Assessing the Green Space-Mental Health Link in Guangzhou: A Bayesian Geoadditive Quantile Regression Approach

Previous studies have established an association between green spaces and the mental health of residents. However, existing research predominantly relies on mean regression models predicated on the assumption of linear relationships, which may overlook potential nonlinear associations between these two variables. Furthermore, the impact of green spaces on mental health might vary across the upper, central, and lower quantiles of the response variable and could be modulated by other environmental and individual factors. In this study, the nonlinear relationship between green spaces and the mental health of residents across different quantiles (i.e., low, medium, and high mental health levels) was explored using questionnaire survey data collected in Guangzhou in 2021 and a Bayesian geoadditive quantile regression model to control for other individual and environmental characteristics. The effects of individual socioeconomic attributes, green space quality, and green space use behavior of the residents on mental health at varying levels were further analyzed. Additionally, spatial correlation was incorporated to capture unexplained geographical associations. The following results were found: (1) Gender showed a significant positive correlation with the low level of mental health. Educational attainment showed a significant negative association with mental health across all three quantiles. Age was positively correlated with mental health at the low, medium and high levels. Income showed a positive association with low and medium levels of mental health.The duration of green space use showed a significant positive correlation at the low mental health level. The frequency of green space use correlated significantly with mental health across different levels. (2) Among the green space quality factors, cleanliness was positively correlated with a low mental health status, whereas security showed positive associations across all mental health levels, with a stronger correlation at the high level. (3) A nonlinear relationship between green spaces and mental well-being was evident. At moderate greenness values (0.36~0.6), green spaces demonstrated a weak positive trend in their impact on mental health, whereas a certain negative trend in this effect was observed at high greenness values. The mean curve suggested an implicit threshold (0.6) at which the health benefits of green spaces become apparent. (4) The low mental health group exhibited a significant negative spatial effect, indicating the presence of unobserved regional risk factors. By contrast, no significant spatial effects were detected in the medium and high mental health groups, suggesting that the model covariates sufficiently explained the observed associations. Overall, BQR enables researchers to integrate prior knowledge via prior distributions and provides comprehensive probabilistic interpretations of model parameters. This allows for posterior distributions of parameters to be obtained, thereby enhancing our understanding of parameter uncertainty and complexity. Second, the method offers valuable insights into the alternative distributional characteristics of the response variable, information that mean regression fails to capture, thus enabling exploration of the impact of green spaces on different mental health levels beyond the average effect. This facilitates the identification of populations who will be the most adversely and favorably affected by green spaces in terms of mental health. The findings of this study have significant implications for promoting public green space utilization and formulating community-level health planning policies.