Evaluating COVID-19 Information and Risk-Averse Behaviours: Insights from Conjoint and Clustering Analyses in the UK, Japan, and Taiwan

Introduction

The threat of health risks from new infectious diseases has been a challenge throughout human history. The COVID-19 pandemic has caused significant health, economic, and social damage worldwide. The issue of new infectious diseases in modern society is affected by the outcomes of the infectious diseases and people’s behaviour in response to the influence of culture and policy in the country and disseminated information.

During the pandemic, experts emphasised the importance of the message, ‘fear rightly and act rightly’ (Terada, 1948), and government officials and experts frequently disseminated information such as the number of new cases per day, the level of bed congestion, and vaccination updates. Citizens decide whether to take preventive actions against the spread of infection by considering such information and their circumstances (e.g., whether they live with older adults or alone, whether they have a job that allows them to work remotely, or whether they have been vaccinated). Although many experts are interested in the manner in which disseminated information and individuals’ circumstances influence citizens’ behavioural change, there are countless combinations of these variables, making it almost impossible to empirically investigate when considering all such factors. The current study attempts to tackle this issue by extracting primary factors and obtaining approximating solutions incorporating a conjoint experiment with an orthogonal design (see Appendix 1).

This approach can provide clearer visions of behavioural change related to new infectious diseases through empirical evidence regarding how expert information provision regarding a new infectious disease and citizens’ respective situations jointly affect their behaviours during the pandemic. The current study reports results from the United Kingdom (UK), Japan (JP), and Taiwan (TW) based on identically-designed survey data. Consequently, the observed effects can be explored regarding cultural differences and similarities.

This study aims to provide fundamental information that contributes to medical policy and public health development concerning new infectious diseases by reporting the empirical results of an international comparative survey of risk-averse behaviour determinants in the context of the COVID-19 pandemic. This study established two goals: (1) to examine how the context of COVID-19 information elicited risk-averse behaviours by empirically identifying the relative most important variables among such information, and (2) to extract subgroups that shared similar patterns of risk-averse behaviour during the pandemic by statistically examining interactions of respondents’ attributes and the information provided on risk-averse behaviours.

For the above two purposes, the current study conceptualised ‘Information’ that affects people’s decision-making by categorising it into (a) content composed of several factors, and (b) factors composed of several levels or degrees. Based on this conceptualisation, an experimental method capable of presenting information regarding the COVID-19 situation including the above factors and levels and assessing decision-making responses to risk-averse behaviours was applied. Therefore, a conjoint experiment (Louviere, 1988; Rao, 2014) was incorporated into the survey design to quantitatively measure the degree of risk-averse behaviour to multiple combinations of factors and levels. A conjoint experiment is a method that prepares several cards (information) that include a hypothetical scenario comprising multiple factors and categories (levels) of each factor to present to respondents and statistically demonstrates which combinations of factors and levels influence people’s decision-making.

Regarding the relationship between information on the COVID-19 situation and risk-averse behaviour, the factors refer to the elements of information disseminated by experts and governments, such as ‘number of daily new COVID-19 infections across the country’ or ‘restrictions on audience numbers in large-scale events’; surrounding information, such as ‘the urgency related to the number of available hospital beds’ or ‘whether people at one’s workplace are infected’; and individual information, such as ‘whether vaccinated or unvaccinated’. The levels refer to the degree or kinds of such factors. For instance, the levels of the factor of ‘number of daily new COVID-19 infections across the country’ can be set as 5,000 or approximately 25,000, and 100,000. The levels of the factor of ‘the urgency related to the number of available hospital beds’ can be set as ‘Plenty of hospital beds’, ‘Approximately 50% occupation of hospital beds’, ‘Beds are stretched, and you cannot be admitted to hospital immediately.’ By setting such experimental conditions, the combination of important factors and levels that are the determinants of risk-averse behaviours can be statistically identified to reveal the determinants of decision-making related to new infectious diseases (See also Appendix 1 and Appendix 2).

Previous studies have identified that engaging in risk-averse behaviour prevents the spread of the COVID-19 infection (e.g., Gao et al., 2020). Therefore, the current study quantitatively explores which factors people place the most importance on when engaging in risk-averse behaviour in response to different combinations of various information and situations related to COVID-19 and what characteristics are observed in subgroups with similar behavioural tendencies. The results can provide essential evidence for preparing for new pandemics in the future.

