Wealth gradients in healthy aging: evidence from the 2011 and 2013 waves of the China Health and Retirement Longitudinal Study

Abstract

Background

It is crucial to understand how individuals accumulate wealth over their lifespan and explain the relation with the changes in health by age cohort. This study examines wealth-to-health causality as an explanation for the health-wealth gradient.

Methods

Using the 2011 and 2013 waves of individuals aged 45 and above from the China Health and Retirement Longitudinal Study, this study tests for causality by employing various econometric models and statistical strategies, conducting longitudinal and cross-sectional data analyses. The validity of the causal mechanisms is assessed by comparing the fitting results from different models.

Results

The fixed effects model reveals that a 1% increase in wealth results in a statistically significant decrease in the constructed health index by approximately 0.00032 units, at a 1% significance level. Similarly, in the instrumental variables model, the coefficient for wealth is -0.229 and is also significant at the 1% level. The results for other control variables, such as demographic, socioeconomic status, geographical, and childhood health status, remain stable and align with expectations across different models.

Conclusion

This study demonstrates a causal relationship between wealth and health, with permanent income as a key determinant. It highlights the need for poverty alleviation policies that promote long-term wealth accumulation, strengthen social welfare, and consider the indirect health effects of economic and environmental reforms.

Text box 1. Contributions to the literature
1. National survey data shows a causal pathway from wealth to health within the middle-aged and elderly population in China.
2. Net wealth and income play different roles in health variations.
3. Highlights the role of stable, long-term income and strengthened social welfare systems in addressing wealth-driven health variations and safeguarding health among impoverished populations.

Background

The phenomenon of population aging has become a global trend. By 2030, 1 in 6 people in the world will be aged 60 years or over. At this time the share of the population aged 60 years and over will increase from 1 billion in 2020 to 1.4 billion. By 2050, the world’s population of people aged 60 years and older will double (2.1 billion) [1]. China faces a more severe aging situation. Its’ elderly population aged 60 and above reached 296.97 million by the end of 2023, accounting for 21.1% of the total population [2]. It is also projected that the elderly population aged 60 and above in China will increase from 16% to 35% between 2017 and 2050 [3]. The rapid growth in the scale of population aging, coupled with declining health conditions and increasing demand for services to manage non-communicable diseases, disabilities, and caregiving, poses even more significant challenges to the already strained healthcare services and social security systems [4]. Therefore, achieving “healthy aging” has become an essential issue for policy-making, and health and social care research.

Over the past two decades, literature has shown a transparent health-wealth gradient, indicating that health tends to improve with increasing wealth [5]. This correlation was first recognized in highly industrialized countries [6, 7, 8] and has gradually been observed in some developing countries and regions [9, 10]. As life progresses, this association will impact various aspects of individual lives. Explaining the relationship between wealth and health, including the causal pathways between the two, is a central focus in current research on healthy aging [11]. Economists and social researchers have summarized their viewpoints into three possibilities. The first possibility is that economic resources influence health status, which Grossman’s health demand model proposed in the dissertation [12]. For example, individuals with higher economic resources can afford better healthcare services, live in better community environments, and consequently maintain better health. This viewpoint has received the most empirical support [13, 14, 15]. The second possibility is that health levels influence economic resources. Healthier individuals tend to work longer than those in poor health, leading to higher incomes and accumulated wealth. With the establishment of large-scale longitudinal databases, some evidence has been accumulated to support this reverse causal relationship [16, 17]. For instance, this possibility often occurs among groups with poor mental well-being. As their health deteriorates, these individuals are more likely to exit the labor market earlier, and some may lose their sources of income partially or entirely [18]. The third possibility suggests that there may be a third factor that simultaneously determines wealth and health. For example, according to time preference theory, individuals with lower discount rates are more inclined to invest in human capital, yielding future economic benefits and improved health [19].

Exploring the essence of the health-wealth gradient is crucial for understanding individual wealth accumulation processes throughout the lifespan and explaining the age-health cohort changes that arise. Furthermore, distinguishing these relationships helps to understand the underlying causes of health disparities and design policies to improve individual wealth, health, and well-being [20]. Suppose the widely accepted wealth-to-health causality holds. In that case, the key to improving health disparities lies in poverty alleviation rather than solely increasing support for healthcare among the impoverished population. Suppose the health-to-wealth causality is indeed true. In that case, health not only directly impacts labor participation rates and wealth accumulation, but it also has a negative relationship with the utilization of social security in old age. Therefore, the series of welfare benefits provided by the country to improve national health can also directly stimulate the country’s macroeconomic development. As a result, the impact of national policies on healthcare expenditure and healthcare service efficiency often undergoes transformations based on different understandings of the wealth-health gradient.

