In this study, we used various instruments to investigate mental health problems among the general population, especially migrant workers and those who work in nonregistered locations for 3 months or more [31]. The first instrument was a discrete choice experiment (DCE) questionnaire, which had a survey-based experiment design that solicited and quantified respondents’ utilities and preferences toward a set of attributes and levels of PHSMs. Following the DCE questionnaire, a Likert psychometric scale of pandemic fatigue was used to measure the respondents’ perceptions of the current PHSMs and levels of pandemic fatigue. Additionally, we conducted a subgroup analysis [32] to explore the heterogeneities based on demographic information and socioeconomic status, and a comparison was conducted of the preferences of respondents with relatively low pandemic fatigue levels and those with high fatigue levels.

The inclusion criteria of this study were age of at least 18 years and absence of cognitive impairments (self-report). Respondents were recruited and selected through an online social media advertising platform (Credamo Inc), which has over 3 million samples and covers all provincial administrative regions in China [33-36]. Credamo randomly distributed the survey in 31 provinces of China (excluding Hong Kong, Macao, and Taiwan). Specifically, we provided a quota size of 350 per sex group and 140 per age group (oversampled). According to this census, the population of China was approximately 1.411 billion, with 51.24% male individuals (723.34 million) and 48.76% female individuals (688.44 million) [37,38], leading to a sex ratio of approximately 105.07 male individuals for every 100 female individuals. We also referred to the National Bureau of Statistics of China [37] for age-specific quota design, with 63.35% in the age group of 15 to 59 years and 18.70% in the age group of 60 years or older. However, due to budget restrictions and practical issues during the pandemic, we considered a 1:1 ratio per sex group and 140 individuals per age group for the data collection platform. No personally identifiable information was collected as the survey was anonymous. Consent was obtained when the respondents actively pressed the button marked “I have been informed with sufficient information of the study and agree to participate in this study” after viewing the introductory section of the questionnaire where the background and objectives of the study were presented. Respondents could only access the questionnaire if they consented and reported that they were 18 years or older and did not have cognitive impairments. The translation of the original survey has been provided in Multimedia Appendix 1. Respondents received RMB 20 (US $2.76) as an incentive for participation.

An anonymous self-administered survey created using Lighthouse Studio (version 9.9.1; Sawtooth Software) was distributed from July 01, 2022, to September 30, 2022, and respondents from the entire country were considered for inclusion. The minimum sample size requirement of this study was calculated using the rule of thumb approach proposed previously [39]. Specifically, the equation for sample size calculation was as follows:

where t refers to the choice tasks in the survey, a refers to the number of alternatives, and c refers to the number of analysis cells. Specifically, the number of analysis cells c in this study refers to the largest number of levels for any of the attributes. As such, the minimum sample size in this study should be 125 respondents. Moreover, according to the standard parametric approach [40] of sample size calculation, the minimum sample size is 267 (Multimedia Appendix 2).

All the questions were close-ended, with tick boxes provided for responses and no question skipping allowed. No data were stored if the questionnaire website was closed before the completion of the survey.

The survey of this study had 4 main sections. Specifically, in the first section, we aimed to solicit respondents’ demographic information, including socioeconomic information (age, sex, education level, religion, marital status, occupation, income level, current residence, and registered permanent residence city).

In the second section of the survey, respondents’ vaccination history and medical history were collected. Respondents were asked how many doses of COVID-19 vaccination they have received, whether they have ever been diagnosed with or are currently experiencing psychological diseases (eg, depression, anxiety, obsessive-compulsive disorder, phobia, bipolar disorder, neurasthenia, schizophrenia, and personality disorder), and whether they have ever been infected with COVID-19. If respondents answered that they had been diagnosed with or are currently experiencing psychological diseases, they were required to answer what specific disease they encountered, the severity of the disease, and whether they have ever received or are currently receiving treatment. In addition, if respondents indicated that they had been infected with COVID-19, they were required to provide information about how they found out that they had been infected, their symptoms and complications, and their date of hospitalization and discharge.

The third section of the survey was the DCE. Respondents were presented with 9 sets of scenarios, and in each scenario, respondents faced 3 hypothetical responses, namely, “option A,” “option B,” and “neither.” Respondents were required to select the measure that they felt most satisfied with. The attributes and levels of different measures of the DCE were determined by a literature review [32,41,42] and consultation with local epidemiologists and experts, and according to design guidelines for DCE [43]. As a result, we determined 8 attributes in our study: (1) Risk of COVID-19 infection within 3 months; (2) Closure of social occasions; (3) Suspension of on-campus educational activities; (4) Suspension of public transportation; (5) Contact tracing, isolation, and quarantine; (6) Nucleic acid screening program; (7) Mandatory booster vaccination; and (8) Loss of income in 3 months. All the attributes and levels selected in the study have been summarized in Table 1.

