Based on the characteristics and functions of social capital, Putnam distinguished between bonding social capital and bridging social capital [12]. Bonding social capital is generated in dense networks within a group, community, or organization. Its main function is to provide emotional support for individuals. Bridging social capital is generated from a sparse network that crosses group boundaries and connects many individuals with different statuses and attributes. It provides members with heterogeneous information [13]. Similarly, scholars have divided online social networking into two basic types. The first is online social support. Offline interpersonal connections are enhanced through the internet, and comfortable communication with close acquaintances is maintained online. Online social support provides companionship and alleviates loneliness [14]. The second type is online information sources, which involve extensive network communication in online spaces such as virtual communities and social media platforms. Their main function is to obtain and extend heterogeneous information resources [15].

Regarding online social support, relationships that offer social support are often marked by depth, mutual benefit, and a strong sense of mindfulness [16]. Typically, family members and close friends constitute the most essential components of an individual’s core social network. These types of connections tend to foster communication, promote harmony, and reinforce cohesiveness within a group. Data analyses from numerous countries and regions have consistently demonstrated that more formal and informal connections tend to make people healthier and happier [17]. Besides, dense social networks can help individuals regulate their emotions, cope with stress, and maintain mental resilience during difficult situations [18]. Nonetheless, it is important to investigate whether the emotional impact of online social support during COVID-19 quarantine may differ from the conclusions drawn from studies on offline social support.

Regarding online information sources, during public health crises, the richness and heterogeneity of the online information to which individuals have access can significantly impact their emotional status. Efficient information transfer is a crucial element of risk management during such times. Studies conducted during the pandemic found that the clearer the subjects’ understanding of the main symptoms and the progress of COVID-19, the smaller the fluctuations in their physical and mental health status [19]. Other studies on Chinese Internet users also found that high heterogeneity in information sources during the pandemic made it easier for people to maintain their health and positive emotional status [20]. We intend to test these conclusions by focusing on the effect of online information sources on emotional status during the quarantine in China.

(1) Micro level: Social status

The ability to deal with, avoid, and compensate for risks in the face of a pandemic is unequally distributed among different groups [21]. An individual’s online social behavior happens in the digital space, where physical subjects are absent. However, the behavioral characteristics, patterns, and psychological effects of online behavior are still shaped by the individual’s social status in the offline space [22]. Access to the Internet and how the Internet is used are affected by social status, personality, abilities, and other factors in the offline space [23,24]. Thus, how the use of social media affects emotional status might be moderated by an individual’s social status [25].

(2) Macro level: Risk management

The positive effect of social capital on individuals’ response to COVID-19 is shaped by community-level factors and macro-level institutional factors [26]. Social networking has an impact on emotional status through its interaction with risk management. For instance, individuals who perceive others as non-compliant with control measures or view government actions as ineffective are more likely to display poor mental health; conversely, those who perceive that the government has taken decisive action tend to have better mental health [27]. China’s epidemic prevention and control measures are issued directly by provincial governments under a unified policy framework provided by the central government. Therefore, the pandemic control practices of the provincial governments may affect individuals’ emotional status during a pandemic. We will focus on three aspects of risk management practice.

The first is the relative difficulty of risk management, which is measured by the population inflow into the local area before a lockdown. During the early stages of the pandemic, provinces with high inter-provincial population inflow from Wuhan faced stricter prevention and control measures. This influx of people from areas with high rates of infection not only directly impacted residents’ emotional status but also perpetuated anxiety through continuous sharing of this information in both online and offline interactions.

The second research focus is the effect of provincial governments’ risk management, which can be measured by the cumulative number of confirmed cases in each province. In public health, the number of confirmed cases has been widely used to evaluate the effectiveness of prevention and control measures [28,29]. The effectiveness of prevention and control measures determines the public’s understanding, acceptance, and confidence in these measures, as well as their attitude toward future epidemic control measures. It also directly affects their attitude toward future epidemic control measures.

The third research focus is recovery from the pandemic, which can be measured by the relaxation of mobility restrictions once the pandemic was under control. After the pandemic was brought under control, most citizens wished to return to work as soon as possible due to economic pressure [30]. During that time, whether the provincial governments could timely adjust interventions and facilitate the orderly resumption of work proved crucial in reestablishing normalcy and shaping residents’ positive attitudes. Consequently, risk management practices at the provincial level might have moderated the impact of online social networking on individuals’ emotional status.

