Government dissemination of epidemic information as a policy instrument during COVID-19 pandemic: Evidence from Chinese cities.

COVID-19 has given rise to a surge in the number of policy instruments used to deal with the pandemic at different levels of governments globally. While much attention has been placed on travel bans, lockdown, social distancing, and economic stimulus packages, government dissemination of epidemic information as a policy instrument has received less attention. Based on 14,637 news items collected from the portals of 79 municipal governments in China, this study aims to 1) conduct a content analysis of news items and construct three key attributes of governments' practices of epidemic information dissemination, namely, comprehensiveness, responsiveness, and the protection of privacy, and 2) study the patterns and determinants of the dissemination of epidemic information. Our results show that these cities vary substantially in how they disseminate statistical data and information on individual cases of COVID-19 infections within their localities, which are shaped by government performance in open data, severity of the pandemic, cities' administrative level, population, and health sector capacities. The findings generate theoretical and policy implications for government dissemination of epidemic information.

Introduction

COVID-19 has given rise to a surge in the number of policy instruments used to deal with the pandemic at different levels of governments globally (Weible et al., 2020). While much attention has been placed on travel bans, lockdown, social distancing (Hale et al., 2020), and economic stimulus packages (Hepburn et al., 2020); the dissemination of epidemic information as a policy instrument has received little attention, despite consensus on the critical importance of having timely releases of COVID-19 information (UNESCO, 2020).

Government dissemination of epidemic information can play a key role in combating public health crises such as COVID-19. First, such information can help the public to make informed decisions in their daily lives during the pandemic. For example, it has been shown that disclosures of the first confirmed COVID-19 cases are one of the most effective measures in dissuading public congregation and motivating people to stay at home (Gupta et al., 2020). Second, timely release of key information can help contain the virus infection by augmenting contact tracing. For example, disclosure of details about individual confirmed cases such as their travel history, activities undertaken, and visible symptoms; could alert people in close contact and prompt them to undergo testing and self-quarantine (Smith et al., 2020). Third, the dissemination of such information helps to dispel rumors and disinformation that can erode public trust during an unprecedented pandemic (Shaw et al., 2020). Effective dissemination of such information can boost public trust, which is critical in dealing with the pandemic (Moon, 2020). These were more than just theoretical prescriptions, but are shown via empirical studies that different level of work when governments disseminate information does correlate with the counter-pandemic performance of the country (Lilleker et al., 2021).

Despite its importance, the dissemination of epidemic information differs considerably across countries. A review of the national portals of the 193 United Nations Member States showed that, as of 25 March 2020, when 438,747 confirmed cases had been reported across 208 countries/territories/areas; 110 countries (57%) publicly disclosed some kind of information on COVID-19, while 83 countries (43%) did not provide any information. The types of information provided by the governments also vary. Some countries only provide basic information such as data about the outbreak, travel restrictions, practical guidance on protection, and governmental response, while others release more detailed statistical information about the epidemic. These disparities exist despite guidelines from the World Health Organization (WHO) that states should disseminate information in well-structured and easily understandable formats to reach the people who need them fast (WHO, 2017).

Literature is scarce when it comes to analyzing the city-level information dissemination amid a major public health crisis, as Machado et al. (2021) observed in a systematic literature review. The great variation in the dissemination of epidemic information found at the city level (Department of Economics and Social Affairs, 2020, pp. 346–348) urgently calls for scholars' attention. The observable variances among cities in the same country already provide ample evidence serving a timely investigation. For example, the information provided by the portal operated by Shenzhen Municipality of China is both detailed and updated frequently, illustrated with clear graphs and tables, and available in both Chinese and English. In addition, consistent effort was made to disclose information about all individual cases of infection. In comparison, the dissemination of pandemic information by many other municipal governments in China is much less extensive and far from consistent. Similar degree of variation is also reported elsewhere such as the U.S. (Wang et al., 2021). There are only a limited number of studies documenting the differences in practices of information dissemination, and none seeking to understand what difference matters, and why. For example, does timely information dissemination during COVID-19 reflect higher degree of transparency? If so, what could explain the higher transparency? Could the answer be linked to related literature, such as the literature on Open Government Data, which is also tightly related to government transparency (Ali Hassan & Twinomurinzi, 2018)?

Our results show that cities vary substantially in how they disseminate statistical data and information on individual cases of COVID-19 infection in their localities, and that practices in epidemic information dissemination may be shaped by government performance in open data, severity of the pandemic, city's administrative level, and health sector capacities. To our knowledge, this paper is among few empirical studies of government dissemination of pandemic information.

The main contributions of this article are as follows. First, it conributes to the existing literature on information dissemination practices by governments during major public health crises as the first empirical study focusing the city-level government. Second, it offers a preliminary assessment of the linkages between the information dissemination practices of cities in China and several factors that are known to influence government transparency in the existing literature. Third, this study joins the rapidly-growing collection of research that focuses on subnational governmental actors and their policies against COVID-19 (Dzigbede et al., 2020; Hu et al., 2020; Kerr et al., 2020; White & Hébert-Dufresne, 2020) by adding information dissemination as a key policy instrument.