A sample of a hypothetical choice task is shown in Figure 1. The levels in task choices of different versions were designed according to the principles of (1) orthogonality and (2) balance. The task choices in the DCE section were 8 random choices and 1 fixed choice. We used the fixed choice for further data quality control. Since the DCE questionnaire is relatively more complicated for respondents to understand and such cognitive burden imposed on respondents may lead to some bias in their selection, we added specific text and forced respondents to stay on the questionnaire page for at least 1 minute and carefully read the text to help them better understand what discrete choice tasks are and how to select the choices subsequently. Details are provided in Multimedia Appendix 1.

The description was as follows:

According to the full factorial design, there were 11,664 (6×2×2×3×3×3×3×6) policies and 11,664×11,663 task choices. To significantly reduce the complication of the design in order to ensure that respondents could complete the tasks, we applied a fractional factorial design based on balance (the frequencies of attribute levels are roughly equal across all tasks) and orthogonality (the frequencies of attribute pairs are roughly equal across the tasks) principles.

Psychological Likert scales were included in the fourth section of the survey, and pandemic fatigue was assessed on a 5-point Likert scale. The reliability (α=.885) and validity (Kaiser-Meyer-Olkin measure=0.737) of the Likert scale for pandemic fatigue were tested. The pandemic fatigue model designed by Lilleholt et al [29] was used to ask about demotivation toward COVID-19 PHSMs and the desire to know the development of the epidemic. We adjusted the pandemic fatigue model in our study. The adjusted pandemic fatigue model contained a series of questions on public attitudes or views on the strengthening of relevant measures for epidemic prevention and control at the current stage and a series of scales to measure the public’s pandemic fatigue. The first question asked about local confirmed cases in the respondent’s living area (town, county, and district), and it was followed by a question that asked about the current PHSMs in that area. The third question contained a scale on the respondent’s perceived risk of being infected with COVID-19. The fourth question contained a scale to measure the respondent’s perceptions of the current measures. Subsequently, the fifth question assessed the epidemic prevention fatigue situation under the current situation of strengthened epidemic prevention measures.

Descriptive analysis was performed to describe respondents’ demographic information and socioeconomic information, as well as information regarding migration, COVID-19 vaccination history, mental health disease history, exposure to COVID-19, infection with COVID-19, and experience of closed-off community management (entrance numbers were minimized, checking points were set up in communities, entry permits were limited, face mask wearing was required, health monitoring was enhanced, and only registered personnel and vehicles were allowed to pass through).

We used the multinomial logit (MNL) model to quantify respondents’ relative utilities among all respondents. The MNL model of this study followed the random utility maximization theory [44]. We calculated the odds ratio (OR) and 95% CI based on respondents’ relative utilities among levels and attributes to further measure respondents’ preferences. In addition, we applied the latent class model (LCM) to determine how respondents’ preferences differed according to group membership. We used the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) to determine the appropriate number of groups among the respondents. The Mann-Whitney test was applied for the analysis of quantitative variables. The scale data were analyzed using SPSS version 23 (IBM Corp), and the MNL model and LCM were adopted in Lighthouse Studio (version 9.9.1).

The LCM was robust in identifying unobserved heterogeneity within the data, and this method allowed for the identification of latent classes of individuals who exhibited similar preferences or characteristics. However, the LCM was not appropriate for investigating the association between pandemic fatigue and preference heterogeneities, as covariates, such as sex, age, and mental health status, were not controlled. Therefore, to further explore the preference heterogeneities among the respondents on controlling sex, age, and mental health status, we also conducted subgroup analyses based on respondents’ demographic information, including sex, age, monthly income, and mental health status. Moreover, we conducted a subgroup analysis based on respondents’ levels (high level or low level) of pandemic fatigue according to the results of the pandemic fatigue scale.

We collected consent from respondents through an online consent form in the survey. This study has been approved by the Institutional Review Board (IRB) at the City University of Hong Kong (reference number: 11-2022-65-E). We adhered to the International Society for Pharmacoeconomics and Outcome Research (ISPOR) reporting guidelines for designing and reporting the research questions, assessing attributes and levels, and performing statistical analysis for the DCE.

