Intergenerational contact frequency and health outcomes among empty-nest older adults in China: a mediation analysis through physical exercise Authors: Hui Wang, Yan Jiang, Jinghang Cui Published: 2026 ABSTRACT This study examines the associations between intergenerational contact frequency and health outcomes among empty-nest older adults in China, and further investigates whether physical exercise serves as a behavioral pathway underlying these associations. Using data from the 2020 wave of the China Health and Retirement Longitudinal Study (CHARLS), we analyzed 3117 community-dwelling individuals aged 65 years and older who did not co-reside with their children. Intergenerational contact frequency was measured as a five-level ordinal indicator based on the frequency of in-person visits by children during the past year, using the maximum contact frequency across all children as the primary aggregation rule. Self-rated health and physical exercise frequency were analyzed using ordered logistic regression, while depressive symptoms measured by the CES-D-10 were modeled using negative binomial regression. Mediation effects were decomposed using the Karlson–Holm–Breen (KHB) method with 5000 bootstrap resamples. Higher intergenerational contact frequency was significantly associated with better self-rated health (OR = 1.18, 95% CI 1.10–1.27) and fewer depressive symptoms (IRR = 0.92, 95% CI 0.88–0.96), after adjusting for age, sex, education, residence, marital status, income, and chronic disease count. More frequent intergenerational contact was also positively associated with physical exercise frequency (β = 0.21, p < 0.001). Physical exercise partially mediated both associations. The indirect effect through exercise accounted for 18.6% of the association between intergenerational contact and self-rated health (indirect effect = 0.088, 95% CI 0.043–0.136), and 13.4% of the association with depressive symptoms (indirect effect = − 0.12, 95% CI − 0.20 to − 0.06). Subgroup analyses suggested that the mediating role of physical exercise was somewhat stronger among rural residents and male older adults. More frequent in-person contact between empty-nest older adults and their children is associated with better physical and mental health outcomes in China, with physical exercise partially explaining these associations. These findings suggest that intergenerational relationships may contribute to healthy aging partly by promoting more active lifestyles among older adults. Due to the cross-sectional design, causal relationships cannot be established. Longitudinal analyses using multiple waves of CHARLS data are needed to clarify temporal ordering and causal mechanisms. Public health strategies that integrate family engagement with community-based exercise promotion may help support healthy aging in China’s growing empty-nest older population. Introduction As populations worldwide age, the demographic shift toward older age groups is becoming increasingly pronounced. For example, United Nations projections indicate that by 2050 roughly one in six people globally will be aged 65 or older1. China, in particular, is experiencing a very rapid aging trend: official data show that by the end of 2022, nearly 19.9% of China’s population was aged 60 and above1, and 65+ older persons already constitute over 12% of the population. Projections suggest China’s 65+ population will more than double from 172 million (12.0%) in 2020 to 366 million (26.0%) by 20502. This “graying” of China’s population poses profound public health challenges, including greater burdens of chronic disease and increased demands on social and long-term care systems. In response, the Chinese government has emphasized policies on healthy aging, expanded pension and healthcare coverage, and new models of older adults’ care. One notable feature of China’s aging is the rise of the “empty-nest” phenomenon. Decades of rapid urbanization, internal migration of younger workers, and historical fertility constraints (such as the one-child policy) have meant that many older Chinese no longer live with their adult children. “Empty-nest” older adults are defined as those who live alone or only with a spouse, without children in the household1. By 2023, empty-nesters are projected to make up over half of all Chinese older adults. Although Chinese families traditionally valued multigenerational co-residence, recent trends show a shift toward “spouse-only” or solo living arrangements. Importantly, filial piety norms remain strong: children, whether nearby or far away, often strive to provide financial, instrumental, and emotional support to aging parents. Nonetheless, empty-nest older adults tend to face specific psychosocial and health challenges. Empirical studies in China have generally found that empty-nest older adults are more likely to experience loneliness, depression, and other psychological difficulties than their non-empty-nest peers3. Empty-nesters often report poorer self-rated health and are at elevated risk of chronic conditions, functional decline, and even increased mortality risk, apparently in part due to the sustained absence of day-to-day companionship and support from children4. Indeed, a recent national Chinese survey found that empty-nest older adults had higher unadjusted rates of dyslipidemia, diabetes, chronic lung disease, and heart disease compared to older adults living with children. These findings have raised concerns that without adequate support, empty-nest older adults’ physical and mental health could be compromised. In any case, understanding the determinants of health among China’s empty-nest population is a pressing public health priority. Family and social ties are widely recognized as determinants of older adults’ well-being. The convoy model of social relations5 conceptualizes an individual’s network as a set of close ties that travel with them across the life course; the structure and