This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Public Health and Surveillance, is properly cited. The complete bibliographic information, a link to the original publication on https://publichealth.jmir.org, as well as this copyright and license information must be included.
The COVID-19 pandemic led to the necessity of immediate crisis communication by public health authorities. In Germany, as in many other countries, people choose social media, including Twitter, to obtain real-time information and understanding of the pandemic and its consequences. Next to authorities, experts such as virologists and science communicators were very prominent at the beginning of German Twitter COVID-19 crisis communication.
The aim of this study was to detect similarities and differences between public authorities and individual experts in COVID-19 crisis communication on Twitter during the first year of the pandemic.
Descriptive analysis and quantitative content analysis were carried out on 8251 original tweets posted from January 1, 2020, to January 15, 2021. COVID-19–related tweets of 21 authorities and 18 experts were categorized into structural, content, and style components. Negative binomial regressions were performed to evaluate tweet spread measured by the retweet and like counts of COVID-19–related tweets.
Descriptive statistics revealed that authorities and experts increasingly tweeted about COVID-19 over the period under study. Two experts and one authority were responsible for 70.26% (544,418/774,865) of all retweets, thus representing COVID-19 influencers. Altogether, COVID-19 tweets by experts reached a 7-fold higher rate of retweeting (
Twitter data are a powerful information source and suitable for crisis communication in Germany. COVID-19 tweet activity mirrors the development of COVID-19 cases in Germany. Twitter users retweet and like communications regarding COVID-19 by experts more than those delivered by authorities. Tweets have higher coverage for both authorities and experts when they are plain and for authorities when they directly address people. For authorities, it appears that it was difficult to win recognition during COVID-19. For all stakeholders studied, the association between number of followers and number of retweets was highly significantly positive (authorities
The occurrence of the SARS-CoV-2 virus and the ongoing COVID-19 pandemic have made crisis communication inevitable. Across countries, people use social media as their source of information about development of the pandemic [
The number of scientific publications regarding COVID-19 is enormous, including several studies that investigated COVID-19 crisis communication on social media platforms such as Twitter. Crisis communication via Twitter was also studied before the COVID-19 pandemic. For example, surveys investigated Twitter data in regard to crisis communication with a focus on health crises [
COVID-19 Twitter crisis communication studies have shown that different stakeholders such as scientists, governmental authorities, and politicians, as well as health care professionals tweeted more during COVID-19 [
In the aftermath of the Japan earthquake in 2011, the crisis communication and leadership of the government were neither clear nor apparent on Twitter [
Given this context, the aim of this study was to describe and analyze COVID-19–related tweets by authorities and experts in Germany. This seems to be a necessary task, as COVID-19 is the first health crisis digitally explained and discussed directly by experts, and as such, it competes with the official crisis communication of authorities for attention and coverage. Tweet spread, as our variable of interest, was measured by retweet and like counts. To compare the content of stakeholder tweets and to explain the spread of tweets, the intrinsic message features of “structure,” “content,” and “style” of COVID-19 tweets were compared. Variables related to structure are those capturing whether tweets consist of hashtags, images, URLs, and mentions. With regard to the content features of tweets, Vos et al [
The data set consisted of tweets from 39 German public authorities and experts. The selection of the 39 German authorities and experts was informed by the importance of authorities and experts during the COVID-19 crisis and their visibility in the German discussion. Additionally, these public authorities and experts are active users on Twitter. These stakeholders included 21 authorities and 18 experts. The Federal Ministry of Health and The Robert Koch Institute (RKI) in Berlin were included among the authorities. The expert group consisted of virologists, science communicators, physicians, and other scientists.
Tweets of the 39 Twitter accounts were retrieved with a Twitter application programming interface (API) account of the authors. Using tokens, timelines of all stakeholders were retrieved on January 15, 2021. Data analysis was carried out using RStudio (version 1.31093 for Windows) and additional code packages as well as Excel for Microsoft 365 MSO. When creating a Twitter account, users confirm that their tweets are public and can be analyzed by third parties [
Data retrieval led to 81,455 tweets from authorities and experts. This first data set of 81,455 tweets needed further adjustments.
PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) diagram of Twitter data adjustment steps. API: application programming interface.
The time frame of tweet analysis was set from January 1, 2020, to January 15, 2021, to capture the first calendar year of the pandemic. However, as vaccinations had just started in Germany at the end of December 2020, and discussions about vaccinations were at a high, the time frame was expanded to January 15. Therefore, as a second step, the data set was adjusted by time so that only tweets from January 1, 2020, to January 15, 2021, were included for the analysis of COVID-19 crisis communication within the first year of the pandemic, leaving 50,100 tweets.
