*4.1. Measurement Model*

Confirmatory factor analysis (CFA) was first conducted to confirm the fitness of the measurement model to the research data before structural model testing. The measurement model included five latent constructs and 22 observed indicators. In the CFA, we allowed the latent variables to correlate with each other, and the observed indicators were restricted to load only on their associated constructs. The CFA results showed that all the fit indexes were within acceptable ranges, except the *χ*2/*df* value (*χ*<sup>2</sup> = 2626.013; *df* = 199; *χ*2/*df* = 13.196; GFI = 0.935; CFI = 0.955; IFI = 0.955; TLI = 0.947; SRMR = 0.046; and RMSEA = 0.057 (90% CI: 0.056, 0.059)). However, given the large sample size of the current study, the model fit was considered satisfactory [55]. In addition, the standardized factor loadings of all the indicators were significant and larger than the benchmark of 0.50 (from 0.609 to 0.963) [58]. Figure 2 displays the results of the measurement model.

**Figure 2.** Results of the CFA.

Furthermore, we ran Harman's one-factor test to examine the common method variance in the data [59]. We compared the fit of a single (common method) factor model with the proposed five-factor model. The results showed that the single factor model (with all the items loaded onto one latent construct) had an unsatisfactory fit to the data (*χ*<sup>2</sup> = 28,647.504; *df* = 209; *χ*2/*df* = 137.069; GFI = 0.457; CFI = 0.469; IFI = 0.469; TLI = 0.413; SRMR = 0.186; and RMSEA = 0.192 (90% CI: 0.190, 0.194)). The chi-square statistic (Δ*χ*<sup>2</sup> = 26,021.491, Δ*df* = 10, *p* < 0.001) also revealed that the measurement model provided a significantly better fit to the data than the single-factor model. Thus, common method variance was not significant in the present study.

Reliability and validity were assessed after the CFA analysis. As is presented in Table 3, the results for the composite reliability (CR) were between 0.724 and 0.944, which was higher than 0.7, indicating an acceptable level of internal consistency [60]. Additionally, the average variance extracted (AVE) scores ranged from 0.471 to 0.772 and were greater than the threshold value of 0.40, suggesting adequate convergent validity [61].


**Table 3.** Standard factor loading of items and reliability of the scales.

Note: CR = composite reliability; AVE = average variance extracted.

As can be seen in Table 4, all of the correlation coefficients among the variables were significant and had the anticipated sign. Specifically, the institutional climate was positively correlated to preventive behaviors (r = 0.343, *p* < 0.001). Attitudes (r = 0.192, *p* < 0.001), subjective norms (r = 0.405, *p* < 0.001) and perceived behavior control (r = 0.407, *p* < 0.001) were each significantly associated with preventive behaviors. The institutional climate was

also significantly correlated with attitudes (r = 0.055, *p* < 0.01), subjective norms (r = 0.317, *p* < 0.001) and perceived behavior control (r = 0.352, *p* < 0.001). These correlations met the conditions for mediation suggested by Baron and Kenny [62]. Moreover, as the square roots of the AVEs for all of the constructs were higher than the correlations among them, the discriminant validity of the measurement was confirmed [61].

**Table 4.** Discriminant validity and correlation.


Note: Diagonal elements (in italics) are the square root of the average variance extracted (AVE). \*\* *p* < 0.01. \*\*\* *p* < 0.001.

### *4.2. Structural Model*

SEM analysis was employed to evaluate the hypothesized paths in the structural model. The analysis revealed an acceptable fit of the proposed structural model to the data (*χ*<sup>2</sup> = 3065.544; *df* = 202; *χ*2/*df* = 15.176; GFI = 0.928; CFI = 0.946; IFI = 0.947; TLI = 0.939; SRMR = 0.069; and RMSEA = 0.062 (90% CI: 0.060, 0.064)). Then, the statistical significance of the path coefficients among the constructs was estimated. As is demonstrated in Figure 3, all the direct paths were statistically significant. First, the institutional climate had significant effects on the attitudes (*β* = 0.057, *t* = 3.066, *p* < 0.01), subjective norms (*β* = 0.329, *t* = 19.026, *p* < 0.001) and perceived behavioral control (*β* = 0.437, *t* = 19.417, *p* < 0.001). Second, the direct effect of the institutional climate on preventive behaviors was significant (*β* = 0.148, *t* = 7.980, *p* < 0.01). Third, the attitudes (*β* = 0.163, *t* = 9.521, *p* < 0.001), subjective norms (*β* = 0.243, *t* = 13.351, *p* < 0.001) and perceived behavioral control (*β* = 0.308, *t* = 13.158, *p* < 0.001) exerted significant impacts on preventive behaviors.

