Metaphors about the COVID-19 pandemic have spread as quickly as the virus itself (Semino, 2021). For instance, policymakers around the globe have made use of militaristic metaphors to emphasize the dangers of the SARS-CoV-2 virus and to prompt citizens to behave in a careful fashion (e.g., Heffernan, 2020). In order to test the performative effects of different metaphors about the pandemic, we conducted several cross-cultural framing experiments across different language contexts (see Schnepf & Christmann, 2022). In doing so, we also exploratively collected data on a newly developed scale aiming to quantify people’s paradoxical attitudes towards preventive measures. In this article, we aim to present the whole dataset of our framing project and, in more detail, how this data and the data of the newly developed scale could be used by researchers who are interested in the interplay of person-level variables and prevention intentions.
Especially at the beginning of the COVID-19 pandemic in Europe in 2020, many popular opponents of the strict COVID-19 measures were under the fallacy that the consequences of the COVID-19 pandemic could be mitigated primarily by protecting high-risk groups such as the elderly or the pre-diseased (e.g., Reiss & Bhakdi, 2020; Helsloot et al., 2020). This point of view clearly neglected the rationale of the epidemiological prevention paradox in the COVID-19 context and underestimated many relevant factors influencing the exponential infection rates and the risk of overburdening the health care systems (Science Media Center, 2020).
When the strict COVID-19 policies had led to significant reductions in infection rates in many European countries, the very same opponents of these measures interpreted their effects (i.e., low infection rates) as “proof” that they had not been necessary at all because the catastrophe had not occurred. In this vein, two of the most popular proponents of the German Querdenker movement and anti-vaccinationists, Karina Reiss and Sucharit Bhakdi (2020), wrote about the significantly reduced infection rates after the implementation of the COVID-19 measures that these were “not a success of distancing and hygiene rules, [but that] the Corona viruses retreated, as it does every year in May” (p. 72). Moreover, they accused the government of having declared a “(…) completely useless lockdown when it was all over” (Reiss & Bhakdi, 2020, p. 82).
These and similar attitudes have been often referred to as the “COVID-19 prevention paradox” (e.g., Schreiner, 2020). The term “prevention paradox” initially stems from epidemiologist Geoffrey Rose (1981) who used it to describe the paradoxical situation that a preventive measure that has a high benefit for the population as a whole has often a barely noticeable benefit for the individual – and vice versa. More precisely, Rose (2001) found that preventive behaviors of the majority help to reduce the incidence of coronary heart diseases better than do preventive behaviors of the affected high-risk groups only. This epidemiological relationship is explained by the fact that the overall incidence of coronary heart diseases is higher in the low- and moderate-risk groups than in the high-risk group (Rose, 2001, p. 430). The prevention paradox holds for several health issues such as alcohol abuse (Rossow & Romelsjö, 2005), diabetes (Bernstein et al., 2018), or hypertension (Dobe, 2013).
In psychological science, this phenomenon is more in line with the rationale of a self-defeating prophecy, which can be seen as the counterpart of the self-fulfilling prophecy (see Merton, 1936). A self-defeating prophecy describes the situation that the occurrence of a predicted event becomes less likely just by predicting the event (e.g., Sabetta, 2019). This relationship mainly refers to the prediction of negative events and is due to the fact that many people behave in a preventive way to avoid the negative prediction. The overall result is that the prophesied negative event does not occur – or if so – to a much lower probability.
With regard to the above-mentioned belief that COVID-19 measures had been unnecessary, both phenomena might come into play. The ironic paradox of this situation thus is that believers in a negative event (e.g., high infection rates and overburdening of hospitals) successfully prevent this event from occurring, while non-believers consider the non-occurrence of the same event as proof that the preventive measures would not have been necessary in the first place. There is neither a fully appropriate epidemiological term for this attitude nor one from psychological science. We call this bundle of self-contradicting, anti-prevention attitudes the “paradox of prevention”. Such inherently inconsistent attitudes towards measures to combat the pandemic have been put forward in particular by opponents against COVID-19 policies (e.g., Querdenker and anti-vaccinationists/anti-vaxxers). This is a serious problem, because much research on effective pandemic control shows that collective compliance with the measures adopted to fight the virus are crucial for their success (see, e.g., Bierwiaczonek et al., 2020; Dohle et al., 2020). Previous research on mask-wearing, for example, was able to show that face masks were worn more frequently by individuals who reported high belief in scientific findings (Stosic et al., 2021). Persons with low mask-wearing compliance, on the other hand, were more likely to use it incorrectly, e.g., by not covering their nose (Machida et al., 2020). Similarly, restrictions in private life such as voluntary contact reductions or self-testing require a high level of civic compliance. Paradoxical attitudes toward preventive measures, however, can lead to a reduced endorsement of and compliance with these measures.