The rest of this article is structured as follows. The second section presents a brief literature review, highlighting the gap where there is a lack of literature on crisis-period policy actions by city-level governments. The third section describes the dataset, key measurements, and model specifications. The fourth section reports the empirical results. The fifth and sixth sections discuss the key findings, interpretation of the results, as well as the policy implications. The last section concludes.

Literature review

The disclosure of epidemic information to the public in timely manner is critical for crisis risk communication during public health crises (Glik, 2007; WHO, 2019). Such information may include the number of confirmed cases and their medical treatment, as well as their travel and contact history (Liu et al., 2020). Studies of previous public health emergencies such as SARS, H1N1, and Ebola have shown that information disclosure (ID) can enhance public awareness and lead to increased public support for various containment policies and measures (French, 2011; O'Malley et al., 2009; Schnell & Jo, 2019). A recent study found that disclosures of the first confirmed cases COVID-19 serve as one of the most effective measures in dissuading public congregation and motivating people to stay at home (Gupta et al., 2020).

ID can also be a part of the containment measures during public health crises (Fu et al., 2020). In addition to aggregate statistics such as the total number of infected cases and the death toll, information about individual cases of infection can be released to the public with varying degrees of details (Fu et al., 2020; Jung et al., 2020; Xu et al., 2020). In the case of COVID-19, one study shows that the disclosure of details about confirmed cases such as their travel history, activities, and visible symptoms may alert and prompt people with close contact to undergo testing and self-quarantine (Smith et al., 2020).

Furthermore, disclosure of epidemic information by authorities can be an effective policy instrument in dealing with rumors and disinformation, which is prevalent during public health crises (OECD, 2020). Oftentimes rumors could be locations having unconfirmed infection flooding social media and cause panic (ibid.). Disinformation cause tension, chaos, and decrease public trust precipitously (Shaw et al., 2020), while effective dissemination of epidemic information can boost public trust (Moon, 2020; OECD, 2020).

There is growing interest in the literature in empirical studies of disclosures of epidemic information as governments have begun to pay more attention to this as a policy instrument in dealing with COVID-19. For example, Vasudevan et al., 2020, Vasudevan et al., 2021 find significant variations in the disclosure of COVID-19 data across different states in India, and similar results are found in China (Hu et al., 2020; Liu et al., 2020). Perhaps due to time constraints, these pieces generally stopped at observing the variations without seeking analysis for explanations. More recently, Wang et al. (2021) reviewed the twitter messages released by federal and state governments, and agencies within government based on sufficiency, timeliness, congruence, consistency and coordination. Noting that the existing literature on risk communication “lacks timelier, domain- and context-specific research in understanding the communication dynamics”, they report generally dismal performance for the chosen indicators. The article also has a timeframe covering the first wave from January to April of 2020, allowing them to demonstrate dynamics of ID practices of actors. Our paper takes further improvements as follows: we conduct zoom-in text analysis to investigate the contents of ID, feeding into a more systematic, literature-inspired set of indicators and more importantly, to include privacy concerns of the ID practices by city-level governments which have been highlighted as an issue of concern (Jung et al., 2020). Their study, is also notably short of analytical rigor and absent of comparative design.

Responsiveness is the first key attribute used to assess such disclosures. There has been longstanding consensus in the literature that people need fast and accurate information in emergencies (Norris et al., 2008). Such information should be transparent and convey both good and bad news honestly (Gray et al., 2012). For quickly evolving crises, timely communication is essential as even small delays could cause serious damage to the public's right to know and in the formulation of the correct responses (Heath & O'Hair, 2020). Responsiveness can be assessed by counting the median frequencies of daily updates to epidemic statistics; More frequent disclosures indicate timelier data.

The second key attribute is comprehensiveness. For crises with high degrees of uncertainty, it is desirable to comprehensively disseminate “what is known” to build public trust (Moon, 2020). The lack of comprehensive disclosures will brew rumors to fill the “gap”, raising public anxiety (Kwon et al., 2016; Starbird et al., 2014). Comprehensive coverage of the epidemic information can provide a sense of control that maintains the psychological safety of the general public (Gómez-Salgado et al., 2020; Sun et al., 2020), and the disclosure of detailed information about confirmed cases can highlight to receivers what is important, potentially leading to behavioral changes that ensure residents' safety (Gupta et al., 2020).

While comprehensiveness is a key attribute of the quality of epidemic ID by the government, demands by the public for more detailed information needs to be balanced against the protection of individual privacy (DMG Lab of Fudan University, 2020; Fahey & Hino, 2020; Jung et al., 2020), this protection being another key attribute of such disclosures. For example, in many cities around world, detailed information about individuals confirmed to be infected such as their communities, travel history, and medical treatment have been released to the public in order to identify high-risk areas and prevent more infections (Hu et al., 2020; Jung et al., 2020). However, disclosure of such information may create privacy concerns (ICO, 2020). In China, for example, there were at least 22 cases in which the release of overly detailed information such as travel history enabled identification of the individuals involved, resulting in bullying and public controversy (Ge et al., 2021). What are the best practices when it comes to balancing privacy concerns and ensure public's right to know? What factors may contribute to the tendency to unnecessarily disclose privacy information to the public? The literature has produced few empirical analyses of privacy protections of disclosure practices to address these questions.

