The China Health and Nutrition Survey (CHNS) is an ongoing large-scale study aimed at determining how China's social and economic development affects the health and nutritional status of the country's population. It is conducted by the Carolina Population Center at the University of North Carolina in Chapel Hill and the National Institute for Nutrition and Health (formerly the National Institute of Nutrition and Food Safety) at the Chinese Center for Disease Control and Prevention. A multistage, random cluster process was used to draw a sample of ~7,200 households comprising more than 30,000 individuals across 15 provinces and municipal cities. The CHNS data-collection rounds were completed in 1989, 1991, 1993, 1997, 2000, 2004, 2006, 2009, 2011, and 2015. Our analysis draws on four waves-2006, 2009, 2011, and 2015—during which data on mental health issues and sedentary behavior were available. We focus on adults who have completed their formal education, resulting in a total of 23,693 observations.

We choose the ordinary least squares (OLS) method to test the relationship between leisure sedentary time and mental health issue in Hypothesis 1. The equation for the method is as follows (see Equation 2):

Where Y_it is dependent variable for individual i at wave t, and ST_it represents individual i's leisure sedentary time at wave t. X_it is a set of control variables. δ_i, δ_f, and δ_commid, represent the fixed effects at individual-level, household-level, and community-level, respectively. τ_t represents year-fixed effect. β₁ is the main coefficient of interest, β₁>0 indicates that heightened leisure sedentary time negatively affects mental health issues.

Prior studies suggest that inconsistent findings regarding the relationship between sedentary time and mental health may stem from confounding factors, causality issues, and measurement errors (72). This necessitates a comprehensive review of endogeneity concerns.

One potential source of endogeneity, in the context of our study, is omitted variable bias. There are several unobservable factors that are likely to be correlated with both leisure sedentary time and mental health issues and, thus, in a multivariate regression framework, it is difficult to rule out more than one omitted variable, making it impossible to clearly predict the direction of bias. For example, adverse childhood experiences are associated with both higher sedentary behaviors, particularly greater screen time, and worse health conditions (73, 74).

While the fixed effect model is generally known to be effective in addressing omitted variable bias, endogeneity may also emerge from potential measurement error and simultaneity bias. Regarding measurement error bias, previous literature has found that leisure sedentary time was collected through questionnaires, inherently resulting in an underestimation of leisure sedentary time (75). Therefore, β1^ will underestimate β₁ if β₁ is positive. Concurrently, emerging evidence suggests bidirectional associations between sedentary behaviors and mental health issues (76). Individuals with mental health issues are generally less active, and sedentary behavior increases the risk of mental health issues. Thus, simultaneity bias should be considered.

Good instrumental variables must satisfy two assumptions. The first assumption is that they should be strongly correlated with the variable being instrumented. The second assumption is that the instrumental variables are orthogonal to the outcomes. To meet the above assumptions, this paper employs three instrumental variables.

First, building on Munasib and Bhattacharya (77), we adopt Lewbel's higher-moment approach (78). This method, initially designed for measurement error models, is widely applicable in contexts with general correlated-regressor error and multilevel structures (79). It constructs instrument using the third-order centered moment of leisure sedentary time. By design, this variable is strongly correlated with leisure sedentary time but unlikely to be correlated with mental health issues unless a non-linear relationship exists. This approach has been extensively validated in prior studies [e.g., (80, 81)].

Second, Duflo and Saez suggest using the average exogenous characteristics of group-level variables as instruments for endogenous independent variables (82). Therefore, inspired by the peer group effect theory (83), we employ the average leisure sedentary time in the household of individual i as second instrumental variable (group-level IV). The average leisure sedentary time in the household of individual i is strongly influence individual i's leisure sedentary time under the family norms. However, this variable is unlikely to correlate with individual time-varying unobservable characteristics once fixed effects are accounted for (84).

The third instrument is the number of TV channels available to a household. This variable is exogenous, as the availability of television signals in China is determined by national and local broadcasting authorities. At the same time, it is associated with the amount of time spent watching television, making it a valid instrument for leisure sedentary time.

These IVs collectively address the potential endogeneity issues in the relationship between sedentary time and mental health outcomes.

Despite we employ instrumental variable method to address endogenous concerns, the results are still affected by unobserved factors. Therefore, we draw from National Digital TV Development Plan as an exogenous variation to examine the effect of leisure sedentary time on mental health issues under the continuous difference-in-differences (DID) method. The logic of DID method is that the prevalence of high-definition digital television after policy greatly affects the screen time, thereby affecting mental health.

