Background

Incorporating overall health condition measures in national surveys is crucial for public health surveillance, communication, and policymaking. Assessing social function provides insights into social well-being, while monitoring depressive symptoms supports the diagnosis and management of depression [1,2,3,4]. Meanwhile, ethnic disparities in health outcomes are a well-documented phenomenon across various societies, with minority populations frequently experiencing poorer health compared to majority groups [5, 6]. These disparities highlight systemic inequities that perpetuate marginalization and disadvantage. Social epidemiological theories, such as ecosocial theory, provide robust frameworks for analyzing the multilevel factors contributing to health inequalities [7]. Ecosocial theory emphasizes the importance of considering individual-level factors (e.g., socioeconomic status, behaviors, living conditions) alongside institutional- and societal-level factors (e.g., policies, social norms, access to resources) in shaping health outcomes [8]. It acknowledges the dynamic interactions between these factors, recognizing that their impact may be synergistic or context dependent [9]. In the United States, this theory has been applied to understand how societal factors such as systemic racism, economic inequality, and healthcare access contribute to poorer health outcomes among racial and ethnic minorities [7, 9]. Similar approaches in Brazil and South Africa have highlighted how historical and structural inequalities shape health disparities, with marginalized groups facing greater exposure to environmental hazards, limited healthcare access, and reduced economic opportunities [9]. Despite its widespread use, there is a lack of research applying ecosocial theory to health disparities among China’s ethnic minority populations, underscoring the need for further study to understand how societal and structural factors influence health outcomes in this context.

China officially recognizes 56 ethnic groups, including the Han majority and 55 ethnic minority groups. According to the 2020 census, ethnic minorities comprise 125.47 million people (8.89% of the total population), including the Zhuang (16 million), Manchu (10 million), Miao (9 million), Hui (9 million), and Uyghur (8 million), among others [10, 11]. These groups exhibit considerable heterogeneity: ethnic minorities are more likely to live in rural areas compared to Han [12]; they speak over 120 distinct languages [13]; practice diverse religions including Islam, Buddhism, and indigenous beliefs [14]; and generally show lower educational attainment and formal employment rates than the Han majority [15, 16]. Most ethnic minorities reside in southern and western China, particularly in resource-limited mountainous and pastoral areas with distinct environmental challenges [17]. Chinese ethnicity minorities warrant focused study because: they comprise a large population of approximately 125 million [11]; they preserve distinctive cultural practices and traditional medicine systems that shape health behaviors [18]; and they are integral to achieving China’s health equity goals and the Healthy China 2030 initiative [19].

China has implemented preferential policies for ethnic minorities including bonus points on the college entrance examination, exemptions from the former one-child policy, tax reductions, business financing preferences, and subsidized healthcare in ethnic autonomous regions [20,21,22,23]. However, these populations still face structural barriers to healthcare access due to geographic isolation, language differences, and limited health infrastructure in minority-concentrated areas [24, 25]. Recent policy shifts since 2019 have scaled back certain affirmative actions for Chinese ethnic minorities, underscoring the importance of understanding their current health status and needs [26].

Despite the significance of ethnic minority health, few studies have used national-level data to quantify health disparities between ethnic minorities and the Han majority within an ecosocial framework. Previous studies have produced mixed findings, with some indicating poorer health among ethnic minority groups and others suggesting potential protective factors associated with certain minority cultures and lifestyles [27,28,29,30]. However, many of these studies have been limited in scope or relied on localized data sources. For instance, Tang et al. (2015) studied health services among ethnic minorities in rural western China and found low utilization due to socio-economic barriers, focusing on a few provinces [31]. Gong et al. (2016) used the China Health and Retirement Longitudinal Study to investigate healthcare access among older adults, finding health disparities between Han and ethnic minority populations, with rural minorities facing greater challenges, but the study does not include all age groups or provinces [32].