There are substantial variations in such practices among local governments. Empirical assessments of the determinants of local government transparency and ID practices point to factors such as the population size of the municipality and the resource capacity of municipal governments, usually approximated using the GDP per capita of the municipality (Araujo & Tejedo-Romero, 2016; Sol, 2013; Tejedo-Romero & de Araujo, 2018); the public administration styles (Anglo-Saxon versus continental European) (Pina et al., 2007); and the degree of openness of decision-making processes to the public (Araujo & Tejedo-Romero, 2016; Muñoz & Bolívar, 2015; Saez-Martin et al., 2017; Tejedo-Romero & de Araujo, 2018) and more recently, the variation of performance of Open Government Data (Ganapati et al., 2019). In addition, government resources for data collection and publication could also affect transparency, especially for smaller cities (Bearfield & Bowman, 2017). On the other hand, there has been no empirical research on the determinants of the practices of ID during an unprecedented public health crisis such as COVID-19, for example, whether findings from other related literature would hold. This paper seeks to extend the works of Hu et al. (2020), Jung et al. (2020), Vasudevan et al. (2020) and Wang et al. (2021), which investigated ID practices of small samples of local governments in China, South Korea, India and federal & state governments in the U.S, to contribute to closing the gap in the literature as identified and discussed above.

The dataset creation and research design

Data sources

Epidemic information dissemination has gained more prominence in China in the last two decades as the country encountered a number of pandemics such as SARS and H1N1 (Zhang, 2010). The Chinese government made numerous public interventions to effectively release information to the public at the national, provincial, and municipal levels (Yang & Cui, 2020). First, having learned from past experience, the Chinese government made a well-founded commitment to information dissemination, the purpose of which was to promote a better public health response (Ma, Deng, & Wu, 2020). Second, past and current crises have caused the government to put in place interdepartmental coordination. For example, through coordination between the National Health Commission and departments managing railway, highway, and aviation transportation networks; these departments have further improved the information collection, disclosure, and linkage mechanism (CCDI, 2020). Third, supplementary laws are passed to further clarify the responsibilities of governments at different levels for different classes of epidemic information produced by national, provincial, and local authorities (Christensen & Ma, 2020).

The main data sources for this study are government websites and portals of prefecture-level cities in China. According to the white paper “Fighting COVID-19: China in Action” published by China's State Council Information Office, there are five stages in China's Fight against the Epidemic in the first wave of COVID-19. Our study period covers the ID practices by city governments for all five stages from initial outbreak to the normalization of the epidemic between December 27, 2019 and April 30, 2020. Given the sample representativeness and data availability, we only consider cities larger than 1.5 million population. Altogether, 79 cities from 21 provinces are included in our database. We collated all news items released from government websites and portals directly related to COVID-19. Such news items are typically found in the websites of the health commissions or health bureaus of the cities in our study. In fact, many municipal governments have set up special portals dedicated to the prevention and control of this virus infection. Over 14,637 items are included in our database. Most of cities in our study are relocated in Eastern or Southern parts of China, and their geographical distribution can be seen in Fig. 1 .

Fig. 1

Map of the prefecture-level cities included in this study.

Note: This map shows the geographical distribution of the 79 cities included in this study.

Content analysis

A content analysis was conducted of the collected news items in order to surface the key indicators that capture the attributes for assessing government dissemination of epidemic information: comprehensiveness, responsiveness, and privacy protection. The inspiration of these three attributes comes from a recent report by the DMG lab of Fudan University. This report was created to assist early diagnosis of possible problems with the ID practices of provincial governments and it only used data from Jan 20th to Feb 15th, which is an early stage in China's response to the COVID-19 pandemic. In this report, the author proposed the following benchmarks for ranking the data disclosure performances of local governments in China: comprehensiveness, responsiveness, and protection of privacy (DMG Lab of Fudan University, 2020). The definition and reasons for inclusion of each benchmark in our paper is presented in Table 1 .

Table 1

The benchmarks of the empirical analysis and supporting reasons.

Benchmarks	Definition	Reasons for inclusion
Comprehensiveness	Conceptual definition: the degree to which the government comprehensively disseminates information to counter rampant rumors and misinformation.Statistical definition: the total number of indicators of statistic information and case-related information each news item released or reported.	The indicators are full-spectrum reflection of COVID-19 patients in a snapshot. Full inclusion of them may help dispelling rumors such as the government hides the true number of deaths from COVID-19.
Responsiveness/timeliness	Conceptual definition: frequencies of daily updates to epidemic statistics.Statistical definition: the time lag between the first case confirmed and the first news item disclosing epidemic information. We take the reciprocal of this variable.	Disseminating information about COVID-19 timely is arguably the most important benchmark evaluating the ID practices of local governments.
Protection of privacy	Conceptual definition: risk of inflected individuals being identified if the certain information is disclosed in combination.Statistical definition: We use the concept of privacy risk which is measured by whether the news item have one of the five scenarios under which there is a risk of inflected individuals being identified.	Whether to pose more identifiable information poses a dilemma to city administrators: more personal information may stimulate higher public awareness but at the same time the patients may face more negative public pressure which could scare people having exposed to them to delay seeking medical attention. Therefore, what choice was taken by each city-level government is worthy of attention.

