COVID-19 Risk Perceptions After the End of the Public Health Emergency

In our experiment, we study how pessimistic or optimistic narratives about infection outlook affect people's COVID-19 risk perceptions.

In addition to the control group, we consider the following four information treatments.

1. Clinic in Tokyo [Pessimistic]

``In early July, the director of a clinic in Tokyo said that `the number of patients is on par with the peak of the eighth wave (the number of infections exceeded 20{,}000 cases per day in Tokyo), and an invisible collapse of the medical system has begun."

2. Hospital in Okinawa [Pessimistic]

``A hospital in Okinawa held an emergency press conference on July 11 to announce the risk of collapse of the medical system due to a resurgence of infections, urging citizens not to relax basic preventive measures.''

3. Expert communication [Pessimistic]

``On July 16, Shigeru Omi, the head of the government's COVID-19 countermeasures subcommittee, stated that `not only the number of new cases but also the number of hospitalized and severe cases are increasing. This could be due to an increase in contact opportunities following the reclassification of COVID-19 and to diminished immunity from natural infections and vaccines over time. We do not know the full extent of the increase in the number of infections, but we should anticipate that this upward trend will continue.''

4. Government Information [Optimistic]

``At a press conference held in early July, Shigeyuki Goto, the minister in charge of measures for novel coronavirus disease, stated that the number of patients had not increased sharply, and it was inappropriate to consider the current period as the ninth wave of infections in Japan.''

2023-08-14

2023-08-28

COVID-19 Risk Perceptions: The probability that a respondent will get infected with COVID-19 within one month.

We conducted our survey online from August 14 to August 28, 2023. We hired a survey company, Cross Marketing Inc., to recruit participants and collect data. The target participants are men and women aged 20 to 79. To ensure that the survey is nationally representative, the distribution of participants' age and gender is matched to the 2020 Population Census in Japan.

---Prior to Information Provision---

In the first stage of our experiment, we provide participants with a nominal anchor---the statistics on the number of confirmed cases and the infection rate in July and October 2022:

``In July 2022, the number of newly confirmed cases was 3,463,299 (accounting for 2.77\% of the total population), while in October 2022, the infection was relatively calm with 1,031,436 newly confirmed cases (0.83\% of the total population).''

We then elicit prior risk perceptions by asking participants to rate the probability of being infected with COVID-19 within the next month, using the following options: (1) less than 0.001\%, (2) 0.001\% to less than 0.01\%, (3) 0.01\% to less than 0.1\%, (4) 0.1\% to less than 1\%, (5) 1\% to less than 5\%, (6) 5\% to less than 10\%, (7) 10\% to less than 20\%, and (8) 20\% or higher.

To capture the individual attributes of all participants, we gather background information, namely age, gender, place of residence, education level, income class, health status, and the primary source of media (television, newspaper, internet, SNS, or others). We also ask about COVID-19-related experiences, including the number of vaccinations and past infections, the severity in case of being infected, and whether participants have any acquaintances who died from the virus. Additionally, we ask all participants to rate how often they took preventative measures such as hand washing and disinfecting, mask-wearing, ventilating, and avoiding the 3Cs (closed spaces, crowded places, close-contact settings) in the recent month. The answer to this question is given on a five-point Likert scale, ranging from ``frequently'' to ``never''.

---Information Treatment---

In the second stage of our experiment, we present all participants with the following overview of the COVID-19 situation in 2023:

``COVID-19 has been reclassified to Category V Infectious Disease on May 8, 2023. At the same time, the system has been changed from recording the number of daily new infections (``notifiable disease surveillance'') to recording the number of new infections in certain medical institutions (``sentinel surveillance'').

Before changing to the sentinel surveillance system, \textbf{265,404 new infections (0.21\% of the total population) were reported in April 2023. According to the weekly sentinel surveillance records after May 8, the number of new infections has been increasing after the reclassification of COVID-19 to Category V, and the number of new infections in the third week of July is approximately 4.2 times that in the second week of May.''

We then randomly divide respondents into five groups and provide additional COVID-related information that varies by group: one group receives no additional information (control group); one group receives a comment by a clinic in Tokyo warning about the potential collapse of the medical system; one group receives a statement from a hospital in Okinawa made at a press conference, also warning about the potential collapse of the medical system; one group receives a comment by a COVID-19 expert stating that the spread of infection will likely continue; and one group receives a statement by a government official indicating that Japan is currently not in the middle of the ninth infection wave. The exact wording of the information provided is given above in the "Intervention" section.

---Post Information Provision---

After the information provision, we elicit respondents' posterior risk perceptions by asking them again about their subjective probability of becoming infected with COVID-19 within the next one month. We use the same question format and response categories as in the pre-information stage. We also ask participants how often they plan to take infection preventive measures in the upcoming month to assess whether the information intervention affects the willingness to implement preventive measures.

---Two Alternative Setups---

For robustness, we conduct the same information-provision experiment described thus far in two slightly modified setups. The first alternative setup is identical to the baseline setup, except that we do not provide our respondents with nominal anchors before eliciting the prior subjective risk perceptions. This allows us to rule out the possibility that nominal anchors substantially shift beliefs and affect our treatment effects.

The second alternative setup is identical to the first alternative setup, except that we do not elicit prior subjective risk perceptions. This allows us to rule out experimenter-demand effects from eliciting priors---that is, respondents may infer the importance of risk perceptions and adjust their subsequent responses.

Randomization done in office by a computer.

Individual

No

15,000 individuals

15,000 individuals. We have 15 groups of 1,000 individuals: 5 groups (one control and four treatments) for each of the 3 setups.

1,000 individuals per treatment group.

Following the experimental design described in the preceding section, we have a total of 15 subgroups---5 groups for the baseline, 5 for the first alternative setup, and 5 for the second alternative setup. We select a total sample size of 15,000, with 1,000 participants in each subgroup. We perform a power analysis to compute the minimum detectable effect at a given level of power. Specifically, we applied a two-sample t-test, assuming equal and unknown standard deviations. Using a significance level of $\alpha$ = 0.05 and a power (1-$\beta$) = 0.8, we obtain the minimum detectable effect of 0.1254 standard deviations. This implies that with 1,000 respondents per arm, there is an 80\% power to identify an effect size as small as 0.1254 standard deviations between a pair of groups. When increasing the power to 0.9, the smallest effect size that can be detected is 0.1450 standard deviations. Because the minimum detectable effect of less than 0.2 standard deviations has been widely accepted in the literature, the power analysis confirms that our chosen sample size is appropriate.