frequency of contact within this convoy shape health through both psychological and behavioral pathways. In China, intergenerational ties between parents and adult children form the most central convoy for older adults, reinforced by the cultural framework of filial piety (xiào, 孝), which prescribes that children demonstrate respect and care through visible, sustained behaviors toward aging parents. With the rise of the empty-nest phenomenon, in-person visits by adult children have become a particularly important observable form of intergenerational engagement. Empirical research in China indicates that older adults with more frequent intergenerational contact report better mental health and lower depressive symptoms than those with limited contact3,6. It is important to distinguish, however, between enacted contact behavior (e.g., visit frequency) and perceived emotional support (e.g., feeling cared for); these are conceptually distinct constructs that may operate through different mechanisms. The present study focuses on intergenerational contact frequency as an enacted, behaviorally observable indicator, while explicitly acknowledging that it does not equate to perceived support quality. Alongside psychosocial factors, lifestyle behaviors strongly influence older adults’ health. A large body of evidence shows that regular exercise helps prevent and manage chronic diseases, preserves mobility and functional independence, and enhances mental health in later life. For example, a recent systematic review and network meta-analysis reported that, across more than 100 randomized trials, exercise interventions significantly improved a range of physical function measures in older adults living in care facilities7. Such improvements are not limited to institutional settings: community-dwelling seniors also gain both physical and psychological benefits from exercise. In China, systematic reviews indicate that virtually any form of regular exercise (from walking to Tai Chi to resistance training) can improve older adults’ overall quality of life8. In particular, traditional mind–body exercises (e.g. Tai Chi, Baduanjin, Liuzijue) have been found to improve both physical health and mental health of older adults practitioners8. Empirical Chinese studies echo these findings: older Chinese adults who engage in leisure exercise report better self-rated health, greater confidence, fewer depressive symptoms, and better management of chronic conditions than sedentary peers9. Internationally, the physical activity guidelines emphasize at least 150 min per week of moderate exercise for seniors, because this lowers risks of cardiovascular disease, diabetes, dementia, and depression. Although the health benefits of regular physical exercise in later life are well established7,10, comparatively less is known about how intergenerational contact may shape older adults’ physical activity engagement. Three theoretical perspectives suggest plausible behavioral pathways. First, social learning theory11 holds that the physical presence of adult children creates occasions for joint or modeled activity—walking, agricultural or household work, accompanying parents to community spaces—that translate directly into measurable physical activity. Second, self-determination theory12 posits that autonomy-supportive relational climates foster intrinsic motivation for health behaviors; warm, regular intergenerational contact may provide such a climate without requiring explicit instruction. Third, behavioral activation perspectives suggest that scheduled visits may contribute to more regular social engagement patterns, anchoring activity in social occasions rather than relying on solitary discipline. We focus on physical exercise as the primary behavioral mediator for three reasons: it is directly observable and reliably measured in CHARLS; it has well-documented dose–response associations with multiple health outcomes in older adults; and it represents a tractable policy lever for healthy aging interventions, in contrast to less tractable mediators such as self-efficacy or cognitive function. The mediating role of physical exercise in linking intergenerational contact to health has not, to our knowledge, been directly tested within China’s empty-nest population—a gap this study addresses. The present study addresses these gaps. Using nationally representative data from CHARLS 2020 on Chinese older adults, we examine the associations of intergenerational contact frequency with self-rated health and depressive symptoms among empty-nest older adults, and we test whether physical exercise statistically mediates these associations. Our objectives are threefold: (1) to estimate the associations between in-person visit frequency from children and indicators of physical and mental health among empty-nest seniors; (2) to assess the extent to which physical exercise frequency accounts for these associations through a formal mediation decomposition; and (3) to draw out implications for China’s healthy aging policies. By integrating behavioral (contact) and lifestyle (exercise) variables in a single analytical framework, our work offers an integrated perspective on the social determinants of older adult health. Methodologically, our use of measurement-appropriate models (ordered logistic and negative binomial regression) and the Karlson–Holm–Breen decomposition for non-linear mediation strengthens inference relative to OLS-based approaches commonly applied to ordinal and count outcomes in this literature. The remainder of the paper is organized as follows. In “Materials and methods”section we describe the data source, key variables, and analytical methods. “Results” section presents the results of our descriptive and mediation analyses. “Discussion” section discusses the findings in light of existing literature, emphasizes the public health implications, and acknowledges limitations. Finally, “Conclusions” section