**3. Results**

Table 1 summarizes the means (standard deviations) and correlations of all variables. We found that people in urban areas reported slightly better mental health status than their rural counterparts as indicated by mean values of 0.05(SD = 0.97) and –0.07(SD = 1.04), respectively. Urban individuals also outperform rural ones in terms of wealth index, household income, and education. However, urban individuals reported lower social-economic status, smaller household size and higher unemployment rate. In terms of social capital, it seems that urban people reported much higher civic participation (0.37 vs. −0.53) but much lower civic trust (−0.18 vs. 0.25), political participation (−0.21 vs. 0.3), and political trust (−0.07 vs. 0.10) than rural ones. In addition, it seems that mental health status is positively associated with all four types of social capitals and married, education, household income, household wealth and social-economic statute. However, mental health status is negatively associated with female and age.

Regression results of Model 0, Model 1, and Model 2a–2e are summarized in Table 2. Both the individual-level variances *σ*2*ε* and the community-level variance *σ*2*μ* were all statistically significantly different to zero. The VPCs for the rural subsample ranged from 9.76% to 14.56%. And the VPCs for urban subsample were range from 9.76%. The VPCs indicate that substantial amount of vairance in mental health can be attributed to differences between communities. We then conducted the likelihood tests (see Table 3). We compared the loglikelihoods between Model 1 and 0. The differences ( Δ-2ll) with 18 degrees of freedom were statistically significant under the Chi-square tests, suggesting that human capital variables as a whole are important determinants of mental health status in both rural and urban areas. Likewise, we compared Model 2a–2e to Model 1-Rural/1-Urban, respectively. The differences ( Δ-2ll) with two degrees of freedom were statistically significant under the chi-squared tests in most models, except the one between Model 2b-Rural and Model 1-Rural. The results sugges<sup>t</sup> that social capital as a whole and CT, PP, and PT are important determinants of mental health for both rural and urban individuals; while CP is an important determinant for urban individuals.

On closer inspection of Model 2a–2e, we noticed that the individual-level aggregated SCI was positively associated with mental health (0.01, *p* < 0.01) in both subsamples. However, the community-level aggregated SCI was positively associated with mental health only in the rural subsample (0.02, *p* < 0.05). In urban areas, the impacts of CP on mental health were significantly positive at the individual level (0.01, *p* < 0.01) but negative at the communitive level ( −0.03, *p* < 0.01). The impacts of individual-level CT (0.1, *p* < 0.01) was statistically significant in rural area. The community-level CT had statistically significant impacts in both rural and urban areas (0.06 and 0.07, *p* < 0.01, respectively), which justified the contextual effects in [39].

What is more, the individual-level PP was positively associated (0.03, *p* < 0.01) with mental health in both rural and urban areas. Finally, the individual level PTs (0.02, *p* < 0.01; 0.02, *p* < 0.01) were positively associated with mental health in both rural and urban areas.



participation; PT: political trust; VPC: variance partition coefficient. Indi: Individual-level. Dist: District-level.


**Table 3.** The likelihood chi-squared tests.

Δ-2ll = The –2ll value of the first model minus the –2ll value of the second model.