Many studies have been conducted on this gradient relationship. In addition to the essential characteristics summarized above, some aspects are worth further exploring. Firstly, most studies have used a single-dimensional measurement in selecting health indicators, focused on specific groups with certain disease states, or used general health indicators without considering the elements representing various dimensions of health comprehensively. However, using these traditional health measurements may conceal the actual inequality status of specific groups in the economic environment [21]. For example, the correlation among different aspects of health state, individuals with certain chronic diseases may also have functional disability issues and experience a lower quality of life. Secondly, although current large-scale health surveys generally include wealth-related information in their collection scope, the effectiveness is not ideal. On the one hand, collecting wealth data is a professional and challenging task, and finance topics can be very personal, with individual assets and debt values constantly changing, requiring the investigators to possess professional evaluation skills [22]. On the other hand, wealth is one of several related components of socioeconomic status, including education, income, and occupation type [23]. Neglecting the interplay between these socioeconomic factors can lead to significant biases in results. Lastly, some scholars have started to focus on the stability of the causal relationship between wealth and health, and have suggested possible heterogeneity among populations in conclusions [24]. For example, the gradient relationship may vary throughout the life cycle. Individuals in early life and childhood may have a unidirectional and unquestionable impact of family economic status on health due to their inability to care for themselves [25]. Additionally, the gradient is more likely to be observed in developed countries and has some connection with national economic resources and income distribution systems [23]. However, in some developing countries, the relationship may no longer be significant [9]. Furthermore, a large portion of evidence from developed countries comes from the United States and Europe, and the applicability of the conclusions to other regions may be limited. Finally, considering the endogeneity issues caused by omitted variables, measurement errors, sample selection, and inappropriate research design, it is worthwhile to re-examine the causality test of the relationship between wealth and health in the context of Chinese samples. Therefore, this study uses the China Health and Retirement Longitudinal Study (CHARLS) 2011 and 2013 data to analyze the impact of wealth on the health status of the middle-aged and elderly population. Considering the rigor of causal inference, this study employs both static panel models and instrumental variable methods to identify the wealth-to-health causality and robustness tests are conducted to confirm the core results.

Methods

Data and sample

The data used in this study are derived from CHARLS, organized and implemented by the National School of Development at Peking University. The survey targeted individuals aged 45 and above, including their spouses. The survey comprises over ten thousand households and nearly twenty thousand individuals, providing a geographically representative and large-scale sample. It covers various aspects such as population characteristics, education, health, and medical care, which can offer data support for interdisciplinary studies in economics, sociology, and other fields [26]. The research sample represents the retired population in China during a critical period of their lives, characterized by stabilizing their assets, income, and long-term health conditions as they enter or have entered the aging stage.

During the 2011 and 2013 surveys, 15,186 individuals participated in both periods, with 2,522 only participants in 2011 and 3,426 new participants in 2013. Regarding the critical variables in this study, the two surveys remain consistent and do not show any significant structural changes. When conducting data cleaning, this study adopted a method similar to the Health and Retirement Study to interpolate missing financial variables in the CHARLS dataset, ensuring the optimal enhancement of original income and wealth-related information [27, 28]. This study selected individuals aged 45 and above as the research sample, excluding observations with missing key variables. The resulting sample sizes are 27,381 for panel data analysis and 10,954 for cross-sectional data analysis. The study utilized the 2011 and 2013 waves of the CHARLS dataset due to the availability of comprehensive and consistent measures of wealth and health, which are essential for examining the causal relationship between the two.