Our study aims to address this gap by utilizing the repeated cross-sectional data from the Chinese General Social Survey (CGSS), a nationally representative survey, and the National Bureau of Statistics of China, to examine the association between the ethnic minority group status and self-reported health outcomes, including overall health, social functioning, and depression. Drawing upon social epidemiological theories, we aim to: (1) Examine disparities in health outcomes between Han and ethnic minority populations in China; (2) Investigate the relative contributions of individual-level factors (e.g., sociodemographic characteristics) and institutional- and societal-level factors (e.g., healthcare resource availability) in explaining variations in self-reported health outcomes. (3) Assess ecosocial theory in the Chinese context, contributing to a clearer understanding of the interplay between ethnicity, social determinants, and health disparities. Given favorable policies in education, tax reduction, and business empowerment for Chinese ethnic minority populations [18, 33], combined with limited access to healthcare resources and health information, we hypothesized that ethnic minority populations would have similar health conditions compared to the population of Han ethnicity.

Methods

Data sources

We used two data sources: the Chinese General Social Survey (CGSS) for individual-level data and the National Bureau of Statistics of China for institutional- and societal-level data to examine the interaction between these factors. The CGSS, established in 2003, is China’s premier national academic survey, designed by Renmin University of China and the Survey Research Center of Hong Kong University of Science and Technology to provide a comprehensive overview of the country’s social structure, quality of life, and social changes. The CGSS utilizes a multi-stage stratified sampling method to collect nationally representative individual-level data through face-to-face interviews conducted by trained surveyors. The CGSS interviews were conducted primarily in Mandarin Chinese; however, when participants did not understand Mandarin, trained interviewers could translate questions and response options into local languages [34]. The target population includes individuals aged 18 and above across the country. The CGSS obtained informed consent from all participants. Detailed methodology and the complete survey of the CGSS are available in other publications [35, 36]. Our study leveraged data from seven waves of the CGSS conducted in 2010, 2012, 2013, 2015, 2017, 2018, and 2021. As the CGSS dataset is de-identified, it does not require Institutional Review Board (IRB) approval. The CGSS dataset is freely available for academic research after registration at the China National Survey Data Archive website [37].

The institutional-level data were obtained from the National Bureau of Statistics of China for the years 2010, 2012, 2013, 2015, 2017, 2018, and 2021, corresponding to the CGSS survey years. The National Bureau of Statistics of China data are publicly accessible without registration through their official website [38]. We overlayed individual-level data with institutional-level data using the individual’s province of residence, in order to measure the health resources for individuals available in each province.

Exposure

The study aimed to compare health outcomes between Han and ethnic minority populations using cross-sectional data. The exposure variable was self-reported ethnicity status, aiming to capture the social stratifying effects of systemic marginalization [39,40,41]. Possible responses included Han, Mongol, Manchu, Hui, Tibetan, Zhuang, Uyghur, and other. The raw CGSS dataset thus captured 7 specifically named ethnic minority groups plus an “other” category that included all remaining minority groups. Han ethnicity was classified as the reference group, while all other ethnicities were classified as the comparison group. There were very few participants who were Tibetan (Range: 2–10) and Uyghur (Range: 0–76) across the survey years. To ensure sufficient statistical power, and because our study aim was to examine minoritized status overall as our exposure of interest, we combined Mongol, Manchu, Hui, Tibetan, Zhuang, Uyghur, and other ethnicities into a single group.

Outcome

The outcome was self-reported health status, measured using five-level Likert scales. For overall health conditions, respondents were asked, “How do you perceive your current state of health?” with possible responses: 1 (Very unhealthy), 2 (Somewhat unhealthy), 3 (Average), 4 (Somewhat healthy), and 5 (Very healthy). For social function, respondents were asked, “In the past four weeks, how often have health problems affected your work or other daily activities?” with responses: 1 (Always), 2 (Often), 3 (Sometimes), 4 (Rarely), and 5 (Never). Higher scores indicated better health for overall health conditions and social functions.

Frequency of depressive symptoms was measured on a five-level Likert scale. Respondents were asked, “In the past four weeks, how often have you felt depressed or downhearted?” with responses: 1 (Never), 2 (Rarely), 3 (Sometimes), 4 (Often), and 5 (Always). Higher scores indicated more severe depressive symptoms. These single-item measurements have been validated in various studies for their simplicity and efficiency [1,2,3,4, 42].

