Parental marriage transfer and old age
support: evidence from China
Ziwei Rao
∗August 11, 2018
Abstract
This study uses data from China to explore whether and how children repay
marriage gift by providing old age care and/or material transfer. Instrumental
help is found to be positively correlated with marriage transfers when parents have heavy limitation in activities of daily living (ADL) or experience heavy deterioration in ADL. A higher amount of monetary marriage gift is associated with a higher probability of living ideally close to parents. The incidence of material support is not affected by any marriage transfer in any empirical specification, suggesting a social norm nature of transferring money or in-kind. Receiving a house as a marriage gift surprisingly decreases the amount of material support.
JEL: D13, J13, J14
Keywords: marriage transfer, old age support, ADL
∗I am sincerely grateful to my supervisors, Prof Dr. Rob Alessie and Dr. Max Groneck, for their
1
Introduction
In the absence of formal care and a public social security system, a strong intergenerational
linkage sustains the family machine and ensures old age well-being. Parents provide
support, such as education and marriage transfers, to young children, and children take care of their parents when the parents are old. This pattern exists in both developed and developing regions, while it is more prevalent in the latter. Children’s education and marriage have been found to be the primary reasons of parental saving in China (Wei & Zhang, 2011). On the other hand, the elderly population in China rely heavily on money and time transfers from descendants. There are 49% of urban elderly parents and 73% of
rural parents in China receive either care or money from children (Lee & Xiao, 1998).1
Old age support from children is especially prevalent in China and other Southeast Asian countries also due to cultural reasons. The widely accepted Confucian filial piety refers to taking care of one’s parents and the patriarchal tradition emphasizes sons’ role as old age security. According to Sung (1998), filial piety requires children to provide material support and care to parents in a manner which conveys respect. The patriarchal tradition is well reflected by parents’ reliance on sons, for instance, parents without son(s) are more likely to enroll in China’s rural pension program (Ebenstein & Leung, 2010).
In recent years, there are concerns that the traditional intergenerational exchange pat-tern could be crucially altered by recent demographic changes and economic development (Cheung & Kwan, 2009). For instance, lower birth rates caused by one-child policy and the higher life expectancy jointly raise children’s burden to support parents. Sin (2005) shows that the old age dependency rate in China rises from 29.7% in 2001 to 37.2% in 2010, and is expected to continue increasing. The intergenerational linkage could be weakened by the introduction of pension and long-term care products. The decline of Confucian impact and its core values would further diminish the linkage (Cheung & Kwan, 2009).
To understand the intergenerational exchange in contemporary China, which is shaped by both traditions and recent social changes, this paper explores the question whether and how children pay back parental early life transfers with old age support. Ho (2017) provides the first evidence of the repayment by analyzing whether parental long-term investments, including investments in college education and marriage, stimulate instru-mental and material support from children. In her model, money support and care are two public goods of which provision is stimulated by parental investments through an income effect. With an emphasis on gender, she finds that higher investments in daugh-1The difference between rural and urban regions could be explained by the employment share in
ters significantly induce more instrumental support. Build on Ho (2017), I focus on the link between marriage transfers and old age support and explore the link under different circumstances.
I first estimate the correlation between marriage transfers and old age support with the differentiation between marital money transfer and marital house transfer. Instead of using the total amount of marriage transfers as Ho (2017), I consider the heterogeneity of marriage transfers. It is possible that only when parents have strong needs of support, marriage transfers stimulate old age support provision. Thus, I extend her paper by exploring whether the repayment differs by parental health status. Furthermore, parents’ health status deteriorate when they age, and children may provide increasing upward transfers in response to parents’ raising needs. However, the rise of transfers might differ from person to person. Therefore, another extension to Ho (2017) is that I explore the panel dimension of the data which allows me to study whether changes in the parental need of long-term care are associated with more help and whether this interaction is influenced by past marital gifts.
Family and child fixed effect models are employed for the analyses. The family fixed effect model rules out unobservable family effects which might cause marriage transfers to be endogenous. The child fixed effect model further differences out child factors such as innate ability and reciprocal personality, which would influence both parents’ transfer de-cision and children’s old age support. Same with Ho (2017), I use data from China Health and Retirement Longitudinal Study (CHARLS). Retrospective questions of marriage gifts provided to each child when the child get married, offers excellent opportunity to proxy early life transfers. The comprehensive old-age support received by respondents from each child, including monetary or in-kind transfers, instrumental help and living arrangement, enable the study of intergenerational support in different dimensions. Parental health is measured by their limitations in activities of daily living (ADL), which reflects their needs for support. Rich demographic information on respondents (as the parental generation) and all their children (as the child generation) makes it possible to conduct a child level analysis and observe the pattern between siblings.
and both kinds of marriage gift enlarge the impact when the parental ADL limitation level changes from light to heavy. The heterogeneous impacts of marriage transfers by parental health status are only found in providing instrumental help.
Overall, the findings suggest that the repayment of marriage transfers differ by gift type and parental status. Marital money gift and marital house gift are found to have differed associations with supports, which suggests that mixing these two types would be misleading. A possible explanation is that different types of marriage gift symbolize different responsibilities in providing old age support. The influence of marriage transfers on providing the most help to parental ADL seems to be status dependent. Only a heavy parental ADL limitation level induces a positive repayment of marriage transfers, and children repay more when the parental health deterioration is heavier. The incidence of material support is not affected by any marriage transfer in any empirical specification, suggesting a social norm nature of transferring money or in-kind.
There are several reasons for analyzing marriage transfers. Large parental transfers are often related with specific events, for instance, divorce, buying a house, unemployment, child birth and marriage. Marriage related downward transfers are common in developing countries (Anderson, 2007), and are especially relevant in China (Y. Yan, 2003). Brown (2009) observes that brideprice amounts to 82 percent of the value of household durables in rural China on average. Such a large transfer is representative of downward transfers, and a larger amount is more likely to induce responses of children in old age support. Apart from its large magnitude, marriage transfers might influence old age support due to its constitutional property. Marriage symbolizes the formation of a new household, and old age support would become a joint decision of the new household. Marriage transfers not only influence the marriage market matching, but also affect the bargain between spouses (Brown, 2009).
Section 2 discusses the conceptual framework, including relevant theories and hy-potheses. Section 3 presents data and descriptive statistics, which gives an overview of the transfers. Empirical models and results are presented in Section 4. I conclude the discussion in Section 5.
2
Literature and Main hypotheses
2.1
Models for intergenerational family transfers
Economists have developed three main streams of theories on family transfers, namely
altruism model, exchange model and mutuality model (Laferr`ere & Wolff, 2006).
Chinese context in the following subsection.
Altruism model
Becker (1974, 1991) proposes a classic model where an altruistic family head optimally allocates resources to children. This family head controls family wealth, has the power to dominate the allocation and cares about the welfare of all family members, including herself. Within such a model, even selfish children are incentivized to maximize family’s total welfare since their post-transfer consumption solely depends on the total family
income2. A prediction of the model is that the amount of parental transfer is negatively
correlated with the child’s income3. If there is only one child, a lower transfer is needed
for the recipient to achieve the optimal level of consumption when her income increases. When there are multiple children, the parent is equally altruistic towards all children, implying a lower transfer to the better off child with the aim to achieve an equal
post-transfer child income. This is the key of many empirical tests on parental altruism4. In
the corporate group, children provide supports to parents when parents reach their old age and desire support. Children provide old age support since their preference has been manipulated and guilt has been introduced by early life investments (Becker, 1993).
