This study utilized publicly available data collected by the National School of Development at Peking University from the China Health and Retirement Longitudinal Study (CHARLS), an ongoing longitudinal investigation targeting individuals aged 45 and older in China. The CHARLS baseline survey employed a multi-stage probability proportional to size (PPS) random sampling method, incorporating implicit stratification based on indicators such as region, urban/rural characteristics, and GDP per capita. In the first stage, all counties and districts in the country, with the exception of Tibet, were ranked according to urban–rural attributes and GDP per capita within each of the eight regions. Subsequently, counties or districts were selected with a probability proportional to their population size. In the second stage, three secondary sampling units (village committees or neighborhood committees) were randomly chosen, again with probability proportional to population size, within each sampled county. Following the sampling process outlined above, the baseline sample of CHARLS was distributed across 450 villages and neighborhoods in 28 provinces and 150 districts and counties (15). Data were collected through face-to-face, computer-assisted personal interviews, which is a method that ensures high-quality data collection and participant engagement. CHARLS utilizes international data collection methods to ensure comparability, referencing studies such as the Health and Retirement Study (HRS) in the United States. As a result, CHARLS is representative both nationally and regionally. The national baseline survey was carried out in 2011 (wave 1), followed by subsequent follow-up surveys employing standardized questionnaires in 2013, 2015, 2018, and 2020 (waves 2–5). In each follow-up survey, the original participants were revisited, and new participants were recruited using the same sampling methods to compensate for attrition and maintain representativeness. Comprehensive details regarding the objectives, design, sampling procedures, and questionnaires of the CHARLS are available in other articles (15). The study was conducted following the Declaration of Helsinki, and approved by the Institutional Review Board of Peking University. The IRB approval number for the main household survey is 00001052–11,015, and all study subjects signed an informed consent form. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline for cohort studies (see Supplementary Table S1).

The follow-up period for the CHARLS data utilized in this study spanned from July 2018 to June 2020. Physical function was assessed as the primary outcome variable, while education level was examined as a moderating variable. The baseline population of CHARLS 2018 was subject to the following exclusion criteria: (1) individuals aged less than 45 years (n = 274); (2) individuals with at least one mental illness, including depression, schizophrenia, anxiety disorders, or bipolar disorder (n = 658); (3) individuals with at least one memory-related disease, such as Alzheimer’s disease or dementia (n = 642). After removing missing values (n = 6,959), a total of 11,283 samples were ultimately included. Figure 1 shows a comprehensive overview of the sample processing flow. A total of 9,497 samples from 2018 to 2020 were ultimately included in the construction of the Cox proportional hazards model and the analysis of moderation effects.

Physical function was measured by the Activities of Daily Living (ADL) and the Instrumental Activities of Daily Living (IADL) scales. ADLs include dressing, bathing, eating, getting in and out of bed, using the toilet, controlling bowel movements, and taking medication. ADLs include dressing, bathing or showering, eating, getting in and out of bed, using the toilet, and controlling urination and defecation: IADLs include doing household chores, preparing hot meals and shopping for groceries, making phone calls, taking medications and managing money. All questions have 4 options: (1) Do not have any difficulty; (2) Have difficulty but can still do it; (3) Have difficulty and need help; (4) Can not do it. If the answer to any of the ADLs and IADLs is one of the options 2, 3, or 4, it is defined as poor functional physical health; if the answer to any of the ADLs and IADLs is option 1, it is defined as well functional physical health. The scale has been used in research in China and abroad and has shown good reliability and validity.

Social capital is measured by four dimensions: social trust, social support, social participation, and reciprocity (16). According to previous studies, China is characterized as a “society of acquaintances,” particularly in rural areas where social capital plays a more significant role in the social economy due to the delayed development of formal institutions (17, 18). If a person’s place of birth is the same as their current place of residence, they may have developed deeper social connections and networks in that area, which can foster a higher level of social trust. Social trust is often linked to the closeness, familiarity, and interdependence within a community. In familiar social environments, individuals are more likely to trust one another due to shared cultural backgrounds, values, and experiences. Therefore, based on this context, place of birth serves as a proxy variable for social trust. Participants whose place of birth matches their current residence are classified as having high social trust, while those whose places differ are classified as having low social trust.