concludes with a summary of contributions and suggestions for policy and future research. By elucidating how family and lifestyle factors combine to affect health, we aim to shed light on strategies to promote healthy, active aging among China’s empty-nest older adults. Materials and methods Samples and data sources This study utilizes data from the 2020 wave of the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative household survey of individuals aged 45 years and older in China, which included a total of 19,395 subjects13. CHARLS adopts a multistage probability sampling design and collects detailed information on demographics, family structure, health status, health care utilization, and socioeconomic conditions (https://charls.pku.edu.cn/). We restrict our analysis to respondents aged 65 years and above, and define “empty-nest older adults” as those who live alone or only with a spouse, without children in the household. After excluding respondents with missing information on the key variables, the final analytical sample consists of 3,117 individuals. The detailed exclusion process is shown in Fig. 1.Fig. 1Flow chart of the selection process of participants. Flow chart of the selection process of participants. Independent variable: intergenerational contact frequency The primary independent variable is intergenerational contact frequency, operationalized as the frequency of in-person visits by children to the respondent over the past year. This is an enacted, behaviorally observable indicator of intergenerational engagement and should not be interpreted as a direct measure of perceived emotional support, which is a conceptually distinct construct addressed by separate measurement traditions. We select in-person visit frequency for three reasons. First, it is the most directly observable form of intergenerational contact and is reliably reported in CHARLS. Second, within the convoy model framework5, the frequency of contact within close network ties is a structural feature of social networks that shapes health through behavioral pathways such as activity modeling, autonomy support, and routine structuring (see “Introduction” section). Third, in the empty-nest context where daily co-residence is absent, visit frequency reflects a substantively meaningful dimension of family engagement that other modes of contact (e.g., telephone calls, digital messaging, financial transfers) do not fully capture. Visit frequency is recorded in CHARLS as a five-level ordinal categorical variable: (1) almost never; (2) about once a year or less; (3) several times a year; (4) about once a month; (5) about once a week or more. We treat this ordinal scale as the primary measure, consistent with its categorical recording in CHARLS. CHARLS records the visit frequency of each child separately. We use the maximum visit frequency across all children as the primary aggregation rule, reasoning that any one child’s sustained high-frequency presence may suffice to provide the behavioral conditions relevant to our hypothesized pathway (e.g., periodic joint activity, scheduled routine, modeled engagement). Although the conceptual focus of our analysis is on enacted behavioral contact rather than perceived support quality, we acknowledge in our “Limitations” section that these are conceptually distinct constructs. Mediating variable: physical exercise frequency Physical activity was measured using the 2020 wave of the China Health and Retirement Longitudinal Study (CHARLS). Respondents reported the frequency and intensity of moderate or vigorous activities (e.g., heavy household work, running, calisthenics) in a typical week. Following CHARLS coding, these responses were combined into a five-level ordinal scale: Level 1 (inactive), Level 2 (low activity), Level 3 (moderate activity), Level 4 (high activity), and Level 5 (very high activity, including daily vigorous exercise or heavy labor). This measure captures both intensity and frequency, with higher levels indicating greater engagement14. When physical exercise frequency serves as a dependent variable, it is modeled as an ordinal outcome (“Analytical strategy” section). When it serves as a mediator in the Karlson–Holm–Breen decomposition, the ordinal scale is preserved, allowing the dose–response interpretation to be retained while respecting the measurement scale of the variable. Dependent variables: health outcomes Two types of health outcomes are examined. First, self-rated health (SRH) is measured on a five-point scale ranging from 1 (“poor”) to 5 (“excellent”). For ease of interpretation, we construct both the original ordinal variable and a binary indicator (good/very good vs. fair/poor). Second, psychological health is assessed using the 10-item Center for Epidemiologic Studies Depression Scale (CES-D10). We calculate a total score ranging from 0 to 30, with higher scores indicating more depressive symptoms. We employ a cutoff score of 10, a threshold commonly used in Chinese public health research, to classify individuals at risk of depression. Covariates A wide set of covariates is included to reduce confounding. Individual-level controls include age, gender, marital status, education, smoking, and drinking behaviors. Household-level controls include per capita household income, urban/rural residence, number of children, and whether the respondent co-resides with a spouse. Community-level controls (where available) such as medical facilities and neighborhood environment are also included. Analytical strategy Our analytical strategy proceeds in four steps. First, descriptive statistics summarize the characteristics of the analytical sample and the distributions of key variables across subgroups defined by intergenerational contact frequency. Spearman rank-order correlations are reported among the principal continuous and ordinal variables. Second, we estimate baseline regression models matched to the measurement scale of each