Cox (1987) and Cox and Rank (1992) modify Becker’s one-side altruism model to a mutual altruism model and incorporate service as a kind of upward transfers. Parents transfer income to children, while children provide service, which is more likely to be attention or companionship and has no clear market substitute. The parent dominates the decision of transfers as in Becker’s model, and her utility is determined by own
con-sumption, service received5 and the child’s utility. Since assuming that providing service
is costly to children, such transfers would only take place if they increase children’s utility when the parent is strongly altruistic. Cox, Eser, and Jimenez (1998) further extend this altruism model to a life-cycle model in the context of imperfect credit markets. Parents care about children’s welfare, thus transfer income to their children when the children are poor in the early life-cycle. Besides, when the parents enter retirement phase and have low income, children transfer income to parents out of altruism. A general implication of 2This refers to Becker’s rotten kid theorem. Conditions for the rotten kid theorem to hold are discussed
in Bergstrom (1989).
3When there is a positive parental transfer, the amount decreases with the child’s income. There are
also situations where the parent is at the corner thus does not transfer income to children.
4However, such negative correlation might not be observed in reality for many reasons even if the donor
is altruistic. Stark and Zhang (2002) argue that counter-compensatory inter-vivos transfer may coexist with parental altruism given endogenous income and inter-sibling transfer. They find that when children’s income is not exogenous but an outcome of parental investment (early life transfers), an altruistic parent may invest more in the high-income or high-educated child. The rationale is that such children have higher earning capability and increase the future family income, they would also share the return with the less endowed siblings.
altruism models is that public transfers would crowd out private transfers, for example, parents would transfer less to a child when he receives government subsidies.
Exchange model
Exchange motives are also discussed by Cox (1987), Cox and Rank (1992) and Cox et al. (1998), where the transfer behavior is not a result of altruism but a way to exchange service or future repayment. Exchange is determined by Nash bargaining between parents and children. In the transfer-service model (Cox, 1987; Cox & Rank, 1992), the transfers would occur even if children have no gain in utility under exchange motive. The transfer amount is determined by the implicit price of service when the parent has an exchange motive. A higher child income implies a higher price of service from this child, thus the
parent needs to transfer more to compensate the child’s loss incurred by service provision6.
In the transfer-repayment exchange model (Cox et al., 1998), the relationship between transfer amount and recipient’s pre-transfer income is ambiguous. A higher pre-transfer income of the recipient on the one hand eases liquidity constraints and thus reduces the transfer amount, while on the other hand it also raises the bargaining power which increases the transfer amount. The incidence of transfers is negatively correlated with recipients’ income, since intergenerational lending is less likely to generate mutual benefit when parents and children are richer.
However, enforcement issues are not directly modeled in Cox’s inter-temporal exchange models (Cox, 1987; Cox & Rank, 1992; Cox et al., 1998). Since upward supports take place at a point later than the time when the contract is agreed on, children are possible to defect their commitment. Apart from Becker’s preference manipulation, bequest could be another way to strategically motivate children to provide support (Bernheim, Shleifer, & Summers, 1986). Empirical evidences show that bequest is positively correlated with care provision (Groneck, 2016). Reputation threat would also regulate children’s behavior since failure of repayment harms children’s reputation in the community and reduces the probability of obtaining future informal credit.
Mutuality model
The mutuality model or family constitution model focuses on the enforcement issue within the transfer model. A three generation structure is employed by Cigno (1993). Each in-dividual lives three periods (childhood, middle age and retirement period) and only cares about her own consumption. In period 1, the child has no income and receives parental transfer; in period 2, the middle aged adult earns income, transfers money to children and repays previous parental transfer; in period 3, the old retiree receives repayment from her children. The transfer amounts are fixed and people only make fertility decisions, which 6In Cox (1987), the no market substitute assumption of service ensures the price to be high and drives
are the major differences with previous family models. A self-enforcing family constitu-tion is the key implicaconstitu-tion of the model. If a person does not pay back early parental transfer to her parents in middle age, she would not receive any repayment from her children in her old age. Thus, the defector is exempted from the family constitution. A similar prediction could be obtained from the demonstration effect theory (Cox & Stark, 2005). The mutuality model is also based on a non-altruistic assumption as the exchange model, and these two models are not contradictory.
The motivation is sometimes indistinguishable due to data limitation, whose identification would need a comprehensive consideration of all transfers between all family members (Stark & Zhang, 2002). There are evidences support all three models and not a single motivation or theory is prevalent in our context. The intergenerational transfer pattern in China seems to need more than one theory to explain. Therefore, I do not attempt to differentiate motives or models behind the transfers, but only analyze certain transfer behaviors with possible explanations.
2.2
Family transfers in China
Families in China have traditionally been characterized as corporate kin groups where family members share a common interest and take care of each other (Greenhalgh, 1985). A close tie exists between generations. Apart from education, health and other expenses related to parenting, parents also support adult children’s marriage even if it causes
economic burden and stress (Wei & Zhang, 2011; X. Chen, 2017).7 On the other hand, it
is a social norm for children to support old parents. Although the lack of social security for the elderly is a crucial reason for the prevalence of children’s support, culture also plays an important role. Confucian filial piety emphasize the obligation of children to support old parents, and children learn about the value since a young age (Lin & Fu, 1990). The children-parents exchange contract is also enforced by the filial piety, since disobedience to such social norms would cause both guilt and reputation harm. Children’s supporting obligation is even formalized in law (Zimmer & Kwong, 2003). Therefore, children are often perceived as a reliable old age security or investment asset.
However, rapid social developments have brought some changes to the traditional Chinese pattern. The increasing old age dependency rate caused by lower birth rates and higher life expectancy, imposes a heavier burden on children. It raises the concern that whether the support from less children would be sufficient for parents’ needs. The erosion of Confucian filial piety is likely to undermine the enforcement of the family contract 7In the traditional pattern, an extended family which consists of parents, the son and the
(Y. J. Chen, Chen, & He, 2017). There is also a soaring trend in the amount of marriage transfers, which brings a considerable change to parental saving and healthy behaviors (Wei & Zhang, 2011; X. Chen, 2017). Higher marriage transfers indicate that it might become less efficient for parents to invest in children’s marriage.
A large literature focuses on the intergenerational relations in China. Li, Rosenzweig, and Zhang (2010) find that parents give more gifts to low income and less educated children by using a Chinese twin data set. Single and sick parents are found to receive more child support (Liu, Lu, & Feng, 2017; Wu & Li, 2014), which is consistent with the altruism assumption since altruistic children shall provide supports to parents according to parental needs. These findings support the altruism model. However, there are also evidences support non-altruistic motives. Choukhmane, Coeurdacier, and Jin (2013) show that parents perceive children as old age security and the fertility restriction encourages parents to save more on their own. A positive correlation between old-age support and fertility has been found (Oliveira, 2016). The subjective question of ”What is the purpose of having a child” was asked in a Chinese survey, and 57.8% of the full sample chose ”for old age support” as a reason (Y. J. Chen et al., 2017). Das Gupta et al. (2003) argue that more old age support from sons is the reason for the persistent son preference in the Asian area, which suggests an exchange motive of parental transfers. Zhang (2012) suggests that out-migrant children transfer more money back home to ensure that parents receive care from other siblings. This could be the case that wealthier siblings with higher bargaining power ”buy” service from less well-off siblings, thus exempt themselves from
time-consuming supports8.
