The intergenerational transmission from adult Children's education to parents' health in China: Nonlinearity and mechanisms

Abstract

Current studies suggest the intergenerational transmission of educational advantages is bidirectional over the life course. However, results from causal analysis studies do not consistently support the beneficial effect of adult children's education on aging parents' health. The conflicting evidence indicates a complex relationship, which may be nonlinear or only prominent in certain settings or explained by specific pathways, but remains unexplored. Using the 2016 and 2018 China Family Panel Studies data and instrumental variable estimation, we examine the effect of adult children's education on parents' health and systematically explore its heterogeneity and underlying mechanisms. Our study finds that adult children's education significantly improves parents' health in middle and older ages using instrumental variables estimation with two-stage least squares (IV/2SLS), but the effect may be nonlinear. The beneficial intergenerational transfer of health may slightly weaken when adult children's educational attainment exceeds the middle school education level. The effect of adult children's education on parents' health may be more notable in less developed regions and among younger parents and parents living with or less educated than their adult children. The mechanism analyses results suggest that adult children's education may enhance parents' health through both stress-based pathways (i.e., family economic hardship) and resource-based pathways (i.e., emotional support from children, housework support from parents, and improving parents' access to the resources), but not via the analyzed health habits. Our findings suggest that promoting children's education may improve parents' health over the life course, especially at least graduating from middle school. Our findings imply that prioritizing basic education policy in less developed regions, and providing buffers for economic stressors or enhancing daily intergenerational interactions within families are important for healthy aging in developing societies.

Highlights

• •Improved adult children’s education enhances their aging parents’ health, including mental, physical, and cognitive health.
• •The health benefits for parents may be nonlinear and diminish when adult children’s education extends beyond middle school.
• •This positive intergenerational transfer may be more notable in less developed regions with lower social security levels.
• •Resource-based and stress-based mechanisms can explain this intergenerational health advantage from offspring to parents.

1. Introduction

Education plays a crucial role in determining an individual's health, not only benefiting one's health (Cutler & Lleras-Muney, 2006) but also their family members' health (Ma, 2019). While early studies have focused on the impact of parents' education on children's health from the intergenerational perspective (Balaj et al., 2021), studies of the intergenerational transmission from adult children to their aging parents emerged recently. Some recent developing countries' studies using the instrumental variable (IV) estimation based on local compulsory education law reform have unveiled significant positive effects of adult children's educational attainment on their parents' health, including mental health (Gutierrez et al., 2024), cognitive health (Lee, 2018) and mortality (De Neve & Harling, 2017). Advancements in individuals' socioeconomic status, including improved educational attainment, may enhance their capacity to support their families (Furstenberg, 2019; Wainwright & Marandet, 2010) and thus affect the health of close family members according to a life course perspective (Elder, 2003). As parents age, the intergenerational transfer of education advantage may switch from a parents-to-children dominant pattern to a children-to-parents dominant model (Silverstein & Bengtson, 1994). However, the causal evidence on how adult children's education affects parents' health is still inconsistent. Unlike developing countries, recent studies from developed countries, such as England, Wales, and Sweden, based on local compulsory education law reform (e.g., 1972 educational reform in England and Wales increased the minimum school leaving age from 15 to 16 years) do not support that children's education has positive effects on a range of parent mortality indicators (Lundborg & Majlesi, 2018; Potente et al., 2023). The inconsistent evidence suggests the relationship between adult children's education and the parents' health may be complicated and needs more comprehensive analyses.

One aspect of the complexity may come from the potential nonlinear relationship between adult children's education and parents' health, which is understudied. As children gain higher degrees, the additional increase in the beneficial effect of education on their parents' health may decline. Previous studies already demonstrated that the education effect on one's own health might not be strictly linear due to diminishing returns (Cutler & Lleras-Muney, 2006; Liu et al., 2018), but none have examined the nonlinear intergenerational transmission process of adult children's education to parents' health or possible threshold effect.

1.1. Heterogeneous effect

The heterogeneous effect of adult children's education on parents' health in different settings may also complicate the study results. Some studies have discussed that the differences in regional socioeconomic development conditions may explain the inconsistent results from developed and developing countries. First, in developed countries, the social security system was launched in the 1880s and grew massively after World War II, while it debuted relatively late in developing countries (Dethier, 2009). In China, the modern social security system was formally established with the 1951 “Regulations of the People's Republic of China on Labor Insurance” and developed in the 1980s (Leung, 2003). Second, in developed countries, social security systems are complex and large, and the level of social security is relatively higher than that in developing countries (Dethier, 2009). Although the Chinese basic healthcare insurance coverage rate exceeded 95 % since 2013 (Yip et al., 2019), and the Chinese basic old-age insurance (social pension) covers 1.05 billion population in 2022 according to data from the Chinese Ministry of Human Resources and Social Security, social security expenditure accounted for only around 13 % of the Chinese GDP in 2022, which is lower than the 20–34 % in developed countries (Zhao et al., 2025). This may limit the regulation of social equity, according to the analyses on the social equity index of social security (Zhao et al., 2025). Compared with developed areas, the elderly who live in less developed areas face a greater pension dilemma due to the weak social security system (Chen & Liu, 2009; Potente et al., 2023). The relatively weak social security system makes family support the primary source of care for older persons in less developed areas (Cai et al., 2012; Giles et al., 2010; Pei & Cong, 2020). Consequently, the benefits of adult children's education to their parents' health in less developed regions may be more potent and evident than those in developed ones, and such a phenomenon could also be observed within a large country like China. Unfortunately, whether across countries or within one country, there is a lack of empirical evidence to examine the possible heterogeneity in the causal effects of adult children's education on parents' health at the different levels of regional socioeconomic development.

In addition to the regional socioeconomic development level, parents' and children's characteristics contributing to the heterogeneous effects of adult children's education on parents' health are neither sufficiently discussed. Previous studies primarily focused on children's and parents' demographic characteristics (e.g., gender of parents and adult children) (Friedman & Mare, 2014; Jiang, 2019; Liu, Ma, & Smith, 2022), but paid little attention to parents' age or birth cohort, parents' socioeconomic status (e.g., parents' educational attainment and the educational gap between adult children and their parents) and living arrangements for heterogeneous effects examination, which may provide important conditions to understand the adult education effect on parents' health. First, previous studies have found that the education effect on one's health weakens with age and is stronger in later-born cohorts (Cutler & Lleras-Muney, 2006; Lauderdale, 2001), but few have examined if such a pattern exists for adult children's education effect on parents' health (Liu, 2021; Thoma et al., 2021). Second, the positive effect of adult children's education on the parents' health is more substantial on parents with lower levels of educational attainment or less educated than their adult children compared to their counterparts (Wang, 2024), because of higher dependency on their children's resources (Goldman & Cornwell, 2018). Living with adult children may enhance the child-parent interactions and the positive influence of adult children's education on their parents' health. The phenomenon of parents living with their adult children is prevalent in China and other developing countries (Cameron & Cobb-Clark, 2008). This living arrangement strengthens intergenerational ties and mutual support within families (Meng et al., 2023). Consequently, living with parents helps them expand their investment in health by providing timely and widespread access to adult children's resources. However, few studies have investigated the moderating role of the parent-child education gap and living arrangements.