outcome. Self-rated health (5-category ordinal) is modeled using ordered logistic regression. The proportional odds assumption is tested using the Brant test; where the assumption is violated, we report partial proportional odds models that relax the assumption for the violating covariates. Depressive symptoms (CES-D-10 score, count variable ranging 0–30) are modeled using negative binomial regression; the choice between Poisson and negative binomial is based on a likelihood-ratio test for overdispersion. Physical exercise frequency (5-level ordinal) is modeled using ordered logistic regression. All models adjust for age, sex, marital status, educational attainment, urban/rural residence, household income per capita, smoking status, drinking status, number of children, and chronic disease count. Third, we decompose the total association of intergenerational contact frequency with each health outcome into direct and indirect components using the Karlson–Holm–Breen (KHB) method15,16. The KHB method is well suited for mediation analysis in non-linear models because it yields comparable coefficients across nested specifications and correctly attributes the difference between total and direct associations to the mediator, addressing the rescaling issue that arises when applying the product-of-coefficients approach to ordered logit or negative binomial models. Confidence intervals for indirect components are obtained from bootstrap resampling with 5000 replications. Fourth, we conduct subgroup analyses (“Subgroup heterogeneity analyses” section) by re-estimating the mediation models stratified by urban/rural residence and sex to assess heterogeneity in the mediation pathway. All analyses are conducted in Stata 17 (StataCorp, College Station, TX). The KHB decomposition is implemented using the khb user-written command. Statistical significance is evaluated at the conventional two-sided α = 0.05 level. Reported confidence intervals are nominal 95% bootstrap percentile intervals. Results Descriptive statistics Table 1 summarizes the descriptive statistics of all variables used in the analysis, presenting the descriptive characteristics of the sample, including key demographic variables, intergenerational contact frequency, physical exercise frequency, and health outcomes. The final sample consisted of 3117 empty-nest older adults aged 65 and above. Among them, 52.3% were female, and the average age was 67.4 years (SD = 6.8). Approximately 61% lived in rural areas, and 72% were currently married. Most respondents had completed primary education or below.Table 1Descriptive statistics of key variables (N = 3117).VariableMeanSDMinMaxAge (years)67.46.86594Female (%)52.3Married (%)72.0Rural residence (%)61.0Education (Primary or below, %)68.0Intergenerational Contact Frequency (1–5)3.81.115Physical Exercise Frequency (1–5)2.91.315Self-rated Health (1–5)2.90.915Depression (CES-D score)9.35.8030 Table 1Descriptive statistics of key variables (N = 3117).VariableMeanSDMinMaxAge (years)67.46.86594Female (%)52.3Married (%)72.0Rural residence (%)61.0Education (Primary or below, %)68.0Intergenerational Contact Frequency (1–5)3.81.115Physical Exercise Frequency (1–5)2.91.315Self-rated Health (1–5)2.90.915Depression (CES-D score)9.35.8030 Descriptive statistics of key variables (N = 3117). For the core study variables, the mean intergenerational contact frequency score was 3.8 (SD = 1.1) on a 5-point ordinal scale, indicating a moderate-to-high frequency of in-person visits from children. The average physical exercise frequency score was 2.9 (SD = 1.3), suggesting a general engagement in low-to-moderate physical activity among the empty-nest elderly. Regarding health outcomes, the mean self-rated health score was 2.9 (SD = 0.9), indicating a slightly below-average perceived health status. In contrast, the mean CES-D-10 depression score was 9.3 (SD = 5.8), which is close to the clinical cutoff of 10, suggesting a notable baseline risk of depressive symptoms in this population. Preliminary t-test results revealed that urban older adults reported significantly higher contact frequency and physical activity levels compared to their rural counterparts (p < 0.01). Furthermore, female respondents reported significantly higher contact frequency from children (p < 0.05) but tended to engage less in vigorous exercise than males. These descriptive findings suggest potential associations among intergenerational contact frequency, physical exercise, and health outcomes, which are further examined in the subsequent regression and mediation analyses. Correlation analysis Table 2 shows the Spearman rank-order correlation coefficients among the key variables. Intergenerational contact frequency was positively correlated with physical exercise frequency (r = 0.24, p < 0.01) and self-rated health (r = 0.26, p < 0.01), but negatively correlated with depressive symptoms (r = -0.23, p < 0.01). Physical exercise frequency also exhibited a significant positive relationship with self-rated health (r = 0.35, p < 0.01) and a negative correlation with depression (r = -0.25, p < 0.01). Age was slightly and negatively correlated with contact frequency, exercise, and health status, while showing a weak positive association with depression. These correlations suggest that more frequent intergenerational contact and more frequent physical exercise are associated with better physical and mental health among older adults.Table 2Spearman rank-order correlation matrix of key variables.Variable123451. Intergenerational contact frequency12. Physical exercise frequency0.24***13. Self-rated health0.26***0.35***14. Depression (CES-D)–0.23***–0.25***–0.48***15. Age–0.10***–0.14***–0.11***0.09***1***p < 0.01, *p < 0.05. Table 2Spearman rank-order correlation matrix of key variables.Variable123451. Intergenerational contact frequency12. Physical exercise frequency0.24***13. Self-rated health0.26***0.35***14. Depression (CES-D)–0.23***–0.25***–0.48***15. Age–0.10***–0.14***–0.11***0.09***1***p < 0.01, *p < 0.05. Spearman rank-order correlation matrix of key variables. ***p < 0.01, *p < 0.05. Regression results Direct associations between intergenerational contact frequency and health outcomes We first estimated an ordered logistic regression model for self-rated health and a negative binomial regression model for depressive symptoms, examining the direct associations of intergenerational contact frequency with each outcome while controlling for demographic and socioeconomic covariates. As shown in Table 3, higher intergenerational contact frequency was positively associated with self-rated health (OR = 1.18, 95% CI 1.10–1.27, p < 0.001), after adjusting for age, sex, education, residence, marital status, income, and chronic disease count. Older adults receiving more frequent in-person visits from their children were more likely to report better overall health. Contact frequency was also negatively associated with depressive symptoms (IRR = 0.92, 95% CI 0.88–0.96, p < 0.001), indicating that respondents with more frequent intergenerational contact reported fewer depressive symptoms on the CES-D-10. The pattern of associations was robust across alternative model specifications.Table 3Ordered logistic regression of intergenerational contact frequency on self-rated health and negative binomial regression on depressive symptoms.VariableSelf-Rated Health OR (95% CI)p-valueDepressive Symptoms IRR (95% CI)p-valueIntergenerational Contact Frequency1.18 (1.10–1.27) < 0.0010.92 (0.88–0.96) < 0.001Age0.97 (0.96–0.99)0.0031.02 (1.01–1.04)0.002Male (ref: Female)1.11 (0.99–1.24)0.0650.89 (0.77–1.04)0.120Education (years)1.08 (1.03–1.13)0.0010.97 (0.93–1.01)0.080Urban residence (ref: Rural)1.13 (1.02–1.26)0.0250.91 (0.80–1.04)0.140Married (ref: Not married)1.16 (1.05–1.29)0.0040.84 (0.72–0.98)0.030Pseudo R20.120.15OR = odds ratio; IRR = incidence rate ratio; CI = confidence interval. Models adjusted for age, sex, education, residence, marital status, income, and chronic disease count. Table 3Ordered logistic regression of intergenerational contact frequency on self-rated health and negative binomial regression on depressive symptoms.VariableSelf-Rated Health OR (95% CI)p-valueDepressive Symptoms IRR (95% CI)p-valueIntergenerational Contact Frequency1.18 (1.10–1.27) < 0.0010.92 (0.88–0.96) < 0.001Age0.97 (0.96–0.99)0.0031.02 (1.01–1.04)0.002Male (ref: Female)1.11 (0.99–1.24)0.0650.89 (0.77–1.04)0.120Education (years)1.08 (1.03–1.13)0.0010.97 (0.93–1.01)0.080Urban residence (ref: Rural)1.13 (1.02–1.26)0.0250.91 (0.80–1.04)0.140Married (ref: Not married)1.16 (1.05–1.29)0.0040.84 (0.72–0.98)0.030Pseudo R20.120.15OR = odds ratio; IRR = incidence rate ratio; CI = confidence interval. Models adjusted for age, sex, education, residence, marital status, income, and chronic disease count. Ordered logistic regression of intergenerational contact frequency on self-rated health and negative binomial regression on depressive symptoms. OR = odds ratio; IRR = incidence rate ratio; CI = confidence interval. Models adjusted for age, sex, education, residence, marital status, income, and chronic disease count. These direct associations are consistent with the convoy model of social relations5, which posits that the frequency and structure of close network contacts shape older adults’ well-being through both behavioral and psychological pathways. They are also consistent with cross-cultural evidence that intergenerational contact is a particularly important component of older adults’ social convoys in collectivist contexts. In the following sections, we examine whether physical exercise statistically accounts for part of these associations, consistent with the behavioral pathway specified in “Introduction” section. Association between intergenerational contact frequency and physical exercise To examine whether intergenerational contact frequency is associated with physical activity engagement among empty-nest older adults, we estimated an ordered logistic regression of exercise frequency on contact frequency, controlling for demographic and socioeconomic characteristics. As shown in Table 4, this model employed exercise frequency as the dependent variable and intergenerational contact frequency as the main predictor, while controlling for demographic and socioeconomic characteristics.Table 4Ordered logistic regression for the association between intergenerational contact frequency and physical exercise frequency.VariableOR (95% CI)p valueIntergenerational Contact Frequency1.23 (1.16–1.31) < 0.001Female (ref: Male)0.89 (0.82–0.97)0.006Age0.95 (0.92–0.98)0.001Education (years)1.09 (1.03–1.15)0.004Married (ref: Not married)1.04 (0.98–1.12)0.218Rural residence (ref: Urban)0.86 (0.79–0.93) < 0.001Pseudo R20.09Observations3,117***p < 0.001, **p < 0.01, *p < 0.05. Table 4Ordered logistic regression for the association between intergenerational contact frequency and physical exercise frequency.VariableOR (95% CI)p valueIntergenerational Contact Frequency1.23 (1.16–1.31) < 0.001Female (ref: Male)0.89 (0.82–0.97)0.006Age0.95 (0.92–0.98)0.001Education (years)1.09 (1.03–1.15)0.004Married (ref: Not married)1.04 (0.98–1.12)0.218Rural residence (ref: Urban)0.86 (0.79–0.93) < 0.001Pseudo R20.09Observations3,117***p < 0.001, **p < 0.01, *p < 0.05. Ordered logistic regression for the association between intergenerational contact frequency and physical exercise frequency. ***p < 0.001, **p < 0.01, *p < 0.05. Higher intergenerational contact frequency was positively associated with physical exercise frequency (OR = 1.23, 95% CI 1.16–1.31, p < 0.001). Older adults receiving more frequent in-person visits from children were more likely to engage in moderate or vigorous physical activities. The association remained robust after adjustment for sex, age, education, marital status, place of residence, income, and chronic disease count. These findings provide