I specifically focus on health risks of the elderly in China, since the risks largely influence the intergenerational relationship. The frail elderly in China have to rely on their spouse, children or other relatives for informal care. According to the Research group of China Research Centre on Aging, over 60% of the Chinese elderly who have chronic diseases or disability rely on help from other people. However, they could hardly obtain care outside the family. Institutional care and the recently initiated community-based care in the Chinese long-term care (LTC) system could hardly fulfill needs of the elderly. The lack of formal care facilities, such as nursing house and geriatric rehabilitation center, and insufficient public spending together make the institutional care unreliable (Keating, Otfinowski, Wenger, Fast, & Derksen, 2003). The quality of service from community-based care has also been concerned since many care providers are unprofessional (Xu & Chow, 2011). When parents have bad health or experience health deterioration, the support behavior of children would be altered.
8If we assume that parents desire service from every child instead of a total amount from all children,
2.3
Main hypotheses
The main purpose of this paper is to examine whether and how children pay back marriage transfers to their parents with old age support. The link between the past downward transfer and current upward supports underlies family transfer models. For instance, in the transfer-service exchange model (Cox, 1987; Cox & Rank, 1992), transfers are a compensation for children’s utility cost of providing service, which implicitly incorporates
a positive link as the basic idea.9 If children are altruistic or manipulated to be considerate,
they would feel the obligation to pay back early life transfers. The mutuality model also predicts a positive link between marriage transfers and the incidence of repayment. Larger marriage transfers are more representative for the total parental transfer, thus children are more likely to pay back such transfers with old age support. Therefore, I expect a positive correlation between marriage transfers and old age support.
Although family transfer models have been examined and differentiated by many em-pirical studies, there are few researches focus on the link between long-term parental transfers and old age support from children. The most relevant study is Ho (2017), on which this paper build. Apart from Ho (2017), Ciani and Deiana (2017) provide Italian evidence and find that parents who helped their children in house purchasing when chil-dren get married are more likely to receive informal care in later life. While Cunningham, Yount, Engelman, and Agree (2013) find that neither economic assistance nor instrumen-tal support is associated with investment in children’s education and marriage by using Egyptian data. Most empirical research explore the relation between contemporary bidi-rectional transfers or between the bequest and child service. Literature on these two links is abundant. For instance, Norton, Nicholas, and Huang (2013) use US data and find that children who provide care to parents have higher probability of receiving money trans-fers from parents; Groneck (2016) provides evidence that children’s caregiving behavior increases the incidence and amount of bequests.
According to the altruism model, children provide support when parents desire sup-port, which indicates that old age support from children is need-based. Thus, children may provide significantly more direct care or paying more for care purchase if parents have bad health and are truly in need of care. It is also true that parents with worse health sta-tus and higher needs for care are expected to receive more care and money from children according to the traditional view (Zimmer, 2005). When parents’ limitation is not heavy,
social contact might be more important and support from children could be customary.10
9Although the model does not have clear predictions on the correlation between downward transfers
and upward service. The incidence and amount of exchanges are determined by the bargaining power of both sides, which is reflected by their income.
10Contacts and visits are excluded from old age support in this study, since it is not clear whether
The support behavior might be solely determined by the child’s own capacity and social norms. Thus, marriage transfers may not yet stimulate old age support when parents’ limitation is not heavy. When parents have heavy limitations in daily living, there is a room for the reciprocity of early life transfers to work. Among siblings, those who received higher early transfers are expected to take the responsibility of supporting more for the bad health of parents. Therefore, the repayment of marriage transfers should depend on parental status and it should be positive when the parental health is bad.
When parental health deteriorates as they age, their needs of care and sometimes money for care services would increase correspondingly. As a result, children who care
about their parents would provide increasing upward transfers. However, the rise of
transfers may not be the same for everyone. The interaction between health deterioration
and supports could also be influenced by the early transfers she received. Altruistic
children who received larger transfers would feel guilty if they do not help more. Thus, children who received larger early transfers would increase their supports by a larger magnitude when parental health becomes worse. Furthermore, it would be more efficient for children with an exchange motivation to pay back past transfers when parents are in a worse health situation, since parents place a higher value on help from children when there is a heavier negative health shock. Therefore, marriage transfers may have a larger impact when there is a heavier deterioration in parental health.
No matter whether old age support is a forced result since the child is feared of re-ceiving no support from her own children, or a result of altruism that manipulated by parental transfer, or an execution of the exchange contract with parents, I expect to ob-serve a reciprocity of marriage transfer. Considering the impact of health risks, the main hypotheses are presented below:
Hypothesis 1: old age support increases with marriage transfers.
Hypothesis 2: old age support increases with marriage transfers when the parental ADL limitation level is heavy.
Hypothesis 3: the effect of marriage transfers on old age support becomes larger when the parental ADL limitation is heavier.
3
Data and Descriptive Statistics
3.1
Data
This paper uses data from China Health and Retirement Longitudinal Study (CHARLS), an ongoing micro-longitudinal survey which is nationally representative of Chinese older population aged 45 and above. CHARLS is the sister dataset of Health and Retire-ment Study (HRS), Survey of Health, Aging and RetireRetire-ment in Europe (SHARE) and English Longitudinal Study of Aging (ELSA). CHARLS contains rich information on demographic, health status and functions, health care and insurance, retirement and pen-sion, work and income, family structure and interpersonal transfers. The main surveys are conducted biennially by face-to-face computer-assisted interviews, there are 3 waves of main surveys in 2011, 2013 and 2015. One life history survey in 2014 is also available. The 2011 baseline survey covers 17708 individuals from 10257 families across 450 com-munities, the response rate is 80.5%. Details of the survey design, sampling procedure and samples please see Zhao et al. (2013) and X. Chen, Smith, Strauss, Wang, and Zhao (2017).
I use data from the three main surveys, and treat respondents as parents. There is one family respondent per household in each interview and the respondent’s spouse (if present) is included as a survey respondent. Questions related to children information, intergenerational transfers between parents and each child and past marriage gifts offered to each child (if the child has ever married) are asked to the family respondent. Informa-tion of both coresident and non-coresident children is available, thus there is a complete family (parent-children) structure. Data contained in CHARLS enables child-level and between-sibling analyses.
There are some people exit and enter the sample during the two latter surveys, thus the panel data is unbalanced. The original sample contains 13738 families with 38674 children and 90079 person-year observations. The match of child across waves depends on household and child ID, however, many mismatches could be found by inconsistent gender or age. Data cleaning of child mismatch reduces the sample to 34323 children and 79650 person-year observations. Since marriage gifts questions are asked for the child’s first marriage, I further exclude children who were never married in all waves or got married between waves. With these selections, I have a sample of 60151 person-year observations of 25128 children from 10133 families. Exact sample varies from analysis to analysis, which depends on the availability of variables.
Table 1: Sample selection
Selection family child person-year obs.
original 13738 38674 90079
child correctly matched 13356 34323 79650
child ever married in all waves 10133 25128 60151
wives are younger than husbands. There are averagely 57.6% children aged between 31 and 45 in the three waves, this majority ensures that the sample is representative of the main group providing support to old parents.