1.2. Potential pathways

The underlying mechanisms of the positive effect of adult children's education on parents' health are also not fully explored. Previous studies have mainly focused on the potential paths through financial support, emotional support, health behaviors, and access to resources according to the Social Determinants of Health (SDOH) framework (Braveman et al., 2011; Marmot et al., 2012). Some studies indicate that higher-educated adult children are more likely to provide money or emotional support to their parents (Ma, 2019; Ma & Wen, 2016; Umberson, 1992), help their parents maintain healthier lifestyles (i.e. exercising, reducing tobacco use and alcohol drinking) (Friedman & Mare, 2014; Liu, 2021) and more likely to help their parents improve their living environment (including access to clean water and access to clean fuels) (Ma, 2019), thus reduce exposure to hazards and risk of diseases and improve their parents' health (Marmot et al., 2012; Wang, 2024).

Although a series of potential mechanisms linking adult children's education to parents' health has been discussed before, some key potential paths are understudied. First, previous studies mainly focus on resource-based mechanisms but rarely test stress-based mechanisms. Parents of adult children with lower levels of education may be more likely to face some stressors, such as excessive debt or expenditure and family economic hardship (Conger et al., 1994; Zhang et al., 2020). Second, mechanisms of physical activities include both physical exercise and unstructured daily activities like housework (Seol et al., 2021), but the latter is underexplored. Due to the high opportunity cost of doing housework for well-educated adult children (Ma & Wen, 2016), parents may help them with housework or take care of grandchildren (Hämäläinen & Tanskanen, 2021; Huo et al., 2019), which can help the middle-aged and elderly increase physical activity and memory (Adjei & Brand, 2018), and relieve depression symptoms (Chu et al., 2023). Third, in addition to access to a good living environment, mechanisms of access to resources may also include access to the Internet, which hasn't been applied to explain the effect of adult children's education on parents' health. Higher-educated adult children may be more likely to help their parents solve the problem of access to health information by utilizing Internet resources (Lei et al., 2023; Papp-Zipernovszky et al., 2021). The lack of studies on the role of parents' family economic hardship, housework activities, and access to the Internet will limit our understanding of how children's educational advantages transfer to parents' health.

1.3. The present study

Understanding how adult children's education would improve parents' health is important in identifying critical family-level resources and processes for preventing or slowing middle-aged and older parents' health decline. In this study, applying the date of compulsory education being universalized in each county as an IV for adult children's education, we employed instrumental variables estimation with two-stage least squares (IV/2SLS) to examine the causal effect of adult children's education on their parents' multidimensional health using two waves of nationally representative survey data in China. Additionally, we verified whether there is an inflection point in the causal effect of adult children's education on their parents' health. Guided by the SDOH framework and previous literature, we conducted a series of heterogeneity and subgroup analyses to display how the above intergenerational transfer of health varies by the extent of provincial socioeconomic development and children's and parents' characteristics. Finally, we further examined the mediating roles of family economic hardship, family support (financial and emotional support), parents' health behaviors (physical activities, tobacco use, and alcohol drinking), and parents' access to resources (access to the Internet and access to a good living environment).

2. Methods

2.1. Data and sample

This study utilizes data from the China Family Panel Studies (CFPS) project, a nationwide survey conducted by the Institute of Social Science Survey at Peking University. The CFPS project began in 2010 and has since been conducted every two years, resulting in six waves of data. Because of the relatively high proportion of young parents in the previous survey and the COVID-19 outbreak in 2019, the latest two waves of data before COVID-19, collected in 2016 and 2018, were included for analysis. Furthermore, participants with surviving middle-aged or elderly parents aged 45 years and above and living adult children aged 18 years and above were included in the analytical sample. Observations with missingness of dependent, independent, or control variables were excluded. Finally, the main analyses contain 10,412 pairs of parents and children.

2.2. Measures

2.2.1. Children's education

Adult children's education is measured by years of schooling, i.e., years the highest-educated adult children take to obtain their highest degree in standard academic duration. A family may have multiple surviving children aged 18 years and older, while the highest level of education of all living children is mostly strongly associated with parents' health than lowest-educated children and the children living closest to their parents (Zimmer et al., 2007). Therefore, we use the data of the highest-educated child (and the eldest child among the highest-educated children) for each parent. The original survey classifies education into eight groups, from illiteracy/semiliteracy to doctoral degree. We recode their highest education degree into adult children's schooling years using the standard academic duration for each category in China, following the methods provided by previous studies (Kemptner et al., 2011; Ma, 2019).

2.2.2. Instrumental variables for children's education

We use the date of compulsory education being universalized in each county as an IV. China enacted the “Nine-Year Compulsory Education Law” in 1986, providing primary and middle school education with no tuition or miscellaneous fees for school-age children and teenagers nationwide (Ming, 1986; Sun, 2022). In 1992, the Chinese government set specific county-level goals, with a detailed timeline for enforcement and inspection, to accelerate the execution process (Fang et al., 2023). The date for each county to universalize nine-year compulsory education and eradicate illiteracy will affect the educational attainment of school-age children who live in the county, but is independent of parents' health. Therefore, the date of compulsory education being universalized in each county can be an effective IV of adult children's education to estimate the causal effect of adult children's education on their parents' health.

2.2.3. Parents’ health

The multidimensional parents' health is measured by three aspects of health: mental health, physical function, and cognitive function. Mental health is measured by an 8-item version of the Center for Epidemiological Studies Depression Scale (CES-D8) for parents’ self-reported depressive symptoms. CES-D8 is one of the most widely used self-evaluation scales to measure depressive symptoms in many large-scale surveys (Alvarez-Galvez & Rojas-Garcia, 2019; Zhou et al., 2022). The scale scored from 0 to 32, with a higher score indicating more severe symptoms.

Parents’ physical function is indicated by the instrumental ability of daily living (IADL). IADL measures the level at which respondents can perform basic daily tasks independently, which has been widely used in previous studies (Li & Zhou, 2021; Takele et al., 2024). The scale used in CFPS is derived and modified from the ADL scale in European Longitudinal Studies on Aging (ELSA), the WHO IADL scale, and the Lawton IADL scale (Ferrucci et al., 1998; Lawton & Brody, 1969; WHO, 1980). The scale contains seven items, including whether parents can independently go outside, eat, prepare food, use public transportation, go shopping, do cleaning work, and do laundry. For each activity respondents report being able to do independently, one score will be added to their IADL scores. The total scores ranged from 0 to 7; a higher score indicates better ability in daily living and lower functional impairment.

We measure parents' subjective memory through the item “How many major things can you remember that happened to you in a week?” in the CFPS survey. Responses to this question were rated on a 5-point scale, which equals 1 = everything, 2 = most of them, 3 = half of them, 4 = a few, and 5 = a little. According to the responses, parents' reporting that they can remember only a little or a few things indicates that parents may have memory impairment, while reporting remembering most major things or everything means parents may have a good memory (Yuan, Gong, & Han, 2020). Parents’ subjective memory status is coded as a three-point variable, indicating having good subjective memory (remembering everything or most of the major things), ordinary subjective memory (remembering half of the major things), or not good subjective memory (remembering a few or little major things); the higher the score, the better the memory. (Xu et al., 2022).