preliminary support for the hypothesized behavioral pathway from intergenerational contact to physical activity (“Introduction” section), consistent with social learning, autonomy-support, and behavioral activation perspectives. The formal mediation decomposition is presented in “Mediation analysis results” section. Mediation analysis results To test whether physical exercise frequency statistically mediates the associations of intergenerational contact frequency with health outcomes, we applied the Karlson–Holm–Breen (KHB) decomposition with bootstrap confidence intervals (5000 resamples), using ordered logistic regression for both the mediator and the self-rated health outcome, and negative binomial regression for the depressive symptoms outcome. Intergenerational contact frequency served as the independent variable, physical exercise frequency as the mediator, and each health outcome (self-rated health, CES-D-10 score) as the dependent variable in separate models. Table 5 presents the decomposed components for self-rated health. The total association between intergenerational contact frequency and self-rated health was 0.47 (95% CI 0.35–0.58, p < 0.001), which decomposed into a direct component of 0.38 (95% CI 0.28–0.49, p < 0.001) and an indirect component through physical exercise of 0.088 (95% CI 0.043–0.136, p < 0.001). The indirect component accounted for approximately 18.6% of the total association, consistent with partial statistical mediation.Table 5KHB Decomposition of Mediation Analysis: Intergenerational Contact Frequency → Physical Exercise → Self-Rated Health.ComponentCoefficient95% CIp-valueTotal Effect0.470.35–0.58 < 0.001Direct Effect0.380.28–0.49 < 0.001Indirect Effect0.0880.043–0.136 < 0.001Mediated Percentage18.6%––***p < 0.001, **p < 0.01, *p < 0.05. Table 5KHB Decomposition of Mediation Analysis: Intergenerational Contact Frequency → Physical Exercise → Self-Rated Health.ComponentCoefficient95% CIp-valueTotal Effect0.470.35–0.58 < 0.001Direct Effect0.380.28–0.49 < 0.001Indirect Effect0.0880.043–0.136 < 0.001Mediated Percentage18.6%––***p < 0.001, **p < 0.01, *p < 0.05. KHB Decomposition of Mediation Analysis: Intergenerational Contact Frequency → Physical Exercise → Self-Rated Health. ***p < 0.001, **p < 0.01, *p < 0.05. The KHB decomposition indicates that intergenerational contact frequency is associated with self-rated health both through a direct pathway and indirectly through its association with physical exercise. The indirect component is statistically distinguishable from zero, supporting the hypothesized behavioral pathway. We caution, however, that statistical mediation in cross-sectional data does not establish causal mediation: the analysis assumes a directional ordering that the data themselves cannot verify. Subgroup heterogeneity analyses To ensure the robustness of the main findings, we re-estimated the mediation model using depressive symptoms (CES-D-10 score) as an alternative indicator of health, applying KHB decomposition with negative binomial regression for the outcome and ordered logistic regression for the mediator. The results remained consistent with the main analysis (Table 6): the total association between intergenerational contact frequency and depressive symptoms was − 0.89 (95% CI − 1.12 to − 0.66, p < 0.001), decomposing into a direct component of − 0.77 (95% CI − 0.99 to − 0.55, p < 0.001) and an indirect component through physical exercise of − 0.12 (95% CI − 0.20 to − 0.06, p < 0.001). The indirect component accounted for approximately 13.4% of the total association, confirming that the mediation pattern is not specific to a single health indicator.Table 6KHB Decomposition of Mediation Analysis: Intergenerational Contact Frequency → Physical Exercise → Depressive Symptoms.ComponentCoefficient95% CIp-valueTotal Effect − 0.89 − 1.12 to − 0.66 < 0.001Direct Effect − 0.77 − 0.99 to − 0.55 < 0.001Indirect Effect − 0.12 − 0.20 to − 0.06 < 0.001Mediated Percentage13.4%–– Table 6KHB Decomposition of Mediation Analysis: Intergenerational Contact Frequency → Physical Exercise → Depressive Symptoms.ComponentCoefficient95% CIp-valueTotal Effect − 0.89 − 1.12 to − 0.66 < 0.001Direct Effect − 0.77 − 0.99 to − 0.55 < 0.001Indirect Effect − 0.12 − 0.20 to − 0.06 < 0.001Mediated Percentage13.4%–– KHB Decomposition of Mediation Analysis: Intergenerational Contact Frequency → Physical Exercise → Depressive Symptoms. We further examined whether the mediation pattern held across different demographic groups. Separate KHB analyses were conducted for urban and rural residents as well as for males and females, with physical exercise frequency serving as the mediator and intergenerational contact frequency as the independent variable. Bootstrap resampling (5000 iterations) was used to estimate indirect components for each subgroup. Figure 2 visually presents the direct, indirect, and total associations of intergenerational contact frequency with depressive symptoms across different subgroups (urban vs. rural; male vs. female). Each point represents the estimated coefficient, and the vertical line represents the 95% confidence interval. Since the depressive symptom score is negatively signed (a lower score indicates better mental health), a larger negative coefficient signifies a stronger inverse association. The pattern is consistent across all groups, but with notable heterogeneity: the indirect pathway through physical exercise is strongest for rural residents and male older adults.Fig. 2Comparison of indirect components of intergenerational contact frequency on depressive symptoms via physical exercise across subgroups. Comparison of indirect components of intergenerational contact frequency on depressive symptoms via physical exercise across subgroups. Discussion Comparison with prior literature Our findings are broadly consistent with existing literature linking intergenerational contact to mental and physical health outcomes