Next, it was necessary to filter COVID-19–related tweets and exclude tweets with other content. To achieve this, COVID-19–related tweets were identified with filter words based on COVID-19 keywords and hashtags. At first glance, tweets of authorities and experts appeared to be very different, and therefore a different filter strategy was used for the two groups. As authorities’ tweets were characterized by the use of hashtags and keywords, COVID-19 tweets were filtered in one step. A COVID-19 word and hashtag list was compiled with 1282 words and hashtags. This list contains a wide spectrum of COVID-19–related words, including different spellings (eg, “corona case,” “corona cases”).
Comparatively, few of the experts’ tweets used hashtags and keywords. As the group of experts’ main expertise is COVID-19, it was assumed that most of these tweets were related to the pandemic; therefore, COVID-19 tweets were identified in two filter steps. In the first filter step, non-COVID-19 tweets were filtered with a non-COVID-19 list of words and hashtags, containing 5789 non-COVID-19–related words (eg, “Navalny,” “Neanderthals”). In this way, most of the COVID-19 experts’ tweets remained in the sample. In the second filter step, the excluded experts’ tweets were filtered again using the COVID-19 filter to capture the experts’ tweets dealing with non-COVID-19 and COVID-19–related topics in one tweet (eg, cancer and COVID-19). Overall, 35,645 COVID-19–related tweets were obtained, 14,624 of which were published by authorities and 21,021 of which came from experts. Authorities tweeted more about non-COVID-19 topics with 13,100 tweets compared to 1157 non-COVID-19 tweets from experts. The complete COVID-19 and non-COVID-19 word lists to separate the tweets are available in
For the content analysis, this study focused on original tweets only. These were tweets written and posted directly from the stakeholders, which excluded replies, retweets, and quotes, as well as tweets in the English language. Thus, COVID-19 tweets were subsequently reduced by quotes, retweets, and replies, leaving 8251 original COVID-19 tweets for content analysis. Of these original COVID-19 tweets, 5432 were from authorities and 2819 were from experts. This last filter step of reducing the data to original tweets illustrated that experts were extremely involved in the COVID-19 discussion on Twitter, with a high share of replies, quotes, and retweets in the broader data set.
Based on the text corpus of all COVID-19 tweets by authorities and experts published within the time frame, we first analyzed descriptive statistics of the tweets (eg, their retweets and likes). Content analysis was then performed. German tweets of authorities and experts were used in text form. For regressions explaining the spread of tweets, retweet and like counts were used as dependent variables.
Some of the explanatory model variables to explain the number of retweets and likes were generated following a previous study [
The four content categories of severity, efficacy, susceptibility, and technical information from a previous study [
Tweets were assigned to specific content categories by comparing single words in the tweets with word lists reflecting specific content. Whether or not a tweet contained information about severity was established by comparing tweet words with a list of 521 specific severity words and hashtags. The list to detect susceptibility tweets contained 99 typical words and hashtags in the context of COVID-19 susceptibility. The efficacy word and hashtag list included 451 typical words. Technical information tweets were generated based on a list of 173 COVID-19 technical information words and hashtags. The word list for politics had 124 words and hashtags. Due to the nature of the word lists, tweets could simultaneously be categorized into different categories. The word lists for all categories are provided in
For the style intrinsic tweet feature, we again followed Vos et al [
As it can be expected that retweets and likes would be higher for users with many followers, the followers count was included as another explanatory variable.
Negative binomial regressions were used to explain the spread of a tweet. Negative binomial regressions are suitable when the dependent variable is a count variable with many zero observations. Tweet spread (ie, the dependent variable) was measured by either the retweet or like count. These counts were retrieved together with the tweets. In the context of this study, a retweet meant that a follower of the 39 stakeholders republished the stakeholder’s tweet. The tweet content was therefore distributed by all users who retweeted that tweet. The more the retweets, the more the original tweet was spread. The like count was the second indicator used to measure the spread of tweets. Likes are an indicator of popularity. It is argued that retweets reflect more engagement as they indicate more interaction compared to likes. A retweet means that the user is sharing the stakeholder’s content with the possibility to comment. This is not the case for likes. Overall, four different regressions were executed: four negative binomial regressions to explain the number of retweets (and likes), separated by authorities and experts. For the retweet and like counts, the dependent variables did not follow a normal distribution and the variance exceeded the mean. The data can be regarded as overdispersed, and therefore the negative binomial regression was the best choice to explain the count variables.