Measurement

Health status index

This study employed various health measurements to identify the characteristics between single-dimensional and multi-dimensional health indicators. CHARLS contains sufficient information that can be used to describe individual health status. We selected 34 questions that reflect the following six single-dimensional aspects: (i) self-rated health status, indicating whether the respondents consider themselves to be in poor health; (ii) difficulties in instrumental activities of daily living (IADL), including housework, cooking, shopping, taking medication, and handling money, totaling five activities; (iii) difficulties in activities of daily living (ADL), including dressing, bathing, eating, getting in and out of bed, using the toilet, and controlling urination and defecation, totaling six activities; (iv) physical functional impairments, such as walking 1 km, sitting or standing for a long time, climbing stairs continuously, bending, kneeling, or squatting, stretching arms upwards along the shoulders, lifting a 10-pound object, and picking up a small coin, totaling seven activities; (v) depressive mood, measured using the CESD-10 scale (The Center for Epidemiologic Studies Depression Scale); and (vi) physician-diagnosed chronic diseases, including hypertension, dyslipidemia, diabetes, cancer, chronic lung diseases, liver diseases, heart diseases, strokes, kidney diseases, digestive system diseases, psychiatric disorders, memory-related diseases, arthritis, and asthma, totaling 14 diseases.

After standardizing the 34 data items, we used principal component analysis to reduce dimensionality and took the first principal component as the health status index [29, 30]. The first principal component is the weighted average of these health indicators, where the weights are chosen to maximize the proportion of the variance of indicators that this weighted average can explain. This index follows a standard normal distribution, where a higher score indicates poorer individual health status. Our health index is based on the deficit accumulation model, which assumes that individual health deteriorates as deficits increase (e.g., chronic diseases, physical decline, mental health issues) [31]. The index for different years was computed separately. The weights of various health indicators are shown in Table 1, and the structure of each index constructed from two periods is relatively consistent.

Table 1 Weighting of indicators for constructing health index using first principal component scores in 2011 and 2013 waves from the health status and functioning module

Wealth variables

This study referred to the definition of wealth assets by CHARLS and categorized them into six parts, as shown in Table 2: liquid assets, debts, real estate assets, household durable goods, productive assets, and land assets. The financial data from these two periods are stable, with real estate assets dominating the absolute proportion [32]. Based on the attributes of assets, the CHARLS questionnaire measures at both the household and individual levels. In calculating the total wealth value, this study first consolidated the individual assets obtained from the personal questionnaire module. Then, the jointly-owned assets at the household level were distributed to individuals based on the proportions of each household member’s ownership. If the distribution proportions were unknown, this asset was evenly distributed among all family members. Finally, the individual-level assets were added to the assets obtained from the household-level distribution. Then, the debts were subtracted to obtain the net wealth value for individuals, which was transformed logarithmically. When calculating the wealth value for 2013, it was converted to the price level of 2011 based on the 2011 price index. Additionally, to avoid erroneous logarithmic transformations caused by zero and negative values, this study truncated the net wealth values at 10, meaning values less than ten were set to 10. The model employs a log-level specification, where the coefficient of log-transformed wealth indicates that a 1% change in wealth corresponds to an average change of 0.01 coefficient units in the health index.

Table 2 The composition of wealth assets in 2011 and 2013 waves from the income, expenditures, and assets module

Instrumental variables and other explanatory variables

This study selected two instrumental variables for wealth: whether inheritance has occurred and the lagged value of present wealth. Considering the occurrence of death events among family members, which may be related to factors such as heredity or early-life events, it can still be assumed that it is not directly associated with the current health status of the respondents but only through increasing the level of wealth, thereby establishing a link with health status [33, 34, 35]. Therefore, inheritance may serve as an appropriate instrument for wealth changes. Using lagged terms as instruments for current variables has some precedent, especially in some dynamic panel data models [36]. As for the lagged value of present wealth, it is generally believed that the lagged term of a particular variable can only affect that present variable, thus influencing other future variables. This assumption is based on the Grossman health demand model, which suggests that health changes mainly depend on current resources and initial endowments, with a weaker effect of lagged variables. We adopted this idea, viewing lagged wealth as a pre-accumulated variable, which indirectly affects health only through current wealth rather than directly impacting health. This causal chain satisfies the exclusion criterion. In addition, this study also selected the following indicators as control variables: (i) demographic variables: gender (0-female; 1-male), age (age at the time of the survey), marital status (0-living alone; 1-having a partner); (ii) socioeconomic status variables: personal annual income (treated in the same way as wealth) and educational attainment (calculated based on the actual years of education received); (iii) geographical variables: living region (0-western region; 1-central region; 2-eastern region), household registration status (0-urban; 1-rural); (iv) childhood health status (0-excellent; 1-very good; 2-good; 3-fair; 4-poor).