2.2.4. Characteristics of parents, adult children, and families

We accounted for an array of individual and family characteristics as control variables. Adult children's and parents' characteristics include their age, gender (1 = male, 0 = female), marriage status (1 = married/cohabitation, 0 = unmarried), parental reports of whether the highest-educated children live with parents at the time, parents' education years, whether the parent is less educated than the indicated adult child (1 = yes, 0 = no), percentage of parents who had any chronic diseases over the past 6 months and parents' medical cost over the past 12 months (10,000 Yuan). Families' characteristics include the number of family members (including non-immediate family members with blood/marriage/adoption relationships who live at home for three months and immediate family) and annual household income per capita (10,000 Yuan).

This study further examines partial parents' and adult children's characteristics, including parents' and adult children's demographic characteristics, parents' socioeconomic status, and adult children's living arrangements, as moderators in exploring the heterogeneity. The parents' and adult children's demographic characteristics considered in this study are whether parents live in urban areas and parents' and adult children's gender. Parents' socioeconomic status is indicated by whether parents graduated from middle school or above and whether the parents are less educated than their adult children. Education is one of the indicators to assess socioeconomic status, and education may affect other indicators of socioeconomic status (e.g., income and occupation) (Oreopoulos & Salvanes, 2011; Yang & Gao, 2018). Lastly, whether the stated child is living with parents indicates the adult children's living arrangement.

2.2.5. Characteristics of regions

This study also examines the heterogeneous effect of adult children's education on their parents' health at different levels of provincial socioeconomic development. The regional socioeconomic development level considered in this study reflects regional economic development and social security level. Social security involves aspects such as unemployment insurance, the pension system, and the medical system, aiming to cover people's old age, unemployment, health, and other risks (Dethier, 2009). According to prior studies, provincial socioeconomic development includes three indicators in this study: whether provincial GDP is higher or equal to the national average (1 = yes, 0 = no), whether the proportion of provincial social security expenditure (including subsidies and relief that may improve the employment situation and enhance social security and social welfare) in GDP is higher or equal to the national average (1 = yes, 0 = no), and whether the proportion of provincial health expenditure in GDP is higher or equal to the national average (1 = yes, 0 = no) (Bradley et al., 2011; ILO, 2021). We use data from the China Statistical Yearbooks to develop the above three indicators (National Bureau of Statistics of China, 2019, National Bureau of Statistics of China, 2017). We combined the provincial socioeconomic development indicator data set with the CFPS data set using the administrative division code of each province.

2.2.6. Family economic hardship

The first potential pathway we examine in this study is via family economic hardship. Family economic hardship is measured by the family debt-to-asset ratio and family expenditure-to-income ratio, following previous studies (Liu et al., 2021; Zhang et al., 2020). Family assets include land, housing, financial assets, productive fixed assets, and durable goods assets; family debt includes housing and nonhousing liabilities. Family expenditure and income refer to the household's total living expenditure (including consumer, transfer, security, and house purchase expenditure) and income (including wage, operating, property, transfer income, and others) over the past 12 months.

2.2.7. Family support

Family support, another potential pathway, includes financial and emotional support. It is worth noting that CFPS only inquired of respondents aged 60 and above about the frequency of family support received from their children. Financial support refers to financial support from adult children to their parents. The amount of financial support from child to parent is measured by the answer to “How much money does your adult child give you each month on average?” and the reported amount of money given by the most educated adult children.

Emotional support contains three variables: the relationship between parents and their adult children, the frequency of remote contact between parents and their adult children, and the frequency of meetings between parents and their adult children. First, we use a 5-point scale ranging from “not close” (1 point) to “very close” (5 points) to measure the relationship between parents and their adult children. Second, we measure the frequency of remote contact and meetings between parents and their adult children using a 7-point scale ranging from never to almost every day. Remote contact in this study means communicating with parents by phone, text, letter, or email.

2.2.8. Health behaviors

We use two sets of variables to detect the mediating role of health behaviors: physical activities and substance use, which are common indicators of health behavior (Noble et al., 2015; Steptoe et al., 1996). First, physical activities include self-reported hours of exercise last week and average hours of housework per day for each parent. Second, substance use includes tobacco smoking and alcohol drinking, which are indicated by “whether parents smoked last month” and “whether parents drank three times or more a week last month”, respectively.

2.2.9. Access to resources

Access to resources includes access to the Internet and a good living environment. First, we use “whether parents can access the Internet” to indicate the parents’ ability to access information through the Internet. In detail, for the item “Do you use a mobile device to access the Internet, such as a mobile phone or tablet?”, the answer “yes” is coded as 1, indicating the respondent can access the Internet to gain health information, and the answer “no” is coded as 0. Second, we use a set of dummy variables to measure whether parents can access clean water and clean fuels from the survey item asking individuals about their cooking water and fuel usage. If individuals mainly use tap or pure water (filtered tap water) for cooking, we define it as having access to clean water. If individuals mainly use gas, electricity, or solar energy for cooking, we code them as having access to clean fuels.

2.3. Statistical analyses

To explore the effect of adult children's education on their parents' health, the IV/2SLS regression models with fixed effects are used. The regression model is specified as follows:

$${parent{s}^{\prime }health}_{ij}={\alpha }_0+{\beta }_{1}adult{childre{n}^{\prime }seductaion}_{ij}+{\beta }_2{X}_{ij}+{\omega }_{ij}+{\epsilon }_{ij}$$ (1)

$$adult{childre{n}^{\prime }seductaion}_{ij}={\alpha }_0+{\gamma }_1{IV}_{ij}+{\gamma }_2{X}_{ij}+{\omega }_{ij}+{\epsilon }_{ij}$$ (2)

Where $X_{ij}$ represents the characteristics of parents $i$, adult children of parents $i$, and their families in survey year $j$. ${IV}_{ij}$ represents the date of compulsory education being universalized in each county, which is an IV for adult children's schooling years. ${\omega }_{ij}$ indicates the regional fixed effect, survey year fixed effect, and parents' birth cohort fixed effect. ${\epsilon }_{ij}$ is the error term.

Nonlinear relationships are often tested by three methods: 1) transforming the continuous independent variable into a discrete one and identifying each category of values' effect (Zhao et al., 2022); 2) jointly modeling linear and quadratic scales of the independent variable (Anand et al., 2018); and 3) estimating the effect of the interaction of the independent variable and its value as a threshold dummy variable (Li et al., 2022; Lynch, 2003). The first and second methods are more suitable for continuous variables. The independent variable considered in this study, adult children's education, is categorical. To facilitate the interpretation of the marginal health benefits for parents per additional year of adult children's education, we transformed adult children's highest educational qualifications into schooling years, referring to the Chinese standard academic durations per educational tier (Kemptner et al., 2011; Ma, 2019). Considering the discrete distribution feature of adult children's schooling years and the appropriate explanation of results, we chose the third method to explore the nonlinear effect over different educational categories. A series of threshold dummy variables are constructed based on the distribution of adult children's educational attainment in this study, starting from primary school or below degree to master and above degree; if adult children's educational attainment equals to or exceeds the threshold, the dummy variable is coded as 1; otherwise, it is coded as 0. The coefficients of interaction terms between adult children's education years and the threshold dummy variables are examined, representing the difference in the intergenerational transmission effect estimated at children's education is equal to or above the threshold versus at children's education is below the threshold. All the interaction terms are tested one by one.

