A cross-sectional design survey with national sample was conducted.
Participants and procedure
A snowball sampling was used to recruit participants from different regions in China. A total of 4826 Chinese visited the online survey between 2 February and 9 February, 2020. Sixty-nine participants were not willing to participate in the study, leaving 4757 participants to take part in the survey, resulting in a response rate of 98.6%. Then, we removed a number of participants because (1) they indicated that they were under 16 years old (N = 99), a cutting age that parent consent is optional, or (2) they were inclined to respond to the items in a similar pattern (e.g., chose the same answer across multiple consecutive items or within the whole questionnaire, N = 51). Finally, 4607 participants provided complete data and were included in the analyses. Participants’ age ranged from 17 to 90 years old (Mean age = 23.71 years old, SD = 7.29). They were from 31 provinces / centrally-governed cities / autonomous regions / special administrative regions, with the sample size ranging from 16 (0.3% of the total sample, Ningxia Hui Autonomous Region) to 1386 (30.1% of the total sample, Guangdong Province). The sample covered a wide range of demographics. Detailed demographics are summarized in Table 1.
The study was reviewed and approved by Guangzhou University. The whole study was conducted online in compliance with the ethical standards for research outlined in the Ethical Principles of Psychologists and Code of Conduct . Over 200 student helpers who majored in psychology voluntarily distributed the online survey link on various internet platforms, including WeChat (the most popular APP for instant message in mainland China), Weibo, QQ, etc. By clicking the hyperlink, participants were directed to an online survey website. An information sheet stating the goal and the procedure of the study was presented to participants on the first page of the survey. If participants checked the “I understood the study and am willing to participate” box at the bottom of the information sheet, they would entered the survey and fill in the questionnaires. If participants were not willing to participate, they could check the “I understood the study but am not willing to participate” box and then the survey ended. Participation was voluntary and no incentive reward was given. Anonymity was emphasized and no identifiable information was collected. It took participants about 20 minutes to complete the survey.
Emotional and behavioural reactions. Participants’ emotional and behavioural reactions were measured with 20 items. These items cover a number of dimensions, including negative emotion (8 items, anxiety, worry, depressive, panic, lonely, nervous, sad, and angry), positive emotion (3 items, happy, joy, and excited), sleep problems (4 items, insomnia, shallow sleep, have nightmares, and insufficient sleep), aggression (2 items, argue with others and physical fight with others), substance use (2 items, smoking and drinking), and mobile phone use (1 item). Participants were asked to indicate the differences in the emotional and behavioural reactions listed above before and after the outbreak of COVID-19 on a five-point scale (from “1 = much less compared to the days before the outbreak” to “5 = much more compared to the days before the outbreak”). To align with other dimensions, the items for positive emotion were reversely scored. A higher score indicates COVID-19 causes more negative emotion, sleep problems, aggression, substance use, mobile phone use, and less positive emotion.
Social participation. The Social Participation Scale used in prior research  was adapted to measure participants’ social participation regarding COVID-19. Participants were asked to indicate how often they participated in different social events since the outbreak of COVID-19 on a five-point scale (from “0 = never” to “4 = very often). A higher score indicates that participants participated in the social events more actively. A sample item is “How often do you help those who need help in the community since the outbreak of COVID-19?”
Precautionary behaviour. Participants’ precautionary behaviour was measured with 19 items written by authors following the precautionary guideline issued by the Chinese government. Participants were asked to indicate how often they show various precautionary behaviour since the outbreak of COVID-19 on a five-point scale (from “0 = never” to “4 = very often). A higher score indicates that participants display more precautionary behaviour. Sample items include avoiding travelling to regions affected by COVID-19, wearing a facemask, regularly changing a facemask, and washing hands.
Knowledge about COVID-19. Participants’ perceived knowing of various aspects of COVID-19 (e.g., cause, ways of transmission, symptoms, diagnostic criteria, etc.) was measured with 11 items. Participants indicated how much they know each item on a five-point scale (from “1 = totally not know” to “5 = totally know”). A higher score indicates participants perceived they know more about COVID-19.
Perceived severity. Participants’ perceived severity about COVID-19 was measured with 5 items. Participants indicated their perception of how severe is the infection rate, morbidity, mortality, the negative influence on social order and the negative influence on the economics on a five-point scale (from “1 = not severe at all” to “5 = very much severe”). A higher score indicates participants perceived COVID-19 to be more severe.
Perceived controllability. Participants’ estimation of how much can the various aspects of COVID-19 be controlled was measured with 9 items on a five-point scale (from “1 = totally uncontrollable” to “5 = totally controllable”). A higher score indicates participants perceived COVID-19 to be more controllable. A sample item is “How controllable do you think the etiology of COVID-19 is?”
Demographic variables. We also collected a number of demographic variables of participants, including their biological sex (0 = male, 1 = female), age, education (1 = junior middle school or below, 2 = high school or equivalent, 3 = college, 4 = bachelor degree, 5 = master degree, 6 = doctoral degree), current residential location (referred to province and city/district), their relationship with COVID-19 (1 = healthy, 2 = other, including suspicious case, diagnosed case, relatives or friends of suspicious/diagnosed case, etc.), the history of chronic physical diseases and psychiatric/psychological disorder (1 = yes, 2 = no), and their current physical health condition (from “1 = very poor” to “5 = very good”).
We analysed the data in SPSS and Mplus 7.0 in several steps, with .05 as the significant level across all analyses. We conducted preliminary analyses prior to carrying out formal statistical analyses. First, we examined the psychometric properties of the measures used in this study, including internal consistency reliability, item discrimination and confirmatory factor analysis. Second, given that only self-report questionnaires were used in this study and this might cause common method variance, we examined whether common method variance should be of a concern in this study. Subsequently, we then continued performing formal analyses. First, we conducted the descriptive statistics to capture the centrality of the variables. Second, we conducted correlation analysis to capture the association between participants’ knowledge about COVID-19, perceived severity, and perceived controllability and emotional and behavioural reactions, social participation, and precautionary behaviour. For the correlation analysis, we employed Cohen’s (1992) standard to determine whether the correlation coefficients were substantial, with r = .01, .03, and .05 representing small, medium, and large effect size . Last, given that the data structure is hierarchical in nature (i.e., participants nested in provinces), multilevel regression analysis should be used. Prior to using multilevel model, we examined the intraclass correlation (ICC) for each outcome variable. We found that the ICCs were trivial, ranging from .008 (sleep problem) to .032 (negative emotion). Given that the ICCs were smaller than .05, we believed that treating the data as individual data would be appropriate and thus would not choose multilevel regression model. We employed Mplus 7.0 to conduct regression analysis with latent variables. We used the items from each measure to construct the latent variables except for perceived controllability, knowledge about COVID-19, negative emotions, and precautionary behaviour because these four measures had a number of items, which may render poor model fit. To construct latent variables for these four measures, we used an item-to-construct balance parceling technique  to create three indicators for each measure. We controlled for a number of demographic variables when we examined the associations between independent variables and the dependent variables. The values of RMSEA (< .08), CFI and TLI (> .90) indicate the model fit is acceptable [29, 30].
 Because the online survey website automatically set each question as “required to be answered”, participants who undertook the study needed to finish the survey prior to submission. Participants who did not want to continue could quit the survey by closing the window. However, the online survey website did not record the response of participants who quitted the survey in the middle and therefore participants included in the analysis were the ones who provided complete data.