A total of 1183 respondents clicked on the link of our survey, and of these, 855 completed the survey. After the control process for data quality involving the exclusion of respondents who wrongly answered a trap question, 689 respondents were included in the final analysis. Among the 689 respondents, 341 (49.5%) were male and 348 (50.5%) were female. Additionally, 286 (41.5%) respondents were aged 35 years or younger. Most respondents (509/689, 73.9%) had a monthly income equal to or less than RMB 10,000 (US $1378.15), and 30.5% (210/689) of respondents were migrants (Table 2).

The attribute that had the most weighted importance in respondents’ decision-making was the risk of being infected with COVID-19 in 3 months (45.53%), followed by the loss of income due to COVID-19 measure (30.69%). Suspension of on-campus educational activities (1.29%) had the weakest weighted preference (Figure 2). Weaker preferences were observed when increasing the risk of infection with COVID-19. Compared with the full suspension of public transportation, respondents believed that suspension in only high-risk areas (areas with 10 or more local confirmed cases were designated as high-risk areas) would be associated with larger utility (OR 1.168, 95% CI 1.106-1.234; P=.002). Moreover, compulsory contact tracing was favored by respondents compared with voluntary contact tracing (OR 1.294, 95% CI 1.225-1.366; P<.001). In addition, respondents were willing to accept booster doses of COVID-19 vaccines, and their utility decreased along with loss of income within 3 months due to PHSMs (Table 3).

To better trace the heterogeneities of the preferences, subgroup analyses were conducted in terms of age and different monthly income levels (Figure 3). Compared with respondents having high monthly income, those having low monthly income were less sensitive to the risk of infection with COVID-19 within 3 months but more sensitive to the loss of income due to the measure within 3 months. Moreover, low-income respondents cared more about nucleic acid test screening for all and preferred the suspension of public transportation in only high-risk areas. Additionally, low-income respondents preferred not to be close to social and living places, which was in contrast with the findings for high-income respondents.

Similarly, respondents older than 35 years were more sensitive to the risk of COVID-19 infection and less sensitive to the loss of income within 3 months. Moreover, compared with older respondents, younger respondents preferred nucleic acid test screening for only high-risk units, workplaces, and vulnerable public, while older respondents preferred screening for all. Furthermore, younger respondents preferred not suspending on-campus educational activities and not closing social and living places.

Respondents diagnosed with mental health diseases did not favor contact tracing, isolation, and quarantine, as well as closure of social and living places compared with those without mental health diseases. The subgroup analysis for migrants and nonmigrants indicated that migrants had less acceptance of the mandatory booster vaccination and accepted the suspension of transportation in high-risk areas or normal operations.

A higher pandemic fatigue level was observed in female respondents, younger respondents, migrants, and relatively lower-income respondents (COVID-19 Pandemic Fatigue Scale [CPFS] correlation with age: r=–0.274, P<.001; correlation with monthly income: r=–0.25, P<.001) (Table 4). Based on the results of the CPFS, some preference heterogeneities were also found among respondents with a lower or higher level of pandemic fatigue (Figure 4). Respondents with a higher level of fatigue tended to be less sensitive to the risk of COVID-19 infection within 3 months and more sensitive to income loss within 3 months. Additionally, compared with respondents with a lower level of pandemic fatigue, those with a higher level of fatigue preferred the nonsuspension of social places and nonsuspension of on-campus educational activities. Mandatory booster COVID-19 vaccination was also not preferred by respondents with a higher level of pandemic fatigue, while universal COVID-19 booster vaccination was preferred by respondents with a lower level of pandemic fatigue.

According to the AIC and BIC of the LCM, 4 latent groups of respondents were determined, with the lowest BIC value of 9680.29 and AIC value of 9104.82. All the other model fitting values have been presented in Multimedia Appendix 3. The segmented sizes were 15.1%, 16.5%, 34.6%, and 33.8% for groups 1, 2, 3, and 4, respectively. As shown in Figure 2, respondents in groups 1 and 4 attached the most importance to the risk of COVID-19 infection within 3 months, while respondents in groups 2 and 3 attached importance to the loss of income within 3 months. Additionally, following the risk of COVID-19 infection and loss of income within 3 months, groups 2 and 4 considered contact tracing and nucleic acid test screening to be the third and fourth most important attributes, respectively. Group 3 believed that nucleic acid test screening and suspension of public transportation were very essential.