The heterogeneous effect of adult children's education on parents' health among parents' birth cohorts is examined by grouped regression analyses. Because this study only contains two waves of survey data, the birth cohort grouping can be representative of the age grouping to some extent. For the heterogeneous effect among other moderators, we add the interaction term between adult child education and each moderator (level of provincial socioeconomic development, and parents' and adult children's characteristics) to explore the heterogeneity in the relationship between adult children's educational attainment and their parents' health status. All the interaction terms are tested one by one.

As for the mechanisms analysis, firstly, we explore which potential pathways can be affected by adult children's schooling years. Secondly, after identifying which key mediating variables are affected by adult children's education, we apply a causal inference approach to directly examine the hypothesized pathways from adult children's education to their parents' health through these key intermediate variables (Emsley & Liu, 2013; Valeri & VanderWeele, 2013).

3. Results

3.1. Summary statistics of characteristics of parents, adult children, and family

Table 1 presents the characteristics of parents, their adult children, and their families. Among adult children, we find that the average age of the sampled adult children is 30.62, and nearly two-thirds are male. Over half of the highest-educated children are married or cohabiting, and the average schooling years of adult children are 12.04 years. Among the parents, the average age of all the sample parents is 56.47, 52.13 % of them are female, and most of them are married or cohabiting. The average schooling years is 6.29 years for parents, and more than 90 % of them are less educated than their children. For family characteristics, each family has close to four members on average. The average annual household income per capita is 23,000 CNY in our sample.

Table 1.

Characteristics of samples.

	Mean/%	SD
Characteristics of the highest-educated adult children
Education years of the child	12.04	3.60
Age of the child	30.62	7.39
Male child (%)	63.82
Unmarried child (%)	41.53
Living with parents (%)	52.87
Characteristics of parents
Urban parent (%) a	47.68
Age of parent	56.47	8.18
Male parent (%)	47.87
Education years of parent	6.29	4.57
Parents graduated from middle school or above (%)	17.41
Less educated than the indicated adult child (%)	95.55
Unmarried parent (%)	9.02
Had any chronic diseases over the past 6 months (%)	28.64
Medical cost over the past 12 months (1000 Yuan)	3.26	9.48
Characteristics of families
Number of family members	4.35	1.96
Annual household income per capita (10,000 Yuan)	2.30	4.66
Characteristics of regions
Provincial GDP above or equal to the national average (%)	38.99
Provincial social security expenditure level above or equal to the national average (%)	41.74
Provincial social health expenditure level above or equal to the national average (%)	34.18
County-level GDP per capita (10,000 Yuan) b	4.45	4.05
Characteristics of parents' health status
Depression score (CES_D8)	13.89	4.24
IADL score	6.73	0.93
Subjective memory score	1.98	0.83

Notes: N = 10,412. SD means Standard Deviation. GDP means Gross Domestic Product. IADL means the Instrumental Activities of Daily Living. CES_D8 means the 8-item Center for Epidemiological Studies Depression Scale.

^aNumber of observations of urban parents is 10,375.

^bNumber of observations of county-level GDP per capita is 7607.

Table 1 also presents the parents’ health distribution, including mental health, IADL scores, and subjective memory. The average depression score of sample parents is 13.89, less than half the total score of 32. The average IADL score is 6.73 on a scale of 0–7. Additionally, the average subjective memory score is 1.98, with less than a third of parents reporting having good memories.

3.2. Relationships between children's education and parents' health

The OLS and IV/2SLS estimates in Table 2 support that elevated adult children's educational attainment can improve their parents' health status. Specifically, when controlling for characteristics of parents, adult children, and families, as adult children achieve higher educational attainments, their parents' depression score is significantly reduced (p < .001), and their parents' IADL score (p < .001) and subjective memory status are improved (p < .05). The results of IV/2SLS effect estimates are considerably larger than the corresponding OLS effect estimates, which may be because OLS, a method with no control for endogeneity bias, may underestimate the true effect of adult children's education on their parents' health (Chen & Hamori, 2009; Ishimaru, 2024). According to the weak identification test and under-identification test (the regressions' Cragg-Donald Wald F-statistics are above 10, p-values of the LM statistic are below 0.001), the instrumental variable used in this study is effective and suitable, and the results in this study are reliable.

Table 2.

Effects of the highest-educated adult children's education on their parents' health.

	Depression score (CES_D8)		IADL score		Subjective memory score
	OLS	IV	OLS	IV	OLS	IV
Adult children’s education years	-0.074∗∗∗	-0.538∗∗∗	0.018∗∗∗	0.159∗∗∗	0.011∗∗∗	0.051∗
	(0.014)	(0.136)	(0.004)	(0.034)	(0.003)	(0.025)
Characteristics of the highest-educated adult children
Age of the child	-0.027∗	-0.053∗∗∗	-0.001	0.007	0.005	0.007∗∗
	(0.013)	(0.015)	(0.004)	(0.004)	(0.003)	(0.003)
Male child	0.229∗∗	-0.113	0.019	0.123∗∗∗	-0.009	0.020
	(0.084)	(0.134)	(0.023)	(0.032)	(0.020)	(0.025)
Living with parents	-0.369∗∗∗	-0.473∗∗∗	-0.029	0.002	0.030	0.039∗
	(0.088)	(0.098)	(0.021)	(0.022)	(0.019)	(0.018)
Unmarried child	-0.481∗∗∗	-0.714∗∗∗	0.021	0.092∗∗	-0.018	0.002
	(0.101)	(0.124)	(0.026)	(0.031)	(0.022)	(0.023)
Characteristics of parents
Age of parent	0.062∗	0.101∗∗	-0.013	-0.024∗∗	-0.028∗∗∗	-0.031∗∗∗
	(0.030)	(0.034)	(0.008)	(0.008)	(0.006)	(0.006)
Male parent	-0.827∗∗∗	-1.103∗∗∗	0.013	0.096∗∗∗	0.117∗∗∗	0.141∗∗∗
	(0.083)	(0.118)	(0.020)	(0.028)	(0.017)	(0.022)
Education years of parent	-0.105∗∗∗	0.032	0.011∗∗∗	-0.031∗∗	0.037∗∗∗	0.026∗∗∗
	(0.011)	(0.041)	(0.003)	(0.010)	(0.002)	(0.008)
Less educated than the indicated child	0.053	2.461∗∗∗	0.009	-0.722∗∗∗	-0.010	-0.218
	(0.207)	(0.736)	(0.064)	(0.188)	(0.044)	(0.135)
Unmarried parent	-1.707∗∗∗	-1.339∗∗∗	0.053	-0.059	-0.018	-0.050
	(0.170)	(0.211)	(0.047)	(0.050)	(0.033)	(0.035)
Had any chronic diseases over the past 6 months	1.430∗∗∗	1.471∗∗∗	-0.097∗∗∗	-0.110∗∗∗	-0.071∗∗∗	-0.074∗∗∗
	(0.093)	(0.099)	(0.024)	(0.024)	(0.018)	(0.018)
Medical cost over the past 12 months (1,000 Yuan)	0.032∗∗∗	0.033∗∗∗	-0.016∗∗∗	-0.016∗∗∗	-0.001	-0.001
	(0.005)	(0.006)	(0.002)	(0.002)	(0.001)	(0.001)
Characteristics of families
Number of family members	-0.061∗	-0.131∗∗∗	0.003	0.025∗∗	-0.007	-0.001
	(0.025)	(0.034)	(0.006)	(0.008)	(0.005)	(0.006)
Annual household income per capita(10,000 Yuan)	-0.033∗∗	-0.004	0.004∗	-0.005	0.010∗∗∗	0.007∗∗∗
	(0.012)	(0.012)	(0.002)	(0.003)	(0.002)	(0.002)
Province fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
Survey year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
Parents’ cohort fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
p-value of LM statistic		<0.001		<0.001		<0.001
F-statistic for weak identification		119.397		119.397		119.397