**Figure 3.** Results of the SEM.

We ran a bootstrapping analysis to further verify the mediation effects in the hypothesized model. As is revealed in Table 5, both the direct and indirect effects of the institutional climate on preventive behaviors were significant (all 95% bias-corrected CI did not include 0), suggesting that the link between the institutional climate and preventive behaviors was partially mediated by attitudes, subjective norms and perceived behavioral control. The results indicated that university students with high perception of the institutional

climate tended to express more favorable attitudes, stronger subjective norms and greater perceived behavioral control toward COVID-19 prevention, which could promote the development and performance of preventive behavior. Thus, H1, H2a, H2b and H2c were supported.

**Table 5.** Results of bootstrapping.


Note: IC = institutional climate; AT = attitudes; SN = subjective norms; PBC = perceived behavioral control; BE = preventive behaviors; LLCI = lower level confidence interval; ULCI = upper level confidence interval. \*\*\* *p* < 0.001.

#### *4.3. Moderating Effects*

Multigroup SEM analyses were employed to examine the moderating effects of perceived risk in the structural model. The sample was divided into two subgroups of high and low risk perception students using the median split approach. Next, we conducted a chisquare difference test to compare a constrained model (all the paths were restricted across the two subgroups) with an unconstrained model (all the paths were not constrained across the two subgroups). If the constrained model presented a significantly larger chi-square value than the constrained model, then this implied a potential moderating effect [60]. In each model, factor loadings between the two groups were held equivalent to ensure that the variables were measured similarly across groups; however, error variances were permitted to vary between groups [63]. The chi-square statistic demonstrated that the constrained (*χ*<sup>2</sup> = 3131.722, df = 428) and unconstrained models (*χ*<sup>2</sup> = 3037.387, *df* = 421) were significantly different (Δ*χ*<sup>2</sup> = 94.335, *df* = 7, *p* < 0.001), supporting the moderation effect of perceived risk on structural relationships.

To accurately detect the moderating effects of perceived risk on specific paths in the proposed model, a battery of chi-square difference tests was applied to compare the constrained models with seven diverse models separately, each retaining only one of the structural paths to be freely estimated. As is illustrated in Table 6, perceived risk significantly moderated four of the seven structural relationships. Specifically, the effect of the institutional climate on preventive behaviors was stronger for high risk perception students (*β* = 0.251, *t* = 8.594, *p* < 0.001) than for low risk perception students (*β* = 0.093, *t* = 3.831, *p* < 0.001). The effect of the institutional climate on subjective norms was stronger for high risk perception students (*β* = 0.379, *t* = 13.115, *p* < 0.001) than for low risk perception students (*β* = 0.292, *t* = 12.129, *p* < 0.001). The influence of the institutional climate on perceived behavioral control was significantly stronger among high risk perception students (*β* = 0.528, *t* = 15.177, *p* < 0.001) than among low risk perception students (*β* = 0.376, *t* = 13.045, *p* < 0.001). Moreover, high risk perception students (*β* = 0.281, *t* = 10.011, *p* < 0.001) exhibited a larger path effect than low risk perception students (*β* = 0.200, *t* = 8.159, *p* < 0.001) in the influence of subjective norms on preventive behavior. However, the results did not suggest the existence of significant differences between high and low risk perception groups regarding the effect of the institutional climate on attitudes, as well as the effect of attitudes and perceived behavioral control on preventive behaviors. Thus, H3 was partially supported.


**Table 6.** Results of the multigroup analysis.

Note: IC = institutional climate; AT = attitudes; SN = subjective norms; PBC = perceived behavioral control; BE = preventive behaviors; PR = perceived risk. \* *p* < 0.05. \*\* *p* < 0.01. \*\*\* *p* < 0.001.

