The Internal Mechanism of Academic Performance’s Influence on Adolescents’ Subjective Well-being

To test the research model, this study used a partial least squares structural equation model (Lohmoeller, 1989) and estimated using SmartPLS4.1 software (Ringle et al., 2015). Compared to traditional covariance based structural equation models, variance based structural equation models have multiple advantages in terms of methodology. Firstly, as a component-based estimation method, it can obtain stable parameter estimation results even in small sample sizes (Hair et al., 2017). Secondly, this method does not require data to follow a multivariate normal distribution and has good adaptability to skewed indicator distributions and formative latent variables (Henseler et al., 2009). In addition, the model can effectively handle situations where the paths between variables are complex, latent variables are highly correlated, and formative constructs coexist (Hair et al., 2017), while alleviating the problem of multicollinearity by projecting the predictor variables onto the latent space with the highest covariance with the outcome variables (Tenenhaus et al., 2005). In recent years, researchers have proposed some fit indicators for model setting tests, but overall, the overall fit evaluation of variance type structural equation models is still developing. Therefore, this study mainly focuses on the predictive validity and measurement quality of the model, and only serves as an auxiliary reference for the model fitting index (Hair et al., 2019). This study follows the two-stage logic of structural equation modeling analysis (Anderson&Gerbing, 1988). In the first stage, the reliability and validity of the external measurement model are tested; In the second stage, evaluate the path relationship and effect strength of the internal structural model. Before entering the second-order model analysis, test the measurement models of the four first-order dimensions of psychological capital - hope, optimism, resilience, and self-efficacy. The results showed that the reliability and validity of each dimension were good, and no items were deleted to maintain theoretical integrity and content validity. Subsequently, using a two-stage analysis method, the factor scores of the four first-order latent variables were used as reflective indicators of the second-order psychological capital variable and included in the subsequent structural model analysis. In the modeling process, consider academic performance as a multi class independent variable. According to the multi class variable modeling standard, two dummy variables were generated based on the "average score group" and included as formative latent variable paths in the model to examine the impact of different academic levels on the main structural path. In the structural model analysis, this study used 5000 self sampling methods to test the significance of path coefficients and calculated the 95% bias corrected accelerated confidence interval (95% BCa CI) for direct effects and specific indirect effects. When the upper and lower limits of the confidence interval cross zero, the effect is considered insignificant (Henseler et al., 2014). In addition, the path coefficient and determination coefficient were reported, and the model fitting indicators were supplemented to assist in evaluating the robustness and theoretical explanatory power of the model.