Covariates

The selection of covariates was based on ecosocial theory, which emphasizes that health conditions are influenced by both individual-, institutional- and societal-level factors [9]. Individual-, institutional- and societal-level factors are classified based on their scope and influence on health outcomes, with the individual-level factors directly influencing an individual’s health and being specific to each person, while the latter two level factors encompass broader influences affecting populations and reflecting collective resources that impact health outcomes at the institutional or societal level. It is important to note that individual, institutional, and societal influences can’t be separated by the type of construct measured, since, as ecosocial theory points out, there is dynamic interplay between these factors. For the current study, individual-level factors included gender, education level, marital status, residence type, income, socioeconomic position, and religious belief [31, 43, 44]. Gender was self-reported and coded as a binary variable, with female as the reference group. Education level was treated as an ordinal variable ranging from no formal education (reference), informal or traditional education, elementary education, secondary education, to higher education. Marital status was classified as a categorical variable with categories: single (reference), cohabiting married, separated but not divorced, divorced, and widowed. Residence type was also a categorical variable with categories: rural (reference), urban, and other. Self-reported household income position was measured by asking participants: “What level is your family’s economic situation in your local area?” with response options on a 5-point scale: 1 (far below average), 2 (below average), 3 (average), 4 (above average), and 5 (far above average). This measure captures perceived relative economic standing within participants’ local communities rather than absolute income levels. Self-reported socioeconomic position was measured by asking participants: “What level do you think you are currently at?” on a 10-point scale where 1 represented the lowest level and 10 the highest level. To align with our analytical framework in family income and increase statistical power, we recoded responses into five categories: scores 1–2 were recoded as 1 (very low), 3–4 as 2 (low), 5–6 as 3 (medium), 7–8 as 4 (high), and 9–10 as 5 (very high). This subjective measure captures participants’ perceived social standing. Religious belief was assessed with 11 response options: no religious belief, Buddhism, Taoism, folk religion, Islam, Catholicism, Christianity (Protestantism), Eastern Orthodox, other Christianity, Judaism, and Hinduism. We dichotomized this variable into “no religious belief” (reference category) versus “any religious belief” for analytical purposes. This binary approach was chosen because: (1) religious belief was not a primary variable of interest in our study; (2) some religious categories had very small sample sizes which could lead to unstable estimates; and (3) preserving multiple categories would substantially reduce degrees of freedom without contributing meaningful insights to our research questions about ethnicity and health outcomes.

Institutional-level factors included health-related variables across provinces and years, including the number of licensed doctors per 10,000 people, number of assistant physicians per 10,000 people, number of nurses per 10,000 people, number of health technicians per 10,000 people, and number of medical beds per 10,000 people. Living region was conceptualized as a societal-level factor and categorized into seven regions: North, Northeast, East (reference), Central, South, Southwest, and Northwest.

We utilized a directed acyclic graph to delineate the assumptions regarding the relationships between the minority group ethnicity, health outcomes, and covariates (Fig. 1).

Fig. 1
Fig. 1
Full size image

Directed acyclic graph for the relationship between variables. Note: Individual-level factors are treated as covariates to estimate controlled effects of ethnicity on health outcomes (overall health, social function, frequency of depressive symptoms), rather than examining mediation pathways. While ethnicity may influence individual-level factors (e.g., education, income) in reality, these pathways are not shown as our approach adjusts for these factors rather than exploring indirect effects. Institutional and societal-level factors account for geographic and healthcare access variations

Statistical methods

To examine the distinctiveness of our three health outcome variables, we conducted Spearman correlation analyses and a Friedman test with post-hoc Wilcoxon signed-rank tests. These analyses assessed whether the outcomes measured distinct health dimensions to warrant separate analyses. Bonferroni correction was applied for multiple comparisons. For the main analysis, we applied ordinal regressions to examine the association between the minority group ethnicity and health outcomes. These models included the residence province as a random effect with random slopes to account for the clustering of participants by province. For each health outcome (overall health condition, social function, frequency of depressive symptoms), we fitted five models: Model 1 did not adjust for any covariates; Model 2 adjusted for individual factors; Model 3 adjusted for institutional- and societal-level factors; Model 4 adjusted for individual-, institutional- and societal-level factors; and Model 5 further adjusted for two sets of interaction terms: the interactions between the minority group ethnicity and institutional-level factors and the interactions between the living region (societal-level) and institutional-level factors. The regression framework is illustrated using the overall health condition as an example of the outcome.