Analytical strategy

This study employed two research designs (see Figure S1 in Supplementary material): a longitudinal design, corresponding to the static panel model, and a cross-sectional design, corresponding to the Instrumental Variable (IV) model. We aimed to enhance the findings’ robustness by conducting cross-comparisons between these two research designs. Based on this strategy, the econometric model chosen was divided into two parts according to the datasets. Firstly, the two-period tracking data were estimated using three methods: pooled Ordinary Least Squares (OLS) model, Fixed Effects (FE) model, and Random Effects (RE) model. These three models offer distinct perspectives and estimation effects that enhance our understanding of data characteristics and allow testing different hypotheses. The pooled OLS served as the baseline model, while the FE and RE models were employed to mitigate endogeneity issues. Secondly, although the static panel model addresses the impact of fixed time effects and partially resolves the endogeneity issue, the results may still be biased. Therefore, this study employed the IV method and applied two-stage least squares to handle the cross-sectional data. The chosen instrumental variables were also tested for weak instruments and overidentifying restrictions. This model addressed concerns of reverse causality and endogeneity resulting from omitted variables and served as a cross-verification with the results obtained from the static panel analysis. Furthermore, this study also measured the robustness of the findings by changing the health measurement and using different models.

Results

Data sets

The descriptive analysis of all variables can be found in Table 3. The two sets of samples are the two-period tracking data used for the panel model and the cross-sectional data used for IV estimation. The 27,381 observations include data respectively from two waves (2011 and 2013), with 10,954 individuals present in both waves. The remaining 5,473 observations are from individuals with a single record due to attrition after 2011 (3,261) or new participants added in 2013 (2,212), reflecting the panel’s unbalanced nature. This aligns with the CHARLS survey, where new respondents are added in each wave to offset attrition. Overall, the data structures between different samples are similar, and there are no cases of data mutation. Taking the IV model dataset as an example, a description is as follows. The average age of the sample is 61.13, 88.8% have partners, 48.5% are male, the average education years is 5.69, the majority comes from the central region, accounting for 38.65% of the total sample, and 4.5% of the respondents had inherited wealth before 2013. Between the two data periods, wealth and annual income show an increasing trend year by year.

Table 3 Descriptive statistics of variables in sample sets: panel data from two waves and cross-sectional data for the IV model

In order to preliminarily identify the correlation between wealth and health, Fig. 1 shows the median wealth trend for groups of different health statuses in the two periods. The health index serves as the basis for health stratification, and the annual samples were evenly divided into five quantiles. Then, the median net wealth was calculated for each health quantile. There is a clear descending trend between different health statuses, with the group at the highest health status having more than three times the median wealth compared to the lowest. Taking the first quantile of 2011 as the benchmark, the difference rates can be calculated for each subsequent quantile, which are: second quantile − 22.86%, third quantile − 42.57%, fourth quantile − 55.17%, and fifth quantile − 68.64%.

Fig. 1

The health–wealth gradient in the 2011 and 2013 waves: median net wealth by health index quintiles

Static panel data analysis

Table 4 presents the fitting results for the pooled OLS, fixed, and random effects models, respectively. We conducted tests on these results to select the best model. For the first two, the p-value for the F-test of the FE model was less than 0.01, indicating a solid rejection of the null hypothesis that individual effects are zero. It suggested that the FE model was superior to the pooled OLS model. Furthermore, applying the Hausman test to the results of the latter two models revealed a high level of statistical significance difference in the coefficients of wealth across the models(t = 0.0413/0.0057 = 7.245). Additionally, the chi-square statistic for the overall test was chi2(5) = 179.64, corresponding to a p-value less than 0.01. These tests rejected the null hypothesis that the RE model provided consistent estimates. Hence, for static panel models, the FE model in the third column of Table 4 is deemed most suitable.

Table 4 Static panel model for individual wealth and health index using two waves of data

The first row in Table 4 presents the semi-elasticity coefficient estimate of wealth on health, showing a significant negative correlation. Taking the FE model as an example, the coefficient of wealth (log-transformed) is -0.032, indicating that a 1% increase in wealth will result in an average decrease of approximately 0.00032 units in the health index. Age is also significantly negatively correlated with health in all three models, indicating that higher age is associated with weaker health status. The associations between annual income, marital status, household registration status, and health are no longer significant in the FE model, and the coefficient signs are also different from the results of the other two. This situation indicates the need to address endogeneity issues further, for which this study employs the IV method.