One aim of this article, therefore, is to propose a measurement for such paradoxical attitudes towards preventive behaviors in the COVID-19 pandemic. In this open data article, we present a dataset from a larger research project on framing effects during the COVID-19 pandemic in which we, among other scales, exploratively collected data of the newly developed Paradox of Prevention Scale in five samples across different country contexts. In addition, the introduced dataset also provides many other scales such as a conspiracy mentality measure, individualism-collectivism or risk aversion which researchers can use to assess which attitudes and personality traits will predict COVID-19 policy support, as well as preventive attitudes and behaviors. The data is openly available for further use and its reuse potential for researchers of different disciplines is discussed at the end of the article.
2.1 Study Design
The data used and presented in this manuscript stems from a larger research project on the effects of metaphoric framing on attitudes and behaviors in the COVID-19 pandemic. All data in this dataset is cross-sectional and was collected via online questionnaires. Participants of samples 1 and 2 participated in a cross-cultural framing experiment in the United States (U.S.) and Germany, which tested the performative effect of the WAR metaphor in contrast to the concept of STRUGGLE on participants’ cognitive, emotional, and attitudinal reactions towards the pandemic (see Schnepf & Christmann, 2022). We also assessed the interaction between both frames and the reporting style of the presented vignette, which was either written in a sober, matter-of-fact style or in an emotionalized manner. In both studies, we found very limited evidence of the effectiveness of the metaphorical framing of the COVID-19 measures.
The data of samples 3–5 was collected in Germany, Austria, and the United Kingdom (U.K.), and were part of the same research project as samples 1 and 2. In these studies, it was experimentally varied whether dealing with the COVID-19 pandemic was framed as a WAR, STRUGGLE, CONFRONTATION, WILDFIRE or JOURNEY. This time, we did not manipulate the reporting style in which the short text excerpts were presented. There was no support for any metaphor framing effects in samples 3–5 (see Schnepf, under review).
We think that the overall dataset can be of relevance for researchers who are interested in the interplay of different person-level variables such as conspiracy mentality or distrust in science and the endorsement of preventive attitudes and behaviors.
2.2 Time of Data Collection
Data of samples 1 and 2 was collected in the U.S. and Germany in July 2020, at a time when the infection rates in both countries had fallen compared to the beginning of the COVID-19 pandemic in March and April 2020. The exact time periods of the studies are displayed in Table 1. Data of studies 3–5 was collected in the context of another cross-cultural study on metaphor framing effects during the COVID-19 pandemic. The samples were recruited in February 2021, at a time when infection rates in all three countries had rapidly risen again, and national lockdowns have been adopted to deal with the second wave (see Austrian Ministry for Digital and Economic Affairs, 2021; Kodzo & Imhöhl, 2022; Schulz, 2021). At this time, the U.K. in particular struggled with an extremely high infection level due to the B.1.1.7 mutation (see WHO Coronavirus Dashboard, 2022).
|Field time||Jul 1–4, 2020||Jul 6–10, 2020||Feb 16–22, 2021||Feb 17–23, 2021||Feb 17–23, 2021|
|Age||M = 37.45, SD = 11.35||M = 46.11, SD = 14.24||M = 43.61, SD = 15.18||M = 42.95, SD = 14.51||M = 42.45, SD = 14.54|
|finished school without qualification||3.7%||0.4%||0.5%||1.0%||4.8%|
|secondary school-leaving certificate||14.9%||16.6%||12.0%||19.8%||6.3%|
|general certificate of secondary education (GCSEs)/|
2.3 Location of Data Collection
The data of sample 1 was collected in the U.S., samples 2 and 3 were acquired in Germany, and data of samples 4 and 5 was collected by the same platform, but in Austria and the U.K., respectively.