A list of the Twitter stakeholders analyzed and descriptive statistics for their numbers of followings and followers are given in
Number of original COVID-19 tweets over time (January 1, 2020, to January 15, 2021) in Germany (N=8251).
Strikingly, the number of COVID-19 tweets from January 1, 2020, to January 15, 2021, paralleled the development of COVID-19 cases in Germany, including the first two waves (March/April and beginning of October) during the sample period. The time of lower COVID-19 cases was also mirrored in the number of tweets, with a smaller number of tweets in the summer of 2020. This effect, namely the parallelism of tweet activity and (COVID-19) incidences, has already been documented elsewhere [
The descriptive statistics in
On average, experts had 814 followings, representing 1.21 times the number of followings compared with that of authorities, with an average of 671. Although these numbers show that experts had a bigger network, the mean of followings did not differ significantly between the two groups (t37=0.54,
The number of followers exceeded the number of followings to a great extent. Among the authorities, RKI had the highest number of followers, with more than 430,000. In the group of experts as well as overall, expert E8 had the biggest follower network, with more than 650,000 followers (as of January 15, 2021). Less than 1000 followers had the Max Planck Institute for Infection Biology in the group of authorities. In the experts’ group, expert E1 had the lowest number of followers (1270). On average, experts had 99,059 followers, representing 1.31 times the number of followers for authorities with an average of 75,817 followers. However, this differences in mean followers was also insignificant (t37=0.52,
Experts were considerably more frequently retweeted with 7-fold (7.03) more retweets than authorities. On average, a tweet by authorities was retweeted 30.7 times, whereas tweets originating from COVID-19 experts were retweeted 215.7 times. This difference in mean retweets by stakeholder group was statistically significant (
Descriptive statistics for variables used in the
Variable | Description | Authorities (n=5432), mean (minimum, maximum) | Experts (n=2819), mean (minimum, maximum) |
Retweet count | Metric variable for the number of retweets (dependent variable in negative binomial regression) | 30.70 (0, 9982) | 215.70 (0, 8457) |
Like count | Metric variable for number of likes (dependent variable in negative binomial regression) | 90.80 (0, 44,274) | 1257 (0, 63,002) |
Hashtag | Structural dummy variable that is 1 if a tweet has a hashtag | 0.92 (0, 1) | 0.19 (0, 1) |
Images | Structural dummy variable that is 1 if a tweet uses images | 0.69 (0, 1) | 0.17 (0, 1) |
URL | Structural dummy variable that is 1 if a tweet uses a URL | 0.71 (0, 1) | 0.71 (0, 1) |
Mention | Structural dummy variable that is 1 if a tweet carries a mention | 0.40 (0, 1) | 0.24 (0, 1) |
Severity | Content dummy variable that is 1 if a tweet contains words describing COVID-19 severity | 0.75 (0, 1) | 0.52 (0, 1) |
Susceptibility | Content dummy variable that is 1 if a tweet contains words describing susceptibility to COVID-19 | 0.22 (0, 1) | 0.05 (0, 1) |
Efficacy | Content dummy variable that is 1 if a tweet contains words describing efficacy measures | 0.35 (0, 1) | 0.28 (0, 1) |
Technical information | Content dummy variable that is 1 if a tweet contains a word related to technical virus information | 0.04 (0, 1) | 0.07 (0, 1) |
Social | Content dummy variable that is 1 if a tweet contains words describing the social consequences of COVID-19 | 0.13 (0, 1) | 0.08 (0, 1) |
Politics | Content dummy variable that is 1 if a tweet contains words describing the political consequences of COVID-19 | 0.11 (0, 1) | 0.04 (0, 1) |
Other | Content dummy variable that is 1 if a tweet contains a word that cannot be attributed to other content variables | 0.03 (0, 1) | 0.31 (0, 1) |
First person | Style dummy variable that is 1 if a tweet uses first-person words | 0.27 (0, 1) | 0.44 (0, 1) |
Second person | Style dummy variable that is 1 if a tweet uses second-person words | 0.04 (0, 1) | 0.06 (0, 1) |
Followers count | Metric variable as the number of followers per Twitter user | 75,817 (971, 435,392) | 99,059 (1270, 657,292) |
Followings count | Metric variable as the number of followings per Twitter user | 671 (38, 3424) | 813.90 (2, 3293) |
Considering the top users in crisis communication, previous studies evaluated the concentration of retweets among single users. For COVID-19 in Germany, two experts and the Federal Ministry of Health were responsible for 70.26% (544,418/774,865) of all retweets, and as such were the COVID-19 influencers in Germany on Twitter.