To meet the demand for high-definition television broadcasts of the 2008 Beijing Olympics, the State Administration of Radio, Film, and Television (SARFT) accelerated the promotion of high-definition digital television. High-definition TV programming and event broadcasts required compatible digital TV and set-top boxes, which significantly boosted demand for set-top boxes. As early as 2003, SARFT released the Digital TV Development Plan and accelerated its implementation in 2008. The plan mandated that by 2010, the digitalization of cable television networks should be largely completed. This target pushed local television stations to speed up the rollout of digital TV. Meanwhile, SARFT reinforced the requirement for the gradual shutdown of analog television signals in 2008, promoting the replacement of analog TV with digital TV. This policy further drove demand for set-top boxes since consumers needed them to receive digital signals. Financially, to accelerate the adoption of digital TV, some local governments introduced subsidy policies for set-top boxes in 2009. Consumers could receive price discounts when purchasing set-top boxes, which further boosted sales. Considering the timing of the plan, we choose 2009 as the cutoff. The regression framework of the continuous DID design is typical written as Equation 3 as follows.

Where the dummy variable D_it is equal to 0 for years from 2009 onwards and 1 otherwise, and other specifications are similar to Equation 2.

To examine the relationship between leisure sedentary time and mental health issues, we estimate Equation 2, with results presented in Table 2. Column (1) uses aggregated mental health scores as the dependent variable, while column (2) uses standardized mental health scores. Both coefficients are positive and statistically significant. Specifically, we find that, on average, one additional minute of leisure sedentary time increases aggregated mental health scores by 0.0010 and standardized mental health scores by 0.0001. A back-of-the-envelope calculation indicates that one day of sedentary behavior increases aggregated mental health scores by 0.1430 and standardized scores by 0.0143. These findings suggest a consistent conclusion: increased leisure sedentary time is associated with mental health issues, supporting Hypothesis 1. However, these results should be interpreted cautiously, as the baseline estimation may be affected by endogeneity issues.

Our analysis also shows that marriage is strongly associated with better mental health. This aligns with existing evidence that marriage improves health through mechanisms such as social support, health behavior regulation, and access to resources that protect health (85). Similarly, physical health, higher family income, and better health scores are positively associated with improved mental health. In contrast, higher housing scores are negatively associated with mental health. One possible explanation is that rapidly rising housing prices in China during the survey period increased stress and mental health risks. Furthermore, individuals diagnosed with mental health issues or genetic disorders tend to report poorer mental health outcomes. Interestingly, we find that smoking is positively associated with mental health. While most studies highlight the adverse impact of smoking on mental health, while some smokers report that smoking helps improve mood, reduce stress, or act as a form of coping (86). We also observe that occupational sedentary time is positively linked to mental health. This may be explained by a shift in sedentary behavior patterns during the survey waves, where leisure sedentary time increased while occupational sedentary time decreased (87). Lastly, dietary characteristics are significantly associated with mental health. For instance, individuals who prefer vegetables tend to have better mental health, while those with a preference for sweets report poorer mental health outcomes.

To alleviate endogenous concerns, Table 3 shows the instrumental variable regression results. Column (3) shows the first-stage IV estimates, indicating a significant positive relationship between the instruments and the endogenous independent variable. The first-stage F-test yields an F-statistic of 1,541, which exceeds the threshold for weak instruments. Additionally, the p-value of the Kleibergen-Paap rk LM statistic is 0.000, indicating that the null hypothesis is rejected, and the instrumental variable passes the unrecognizable test. Concurrently, Sargan test p-values exceed 0.1, confirming the exogeneity of the instrumental variables. Hence, our instrumental variable selection is reasonable. Moving to column (1) and column (2), the result show that the regression coefficients are 0.0017 and 0.0001, respectively, and are significant at the 1% level, which are consistent with the baseline regression. Specially, the coefficients in column (1) and column (2) are larger than those in Table 2, suggesting a potential downward bias in the estimates when endogeneity is not addressed. To examine the valid of Lewbel's IV, Table 4 further reports the results of non-linear relationship between leisure sedentary time and mental health issues. The results show this non-linear relationship is unlikely to be hold, confirming the validity of Lewbel's IV.

Table 5 further presents the results of DID method. The result of the first two columns show that heightened leisure sedentary time is linked with worse mental health issues. Similarity, the coefficients of the first two columns are larger than those in Table 2, suggesting a potential downward bias in the estimates when endogeneity is not addressed. A back-of-the-envelope calculation suggests that one-day sedentary behavior significantly increases in aggregated mental health scores by 0.315 on average, and standardized mental health scores by 0.0143. A basic premise for using the DID model to evaluate policy effects is that the development trends of the explained variables for the treatment and control groups should be consistent before policy implementation in the pilot. In other words, there should be no statistically significant difference between the treatment and control groups before policy implementation. Therefore, it is necessary to conduct a parallel trend test for the DID model. In this study, we use the period prior to policy implementation, specifically 2009, as the reference group. As shown in Column (3), the coefficients of the parallel trend are insignificant before the policy. In 2011, the first period after policy implementation, the coefficients of the parallel trend are insignificant, which may be attributed to policy lag. However, in the second period after policy implementation, namely in 2015, the coefficients of the parallel trend become statistically significant. It indicates that the DID model in this study passes the parallel trend test, thereby validating the regression results presented in Table 5 to some extent.

The empirical results show that increased leisure sedentary time negatively impacts mental health. Therefore, mitigating this adverse effect is a critical goal for policy interventions and health-promoting behaviors. In this section, we examine moderating factors that may reduce the negative impact of leisure sedentary time on mental health.