Participants of sample 1 were recruited via Amazon MTurk (Amazon, 2022) and compensated with 2 $ for study participation. To guarantee a high quality of the data and low dropout rates, we have only invited participants with a HIT approval rate of greater than 98 representing the proportion of completed tasks by the participants. Even though, sample 1 is not fully representative for the U.S. context, as it shows a male bias and above-average educational attainments, MTurk samples have been found to be similar to benchmark population samples in terms of political affiliation and related psychological foundations (Clifford et al., 2015).
The remaining samples have all been recruited by the international crowd-sourcing platform Respondi (Respondi, 2022). Study compensation was by the company’s default 2.80 € for an average duration of 15 minutes. Respondi online recruiting was based on strict representativity standards for each country sample. Invitation mails for study participation were sent out weighted by representativeness criteria for age, gender and educational attainment by Respondi. Table 1 shows the socio demographic characteristics of the samples under research.
In this section, we will present the main variables which we covered in the questionnaires of the five framing experiments conducted at different phases of the pandemic and in different countries. First, we will introduce the newly developed Paradox of Prevention Scale as this is the only variable which we consistently measured across all five samples. This is followed by further variables capturing other COVID-19 related attitudes and behavioral intentions. We will briefly discuss the value of each construct to make transparent why we added these variables in our studies.
2.5.1 The Paradox of Prevention Scale
The Paradox of Prevention Scale consisted of 12 items developed on the base of newspaper reports dealing with paradoxical attitudes about COVID-19 prevention behaviors. After the first lockdown was implemented in many countries in 2020, some people have called for relaxations of the COVID-19 measures, claiming that the measures were exaggerated because infection rates were low (e.g., Reiss & Bhakdi, 2020; Wodarg, 2021). In Germany, this attitude was particularly shared by proponents of the so called Querdenker movement. This attitude is paradoxical, since it was the prior counter measures that had caused the decrease of infection numbers in the first place. We took this phenomenon as an impulse to assess a scale that systematically measures such attitudes across different studies.
Across all samples, the scale was measured as an additional exploratory variable at the end of the respective questionnaires. The scale’s instruction read as follows: “Currently, there is much talk about the spread of the Corona or SARS-CoV-2 virus and related countermeasures. The following questions and statements relate to this current topic. Please indicate to what extent you agree with the following statements.” All items are displayed in Table 2. They were presented in a randomized order. Answers were given on a 6-point Likert scale reaching from 1 = “completely disagree” to 6 = “completely agree.” Internal consistency was high across the different samples (Cronbach’s αSample 1 = .93, Cronbach’s αSample 2 = .93, Cronbach’s αSample 3 = .93, Cronbach’s αSample 4 = .94, Cronbach’s αSample 5 = .88). Depending on the study context, we collected several additional variables that allowed us to validate the Paradox of Prevention Scale. We added a syntax file for the scale validation in the project folder on the Open Science Framework (OSF). Researchers who are interested in the factorial structure and validation of the scale can derive the commented syntax on: https://osf.io/b7sxq.
