In this work, we designed a research flow (see Fig. 1) to address the abovementioned research questions. First, we constructed a questionnaire with various items to measure people’s privacy concerns, perceived social benefits, and acceptance of COVID-19 control measures, and their individualist orientation, political views, perceived social tightness, and perceived COVID-19 risk according to previous studies. Then, we conducted an online survey to recruit 4,260 participants from six study areas to collect the dataset. After that, we applied descriptive analyses to address the first research question. Finally, we used multilevel linear regression models and multilevel structure equation models (SEMs) to address the second research question.

Fig. 1Study areas and data collection

We used online surveys to collect data from September 8 to 16, 2022, from a target sample of 4,200 participants in six study areas: New Zealand, the U.K., the U.S., Hong Kong, Japan, and South Korea. These study areas have implemented some COVID-19 control measures that use IGD (e.g., South Korea has used the mobile phone locations of infected persons when conducting contact tracing, and Hong Kong has utilized electronic wristbands to ensure those required to self-quarantine adhere to the requirement). There are considerable cultural differences between the first three study areas (Western) and the latter three study areas (Eastern) (e.g., North American cultures tend to be individualistic while Asian societies tend to prioritize commitment to social norms; Bavel et al. [17]). They also have different experiences with the COVID-19 pandemic and levels of success in controlling it (e.g., New Zealand is considered highly successful, while the U.S. is considered far from satisfactory). It should be noted that the six study areas had experienced large-scale outbreaks of omicron at the time of the survey (i.e., September 8 to 16, 2022), and the control measures had shifted from non-pharmaceutical interventions to vaccination. These six study areas thus entail some diversity in the control measures implemented, cultural contexts, and experiences of the pandemic.

For each of the six study areas, data were collected through an online survey from 700 individuals who are at least 18 years old and have lived in the study area during the COVID-19 pandemic (i.e., the total target sample is 4,200 individuals). The surveys were implemented by an international survey research company (Cint) using quota sampling to ensure the representativeness of the samples. Solicitations to participate in the study were distributed via Cint’s international networks in each of the six study areas. As indicated by the Winton Centre for Risk and Evidence Communication (WCREC 2020) [19], which has completed an international online survey on people’s attitudes towards the risk of COVID-19, 700 participants per study area would achieve a margin of error of + /‒ 4% and yield reasonably reliable results. A total of 4,260 valid responses were finally obtained. The Survey and Behavioral Research Ethics (SBRE) Committee of the authors’ universities reviewed and approved the study protocol and survey questionnaire. Table 1 shows the sociodemographic profile of the participants in the six study areas. Specifically, the distribution of gender represents their respective populations well. In terms of age group, our sample has a higher percentage of young adults (25–45 years old) for the six study areas.

Table 1 Sociodemographic characteristics of survey participants, and comparison with those of the national/urban populations (n = 4260)Variables and factor analysis

Six versions of the survey questionnaire were prepared with the help of the collaborators of the project who are native (or close-to-native) academics and have considerable exposure to the social and cultural contexts in the respective study areas. All versions of the questionnaire asked the same questions. They only have minor variations to ensure that the expressions are socially and culturally appropriate to the study area in question (e.g., the name of the local currency). The survey questionnaire has three sections.

Privacy concerns, perceived social benefits, and acceptance

We collected participants’ data on their views of ten COVID-19 control measures that rely on individual-level georeferenced data (IGD) using the questionnaire. The ten measures can be classified into three broad types: contact tracing, self-quarantine monitoring, and location disclosure (i.e., disclosure of visited places, locations, or venues to the public) (see Additional file 1: Table S1 for details). For each of these ten COVID-19 control measures, the questionnaire collects data from participants concerning: (1) their privacy concerns for the measure, (2) their perceived social benefits of the measure (i.e., the level of social benefits the respondent thinks would be gained by providing the information requested by public health agencies) and (3) their acceptance of the measure. Each of these three response items is measured on a 7-point scale (from 1 to 7). For privacy concerns, “1” indicates “not concerned at all,” 4 indicates “neutral,” and 7 indicates “very concerned.” For perceived social benefits, “1” indicates “not beneficial at all,” while “7” indicates “very beneficial.” For acceptance, “1” indicates “not acceptable at all,” while “7” indicates “very acceptable.” The Cronbach’s alphas of the response items for privacy concerns, perceived social benefits, and acceptance are 0.918, 0.940, and 0.928 respectively, indicating that they have good internal consistency. We further derive a total score for each category of response items (by summing all scores of the response items for privacy concerns, perceived social benefits, and acceptance respectively). A higher total score indicates a higher level of privacy concerns, perceived social benefits, or acceptance for a participant. Additional file 1: Figure S1 shows the statistical distribution of people’s privacy concerns, perceived social benefits, and acceptance of COVID-19 control measures in the six study areas.