Instrumental variable model

We focused on the 2013 data and selected inheritance and the lagged value of present wealth as instrumental variables to estimate the IV model. Afterward, we conducted a series of statistical tests to ensure the reliability. Firstly, the Durbin-Wu-Hausman test was used to measure whether endogeneity issues exist. For the coefficient of wealth, the IV estimation showed significant differences compared to the baseline results, with t = 0.118/0.0275 = 4.29. At the same time, the overall test statistic chi2(13) = 25.14, corresponding to a P-value of 0.0221, rejected the null hypothesis at a 95% confidence level, indicating that the original hypothesis of “all explanatory variables are exogenous” can be rejected and endogeneity problem exists. Secondly, we used the Hansen-J test to measure the presence of an overidentification problem. The test statistic chi2(1) = 2.52 corresponded to a P-value of 0.113. Therefore, the null hypothesis of “all instrumental variables are uncorrelated with the disturbance term” cannot be rejected, satisfying the exclusion criterion. Finally, using the relative bias criterion, the partial R-squared value was 0.2, and both F = 715.56 and the smallest eigenvalue of 715.559 significantly rejected the hypothesis of weak instruments, indicating that the two instrumental variables used in this study are reasonable.

The third column in Table 5 displays the result estimated using the overidentification IV model. Compared to ordinary OLS estimation, it maintains the significance of the wealth variable and shows a substantial increase in the absolute value of the coefficient. This coefficient difference was highly significant in the Hausman test. Therefore, the coefficient of wealth (log-transformed) is -0.229, indicating that a 1% increase in wealth will lead to an average decrease of approximately 0.00229 units in the health index and is statistically significant at the 1% level. The results of other control variables are stable and in line with expectations. For example, individuals have worse health conditions as their age increases; males have better health status compared to females; individuals with partners have better health status compared to those living alone; individuals with higher education have better health compared to those lower educated; individuals living in the eastern region have better health compared to those in the western region. The annual income shows a decrease in the absolute value and significance of the coefficient, consistent with the manifestation of reverse causality bias. Therefore, as many scholars have pointed out, wealth, a permanent income rather than an annual or short-term income, truly affects health. The household registration status has the expected direction of effect on health but is insignificant. Childhood health status is no longer significant but serves its expected purpose as a control variable.

Table 5 IV model for individual wealth and health index using cross-sectional data

Robustness test

Both panel data models and the IV model indicated a significant causal association between wealth and health, demonstrating the robustness of this study in terms of econometric methods and explanatory power. As follows, this study changed the measurement of health and selected suitable econometric models to examine the conclusion’s stability further. This study transformed the health index into quintiles and estimated the ordered logit model, fixed effects ordered logit model, random effects ordered logit model and IV-ordered logit model. Taking the Order Logit_2sls IV from Table S2 (see Supplementary material) as an example, the wealth coefficient of -0.184 indicates that a 1% increase in wealth results in a 16.8% reduction in the odds of health status transitioning from the excellent to a worse health category (Odds Ratio = e^-0.184 = 0.832). This study also used self-rated health status, CESD-10 scale, ADL scale, IADL scale, other physical impairments, and the presence of eight specific chronic diseases as measures, where higher values indicate poorer health status. Table S1 (see Supplementary material) presents the results of each model, and it can be observed that the coefficient and statistical significance of the wealth remain consistent with the previous results. Furthermore, the results for other control variables are also consistent.

Apart from the eight common chronic diseases in Table S1(see Supplementary material), we also attempted to include the occurrence of hypertension, dyslipidemia, diabetes, liver diseases, stroke, and mental illnesses in the same framework. However, although the wealth coefficients align with expectations in those cases, they are no longer statistically significant. We attribute this to the progression and characteristics of specific diseases, where the significant association with wealth may only manifest when the diseases reach a moderate or severe stage. We speculate that as wealth increases, it may mitigate mild symptoms or improve underlying health risks, ultimately suppressing the occurrence of moderate or severe illnesses. Moreover, the significant coefficients observed between wealth and the eight chronic diseases suggest that clinical treatment and disease prevention should focus more on this effect.

