker -0.015 -0.016 -0.099** -0.100**
(0.017) (0.017) (0.039) (0.039)
M anager -0.037* -0.037* -0. 128*** -0.128***
(0.019) (0.019) (0.046) (0.046)
Others -0.029 -0.029 -0.072 -0.072
(0.021) (0.021) (0.055) (0.055)
Urban resident -0.036*** -0.036*** -0.061** -0.062**
(0.011) (0.011) (0.026) (0.026)
The share of plain in the county 0.000 0.000 -0.000 -0.000
(0.000) (0.000) (0.001) (0.001)
M issing indicator: The share of plain in the county 0.002 0.001 0.009 0.008
(0.027) (0.026) (0.058) (0.058)
Ln (lagged GDP p er capita) -0.003 0.000 -0.025 -0.021
(0.010) (0.010) (0.024) (0.024)
M issing indicator: Ln (lagged GD P per capita) 0.025 0.028 -0.007 -0.001
(0.033) (0.033) (0.073) (0.073)
Constant 0.093 0.130 1.268** 1.327**
(0.315) (0.316) (0.644) (0.646)
Wave fixed effect Yes Yes Yes Yes
Province fixed effect Yes Yes Yes Yes
R-squared 0.12 0.12 0.13 0.13
Observations 24558 24558 24558 24558
Feel lonely (5-level)
Feel lonely (dummy)
Source: CLHLS 2008-2014.
Notes: Standard errors in parentheses are clustered by individual. *** p<0.01, ** p<0.05, * p<0.10.

38
Table 3. Heterogeneity analyses
Feel
lonely
(dummy)
Feel
lonely (5-
level)
Feel
lonely
(dummy)
Feel
lonely (5-
level)
Feel
lonely
(dummy)
Feel
lonely (5-
level)
(1) (2) (3) (4) (5) (6)
Subsample
Living alone 0.137*** 0.305*** 0.156*** 0.355*** 0.153*** 0.314***
(0.023) (0.051) (0.023) (0.050) (0.026) (0.055)
Living with the spouse only -0.016 0.008 0.005 0.039 -0.011 -0.023
(0.028) (0.057) (0.018) (0.039) (0.025) (0.051)
Living with others 0.044 0.078 0.159 0.266* 0.166* 0.213
(0.084) (0.176) (0.109) (0.157) (0.099) (0.144)
Ln (number of genealogies per million p eople) -0.012 -0.008 0.001 -0.001 -0.000 0.004
(0.009) (0.021) (0.010) (0.022) (0.010) (0.022)
Living alone × Ln (number of 0.022* 0.036 0.027** 0.051* 0.021 0.016
genealogies per million p eople) (0.013) (0.027) (0.013) (0.027) (0.013) (0.028)
Living with the spouse only × Ln (number of 0.030** 0.050* 0.025*** 0.037** 0.021** 0.028
genealogies per million p eople) (0.012) (0.028) (0.008) (0.018) (0.010) (0.022)
Living with others × Ln (number of 0.004 0.038 0.031 0.027 0.087* 0.119
genealogies per million p eople) (0.063) (0.116) (0.051) (0.088) (0.048) (0.076)
R-squared 0.09 0.10 0.13 0.13 0.12 0.12
Observations 13300 13300 8709 8709 13062 13062
Subsample
Living alone 0.147*** 0.341*** 0.110*** 0.248*** 0.144*** 0.360***
(0.029) (0.061) (0.017) (0.037) (0.023) (0.057)
Living with the spouse only 0.019 0.048 0.024 0.012 0.026 0.103**
(0.020) (0.045) (0.021) (0.046) (0.021) (0.045)
Living with others 0.205* 0.312** 0.109* 0.230* 0.148 0.307*
(0.111) (0.145) (0.057) (0.118) (0.100) (0.167)
Ln (number of genealogies per million p eople) 0.014 0.001 -0.019* -0.031 0.006 0.012
(0.014) (0.029) (0.010) (0.022) (0.011) (0.024)
Living alone × Ln (number of 0.012 0.026 -0.011 -0.031 0.010 0.036
genealogies per million p eople) (0.016) (0.036) (0.011) (0.025) (0.015) (0.031)
Living with the spouse only × Ln (number of 0.016* 0.018 -0.000 -0.016 0.014 0.015
genealogies per million p eople) (0.009) (0.021) (0.013) (0.033) (0.009) (0.023)
Living with others × Ln (number of 0.099** 0.142* 0.046 0.179* -0.058 -0.046
genealogies per million p eople) (0.041) (0.080) (0.042) (0.103) (0.061) (0.144)
R-squared 0.15 0.16 0.06 0.08 0.13 0.14
Observations 11258 11258 15849 15849 11496 11496
Female Age 80 or below Rural resident
Male Above age 80 Urban resident
Source: CLHLS 2008-2014.