To gain a detailed understanding of the correlation between online social networking and emotional status, this study has integrated survey data with provincial-level online big data during the early stages of COVID-19 to address three objectives. Firstly, we examine the relationship between online social networking from two dimensions: online social support and online information sources. Secondly, we investigate how the relationship between online social networking and emotional status is moderated by individuals’ social status. Thirdly, we will investigate the moderating effect of provincial interventions on this relationship. Therefore, we propose the following hypothesis. Since online social networking has two dimensions, each hypothesis can be refined as two sub-hypotheses as below:

Hypothesis 1: Online social networking is positively related to users’ emotional status.

Hypothesis 1.1: The more online social support an individual receives in the context of pandemic isolation, the more positive their emotional status.

Hypothesis 1.2: The more variety in individuals’ online information sources, the more positive their emotional status.

Hypothesis 2: The influence of online social networking on emotional status is moderated by social status.

Hypothesis 2.1: The influence of online social support on emotional status is moderated by social status.

Hypothesis 2.2: The influence of online information sources on emotional status is moderated by social status.

Hypothesis 3: The influence of online social networking on emotional status is moderated by provincial risk management practices.

Hypothesis 3.1: The effect of online social support on emotional status is moderated by provincial risk management practices.

Hypothesis 3.2: The effect of online information sources on emotional status is moderated by provincial risk management practices.

We integrated individual level survey data and provincial level online big data to to conduct our analysis.

1) Individual level: WeChat users online survey

This part of study adopted a cross-sectional research design. In 2020, the number of monthly active users of WeChat reached 1.225 billion [31]. Due to its large user base and the boom in social media usage during the COVID-19 pandemic, many researchers have used data collected from WeChat in behavioral and psychological studies, particularly studies of changes during the COVID-19 pandemic [32]. At the onset of the epidemic, stringent physical distancing measures were enforced nationwide, rendering online surveys almost the only option. We designed a short questionnaire including seven sections: A. Infected respondents or infected acquaintances of respondents; B. Behaviors during the quarantine; C. Social networking during the quarantine; D. Physical and mental health during the quarantine; E. Difficulties and solutions; F. Personal information. We conducted a pilot study with a sample size of 50 and the average survey time was 8 min. After modifications, 60 survey questions were included in the final questionnaire.

We carried out an online survey based on the WeChat sample bank through the data collecting platform named Researcher. The survey was posted online on April 23 and 24, 2020, and the respondents received and completed the questionnaire through the WeChat app on their mobile phones. The data collection platform randomly sampled 30,000 active WeChat users from the Chinese user base, encompassing both urban and rural areas across 31 provinces and over 300 cities. The average age of the users in the sample pool is 30, and they have relatively high levels of education, relatively stable incomes, and are mainly white-collar workers. The survey was completed by 8,019 respondents. To improve the quality of the online survey, three detectors were embedded among survey questions aimed to screen out programs coded to automatically complete questionnaires, multiple entries of a same respondent, and irresponsible answers that were intended to collect a payment for a completed questionnaire. As a result, responses that were careless, incomplete, or repetitive, and automatically completed questionnaires were excluded. Following the screening process, a final sample of 3,004 valid questionnaires was obtained from all provinces in China.

2) Provincial level: Online big data and provincial statistics

This part of study utilized a multilevel research design, incorporating online big data from different sources and survey data. Online big data sets have become increasingly popular in computational social science research due to their immediacy, continuity, authenticity, and large scale. Besides, they can provide objective records independent of subjective reports [24,33]. The online data used in this study was primarily obtained from two sources:

(1) Baidu Migration data. The population flow data used in this research were drawn from the Baidu Migration data set (https://qianxi.baidu.com/#/). Baidu Migration acquires location data of smartphones in real-time through mobile phone base stations and other devices, anonymously processes the data on an individual level, and calculates overall indicators of daily population flow in different regions. This method accurately reflects the population flow between and within provinces and cities in real-time. To obtain the Baidu Migration data, the authors wrote a crawler program based on the open-source crawler framework Scrapy using Python 3.7. The result was a web-scraping program specifically targeting the Baidu Migration platform. The code used for this is provided in the Appendix. During August to September 2020, the Migration data of more than 300 cities in China in the three months before and after the 2019–2020 Spring Festival was captured from the website. Using the Migration data from the same period in 2019 as a baseline, changes in population flow caused by the COVID-19 pandemic in 2020 were calculated. The daily inter-city population flow data and within-city movement index were then measured to gauge the population inflow from epidemic areas and reflect the strength of the lock-downs in various provinces.