Social support is defined as the exchange of resources between at least two individuals, where at least one party perceives the exchange as aimed at enhancing the recipient’s well-being (19). Social support typically encompasses emotional, material, and informational support that an individual perceives from members of their social network, such as family, friends, and colleagues. Based on this definition, financial support was used to measure social support, setting social support as yes or no based on “whether or not they have received in-kind (buying food, clothes or other stuff regularly for you) or money (Regular payment includes providing living expenses, paying for water, electricity or telephone bill, paying for mortgage/rent or other forms of regular expenses) from their children or whether or not they have pension insurance (3). Due to the challenges of directly measuring emotional support and its inherent subjectivity, it was excluded from this study.

Social participation refers to the involvement in activities that provide interaction with others in a social network, fostering a sense of belonging and community (20–22). Reciprocity, on the other hand, involves the mutual exchange of support and assistance, where individuals not only receive help but also provide it to others, thus creating a supportive network (23). Social participation and reciprocity are judged by the participation in 8 activities in the past month: (1) Attended an educational or training course; (2) Interacted with friends; (3) Played Ma-Jong, played chess, played cards, or went to community club; (4) Went to a sport, social, or other kinds of the club; (5) Took part in a community-related organization; (6) Done voluntary or charity work, of caring for a sick or disabled adult who does not live with you; (7) Caring for a sick or disabled adult who does not live with you; (8) Provided help to family, friends, or neighbors who do not live with you. If at least one of the activities numbered 1 through 5 meets the specified conditions, it is classified as exhibiting Social Participation Behavior; otherwise, it is classified as lacking Social Participation. Similarly, if at least one of items 6, 7, or 8 meets the conditions, it is classified as demonstrating reciprocity; otherwise, it is classified as lacking reciprocity.

Based on the structure of the CHARLS questionnaire, we categorized educational attainment into four levels: below primary school, primary school, middle school, and high school or above. Individuals with less than a primary school education are classified as illiterate; otherwise, they are considered non-literate.

Referred to prior research, the covariates in this study were categorized into demographic variables and risk factors. Demographic variables included (1) Age, (2) Gender (Male or Female), (3) Marital Status (Divorced/Separated/Widowed/Never Married or Married/Cohabitated), (4) Residence (Rural or Urban), (5) Economic capacity (Yes or No) if participants have income apart from their family or friends’ support or pension. Risk factors are as follows: Sleep duration is categorized into three groups based on hours: insufficient sleep (less than or equal to 6 h), sufficient sleep (more than 6 h but less than or equal to 8 h), and excessive sleep (more than 8 h). If participants have at least one chronic disease such as Hypertension, Dyslipidaemia and Diabetes, they are classified as having chronic diseases, otherwise, they are classified as having no chronic disease. Smoking and drinking behaviors were self-reported by participants based on their current experience.

All statistical analyses were conducted using Stata 17.0 software (StataCorp LLC, College Station, TX, USA). First, descriptive statistics were used to summarize the baseline characteristics of the study participants. Continuous variables were expressed as mean ± standard deviation (SD), and categorical variables were presented as percentages. Continuous variables were expressed as mean ± standard deviation (SD), and categorical variables were presented as frequencies and percentages. Group comparisons (physical function) were performed using t-tests and chi-square. Group comparisons (physical function) were performed using t-tests and chi-square tests. To investigate the relationship between social capital, educational level, and physical functioning, we conducted a cohort study with 2018 (wave 4) participants as the baseline population. Social capital was categorized into 4 dimensions for risk-proportional Cox regression analysis. The exposure variables included social capital and educational level, while the outcome event was poor physical function. Cox regression, in contrast to logistic regression, was utilized to identify cause-and-effect relationships. To account for the influence of confounders on outcomes, we adjusted the model for demographic information, as well as participants’ smoking, drinking, sleep duration, disability, and chronic disease status.