Notes: The regression specification is the same as that in Columns (2) and (4) in Table 2. Standard errors in parentheses are
clustered by individual. *** p<0.01, ** p<0.05, * p<0.10.

39
Table 4. Placebo tests
Samp le
Feel
lonely
(dummy)
Feel
lonely (5-
level)
Feel
lonely
(dummy)
Feel
lonely (5-
level)
Feel
lonely
(dummy)
Feel
lonely (5-
level)
(1) (2) (3) (4) (5) (6)
Living alone 0.125*** 0.316*** 0.234*** 0.462*** 0.206** 0.423**
(0.039) (0.078) (0.089) (0.166) (0.092) (0.200)
Living with the spouse only 0.008 0.066 -0.021 -0.021 -0.001 -0.036
(0.031) (0.058) (0.068) (0.098) (0.073) (0.109)
Living with others -0.125 -0.178 -0.386*** -0.562*** -0.373*** -0.557***
(0.079) (0.152) (0.101) (0.145) (0.101) (0.149)
Ln (number of genealogies p er million peop le) -0.017 -0.092 -0.226 -0.878 -0.484 -0.629
(0.032) (0.062) (0.367) (0.759) (0.474) (1.109)
Living alone × Ln (number of -0.024 -0.033 0.040 0.068 0.079 0.117
genealogies per million p eople) (0.029) (0.077) (0.065) (0.150) (0.068) (0.197)
Living with the spouse only × Ln (number of 0.008 0.078* 0.005 0.052 -0.023 0.087
genealogies per million p eople) (0.020) (0.042) (0.041) (0.074) (0.051) (0.096)
Living with others × Ln (number of -0.018 -0.039
genealogies per million p eople) (0.069) (0.179)
R-squared 0.12 0.17 0.12 0.18 0.13 0.18
Observations 3869 3869 1073 1073 947 947
Ethnic minority Ethnic minority areas Ethnic minority in
ethnic minority areas
NA NA NA NA
Source: CLHLS 2008-2014.
Notes: The regression specification is the same as that in Columns (2) and (4) in Table 2. NA denotes estimates not available
due to lack of observation in the specific category. Standard errors in parentheses are clustered by individual. *** p<0.01, **
p<0.05, * p<0.10.

40
Appendices
A1. Validity of the genealogy measure
To check whether the genealogy measure is a valid proxy for the traditional family norm,
we examine the associations between the family genealogy density and a series of measures of
older adults’ attitudes. We first leverage a question in the CLHLS, “What kind of living
arrangement do you like best?” Options include (1) Living alone, whether children living
nearby is unimportant; (2) Living alone, children should live nearby; (3) Living with children;
(4) Living in nursing houses; and (5) No idea. Respondents choosing option (3) were classified
into 1- “Valuing living with children,” and those answering other options were collapsed into
0- “not valuing living with children.”.