(2) Confirmed cases data. The data on the number of confirmed cases is publicly available and released by the Chinese Center for Disease Control and Prevention. This ensures the accuracy of the data and facilitates comparison with related studies.

1) Dependent variables

In the WeChat survey, respondents were asked how often they experienced different emotional status, including optimism, calmness, anxiety, and fear. Each state of mind was measured using a 4-point scale with reverse coding for anxiety and fear. Next, we constructed a factor score using principal component factor analysis. The values ranged from 1 to 10, where higher scores represented a more positive emotional status. The Cronbach’s α was above 0.8, indicating high validity of this measurement.

2) Independent variables

(1) Online social networking. Online social networking can be divided into two dimensions: online social support and online information sources. To measure online social support, the respondents were asked “Has your online communication with relatives, friends, and acquaintances increased since the beginning of the pandemic?” The survey question used a 4-point scale (0 = “never,” 1 = “decrease,” 2 = “constant,” and 3 = “increase”). To construct a binary variable, “never,” “decrease,” and “constant” were coded as 0, indicating no increase, and “increase” was coded as 1. For online information sources, the questionnaire asked the respondents about their “sources of information related to the COVID-19,” including WeChat, Weibo, news portals, Douyin, Billibilli and other video platforms, and TV broadcasting. This variable was a count variable with answers ranging from 0 to 5; a larger value indicated access to more information channels.

(2) Cumulative confirmed cases. This variable was derived from online data released by CDC. It represented the daily cumulative number of confirmed cases in each province. The cumulative number of confirmed cases in each province on April 23, the date of the survey, was used to represent the severity of the pandemic in the province during the study period.

(3) Population inflow from the epidemic area: This variable was derived from Baidu Migration’s online data. As inter-provincial movement of populations from areas with outbreaks was strictly restricted after the Wuhan lockdown on January 23, 2020, this indicator was measured as the percentage of the population flowing into the focal province from Wuhan on January 22. This variable depicted varying levels of difficulty in epidemic prevention and control caused by the geographical location of each province.

(4) Post-pandemic recovery: The population flow index within each province on March 17th was used to measure post-pandemic recovery. March 17th marked the end of the third phase (see Fig. 1 for the division of phases). This date marked the point when the pandemic was initially under control. The population flow index measured the relaxation of lock-downs and the extent to which social life had returned to normal in each province.

3) Control variables

We controlled for two dimensions of social status. The first consisted of demographic characteristics, such as gender and age. To ensure a balanced gender ratio, we set a gender quota during the online survey. Respondents ranged in age from 16 to 84 years old with an average age of 29. To control for the nonlinear effect of age, we also added the square term of age as a control variable. The second set of control variables consisted of social status characteristics, such as education level, occupation type, and family income. In terms of education level, 78% of the WeChat sample had obtained or were in the process of obtaining a bachelor’s or junior college degree, while 22% had a high school or lower degree. The majority had middle-class occupations; employers and white-collar workers accounted for 11% and 40%, respectively, and blue-collar workers accounted for 32% of the sample. Economic status was measured by annual household income. In the analysis of moderating effects of social status, annual household income was treated as a categorical variable to eliminate the influence of skewed distribution. Following the calculation methods proposed by Li et al. [34], people with annual household incomes of less than 88,300 yuan in 2020 were assigned to the low-income group, while those with annual household incomes of more than 300,000 yuan were assigned to the high-income group. We also included a control variable that measured the frequency of going out. The questionnaire asked “How frequently did you go out to parties with others, for shopping or for volunteering or community service, or other outdoor activities during the pandemic?” We calculated a principal component factor score for these “going out” variables, and the values ranged from 1 to 100, where a higher value indicated a higher frequency of going out. Descriptive statistics of these variables are shown in Table 1.

In our study, we used Stata version 17.0 to conduct statistical analyses. First, we integrated online big data at the provincial level with survey data at the individual level. The two datasets were linked based on the locations reported by participants in the survey. This integration of datasets facilitated a comprehensive analysis of the research subject. Next, we employed OLS regression to investigate the impact of online social networking on emotional status and determine whether there was an interaction between online social networking and social status.