Table 2: Sample composition by wave
Age 2011 2013 2015
Percentage Son Daughter Percentage Son Daughter Percentage Son Daughter -30 23.85% 22.26% 25.54% 18.03% 15.90% 20.18% 14.91% 12.14% 17.63% 31-35 18.90% 19.05% 18.73% 18.96% 18.93% 19.00% 17.77% 17.72% 17.81% 36-40 21.06% 21.46% 20.63% 20.39% 21.13% 19.65% 19.41% 20.25% 18.59% 41-45 16.77% 17.12% 16.39% 19.22% 19.95% 18.50% 20.40% 21.53% 19.29% 46-50 10.29% 10.38% 10.18% 12.58% 12.76% 12.40% 14.26% 14.50% 14.03% 50+ 9.14% 9.72% 8.53% 10.80% 11.33% 10.26% 13.25% 13.86% 12.65% OBS 18,754 9,678 9,076 20,385 10,243 10,142 21,012 10,425 10,587
3.2
Descriptive Statistics
I first present cross sectional patterns while describing the measures of variables, including past marriage transfers, intergenerational old age support and some characteristics. Then I focus on the changes over time by cohort and by wave.
3.2.1 Measures and cross sectional patterns
The summary statistics for the variables are presented in Table 3. The upper panel re-ports the one-time marriage transfers, the middle panel rere-ports various old age supre-ports and the lower panel describes respondent parent and child characteristics.
Table 3: Summary Statistics
2011 2013 2015 All
n=18754 n=20385 n=21012 60151
Mean/Prop. SE Mean/Prop. SE Mean/Prop. SE Mean/Prop. SE
Marriage gift
Prop. received money gift 0.66 0.47 0.65 0.48 0.64 0.48 0.65 0.48
Prop. received house gift 0.11 0.31 0.10 0.30 0.10 0.30 0.10 0.30
Amount of money gift if > 0 1.41 2.42 1.46 3.35 1.55 3.40 1.48 3.14
Amount of house gift if > 0 9.80 17.20 10.46 22.27 11.74 23.44 10.74 21.42
Total amount received if > 0 2.85 8.17 2.93 10.26 3.23 10.87 3.01 9.96
Intergenerational support
Prop. best living proximity 0.25 0.43 0.23 0.42 0.23 0.42 0.24 0.42
Prop. provide most ADL help 0.02 0.14 0.05 0.22 0.06 0.24 0.05 0.21
Prop. provide material support 0.49 0.50 0.73 0.44 0.73 0.45 0.66 0.47
Amount of material support if > 0 0.22 0.84 0.21 0.81 0.28 0.81 0.25 0.97
Respondent parent characteristics
Age 65.06 10.00 66.37 9.95 67.32 10.11 66.29 10.06
Prop. female 0.53 0.50 0.53 0.50 0.51 0.50 0.52 0.50
Prop. married 0.73 0.45 0.71 0.45 0.69 0.46 0.71 0.45
No. of children 3.84 1.68 3.86 1.72 3.79 1.73 3.83 1.70
Prop. with primary school 0.47 0.50 0.47 0.50 0.48 0.50 0.47 0.50
Prop. with secondary school 0.24 0.42 0.24 0.43 0.25 0.43 0.24 0.43
Prop. with high school 0.08 0.27 0.09 0.28 0.09 0.28 0.08 0.28
Prop. with college degree 0.01 0.12 0.02 0.13 0.01 0.12 0.02 0.12
Prop. living in urban area 0.20 0.40 0.22 0.42 0.26 0.44 0.23 0.42
Prop. without any difficulty in ADL 0.68 0.47 0.59 0.49 0.55 0.50 0.60 0.49
Prop. with light ADL limitation 0.84 0.37 0.73 0.44 0.71 0.45 0.76 0.43
Prop. with medium ADL limitation 0.12 0.32 0.18 0.39 0.19 0.39 0.17 0.37
Prop. with heavy ADL limitation 0.05 0.21 0.08 0.28 0.09 0.29 0.08 0.27
Child characteristics Age 37.84 9.06 39.23 9.01 40.31 9.06 39.17 9.10 Prop. male 0.52 0.50 0.50 0.50 0.50 0.50 0.50 0.50 Prop. married 0.97 0.18 0.96 0.19 0.95 0.21 0.96 0.20 No. of children 1.53 0.84 1.60 0.80 1.65 0.83 1.59 0.82 Birth order 2.33 1.38 2.33 1.39 2.30 1.39 2.32 1.39
Prop. with primary school 0.83 0.37 0.84 0.37 0.82 0.39 0.83 0.38
Prop. with secondary school 0.58 0.49 0.59 0.49 0.57 0.49 0.58 0.49
Prop. with high school 0.23 0.42 0.23 0.42 0.23 0.42 0.23 0.42
Prop. with college degree 0.09 0.28 0.09 0.28 0.09 0.29 0.09 0.29
Note: Statistics are performed on child level. Number of observations differs across variables due to missing values. All monetary values are in 10000 yuan and reported at 2015 prices.
Upward material support related questions were only asked for noncoresident children in 2011 wave, but asked for both coresident and noncoresident children in 2013 and 2015 waves.
Income is a categorical variable, thus the number is not the amount of income.
All the monetary values are converted to 10000 yuan (approximately 160 USD) in 2015 price.
amount of betrothal gifts reaches 15500 yuan in 2015 among those who received betrothal gifts. The magnitude of betrothal gifts is quite impressive given the overall lower income of rural old parents. The proportions of children received money and house gifts do not vary over time, but there are increases in all the means of amount variables. For instance, the mean of amount of money gift among all children increases from 8800 to 9400 yuan. The mean of total amount among those received positive transfers also rises from 28500 in 2011 to 32300 yuan in 2015. Since marriage gift is a past transfer, no difference in marriage gift should be observed if it is a same group of children in all waves. Given the unbalanced panel, there must be some new children, who received higher marriage transfer, enter the sample in later waves. Moreover, it is clear that the amount of house gift varies more than the amount of money gift across children according to the standard deviations. This might due to the highly uneven housing prices among regions.
Intergenerational support There are four measures of intergenerational support, in-cluding instrumental help and material transfers. I adopt the same construction of support variables as (Ho, 2017). whether best living proximity is a dummy variable which equals one when the child lives in the same household, or in the same village/neighborhood as their parents and parents recognize this living arrangement as the best. Living arrange-ment preferences have been asked to the parents: ”Suppose an elderly person has a/no spouse and adult children, and has good relationship with children, what do you think is the best living arrangement for the elderly person? ” with answers including ”Live with adult children” and ”Don’t live with them in the same house, but live in the same commu-nity or village”. The parents need to make a choice between privacy and receiving daily help. Living close to parents not only represents the probability of more instrumental help (S. Yan, Chen, & Yang, 2003), but also fulfills the filial piety and improves parental wellbeing in China (X. Chen & Silverstein, 2000). There are on average 24% of children living ideally close to their parents. The proportion changes little across waves, which indicates that living arrangement is stable.
tra-ditionally provides informal care to parents-in-law, the proportion providing ADL help in this paper should be much smaller than in other papers where help from the child’s family is summed. In the sensitivity analysis, I test whether there is a differencing in using the child’s care provision or the care provision from the child’s family by 2013 and 2015 date, and find no significant difference. The proportion of children who provided most help increases from 0.02 in 2011 wave to 0.06 in 2015 wave. This increase reflects the trend in care provision when the parent ages.
There are also two measures of material transfers. The third measure is whether pro-vide material support to old parents in the year before the interview. This is again a dummy variable which equals 1 when the child transferred money or in kinds to the par-ents; 0 otherwise. The fourth variable is the amount of material support, which is the amount of money and in kind transfers provided over the previous year in 10000 yuan. Questions related to material support are restricted for non-coresident children in 2011 baseline, but they apply to all children in the later waves. These questions are: ”In the past year, how much economic supports did you or your spouse receive from your [child’s name]? Money support such as helping with living expenses and in kind support includes food or clothes.”. This is the reason why there are only 49% of children provided material support in 2011 wave but the proportion rises to 73% in 2015 wave. Variable amount of material support is censored at zero given that many children did not provide material support. Thus, two part models are employed in analyses of material support. There is some complemetarity between best living proximity and providing ADL help (corr.=0.12 in 2015 wave), and some substitution between best living proximity and providing any material support (corr.=-0.09 in 2015 wave). Children live close to parents seem to pro-vide more ADL help and transfer less money or in kinds.