#### **5. Discussion and Implications**

The aim of the current study was to investigate the influencing factors of preventive behaviors for COVID-19 among university students in Beijing, China. With an extended TPB framework, we tested the hypothesized relationships among the institutional climate, three components of the original TPB model and preventive behaviors, as well as the moderating role of perceived risk in the structural relationships. The major research findings are summarized and discussed as follows.

Based on the extended TPB model, we found that the institutional climate was significantly associated with university students' preventive behaviors against COVID-19. Consistent with previous studies [35], the results imply that a positive institutional setting with formal policies, procedures and practices concerning COVID-19 prevention and control could enable university students to adaptively face epidemic challenges and facilitate their preventive actions against COVID-19. In addition, countries with different strengths of social norms (or cultural tightness–looseness) were varied in their effectiveness to combat COVID-19 [64]. Thus, a possible explanation for this relationship may be that an institutional climate creates social norms, duties, obligations and expectations within a specific institution that reinforce the preventive behaviors of students, especially those from tight cultures and collectivist societies such as China [48,64]. Moreover, according to the focus theory of normative conduct [65], the extent to which university students' preventive behaviors are practiced is highly dependent on the saliency and level of HEIs' COVID-19 prevention and control measures.

As expected, the results indicate that the institutional climate was significantly related to the three original TPB components, which in turn yielded a significant effect on preventive behaviors. The mediating effects of university students' attitudes, subjective norms and perceived behavioral control on the relationship between the institutional climate and preventive behaviors were supported via a bootstrapping procedure. Specifically, all three TPB components partially mediated the relationship between the institutional climate and preventive behaviors. These results indicate that attitudes, subjective norms and perceived behavioral control are critical sociopsychological factors that link institutional intervention and students' actual preventive behaviors toward COVID-19. The results suggest that with increasing emphasis on formal policies, procedures and practices concerning the prevention and control of COVID-19 on campus, university students may be expected to adopt more preventive behaviors, which requires them to possess an understanding of not only COVID-19 prevention knowledge, requirements and recommendations but also a positive emotional disposition, strong perception, substantial normative stimuli and the motivation to perform preventive behaviors; that is, the accessibility of external support, resources and information for COVID-19 prevention might lead to the enhancement of preventive behaviors by shaping the positive environment needed for university students' active precautionary beliefs to flourish.

Multigroup SEM analyses indicated that perceived risk significantly moderated several paths in the research model. We found that the impacts of the institutional climate on

both subjective norms and perceived behavioral control were significantly stronger among university students with a higher level of risk perception than among those with a low level of risk perception. Our study also demonstrated that the influences of the institutional climate and subjective norms on university students' preventive behaviors were moderated by the perceived risk of COVID-19. Specifically, compared with students with a low level of perceived risk, those with a high level of perceived risk derived more benefits from the institutional climate in terms of the promotion or maintenance of preventive behaviors. These findings are highly similar to those of a recent study that found a moderating role of risk perception on the relationships among institutional factors, self-efficacy and compliance with prevention measures in Italian residents during the COVID-19 outbreak [46]. This may be explained by the fact that high risk perception students attempted to reduce their uncertainty and anxiety by resolving to accept preventive support, opinions or information from affiliated institutions and important figures and to enact preventive behaviors more strictly, while low risk perception students may have depended more on their own ability and judgment [66]. Moreover, our study revealed that the effect of the institutional climate on attitudes, as well as the influence of attitudes on preventive behaviors, remained invariant across the high and low risk perception groups. It can be concluded that, regardless of the level of university students' perception of the risk related to COVID-19, a higher level of perception of the supportive institutional climate toward COVID-19 prevention stably fostered the formation of a positive attitude toward adopting preventive behavior and, in turn, resulted in increased performance of actual behaviors.

