*3.3. Analytical Strategy*

First, the Statistical Package for the Social Sciences (SPSS) 23.0 was used to obtain descriptive statistics and correlations between the main variables. Second, we conducted mediation analysis using the stepwise regression method proposed by Mackinnon et al. [53] to examine the multiple mediating roles of risk perception and PAR adoption in the relationship between government response and infection rate. In the first step, we tested the effect of government response on risk perception and PAR adoption. Next, we used stepwise regression to compare the changes in the magnitude of the coefficients of the main explanatory variables in the model before and after the addition of the mediating variables, and make a preliminary determination of the possible mediating variables. We used the following regression model:

$$
\Upsilon = \mathfrak{\alpha} + \mathfrak{P}\mathfrak{X} + \mathfrak{S}\mathfrak{C} + \varepsilon \tag{1}
$$

$$\mathbf{M1} = \alpha + \beta \mathbf{X} + \delta \mathbf{C} + \varepsilon \tag{2}$$

$$\mathbf{M2} = \alpha + \beta \mathbf{\mathcal{X}} + \gamma \mathbf{M1} + \delta \mathbf{\mathcal{C}} + \varepsilon \tag{3}$$

$$\mathbf{Y} = \mathbf{x} + \beta \mathbf{X} + \gamma \mathbf{M} \mathbf{1} + \lambda \mathbf{M} \mathbf{2} + \delta \mathbf{C} + \varepsilon \tag{4}$$

where Y is the dependent variable (infection rate), X is the independent variable (government response), M1 is a possible mediating variable (risk perception), M2 is another possible mediating variable (PAR adoption), and C is a set of control variables including gender, age, household registration, years of schooling, health status, urbanisation rate, and region.

Finally, the PROCESS macro was used to examine the multiple mediating roles of risk perception and PAR adoption in the relationship between government response and infection rate. Model 6 from the PROCESS macro in SPSS, as developed by Hayes [54], was used to conduct a multiple mediation analysis and the bootstrapping method (sampling repeated 1000 times) was used to construct a 95% confidence interval.