Considering the potential influence of educational status, we conducted a moderation effect analysis using the Cox regression model described above. Moderation is a significant methodological concept in social science research and serves as a valuable tool for investigating the interplay among multiple variables (24). When the relationship between the dependent variable Y and the independent variable X (specifically, the magnitude and direction of the regression slope) varies in response to changes in a third variable Z, it is posited that Z exerts a moderating effect on the relationship between X and Y. In this study, social capital is designated as the primary effect variable, while education is employed as the moderating variable. The Cox moderated effect model is as follows:

Where h(t, X) is the risk function of an individual with covariate x at time t. t is the survival time, and X = (X₁, X₂,…, X_m) is the relevant factor that may influence survival time, commonly referred to as covariates. h0(t) is the risk function when all covariates are set to zero, known as the baseline hazard function. β = (β₁, β₂,…, β_m) is the regression coefficient of the Cox model, which consists of a set of regression parameters to be estimated.

In conducting a moderated effects analysis, centering (i.e., subtracting the mean from the variable) for independent and moderating variables is necessary to reduce the covariance between the interaction term and these variables. Centering corrects these biases without affecting model estimation. Thus, we centered the main and moderated effect variables to create a centered error correction moderated effects Cox model. Finally, we conducted subgroup analyses based on gender and residence type to examine their influence on the results. Due to the presence of missing values for covariates such as sleep duration, smoking, and alcohol consumption (n = 648), we used multiple imputation techniques to address the missing data and assessed the robustness of the results. All statistical tests were two-sided, with a p-value of less than 0.05 considered statistically significant.

The baseline characteristics of participants in the 2018 wave of the China Health and Retirement Longitudinal Study (CHARLS) are presented in Table 1. The study sample comprised 9,497 respondents, of whom 5,632 (59.3%) were women and 6,132 (64.6%) resided in rural areas. The mean age of participants with poor physical function was 65.84 years (SD = 9.937), whereas those with good physical function had a mean age of 61.51 years (SD = 9.302), indicating a statistically significant difference (t = 481.869, p < 0.001). Among participants with poor physical function, 65.8% were female and 34.2% were male. In contrast, 55.7% of those with good physical function were female and 44.3% were male, reflecting a significant gender difference (χ² = 90.706, p < 0.001). Furthermore, participants with poor physical function were more likely to have lower levels of educational attainment. Specifically, 62.6% had education levels below primary school, 19.6% had completed primary school, 12.7% had completed middle school, and only 5.1% had attained high school education or higher. In comparison, among those with good physical function, 43.1% had education levels below primary school, 24.4% had completed primary school, 21.7% had completed middle school, and 10.8% had attained high school education or higher (χ² = 362.881, p < 0.001).

The baseline data pertaining to the four dimensions of social capital indicate that 84.5% of participants exhibiting poor physical function reported low levels of social trust, in contrast to 82.6% of those demonstrating good physical function (χ² = 5.823, p = 0.016). Furthermore, 91.9% of participants with poor physical function received social support, compared to 88.5% of those with good physical function (χ² = 15.439, p < 0.001). Additionally, 82.9% of participants with poor physical function did not engage in social participation, whereas 75.8% of those with good physical function reported the same (χ² = 64.913, p < 0.001). Moreover, 87.5% of participants with poor physical function did not engage in reciprocity, compared to 83.9% of those with good physical function (χ² = 23.151, p < 0.001). These baseline characteristics underscore significant differences between participants with poor and good physical function across various demographic, social capital, and health-related factors.

To examine the relationship between social capital, educational level, and physical function, we conducted a series of Cox proportional hazards regression analyses. The results are presented in Table 2.