Parental characteristics I adopt the demographic and health information of the family respondent as a proxy for parental status, and correspondingly use child’s care provision to the family respondent parent. Many demographics of the responding parent are included in the analyses. Parental educational attainment is a proxy of parental life time income. In the sensitivity test, household wealth is included for further controlling the potential bequest motive whereas no significant difference has been found. Number of children in the family reflects the family structure. Whether the parent lives in the urban area is used to define rural and urban sample. There are also other characteristics such as age, gender and marital status.
Among them, ADLl, ADLm and ADLh are crucial variables that truly reflect the
calls, taking medications? ” with answers 1 = No, I don’t have any difficulty, 2 = Yes, I have difficulty but can still do it, 3 = Yes, I have difficulty and need help and 4 = I can not do it for each activity. Having no difficulty is defined as answering 1 (No, I
don’t have any difficulty) to all activities.11 The proportion of the elderly who do not
have any difficulty in all daily living activities is decreasing (0.68 in 2011 to 0.55 in 2015). Answering 3 (Yes, I have difficulty and need help) or 4 (I can not do it) to one activity is defined as having one serious ADL limitation. Since the percentage of parents with 3 or more serious ADL limitations is rather small (0.08 on average), I then generate the following group of dummy variables to represent the light, medium and heavy limitation
level: ADLl=1 if there is no serious ADL limitation, ADLm=1 if there are 1 or 2 serious
ADL limitation(s) and ADLh=1 if there are 3 or more serious ADL limitations. A time
trend of ADL limitations could be observed from Table 3 that more parents are having medium or heavy ADL limitations (11% increase from 2011 to 2015).
Child characteristics Educational attainment together with other demographics re-flects the life time income. It is also a proxy of parental education investment, which
is another major early life investment that parents made12. Especially in rural families,
many children dropped out of school due to financial constraints and started to work and support the family in a young age. Thus, although college education expense is available in one wave of CHARLS, educational attainment is more appropriate for proxying edu-cation investment. Birth order is used together with eduedu-cational attainment to capture
some innate ability13, given the findings that birth order affects earnings and intelligence
quotient (Black, Devereux, & Salvanes, 2005; Barclay, 2015). Other child characteristics include age, gender, marital status and number of children.
3.2.2 Changes over time
Cross sectional statistics provided above show that children provide different kinds of old age support to parents and the majority of children provide either care or material support. More children provide ADL help and the amount of material support rises over time. It seems that children contribute more to supporting their parents when parents age. The child level change of parental ADL limitations also suggests that parents are 11Sample whose parent does not have any difficulty is excluded from some of the regressions on whether
providing the most ADL help. When the parent has no difficulty in ADL, there is no ADL help. However, children may do not provide ADL help when the parent needs care. Do not provide the most help to parental ADL under these two circumstances could be different.
12Ho (2017) uses college education expenditure as the other part of parental investment. Brandt,
Siow, and Wang (2015) study the compensation effect of marriage investment to educational investment. Wei and Zhang (2011) also find that Chinese parents save primarily for the education and marriage expenditure of their children.
worse off as they age. In the rest of this section, I first explore the marriage transfers over time by observing cohort patterns, then conduct analyses of ADL limitations and help received on parent level to check its evolution.
Marriage gift In the following, I show the increasing trend of marriage gift. This might merely be a social norm change that all parents support more to their children. It could also be the result of the rising opportunity cost of children’s time, in the sense that parents need to transfer more to exchange old age care. I take the child level sample with one child one observation (N=25128) and stratify them by age in 2015. There are 6 age cohorts with approximately the same size, and they are age under 31, 31-35, 36-40, 41-45, 46-50 and over 50, respectively. The averages of various marriage gift variables across cohorts are shown in the following graphs.
Figure 1: Received money gift Figure 2: Received house gift
Figure 1 and Figure 2 report the proportion of each cohort received money gifts and housing gift, respectively. Overall, there are always over 50% of children received money gifts, while only less than 10% of them received a house gift. The probability of receiving monetary gifts increases with time, while the probability of receiving a house gift changes little. It is noticeable that the proportion received a house gift is smaller in the youngest group. The soaring housing price in China might be too high for parents to afford a new house recently.
Figure 3 and Figure 4 show the changes in the amount of money gift and house gift. The total amount is also higher in the younger cohorts, which is not presented. The means are performed on all children within the cohorts, they are unconditional averages. From both figures we could find that there are prominent increases in the amounts over time. Younger children are more likely to receive a larger amount of marriage gift. An increase in the amount of marriage gift could be driven by many reasons. For instance, parents who want sufficient informal care in the old age would need to invest more in children’s marriage given children’s increasing time cost. The rising sex ratio (male to female) in some regions might boost the marriage transfers from sons’ families, since the marriage market is more competitive for sons and women have more bargaining power (Angrist, 2002).
ADL limitations and help On the responding parent level (13484 responding parents with 24898 person-year observations from 10133 families), I summarize the evolution of ADL limitations and ADL help to show that ADL is indeed a relevant issue for the elderly. Figure 5 presents how the responding parent’s ADL limitation level changes over time. The percentage of parents with a light ADL limitation level decreases, while the proportion with a medium or heavy level increases. The overall level of ADL limitations is rising as parents age, thus old parents need more help in daily life. The rising pattern is more evident if the sample is restricted to those parents stay in all the three waves. Correspondingly, ADL help provided by children is expected to be responsive to parents’ deterioration of living ability. Figure 6 thus shows whether the parent receive any ADL help from children on the parent level. We could see that the proportion of parents received ADL help is more than doubled in 2015 compared with it in 2011.
Figure 5: Changes in ADL limitation level Figure 6: Changes in ADL help received
Figure 7: Changes in ADL limitation level (single parents)
Figure 8: Changes in ADL help received (single
parents)
single parents have medium to heavy ADL limitations. The worse living ability might be the result of older age, which is also positively correlated with widowhood. As shown by Figure 8 and Figure 6, single parents have much larger probability of receiving ADL help from children. The proportion of single parents who received help is one to two times higher than it of all parents. In 2015 wave, there are around 22.5% single parents received most ADL help from children.
More summary graphs could be found in the appendix, which describe the proportion with at least one near coresident child, proportion received material support, proportion
received average amount of material support per child over 1000 yuan14 and mean of the
average amount of material support per child.
Pattern of old age support Table 4 shows the relationships between changes in parent’s number of serious ADL limitations and changes in old age support across any two
consec-utive waves15. Child level data is adopted for the analysis. For children whose parents’
number of ADL limitations increased between waves, 15.96% had an increase in ADL help provision while 1.96% had an decrease in providing help. For children whose par-ents’ number of ADL limitations kept the same, only 1.51% increased ADL help provision. Children’s ADL help are quite responsive to parental limitations’ change. However, no responsiveness of material support could be found. The percentage of children increased material support has no difference across groups of parental ADL limitations change. Al-though there is a higher proportion of children who increased material support than who decreased support, it seems to be a common increasing pattern over time.