among Chinese older adults6,14,18,19. A particularly instructive contrast emerges with Luo et al.17, who, using 2014 CHARLS data, found that life and financial support from adult children were associated with lower depression and loneliness, whereas measures of perceived emotional support showed no statistically significant association. The authors attributed this divergence to an expectation-discrepancy mechanism whereby older adults’ high expectations of perceived support may produce a “sense of loss” when those expectations are unmet. Our results, by focusing on enacted in-person contact behavior rather than perceived support, reinforce the conceptual distinction between these two dimensions and suggest that behavioral indicators, being less subject to expectation-discrepancy effects, may offer a more robust empirical signal in cross-sectional analyses of Chinese empty-nest populations18,19. Mechanisms underlying the contact–exercise–health association The observed association between intergenerational contact frequency and health outcomes—partially channeled through physical exercise—may be interpreted through several complementary theoretical perspectives that highlight different facets of how close-network relationships may shape later-life behavior. Within the convoy model of social relations5,20–22, the frequency of in-person contact may reflect a structural feature of close ties that facilitates engagement in health-promoting activities. From a social learning perspective11,23, visits from adult children may create occasions for joint or modeled activity—accompanying parents on walks, engaging in shared household or agricultural tasks—that may translate observable contact into measurable physical activity. Self-determination theory12 offers a complementary motivational interpretation: warm, recurrent intergenerational contact may reflect an autonomy-supportive relational context that facilitates sustained exercise engagement, potentially through the motivational processes specified in that framework. While our cross-sectional design does not permit adjudication among these mechanisms, the partial mediation through physical exercise—accounting for 18.6% of the contact–SRH association and 13.4% of the contact–depression association—is consistent with a behavioral pathway broadly congruent with all three perspectives20–24. Heterogeneity by residence and sex The indirect pathway through physical exercise was somewhat stronger among rural residents and male older adults. Several contextual factors may help interpret these patterns. In rural settings, baseline physical activity is more closely tied to embodied daily routines—agricultural labor, walking, household tasks—so the marginal behavioral response to a child’s visit may translate more readily into measurable activity than in urban settings, where exercise tends to be more facility-dependent. Rural support networks are also typically thinner; with fewer formal eldercare resources and community organizations, intergenerational contact may represent a relatively higher-leverage input to behavioral activation. For male older adults, the pattern may reflect gendered patterns of social engagement in China: older Chinese women typically maintain denser non-familial networks (e.g., neighborhood groups, square-dancing communities) that may provide alternative pathways to physical activity engagement, whereas older men’s support sources tend to be more concentrated within the immediate family. Father–son co-activity in rural contexts may also more readily translate intergenerational contact into shared physical activity. These heterogeneous patterns suggest that the behavioral pathway from family contact to exercise is not uniformly available across subgroups; complementary investments in community-based exercise infrastructure may be particularly relevant in urban settings and for older women, where this pathway appears comparatively weaker. Methodological considerations Our analysis addresses two methodological limitations common in this literature. By modeling ordinal self-rated health and ordinal exercise frequency with ordered logistic regression—rather than treating these scales as continuous in OLS—we respect the categorical structure of these measures and avoid potentially distorted effect-size interpretations. By applying the Karlson–Holm–Breen decomposition with bootstrap inference, we address the rescaling problem inherent to mediation analysis in nonlinear models, where the conventional product-of-coefficients approach does not produce directly comparable estimates across nested specifications. These methodological choices yield an interpretable decomposition of the total association into direct and indirect components without distorting the underlying probability structure, and may provide a useful template for future mediation analyses using ordinal and count outcomes in survey data on older adults. Limitations Several limitations warrant attention. First, the cross-sectional design of CHARLS 2020 precludes causal inference. Our mediation framework assumes a directional pathway from intergenerational contact frequency to physical exercise to health outcomes, but reverse causal pathways are plausible at multiple points. Older adults with poorer self-rated health may elicit more frequent visits from children—for example, when children travel home in response to a parent’s declining condition; in this scenario, the observed positive association between contact and health could partly reflect children’s caregiving response rather than contact-driven behavioral activation. Similarly, depressive symptoms may both diminish parents’ capacity for physical activity and alter the contact behavior of adult children, who may withdraw in distress or, alternatively, intensify visits in concern. Selection into physical activity is also non-random: older adults with better baseline health are more able to exercise, generating