We supplement our analysis by utilizing questions about subjective attitudes in the
CHARLS (see Appendix A2 for more details of the data). The CHARLS 2011 and 2013 asked
two questions about preferences for living arrangements. Respondents were first asked:
“Suppose an older person has a spouse and adult children and has a good relationship with
them; what do you think is the best living arrangement for the older person?” Options include
(1) Living with adult children; (2) Not living with them in the same house but living in the
same community or village; (3) Not living with them in the same house and the same
community or village; (4) Living in a nursing house; and (5) Others. We constructed a dummy
variable by coding option (1) as 1- “preferring living with adult children,” and 0 for all other
options. Respondents were then asked again about their living arrangement preferences in a
different scenario: “Suppose an older person has no spouse but has adult children and has a
good relationship with them. What do you think is the best living arrangement for the older
person?” Options as the same and we created a dummy variable in the same way with 1
indicating “preferring living with adult children” and 0 “otherwise.”
In addition, the CHARLS 2013, 2015, and 2018 asked respondents “Whom do you think
you can financially rely on when you are old?” Options include (1) children, (2) savings, (3)
pension or retirement salary, (4) commercial pension insurance, and (5) others. We created a

41
dummy variable with 1 indicating “believe to be supported by children in old age” (option 1)
and 0 indicating “otherwise.”
Table A1 presents the results by regressing people’s attitudes on the prefecture-level
genealogy density measure after controlling for a set of individual- and regional-level
covariates, survey wave fixed effects, and province fixed effects. Column (1) reports the results
on “Valuing living with children” using CLHLS data. Columns (2) to (4) report results on
preferences of living with children (two different scenarios with or without spouse) and the
belief of being supported by children in old age using CHARLS data.
Consistent with our expectations, a higher number of family genealogy books per capita
in a prefecture is indeed associated with local older residents’ higher probabilities of valuing
coresidence with adult children and believing in being supported financially by children in the
old age. These results bolster our confidence that the family genealogy density in a prefecture
is a valid measure for the traditional family norm that values coresidence with adult children
in China.

42
A2. Description of additional data
This appendix provides more information on datasets used in the supplement analysis: the
China Health and Retirement Longitudinal Study, the historical data, and the China City
Statistical Yearbooks data.
1. The China Health and Retirement Longitudinal Study
We supplement our analysis by utilizing questions about subjective attitudes on living
arrangements from the China Health and Retirement Longitudinal Study (CHARLS), a biennial
panel survey project that follows a nationally representative sample of the adult population
aged 45 and above in China since 2011. The 2011 and 2013 waves of CHARLS asked
respondents about their preferences for living arrangements and their perceptions of financial
support in their old life. We use these questions to examine the validity of our family genealogy
measure in representing the traditional family culture (see Appendix A1 for more details).
2. The historical data
In the supplemental analysis, we also include several distance measures representing the
historical socioeconomic background of the respondent’ region of residence, i.e., the logged
distances from the respondent’s current residence citiy to (1) the nearest commercial center
during the Ming-Qing dynasties, (2) a silk center during the Ming-Qing dynasties, (3) a tea
center during the Ming-Qing dynasties, and (4) the nearest large city in 1920. The data were
compiled and digitalized by Chen et al. (2020), based on historical information on cities’
economic activities from the provincial gazetteers.
3. The China City Statistical Yearbooks
In the robustness check, we further control for prefecture-level GDP per capita (lagged by
one year and in the logarithm form) to exclude socioeconomic confounding factors that may
simultaneously affect older adults’ loneliness and cultural value. We first obtained the GDP
data from the China City Statistical Yearbooks (edited by the National Bureau of Statistics of
China) and then matched them with respondents in the CLHLS living in different prefectures.