|1. The so-called countermeasures against the virus are nothing but paternalism for adult citizens.|
|2. The fear of a second Corona wave is superfluous.|
|3. The rapid success of the initial anti-Corona measures shows that they would not have been so radically necessary.|
|4. In the end, there will be more deaths from the economic consequences of the restrictions introduced than from the virus itself.|
|5. The relaxation of the lockdown could have been started much earlier.|
|6. The Corona virus is presented by the media as more dangerous than it actually is.|
|7. The Corona pandemic is no worse than a normal wave of flu.|
|8. If people at high risk protect themselves, this is quite sufficient to control the pandemic.|
|9. Everyone should have the right to decide for themselves what risk of illness they are exposed to.|
|10. Too little consideration is given to the negative side effects of anti-Corona measures.|
|11. The following applies to protective measures: Only if everyone participates the weak are really protected. (R)|
|12. It is a moral duty to protect the weak in society by imposing restrictions on all citizens. (R)|
2.5.2 Distrust in Science
Trust in science plays a pivotal role for individuals’ adherence to COVID-19 prevention measures (e.g., Dohle et al., 2020; Plohl & Musil, 2021). We measured participants’ distrust in science by using the Negative Perceptions of Science Scale (Morgan et al., 2018), in samples 1 and 2. The scale consisted of three subscales, namely perceiving science as corrupt (e.g., “Most scientists are politically biased”), onerus (e.g., “Science is too complicated to understand”), and limited (e.g., “The scientific method is limited”), which were measured by five items, respectively. Answers ranged from 1 = “totally disagree” to 7 = “totally agree”. Given the high internal consistency (Cronbach’s αSample 1 = .94, Cronbach’s αSample 2 = .90, Cronbach’s αSample 3 = .90, Cronbach’s αSample 4 = .88, Cronbach’s αSample 5 = .94), all items were collapsed into one mean index, with higher values indicating a higher distrust in science (MSample 1 = 2.95, SD = 0.94, MSample 2 = 2.79, SD = 0.64, MSample 3 = 2.77, SD = 0.70, MSample 4 = 2.73, SD = 0.66, MSample 5 = 2.97, SD = 1.10).
2.5.3 Trust in Politicians
In addition, we used the Trust in Politicians Scale by Halmburger et al. (2019) in samples 1 and 2, which consists of nine items measured on a 5-point Likert scale reaching from 1 = “totally disagree” to 5 = “totally agree” (e.g., “Politicians are all in all honest with their voters”, Cronbach’s αSample 1 = .87, Cronbach’s αSample 2 = .91). A mean index across all items was determined (MSample 1 = 2.90, SD = 0.79, MSample 2 = 2.73, SD = 0.77).
2.5.4 Conspiracy Mentality
During the COVID-19 pandemic, conspiracy theories have been spreading as fast as the virus itself (Muller, 2020). They typically fulfil people’s need for control and security by providing a comprehensive explanation for negative events involving “(…) secret plots by powerful and malevolent groups” (Douglas et al., 2017, p. 538). In doing so, they often reinforce existing friend-foe schemata and blame a scapegoat to be responsible for the situation (van Prooijen, 2021). According to Imhoff and Bruder (2014), individuals differ in the extent to which they tend to believe in conspiracy theories, which is referred to as conspiracy mentality. In samples 3–5, we assessed this variable by the 5-item Conspiracy Mentality Questionnaire by Bruder et al. (2013). Answers were measured on a 11-point Likert scale from 1 = “certainly not” to 11 = “certain” (e.g., “Events which superficially seem to lack a connection are often the result of secret activities”, Cronbach’s αSample 3 = .92, Cronbach’s αSample 4 = .89, Cronbach’s αSample 5 = .89). A mean index across all items was determined, so higher scores indicate higher conspiracy mentality (MSample 3 = 5.56, SD = 2.67, MSample 4 = 5.93, SD = 2.56, MSample 5 = 6.04, SD = 2.27).
In the COVID-19 pandemic, adherence to preventive behaviors was shown to be more likely among individuals with high empathy and care for others and less likely among those with egoistic motives (Christner et al., 2020; Huang et al., 2020; Ospina et al., 2021). We therefore found it important to cover individualist versus collectivist motives in some of our studies.
In samples 1 and 2, we used the Horizontal Individualism Scale adapted from Triandis and Gelfand (1998). The scale consists of 4 items (e.g., “I’d rather depend on myself than on others”) measured on a 5-point Likert scale ranging from 1 = “completely disagree” to 5 = “completely agree” (Cronbach’s αSample 1 = .71, Cronbach’s αSample 2 = .81). Items were collapsed to a mean score (MSample 1 = 3.90, SD = 0.69, MSample 2 = 3.78, SD = 0.73).
2.5.6 Risk Aversion
Risk aversion, i.e., people’s tendency to prevent risks and seek for security, has been shown to be related to a particularly high commitment with restrictive COVID-19 policies and preventive behaviors such as mask-wearing, social distancing (Nikolov et al., 2020) and vaccination intentions (Neumann-Böhme et al., 2020). Previous research has shown that people with higher risk aversion are indeed better at objectively assessing risks than those with low risk aversion. With regard to the COVID-19 pandemic, it was also found that people with a low vaccination willingness significantly underestimate the actual risks of an infection and overestimate the risks of possible side effects due to a vaccination (Betsch et al., 2020).
