RETRACTED ARTICLE: Identify and measure the degree of over-prevention behaviors in the post-COVID-19 era in China

The vaccination was proved to be useful in controlling the COVID-19 [1]. With the spread of vaccines in many countries, the pandemic may be mitigated gradually. Although there are still some lingering impacts on the original business, we have entered into the post-COVID-19 era [2]. Nowadays, many countries have restarted their operation. For example, the Chinese government has been opening schools and workplaces to resume classes and various works since April, 2020 [3]. By doing so, the country tries to strike a balance between reviving human activities and lowering the risk of another wave [4]. China strives to prevent a huge resurgence that, which can cause further losses of health and economy, by not loosening the regular control drastically [5]. However, a contradiction exists between outbreak prevention and economic recovery. For example, Italian governments have closed all schools across the country in a short time. They also restricted the population movement and closed various non-essential business [6]. They have been emphasizing the importance to control the social distance and behaviors. Especially for people with a high exposure risk, such as teachers, governments required them to wear surgical masks at their work time [7]. These prevention measures have played an essential role in handling the first wave of pandemic in Italy successfully [8]. However, when the first wave came to an end, some regions relaxed their previous distance measures to revive the economy. As a result, the second wave hit Italy and resurgences occurred in many regions, causing more deaths [9]. From Italy’s prevention lessons, to preclude the possibility of resurgence and transmission, the government may continue to implement various precautions, such as curbing population flow and disinfecting places that are accessible to people. However, such implementations may also result in some deficiencies. Last year, 150 wild animals in Chongqing, China had died from excessive usage of disinfectant [10]. The United Nations Conference on Trade and Development also reported that plastic wastes from masks, gloves, bottles of disinfectants, and other plastic materials had polluted many beaches and oceans [11]. Aside from the current situation of the environment, the negative effects of social isolation on humans are not negligible. For instance, research finds that in this year, adults have consumed considerable alcoholic drinks and cannabis, underlying latent effects on health [12]. Scientists also revealed some negative psychological effects, such as stress symptoms, frustration, confusion, and anger [13].

To invent an instrument to solve RQ1, We used an exploratory design and analysis method. Firstly we developed an item pool based on a broad literature review and governmental guidelines. The guidelines were introduced by experts in the press conference of authoritative institutions, including WHO, CDC, and the National Health Commission of the People’s Republic of China, to show various items describing over-prevention behaviors. Second, based on a five-point Likert scale, we developed our scale using these items. The scale incorporated a total of 27 items, which have been revised and approved by all the co-authors. Meanwhile, we conducted a pilot test with 100 people aged 20 to 55, collecting their comments to revise the scale further and provide explicit descriptions. Most of behaviors described among these items were excessive in the high-risk and low-risk regions. We defined high-risk regions as districts with an above 50 accumulative confirmed cases and an outbreak in clusters in the past 14 days [20]. Accordingly, we classified the remaining regions as low risk. However, some descriptions could only apply to the low-risk regions. For instance, one behavior describes a situation wherein one wears a mask indoors, such as in an office, chamber, school, and other ventilated rooms, where people can still exercise social distancing from one another. This behavior is excessive in the low-risk regions but not in the high-risk ones [21]. People may show a contrary cognition at different risk levels. Thus, we investigated their degree of agreement on each description in one or two scenarios. For example, one sentence describes that staff settled a disinfection shed at the gate of a community, cleansing the people who entered it thoroughly, which is an over-prevention behavior [20, 22]. Then, we asked Sam how he agrees with the statement concerning a high-risk or a low-risk region. He chose between one to five points, varying from greatly disagree to greatly agree. In accordance with his answers, we averaged the points in each two-scenario item to represent its final grade. X1 to X27 shows all items in Table 1.