14When a child transfers money or in kinds more than 1000 yuan, the child could be viewed as providing
sufficient material support to parents (Ho, 2017).
15Although there are also changes in best living proximity over time, they are few and likely to be
Table 4: Relationship between changes in ADL limitations and changes in old age support
Change in Change in ADL help Change in material support incidence Change in material support amount No. of ADL limitations Decreased No change Increased Total Decreased No change Increased Total Decreased No change Increased Total Decreased: Number 448 3794 173 4415 377 2315 633 3325 1185 465 1675 3325 Percent 10.15% 85.93% 3.92% 12.92% 11.34% 69.62% 19.04% 13.02% 35.64% 13.98% 50.38% 13.02% Same: Number 132 21898 338 22368 1779 10860 4068 16707 5565 2376 8765 16706 Percent 0.59% 97.90% 1.51% 65.47% 10.65% 65.00% 24.35% 65.43% 33.31% 14.22% 52.47% 65.43% Increased: Number 145 6057 1178 7380 615 3667 1219 5501 1920 738 2843 5501 Percent 1.96% 82.07% 15.96% 21.60% 11.18% 66.66% 22.16% 21.54% 34.90% 13.42% 51.68% 21.54%
4
Empirical Analyses
There are two main parts of empirical analyses. In the basic analyses, pooled OLS and family fixed effect model are employed to study the correlation between marriage gift and old age support. The second part is based on family fixed effect model and child fixed effect model with interactions. It further explores how the interaction between parental ADL limitation level and marriage gift affects old age support.
4.1
Basic Analyses
4.1.1 Empirical models
The basic analyses explore how marriage gift influences old age support and test hypoth-esis 1: yiht = β0+ mgihβ1+ ADLmhtβ m 2 + ADL h htβ h 2 + z 0 ihβ3+ wh0β4+ x 0 ihtβ + δ˜ t+ αh+ εiht. (1)
yiht denotes intergenerational old age support provided by child i from family h in year
t to the family responding parent: whether provide most ADL help, whether best living
proximity, whether provide material support and amount of material support. mgih is
the marriage gift variable, which have the following forms: Whether received money gift, Whether received house gift, Amount of the money gift and Total amount of marriage
gift. ADLm
htand ADLhht reflect whether the responding parent has medium or heavy level
ADL limitation.
xiht are time-variant variables that capture child and parent characteristics, including
child’s age, age square, marital status, number of own children, and parental work status,
marital status, age, age square, gender16. z
ih are child time-invariant variables, including
gender, educational attainment dummies, birth order and whether the oldest child among
siblings dummy. wh are family time-invariant variables, including number of siblings,
parental educational attainment, residence type urban or rural and regions dummies. wh
16Household wealth is also one of the time-variant variables, which is included in the models in
will be excluded from the family fixed effect analysis. δt represents time effect which is
controlled by including wave dummies.
αh denotes the unobserved family effect and εiht is an idiosyncratic error term. In
the pooled OLS model, αh is first assumed to be not correlated with the explanatory
variables. In the family fixed effect model, the identification assumption of family effect
αh is relaxed. Family fixed effect αh represents unobserved fixed family shared values
which may determine marriage investment and also influence old age support. In a more altruistic family, parents invest more in children and children tend to repay more. This generosity or altruism not only generate positive correlation between investment and
return, but also reinforce the estimated correlation. Identification assumption that mgih
may only be correlated with αh but not correlated with the error term εiht needs to be
made.
The basic analyses start from a pooled OLS with mgih= total amount of marriage gift
and wave and region dummies. Then all the other variables that mentioned in the model, which are referred to as ”demographics” in the results table, are included in the pooled regressions. This group of analyses show whether marriage gift if associated with old age support and whether the influence could be explained by child and parental characteristics. With the full set of controls, the heterogeneity of marriage gift is then considered in the sense that amount of money gift and whether house gift are employed to represent
mgih. Instead of using the total amount of marriage transfers as Ho (2017), I differentiate
marital money transfer from marital house transfer. It is not clear how money gift and housing gift differently affect support behavior given the rough total measure of marriage gift. Since money gift and house gift are not exclusive, which means that child could receive both kinds, I replace the total amount of marriage gift with the amount of money
gift and whether the child received house gift17.
To settle endogeneity issues induced by family fixed factors, family fixed effect model is lastly employed. Although the effect of a large bunch of controls has been captured in the pooled analyses, there is still a possibility that some family unobserved characteristics
would exaggerate or underestimate the association (mgih may be correlated with αh).
β1 represents the marginal return of marriage gift in terms of old age support.
Hy-pothesis 1 suggests a positive value of β1, which means that child who received more
marriage gift provides more old age support. 18
17The rationale is that housing price vary a lot across areas and time while the indicator meaning of
helping with house might also be important (Ciani & Deiana, 2017), thus the amount of house gift might be a contamination. After the separation, there would be some associations become significant because of the purification.
18Ho (2017) finds that the effects of marriage transfers on providing the most help to parental ADL
and transferring any money or in-kind to parents differ by gender. It indicates gender differences in β1,
4.1.2 Results of pooled OLS and family fixed effect models
Table 5 presents the regression results of pooled OLS with and without the set of controls. All the old age support measures are significantly correlated with the total amount of marriage gift in the models without controls. Once the demographic characteristics are added, marriage gift becomes insignificant, except amount of material support where only the magnitude of the impact of marriage gift decreases. Many demographic characteristics well explain the association between marriage gift and old age support, for example, age, gender and educational attainment of the child.
Sons receive more marriage gift in general and live ideally closer to parents, so marriage gift become insignificant in the regression of best living proximity once gender and other characteristics are controlled. Part of the effect of marriage gift on the amount of material support is explained by educational attainment when comparing model (7) and (8). One possible explanation would be that parents transfer more marriage gift to more educated children, which suggests a counter-compensatory parental transfer. A larger number of siblings seems to discourage child from providing ADL help and transferring more money or in kind. A higher parental educational attainment represents a higher parental life-time income, and it reduces the incidence of material support while increases the material support amount. This finding supports Cox’s exchange model (Cox, 1987) perfectly, although no family fixed effect is considered.
Table 6 shows the heterogeneity of marriage gift and proofs that using the total amount of marriage gift could be misleading, please check Table A2 for the full regression results. Comparing pooled OLS results in the upper panel of Table 6 with Table 5 we could find that indeed different types of marriage gift sometimes show different and significant associations with upward supports. Children received a larger amount of money gift and children who received a house gift both tend to live closer to their parents as the parents desired. 10000 yuan increase in the amount of money gift is associated with 187 yuan increase in the money or in kind transfer to parents. While children who received a house gift are likely to transfer 615 less yuan to their parents. Different with Ciani and Deiana (2017) who find a positive correlation between receiving housing help in marriage and providing informal care by using Italian data, no significant relation has been found between receiving a house gift and providing the most help to parental ADL. Receiving a larger total amount of marriage transfers is significant positively correlated with the amount of material support provided to parents, however, the amount of money gift and whether received a house gift have different signs of correlation. This difference suggests that Ho (2017) might mis-specify the correlations due to the neglect of heterogeneity in marriage transfers.
