4. Results

4.1. Descriptive Statistics of the Content Type and Crisis Lifecycle
To analyze the long-term data more intuitively, we divided the duration from 31 November 2019 to 19 April 2020 into 22 groups (Table 2), each group contained five days. After this step, we calculated the average count of each day’s post of the Wuhan Release of each group. Then, we obtained the count average count of the repost, comments, and likes of each post of each group, and to avoid the influence of abnormal value, we remove the extremum of the repost, comment, and likes. Even so, there were some extreme values in Figure 1, and we will take this into account in our analysis.
Through Figure 1, we could roughly divide the crisis into three stages. Considering the characteristic of the epidemic crisis and the physical truth of the COVID-19, there are some differences with Coomb’s three-stage crisis development model. We describe the pre-crisis as the development period, the crisis as the outbreak period, and the post-crisis as a grace period.
Combining the behavior of the government and the public, we distributed the first 4 groups (31 December 2019 to 19 January 2020) as the development period, between group 5 to group 12 is the outbreak period (20 January 2020 to 28 February 2020). And the period from 29 February 2020 to 19 April 2020, the last 10 groups was regarded as a grace period. At the first stage, the public was aware of this new crisis and showed considerable attention, they tend to repost to tell others some news of the new unknown virus. However, when no more information came out, public attention was not continuable and phased down quickly. This was later seen as a failure of crisis management in the early stage. With the outbreak period, both the government and the public paid much more attention than before, people would like to talk with others through online comments to gather more information, and relieve their emotion or support the government action through thump-up. As the crisis became under control and the situation was improved, although there were some small range fluctuations, the crisis came into a grace period as a whole. However, people in this period were vigilant about the crisis, any sensitive information could trigger a large-scale spread. For example, in group 19, there was a rumor about new cases in Wuhan, people panicked again, and when the government quickly clarified the fake news, the effect was stopped immediately.
The most popular content types among the Wuhan Release in our sample are shown in Table 3. The results show that the local government tends to use their Weibo account to thank the worker (n = 823, 22.89%) and report news (n = 795, 22.11%). Except for these two types, there also includes government measures (n = 536, 14.91%), scientific guidance (n = 446, 12.40%), epidemic data release (n = 396, 11.01%). Other categories which were less than 10% include: notice release (7.73%), encouragement (6.15%), and dispel rumors (2.81%).
Table 3 shows that the most frequently used media type is website links (35.68%). After that, the statistical results show that Wuhan local governments tend to use visual tools, such as photos (16.04%) and videos (16.04%). Moreover, 15.98% (n = 205) posts contained forwarding and 10% (n = 305) of @posts from other users communicated with the public. Finally, the total amount of traditional plain text posts is only 8.63% (n = 50), indicating that this is different from Bonsón et al.’s (2019) research results. That is to say, in the context of the COVID-19 crisis, Wuhan local government attaches great importance to promoting citizen participation through multiple media forms, rather than plain text.

4.2. Median Statistics of Content Type
To determine whether different categories of content influent the public engagement during the crisis, we calculated median repost, reply, and likes across each group. These figures below (Figure 2, Figure 3 and Figure 4) show that public engagement is different across different time-line and categories. Whether in repost, reply, and likes, dispel rumors garners much more interaction than other categories during the period. Because the spread of rumors would damage the crisis prevention and control effects, and mislead the public, the authorities and the public pay much attention to dispel rumors. At the early stage of the crisis, on account of the unknowability of the virus, people tend to engage more in posts about scientific guidance, epidemic data release, and news report, to protect themselves effectively and repost this kind of message to let more people be aware of it. Thank the workers and encouragement content elicit repost and likes from the public in the middle and the later period. The funding could be explained that the crisis was toughly under control, people need to release their emotions and develop gratitude spontaneously. Research shows that these types of government measures generate considerable engagement during the middle stage, including repost, reply, and likes. Except the dispel rumors and encouragement, other categories cannot trigger too many citizens’ interest to reply and likes during the last period of the crisis.
Back to Hypothesis 1, we found that with time-varying, the public most interested content type has indeed changed. At the early stage of the crisis, most people were concerned about the introduction of the virus and the method to prevent this new epidemic disease. During the development of the crisis, citizens paid much attention to the government’s employment and relevant notice. At the end of the crisis, the release of emotion and gratitude became mainstream. Control rumors were the mainstream during the whole crisis.

4.3. Multivariate Statistics
Table 4 shows the results of our correlation model predicting Wuhan’s local government and citizen engagement during the COVID-19 crisis. Hypothesis 2 focused on whether the level of citizen engagement depends on the dialogic loop. The correlation model showed posts related to the COVID-19 crisis that contain “@other accounts” positively improve citizens’ reposts and likes, which indicated more interactions between different accounts. However, at the same time, none of the indicators of hashtag were significant. In summary, the dialogic loop factors are partially associated with citizen engagement.
Hypothesis 3 proposed that whether Weibo posts presented with media richness features of Wuhan Release tended to lead to more citizen engagement. Among them, the increase of pictures brought a significant increase in comments, because the pictures could deliver more information than text could convey, people were willing to express when they saw one or more pictures. However, the existence of video led to a decrease in comments. The difference between picture and video could explain this phenomenon, unlike the picture’s simply and visualized, learning the message of a video could waste more time, some people were unlikely to open the video or watch the full video, so they also showed no interest to discuss the video. The existence of links led to a significant increase in of likes, reposts, and comments, which means people tend to engage with information with reliable evidence. On the other hand, only important information needed to be explained further by linking. In consideration of the relationship between media videos and citizen engagement had a negative correlation. Thus, the media richness partially influent citizen engagement.
Hypothesis 4 mainly asked that whether the longer posts of Wuhan’s local government information release would result in a higher engagement rate. As the model result showed, text length was significantly positively associated with citizen engagement, especially with comments and likes, which means people tend to discuss the long text post.
Hypothesis 5 assumed that people tend to engage with original posts than non-original. The exam result is that posts containing sources negatively affected the three measures of citizen participation. It could be explained that people prefer to get involved in the original information.