Few empirical studies deal with the relationship between information related to COVID-19 and behavioural changes such as taking risk-averse behaviour. For example, many papers published in the 2020s at the beginning of the spread of the COVID-19 infection mainly focus on qualitative considerations of risk awareness of new infectious diseases (Cori et al., 2020). Later, empirical studies gradually emerged, such as those utilising social big data from travel booking sites to examine the relationship between the published number of infections and individuals’ travel-related decisions (Ito et al., 2022; Im et al., 2021; Omori et al., 2024). Ito et al. (2022) have identified that the paucity of empirical studies on this issue might be due to a lack of suitable methods to observe decision-making for future behaviour. The current study fills this gap by incorporating a conjoint experiment method capable of assessing peoples’ decision-making regarding hypothetical situations.

People’s risk-averse behaviour regarding the COVID-19 pandemic has been investigated worldwide based on social surveys (Bergeot and Jusot, 2024; Dai et al., 2022; Rayani et al., 2021) and international comparative surveys (Dryhurst et al., 2020). Such studies have identified that people who perceived greater risks were more likely to have more frequent risk-averse behaviours during the COVID-19 pandemic (Bruine de Bruin and Bennett, 2020). Further, some studies have explored risk-averse behaviour in conjunction with psychological aspects including trust in government or medical professionals (Schneider et al., 2021).

Several social science studies regarding COVID-19 have incorporated conjoint experiments; however, most are choice experiments focusing on the specific condition of a region dealing with triage, who deserves immediate healthcare (Reeskens et al., 2021), or willingness to accept vaccination concerning COVID-19, (Wang et al., 2023; Igarashi et al., 2022), thus, their focus and set conditions are different from that of the current study.

Further, by clustering people with similar behavioural decision-making in response to the information, the current study pays unique attention to people’s scientific knowledge. How the general scientific knowledge or science literacy of the public influences their attitudinal change regarding provided information has been a critical issue in the fields of science and risk communication (Kato-Nitta et al., 2019; Simis et al., 2016). People use available information for attitudinal decision-making toward the application of novel science to their everyday lives (Finucane et al., 2002; Slovic et al., 1981). The traditional ‘deficit model’ framework proposes that higher ‘proper’ scientific knowledge among the public leads to a ‘proper’ response to the information; accordingly, people’s science literacy should have some influence on clustering people into subgroups when considering the relationships of COVID-19 information and risk-averse behaviour.

As Marivate and Combrink’s (2020) case study on using data to inform South Africa’s COVID-19 response claimed, enabling the public to engage with open data during a pandemic is significant. They also emphasised that prioritising data management practices by getting feedback from the public who use such information is a proactive approach that helps prepare for the future risks. Considering this, the current study empirically examining how data influences people’s public health behaviours in the different regions by focusing on the intersection of data science and cultural contexts should elucidate the manner in which different communities would interpret and respond to health information in varying ways. By doing so, it should contribute to the previous discussion on cultural differences in risk perceptions (Schaller and Murray, 2010; van Doren et al., 2024). It further extends the previous literature (Kato-Nitta et al., 2023) exploring how understanding and application of data can be shaped by regional, cultural, and societal contexts, and is thus relevant to not only medical experts but also data managers related to science, technology, and health.

This study aims to answer the following two research questions.

The current study explores the above two research questions by focusing on three regions—the UK, Japan, and Taiwan—selected for their similarities as high-income economies (above $13,205 GNI per capita) according to the World Bank (2023) and being insular states capable of undertaking relatively strict border measures. This approach enables us to examine the statistical results about whether regional or cultural similarities are observed, while broadly controlling for economic and geographical conditions.

Methods

Results

Information with the largest effect on risk-averse behaviour based on utility range (importance) in the UK, Japan, and Taiwan

Conjoint analyses based on experiments with 18 sets of conjoint cards presented to each respondent were conducted in the UK, Japan, and Taiwan. Before calculating the importance scores (utility range) of the eight types of factors, we confirmed the measurement reliability and validity by examining the correlation coefficients between the observed and estimated values of the utilities and obtained sufficient results (r > 0.95). We further confirmed that the utilities presented no inverse results; for example, if the level of a smaller number of infections exhibited a higher utility value than that of a larger number of infections. Some inversions were observed at the individual level, but none were observed in the overall regional-level average values for the respective regions. Thus, the measurement validity and reliability were confirmed. Appendix 5 provides details regarding the utility values.