Model 1 is specified as:

$$\:{Y}_{it}={\beta\:}_{0}+{\beta\:}_{1}\:\mathbf{I}\:Minority\:group\:ethnicity+\:{u}_{j}+{\epsilon\:}_{it}$$

where \({Y}_{it}\) represents individual \(\:i{\prime\:}s\) overall health condition at time \(\:t\). \(\:\:\mathbf{I}\:Minority\:group\:ethnicity\) is an indicator variable for having a minority group ethnicity (1 if is ethnic minority, 0 if not). \(\:{u}_{j}\) is the random intercept for province \(\:j\). \(\:{\epsilon\:}_{it}\) is the error term. \(\:{\beta\:}_{1}\)represents the main effect of the minority ethnicity, similarly denoted in Models 2, 3, 4, and 5.

Model 2 is specified as:

$$\begin{aligned}{Y}_{it}&={\beta\:}_{0}+{\beta\:}_{1}\:\:\mathbf{I}\:Minority\:group\:ethnicity\\&+\:{\gamma\:X}_{it}+\:{u}_{j}+{\epsilon\:}_{it}\end{aligned}$$

where \(\:{X}_{it}\:\)is a vector of individual level covariates for individual \(\:i\) at time \(\:t\) (gender, education level, marriage status, residence type, income position, socioeconomical position, and religious belief).

Model 3 is specified as:

$$\begin{aligned}\:{Y}_{it}&={\beta\:}_{0}+{\beta\:}_{1}\:\:\mathbf{I}\:Minority\:group\:ethnicity\\&+\:\lambda\:{Z}_{t}+\:{u}_{j}+{\epsilon\:}_{it}\:\end{aligned}$$

where \(\:{Z}_{t}\) ​is a vector of institutional- and societal-level covariates (living region, year fixed effects, number of licensed doctors per 10,000 people, number of assistant physicians per 10,000 people, number of nurses per 10,000 people, number of health technicians per 10,000 people, and number of medical beds per 10,000 people).

Model 4 is specified as:

$$\:{Y}_{it}={\beta\:}_{0}+{\beta\:}_{1}\:\:\mathbf{I}\:Minority\:group\:ethnicity+\:{\gamma\:X}_{it}+\lambda\:{Z}_{t}+\:{u}_{j}+{\epsilon\:}_{it}$$

Model 5 is specified as:

$$\:{Y}_{it}={\beta\:}_{0}+{\beta\:}_{1}\:\:\mathbf{I}\:Minority\:group\:ethnicity+\:{\gamma\:X}_{t}+\lambda\:{Z}_{it}+\partial\:\:\mathbf{I}\:Minority\:group\:ethnicity\:\times\:{Z}_{t}+\phi\:\:{Living\:Region\:\times\:Z}_{t}+\:{u}_{j}+\:{\epsilon\:}_{it}$$

where \(\:{\:\mathbf{I}\:Minority\:group\:ethnicity\:\times\:Z}_{t}\)​are interactions between \(\:\:\mathbf{I}\:Minority\:group\:ethnicity\) and institutional-level factors \(\:{Z}_{t}\), and \(\:{Living\:Region\:\times\:Z}_{t}\)are interactions between \(\:Living\:Region\) and institutional-level factors. We included the first set of interactions because ecosocial theories emphasize institutional factors as potential mechanisms of marginalization for minority ethnic populations. The second set of interactions was included due to the variation in medical resources across the different regions where participants resided.

We calculated McFadden’s pseudo R-squared to assess the relative contributions of four components: ethnicity, individual-level factors, institutional- and societal-level factors, and their interactions, with higher values indicating a greater proportion of variance explained by the model [45]. The formula for calculating McFadden’s pseudo-R-squared is as follows:

$$\:{R}^{2}=1-\frac{\text{l}\text{n}\left({L}_{model}\right)}{\text{l}\text{n}\left({L}_{model0}\right)}$$

where \(\:\text{l}\text{n}\left({L}_{model}\right)\) is the log-likelihood of the fitted model, and \(\:\text{l}\text{n}\left({L}_{model0}\right)\) is the log-likelihood of the null model (a model with only the intercept), and it is specified as (Model 0):

$$\:{Y}_{it}={\beta\:}_{0}+\:{u}_{j}+\:{\epsilon\:}_{it}$$

To evaluate potential overfitting due to the inclusion of additional variables, the Akaike Information Criterion (AIC) was computed for each model, with lower values indicating better model fit.