Our study has the following theoretical implications. First, it broadens the research on individuals' preventive behaviors against COVID-19 from an institutional impact perspective with an expanded TPB model within the context of higher education. Although the institutional climate is known to be a key contextual factor for promoting individuals' disease prevention actions, empirical evidence on the association between the institutional climate and preventive behaviors for COVID-19 is limited. We examined the direct influence of the institutional climate on the preventive behaviors of university students in Beijing, China to fill this gap in the literature. Second, to the best of our knowledge, this is the first attempt to explore quantitative evidence in the potential role of TPB core constructs for bridging the relationship between institutional factors and university students' preventive behaviors toward COVID-19. Third, our study incorporates perceived risk as a moderator into the TPB model, thus providing more comprehensive insights into the influence mechanism of the institutional climate and TPB components on preventive behaviors. Moreover, our study verifies the scalability and versatility of the extended TPB model as a powerful theoretical basis for future studies of the COVID-19 preventive behaviors of other groups of people from diversified organizations around the world.

Regarding the practical implications, the findings of our study contribute to supporting HEIs' vital functions in the "new normal" period of COVID-19 in China and offer meaningful information for authorities and HEIs to encourage the adoption of preventive actions among the general public and to prevent the spread of COVID-19. First, by making COVID-19 an urgent and vital political issue, institutional actors can play a powerful and effective role in shaping the social norms of epidemic prevention [67], because political engagement and social norms represent crucial factors in facilitating prosocial behavior [68]. Accordingly, HEIs could prompt the creation of an institutional climate for COVID-19 prevention via a series of institutional interventions, including establishing effective prevention and control measures and demonstrating commitment and concrete efforts to ensure the physical and mental health and safety of students and staff on campus and to maintain the normal functions of the institutions. Second, HEIs should contribute to the management and intervention of students' positive psychological states, which will guide students in deciding which behaviors and protocols to pursue. Thus, we suggest that HEIs configure platforms to provide positive psychological interventions to students to stimulate them to enhance their knowledge, attitudes, norms and behavioral control toward COVID-19 prevention. Moreover, specific institutional interventions might be more

efficient for individuals with a high level of risk perception. We propose that HEIs emphasize that more risk and crisis education is especially helpful for enhancing students' beliefs regarding the obligations of the country, institutions and themselves to make successful efforts to defeat COVID-19.

### **6. Conclusions**

Overall, the present study demonstrated that the main variables in the research model, including the institutional climate, attitudes, subjective norms, perceived behavioral control and perceived risk, played critical roles in predicting university students' preventive behaviors against COVID-19. Thus, the TPB-based expansion model could be functionalized as an effective framework for understanding university students' preventive behaviors on campus. Although promising, there are limitations that should be noted in subsequent research. The results of our study are limited by its generalizability to HEIs and university students in other parts of China and the world because the sample data were collected from university students in Beijing. Therefore, cross-regional and cross-country studies involving university students from a broader scope of HEIs are needed in the future to enhance the generalizability and validity of research findings or revise the framework utilized to understand the influential mechanism of contextual and psychological factors on university students' preventive behaviors. Moreover, future studies should consider other potential mediation and moderation mechanisms of multiple cultural and psychological factors, through which HEIs can foster the preventive behaviors of university students due to the complexity and heterogeneity of COVID-19 spread and control around the world [69], thus producing valuable and creative theoretical and practical outcomes for combating COVID-19.

**Author Contributions:** Conceptualization, X.L. and J.L.; methodology, Y.Z.; software, Y.Z. and X.L.; formal analysis, X.L., J.L. and Y.Z.; investigation, J.L., M.Z., Y.D., Y.Z., M.Y., D.W., H.Z. and X.Z.; resources, D.W., H.Z. and X.Z.; data curation, J.L., M.Z. and X.L.; writing—original draft preparation, J.L. and X.L.; writing—review and editing, X.L. and Y.Z.; supervision, X.L.; project administration, X.L. and M.Y.; funding acquisition, X.L., Y.Z. and M.Y. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Project of the Social Science Plan of the Beijing Municipal Education Commission (SM201910005004), Fundamental Research Funds for the Central Universities (FRF-BR-20-08A) and the Research Start-up Fund of the Capital University of Economics and Business (XRZ2021005).

**Institutional Review Board Statement:** The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the institutional review board (or ethics committee) of the Research Centre for Capital Engineering Education Development at the Beijing University of Technology (approval number: 2021001).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data used or analyzed during the current study are available from the corresponding author on request.

**Acknowledgments:** We would like to thank the university students for participating in the survey and the assistance of the student activities directors or advisors of each targeted HEI in data collection.

**Conflicts of Interest:** The authors declare no conflict of interest.