We take the following steps in the content analysis:

Classify the news items into four categories based on the type of information disclosed: statistical data about inflection cases locally, detailed information about individual cases of infection, news about the pandemic, and announcement and advisory about anti-epidemic measures. The primary difference between the first two categories is that statistical data summarize the overall information of the city and its districts in terms of the number of daily new cases, total confirmed cases, cases that are severe, receiving ICU treatment, stable, recovered or deceased, etc. In contrast, information on individual confirmed cases reports travel history, prior exposure to other confirmed cases (if any), areas where confirmed cases have been detected, modes of transportation used, time when symptoms became visible or when treatment was sought, etc. The types of information listed above serves as indicators for accessing the comprehensiveness of each news items.

Manually code the information contained in the news items. For example, for the first category, the statistical information, there are 11 indicators and for the second category, information on individual confirmed cases, we have 22 indicators. All of the indicators were binary assigned a value of 1 if a news item reported this indicator, otherwise 0. Comprehensiveness are therefore calculated separately for statistical information and information on individual confirmed cases based on the above indicators.

Table A1 shows the selection and detailed explanation of the indicator used in this paper. During the actual process of data coding, the research team was divided into two teams for coding, and they cross-checked each other's coding results to ensure accuracy.

The measurement of key variables

The key dependent variable is epidemic information dissemination practices by local governments which is measured by the three attributes: comprehensiveness, responsiveness, and the protection of privacy. For statistics, comprehensiveness is captured by the total number of indicators each news item released or reported. Responsiveness is captured at the city level and measured by the time lag between the first case confirmed and the first news item disclosing epidemic information. We take the reciprocal of the total number of lagging days as the variable for responsiveness (see Table 1). While information about contact history and the use of public transport can be of practical importance, the inclusion of detailed information about place of residence, place of work, and car license number of infected individuals can lead to concerns over privacy because such information may allow individuals to be identified. We therefore consider the issue of privacy in our analysis and construct five scenarios under which there is a risk of inflected individuals being identified if the certain information is disclosed in combination (see Table 2 ). We measure it by the number of types of privacy concerns per news item.

Table 2

Scenarios with potential privacy concern.

Variable	Type 1	Type 2	Type 3	Type 4
Surname	Yes	Yes	Yes	Yes
Gender	Yes	Yes	Yes	Yes
Age	Yes	Yes	Yes	Yes
Work unit		Yes
Type of work				Yes
Car license number			Yes
Residential community	Yes	Yes	Yes	Yes
Hospital name	Yes

We include two key explanatory variables—government performance in Open Government Data and general government efficiency—in the multivariate analysis. Since open data performance is a key dimension highly relevant to information availability (Ali Hassan & Twinomurinzi, 2018), we hypothesize that government performance in pandemic ID, which can be proxied by combining indicators shown in the sections above, is positively correlated to generic government performance in open data during normal times. This is based on two assumptions: First, local governments that are proactive in open data are more willing to disclose information about the pandemic and individual cases of COVID-19 to respond to public concerns; second, local governments that are more experienced in open data practices should be able to use their skills for ID, from collecting to reporting. We operationalize government performance in open data using China Open Data Index (Fudan University, 2019),1 which is the first professional index in China measuring the data opening level of local governments. For the measurement of local government efficiency, we utilize a specific ranking list from Research Report of Local Government Efficiency in China (2020) provided by Beijing Normal University and Jiangxi Normal University. This rank indicates the ability of local governments to achieve optimal government output and predetermined administrative goals at lower costs and using fewer resources.

The next set of factors we consider relates to the nature of the crisis. Previous studies have shown that, rapid, severe and transnational crises are more likely to attract the attention of organizations and leaders (Boin, 2019; Kehinde, 2014). We expect that the severity of the pandemic is positively correlated to government information dissemination. We use the daily number of confirmed cases in each city to measure government challenges in addressing the pandemic.

We also look at specific health-sector capacities, a factor we compute using statistics on the number of hospitals, hospital beds, and doctors practicing in the city published by the China City Statistical Yearbook 2020. We expect to see a positive relationship between health-sector capacities and information dissemination as the former should give city governments more resources to treat COVID-19 patients, find and quarantine their close contacts, and collect epidemic information for dissemination.

In addition, we hypothesize a positive relationship between the city administrative level measured by whether the city is a province capital city or not, GDP per capita, population, and performance in ID, since large cities are expected to have more resources in handling the crisis.