Figure 1 presents the conjoint analysis results of the UK, Japan, and Taiwan.

Figure 1

The impact of information on risk-averse behaviour differs among the three regions. Nevertheless, the daily number of new COVID-19 infections across the country, availability of hospital beds and COVID-19 vaccination rates were of relatively high importance in each region. Respondents from the UK placed the highest importance on infection status in one’s surrounding living environment, that is, ‘someone infected near you’; however, respondents from Taiwan reported prominently high importance for the number of new infections per day despite the distinctively small number of infections established in the levels of conjoint cards among the three regions.

Overall, people in Japan and Taiwan were more likely to decide to engage in risk-averse behaviour related to COVID-19 information based on the number of new daily infections in the country as compared to those in the UK. On the contrary, people in the UK placed greater importance on direct contact via the occurrence of infections in their workplace or home.

Discussion and Conclusion

The current study extended previous research on the manner in which data informs public health responses, particularly in the context of the COVID-19 pandemic. It presents a novel direction in the field of research by integrating a comparative regional approach to understanding risk-averse behaviour in response to COVID-19 information across three distinct cultural and geographical contexts—the UK, Japan, and Taiwan. It contributes to prior discussion on the relationship between information and risk-averse behaviours by uniquely employing a conjoint experimental design to quantitatively measure decision-making, and identified key factors that influence individuals’ responses to COVID-19-related information. The incorporation of a finite mixture model to identify latent subgroups based on respondents’ characteristics, including science literacy, introduced an innovative approach of segmenting populations according to public health- behavioural tendencies. Hence, the current study provides important findings that are imperative in preparing for future pandemics.

Regarding RQ1, the utility ranges obtained from the conjoint analyses revealed that citizens in Japan and Taiwan were most sensitive to the number of daily new infections when making a risk-averse behaviour. This result is concordant with a previous Japanese study using social big data (Ito et al., 2022; Omori et al., 2024). By contrast, those in the UK were most sensitive to the presence of familiar infected persons. The importance of the number of new infections daily was outstandingly large in Taiwan, which experienced almost no major peaks in infections at the time of the survey (the values of the levels set in the numbers in Taiwan were 1/250 of that in the UK). This implies that the number of infections may not directly influence differences in behaviour. Furthermore, people in dissimilar cultural contexts or situations may have reacted differently to the same COVID-19 information. Such a phenomenon has also been seen in a previous study that explored the relationship between information on gene-edited food and people’s risk perceptions in the US, Japan, and Germany (Kato-Nitta et al., 2023). In the current study, this may be attributable to cultural differences, differences in habitual behaviours or social norms between East Asia and Europe or differences in how COVID-19 and related policies were reported in these countries. This underscores the importance of considering cultural and societal contexts when designing public health interventions, as people’s risk-averse behaviours may be shaped by their unique cultural and social norms.

Certainly, a myriad of combinations of information and conditions exists in individuals’ surroundings, which can evoke behavioural changes in them. Therefore, considering all the information or situations, including social distancing, wearing masks, and abstaining from travel, was nearly impossible. To address this issue, we used a conjoint experimental design. As the observed and estimated values were sufficiently correlated and no overall reversal of utility values across the three regions was observed, the validity of the current study’s approach was confirmed. However, this study has several limitations.

First, it is unclear whether the hypothetical situations used in the conjoint experiment reflect real-life situations. This is a weakness as the experiment involves recalling and responding to a hypothetical situation. Second, the current research employed the approximation with 7 factors (attributes) × 3 levels and 1 factor (attribute) × 2 levels; it is debatable whether the factors and levels established by the current study are optimal for statistically exploring the relationship between information and risk-averse behaviour under the COVID-19 situation. If any factors had a greater influence than the eight factors utilised in the current study, this approximation would not be optimal; however, since this research was conducted during the spread of COVID-19 when few previous studies had adopted the same survey methodology based on similar research questions and themes, this research had to be exploratory. There may have been other important variables that should have been considered as factors; therefore, we hope the current study contributes as a basis for further discussions. Third, this study has a limitation common to all surveys applying hypothetical experiments regarding the COVID-19 situation: we cannot rule out the possibility that respondents’ responses may have been influenced by the situation in their area when the survey was administered. As time passed and the situation of infections in the areas changed, the responses may have also changed (Qin et al., 2021). Nevertheless, by using identical experimental designs in the three regions, this study provides more information than a one-shot survey administered in a single region. To strengthen these results, similar surveys should be administered in other regions, and multiple surveys should be administered in the same region over time.