Table 5: Pooled OLS: whether the total amount of marriage gift influence old age support
Whether most ADL help Whether best living proximity Whether material support Amount of material support if > 0 (1) (2) (3) (4) (5) (6) (7) (8) total amount of marriage gift −0.000475∗∗∗ −0.000134 0.00216∗∗∗ −0.000448∗ −0.00138∗∗∗ −0.0000930 0.00480∗∗∗ 0.00155∗∗
(−4.20) (−1.14) (8.73) (−1.70) (−5.16) (−0.32) (8.45) (2.41) medium ADL 0.108∗∗∗ 0.00330 −0.00435 −0.0272∗∗ (43.85) (0.60) (−0.71) (−2.03) heavy ADL 0.194∗∗∗ 0.00771 −0.00555 0.00728 (53.05) (0.91) (−0.61) (0.36) age −0.00267∗∗∗ −0.0137∗∗∗ 0.0164∗∗∗ 0.00565 (−2.70) (−6.13) (6.69) (1.01) age2 0.0000244∗∗ 0.000131∗∗∗ −0.000166∗∗∗ −0.0000926 (2.10) (4.90) (−5.81) (−1.43) son 0.0265∗∗∗ 0.264∗∗∗ −0.0743∗∗∗ 0.0911∗∗∗ (14.66) (67.05) (−16.65) (9.05) 1.edu −0.000291 −0.0182∗∗∗ 0.0295∗∗∗ 0.0367∗∗∗ (−0.14) (−4.16) (5.99) (3.32) 2.edu −0.00150 −0.136∗∗∗ 0.101∗∗∗ 0.331∗∗∗ (−0.40) (−16.64) (11.13) (16.47) marriage −0.00904∗ −0.0414∗∗∗ 0.0924∗∗∗ −0.0267 (−1.96) (−4.00) (7.97) (−0.97) num children −0.000384 0.0165∗∗∗ −0.00370 0.00981 (−0.30) (5.90) (−1.18) (1.39) birth order 0.000380 −0.000616 0.00542∗ −0.000808 (0.31) (−0.22) (1.79) (−0.12) oldest child 0.00124 0.00530 −0.00659 0.0224 (0.47) (0.93) (−1.03) (1.56) number sibling −0.00427∗∗∗ −0.0119∗∗∗ 0.00501∗∗ −0.0197∗∗∗ (−5.18) (−6.53) (2.49) (−4.36) pedu −0.00131∗∗ 0.000259 −0.00780∗∗∗ 0.0146∗∗∗ (−2.23) (0.20) (−5.40) (4.41) pwork −0.00507∗∗ −0.0190∗∗∗ 0.0108∗∗ −0.0659∗∗∗ (−2.36) (−4.07) (2.06) (−5.65) pmarried −0.0284∗∗∗ 0.00211 −0.00879 0.0136 (−12.28) (0.41) (−1.55) (1.09) page −0.00572∗∗∗ −0.0000920 0.0208∗∗∗ −0.0137∗ (−4.32) (−0.03) (6.25) (−1.83) page2 0.0000502∗∗∗ 0.00000496 −0.000146∗∗∗ 0.0000697 (5.19) (0.22) (−6.05) (1.29) pfemale 0.00884∗∗∗ 0.0232∗∗∗ 0.0269∗∗∗ −0.0102 (4.23) (5.10) (5.24) (−0.88) purban 0.00326 0.0356∗∗∗ −0.0847∗∗∗ 0.0654∗∗∗ (1.32) (6.48) (−13.86) (4.70) Waves, regions X X X X X X X X N 52372 47358 46626 43676 46911 42803 31677 28834
Waves and regions are dummy variables; Controls are child’s characteristics: age, age square, gender, educational attainment dummies, marital status, number of own children, birth order, dummy whether the oldest child among siblings, and parental characteristics: ADL limitation level dummies, number of children (No. of siblings of the child), educational attainment, work status, marital status, age, age square, gender, residence type urban or rural.
1.edu equals one when the child has completed middle or high school and 2.edu equals one when the child has degrees of college or above. t statistics in parentheses
∗
p < 0.1,∗∗
p < 0.05,∗∗∗
p < 0.01
Table 6: Pooled OLS and family fixed effect: whether received house gift and the amount of money gift influence old age support
whether most ADL help whether best living proximity whether material support amount of material support if > 0 Pooled OLS
(1) (2) (3) (4)
amount of money gift -0.000294 0.00397∗∗∗ -0.000908 0.0187∗∗∗
(-0.64) (3.96) (-0.74) (6.42)
whether house gift 0.00225 0.0241∗∗∗ 0.0108 -0.0615∗∗∗
(0.72) (3.56) (1.37) (-3.37)
N 47358 43676 42803 28834
The upper panel of pooled regressions contain the full set of control variables including wave and region dummies, which are comparable with model 2, 4, 6, 8 in Table 5.
Family fixed effect
(5) (6) (7) (8)
amount of money gift -0.00134∗∗ 0.00733∗∗∗ -0.00201 0.0107
(-2.04) (3.04) (-1.21) (1.60)
whether house gift -0.00588 0.0105 0.00405 -0.0909∗∗∗
(-1.08) (0.69) (0.40) (-2.96)
N 50616 45655 45642 30856
The lower panel of family fixed effect exclude family invariant variables, and contain waves dummies, child demographics (age, age square, gender, educational attainment dummies, marital status, number of own children, birth order, whether the oldest child among siblings dummy) and parental characteristics (work status, marital status, age, age square, gender, ADL limitation level dummies).
t statistics in parentheses
∗p < 0.1,∗∗p < 0.05,∗∗∗p < 0.01
Different with Ho (2017), who finds inconclusive evidence of the relationship between marriage investment and the amount of material support across models, both pooled OLS and family fixed effect in this paper show significant negative associations between receiving a house gift and the amount of upward material support. It is also possible that receiving a house gift denotes a higher bargaining power of the daughter-in-law or son-in-law (mostly daughter-in-son-in-law since few daughter received house gift), which restricts the transfer of money or in kind to parents from the child. However, this explanation builds on the matching hypothesis that men with higher marriage gift marry even ”better” women with higher bargaining power, which is still unverified in the China context(Wei & Zhang, 2011). Only the relationship between best living proximity and amount of money gift is consistent with hypothesis 1 in both models.
4.2
Family fixed effect interaction model
In order to test hypothesis 2 and 3, I first extend the basic model to include the interaction terms between ADL limitation and marriage gift in the family fixed effect model. The family fixed effect interaction model explores how marriage gift influences old age support provision when the parent is at different status. The association between marriage gift
and old age support may vary across parental status, such as ADL limitation level19.
19The change in marital status might also cause heterogeneity. Interactions between parental marital
When parents have light limitations, the marriage transfer may not yet stimulate old age support.
4.2.1 Empirical model
Family fixed effect model is employed to eliminate the endogenous influence of unobserv-able family characteristics, for instance, generosity or altruism. Adding the interaction term into previous model, we have the interaction family fixed effect model:
yiht =β0+ mgihβ1 + (ADLmht· mgih)β1m∗+ (ADL
h ht· mgih)β1h∗+ ADL m htβ m 2 + ADL h htβ h 2 + zih0 β3+ x 0 ihtβ + δ˜ t+ αh+ εiht. (2)
The identification assumption is still that mgih may only be correlated with the family
fixed effect αh but not correlated with the error term εiht. Educational attainment and
birth order are assumed to capture the innate ability.
β1 represents the marginal return of marriage gift in the reference group where the
parental ADL limitation is light. βm∗
1 and β1h∗ show the heterogeneous effects of ADL
limitation level on the marginal return. The sum of β1 and β1m∗ (β1h∗) gives the marginal
return of marriage gift when the parent has medium (heavy) ADL limitation. Hypothesis
2 refers to a positive value of the sum of β1 and β1h∗. Hypothesis 3 suggests that β1h∗ is
larger than βm∗
1 .