Empirical analyses

Descriptive analysis

As shown in Table 3 , cities vary significantly in the number of news items across these four categories. For example, 50.6% of cities have released 101–200 pieces of statistical information about the epidemic, while there are fewer items in the other categories.

Table 3

Distribution of frequency of the information dissemination by all cities and type of information.

Type of information	None	1–50	51–100	101–200	>200	Total
Statistical data about local epidemic	0	16	21	40	2	79
	(0%)	(20.3%)	(25.3%)	(50.6%)	(2.5%)	(100%)
Information disclosure about individual cases	0	42	11	22	4	79
	(0%)	(53.2%)	(13.9%)	(27.9%)	(5.1%)	(100%)
News items about pandemic	19	42	8	6	4	79
	(24.1%)	(53.2%)	(10.1%)	(7.6%)	(5.1%)	(100%)
Announcement of anti-epidemic measure	23	46	6	3	1	79
	(29.1%)	(58.2%)	(7.6%)	(3.8%)	(1.3%)	(100%)

Note: This table presents the number of cities reporting 0, 1–50, 51–100, 101–200 pieces of news items for four types of information respectively, and the percentage of total 79 cities is presented in parenthesis.

Comprehensiveness of ID

Table 4 presents the comprehensiveness of ID of two categories of news items. The majority of news items disclosed statistical information about local epidemic reported 4 to 6 indicators (41.0%), followed those reported 6 to 8 indicators (26.6%), and only less than 1% reported more than 8 indicators. For the news items on information about individual cases, of all 21 indicators in this category, about 68.6% reported 0 to 4 indicators, 19.8% reported 4 to 8, and less than 1% reported more than 16 indicators.

Table 4

Comprehensiveness of ID.

Type of information	Number of indicators	%
Statistical information about local epidemic	0–2	8.6
	2–4	23.4
	4–6	41.0
	6–8	26.6
	>8	0.4
Information about individual cases	0–4	68.6
	4–8	19.8
	8–12	8.5
	12–16	3.1
	>16	0.1

Analysis is also performed on the coverage of specific indicators and the results are shown in Table A1. We can see that 94.6% of epidemic statistics include information on the number of newly confirmed cases. This is followed by 84.7% of epidemic statistics including information on cumulative confirmed cases. Statistical information on the number of medical observations and the number of the recovered cases also are also widely reported by prefecture-level cities in China, with 83.6% and 77% of items covering them respectively. In comparison, cities publish the number of people who are no longer classified under suspected cases, and those who are no longer receiving intensive care, with less frequency. Fig. 2 shows how an infected case would move through the system and the key statistical indicators that can be reported by the government in disseminating pandemic information. In addition to new cases and deaths, key statistical indicators on the pandemic situation in cities include accumulated cases, quarantined cases, recovered cases, cases in ICU, new suspected cases, accumulated suspected cases, stable cases and cases discharged from ICU.

Fig. 2

Reporting of epidemic information.

Note: Daily increase of the confirmed cases, accumulated number of the confirmed cases, number of cases receiving intensive care, the number of people under quarantine, number of cases that have recovered, number of cases that are under stable condition, number of cases that have deceased, total number of suspected cases, daily increase of the suspected cases, number of suspected cases that are confirmed as not infected, and finally number of cases that are no longer receiving intensive care.

While numbers of newly confirmed cases and deaths are widely reported and quoted, many other epidemic statistics can be of great importance to the public. For example, the number of suspected cases can be quite informative when testing may not be available. The number of recovered patients can also be important for the public to understand the lethality of the illness.

We also find substantial variation in the coverage of statistical information, both across cities and within cities over time (see Table A2). For example, in Zunyi the average number of statistical indicators reported is about 7, while it is less than 1 in Guiyang; despite both being in the same province. The fact that cities differ considerably from one another despite being in the same provinces may suggest that the dissemination of the epidemic information is decided by city governments, and not dictated by directives from upper-level governments.

There is also significant variation over time. For example, in Taiyuan the lowest number of statistical indicators reported is 0, while the highest number is 10. Large variations are found in a number of cities such as Nanchong, Qiqihar, Huhehaote, Baotou, and Taian. As shown in Fig. 3 , the average number of indicators of statistic information in each news item increased rapidly firstly and then stayed at a stable level. While for the case-related information, it shows a different trend that the number of indicators increased first, then decreased, and finally stayed at a relatively low level. What the above analysis shows is that a city would typically start its reporting with a narrow range of statistical indicators, gradually increasing the scope of statistical data disseminated. This could be evidence of increasing demand from the public for more information, and/or evidence of horizontal and vertical policy learning or diffusion.

Fig. 3

Time trend of the average number of indicators of statistic information (a) and case-related information (b) in each news item.

Note: This figure shows the changes of the average number of indicators per news item for statistic pandemic information and case-related information during our study period respectively.