Notes: N = 10,412. Regression coefficients are given in the table. Standard errors are in parentheses. IADL means the Instrumental Activities of Daily Living. CES_D8 means the 8-item Center for Epidemiological Studies Depression Scale.

†p < .10, ∗p < .05, ∗∗p < .01, ∗∗∗p < .001.

3.3. Non-linearity effect of adult children's education on parents' health

We further explore the nonlinear causal effect of adult children's education on parents' health. The results in Fig. 1 show that the estimated effect of adult children's education on their parents' health is not the same size for different educational degrees. The interaction item coefficients indicate that the effect size of education for adult children with at least middle school degrees on parents' health is larger than that for those below middle school degrees, though not statistically significant. However, the effect size for adult children with at least a high school degree is significantly smaller than that for children with middle school degrees or less (i.e., below high school). As adult children's education thresholds exceed middle school, the positive effect of each additional year of schooling on their parents' health will decline and weaken.

Fig. 1.

Differences in effects of adult children's education on their parents' health over adult children's education thresholds.

Notes: IADL means the Instrumental Activities of Daily Living. CES_D8 means the 8-item Center for Epidemiological Studies Depression Scale.

The error bar illustrates the 90 % confidence intervals for the estimated coefficients. Thresholds are defined by whether the education of the highest educated child equals to or exceeds these education levels, if adult children's educational attainment equals to or exceeds the threshold, the threshold dummy variable is coded as 1; otherwise, it is coded as 0. Coefficients are interaction terms between adult children's education years and the threshold dummy variables, indicating how much the intergenerational health effect of adult children's education would change when children's education is equal to or above the threshold versus below the threshold.

3.4. Results for heterogeneity and subgroup analyses

To gain a deeper understanding of how adult children's education affects the health of their middle-aged or elderly parents, we perform its heterogeneity for various levels of demographic characteristics, socioeconomic status, and living arrangements of both parents and adult children. We also present the results of the under-identification test and weak identification test for IV.

Parents' and adult children's demographic characteristics include whether parents live in urban areas, parents' age, parents' gender, and adult children's gender. Parents' socioeconomic status indicates parents' education and the parent-child education gap. First, Fig. 2 demonstrates that children's education may exert a larger effect on parents' health for the later-born birth cohorts, but the pattern is not consistent over different health outcomes. For depression, parents aged 52–59 years had larger effects compared to other age groups, but not statistically significant. For IADL, parents aged 77–91 years had a relatively larger but not statistically significant effect, while parents aged 57–74 years had smaller but significant effects. For subjective memory, parents under 59 years had larger effects than other age groups, but statistically insignificant.

Fig. 2.

Estimated effects of adult children's education on their parents' health by different parents' birth cohort group.

Notes: IADL means the Instrumental Activities of Daily Living. CES_D8 means the 8-item Center for Epidemiological Studies Depression Scale.

The graph illustrates the subgroup regression coefficient and their 90 % confidence interval estimates for parents of different birth cohorts.

Moreover, as shown in panel A of Table 3, the effect of adult children's education on parents' health may not vary over parents' or adult children's gender as a whole but differ by children's gender when limited to the rural population. Second, the IV/2SLS results demonstrate that compared to parents who are the same as or more educated than their children, adult children's education may exert a greater effect on the mental health and IADL of parents who are less educated than their children (p < .05). Third, for living arrangements, we find that the education of adult children who live with their parents has a greater positive effect on their parents' IADL than adult children who do not live with their parents (p < .01).

Table 3.

Heterogeneous effects by parents and the highest-educated adult children's characteristics and provincial socioeconomic development conditions.

	Depression score (CES_D8)		IADL score		Subjective memory score
	OLS	IV	OLS	IV	OLS	IV
Panel A. By parents' and adult children's characteristics
Adult children's education years $\times$ Urban parent	0.001	−0.101	−0.016∗∗	−0.046	0.005	−0.006
	(0.024)	(0.155)	(0.006)	(0.036)	(0.005)	(0.028)
Adult children's education years $\times$ Male parent	0.006	0.033	−0.013∗	−0.035	−0.012∗∗	−0.004
	(0.022)	(0.110)	(0.006)	(0.028)	(0.004)	(0.020)
Urban: Adult children's education years $\times$ Male parent	0.026	0.130	−0.008	−0.062	−0.004	0.056
	(0.036)	(0.237)	(0.007)	(0.047)	(0.007)	(0.040)
Rural: Adult children's education years $\times$ Male parent	−0.002	−0.090	−0.016a	0.016	−0.014∗	−0.029
	(0.030)	(0.179)	(0.009)	(0.049)	(0.006)	(0.033)
Adult children's education years $\times$ Male child	0.045	−0.376	−0.048∗	0.008	−0.029∗	−0.023
	(0.063)	(0.367)	(0.019)	(0.090)	(0.013)	(0.070)
Urban: Adult children's education years $\times$ Male child	0.151	0.837	−0.013	0.161	−0.008	−0.208
	(0.097)	(1.198)	(0.019)	(0.175)	(0.019)	(0.214)
Rural: Adult children's education years $\times$ Male child	−0.064	−1.161a	−0.078∗∗	0.072	−0.049∗∗	0.076
	(0.087)	(0.644)	(0.029)	(0.172)	(0.018)	(0.116)
Adult children's education years $\times$ Parent graduated from middle school or above	−0.072a	0.263	−0.001	0.181	0.009	0.083
	(0.040)	(0.375)	(0.008)	(0.119)	(0.008)	(0.078)
Adult children's education years $\times$ Parent who is less educated than the indicated child	−0.003	−0.565∗	0.040∗∗∗	0.129∗	−0.019	0.019
	(0.054)	(0.258)	(0.010)	(0.064)	(0.010)	(0.058)
Adult children's education years $\times$ Living with parents	0.007	−0.097	0.026∗∗∗	0.077∗∗	−0.002	−0.043
	(0.022)	(0.114)	(0.006)	(0.027)	(0.004)	(0.025)
Panel B. Characteristics of regions
Adult children's education years $\times$ Provincial GDP is higher or equal to the national average	0.001	0.228a	0.001	0.033	0.009∗	0.025
	(0.024)	(0.120)	(0.006)	(0.034)	(0.005)	(0.023)
Adult children's education years $\times$ The proportion pf provincial social security expenditure in GDP is higher or equal to the national average	−0.021	0.960∗∗	0.023∗∗∗	−0.110	−0.009a	−0.077a
	(0.023)	(0.307)	(0.006)	(0.067)	(0.004)	(0.045)
Adult children's education years $\times$ The proportion of provincial health expenditure in GDP is higher or equal to the national average	−0.013	0.447∗	0.006	−0.154∗∗	−0.017∗∗∗	−0.116∗∗∗
	(0.023)	(0.190)	(0.006)	(0.051)	(0.004)	(0.035)