Discussion

This study aimed to answer a fundamental question: How do we explain the well-known wealth-health gradient? This study used the 2011 and 2013 CHARLS data to identify significant health differences associated with varying wealth statuses. Our results supported the wealth-to-health causality hypothesis, indicating a causal relationship where economic resources influence health. Additionally, with the process of life, using a single indicator such as self-rated health alone cannot fully reflect an individual’s healthy aging condition. Daily life information based on comprehensive dimensions can provide a more sufficient view of health. Finally, there were differences based on our results regarding how short-term and long-term income affect health status. Permanent income plays a decisive role in health, while short-term income is a proxy measure for long-term income. The causal mechanisms between short-term income and health are worth further exploration. When individuals experience sudden health crises, leading to significant fluctuations in short-term finances, these fluctuations may not threaten the overall wealth situation in the long run. This study has the potential to make several contributions to the existing literature. Firstly, we use the CHARLS data, which is representative and internationally comparable, allowing for comparing the main findings with existing literature from other developed countries. This study reflects changes in wealth and health variations among older generations in the context of rapid economic development, providing valuable insights for many developing countries. Secondly, we employ an innovative analytical strategy by utilizing two study designs: static panel data analysis and IV estimation methods. This approach enhances the robustness of our conclusions and allows for a more accurate identification of the relationship between wealth and health compared to traditional single-analysis methods. Thirdly, our results provide insights into national policy and practice. These empirical results have significant practical implications for policymakers in other developing countries and regions, facilitating optimal resource allocation and improving population health in aging populations.

Both theory and empirical research have identified multiple possible explanations for the association between health and wealth. This study argues that these results emerge due to the different construction methods of health indicators. In the literature, different indicators reflecting health conditions are chosen and examined with wealth, often leading to entirely different conclusions. For instance, a causal path from health to wealth is found when psychological illness is measured. However, this effect may disappear or show the opposite direction when using common chronic diseases [18, 21, 37]. Therefore, we first adopt the second approach from existing literature, constructing a comprehensive indicator of health status [16, 23, 29]. Compared to using single-dimensional health indicators or “0–1” discrete variables, this measurement can reflect more health-related information and is more in line with our topic of healthy aging [30]. Secondly, this study also incorporates different single-dimensional health indicators into the same framework to examine other conclusions presented under specific health conditions. The core findings of this study are supported under both measuring modes. With the weight distribution of the 34 health indicators in Table 1, the constructed health index mainly reflects the daily functional activities and self-perception of health among the middle-aged and elderly population, which is more consistent with the World Health Organization’s definition of healthy aging [3]. Our log-level model’s coefficient of -0.229 indicates that a 1% increase in wealth is associated with a 0.00229 units reduction in the health index. While this may appear small, it reflects a significant effect when considering the long-term accumulation of wealth disparities. Particularly for individuals with lower wealth levels, small increases in wealth may have a more pronounced impact on health, especially when considering the health index’s standardized nature. It highlights the importance of wealth in improving health, particularly in the elderly population, where even modest health improvements can result in substantial benefits in disease prevention and quality of life.

In Grossman’s health demand theory, wealth and health are two fundamental resources individuals need to sustain their lives. They are determined by various current state and production efficiency parameters and influenced by common initial endowments or variables such as time preference [38, 39]. Therefore, discussing the causal mechanisms between wealth and health has always been the focus of health economics research. The empirical evidence in this study supports a causal pathway from wealth, a core aspect of socioeconomic status, to variations in health outcomes within the population. This causal effect is reflected in the direct and indirect impacts on health. This study employs various econometric models and statistical strategies, conducting longitudinal and cross-sectional data analyses. The causal mechanisms are gradually validated by cross-comparing the fitting results among different models. Throughout the IV analysis, the proper selection of instruments is crucial. In our study design, we chose a sample set using only the 2013 data and used the lagged value of present wealth and inheritance as IVs. While the post-estimation test results provide robust support, it is crucial to consider the following key characteristics of these two instruments, which warrant attention in future research. The inheritance variable may not be suitable in dynamic panel data models. On the one hand, the inheritance phenomenon is not frequent, leading to minimal changes over time and making it more suitable as a non-time-varying variable. On the other hand, the relationship between inheritance and age strengthens over the long term, rendering this instrument less reasonable. Therefore, in the long-term tracking data structure, when selecting appropriate instrumental variables for dynamic panel models, the lagged term of wealth variable may be a better choice. Of course, this research design also requires higher expectations for the continuity and stability of the survey.