In conclusion, the impact of information on behavioural change pertaining to the COVID-19 situation across regions had both differences and similarities. The main difference was that while people in the UK placed the highest importance on the extent of direct contact for deciding behavioural change, people in Japan and Taiwan were most concerned with the number of daily new infections across the country when determining their actions.

Regarding RQ2, the study identified common latent subgroups across the three regions, specifically the risk-taking and prudent classes. These subgroups shared certain characteristics, such as age, vaccine hesitancy, science literacy, trust in government, and trust in medical institution measures against COVID-19, which have significant implications for public health policies. Better understanding these latent groups can help policymakers develop more targeted interventions, especially to engage those in the risk-taking class who are less likely to change their behaviour in response to information. The result that the risk-taking class contained higher proportions of vaccine hesitancy may contradict the intuition that vaccination would have fostered feelings of safety and led to going out more actively. Those who answered that they could not get vaccinated for medical reasons exhibited the highest mean value for the dependent variable of risk-averse behaviour across the regions (see Appendix 7), highlighting the distinctiveness of this subgroup’s behaviour. Nevertheless, as the responses of the risk-taking classes commonly exhibited positive and negative coefficients to the levels of ‘Vaccinated’ and ‘Unvaccinated’, respectively, the orientations of risk-averse behaviour frequency are the same as the other classes. Therefore, participants in the risk-taking classes may have simply been confident in their physical robustness.

In contrast, the prudent subgroups exhibited proportionately high levels of trust in government and medical institution measures against COVID-19, which was similar across the three regions. This could be an important implication for health policy experts and healthcare professionals because it indicates that restoring people’s trust significantly influences the decision-making of those in risk-taking classes, which could result in behavioural changes. This result is concordant with previous studies noting that trust is closely related to people’s risk perceptions (Lobb, 2005; Slovic et al., 1981).

Scientific knowledge was another similar issue across the UK, Japan, and Taiwan. The prudent class existed in all three regions and exhibited the highest frequency of risk-averse behaviour among the four sub-groups. They tended to score higher in science literacy as well as higher in health numeracy in the three regions. Being prudent and performing risk-averse behaviour (preventing infection spread) in the context of a new infectious disease outbreak implies correct or appropriate fear. Hence, increasing people’s general scientific knowledge and health literacy (numeracy) could be an effective strategy to promote positive health behaviors, particularly in the context of infectious disease outbreaks. Furthermore, this confirms the information deficit model in science communication studies regarding the assumption that an increase in ‘correct’ scientific knowledge among the public enhances positive attitudes toward new science and technology applications in everyday life, thereby contributing to further elucidating the deficit model’s boundary conditions (Ahteensuu, 2012; Kato-Nitta et al., 2019).

Moreover, the current study’s results further contribute to the discussion on the manner in which data-sharing systems can be adapted to meet the needs of diverse populations, such as those with varying levels of data literacy and science literacy or different health risks. By integrating cultural dimensions with data science approaches, the current study offers a more holistic understanding of how data can be used in public health, and thus makes a novel contribution to the ongoing discussions in data science research and public health.

Data Accessibility Statement

The datasets generated and/or analysed during the current study are not publicly available owing to the regulations of the Japanese privacy protection law; however, they are available from the corresponding author upon reasonable request.

Additional File

The additional file for this article can be found as follows:

Ethics and Consent

The surveys used in the current study were administered in accordance with the General Data Protection Regulation, EU Regulation 2016/679 and the Japanese Privacy Information Protection Law and were approved by the Institute of Statistical Mathematics Ethical Review Board beforehand (ISM21-005). Participation was voluntary, and participants could withdraw at any time. Informed consent was obtained from all survey company participants.

Competing Interests

The authors have no competing interests to declare.

Author contributions

NKN conceptualised and designed the current research, obtained funding, designed and organised the experimental questionnaire surveys, curated the data, developed the statistical analysis plan, conducted the statistical analyses, and wrote the paper. YI designed the experimental questionnaire surveys and curated the data. TM designed the experimental questionnaire surveys and curated the data. All authors have reviewed and approved the final manuscript.