We conducted a complete case analysis, with a comparison between respondents with and without missing values documented in the Supplementary Material (Table S1). Cumulative logit plots were generated to examine the proportional odds assumption for ordinal regressions. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated, with statistical significance determined by 95% CIs excluding 1. The regressions were weighted using survey weights provided by the CGSS to ensure national representativeness. We performed several sensitivity analyses: (1) We used the random forest imputation and reperformed the analysis using the imputed data set to examine the possible effects of missing values on the estimate. Imputation of missing values was performed using the missRanger algorithm in R, with the number of trees set to 500, and predictive mean matching with \(\:k=5\). The imputation process utilized 4 threads and was run for a maximum of 10 iterations, ensuring robust and efficient handling of missing data. (2) For the full model (Model 5), we dropped interaction terms between the region and institutional-level factors to assess the model’s robustness. (3) We generated 9 random variables as the negative outcome to examine if the observed effect was due by chance following Rehkopf et al. [46].

All analyses were performed using R (Version 4.1). The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (Supplementary Material STROBE Statement).

Results

Participants

The study examined the association between minority group ethnicity and self-reported health outcomes. The raw dataset included 79,471 individuals. After participants with missing values in key variables were excluded, the final sample size for complete case analysis was 66,057 (Fig. 2). A comparison between respondents with and without missing values is provided in the Supplementary Material Table S1, and the analytical flow chart is depicted in Fig. 2. We observed no systematic differences in the sociodemographic characteristics between participants with and without missing values.

Fig. 2
Fig. 2
Full size image

Analytic sample flowchart for the Chinese General Social Survey (2010–2021)

Descriptive data

Table 1 presents the sample characteristics. Most participants resided in the Eastern China throughout the survey years (Range: 28.5% to 35.6%). Most participants were of non-minority ethnicity (Range: 91.43% to 93.71%). The majority reported a medium household income position (Range: 50.2% to 58.3%) and identified their socioeconomic position as medium (Range: 40.7% to 47.6%).

Table 1 Sample characteristics: the Chinese general social survey (2010–2021)
Full size table

After stratifying by the minority group ethnicity status, ethnic minorities were more likely to reside in rural areas (51.6–62.6% vs. 26.8–39.5% for Han) and report religious belief (30.1–43.0% vs. 5.4–11.3% for Han). Geographically, while Han participants concentrated in eastern China (30.7–37.5%), minorities were distributed across southwest, northwest, and south regions. Educational attainment was lower among minorities, with higher proportions having elementary education or below (34.2–46.6% vs. 28.2–34.8% for Han).

Regarding health outcomes, we observed that most participants reported their overall health status as “quite healthy,” with proportions ranging from 31.0% to 40.3% (Fig. 3). For social function, most participants indicated they were “very healthy,” except in 2015, when the majority reported being “quite healthy” (Han ethnicity: 37.5%; Minority ethnicity: 36.2%) (Fig. 4). Regarding self-reported frequency of depressive symptoms, there was an increasing trend in participants reporting “never” having depressive feelings over the survey years (Fig. 5). The Spearman correlations among health outcomes showed moderate associations (overall health-social function: r = 0.587; overall health-frequency of depressive symptoms: r=−0.393; social function-frequency of depressive symptoms: r=−0.502), indicating related but distinct constructs. The Friedman test revealed significant differences in how participants rated themselves across the three dimensions (χ² = 50641, df = 2, P < 0.001), with all pairwise comparisons significant after Bonferroni correction. These results support analyzing the three outcomes separately (Supplementary Table S2).