Notes: The number of observations for all interaction terms in this table is 10,412, in addition to the interaction term of urban. Due to the missing data, the number of observations for the interaction item of urban is 10,375. Number of observations of urban subgroup of gender heterogenous analyses is 4,947, of rural subgroup of gender heterogenous analyses is 5428. Regression coefficients are given in the table, and coefficients for each set of interaction terms are from separate models. Standard errors are in parentheses. IADL means the Instrumental Activities of Daily Living. CES_D8 means the 8-item Center for Epidemiological Studies Depression Scale. GDP means Gross Domestic Product.

Province, survey year, and parents' birth cohort fixed effects are added in all the regressions. All regressions also include control variables of adult children's characteristics (education years of the highest educated child, age of the highest educated child, indicators for males, whether the highest educated children live with their parents and marital status), control variables of parents' characteristics (age and indicators for male, highest educational attainment, whether the parents are worse educated than their adult children, marital status, whether had any chronic diseases in the past 6 months and medical cost) and control variables of families' characteristics (number of family members and annual household income per capita).

^ap < .10, ∗p < .05, ∗∗p < .01, ∗∗∗p < .001.

This study also explores the heterogeneity at the level of provincial socioeconomic development. For provincial GDP, the results in panel B of Table 3 indicate that depression scores will decline more for parents residing in provinces with GDP below the national average as adult children's schooling years increase (p < .10). For social security expenditure, the results demonstrate that with the increased proportion of provincial social security expenditure in GDP, the positive effect of adult children's education on parents' mental health and cognitive function is weakening (p < .10). For public health expenditure, the results reveal that adult children's education may more evidently improve the health of parents who live in the province with a lower proportion of public health expenditure in GDP (p < .05). These results support the hypothesis that the positive effect of adult children's education on parents' health may be greater in provinces with lower levels of socioeconomic development.

3.5. Results for mechanism analyses

The IV/2SLS estimation results for the indirect effect of adult children's education on their parents' health are reported in Table 4, Table 5, Table 6, Table 7. TableA6. Table 4, Table 5 indicate that parents of higher-educated adult children will face less family economic hardship (p < .05), and their children will remotely contact or see them more often (p < .01 and p < .001, respectively). For parents' health behaviors, adult children's education will make parents spend more time doing housework (p < .05), but has no significant effect on parents' tobacco use and drinking alcohol behaviors. For access to resources, as shown in Table 7, as adult children become more educated, their parents are more likely to access the Internet or better clean water and fuel (p < .01). Among all potential pathways, the above results support that family economic hardship, emotional support, proper housework, and access to resources may be the key mediators to explain the causal effect of adult children's education on their parents' health. The OLS estimation results are generally consistent with the IV/2SLS estimations (Tables A2-5).

Table 4.

IV/2SLS estimation for the effect of the highest-educated adult children's education on family economic hardship.

	Family debt-to-asset ratio	Family expenditure-to-income ratio
Adult children's education years	−0.026∗∗	−0.034∗
	(0.008)	(0.015)
Province fixed effect	Yes	Yes
Survey year fixed effect	Yes	Yes
Parents' birth cohort	Yes	Yes
p-value of LM statistic	<0.001	<0.001
F-statistic for weak identification	118.199	113.870
Number of Observations	10,013	10,022

Notes: Regression coefficients are given in the table. Standard errors are in parentheses.

All regressions also include control variables of adult children's characteristics (age of the highest educated child, indicators for male, whether the highest educated children live with their parents and marital status), control variables of parents' characteristics (age and indicators for male, highest educational attainment, whether the parents are worse educated than their adult children, marital status, whether had any chronic diseases in the past 6 months and medical cost) and control variables of families' characteristics (number of family members and annual household income per capita).

∗p < .05, ∗∗p < .01.

Table 5.

IV/2SLS estimation for the effect of the highest-educated adult children's education on family support.

	Family financial support	Family emotional support
	Amount of financial support from the child	Child-parent closeness	Child-parent remote contact frequency	Child-parent meeting frequency
Adult children's education years	−0.001	−0.031	0.176∗	0.261∗∗∗
	(0.023)	(0.026)	(0.069)	(0.073)
Province fixed effect	Yes	Yes	Yes	Yes
Survey year fixed effect	Yes	Yes	Yes	Yes
Parents' birth cohort	Yes	Yes	Yes	Yes
p-value of LM statistic	<0.001	<0.001	<0.001	<0.001
F-statistic for weak identification	80.535	90.413	90.603	90.001
Number of Observations	2504	3355	3352	3353

Notes: Regression coefficients are given in the table. Standard errors are in parentheses.

Family support is only applied to participants aged 60 and above at the time of data collection. All regressions also include control variables of adult children's characteristics (age of the highest educated child, indicators for male, whether the highest educated children live with their parents and marital status), control variables of parents' characteristics (age and indicators for male, highest educational attainment, whether the parents are worse educated than their adult children, marital status, whether had any chronic diseases in the past 6 months and medical cost) and control variables of families' characteristics (number of family members and annual household income per capita).

∗∗p < .01, ∗∗∗p < .001.

Table 6.

IV/2SLS estimation for the effect of the highest-educated adult children's education on their parents' health behaviors.

	Hours parents exercised in the last week	Average hours parents spent on housework per day	Whether parents drank three times or more a week last month	Whether parents smoked last month
Adult children's education years	0.411	0.317∗	0.007	0.005
	(0.277)	(0.154)	(0.010)	(0.011)
Province fixed effect	Yes	Yes	Yes	Yes
Survey year fixed effect	Yes	Yes	Yes	Yes
Parents' birth cohort	Yes	Yes	Yes	Yes
p-value of LM statistic	<0.001	<0.001	<0.001	<0.001
F-statistic for weak identification	120.922	25.174	119.397	119.397
Number of Observations	10,396	2721	10,412	10,412

Notes: Regression coefficients are given in the table. Standard errors are in parentheses.

All regressions also include control variables of adult children's characteristics (age of the highest educated child, indicators for male, whether the highest educated children live with their parents and marital status), control variables of parents' characteristics (age and indicators for male, highest educational attainment, whether the parents are worse educated than their adult children, marital status, whether had any chronic diseases in the past 6 months and medical cost) and control variables of families' characteristics (number of family members and annual household income per capita).

∗p < .05.

Table 7.

IV/2SLS estimation for the effect of the highest-educated adult children's education on their parents' access to resources.