Risk aversion was assessed in samples 1 and 2, by the risky choice paradigm adopted from Garbarino et al. (2011). Participants had to choose one out of six decisions across three fictitious lottery games. Choices were ranked dependent on the riskiness of the decision, with lower options indicating riskier decisions. A mean index across all three lotteries was determined. Higher values indicate higher levels of risk aversion (MSample 1 = 2.73, SD = 1.15, MSample 2 = 2.91, SD = 1.48).
2.5.7 Support for COVID-19 Policies
In the research project, one aim was to be able to predict behavior-like measures in the context of the COVID-19 pandemic. For this purpose, in samples 1 and 2, we measured participants’ support for COVID-19 policies in both country contexts by asking them whether they perceived the implementation of selective curfews to be 1 = “not at all useful” to 9 = “very useful” (MSample 1 = 7.14, SD = 1.81, MSample 2 = 6.73, SD = 2.16). In the German sample, we added the additional question of whether they perceived local entry restrictions to be useful or not, measured on the same scale range (MSample 2 = 6.73, SD = 2.14). For the German context, both items were collapsed to a mean score (ρ = .85, MSample 2 = 6.73, SD = 2.00).
2.5.8 Vaccination Willingness
As at the field time of data collection of samples 3–5, support for COVID-19 policies was not a big issue anymore, we assessed participants’ willingness to get vaccinated against the SARS-CoV-2 virus in these samples. Participants were asked whether they intend to get vaccinated as soon as possible. Answers reached from 1 = “no, definitely not” to 5 = “yes, definitely” (MSample 3 = 3.65, SD = 1.43, MSample 4 = 3.79, SD = 1.44, MSample 5 = 4.49, SD = 1.00). Participants also had the opportunity to answer “Don’t know yet, depends on the vaccine”.
2.6 Quality Control
The first stage of quality control of the data was undertaken by the crowd-sourcing platforms which we used for the sampling. The data of sample 1 was collected via Amazon MTurk (Amazon, 2022) and only participants with a HIT approval rate of greater than 98 were invited to participate in the study, which represents the proportion of completed tasks by the participants. Data of all other samples was collected via Respondi (Respondi, 2022), a German crowd-sourcing platform which uses its own (non-published) quality criteria for their samples.
In addition, as all data was collected through online questionnaires via SoSci Survey (Leiner, 2014), all data files contain the same standard quality indicators provided by the company to control data quality. The variable MISSING covers the percentage of answers which have been omitted by a respondent. MISSREL weights the percentage of missing answers of a participant by the answering behaviors of the other participants. DEG_TIME indicates whether a respondent has completed the questionnaire very fast, assuming a low data quality for values above 100. However, Leiner (2019) argues that a more elaborate indicator of a participant’s relative answering behavior is their relative speed index, which is labelled as TIME_RSI. Values in the range of 2.0 and higher are seen as indicators of low data quality and can serve as an exclusion criterion (Leiner, 2019). No exclusion criteria were applied to the descriptive analyses reported above, as the data was overall of high quality (RSI < 2).
2.7 Data Anonymization and Ethical Issues
In all studies, participation was voluntary, anonymous, and financially compensated by the minimum wage of the respective country. Only persons aged 18 and older were allowed to participate. Participants could stop the study at any time without giving a reason. The studies were performed in accordance with the ethical standards of the German Association of Psychology (DGPS) and with the 1964 Helsinki Declaration and its later amendments. The study was approved by the local ethics committee of the University of Koblenz-Landau (application number LEK-233). Informed consent was obtained from all participants (see Appendix 1).