Fig. 3
Fig. 3
Full size image

Percentage of people selecting each overall health condition category by year, stratified by minority group ethnicity status: The Chinese General Social Survey (2010–2021) (a) Overall; (b) Stratified by minority group ethnicity. Note: The shaded region around each line indicates the standard error of the point estimate. For overall health conditions, respondents were asked, “How do you perceive your current state of health?” with possible responses: 1 (Very unhealthy), 2 (Somewhat unhealthy), 3 (Average), 4 (Somewhat healthy), and 5 (Very healthy). Higher scores indicated better overall health

Fig. 4
Fig. 4
Full size image

Percentage of people selecting each social function condition category by year, stratified by minority group ethnicity: The Chinese General Social Survey (2010–2021) (a) Overall; (b) Stratified by minority group ethnicity. Note: The shaded region around each line indicates the standard error of the point estimate. For social function, respondents were asked, “In the past four weeks, how often have health problems affected your work or other daily activities?” Response options included 1 (Always), 2 (Often), 3 (Sometimes), 4 (Rarely), and 5 (Never). To align with the coding of overall health, these responses were relabeled as 1 (Very unhealthy), 2 (Somewhat unhealthy), 3 (Average), 4 (Somewhat healthy), and 5 (Very healthy). Higher scores indicated better social function

Fig. 5
Fig. 5
Full size image

Percentage of people selecting each frequency of depressive symptoms category by year, stratified by minority group ethnicity status: The Chinese General Social Survey (2010–2021) (a) Overall; (b) Stratified by minority group ethnicity. Note: The shaded region around each line indicates the standard error of the point estimate. Frequency of depressive symptoms was measured on a five-level Likert scale. Respondents were asked, “In the past four weeks, how often have you felt depressed or downhearted?” with responses: 1 (Never), 2 (Rarely), 3 (Sometimes), 4 (Often), and 5 (Always). Higher scores indicated more severe depressive symptoms

Main results

The cumulative logit plots were used to examine the proportional odds assumption and are shown in Supplementary Material Figure S1 to Figure S3. The parallel trend across cumulative logits demonstrates that the proportional odds assumption is met, allowing us to use a single OR to describe the relationship between predictors and the odds of choosing a higher category of health outcomes. The AIC showed that the full models (Model 5) had better fitness across all health outcomes even with additional predictors in the models (Fig. 8).

For overall health condition, the minority group ethnicity is associated with 14% higher odds of being self-reported healthier compared to non-minority ethnicity, after adjusting for individual-, institutional-, and societal-level factors (OR: 1.14, 95% CI: 1.07–1.22). This difference becomes 2.15 times after further adjusting for the interactions between minority ethnicity status and institutional factors and the interactions between region and institutional-level factors (OR: 2.15, 95% CI: 1.52–3.05). For social function, the minority group ethnicity is not significantly associated with being healthier compared to non-minority ethnicity after adjusting for individual factors alone, institutional- and societal-level factors alone, and both. The difference is 1.68 times after further adjusting for the interactions between minority ethnicity status and institutional-level factors and the interactions between region and institutional-level factors (OR: 1.68, 95% CI: 1.18–2.39). For the frequency of depressive symptoms, the minority group ethnicity is associated with a 12% decrease in the odds of being self-reported less depressed compared to non-minority ethnicity after adjusting for individual-, institutional-, and societal-level factors (OR: 0.88, 95% CI: 0.82–0.94). This difference is 36% lower odds after further adjusting for the interactions between minority ethnicity status and institutional-level factors and the interactions between region and institutional-level factors (OR: 0.64, 95% CI: 0.47–0.86) (Fig. 6). These results remained robust through sensitivity analysis (Fig. 6). Models using randomly generated negative outcomes as dependent variables produced estimates closer to the null compared to the primary analysis (Fig. 7), indicating that the observed disparity might not be due to chance.

Fig. 6
Fig. 6
Full size image

The association between minority group ethnicity and self-reported health conditions: The Chinese General Social Survey (2010–2021). Note: Error bars represent 95% confidence intervals. Estimates are statistically significant when confidence intervals do not cross 1.0. Model 1 included only minority ethnicity as a predictor. Model 2 adjusted for individual-level factors in addition to the minority group ethnicity. Model 3 only adjusted institutional- and societal-level factors. Model 4 adjusted for individual-, institutional- and societal-level factors. Model 5 was built on Model 4 by including interaction terms between the minority group ethnicity and institutional-level factors, and the interactions between region and institutional-level factors. Primary analysis was based on the complete case data. Sensitivity analysis 1 was based on the random forest imputed data. Sensitivity analysis 2 did not adjust for the interactions between region and institutional-level factors in Model 5