As seen in
There were also differences between the stakeholders under study with regard to the content of the tweets, with the strongest difference observed for severity. Overall, 75.00% (n=4074) of authorities’ tweets referred to the severity of COVID-19, such as with reference to symptoms, whereas only 51.47% (n=1451) of experts’ tweets were categorized into severity. In 22.22% (n=1207) of the authorities’ tweets, there were words referring to susceptibility and regional information related to COVID-19, whereas this was the case in only 4.68% (n=132) of the experts’ tweets. With regard to efficacy information, the tweets of experts and authorities were similar with 34.70% (n=1885) and 27.49% (n=775) of tweets including this content, respectively. Tweets bearing technical information of the spread of the virus and references to scientific findings were more frequent for experts (n=188, 6.66%) than for authorities (n=224, 4.12%).
More of the authorities’ tweets referred to the social consequences (n=719, 13.24%) and political consequences (n=617, 11.36%) of COVID-19 than the experts’ tweets (n=225, 7.98% and n=104, 3.68%, respectively). The content category “other,” which refers to words in the tweets that cannot be categorized by any of the category word lists, explains the differences observed: 31.80% (n=896) of experts’ tweets and only 3.73% (n=203) of authorities’ tweets were related to content that was not captured by the previous categories. Thus, the tweets of the experts were much simpler and did not use that many clearly detectable keywords, favoring more colloquialisms than authorities. An example is the tweet of a German expert “Sehr gut” (“very good”) with additional links to external information. Categorization of these tweets by quantitative text analysis was impossible. Taken together, the findings from analysis of the structural variables show that authorities were much better in using structure, hashtags, and keywords.
The tweets of experts and authorities also differed with regard to the style variables: 26.82% (n=1457) of authorities’ tweets and 43.70% (n=1332) of experts’ tweets used first-person words, whereas 3.73% (203) of authorities’ tweets and 5.57% (n=157) of experts’ tweets used second-person words. The fact that experts used the second person slightly more often than authorities indicates that they interacted more with other Twitter users.
The results for the like count regressions were similar, which are shown in
Negative binomial regression to explain the retweet count of COVID-19 tweets for authorities and experts (N=8251 tweets).
Variables | Authoritiesa | Expertsb | |||||
|
IRRc |
|
IRR |
|
|||
Model variable: constant | 16.69 | 30.57 | <.001 | 71.95 | 61.37 | <.001 | |
|
|||||||
|
Hashtag | 0.64 | –6.92 | <.001 | 1.11 | 1.56 | .12 |
|
Images | 1.06 | 1.32 | .19 | 1.06 | 0.87 | .38 |
|
URL | 0.82 | –4.81 | <.001 | 0.76 | –4.27 | <.001 |
|
Mentions | 0.81 | –5.45 | <.001 | 0.73 | –5.27 | <.001 |
|
|||||||
|
Severity | 1.40 | 8.09 | <.001 | 1.18 | 3.34 | <.001 |
|
Susceptibility | 1.02 | 0.46 | .65 | 1.15 | 1.21 | .23 |
|
Efficacy | 1.34 | 8.63 | <.001 | 1.10 | 1.71 | .09 |
|
Technical information | 1.45 | 4.23 | <.001 | 1.00 | 0.06 | .95 |
|
Social | 1.24 | 4.05 | <.001 | 1.27 | 2.69 | .01 |
|
Political | 0.71 | –6.12 | <.001 | 0.87 | –1.12 | .26 |
|
|||||||
|
First person | 0.93 | –1.80 | .07 | 1.10 | 0.02 | .99 |
|
Second person | 1.88 | 6.96 | <.001 | 1.03 | 0.23 | .82 |
Other: followers count | 1.00 | 28.74 | <.001 | 1.00 | 25.99 | <.001 |
aAuthorities: –2 log-likelihood=–44365.18; Akaike information criterion=44,395; null model logistic regression
bExperts: –2 log-likelihood=–33,752.49; Akaike information criterion=33,782; null model logistic regression
cIRR: incidence rate ratio.