	Whether parents can access the Internet	Whether parents can access clean water	Whether parents can access clean fuels
Adult children's education years	0.046∗∗∗	0.052∗∗∗	0.180∗∗∗
	(0.012)	(0.016)	(0.023)
Province fixed effect	Yes	Yes	Yes
Survey year fixed effect	Yes	Yes	Yes
Parents' birth cohort	Yes	Yes	Yes
p-value of LM statistic	<0.001	<0.001	<0.001
F-statistic for weak identification	119.397	119.397	118.593
Number of Observations	10,412	10,412	10,409

Notes: Regression coefficients are given in the table. Standard errors are in parentheses.

All regressions also include control variables of adult children's characteristics (age of the highest educated child, indicators for male, whether the highest educated children live with their parents and marital status), control variables of parents' characteristics (age and indicators for male, highest educational attainment, whether the parents are worse educated than their adult children, marital status, whether had any chronic diseases in the past 6 months and medical cost) and control variables of families' characteristics (number of family members and annual household income per capita).

∗∗∗p < .001.

We draw on VanderWeele's mediation analysis of causal inference to further directly verify the indirect effects of adult children's education on their parents' health through the paths that are influenced by adult children's education. Table A6 presents the results for indirect effects via intermediate variables of family economic hardship, family emotional support, health behavior (i.e., housework), and access to resources when tested separately. The results show that the majority of these indirect effects are statistically significant. Each intermediate variable could explain part of the estimated total effect of adult children's education on parents' health, with child-parent remote contact frequency of family emotional support and proper housework of health behavior explaining the most (11.76 % and 10.17 %, respectively) and family expenditure-to-income ratio of family economic hardship explaining the least (2.67 %).

3.6. Robustness checks

The following robustness checks are conducted to address potential analytical issues. Firstly, socioeconomic status and other unobserved characteristics of the counties where the parent lives may affect the parent's current health status and the time counties achieved the goal of universalizing compulsory education, thus violating the exclusion restriction assumption in IV estimation. Therefore, following Oster's (2019) approach, we test for omitted variables bias in current analyses (Table A7). Furthermore, we add county-level GDP per capita in the model to control for potential confounding (Table A8). The results in Table A7 and A8 suggest that the main findings shown in Table 2 do not omit important unobserved variables and are robust after adjusting for the county's socioeconomic characteristics.

Secondly, this study selects the highest-educated adult children aged over 18 in the family as the representative adult children sample, which may also generate substantial bias in the IV/2SLS regressions due to its focus on a single specific child. To address this issue, we conduct several robustness checks using alternative independent variables or subsamples: 1) controlling siblings' average education years and using all adult children's average education years (Panels A and B in Table A9); 3)using the subsample of cohabitating adult children (Panel C in Table A9); 4) using the lowest-educated adult children's education years (Panel D in Table A9); 5) limiting adult children's age for at least 23 years, to ensure that adult children had completed their education and avoid the influence of censoring (Panel E in Table A9); 6) limiting adult children sample to who lived in the same county from birth until the survey was conducted for analysis (Panel F in Table A9). Only approximately half of the adult children effectively answered whether the county where they were born was where their parents lived at the time of the survey, and 91.68 % of the adult children in this half had not moved from birth to the time of the survey. Regarding clustering, as both parents of the same highest-educated child may belong to the same family unit, we additionally control for family clusters(Panel G in Table A9). The results of the sensitivity analyses shown in Table A9 support the main findings by demonstrating robustness.

Thirdly, the independent variable in the study can be used as an ordered categorical variable as well. The categorical variable of adult children's education can be indicated by educational attainment. Parents' subjective memory scores can be used as an ordered three-point categorical variable, indicating whether their memory is good, average, or poor. Following Roodman's (2011) approach, we use both the variables of adult children's education and parents' subjective memory as categorical variables to test the robustness of the results presented in Table 2 by using the conditional mixed process (CMP) estimator with multilevel random effects and coefficients. Table A10 provides the IV/2SLS estimation results based on the CMP method, confirming that the results of all three estimations in Table 2 are robust.

4. Discussion

This study found a significant positive effect of adult children's education on middle-aged and elderly parents' health in China, and the effect may be nonlinear. In addition, we found heterogeneous effects of adult children's education on their parents' health over different regional socioeconomic conditions, individual socioeconomics, and living arrangements. Examining mechanisms based on resources and stressors, we found intergenerational transmission may be explained by parents' family economic hardship, adult children providing emotional support to their parents, their parents doing housework, and their parents' access to resources for health maintenance.

This study indicates that adult children's education can improve parents' health with a possible threshold effect. Our study is consistent with previous studies from developing countries, including China, which found that adult children's education can relieve parents' depression levels (Ma, 2019) and improve parents' physical activity function and subjective memory status (Pai et al., 2021; Yahirun et al., 2016, 2017) by using both OLS and IV/2SLS estimations. We estimated the effect of adult children's education on parents' health using the highest-educated adult children's education, which may slightly overestimate the effect compared to that using the average education of all adult children, as shown in our robustness check. However, although there is variation, even the most conservative results (using the education of the lowest-educated children) suggest a significant positive correlation. In addition, we provide new evidence that the intergenerational health advantage of adult children's education to parents may be nonlinear, becoming smaller when adult children's educational attainment exceeds middle school, which is similar to the previous studies focusing on the effect of individuals' education on their own health (Cutler & Lleras-Muney, 2006; Liu et al., 2018). Compared with illiterates who have never experienced academic education, completing basic education helps to develop basic cognitive and non-cognitive abilities in human capital, which could help to prevent family economic hardships and enable one to access basic health promotion resources which may make a prominent difference in their parents' health, but incremental benefits brought by subsequent higher-level education may be limited. These findings emphasize the important role of family members' human capital and intergenerational transmission in facing population aging and the need to strengthen intergenerational ties with parents when adult children's education exceeds a certain level.

Our heterogeneous results over regional socioeconomic development levels may contribute to resolving conflicting results from developing and developed countries. As reviewed before, studies in developing countries, such as China, widely observe the health advantages transmitted from adult children's education to their parents, which is not supported by recent studies in developed countries, such as England and Switzerland. Our heterogeneous analyses of different regions in one country confirm the hypotheses that regional socioeconomic development may be a condition for adult children's education to improve their parents' health. Compared with developed areas, the inadequate social welfare and social security systems in developing areas leave a larger proportion of parents largely dependent on their children's resources to increase their health status (Chen & Liu, 2009; Pei & Cong, 2020). Thus, adult children's higher education may be more beneficial to parents' health in less developed areas. We add new evidence that, given large regional differences in China, in the provinces with comparatively lower levels of economic development and social security protection, the improvement of the parents' health is more dependent on the children's education. The in-depth examination of heterogeneity from socioeconomic development conditions helps to understand the complex effect of adult children's education on parents' health and manifests the intergenerational health return of early life education investment in underdeveloped areas.

Different characteristics of parents and their adult children may also explain the varied effects of adult children's education on parents' health. First, we found that the effect may not vary depending on whether parents live in urban or rural areas. This differs from the results of provincial socioeconomic development, indicating that rural-urban disparity within a region may be narrower compared to regional disparity, or rural-urban migration is more prevalent within a region in China. The regional economic and social development remains unbalanced when comparing East and West China, as well as South and North China, while the urban-rural disparity within the same region is narrowing (Deng et al., 2022; Jiang, 2022; Kenneth, 2006; Liu et al., 2022, Liu et al., 2022).