Initially, we investigated the relationship between social capital and physical function. The crude model revealed that social trust was not significantly associated with physical function (HR = 1.096, 95%CI: 0.999–1.203). After adjusting for age, sex, residence, and marital status (Model 1), the association remained non-significant (aHR = 1.090, 95%CI: 0.991–1.198). Further adjustments for smoking, drinking, sleep duration, disability, and chronic disease (Model 2) did not alter the results (aHR = 1.070, 95%CI: 0.973–1.176). In the crude model, social support was significantly associated with better physical function (HR = 1.209, 95 CI: 1.074–1.361). However, this association was not significant after adjusting for demographic variables in Model 1 (aHR = 0.910, 95%CI: 0.805–1.030) and remained non-significant in the fully adjusted model (aHR = 0.912, 95%CI: 0.806–1.032). Social participation was significantly associated with better physical function in all models. The crude model showed a strong association (HR = 0.745, 95%CI: 0.682–0.815), which remained significant after adjusting for demographic variables (Model 1: aHR = 0.826, 95%CI: 0.755–0.905) and in the fully adjusted model (aHR = 0.856, 95%CI: 0.675–0.809). Reciprocity was also significantly associated with better physical function. The crude model indicated a strong association (HR = 0.519, 95%CI: 0.466–0.578), which persisted after adjusting for demographic variables (Model 1: aHR = 0.668, 95%CI: 0.597–0.748) and in the fully adjusted model (aHR = 0.700, 95%CI: 0.626–0.784).

In the study examining the relationship between education and physical function, our findings indicate that participants with a primary school education demonstrated significantly better physical function compared to those with an education level below primary school across all models (Crude model: HR = 0.693, 95%CI: 0.635–0.756; Model 1: aHR = 0.805, 95%CI: 0.753–0.881; Model 2: aHR = 0.841, 95%CI: 0.769–0.921). For participants with a middle school education, the results showed a significant association with improved physical function in all models (Crude model: HR = 0.549, 95%CI: 0.495–0.609; Model 1: aHR = 0.715, 95%CI: 0.640–0.798; Model 2: aHR = 0.745, 95%CI: 0.667–0.832). Participants with a high school education or higher exhibited the most favorable physical function outcomes when compared to those with an education level below primary school (Crude model: HR = 0.466, 95%CI: 0.400–0.544; Model 1: aHR = 0.611, 95%CI: 0.520–0.718; Model 2: aHR = 0.661, 95%CI: 0.563–0.777).

The findings indicate that higher levels of social capital and education are associated with improved physical function among middle-aged and older adults. In particular, social participation and reciprocity demonstrated strong associations with physical function, emphasizing the significance of these dimensions of social capital. In addition, higher educational attainment consistently correlated with enhanced physical function.

The results of the moderating effect of education on the relationship between social capital and physical function are presented in Table 3. The interaction terms for social trust (aHR = 1.120, 95%CI: 1.012–1.240), social participation (aHR = 0.923, 95%CI: 0.840–0.980), and education were significant. However, the main effect of social trust was not significant when compared to social participation. While there is a direct joint effect of education and social trust, it does not suggest a moderating effect of education on the relationship between social trust and physical function.

The analysis demonstrates that education plays a significant moderating role in the relationship between specific dimensions of social participation and physical function. Higher levels of education enhance the positive impact of social participation on physical function. However, the benefits of social support and reciprocity on physical function remain consistent across various education levels.

To further investigate the relationship between social capital, education, and physical function, we conducted subgroup analyses based on gender (male vs. female) and place of residence (urban vs. rural). Table 4 provides a visual representation of the overall findings and highlights the differences in the subgroup analyses that are inconsistent with the overall results. In examining the relationship between social capital and physical functioning, we found that female participants’ social support contributed to maintaining good physical function (aHR = 0.857, 95%CI: 0.737–0.998). Additionally, social trust was identified as a risk factor for better physical function in the urban population compared to the overall participants (aHR = 1.330, 95%CI: 1.135–1.560). Social participation was not significantly associated with physical function in the rural population (Supplementary Tables S2–S5). When exploring the moderating effect of education on the relationship between social capital and physical function, we found that the positive moderating effect of education level did not extend to the rural population (Supplementary Tables S6–S9). In addition, we used multiple interpolation methods to address missing data. The overall conclusions, both before and after interpolation, remain consistent, demonstrating the robustness of our findings (Supplementary Table S10).