3 Dataset Description and Access
3.2 File Names
The data of samples 1–5 can be found under the following file names: Data_Sample1_US.sav, Data_Sample2_DE.sav, Data_Sample3_DE.sav, Data_Sample4_AU.sav, and Data_Sample5_UK.sav, as well as under: Data_Sample1_US.csv, Data_Sample2_DE.csv, Data_Sample3_DE.csv, Data_Sample4_AU.csv, and Data_Sample5_UK.csv. The file name 2022_Prevention_Paradox_Syntax_Regressions.sps refers to the syntax file, which can be used for the validation of the newly developed Paradox of Prevention Scale. Content codebooks including the experimental treatment texts and materials can be found in the Content Codebooks folder and have the file names: 2022 Content Codebook Samples 1+2.pdf for the documentation of Studies 1 and 2, 2022 Content Codebook Samples 3-5.txt and 2022 Questionnaire Samples 3-5.pdf for the documentation of Studies 3–5. The data codebooks of all samples are indicated by the file Specific_Codebook_Samples_1-5.xlsx, as well as by single files in .CSV format: Specific_Codebook_Sample_1.csv, Specific_Codebook_Sample_2.csv, Specific_Codebook_Sample_3.csv, Specific_Codebook_Sample_4.csv, and Specific_Codebook_Sample_5.csv in the Specific Codebooks folder.
3.3 Data Type
All data files contain the primary data from the surveys conducted in the superordinate research project.
3.4 Format Names and Versions
All data files are stored in .CSV and .SAV format. The specific codebooks are available in .XLSX and .CSV format. The syntax of the validation models is only available in .SPV format and is thus only suitable for use by the statistical software IBM SPSS.
The language in which the datasets and the codebook are publicly provided is American English. However, the data of the German and Austrian studies are also available in German language and can also be provided on request.
The open license under which the data has been deposited is CC0 1.0 Universal.
3.7 Limits to Sharing
The authors declare no limits to sharing.
3.8 Publication Date
The documented OSF dataset was published on 24/03/2022 (Schnepf, 2022, URL: https://osf.io/7zgfj/) and updated on 15/08/2022. The undocumented raw data of the project has been previously published on Zenodo on 31/07/2020 and updated on 10/03/2021 (Schnepf, 2020, URL: https://zenodo.org/record/4730815#.YvodyS-23rA). Sharing the raw data immediately via Zenodo was mandatory after having received the EOSC funding. For the reuse of the data, we recommend to use the fully documented OSF files which are referred to in this article.
3.9 FAIR Data
We followed the 15 FAIR data principles recommended by Wilkinson and colleagues (2016) to make our data and meta-data Findable, Accessible, Interoperable, and Reusable. All materials have been translated into American English. By sharing our experimental treatment materials, the content as well as the specific codebooks, and also by the documentation of the dataset within this article, we sought to maximize the transparency of our data and make it usable by other (human) researchers and machines (see Stall et al., 2019 for extended FAIRness guidelines).
4 Reuse Potential
The aim of this data paper was to present data of a cross-cultural framing project which has been conducted at different phases of the COVID-19 pandemic. A specific aim of this paper was to make other researchers aware of a newly developed scale measuring paradoxical attitudes towards prevention behavior in the COVID-19 pandemic, the so-called Paradox of Prevention Scale. Making our data publicly available allows that researchers from other disciplines might benefit from our previous work. Even though the aim of our project was to investigate situational influences such as the effects of different semantic frames of the pandemic situation on preventive attitudes and behavioral intentions, our dataset covers a bunch of different person-level variables which might be of interest for researchers focusing on the interplay of different traits and attitudes regarding preventive behaviors in the pandemic. What was aimed to be a research project in general and social psychology might also have some additional value for differential psychologists.
Thus, our current example corresponds to a core function of open science, namely the use of publicly available data for the generation of further knowledge (Potterbusch & Lotrecchiano, 2018). Furthermore, the data within the dataset may also be of interest to researchers in the broad field of health psychology who are interested in the prediction of preventive behavior intentions, or researchers of related disciplines who aim to connect some of the variables captured in our project to within-country processes.
Last but not least, the public availability of the dataset enables researchers to make use of the data for meta-analyses (Saunders & Inzlicht, 2021), which in this case could be either embedded in the context of framing research or focusing on the relationship between certain person-level variables such as distrust in science, conspiracy mentality or risk aversion with COVID-19 related variables such as vaccination willingness.