a simultaneity bias between exercise and self-rated health that cross-sectional analyses cannot resolve. Accordingly, all associations reported here should be interpreted as conditional statistical relationships consistent with the hypothesized pathway, not as causal effects. Second, our primary measure captures in-person contact frequency rather than perceived emotional support. Although Chinese filial piety norms suggest meaningful overlap between enacted contact and perceived support, the two are conceptually distinct constructs operating through different theoretical mechanisms: enacted contact may or may not generate subjective feelings of being cared for, and perceived support may exist independently of physical co-presence (e.g., through phone, digital contact, or financial transfers). We anchor our interpretation in observable contact behavior rather than perceived support quality and have updated the theoretical framework accordingly (“Introduction” section), but caution that future studies should incorporate validated perceived-support instruments (e.g., the Multidimensional Scale of Perceived Social Support, MSPSS) alongside contact-frequency measures to clarify which dimension of intergenerational support drives the observed associations. Third, all measures are self-reported and may be subject to recall bias and social desirability bias, particularly for self-assessed health and frequency reports. Objective measures of physical activity (e.g., accelerometry) and multi-item scales capturing distinct dimensions of intergenerational support (emotional, instrumental, informational) would enrich future analyses. In addition, the CHARLS physical activity measure combines leisure-oriented exercise with labor-related activities (e.g., household or agricultural work), particularly in rural settings. This may blur distinctions between voluntary health-promoting exercise and necessity-driven physical exertion, and limits our ability to identify which specific forms of activity drive the observed associations25,26. Fourth, we focused on empty-nest older adults aged 65 and above; results may not generalize to non-empty-nest older adults, younger cohorts, or populations outside China. Fifth, other potential mediators—including social engagement, cognitive function, self-efficacy, sleep quality, and diet—were not examined in the present analysis. Future research should test these alternative pathways in parallel, ideally within a unified framework that permits comparison of mediation strength across behavioral and psychological channels25,26. Directions for future research To address several of the limitations noted above—particularly the cross-sectional design and reverse causality concerns—we propose specific extensions that leverage the longitudinal structure of CHARLS. The CHARLS panel spans five waves (2011, 2013, 2015, 2018, 2020), permitting estimation of cross-lagged panel models in which intergenerational contact frequency at wave t − 1 predicts physical exercise at wave t, and physical exercise at wave t − 1 predicts health outcomes at wave t + 1, while simultaneously controlling for prior values of each variable. Such designs can distinguish between contact-to-exercise pathways and health-to-contact pathways operating in the reverse direction. Alternatively, individual fixed-effects models leveraging within-person variation across waves can rule out time-invariant confounders such as personality traits, baseline family closeness, and stable health behaviors. For identification of causal effects, instrumental variable approaches—using, for example, community-level exercise facility access, regional variation in family policy, or geographic distance between parental and child residence as instruments—represent a promising direction. Beyond longitudinal extensions, future work should incorporate objective measures of physical activity (accelerometry, step counts), multidimensional indicators of intergenerational support spanning emotional, instrumental, and informational dimensions, and tests of alternative mediators including social engagement, cognitive engagement, sleep quality, and diet. Mixed-method designs combining survey analysis with qualitative interviews would also help clarify the subjective experience underlying the behavioral patterns we document, particularly the cultural meanings of in-person visits in the context of contemporary Chinese filial piety. Conclusions This study examined the associations of intergenerational contact frequency with self-rated health and depressive symptoms among empty-nest older adults in China and tested whether these associations are statistically mediated by physical exercise. More frequent in-person visits by children were associated with better self-rated health and fewer depressive symptoms. Physical exercise frequency partially mediated both associations, with the indirect pathway through exercise somewhat stronger among rural residents and male older adults. These patterns are consistent with theoretical frameworks—the convoy model, social learning theory, and self-determination theory—that emphasize the behavioral pathways linking close-network contact to health in later life. The findings have implications for healthy aging policy in China: programs that combine family engagement with community-based exercise opportunities may contribute to older adult well-being, with particular relevance for rural and male subgroups where the behavioral pathway is strongest. We emphasize, however, that the cross-sectional design precludes causal inference; reverse causality and selection effects cannot be ruled out, and the contact measure captures enacted behavior rather than perceived support. Future research should leverage the longitudinal structure of CHARLS to estimate cross-lagged and fixed-effects models, and should incorporate objective measures of physical activity and multidimensional support indicators.