43
A3. Additional tables
Table A1. Validation of family genealogy density
CLHLS data
Valuing living
with children
Preferring living
with adult
children
(scenario 1)
Preferring living
with adult
children
(scenario 2)
Believe t o be
supported by
children in old
ages
(1) (2) (3) (4)
Ln (number of genealogies per million people) 0.017** 0.038*** 0.030** 0.018**
(0.008) (0.013) (0.012) (0.008)
Other control variables Yes Yes Yes Yes
Wave fixed effect Yes Yes Yes Yes
Province fixed effect Yes Yes Yes Yes
R-squared 0.10 0.08 0.05 0.22
Observations 24553 5274 5260 17323
CHARLS data
Source: The CLHLS 2008-2014 are used for the valuation on living with children in Column (1), the CHARLS 2011 and 2013
are used for the attitudes towards living arrangements in Columns (2) and (3), and the CHARLS 2013, 2015, and 2018 are
used for the attitude towards old-aged support in Column (4).
Notes: The other control variables in Column (1) include age, age squared, dummies of male, primary school education, middle
school education or above, occupation categories before age 60 (workers, managers, and others), urban resident, the share of
plain on the county level, the logarithm of one-year-lagged GDP per capita on the prefecture level, and indicators of missing
values for the county-level and prefecture-level variables. The other control variables in Columns (2) to (4) are the same as
those in Column (1) except that they do not include the dummies of occupational categories before age 60 and the indicator of
missing values for the county-level variable. Scenario 1 asked respondents, “Suppose an older person has a spouse and adult
children and has a good relationship with them; what do you think is the best living arrangement for the older person?” A
dummy variable with value 1 for Scenario 1 indicates “preferring living with adult children.” Scenario 2 asked “Suppose an
older person has no spouse but has adult children and has a good relationship with them. What do you think is the best living
arrangement for the older person?” A dummy variable with value 1 for Scenario 2 indicates “preferring living with adult
children.” Standard errors in parentheses are clustered by individual. *** p<0.01, ** p<0.05, * p<0.10.

44
Table A2. Robustness checks on sample selection
(1) (2) (3) (4)
Living alone 0.146*** 0.149*** 0.145*** 0.147***
(0.018) (0.013) (0.019) (0.018)
Living with t he spouse only 0.007 0.012 0.006 0.008
(0.017) (0.010) (0.018) (0.016)
Living with ot hers 0.154* 0.141* 0.152 0.154*
(0.090) (0.073) (0.094) (0.090)
Ln (number of genealogies p er million peop le) -0.001 0.003 -0.000 -0.001
(0.008) (0.006) (0.008) (0.008)
Living alone 0.018* 0.014* 0.019* 0.018*
× Ln (number of genealogies per million people) (0.010) (0.007) (0.011) (0.010)
Living with t he spouse only 0.021*** 0.015*** 0.023*** 0.021***
× Ln (number of genealogies per million people) (0.007) (0.005) (0.008) (0.007)
Living with ot hers 0.035 -0.006 0.039 0.034
× Ln (number of genealogies per million people) (0.041) (0.029) (0.043) (0.041)
Living alone 0.333*** 0.351*** 0.331*** 0.334***
(0.040) (0.028) (0.042) (0.040)
Living with t he spouse only 0.038 0.032 0.033 0.038
(0.036) (0.022) (0.037) (0.035)
Living with ot hers 0.268** 0.259** 0.254* 0.266**
(0.131) (0.113) (0.136) (0.131)
Ln (number of genealogies p er million peop le) -0.002 0.005 0.002 -0.001
(0.017) (0.013) (0.018) (0.017)
Living alone 0.031 0.014 0.033 0.031
× Ln (number of genealogies per million people) (0.021) (0.016) (0.023) (0.021)
Living with t he spouse only 0.030* 0.023** 0.035** 0.029*
× Ln (number of genealogies per million people) (0.016) (0.012) (0.018) (0.016)
Living with ot hers 0.062 0.006 0.088 0.061
× Ln (number of genealogies per million people) (0.070) (0.054) (0.068) (0.070)
Samp le CLHLS
2008-2014
CLHLS
2002-2014
Excl.
unmatched
CLHLS
2008-2014
Other control variables Yes Yes Yes Yes
Wave fixed effect Yes Yes Yes
Province fixed effect Yes Yes Yes Yes
Wave by age-group fixed effect Yes
Observations 24558 48165 23107 24558
Feel lonely (dummy)
Feel lonely (5-level)
Source: CLHLS 2002-2014.