Third, we conducted an online survey to validate these items based on the exploratory factor analysis (EFA) and confirmatory factor analysis (CFA). The recommended sample size of the EFA and CFA is at least 300 and 200 [30, 31]. We divided China into seven areas based on a common view formed by Chinese geographical experts. Then, we used stratified random sampling to hand out our questionnaire [32]. The online survey was conducted via the Tencent Questionnaire Platform. It’s a platform wherein editing our questionnaire and making it publicly available is possible. It supports us to specify the characteristics of our participants, and helps to seek our target automatically in its sample database. Figure 1 shows the screenshot of the webpage. We determined some demographic characteristics, including gender, age, region, degree, and marriage. We paid for the service charge, then the platform began to filter to search for the target population. For matching people, it posted the questionnaire to them via SMS or WeChat, and waiting for their answers. When the collected answers reached our specified number, the platform stopped handing out. Then we can download the collected data. Finally, we received a total of 1528 answers from the platform. We tested the reliability of our scale and demonstrated its construct validity by using the Kaiser–Meyer–Olkin (KMO) measure and Bartlett’s test of sphericity (BTS). The Cronbach’s α value was 0.936, demonstrating a good internal consistency of the scale. The KMO value was 0.953, with a BTS result that was statistically significant (P < 0.001). Thus, the collected data was suitable for the EFA. We randomly used 1000 of the 1320 participants to construct the EFA model. We extracted factors on the basis of the principal component analysis (PCA). We followed three criteria to filter invalid items [33]. First, we would delete the one with factor loading less than 0.5. Second, if more than one factor loaded the same item, we would remove it. Third, we would exclude who loaded on unintended factors. Finally, we grouped the remainder and calculated their weight individually. What’s more, for the remaining 320 participants, we used the CFA to evaluate goodness-of-fit indices, including Root Mean Square Error of Approximation (RMSEA), Comparative Fit Index (CFI), Tucker Lewis Index (TLI), and Incremental Fit Index (IFI). We also applied the Average Variance Extracted (AVE) and Composite Reliability (CR) and calculated the correlation coefficient between every two factors to test the convergent and discriminant validity. Besides, we adjusted the model by using the modification indices (MI) reasonably.

To answer RQ2, we firstly collected participants’ basic information, including their gender, age, education degree, employment status, and provincial address. The average answering time was 8 min. We filtered our data by following three rules. First, we filtered 88 instances wherein the recorded time of answering the questionnaire was below 2 min [34]. Then, we set a question instructing the participants to choose number four [34]. Considering that some people may give a wrong choice accidentally or deliberately for fun, we deleted those who did not choose four and those who recorded an answering time of below 4 min, which was half of the average time. At this time, we removed at least 89 questionnaires. Although we considered that the participants must have chosen the same answer because they possess a consistent cognition, those who presented invariant responses were still questionable [35]. For example, John was extremely careful, and he thought that all of the behaviors described in the scale were not excessive. Certainly, he could disagree will all 27 statements. However, we excluded those who gave invariant responses continuously in more than 14 items (half of the total 27), with an answering time below 4 min. Accordingly, we removed 31 more. There are a total of 1320 remaining valid answers. Table 2 shows the sample distribution. Table 3 shows the characteristics of the remaining sample. Among them, 53.1% (701/1320) were males, and 46.9% (619/1320) were females. The distribution in age groups approximated a normality tendency, with people aged 30–39 years accounting for the largest proportion, i.e., 43.7% (577/1320). More than half of the participants had a university degree (66.4%, 876/1320), followed by those with a senior high school degree (17.8%, 235/1320). As for the employment status reported, the majority comprised staff in enterprises (30.1%, 397/1320). Those in governmental agency composed the smallest group, accounting for only 3.8% (50/1320). After developing the model, for all the 1320 participants, we calculated scores of each extracted factor on these items by applying factor score matrix. We integrated the items into cognition variables, representing these factors, respectively. Then, we explored the discrepancy of individual cognition and the relationship between it and demographic characteristics on the basis of the χ² test and Logistic regression. From the statistical results, we identified which group of people tend to behave excessively. We used IBM SPSS 25.0 and IBM AMOS 24.0 to perform all data analyses.

In this study, we developed a measurement model, proving ideal content, convergent, and discriminant validity. We tested our model to fit the investigated data well. We also helped to identify demographic characteristics that can indicate groups of people who should be the focus of decision-makers when promoting health literacy when managing a public crisis. Health literacy for the public is critical because holding appropriate prevention helps reduce the prevalence of infection and harms on human and nature [43].