Furthermore, we adopted a multilevel model to analyze the relationships between provincial-level and individual-level variables. Both the intercept and slope were expressed as functions of the provincial-level independent variables. The model assumes that the intercepts and slopes at the individual level are random and affected by some independent variables at the provincial level. x′ j in Eqs. (2) and (3) represents the provincial-level independent variables in the jth province, which affect the slope of the individual-level independent variable xij and the intercept in Eq. (1). x¯j  is the mean of the individual-level independent variable xij in the jth province.

(1) yij=β0j+β1jxij+ϵij

(2) β0j=γ00+γ01x j′+γ02x¯j+μ0j

(3) β1j=γ10+γ11x j′+γ12x¯j+μ1j

OLS regression was used to analyze the effect of online social networking on emotional status. As shown in Table 2, the two dimensions of online social networking, online social support and online information sources, are positively and significantly related to emotional status. The standardized regression coefficients (B) indicates that after eliminating the influence of value range, the coefficient of online social support is larger than that of online information sources. In other words, individuals’ online social support during the pandemic had a greater effect on emotional status than online information sources. Among the control variables, going out had a significant negative impact on emotional status. During the pandemic, quarantine and mobility restrictions were widely implemented across the country. People who had to go out for any reason faced a much higher risk of infection than others, leading them to feel more insecure and anxious. Males had a more positive emotional status than females. The unemployed group, who faced greater pressure during the pandemic, had a more negative emotional status than the employed.

To investigate group differences in the effect of online social networking on emotional status, we included interactions with three status variables: occupation type, education level, and family income, based on the OLS model (see Table 3). Occupation type was coded as a four-category variable (self-employed, white-collar, blue-collar, and unemployed). Education level was a binary variable, where 1 represented not having attended university and 0 represented having attended university (including junior college). Household income was a three-category variable.

As shown in Table 3, there is no significant moderation effect of education, occupation, or income on the relationship between online social support and emotional status. This means there were no class differences in the effect of emotional support on positive attitudes during the quarantine. However, the interaction terms between online information sources and the three status variables are significantly positive. This indicates that the positive influence of online information sources on emotional status was stronger for lower occupational status, lower-educated, and low-income groups. There is clearly an interaction between online information sources and individual social status.

In Table 4, we tested the influence of provincial-level factors and their interactions with online social networking. The provincial-level independent variables included cumulative confirmed cases, population inflow from epidemic areas, and post-pandemic recovery. Since sample sizes varied between provinces, the average emotional status for each province might have been affected by different sample sizes. To reduce the influence of unbalanced samples from different provinces, a weight was calculated based on the ratio of sample size to the population size in each province. The population weight of each province was then controlled in the multilevel models.

Due to space limitations, we only present coefficients for the main independent variables and interaction terms in Table 4. However, readers interested in obtaining complete results may request them from the authors.

As shown in Table 4, confirmed cases in each province had a significant negative impact (B = −0.08, p-value < 0.001) on the emotional status of Internet users in the province. Inflows of people from Wuhan had a significant negative impact (B = −0.51, p-value < 0.001) on the emotional status of residents in each province. Specifically, the large influx of people from Wuhan created anxiety and fear among the local population. On the other hand, post-pandemic recovery efforts had a significant positive effect (B = 0.02, p-value: 0.01) on individuals’ emotional status across all provinces. This suggests that normalization of travel within provinces has contributed to increased positivity among people.

The significantly negative interaction term between social support and confirmed cases indicates that social support amplified the negative impact of confirmed cases on emotional status. In the early phases of the pandemic, online communication between acquaintances functioned as an amplifier, as people reinforced each other’s anxiety and fear of infection. Likewise, the significantly negative interaction of online social support and population inflow indicates that online networking between acquaintances amplified the negative impact of population inflows from Wuhan on emotional status. The significantly positive interaction between online social support and recovery indicates that online social support enhanced the positive impact of post-pandemic recovery on emotional status.

The interactions between online information sources and provincial factors show that none of the interaction terms are significant. That is, the richness of online information channels and provincial risk management had independent effects on Internet users’ emotional status, and the impact of online information sources on emotional status did not vary because of differences in confirmed cases, population inflows from Wuhan, or the recovery of normal life across 31 provinces.