Second, unlike other Chinese studies that reported no significant age group differences (Liu, 2021), we find that the health of younger parents may benefit more from their adult children's education, which may be because the younger parents are more able to take advantage of their offspring's resources and more sensitive to intervention at some age periods. This may be because other studies only compared differences in effects for parents above versus below the mid-50s, but not by refined age groups. The heterogeneous effects in terms of age category might be complex, which needs future study to confirm.

Third, our findings suggest that the intergenerational transmission of adult children's education to parents health do not vary on all demographics and socioeconomics variables, but those related to parents' health needs and adult children's ability to provide resources. The effect of adult children's education on their parents' health may not differ by parents' gender (Jiang, 2019) because adult children may not respond to their parents' gender but to their older parents' actual health demands, which do not differ by gender (Silverstein et al., 2006). Overall, we also find that the health advantages of adult children's education to parents' health may not vary from adult children's gender (Friedman & Mare, 2014), while in rural China, the male offspring's education is more beneficial for the parents' health than the female offspring's. In rural China's traditional patriarchal family culture, sons are expected to take more responsibility for parental care than daughters (Song et al., 2012). Addionally, we provide new evidence that the effect of adult children's education on their parents' health may not differ by parents' educational attainment (Lundborg & Majlesi, 2018), but may be more notable among parents who are less educated than their adult children (Sabater et al., 2020). On the one hand, parents with lower levels of education have more health needs and may be more dependent on their children's resources (Goldman & Cornwell, 2018; Liu, 2021). On the other hand, when the parents' educational attainment is similar to or higher than their children's, the resources and information that parents can acquire or learn from their children are limited due to the positive correlation between education and the ability to access the resources (Wang, 2024). This means that parents' needs and adult children's ability to provide resources are both essential.

Living arrangement is another important social condition to explain whether the effect of adult children's education on parents' health would vary. Adult children's education may have greater health effects on their parents for those living together with adult children than those living apart. This may be because long-term intimate contact caused by living with parents improves the opportunities for adult children and parents to interact and improves the chances that adult children and their parents help each other, and their education may easily benefit their parents' health (Caputo & Cagney, 2023; Teerawichitchainan et al., 2015). In addition, parents who live with their adult children tend to have poorer health and are more responsive to any supportive resources for health and well-being enhancement (Langenberger et al., 2025; Wang, 2022; Zimmer, 2008).

We find that the positive effect of adult children's education on their parents' health is more likely to be exerted through family economic hardship, family emotional support, housework support, and access to resources. Firstly, we provide new evidence that parents of highly educated adult children often face lower potential expenditure and debt risks, which may reduce family economic stress, thus making parents healthier (Conger et al., 1994; Zhang et al., 2020). Secondly, for resource-based mechanisms, we confirm that higher-educated adult children provide more emotional support and receive more housework support from their parents, thus making their parents healthier. Parents assist their children with housework, which leads to closer parent-child interactions and more frequent parent-child communications (Ma & Wen, 2016), thus stimulating parents' brains (Ellwardt et al., 2015) and helping prevent parents' memory decline (Zahodne et al., 2019). On the other hand, doing housework as a kind of physical activity integrated into daily life helps maintain proper physical activity levels to protect against elderly health decline (Adjei & Brand, 2018). In addition, similar to the prior evidence, the results reveal that higher-educated adult children will be more capable and conscious of helping their parents use the Internet, clean water, and clean fuels, and thus improve their parents' health (Wang, 2024). Our study also shows that adult children have a limited effect on parents' exercise, alcohol abstinence, and smoking cessation (Torres et al., 2022). Health behavior is more influenced by autonomous motivation thus, if parents' motivation for health behaviors is not naturally self-driven but controlled by external influences, they believe that physical activities are not something they want to do but are requested by their children, they may not truly adopt a healthy lifestyle (Solomon-Moore et al., 2017).

Unlike previous studies, we quantified the extent to which various pathways explain intergenerational transmission from adult children's education to parents' health. We find that child-parent remote contact frequency from family emotional support and proper housework from health behavior explain the largest proportions of the positive adult children's education effects on parents' health outcomes (11.8 % and 10.2 %, respectively), while the expenditure-to-income ratio from family economic hardship explains the least proportion (2.8 %). Overall, these intermediate variables can partially explain the effects of adult children's education on parents' health, which is similar to other pathways reported in studies focusing on health outcomes (1 %–25 %) (Hvidtfeldt et al., 2013; Kim & VanderWeele, 2019; Ohrnberger et al., 2017). Therefore, stress-based family economic hardship mechanisms and resource-based daily intervention mechanisms are important in the health benefits of adult children's education to parents. These intergenerational interactions and integrated health interventions are more widely integrated into the daily lives of the general population in developing societies, which may make it easier for the middle-aged and elderly to persevere and can naturally change parents' activities and exposures.

This study has its limitations. First, for the main analyses, this study only includes parents aged 45 years or above from recent surveys in China. For the mechanism analyses of family support, this study only reported the results for the elderly but not middle-aged parents, because the questionnaires related to family support were only for respondents aged 60 and above. The results in this study may not be directly generalized to other developed countries or younger parents. Second, some mechanism analyses in this study may suffer from bidirectional causation issues, like other similar studies. For instance, healthier parents may be more capable of doing physical activities, while weaker parents receive more frequent support from children (Yuan, Gong, & Han, 2020), which potentially overestimates the role of physical activities and family support. We applied both IV estimations and causal mediation analysis in our study to improve the mechanism analyses in this field to some extent. Future studies can utilize multi-wave data of prospective cohort studies or adopt a set of appropriate IV for different mediators to address such reverse causal bias.

In summary, this study further verifies the benefits of adult children's education to their parents' health improvement in China. We resolve the conflicts in the literature and provide a more coherent explanation for the complex relationship between adult children's education and their parents' health by systematically conducting nonlinear effect estimations and exploring the heterogeneity and mechanisms. Our findings suggest that education beyond a certain level has a weaker effect on enhancing intergenerational transmission. Education policies may be tailored based on socioeconomic development: less developed regions need to improve coverage and quality of preschool and basic education, while developed regions need to target children of low family socioeconomic status. Family policies that focus on facilitating multi-generational cohabitation and improving child-parent interactions may help address the social challenge of population aging. Health interventions for the middle-aged and elderly could emphasize the role of strategies that buffer economic stress or can be easily adhered to in daily life settings. Our findings have important implications for policymakers, professionals, and families in seeking strategies for healthy aging and health equity across generations.

CRediT authorship contribution statement

Huan He: Writing – review & editing, Data curation, Conceptualization. Lanxi Peng: Writing – review & editing, Writing – original draft, Data curation. Xuanhan Li: Methodology, Data curation.

Ethics statement

This is a secondary data analysis of publicly available, de-identified data; The database used in this study, CFPS, was reviewed and approved by the Institutional Review Board (IRB) of the Biomedical Ethics Committee of Peking University (IRB00001052-14010).

Declaration of interest

The authors declared no potential conflicts of interest concerning the study, authorship, and publication of the present paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

Data can be obtained by applying on the CFPS official website.