The dissociative hypothesis depicts individuals who achieve social mobility as “socially isolated individuals caught between the margins, experiencing dual exclusion from both their origin and destination social classes” (8). It suggests that any form of social mobility may have adverse effects on health. From a psychological health perspective, individuals experiencing social mobility must adapt to the norms of a new social group, a process often accompanied by stress, which can lead to feelings of isolation and alienation, potentially resulting in anxiety and depression (9). From a physical health standpoint, prolonged psychological stress may activate the stress response system, increasing the risk of cardiovascular diseases and immune suppression (10). The frustrated achiever hypothesis further posits that new entrants are at a disadvantage compared to the original members of their destination class. These achievers may compare their situation to those stable in higher classes, realize their relatively lower status, and consequently feel negative emotions, which can adversely affect their health and well-being (11). Early stress theories and clinical studies have also identified the negative impacts of social mobility on mental and physical health, showing that class transitions can deteriorate the psychological health or well-being of some individuals. This conclusion has been validated in European countries such as Germany, the United Kingdom, and Poland (12–14).

Existing research generally concurs that downward social mobility exerts a detrimental impact on health. For instance, the ‘falling from grace’ thesis posits that such mobility, entailing a loss, often involves involuntary and uncontrollable negative life events such as bankruptcy, unemployment, or divorce, which signify a status, power, or income decline. Individuals may struggle to adapt to their new, deteriorated living circumstances (15). Moreover, downward mobility can be perceived as a failure to meet societal expectations or personal aspirations, potentially giving rise to negative emotions such as self-blame, depression, pessimism, fear, and even anger and feelings of injustice, which can leave lasting scars on an individual’s health (16). Gugushvili and Prag (17) further illustrate that the health toll of downward mobility on participants manifests in increased substance abuse, smoking, alcohol and drug consumption, depression, anxiety, fear, weight gain, exacerbation or onset of chronic diseases, among other health indicators (18).

The rising from rags hypothesis, originating from social psychology research, suggests that the positive impacts of upward social mobility on health outweigh the potential negatives (19). Enabled by the fulfillment of socioeconomic aspirations and enhanced feelings of personal control, upward social mobility is seen as an aspirational process where overcoming adversity to escape disadvantageous socioeconomic positions generates satisfaction and a sense of achievement that can buffer against psychosocial stressors during the mobility process (20). Moreover, it fosters gratitude and endorsement toward the societal structures facilitating the upward leap (21), positively influencing health. Studies have revealed that individuals experiencing upward social mobility exhibit reduced mortality rates (22), lower morbidity (23), improved cognitive function (24), lower depression scores (25), fewer functional impairments (26), better self-rated health (27), heightened happiness, and healthier behaviors (28). Compared to those experiencing downward mobility, upwardly mobile individuals demonstrate superior physical and mental health and higher levels of well-being (29).

The acculturation hypothesis frames social mobility as a process of cultural adaptation to the normative values and lifestyles associated with the destination social status, a matter of assimilation or resocialization, devoid of the psychological distress implied by “separation” (30). Through the lens of acculturation theory, the impact of social mobility on individual health and well-being is largely contingent upon the degree to which individuals integrate into their new class milieu; the lower the level of social integration, the poorer the health outcomes. Hence, the health effects of social mobility are also mediated by the discrepancy between the origin and destination social classes, with both the origin and destination exerting some influence on those undergoing mobility. According to Houle (31), the destination class exerts a stronger influence on health status than the origin class, given the closer relationship between current class position and immediate outcomes. In summary, the acculturation hypothesis emphasizes that the impact of social mobility on health largely depends on the level of integration achieved and the class of origin and destination, with the destination class potentially having a more significant impact on health outcomes.

Existing research exhibits several limitations, mainly in the following areas: Firstly, from the perspective of research subjects, the current studies on the health effects of social mobility are relatively limited, with the majority focusing on developed countries or regions, while research on developing countries or areas is scarce. Secondly, from the perspective of research content, existing studies mostly remain at the superficial level of examining whether social mobility affects health status, lacking in-depth exploration of how social mobility influences health status. Thirdly, from the perspective of variable measurement methods, existing empirical research predominantly characterizes social mobility through changes in objective socioeconomic status indicators such as education, income, and occupation. However, studies incorporating subjective indicators, such as self-perceived social mobility, still need further enrichment.