The regression constant for both authorities and experts was significant and positive. The first set of explanatory variables included four dummy variables that recorded whether a tweet used a hashtag, image, URL, or a mention, and as such captured structural elements of intrinsic message features. The number of retweets for authorities’ COVID-19 tweets that used a hashtag was 0.64 the number of retweets for tweets that did not carry a hashtag, indicating a negative impact of hashtags. This result was highly significant (
When both authorities’ and experts’ tweets had images, the retweet count was not significantly different compared with tweets without an image. Using URLs in an authority’s tweet reduced the success of the tweet with respect to retweet counts: the number of retweets was approximately 0.82 that for authorities compared to tweets without URLs (
As a second category of intrinsic message features, the effect of different content and themes of tweets was analyzed, considering six content variables: tweets containing words referring to the severity of COVID-19, susceptibility, efficacy, tweets containing technical information about the spread of the virus, as well as tweets containing words referring to social and political consequences of the COVID-19 pandemic. As for the structural categories, there were fundamental differences in the success of COVID-19 crisis communication between authorities and experts. The first content category considered was severity. Authorities’ tweets containing words referring to the severity of COVID-19 were associated with 1.40 more retweets than those of authorities’ tweets that did not refer to severity (
The retweet count of authorities’ COVID-19 tweets that contained informative words about susceptibility (eg, regarding SARS-Cov2 at-risk groups) was not significantly different from that of tweets without this information. For experts, there also was no significant impact of susceptibility content on the retweet count. Tweets with efficacy information increased the tweet success for authorities: the retweet count of authorities’ tweets with efficacy content was approximately 1.34 times that of tweets without this content (holding all other model variables constant). This effect was highly significant (
For authorities, tweets with technical information about the spread of the virus and other scientific findings were positively associated with retweet count. For authorities, the retweet count was approximately 1.45 that of tweets without this information (
Tweets that contained words referring to social consequences (eg, lockdown) led to a retweet count for authorities that was approximately 1.24 times (
The third category of explanatory variables considered the impact of style elements of COVID-19 tweets on the retweet count. COVID-19 tweets written in first-person language had no impact on the retweet count for experts or authorities. However, when authorities’ tweets used second-person language, the retweet count was approximately 1.88 times that of tweets that did not use it (
The followers count, as an independent variable, had a highly significant impact on the spread of tweets for both authorities and experts (
These results reveal that there are differences and similarities in the determinants of the retweet success of COVID-19 tweets between authorities and experts.
Overall, this study indicated strong differences between authorities and experts as to what increases the retweet rate of crisis communication regarding COVID-19 on Twitter.
Over the timeframe studied, authorities and experts tweeted increasingly about COVID-19 when considering the number of tweets (see
The results of the negative binomial regression validated the descriptive analysis in that fundamental differences in crisis communication were observed between authorities and experts in Germany. These can be traced back to intrinsic message features such as the structural, content, and style features of tweets [
Referring to the content of COVID-19 tweets, for authorities, the content categories were mostly positively associated with the retweet count. Content covering susceptibility did not significantly affect spread, whereas information about severity, efficacy, technical information, and tweets about social consequences led to a higher retweet count for authorities.
Content about political consequences of the pandemic led to a lower number of retweets compared to tweets without political information, again for authorities only. This result seems remarkable, especially against the background of crisis communication, because political information is the core element of authorities’ communication. Moreover, tweets referring to technical information about the spread of the virus and research results (eg, regarding vaccine development) led to a higher retweet count for authorities than tweets without these references.
Looking at the COVID-19 tweets of experts, it must be noted that fewer content variables were significant in explaining the retweet count. Only two content variables were significantly and positively correlated with the number of retweets. These were tweets dealing with the severity and social consequences of COVID-19. It is within these categories that the experts’ knowledge was valued by Twitter users.
In contrast to the results regarding the structure of COVID-19 tweets, authorities’ and experts’ tweets using severity information were retweeted more often. For both groups, content about susceptibility was not associated with the retweet count. Information about efficacy and technical information increased the retweet count only for authorities’ tweets. Tweets with content about social consequences led to a higher retweet count for both groups of stakeholders. Strikingly, political content in authorities’ tweets was associated with a lower spread of corresponding tweets.
Style elements in COVID-19 tweets considered first- and second-person–specific words. Authorities should consider increasing tweets written in the second person to directly address users, as the relationship with the retweet count was positive. Style variables were not related to significant differences in retweets of experts. Overall, looking at intrinsic message features of COVID-19 tweets reveals that authorities’ tweets appear to be more designed with the strong use of structural elements such as hashtags and URLs compared with the COVID-19 tweet of experts.