4.2.2 Results of family fixed effect interaction model
The coefficients of the interaction terms are only significantly positive on the heaviest ADL limitation level (with whether providing the most ADL help as outcome variable), which testifies hypothesis 3. The finding indicates that a medium limitation level is unable to induce sufficient difference with the reference group of light ADL limitation. Positive return of marriage gift is thus stimulated by a heavy parental ADL limitation. The disparity in marginal return across parental status could be the reason of the insignificant marriage gift effect in the previous model where the coefficient reflects an average effect. Table 7: Family fixed effect: whether the interactions between marriage gift and ADL limitation level influence old age support
(1) (2) (3) (4)
whether most ADL help whether best living proximity whether material support amount of material support if > 0 amount of money gift -0.00796∗∗∗ 0.00756∗∗∗ -0.00191 0.0106
(-4.05) (3.03) (-1.02) (1.25)
whether house gift -0.0504∗∗∗ 0.00740 -0.00286 -0.0960∗∗
(-3.65) (0.46) (-0.25) (-2.43)
medium ADL#amount of money gift 0.00575 -0.000201 -0.00186 -0.00356
(1.46) (-0.05) (-0.42) (-0.39)
heavy ADL#amount of money gift 0.0100∗∗ -0.000964 0.00160 0.00279
(2.35) (-0.20) (0.49) (0.23)
medium ADL#whether house gift 0.0266 -0.00736 0.0193 0.0306
(1.41) (-0.34) (0.81) (0.89)
heavy ADL#whether house gift 0.0906∗∗ 0.0607∗ 0.0405 -0.00990
(2.30) (1.65) (1.19) (-0.22)
N 18625 45655 45642 30856
t statistics in parentheses ∗p < 0.1,∗∗p < 0.05,∗∗∗p < 0.01
Family time-invariant variables are excluded. Controls include waves dummies, child demographics (age, age square, gender, educational attainment dummies, marital status, number of own children, birth order, whether the oldest child) and parental characteristics (work status, marital status, age, age square, gender, ADL limitation level dummies). Note that model 1 with outcome variable whether most ADL help is conducted on the sample of children whose parents have at least one ADL limitation. When the responding parent has no ADL limitation, outcome variable whether most ADL help is 0 by definition. In this case, there should be no return of marriage gift.
4.3
Child fixed effect interaction model
In order to test hypothesis 3 more strictly, I finally estimate a child fixed effect interaction model. By this model, I explore how marriage gift influences old age support provision when parents’ daily living ability deteriorates. Parents are getting older and gradually requiring more help, and they would expect to receive more care or monetary transfer from children.
4.3.1 Empirical model
produces the strictest effect of marriage gift.
The child fixed effect analysis employs a slightly different model:
yit= β0+(ADLmit·mgi)β1m∗+(ADL h it·mgi)β1h∗+ADL m htβ m 2 +ADL h htβ h 2+x 0 itβ+α˜ i+εit, (3)
where αi denotes a child fixed effect. Marriage gift mgi is excluded since it is a past
transfer and time-invariant. Child time-invariant controls zih are further excluded. Child
fixed effect model produces the strictest effect since any heterogeneity among children is
ruled out. Assumption that mgi might only be correlated with the child fixed effect αi
but not with the error term εit needs to be made. The endogeneity issues brought by
innate ability and personality is best solved by child fixed effect.
β2m represents the direct effect of ADL deterioration, from light to medium limitation
level, on old age support. β2h represents the direct effect of ADL deterioration, from light
to heavy limitation level, on old age support. β1m∗ and β1h∗ present the indirect effect
through which marriage gift affects the responsiveness of old age support to parental
ADL limitation changes over time. Hypothesis 3 suggests that β1h∗ should be positive and
larger than βm∗
1 so that the marriage transfer has a larger effect on old age support when
there is a heavy ADL deterioration.
4.3.2 Results of child fixed effect interaction model
Results of the child fixed effect model are reported in Table 820. Marriage gift is omitted
since it is a past transfer. The results are similar with results in the family fixed effect interaction model. Only the interaction between marriage gift and the heavy ADL limi-tation level is significantly positive, and the other three supports seem to be unaffected. The interpretation of interaction terms is slightly different. Within children, when the parental ADL limitation level changes from light to medium over time, the probability of providing the most ADL help increases 10%. Marriage gift not yet stimulates ADL help provision. When the parental ADL limitation level changes from light to heavy, receiv-ing 10000 yuan higher money gift increases the responsiveness of ADL help to 18.01% (0.170+0.0101) and receiving a house gift raises the probability of providing the most ADL help to 23.86% (0.170+0.0686). These results testify hypothesis 3 when providing the most ADL help represents old age support.
Table 9 further reports the results of child fixed effect model of ADL help, where detailed heterogeneity of marriage gift are considered. Regressions on different marriage gift measures are conducted separately. The sample is restricted to children whose parents have difficulty in ADL thus need care. Similar results could be found: the interaction terms are only significantly positive on the heaviest ADL level. For instance, when the
Table 8: Child fixed effect: whether the interactions between marriage gift and ADL limitation level influence old age support
(1) (2) (3) (4)
whether most ADL help whether best living proximity whether material support amount of material support if > 0 medium ADL#amount of money gift 0.00261 0.00149 -0.00469 -0.00876
(0.81) (0.34) (-0.81) (-1.03)
heavy ADL#amount of money gift 0.0101∗∗ 0.000708 0.00262 0.00291
(2.11) (0.22) (0.28) (0.17)
medium ADL#whether house gift 0.0221 -0.0210 0.0239 0.00249
(1.27) (-1.03) (0.78) (0.08)
heavy ADL#whether house gift 0.0686∗ 0.0185 0.0289 0.00100
(1.89) (0.55) (0.59) (0.02) medium ADL 0.100∗∗∗ 0.00340 0.0208∗ -0.00354 (15.74) (0.54) (1.78) (-0.33) heavy ADL 0.170∗∗∗ 0.00175 -0.00697 0.0301 (13.88) (0.16) (-0.35) (1.25) N 50854 45824 45821 30960 t statistics in parentheses ∗p < 0.1,∗∗p < 0.05,∗∗∗p < 0.01
Family time-invariant variables and child time-invariant variables are excluded. Controls include child time-variant demographics (age, age square, marital status, number of own children) and time-variant parental characteristics (work status, marital status, age, age square, gender, ADL limitation level dummies).
parental ADL limitation level changes from light to heavy, receiving marital money gift in the past increases the probability of providing the most help parental ADL by 0.0692. The magnitudes are comparable between whether received money gifts and whether received a house gift, and between the amount of money gift and the total amount of marriage
gift. While surprisingly, the R2s are very similar, which suggests a similar explanation
power of all marriage gift measures.
4.4
Robustness checks
In this section, issues with ADL help provision from the child’s family, parental ed-ucational attainment stratification, contemporaneous downward material transfer and parental wealth are considered. Sensitivity tests are performed mainly on the interaction models, and results of the paper are proved to be robust.
I first substitute the outcome variable whether the child provide the most ADL help with variable whether the child and his/her spouse provide the most ADL help. Since marriage gift is sent to the child’s family and old age support is a family’s joint decision. Given the survey design that ADL provision from the child’s spouse can only be matched to the child in 2013 and 2015 wave, I restrict the sample to observations in the last two waves. The results for family fixed effect and child fixed effect are reported in Table A7, and they are similar to the first column of Table 7 and the first column of Table 8.