First, we examine the moderating role of health literacy. In doing so, we use two measures: whether individuals are familiar with the Chinese Dietary Guidelines (CDG) and whether they have family members who are medical professionals. Familiarity with the CDG reflects a tendency to seek health-related information, while having medical professionals in the family indicates greater health literacy within the household. Based on that, we divide the sample into two groups: those unfamiliar with the CDG vs. those familiar, and households without medical professionals vs. those with medical professionals. Table 6 presents the results. For individuals unfamiliar with the CDG, the coefficient for leisure sedentary time is significantly positive at the 10% level. In contrast, for those familiar with the CDG, the coefficient is positive but not statistically significant at the 10% level. Similarly, households without medical professionals show a significantly positive coefficient at the 5% level, while households with medical professionals exhibit an insignificant coefficient, despite its larger magnitude. These findings indicate that health literacy moderates the relationship between leisure sedentary time and mental health issues, supporting Hypothesis 2a.

Next, we examine whether reducing prolonged sedentary time improves mental health. We first divide the sample based on the number of accessible TV channels, assuming a correlation between leisure sedentary behavior and television availability. Using the National Digital TV Development Plan as a benchmark, we categorize samples into two groups based on the average number of TV channels before and after 2009 (23 vs. 31 channels). Second, leveraging the longitudinal nature of the CHNS data, we group individuals based on changes in sedentary time over survey years, distinguishing between increased and reduced sedentary time. Table 7 presents the regression results. In the first two columns, households with more TV channels show a stronger negative impact of sedentary time on mental health, reflected by larger coefficients and greater statistical significance. In the last two columns, reduced sedentary time is associated with larger coefficients, though not statistically significant, while increased sedentary time has significant negative effects with a greater magnitude than the baseline regression. These results support Hypothesis 2c.

Finally, we focus on older adults and examine the moderating roles of caregiver gender and social interaction. To investigate Hypothesis 3b, we use the gender composition of caregivers. Older adults are assumed to interact more with female caregivers when daughters constitute the majority of caregivers, as the CHNS does not explicitly ask about the primary caregiver's gender. We also examine the role of urban parks as facilitators of social interaction. Parks provide spaces for physical activity, relaxation, and community engagement, which enhance mental health and wellbeing (88). While beneficial for all age groups, parks hold particular significance for older adults (89). Using the question “Does your house have a park nearby?”, we classify older adults into two groups: those with nearby parks and those without. We hypothesize that proximity to parks promotes social activities, reduces sedentary time, and improves mental health. Table 8 reports the results. In households with predominantly female caregivers, older adults experience fewer mental health issues, as evidenced by smaller coefficients in the first two columns. Regarding social interaction, older adults living near parks exhibit insignificant coefficients, suggesting that social activities mitigate the adverse mental health effects of sedentary behavior. In contrast, individuals without park access show significant and larger coefficients, indicating that limited social interaction exacerbates mental health problems. In summary, these findings support Hypotheses 3a and 3b.

The literature suggests that obesity is a potential channel underlying our results. However, it remains necessary to test whether this channel is valid in our context. To address this, we use BMI, overweight status, and waist circumference as mediating factors. In the CHNS, weight is measured to the nearest 0.1 kg using a calibrated beam scale with participants wearing lightweight clothing, while height is measured to the nearest 0.1 cm using a portable stadiometer with participants barefoot (90). BMI (kg/m²) is calculated as a mediating factor, and overweight status is defined as a BMI > 25 based on Chinese criteria (91). Waist circumference is measured at the midpoint between the bottom of the rib cage and the top of the iliac crest at the end of exhalation. Table 9 presents the empirical results. The findings show that increased leisure sedentary time is associated with higher BMI, a greater likelihood of being overweight, and higher waist circumference. These results support Hypothesis 4.

In our analysis, we control for occupational sedentary time to address concerns about omitted variable bias in the relationship between occupational sedentary time and mental health. To further examine the role of occupational sedentary time, we conduct a heterogeneous analysis by occupational type. The sample is divided into three categories: light physical work, medium physical work, and heavy physical work. Table 10 presents the results. Individuals engaged in light and medium physical work show a higher risk of mental health issues compared to those in heavy physical work. Moreover, the coefficients in the first two columns are larger than the baseline results, suggesting that individuals with high levels of both leisure and occupational sedentary time face worse mental health outcomes.

Our analysis focuses on adults who have completed formal education as the main sample. To test whether this sample restriction affects the generalizability of the findings, we perform several subsample analyses to ensure robustness. First, we exclude groups that may bias the results, including the oldest-old population (aged ≥80 years), individuals reporting poor health status, those diagnosed with genetic diseases, recently retired individuals, and individuals diagnosed with muscle diseases. The first three groups are more likely to experience mental health issues, while the last two are more likely to engage in sedentary behavior. Additionally, we randomly select 80% of the sample to further test the model's robustness. Tables 11, 12 present the robustness checks. The results across all subsamples remain consistent with the baseline findings, confirming the reliability of the empirical results.