Although a considerable amount of research has focused on socioeconomic gradients in health during recent decades, only a few have examined the relationship between wealth and health. Even within this subset of literature, there is a significant reliance on proxy variables for wealth measurement, such as long-term expenditures like household consumption [18], wealth indexes constructed from durable goods [40], or indirect measurements using prime assets like housing estates [37]. Considering that these wealth indexes often only reflect a specific type of asset and cannot comprehensively represent the balance of surplus between different income sources within an individual’s total wealth [22, 41], this study primarily focuses on the net personal wealth that represents the characteristics of the Chinese population. Furthermore, existing research commonly uses imputation methods to address the issue of missing data in financial surveys [13, 16, 42]. In a similar vein, this study employs a universal strategy to enhance the core information of wealth. Additionally, we found that in such a strategy, when imputing missing data, covariate sets used to predict missing values often include health information. Therefore, we suggest that future studies should adequately consider this reverse effect caused by this artificial operation when selecting appropriate econometric models and statistical inference methods. Furthermore, the academic community has long advocated exploring the overall mechanisms through which socioeconomic variables impact health [43]. To address this, our research model simultaneously considers the synergistic effects of wealth, income, and education on health. Our conclusion can be viewed as analyzing the causal effects of socioeconomic status on health differences. Throughout the results, the coefficients for wealth and education consistently maintain the same sign and statistical significance level. However, the income variable yields significantly different results across our models. Considering that this study employs a similar data processing approach for income as for wealth, measurement bias can be ruled out as a reason for this discrepancy. This discrepancy is likely due to reverse causality, where individuals experience sudden health emergencies that result in significant short-term financial fluctuations that do not affect their long-term wealth status [44, 45, 46]. Individuals either recover their health and return to normalcy or succumb to nature, leaving no possibility for subsequent observations.

This study offers two critical policy implications. First, our findings emphasize the importance of targeted poverty alleviation policies in improving impoverished populations’ health. Social welfare policies should focus on increasing wealth, mainly by providing long-term, stable income sources to those in poverty. Health outcomes can be improved by addressing wealth disparities. At the same time, addressing the effect of illness-induced poverty is crucial. Strengthening basic medical security and social welfare systems is essential to safeguarding the health of people experiencing poverty [47]. Second, our study highlights the indirect effects of economic reforms on national health outcomes. For instance, environmental protection policies, while improving the living environment, may also lead to adverse health consequences due to their impact on wealth. The costs of environmental reforms may be passed onto workers through long-term income losses, including layoffs, reduced working hours, or lower wages, which could potentially offset the health benefits of environmental improvements [48]. Policymakers should consider these indirect costs when implementing such sustained reforms to avoid unintended negative impacts on health.

This study has several limitations. Firstly, CHARLS is a panel study that collects data through self-reporting, with multiple waves providing opportunities for longitudinal analysis. This type of survey may have potential biases, including misreporting and underreporting. As a result, specific indicators may be underestimated. On one hand, when dealing with more personal information, such as wealth and income-related questions, there may be inaccuracies in reporting. On the other hand, asset valuation is often complex and difficult to calculate precisely, and the reported facts may be based on very vague original information. Secondly, the data from 2011 to 2013 may only reflect patterns of health differences among the middle-aged and elderly in China during this generation and may not capture recent socio-economic changes. Future research could further explore the potential impacts of time changes by incorporating more waves or updated data, thereby expanding the depth and breadth of this study. This conclusion must be extrapolated and applied concerning specific countries’ social development and era-specific characteristics. If this study design is applied to younger populations or different countries, it may reveal more influences on life processes and national institutional designs. Thirdly, the analytical strategy of this study can be further optimized under the conditions of long panel data, enabling a more comprehensive examination of longitudinal health changes. On the one hand, due to the study’s design, changes in the health index over time between the waves were not considered, as the analysis primarily aimed at capturing the cross-sectional relationship and addressing endogeneity issues. On the other hand, our results also indicate the weak instrumental characteristics of inheritance in the long term. Therefore, exploring more reasonable instrumental variables and requiring at least five or more structurally stable data samples accumulated over multiple periods in the CHARLS survey is necessary to conduct dynamic panel model analysis. Lastly, this study did not directly demonstrate the effect of health on economic resources but instead provided indirect feedback by comparing results among different models. It does not mean that this reverse causal path does not exist under other conditions, such as in early-life populations [49] or specific groups [17, 50].

Conclusion

This study employs static panel data models and IV models to examine the causal effects of wealth on the health status of the middle-aged and elderly population in China while controlling for other socioeconomic factors. Our findings demonstrate that this causal effect is evident in both direct and indirect impacts on health. Nevertheless, caution must be exercised when extrapolating and applying these conclusions to particular countries with distinct social development and era-specific characteristics.