Notes: Column (1) reports the benchmark results. Column (2) reports the estimates using an extended CLHLS 2002-2014
sample. Column (3) reports the estimates by excluding the subsample of respondents living in prefectures that failed to match
with the genealogy data. Column (4) further controls for the wave by age-group fixed effects in the baseline specification to
account for survival bias, which are constructed by interacting the indicator of survey waves and age-group dummies. All the
regressions also control for the same set of individual characteristics and regional characteristics as those in Columns (2) and
(4) in Table 2. Standard errors in parentheses are clustered by individual. *** p<0.01, ** p<0.05, * p<0.10.

45
Table A3. Robustness checks on omitted variables
(1) (2) (3) (4)
Living alone 0.146*** 0.155*** 0.143*** 0.149***
(0.018) (0.018) (0.019) (0.019)
Living with t he spouse only 0.007 0.009 0.006 0.007
(0.017) (0.017) (0.017) (0.017)
Living with ot hers 0.154* 0.090 0.114 0.152*
(0.090) (0.055) (0.082) (0.090)
Ln (number of genealogies p er million peop le) -0.001 -0.002 -0.000 0.003
(0.008) (0.008) (0.008) (0.015)
Living alone 0.018* 0.024** 0.018* 0.019*
× Ln (number of genealogies per million people) (0.010) (0.011) (0.010) (0.010)
Living with t he spouse only 0.021*** 0.027*** 0.021*** 0.019***
× Ln (number of genealogies per million people) (0.007) (0.008) (0.007) (0.007)
Living with ot hers 0.035 -0.057 0.034 0.034
× Ln (number of genealogies per million people) (0.041) (0.053) (0.040) (0.043)
Living alone 0.333*** 0.358*** 0.328*** 0.333***
(0.040) (0.040) (0.040) (0.040)
Living with t he spouse only 0.038 0.031 0.037 0.036
(0.036) (0.036) (0.036) (0.036)
Living with ot hers 0.268** 0.168* 0.213* 0.249*
(0.131) (0.090) (0.127) (0.131)
Ln (number of genealogies p er million peop le) -0.002 0.009 -0.009 0.002
(0.017) (0.018) (0.017) (0.032)
Living alone 0.031 0.028 0.029 0.032
× Ln (number of genealogies per million people) (0.021) (0.024) (0.021) (0.021)
Living with t he spouse only 0.030* 0.039** 0.030* 0.033**
× Ln (number of genealogies per million people) (0.016) (0.019) (0.017) (0.016)
Living with ot hers 0.062 -0.048 0.050 0.075
× Ln (number of genealogies per million people) (0.070) (0.089) (0.073) (0.078)
Other control variables Yes Yes Yes Yes
Wave fixed effect Yes Yes Yes
Province fixed effect Yes Yes Yes
Historical variables Yes
Supp ort from other sources Yes
Province-wave fixed effect Yes
City time trend Yes
Observations 24558 22400 24558 24558
Feel lonely (dummy)
Feel lonely (5-level)
Source: CLHLS 2008-2014.