Self-rated health. Self-rated health represents an individual’s subjective perception and psychological appraisal of their health status, which significantly reflects an integrated assessment of one’s health and has demonstrated predictive capacity for both mortality and morbidity (38, 39), with high data accessibility. In this study, self-rated health is used to characterize the health status of the sample population, measured through the CGSS questionnaire item: “How do you feel about your current physical health status?” Responses are categorized into five levels: very unhealthy, somewhat unhealthy, average, somewhat healthy, and very healthy, coded numerically as 1 through 5, respectively.

Upward social mobility. The CGSS questionnaire employs a visual analog scale accompanied by a ladder image, inviting respondents to self-assign their social class positioning at different points in time, ranging from 1 (lowest) to 10 (highest). Drawing from Jindong et al. (49), this study consolidates adjacent classes into five broader categories: very low, low, middle, high, and very high social classes. For the baseline regression analysis, a binary variable (0–1) is utilized to denote social mobility, coded as “1” if the current social class ranking is higher than that of 10 years prior, indicating upward social mobility, and “0” otherwise. In the subsequent advanced analysis aimed at exploring how the direction and magnitude of social class change affect health outcomes, the difference between an individual’s current social class and that of a decade ago is employed as a measure of social class transition.

This study also controls for factors that may give rise to variations in health status, chief among which are gender, age, age squared, Hukou type, marital status, educational attainment, employment status, logarithm of annual household income, number of children, smoking habits, alcohol consumption, participation in medical insurance, and geographic region. A detailed descriptive statistical summary of these variables is provided in Table 1.

The primary independent variable in this study is self-rated health, which encompasses five ordinal categories: very unhealthy, somewhat unhealthy, average, somewhat healthy, and very healthy. Consequently, an ordinal logistic regression model is employed for analysis, given its suitability for handling dependent variables with ordered categories. The ordinal logistic regression equation utilized is formulated as follows:

The unobservable latent variable Yi∗ has an underlying relationship with the observed ordinal outcome of self-rated health, which can be described as follows:

Where β1,β2,β3,β4 represent the estimable parameter, known as the “cut-point” or threshold parameter, and β1<β2<β3<β4. The ordinal logistic regression model is specified as follows:

Let j represent the self-rated health levels, yj denote the intercept, xi be the factors influencing self-rated health, and δi be the regression coefficients for each influencing factor, representing the impact of the explanatory variables on the dependent variable.

Common methods for testing mediation effects, such as the “three-step method,” Sobel test, and Bootstrap test, are premised on the comparability of coefficients in linear regression models. However, when the dependent variable is non-continuous, scale issues render coefficients incomparable, thereby hindering the application of techniques for examining mediating variables in nonlinear regression frameworks. Therefore, this paper employs the KHB mediation analysis method proposed by Kohler, Karlson, and Holm to discuss how social mobility affects health outcomes. Essentially, this method analyzes the underlying linear model behind the nonlinear probability model, using the residual term from the regression of the mediator on the independent variable to replace the mediator itself. This approach eliminates the model scale changes caused by changes in the independent variable within the nonlinear probability model, thereby obtaining the proportion of the mediation effect.

Equation 4 represents the model where the mediator variable is controlled, whereas Equation 5 illustrates the model without controlling for the mediator. In these equations, Xi′ signifies the control variables, and Z denotes the mediator variable. The indirect effect, denoted by Δβ=β1′−β1, captures how the independent variable operates through the mediator to influence the dependent variable. In the estimation of the baseline regression models, we obtain fitted coefficients b1 and b2, b1=β1σ1, b2=β1′σ2, with σ1 and σ2 serving as scale parameters that depend on the standard deviation of the residuals in respective models. Conventionally, scale coefficients may vary across distinct models; however, they adhere to the constraint that b2−b1≅β1′−β1,

The KHB method initially treats the mediator variable Z as the dependent variable, regressing it on Yi as the independent variable, forming the model Z=c+d·Yi+r, where r represents the residual from this regression. Subsequently, this residual r is incorporated as an independent variable in a subsequent model fitting exercise, resulting in:

Equation 6 exhibits a comparable model fit to Equation 5, indicating that ε2i=ε2′i∗, suggesting that Yi and r are not perfectly correlated. Consequently, the indirect effect of variable Z in the baseline model can be represented as follows: b2∗−b1=β1′σ2∗−β1σ1. Thus, the direct effect, indirect effect, and total effect of the independent variable MOBi on the dependent variable Yi in the baseline model are disentangled, and by comparing the ratio of the indirect effect to the total effect, the magnitude of the mediator variable Z’s influence is quantified.

To exclude the potential impact of omitted variables and random factors on the aforementioned research findings, drawing upon the studies of Yuhui, Huan (40) and La Ferrara, Chong (41), this paper conducts placebo tests by constructing fictitious core explanatory variables. The essence of this approach lies in substituting the original explanatory variable with a fabricated dummy variable that theoretically should exert no influence on the dependent variable. If the dummy variable influences the dependent variable, it suggests the presence of omitted variable bias in the baseline regression results. Accordingly, we randomly generate the states of sample’s social mobility and refitted Model (1) to compute the model coefficients. This study employs a Monte Carlo test to repeat the above steps 500 times. The purpose of multiple simulations is eliminate the possibility of randomness in the results, and exclude any systematic biases by comparing the consistency of different runs, thereby obtaining a more stable and reliable distribution of results. We then analyze the regression coefficients of social mobility from the 500 regressions, generate a kernel density plot, and compare them with the coefficients of social mobility in Model (1). As shown in Figure 3, the estimated coefficients obtained from the pseudo-explanatory variables are concentrated around 0 and follow a normal distribution, all of which are much smaller than the true estimated coefficient (0.352) in Model (1). This indicates that fabricated social mobility scenarios cannot support a positive correlation between social mobility and self-rated health, suggesting that our regression model can effectively control for omitted variable bias.

To mitigate the impact of sample selection biases and inter-group heterogeneity on the research outcomes, this study employs the Propensity Score Matching (PSM) technique. The fundamental rationale behind PSM is to identify a counterpart individual “j” in the control group whose observable characteristics closely resemble those of an individual “i” in the treatment group, thereby establishing a counterfactual scenario that facilitates estimation of the treatment effect for individual “i.” Employing neighbor matching, radius matching, kernel matching, spline matching, and Mahalanobis matching, we found that social mobility positively affects self-rated health. The net effect of social mobility on self-rated health ranges between 16.1 and 20.2% (Table 3). The PSM analysis outcomes affirm that the positive association between social mobility and self-assessed health remains robust even after adjusting for sample selection biases.

On one hand, higher social classes typically have greater access to health resources, contributing to better health. On the other hand, good health can enhance an individual’s work efficiency and learning ability, increasing their chances of social mobility. To tackle this potential endogeneity issue where the explanatory and response variables may be mutually causal, this study employs the instrumental variable (IV) method. Following Bentolila, Michelacci (42), we adopt aggregate-level data as an instrument for the lower-level variable, specifically utilizing provincial upward social mobility rates as the instrumental variable.

Table 4 presents the regression results using the Instrumental Variables Two-Stage Least Squares (IV-2SLS) method. As shown in Model (3), the provincial upward social mobility rate has a coefficient of 0.955 on the sample individuals’ social mobility, statistically significant at the 1% level, strongly validating the aforesaid argument about the instrument’s relevance. This confirms that provincial upward social mobility rates positively influence individuals’ ability to achieve social class advancement. In Model (4), the coefficient of social mobility on self-rated health remains positive, suggesting that, after addressing potential endogeneity issues with the IV method, the positive effect of social mobility on self-assessed health found in the baseline regression persists. Additionally, the Cragg-Donald Wald F statistic and the Kleibergen-Paap Wald rk F statistic, both used to test for weak instruments, exceed the critical value associated with a 10% maximal IV size (16.38), demonstrating that this study does not suffer from identification or weak instrument problems.