Further, the larger the network of the stakeholders, the larger the retweet count. This indicates that stakeholders need to make great efforts in expanding and maintaining a large network to disseminate their crisis communication messages.
Comparing the results of this study with previous studies on crisis communication prior to the pandemic such as that of Vos et al [
There are also similarities to the swine flu crisis in 2009, which was studied by Kostkova et al [
Overall, confirming previous studies, there was a strong concentration of retweets in COVID-19 crisis communication. Two experts and the Federal Ministry of Health were responsible for 70.26% (544,418/774,865) of all retweets, and as such were COVID-19 influencers in Germany on Twitter. A study on the nuclear crisis of Fukushima [
This study also confirms previous Twitter-specific crisis communication studies in showing that all stakeholders tweeted more about COVID-19 over the course of time [
Overall, there are several differences in crisis communication between authorities and experts regarding COVID-19 in Germany: experts have a larger network of followings and followers, receive a much higher spread via retweets and likes, and engage to a larger extent more directly with Twitter users about COVID-19 themes compared to authorities (which became more obvious after filtering out quotes, replies, retweets, and non-German tweets). Regarding intrinsic message features, the fact that experts use fewer structural and style elements in tweets than authorities and exceed them by far in spread indicates that other aspects such as sympathy, reputation, publicity, reliability, general media presence, and directness/speed in communication are more important for crisis communication on Twitter.
Both groups should tweet more about specific COVID-19–related topics. Tweets with content about the severity of COVID-19 had more retweets compared with tweets that did not make severity references. However, it seems advisable to prevent alarmism in the public. More research is needed to determine how the results can be translated one by one to authorities’ crisis communication, such as when it comes to directness in crisis communication versus preventing alarmism.
It must be noted that Twitter users are not representative of the overall German population. By contrast, only few, albeit more educated, people use Twitter in Germany [
It is important to note again that the comparison between the groups of experts and authorities is limited. The modes of crisis communication of these two groups are different. For example, authorities have many other communication channels (online and offline) that they can use for crisis communication, such as press conferences. Moreover, it can be questioned to what extent authorities use Twitter to inform the public in general or more specific target groups such as journalists or experts themselves. This study only analyzed tweets written in German. To communicate within the scientific community, experts especially use English; however, these tweets were not analyzed.
The study results further indicate that there might be other determinants for the success of specific stakeholders on Twitter, which cannot be observed by only using Twitter data, such as sympathy, reputation, publicity, reliability, and a general media presence during the study period. Overall, the tweets were downloaded on a specific date and as such represent a snapshot of events. The data, retrieved on January 15, 2021, cover a singular period amidst an ongoing crisis with an unforeseeable ending and are also designed to be analyzed retrospectively.
Twitter data represent a powerful information source and are suitable for crisis communication in Germany regarding COVID-19. Some important results can be highlighted. COVID-19 tweet activity mirrors the COVID-19 case numbers in Germany. Both authorities’ and experts’ COVID-19 tweets have higher spread when they are plain and for authorities when they address the public directly. Experts’ success in crisis communication on Twitter outweighs the spread of authorities by far. Experts are more valued as an information source in the pandemic situation than authorities. For authorities, it appears difficult to win recognition during a crisis when their crisis communication is not only related to the specific crisis. Authorities should consider developing separate accounts on Twitter and using these accounts for more targeted communication.
List of stakeholders and number of followings and followers.
Complete word lists.
Additional binomial regression results on likes count.
application programming interface
incidence rate ratio
Robert Koch Institute
This study was funded under the project “A comparative evaluation of online crisis communication of authorities and independent experts in the context of the COVID-19 pandemic as basis for an improvement of crisis communication of the Federal Office of Radiation Protection” (Eine vergleichende Evaluation der Online-Krisenkommunikation von Behörden und unabhängigen Expert*innen im Zuge der Covid-19 Pandemie als Grundlage für die Verbesserung der BfS-Krisenkommunikation) by the German Federal Office for Radiation Protection. Valuable comments by Maren Gruß, Christiane Pölzl-Viol, and Michael Thieme from the German Federal Office for Radiation Protection, and four anonymous reviewers are gratefully acknowledged.
LD: study design, data adjustments, analysis, interpretation, and manuscript preparation; JR: study design, data retrieval, data analysis, and interpretation; KA: study design, literature review, and interpretation; KD: study design and interpretation; WS: study design and interpretation; AG: study design and interpretation.
The authors are researchers at a private research company and received funding from the German Federal Office for Radiation Protection.