We find that all cities have adopted the practice of disseminating detailed information about individual cases of infection, although the coverage of specific indicators varies a lot (see Table 3). Below is an example from Shenzhen for this type of news item:

Case 37 is a 17-year-old male student of Chinese nationality returning to China from the United States. He flew from New York to Amsterdam, the Netherlands on flight KL646 on April 3, then boarded flight CZ308 for Guangzhou and entered the Chinese mainland via Guangzhou Baiyun International Airport on April 4. His health status declaration information showed no abnormalities, and he was later sent to a designated quarantine point in Shenzhen by a designated vehicle. His preliminary nucleic acid result and reviewed result released on April 6 were both positive. He was confirmed of COVID-19 infection after a group consultation by experts on April 7.

The patient has been transferred to Shenzhen No. 3 People's Hospital for treatment. Primary tracing results show that no person on the Chinese mainland has had close contact with the patient.2

As mentioned in the literature review, the disclosure of such information to the public could slow down the rate of virus transmission as it can help the residents identify areas with higher infection risks. It could also alert them sooner if they are in close contact with infected cases.

Our database contains 5919 news items reporting detailed information about individual cases of infection. Our analysis shows that three types of information are included in such items: basic information (e.g., demographic information, residence), behavioral information (e.g., contact history, use of the public transport), and medical treatment and current condition. Table A1 shows the indicators used, their explanations, and their incidence across these 5919 news items. In addition to information on gender (28.4%), age (29.8%), and district (73.1%); information on current status (68.2%), condition (34.1%), and symptoms (19.0%) are included in the bulk of such news items.

Responsiveness of ID

Responsiveness in the dissemination of epidemic statistical information is assessed by measuring the times of the first news items reporting statistical information relative to the dates of the first confirmed cases in a locality. Fig. 4 shows that 37 out of 74 cities (46.8%) started to disseminate epidemic statistical information on the same day as the first confirmed case, while 19.0% of the cities in this study disseminated such information more than one week later.

Fig. 4

The timing of the epidemic information reporting.

Note: This figure presents the percentage of the time lag between epidemic information disclosure and the first confirmed case among the 79 cities.

Protection of privacy in ID

The presence of the scenarios with potential concern over privacy is found in the news items released by a few cities such as Qingdao, Xuzhou, Guangzhou, and Fuyang (see Table 5 ). While the provision of more detailed information about individual cases can play a key role in combating the spread of the virus, it is critically important for the government to strike the right balance between public health benefits and the protection of privacy. The disclosure of information with privacy implications should be removed from news items released by the government.

Table 5

The occurrence of cases with potential privacy concerns.

City	Type 1	Type 2	Type 3	Type 4	Total
Qingdao	15	1	0	3	19
Xuzhou	12	3	1	1	17
Guangzhou	0	0	7	7	14
Fuyang	8	0	0	0	8
Zhongshan	0	0	0	7	7
Weihai	2	2	0	2	6
Harbin	3	2	0	0	5
Yantai	5	0	0	0	5
Zaozhuang	0	0	4	0	4
Yangzhou	0	1	1	1	3
Qiqihr	0	1	0	2	3
Qingyuan	0	0	0	2	2
Zibo	0	0	0	2	2
Nanyang	0	1	0	1	2
Shangqiu	0	1	0	1	2
Dongguan	0	0	0	2	2
Zunyi	0	0	0	2	2
Jiamusi	0	0	0	1	1
Jinan	0	0	0	1	1
Jining	0	0	0	1	1
Zhengzhou	0	1	0	0	1
Baotou	1	0	0	0	1

Note: The cities are ranked in a descending order by the total number of news items with privacy risk. Overall, 22 of the 79 cities disclosed epidemic information with privacy concern.

Most of the sampled cities (72%) didn't disclose news items with privacy concern in their ID practice during our study period. But we still find that 22 cities had potential privacy issues in the process of ID (see Table 5). Among them are developed cities in the east or southeast coastal areas such as Qingdao and Guangzhou, as well as some economically underdeveloped cities in the central and western China. By analyzing the trend of the percentage of news items with privacy concern over time, we find news items with privacy concern in ID practice increases first and then decreases substantially after February 2020 (see Fig. 5 ). This could be evidence of growing awareness of privacy protection, and/or evidence of horizontal and vertical policy learning or diffusion.

Fig. 5

Trend of the percentage of news items with privacy concern over time.

Note: We calculate the percentage of news items with privacy concern by dividing the number of news items with privacy concern by the total number of news items disclosed each day.

Multivariate analyses

What may have shaped the performance of government information dissemination during the COVID-19 pandemic? As shown in the Descriptive analysis section above, epidemic information dissemination varies across city governments in China in comprehensiveness, responsiveness and protection of privacy. In this section, we ask what may account for the variation in the performance of epidemic information dissemination. We focus on a number of factors related to general governance, the nature of the crisis, and specific health-sector related capacities as well as socioeconomic characteristics.

Because the key dependent variables in our model are all count data variables, we utilize Poisson and Negative Binomial regression models to examine factors associated with the comprehensiveness, responsiveness, and privacy protection of government information dissemination. To choose which model to use, we consider the over-dispersion problem and conducted likelihood ratio test. The descriptive statistics of the variables included in the regression models are reported in Table A3.