Individualist-collectivist orientation

We used 16 items to assess participants’ individualist-collectivist orientation (see Table 2). These 16 items were designed by Triandis and Gelfand [20] and can be used to assess participants’ personal individualist and collectivist orientations. In other words, these 16 items are not used to assess the participants’ perceived individualist and collectivist orientation of the area in which they live (thus, unlike House et al. [21]). Specifically, participants assessed their agreement on the 16 items using a 7-point scale (from 1 to 7), where a higher number stands for a higher level of agreement. Further, these 16 items cover four distinct patterns of individualist and collectivist orientation: vertical individualism, horizontal individualism, vertical collectivism, and horizontal collectivism. A vertical-individualist person tends to be concerned with improving his/her individual status among others via competition and achievement, whereas a horizontal-individualist person tends to view himself or herself as equal to others in status but emphasizes her/her uniqueness and distinctiveness from others in the group. A vertical-collectivist person prioritizes his/her in-group goals over personal goals, while a horizontal-collectivist person values sociability and interdependence within an egalitarian framework. We used exploratory factor analysis (EFA) to evaluate and validate the four-factor structure underlying the 16 items. We expected that people’s vertical and horizontal individualist-collectivist scores would be explained by four underlying factors at the individual level with item loadings in the expected direction. Table 2 presents the items and factor loadings. As expected, the EFA results demonstrated a four-factor solution: one factor with 3 items assessing vertical individualism (eigenvalue = 4.52), one factor with 3 items assessing horizontal individualism (eigenvalue = 1.96), one factor with 4 items assessing vertical collectivism (eigenvalue = 1.57), and one factor with 3 items assessing horizontal collectivism (eigenvalue = 1.13). Notably, a person’s individualist orientation and collectivist orientation are mirror concepts. In other words, the individualist orientation score is strongly negatively associated with the collectivist orientation score. We therefore applied participants’ vertical and horizontal individualism in the study. Higher vertical or horizontal individualism scores mean that the participant has a stronger vertical or horizontal individualist orientation. Additional file 1: Figures S2(a) and S2(b) present the distribution of participants’ vertical and horizontal individualism in the six study areas. They indicate that participants from Western study areas (i.e., the U.S., the U.K., and New Zealand) have lower vertical individualism scores and higher horizontal individualism scores than those from Eastern study areas (i.e., Hong Kong, Japan, and South Korea).

Table 2 Questions and factor loadings for people’s horizontal and vertical individualism and collectivism (n = 4260)Political views

Participants’ political views were recorded on a 7-point scale using the question “Please rate your political views on a 7-point scale,” where “1” indicates “very liberal”, “4” means “neutral”, and “7” indicates “very conservative.” For our analysis, we re-coded the raw political view score into three categories: “liberal” (i.e., a response score of 1–3), “neutral” (i.e., a response score of 4), and “conservative” (i.e., a response score of 5–7). Additional file 1: Figure S2 (c) presents the distribution of participants’ political views in the six study areas. Participants from Western study areas (i.e., the U.S., the U.K., and New Zealand) have more diverse political views than participants from Eastern study areas (i.e., Hong Kong, Japan, and South Korea). In other words, most participants from Eastern study areas reported that they are politically neutral. Meanwhile, the proportion of “liberal” participants is similar to “conservative” participants in the U.S., the U.K., and New Zealand. The proportion of liberal participants is higher than conservative participants in Hong Kong, which is the opposite of Japan and South Korea.

