2.2.5. Statistical Analysis

To examine the hypothesized multiple mediation model for the association between perceived social support and active coping with COVID-19, which was mediated by risk perception or confidence (Figure 1), the following analyses were conducted using SPSS and AMOS version 23.0 for Windows (SPSS Inc., Chicago, IL, USA). We examined bivariate associations among the variables using Pearson's correlation coefficient (*r*), followed by two steps of structural equation modeling (SEM). First, confirmatory factor analysis (CFA) was used to verify the association between latent variables and their indicators in the measurement model. Each question was composed of observed variables (indicators) and latent variables, which indicated perceived social support, active coping with COVID-19, risk perception, and confidence. Factor loading was used as an index to assess the scale reliability between indicators and the corresponding latent variables in the CFA. In addition, Cronbach's α was calculated to examine the internal consistency reliability. The range was considered acceptable if Cronbach's α was >0.5 [27]. To estimate the sample adequacy of "active coping with COVID-19" in factor analysis, the Kaiser–Mayer–Olkin (KMO) measure of sampling adequacy and Bartlett testing were applied. A KMO value of >0.60 and statistically significant value of *p* < 0.05 from Bartlett testing indicated the data was adequate for factor analysis [28]. Then, the total variance explained (%) was also estimated through EFA to estimate the validity of "active coping with COVID-19".

**Figure 1.** The conceptual model of mediating effect.

Latent variable path analysis with maximum likelihood parameter estimations was used to estimate the model adequacy and the direct/indirect effects of perceived social support on active coping with COVID-19 through risk perception or confidence [29]. Bootstrapping method with 5000 samples was applied in the path analysis due to the non-normality of the data (Kolmogorov-Smirnov test; *p* < 0.001). As a multiple mediator model, both mediators were applied into the model to assess and compare the mediating effects. As there was a relatively high proportion of females in the study cohort and as the Kolmogorov-Smirnov test (*p* < 0.001) for age was significant, indicating non-normal distribution, age and gender were also included in the multiple mediators' model as covariates to adjust for their effects on the latent variables. Gender (female, male and transgender) was transformed into two dichotomous dummy variables (male vs. female; and transgender vs. female) for the analysis. The standardized estimates (beta coefficient; β) were reported for the predictive strength explained in the model.

We used the Sobel test to verify the mediating effect [30]. Furthermore, to test the adequacy of the model, multiple indices were applied to verify the goodness of fit. For each of these fit indices, the values indicating an acceptable model fit were as follows: Goodness of Fit Index (GFI ≥ 0.9); Adjusted Goodness of Fit Index (AGFI ≥ 0.9); root mean square error of approximation (RMSEA < 0.08); and standardized root mean square residual (SRMR ≤ 0.08) [31,32].

#### **3. Results**