Significant differences between men and women in physiological, psychological, and behavioral aspects may influence the generation and manifestation of health effects. An analysis of gender disparities in how social mobility impacts self-rated health reveals that the coefficient for the effect of social mobility on men’s self-rated health is 0.462 (p = 0.000), whereas the corresponding coefficient for women is substantially lower at 0.273 (p = 0.002), indicating that social mobility exerts a much stronger positive influence on men’s self-perceived health compared to women’s. The difference in the coefficient of influence between the two groups is 0.189, which passes the Fisher combination test at the 10% level (Table 5).

This study utilizes the social class of the respondent’s family at age 14 to represent early socioeconomic status. Samples are divided into two groups—high and low early socioeconomic status—based on whether their family’s social class at age 14 was above or below the median. Heterogeneity analysis results based on early socioeconomic status indicate that social mobility positively impacts self-rated health outcomes for groups with higher and lower early socioeconomic status. Particularly noteworthy is that for individuals with lower early socioeconomic status, the effect of social mobility on self-rated health is markedly more substantial. The difference in the effect sizes between the two groups is 0.199, passing the permutation test at the 5% significance level (Table 5). Therefore, the health-promoting effect of social mobility is more pronounced among those with lower early socioeconomic status.

This study employs the frequency of consuming fresh fruits or vegetables and the occurrence of preceding necessary medical care due to affordability within the past year to evaluate residents’ material living conditions from the perspectives of dietary intake and potential healthcare demand. According to the results from Model (5) in Table 6, after introducing dietary intake as a mediator in the baseline model, the regression coefficient of the primary explanatory variable—social mobility—significantly diminishes to 0.331 (p = 0.000), while the coefficient for dietary intake stands at 0.290 (p = 0.000). Additionally, referring to the data in Table 7, the percentage of confounding effects attributable to dietary intake is 7.32%. This indicates that the enhancement in self-rated health following social mobility is partially attributable to improved intake of fresh fruits and vegetables post-mobility, accounting for 7.32% of the health effect.

Upon introducing the latent healthcare demand variable into the baseline regression, the impact coefficient of social mobility reduces to 0.345 (p = 0.002). Unlike all other mediators, the coefficient for the latent healthcare demand mediator becomes negative, at −1.194 (p = 0.000), with the corresponding transmission mechanism depicted in Pathway 2 of Figure 4. Hypothesis H2 is thereby substantiated.

This study uses well-being and self-confidence as proxy variables to test the mediating role of psychological support mechanisms in the relationship between social mobility and self-rated health. Well-being is derived from residents’ self-assessment of their happiness, while self-confidence is measured by whether residents have ever lost self-confidence. The results support hypothesis H3, corresponding to mediation pathway 1. Well-being and self-confidence are positive mechanisms through which social mobility affects self-rated health, explaining 32.99 and 16.18% of the total effect, respectively. Among all the mediating variables, well-being has the strongest explanatory power.

Drawing from the methodology of Fuqin and Yuyin (44), this study employs social interaction and social identify to gage the sample’s social capital. Social interaction is measured by the frequency of socializing or visiting others during leisure time, categorized into five levels from “never” to “very frequently.” Social identity is derived from the respondent’s perception of societal fairness, ranging from “completely unfair” to “completely fair” across five categories. As demonstrated in Table 6, Models (9) and Models (10) present the mediating effects of social interaction and social identify, respectively.

Firstly, the coefficients for the impact of social interaction and social identify are significantly positive at the 1% level, suggesting that closer social ties and higher degrees of social identify can positively influence self-rated health to a certain extent. The coefficients for the effect of social mobility stand at 0.338 and 0.330, respectively, and are statistically significant at the 1% level. Compared to the coefficient for the dependent variable in the baseline regression (0.352), these figures show a decline, indicating that social interaction and social identify serve as mechanisms through which social mobility positively affects self-rated health status. In summary, the social capital mechanism by which social mobility impacts self-rated health is illustrated in Pathway 1 of Figure 4, showing that social class transitions improve self-rated health by increasing social interaction and enhancing social identify. Hypothesis H4 is thus supported. According to the KHB mediation effect analysis results in Table 7, social interaction and social identify account for 4.86 and 10.04% of the total effect, respectively, with social identify exhibiting a stronger mediating role than social interaction.