**4. Discussion**

This study is the first Japan-wide study that analyzed behavioral factors associating with SARS-CoV-2 infection in detail. It reconfirmed the effectiveness of mask wearing and hand washing in risk reduction. At least for infection prevention, the study did not show effectiveness of excessive behavior, such as frequent changing of clothes and extreme reduction of outings.

The most notable finding is that remote working and restrictions on outings did not always reduce the risk of COVID-19. Instead, these actions even appeared to increase the risk of the infection. This is contrary to previous analysis that showed effectiveness of lockdown [13,14]. There could be several reasons for this result. One possibility is that remote working and restrictions on outings gave a false sense of security and individuals began to neglect hand washing and mask wearing. A study of one Massachusetts city examined the genetic material of SARS-CoV-2 attached to surfaces to investigate the virus in the environment. PCR was positive in approximately 8% of the samples taken from environmental surfaces, and a particularly high level of virus was attached to the surfaces of trash cans [15]. Thus, even if outings are restricted, individuals cannot completely avoid their contact with environmental surfaces. Therefore, the infection risk could increase, especially if there is inadequate hand washing. Another possibility is that, even if individuals work remotely, they could be engaging in other high-risk behavior such as eating out with multiple individuals.

Our research also revealed that frequent changing of clothes and sanitizing belongings were significantly and positively associated with the infection risk. The result, however, does not mean that wearing and removing clothes increase the infection risk. It instead suggests that individuals who engage in such behavior might have limited knowledge of infection—they could be implementing ineffective preventive measures while neglecting the practice of highly effective ones. It is also possible that frequent changing of clothes

could be a sign of mental disorder triggered by anxiety of infection, which has been reported to increase the COVID-19 risk [16].

In general, a moderate level of exercise is necessary for reduction of health risk. Our study showed walking may have a preventive effect of infection. However, our research also indicated that a high infection risk was correlated with 4 or more days of exercise, 30 min−<sup>1</sup> h duration, and running was associated with a higher proportion of infection. The result suggests that individuals might have increased their contact with the virus by going out to exercise or by the use of a gym. Even so, a moderate level of exercise decreased the risk of severe illness from infection. It also has a preventive effect on other conditions (including diabetes, obesity, and hypertension) which increases the risk of severe illness from infection. Therefore, individuals should not unnecessarily avoid exercising.

The findings of this study strongly suggest that we may need a strategy other than legislation to change behaviors of populations. Epistemic communities, defined as "a network of professionals with recognized expertise and competence in a particular domain and an authoritative claim to policy-relevant knowledge in that domain or issue area [17]", may play an important role in nudging the public to take effective and efficient actions without legislation [18]. This epistemic community may also help citizens act according to expectations independently and voluntarily and may reduce the needs of aggressive interventions by the government. Although there is a study that suggests the efficacy of such a strategy in a specific field [19], further research is needed to elucidate the effective ways to achieve population health in disaster settings.

This study has several limitations. The first limitation is that the study relied only on participant responses to determine whether or not they "experienced COVID-19 infection", which was the primary outcome variable. As of 1 November 2020, there was a cumulative total of 101,368 people who tested positive by PCR test according to the Japanese Ministry of Health, Labour and Welfare. It translates to only 0.1% of the entire population of Japan testing positive. In our study, 0.48% of the total valid respondents said that they had been diagnosed as having COVID-19 infection, which is about three times more than that of the Japanese ministry's. Thus, it is highly likely that there was an upward bias in our study. For example, individuals with an infection experience could have more actively sought to participate in our study because of their increased interest in the significance and content of this online survey, causing an upward bias in participation of this type of subject. The RIETI questionnaire survey used self-reported information on their SARS-CoV-2 diagnosis at a medical facility to establish the presence or absence of SARS-CoV-2 infection experience. The study, therefore, does not include information on individuals who could have had SARS-CoV-2. These individuals might not have received the diagnosis because they were asymptomatic or only had mild infection and recovered without medical intervention. If the individuals with a diagnosis differed from asymptomatic or mild cases in their behavioral pattern or individual characteristics, such differences could have introduced a constant bias into the analysis results.

The second limitation is that the study was cross sectional. Therefore, a causal relationship cannot be determined between infection and behavior: individuals with a SARS-CoV-2 diagnosis could have been more careful in their daily lives. This possibility is supported by our result that the SARS-CoV-2 group had only a few individuals who had an exercise habit. This habit seems to increase the infection risk, as previously mentioned. Considering the likelihood of such bias, interpretation of estimates should be carefully examined (such as the average treatment effect), particularly the interpretation of the level of effect size. The RIETI questionnaire survey is a panel survey. Even if there were biases from active participation of the aforementioned type of individuals, the data might not show newly confirmed SARS-CoV-2 cases in second and later surveys conducted at our scale. Therefore, second and later surveys should also be analyzed in the same way.

Given these considerations of limitation, infection was still more strongly and negatively correlated with hand washing and mask wearing compared with other behaviors. This result is important in devising effective and sustainable infection control in the future.