Fig. 7
Fig. 7
Full size image

Comparison of estimates between true and negative health outcomes: The Chinese General Social Survey (2010–2021). Note: Error bars represent 95% confidence intervals. Estimates are statistically significant when confidence intervals do not cross 1.0. Blue dots represented point estimates from the primary analysis, while grey dots represented point estimates using randomly generated negative outcomes. Grey dots were closer to the null compared to the blue dots

We also calculated McFadden’s pseudo-R-squared to determine the proportions of variance attributable to individual-, institutional-, and societal-level factors, and their interactions. For overall health conditions, individual-level factors accounted for 94.2% of the variance, institutional- and societal-level factors for 3.0%, and interactions for 2.8%. Similarly, for social function, individual-level factors explained 90.0% of the variance, institutional- and societal-level factors 6.1%, and interactions 3.8%. For the frequency of depressive symptoms, individual-level factors accounted for 75% of the variance, institutional- and societal-level for 7%, and interactions for 18% (Fig. 8).

Fig. 8
Fig. 8
Full size image

The proportion of variance of health outcomes explained by individual, institutional, and interaction between individual and institutional factors: The Chinese General Social Survey (2010–2021). Note: AIC: Akaike Information Criterion. Model 1 included only minority ethnicity as a predictor. Model 2 adjusted for individual-level factors in addition to minority ethnicity. Model 4 adjusted for individual-, institutional- and societal-level factors. Model 5 built on Model 4 by including interaction terms between minority ethnicity and institutional-level factors, and the interactions between region and institutional-level factors

In summary, the minority group ethnicity is associated with better self-reported overall health, social function, and the frequency of depressive symptoms. The variance in these health outcomes is predominantly explained by individual-level factors. Meanwhile, institutional- and societal-level factors and the interactions factors contributed.

Discussion

The results show that belonging to an ethnic minority group in China is associated with higher self-reported overall health, better social functioning, and lower levels of depressive symptoms. This observation is intriguing as it contrasts with the common health disparities observed between minority and majority ethnic groups in other contexts.

Notably, ethnic minorities demonstrated 36% lower odds of experiencing frequent depressive symptoms (OR: 0.64, 95% CI: 0.47–0.86) after full adjustment, which was the strongest protective association among our three health outcomes. This mental health advantage is especially significant as it suggests that the health benefits observed among ethnic minorities extend beyond physical health to psychological well-being. Although individuals in China may tend to underreport poor mental health due to social desirability bias and cultural stigma surrounding mental illness [47,48,49], the relative advantage observed among minority groups may still indicate the protective influence of strong community cohesion, cultural resilience, and social support networks, which can buffer against psychological distress despite socioeconomic disadvantages [50, 51].

The geographic and residential patterns observed provide important context for understanding these health advantages. The concentration of ethnic minorities in rural areas (51.6% to 62.6% vs. 26.8–39.5% for Han) and in southwest, northwest, and south regions creates a complex health dynamic. Many of these groups, including Tibetans and Uyghurs, inhabit relatively isolated regions with distinct cultural practices and traditional lifestyles [52, 53]. This geographic isolation strengthens within-group social cohesion - minority communities maintain dense kinship networks and reciprocal support systems that are culturally specific and linguistically accessible [54, 55]. These within-group networks, characterized by shared language, customs, and worldviews, provide more effective emotional and practical support than cross-ethnic interactions, which may involve language barriers, cultural misunderstandings, and reduced trust [56]. In contrast, Han-dominated urban areas often feature more individualistic and dispersed social networks with weaker community bonds [57].While rural residence typically implies reduced healthcare access, it may confer health advantages through these traditional support networks, protection from urban stressors, traditional dietary practices, and higher physical activity from agricultural lifestyles [58,59,60,61].

Several additional patterns warrant exploration. The low rates of religious belief across all participants (5.4%−11.3% for Han) reflect China’s secular socialist history and state policies promoting atheism [62]. However, the substantially higher rates among minorities (30.1–43.0%) likely reflect the integration of religious practice with ethnic identity, particularly among groups like Hui Muslims and Tibetan Buddhists, where religion is inseparable from cultural identity [14, 63, 64]. Religious institutions provide structured community support, regular social gatherings, and shared meaning-making frameworks that may further strengthen within-group social cohesion and promote psychological resilience [65,66,67,68].