Table 6 displays the results of multivariate analysis. In Model (1) and (2), we regress the independent variables on comprehensiveness measured by the number of indicators of statistical information about local epidemic and the number of indicators of information on individual confirmed cases. Model (3) and (4) are regressed on the other two attributes: responsiveness and privacy risk, respectively.

Table 6

Multivariate regression results of government information disclosure.

Variable	Model (1)	Model (2)	Model (3)	Model (4)
	Comprehensiveness of statistic information disclosure	Comprehensiveness of case-related information disclosure	Responsiveness	Privacy risk
Government open data performance	0.003	0.030⁎⁎⁎	−0.002	0.032⁎⁎
	(0.006)	(0.004)	(0.005)	(0.014)
Government efficiency (rank)	0.001	−0.001	0.002	−0.003
	(0.003)	(0.004)	(0.003)	(0.012)
Daily confirmed cases	0.083	−0.107⁎⁎	0.007⁎	−0.781
	(0.059)	(0.048)	(0.004)	(1.281)
Capital city	−0.234	−0.578⁎⁎	−0.017	−1.660⁎
	(0.261)	(0.243)	(0.275)	(0.906)
GDP per capita (yuan)	0.036	0.001	−0.001	0.090
	(0.035)	(0.039)	(0.037)	(0.115)
Population (unit = 10,000 persons)	0.002⁎⁎⁎	−0.001	0.000	0.003
	(0.001)	(0.001)	(0.001)	(0.002)
Health sector capacity	−0.292	0.385⁎	0.093	−0.045
	(0.213)	(0.214)	(0.220)	(0.624)
N	14,637	14,637	79	14,637
Pseudo R2	0.016	0.033	–	0.087
Log Lik.	−31,926.948	−22,358.875	−100.03837	−526.25035

Note: This table shows the results of the multivariate regression analyses for the four dependent variables: comprehensiveness of ID for statistic information, comprehensiveness of ID for case-related information, responsiveness of ID and privacy risk of ID. Models (1) and (2) use negative binomial model, and Model (3) uses Poisson regression model based on the results of Likelihood-ratio test. Model (4) uses ordered logit model for the categorical dependent variable. Health sector capacity is a factor computed using statistics on the number of hospitals, hospital beds, and doctors practicing in the city. Standard errors are clustered at the provincial level.

p < 0.1.

p < 0.05.

p < 0.01.

From a first analysis of the results obtained in Table 6, the coefficient of the government open data variable is statistically significant in Model (2) and (4), showing that better government performance in open data is associated with higher comprehensiveness in ID during the pandemic but also high privacy risk. This result suggests a trade-off between information transparency and privacy protection as existing literatures noted (Moon, 2020; Agozie & Kaya, 2021). In contrast, government efficiency is not significant in any of the six regression models, which suggest that government with higher efficiency in normal period doesn't necessarily perform better in ID practice during a pandemic.

We hypothesize a relation between more severe outbreaks and better performance of ID by governments. Model (2) reveals that the average number of confirmed cases per day has statistically and significantly negative association with the comprehensiveness of ID on case-related information. A possible explanation is that as the number of daily confirmed cases increases, the workload and the difficulty of collect information about individual cases increase; and the government may therefore focus more on disclosing information about the statistic information in general instead of individual cases in order to get attract public attention. The results in model (3) indicate that for cities where there are more confirmed cases, the government is more responsive to disclose the epidemic information. Also, consistent with our assumption, the results suggest that big cities with more population and higher health sector capacities measured by the richness of medical resources such as doctors, beds and hospitals are associated with better performance in comprehensiveness of ID as shown in model (1) and model (2). One interesting result is that capital cities performed worse in comprehensiveness of ID practice while have a higher sense of privacy protection as shown in Model (2) and Model (4).

Discussion

Why do the city-level government actor matter? City-level governments play a pivotal role in responding to public health emergencies, and our research data show that even in China, a country that is known for centralized power, local governments still disseminate COVID-19 information in ways that are most relevant to local populations (Christensen, & Ma, 2021). This plays an important role in the country's fight against the pandemic. The fact that local governments, even those within the same province, vary in their practices of ID shows that they are taking the initiative in their ID practices, thus delivering discrepant performances.

In our framework comparing the information dissemination practices of cities, we focus on comprehensiveness, responsiveness, and protection of privacy. Through both text-based and quantitative analyses, the following discoveries are highlighted in connection with the relevant literature:

Firstly, city-level governments are quick to start disseminating local COVID-19 information once it becomes available. This is valued irrespective of their content. For example, while new confirmed cases and total confirmed cases undoubtedly are important disclosures, the number of discharged patients and recoveries from critical conditions also are valued and reported by many city jurisdictions.