Notes: Column (1) reports the benchmark results. Column (2) reports the estimates from regressions that further control for
the historical variables (including the log distance to the nearest commercial centers, silk centers, and tea centers in Ming-
Qing dynasties and the log distance to the nearest large cities in 1920) and their interaction with the dummies of living
arrangement. Column (3) reports the estimates from regressions by controlling for the support from other sources (including
indicators of “Obtain help from the extended family when being sick,” “Talk with people from the extended family,” and
“Access to support from the community”) and their interaction with the genealogy density measure. Column (4) reports the
estimates from regressions that replace the original province fixed effect and the survey wave fixed effect with the province-
by-wave fixed effect and the city-specific time trends. All the regressions also control for the same set of individual
characteristics and regional characteristics as those in Columns (2) and (4) in Table 2. Standard errors in parentheses are
clustered by individual. *** p<0.01, ** p<0.05, * p<0.10.

46
Table A4. (Ordered-) Probit regression estimates
Feel lonely (dummy) Feel lonely
(5-level)
(1) (2)
Living alone 0.406*** 0.351***
(0.050) (0.043)
Living with the spouse only 0.022 0.044
(0.067) (0.048)
Living with others 0.428* 0.318**
(0.231) (0.137)
Ln (number of genealogies per million people) -0.005 -0.004
(0.023) (0.019)
Living alone 0.060** 0.040*
× Ln (number of genealogies per million people) (0.027) (0.022)
Living with the spouse only 0.086*** 0.038
× Ln (number of genealogies per million people) (0.030) (0.023)
Living with others 0.093 0.065
× Ln (number of genealogies per million people) (0.111) (0.074)
Other control variables Yes Yes
Wave fixed effect Yes Yes
Province fixed effect Yes Yes
Pseudo R-squared 0.10 0.05
Observations 24558 24558
Source: CLHLS 2008-2014.
Notes: We use a probit model when the outcome is the dummy variable and an ordered-probit model for the five-level measure
to estimate the coefficients. The regressions also control for the same variables as those in Columns (2) and (4) in Table 2.
Standard errors in parentheses are clustered by individual. *** p<0.01, ** p<0.05, * p<0.10.

47
Table A5. Multinomial logit regression of living arrangement on lagged loneliness
(1) (2)
L1.Feel lonely (dummy) -0.030 -0.033
(0.133) (0.132)
L1.Living alone 2.501*** 2.503***
(0.171) (0.169)
L1.Living with t he spouse only 1.815*** 1.807***
(0.162) (0.162)
L1.Living with ot hers 1.951*** 1.958***
(0.382) (0.380)
Ln (number of genealogies p er million peop le) 0.029
(0.066)
L1.Feel lonely (dummy) 0.088 0.081
(0.189) (0.189)
L1.Living alone 0.809*** 0.810***
(0.267) (0.267)
L1.Living with t he spouse only 2.096*** 2.093***
(0.130) (0.130)
L1.Living with ot hers 2.261** 2.268**
(0.972) (0.971)
Ln (number of genealogies p er million peop le) -0.060
(0.301)
L1.Feel lonely (dummy) 0.184 0.184
(0.392) (0.393)
L1.Living alone 1.853*** 1.884***
(0.529) (0.517)
L1.Living with t he spouse only 1.472** 1.477**
(0.671) (0.676)
L1.Living with ot hers 4.885*** 4.898***
(0.662) (0.651)
Ln (number of genealogies p er million peop le) 0.231
(0.280)
Control variables Yes Yes
Wave fixed effect Yes Yes
Province fixed effect Yes Yes
Pseudo R-squared 0.42 0.42
Observations 10222 10222
Reference group : Live with adult children
Living arrangement
Living alone
Living with the spouse only
Living with others
Source: CLHLS 2008-2014.
Notes: The table reports the estimates from the multinomial logit regressions of living arrangements on one-wave lagged
loneliness. The reference group of the outcome variable is older people who live with adult children (with or without the
spouse). The other control variables in Column (1) include the dummies of one-wave lagged living arrangements, the same set
of covariants as those in Columns (2) and (4) in Table 2, and wave fixed effect and province fixed effect. Column (2) further
controls the log of the number of genealogies per million people. Standard errors in parentheses are clustered by individual.
*** p<0.01, ** p<0.05, * p<0.10.