The study’s results resonate with the ecosocial theory, which posits that health outcomes are influenced by both individual-, institutional- and societal-level factors [7, 8]. The substantial contribution of individual-level factors to the variance in health outcomes underscores the significance of personal characteristics, behaviors, and circumstances in shaping health [69]. In this study, we observed a notable increase in the ORs across all health outcomes when interaction terms between minority ethnicity and institutional factors, and interaction terms between region and institutional factors were included in Model 5. This finding underscores the importance of considering interactions between individual, institutional and societal-level factors, as emphasized by ecosocial theory [9]. The substantial increase in ORs suggests that the health advantages associated with minority ethnicity are context-dependent and may be amplified by specific societal conditions, such as healthcare resource availability or regional characteristics [31]. These results highlight the complex interplay between ethnicity and broader societal factors, supporting the notion that minority health outcomes cannot be fully understood without considering the broader context in which individuals live. Future research should further investigate these interactions to uncover the underlying mechanisms that may influence health disparities.

These findings also offer new perspectives on the application of the ecosocial theory in a global context. While extensively studied in the United States, applying this theory in different cultural and societal settings may yield varying results. The limited explanatory power of interactions between minority ethnicity and institutional factors in the Chinese context could be attributed to the unique dynamics and historical trajectories of ethnic relations in China. Sociocultural factors, such as the recognition and promotion of ethnic diversity within the Chinese national identity, might have influenced the interplay between ethnicity and institutional factors in shaping health outcomes [70].

We emphasize that the ethnicity and health outcomes are not causally linked; ethnicity itself does not denote biological differences. Health disparities among ethnic groups are primarily driven by societal factors. This study’s use of ecosocial theory underscores the importance of researcher choice in study focus. By examining ethnic health disparities in China, we highlight and analyze the unique health experiences of minority groups, enriching global scientific literature. Including diverse populations in health research is crucial for addressing knowledge gaps and promoting health equity. Researchers must ensure their findings are ethically sound and contextually accurate. Our results suggest several policy implications. First, contrary to assumptions that minorities universally experience health disadvantages, our findings indicate that existing social structures and community practices among Chinese ethnic minorities may provide health-protective factors that should be identified and preserved. Second, given that individual-level factors explained 75–94% of health variance, policies targeting socioeconomic development may be effective in improving population. Third, future research should employ qualitative methods to understand the mechanisms underlying minorities’ health advantages—such as specific social support structures, traditional practices, or community resilience factors—which could then inform interventions to improve health outcomes in the overall population.

Despite these compelling findings, there are several limitations of the study. First, due to sample size constraints, statistical comparisons between the Han majority and specific minority ethnic groups were not feasible. Future research employing oversampling strategies could address this limitation, providing more nuanced insights into the health experiences of individual ethnic minority groups. Second, the self-reported nature of both ethnicity and health outcomes introduces the possibility of reporting biases and discrepancies between self-perceived and externally perceived ethnicity. Self-reported race or ethnicity reflects individuals’ lived experiences and the internalized effects of societal racism, providing valuable insights that interviewer-assigned categories may miss, as argued by Lett et al. (2022) [39]. This perspective aligns with ecosocial theory, which emphasizes how societal forces are biologically embodied through social and economic deprivation, ultimately affecting health outcomes. Third, language and literacy factors may have influenced our findings. Although interviewers could translate questions into local languages, the lack of standardized translations may have introduced variability in responses. Educational disparities—with more minorities having no formal education (11.7–16.2% vs. 8.6–13.4% for Han)—may have affected comprehension of health concepts and scales, potentially contributing to observed ethnic differences. Future surveys should consider developing standardized questionnaires in ethnic minority languages to reduce potential measurement error and improve data quality.

Conclusion

This study enhances our understanding of the complex interplay between ethnicity, individual-, institutional-, and societal-level factors in shaping health outcomes within the Chinese context. While the findings challenge conventional assumptions about ethnic health disparities, they underscore the importance of considering sociocultural nuances and employing a contextualized approach when applying theoretical frameworks across diverse global settings.