Secondly, although they are responsive, cities do not seem to have established any consistent structures for the contents of their ID practices, as demonstrated by fluctuations of indicators reported by some cities. This prompts follow-up questions about the (structural) reasons behind such disparities, especially why some cities release detailed information of individual COVID-19 cases even though most cities do not. Media in Guangdong province ranked the comprehensiveness of travel history information of COVID-19 patients released by local health authorities, and this prompted city governments to make improvements to their ID, and the provincial health authority to follow up with clearer requirements for travel history collection and dissemination (Nandu News, 2020). Such reports are examples of what is required to answer these questions.

Thirdly, the majority of sampled cities have no privacy risk problems in their ID practice during our study period. But we still find 28% of them exposed potential privacy issues in the process of ID. When comes to the time trend, we find this kind of privacy risk in ID practice increases first and then decreases substantially after February 2020. This could be evidence of growing awareness of privacy protection, and/or evidence of horizontal and vertical policy learning or diffusion, which may be interesting for further research.

Lastly, multivariate analysis in this paper also prompts interesting questions for relevant theoretical discussions. In addition to our analyses showing that government performance in open data, pandemic severity, city's administrative level and health sector capacities are associated with city governments' practice of ID; our findings also include two interesting results. One is that local governments that are more advanced in open data also disseminate information with higher levels of risks to privacy. The result echoes concerns raised in literature that more information may not be necessarily virtuous (Argente, Hsieh, & Lee, 2020; Jung et al., 2020). Whether and how exactly the mechanism is remains to be investigated. The other surprising finding is that government efficiency is not significantly correlated with government performance in ID during the COVID-19 pandemic. One possible interpretation may be that government efficiency, which is evaluated based on data of policy actions during routine contexts, might translate quite differently into information dissemination practices during emergencies. City leaders and administrators are suddenly made aware that they need to mobilize cooperation between departments and workers not trained or used to coordinating among each other (Huck et al., 2020). Many did not expect that their performances would be so immediately and intensively scrutinized by their superiors, the (national, central, and local) media, and citizens. This scrutiny occurred both offline and online, and examined their performances both in absolute and relative terms, singly or connected to other capabilities. These two findings obviously challenge existing wisdom and therefore warrant further research.

Policy implications and takeaway for practice

For city administrators, the findings of this paper provide four policy implications. First, cities should realize that enhancing transparency generates much synergy and therefore should be placed at a higher priority; Second, they can use the framework developed in this study to conduct self-diagnosis of local information dissemination, and to improve their interactions with citizens and media for better transparency, possibly sharing techniques to improve Open Government Data. Third, cities should put privacy protection at a higher level when deciding whether to disseminate information about individual cases (Li, Ma, & Wu, 2022). Our findings also reveal that the comprehensiveness and responsiveness of ID practices are not highly related to government efficiency, and local governments should improve their ID even if their efficiency have previously been impressive. Finally, disparate ID practices may warrant higher-level political authorities to specify a framework for ID during public health crisis, in light of their experiences and lessons learned.

Conclusion

We find that cities with higher level of government open data perform better in ID practice in terms of disclosing information comprehensively during the pandemic. Under such emergencies, previous experience and preparedness present a significant contribution to ID practices. However, the level of government open data is positively correlated with privacy risks of ID. We also find that pandemic severity is highly correlated with lower comprehensiveness of case-related information and higher responsiveness of government ID during the pandemic, partly due to the fact that cities exposed to higher risks of infection are occupied with disseminating key information of the general statistic information to citizens to help them make the right decisions timely, instead of focusing on detailed information of individual cases. Lastly, city's administrative level is also correlated with the performance of ID measured by comprehensiveness and privacy risk.

While preliminary, this paper demonstrates that there exists significant under-exploited potential to collate and study subnational governmental actions under COVID-19, which may inspire discussion and prompt re-examinations of subnational, city-level governmental actions in response to emergencies in ways that contrast with existing understanding. We should also emphasize that the results presented here can only infer factors correlated to the government ID about the COVID-19 pandemic rather than explaining their causal effects.

Our study is limited in several ways and we call for future research to deepen our understanding of government openness during crises. First, the sample size of the cities included in this study can be expanded in future studies, both within China and beyond. This will help to include more cities with divergent attributes. Second, our data are mainly from the first wave of COVID-19, and data from later waves may show different patterns. For example, several cities are experiencing second and third waves of COVID-19, and their ID practices in response to these later waves may be added into the dataset. Lastly, more qualitative evidence from in-depth studies can be used to identify the mechanisms performing government information dissemination. Despite these limitations, as one of the first empirical studies to quantify government information dissemination at the city level, our findings contribute to ongoing debates about pandemic responses across cities.

CRediT authorship contribution statement

Xun Wu: Conceptualization, Methodology, and Writing - Reviewing and Editing.

Lei Shi: Data analysis and Writing - Original draft preparation.

Xinyu Lu: Software and Data collection.

Xiaotong Li: Data analysis and Writing - Original draft preparation.

Liang Ma: Conceptualization, Methodology, and Writing - Reviewing and Editing.

Declaration of competing interest

National Natural Sciences Foundation of China (Grant/Award Number: 71774164).