Perceived social tightness

In addition, this section of the questionnaire includes the 6 items shown in Table 3 from Gelfand et al. [22, 23] that measure the cultural tightness of the six study areas (cultural tightness is the strength of norms in a study area and the tolerance for people who violate norms). For these 6 items, participants also assessed their agreement on these items using a 7-point scale (from 1 to 7), where a higher number stands for a higher level of agreement. Then, we also applied EFA to evaluate and validate the one-factor structure underlying the 6 items. We expected that people’s perceived social tightness scores would be explained by one underlying factor at the individual level. Table 3 presents the items and factor loadings. As predicted, the EFA results demonstrated a clear one-factor solution (eigenvalue = 2.36). A higher score means that the participant perceives the social norms to be tighter and the tolerance for deviance in their society to be lower. Additional file 1: Figure S2 (d) shows the distribution of participants’ perceived social norms in the six study areas. It indicates that participants from Eastern study areas (i.e., Hong Kong, Japan, and South Korea) generally have higher perceived social tightness scores than those from Western study areas (i.e., the U.S., the U.K., and New Zealand).

Table 3 Questions and factor loadings for people’s perceived social tightness (n = 4260)Perceived COVID-19 risk

People perceived COVID-19 risk was assessed through the 3 items shown in Table 4: (1) “The COVID-19 pandemic is still not under control.”; (2) “I do not know the risk of the COVID-19 pandemic.”; (3) “The COVID-19 pandemic is catastrophic.” We solicited the level of participants’ agreement on the three items using a 7-point scale. Then, we also applied EFA to evaluate and validate the one-factor structure underlying the 3 items. We expected that people’s perceived COVID-19 risk would be explained by one factor at the individual level underlying the 3 items. The EFA result demonstrated a clear one-factor solution, and the item loadings were greater than 0.5, except for one reverse-coded item that had a loading of 0.149, which followed the expectation. A higher score on perceived COVID-19 risk means that the participant perceived the COVID-19 pandemic to be more severe. Additional file 1: Figure S2 (e) presents the distribution of participants’ perceived COVID-19 risk in the study areas. It indicates that participants from the U.S., Hong Kong, Japan, and South Korea have higher perceived COVID-19 risk scores than those from the U.K., and New Zealand.

Table 4 Questions and factor loadings for people’s perceived COVID-19 risk (n = 4260)Statistical Analysis

We first conducted descriptive analyses (i.e., the mean values and 95% confidence intervals) to chart how the trends of people’s privacy concerns, perceived social benefits, and acceptance of the ten COVID-19 measures varied due to the amount of IGD they use or disclosed to the general public. Specifically, we described how people’s privacy concerns, perceived social benefits, and acceptance of the COVID-19 measures varied across the three types of control measures (contact tracing, self-quarantine monitoring, and location disclosure).

Second, we examined the effects of people’s privacy concerns and perceived social benefits on their acceptance of the ten COVID-19 control measures. To do so, we fitted three multilevel linear regression models that include people’s privacy concerns (PC) and perceived social benefits (SB) as the key independent variables to predict their acceptance (AP). We first used people’s privacy concerns and perceived social benefits respectively as the predictor variable in Models 1a and 1b. In the full model (Model 1c), both people’s privacy concerns and perceived social benefits are included as the predictor variables. We fitted the multilevel models to simultaneously control for the within and between country/region effects on people’s acceptance. Therefore, all models include a set of individual-level (i.e., people’s age, gender, educational status, COVID-19 infection status, perceived COVID-19 risk) and country-level (i.e., case fatality rate and per capita income) covariates as controls. The three multilevel linear regression models were implemented using ‘lme4’ (v.1.1–31) and ‘lmerTest’ (v.3.1–3) packages in R statistical software (version 4.1). Equation (1) summarizes the models, with β denoting the coefficients, ε and π denoting the random effects of the individual-level and the country-level variables respectively:

$$\begin AP_} = \alpha _ + \beta _ PC + \beta _ SB + \beta _ \,Controls + \varepsilon _} + \pi _ \hfill \\ }\,i\,\left( }} \right)} = },} \ldots ,;\,j\,\left( }} \right)} = },} \ldots ,} \hfill \\ \end$$

(1)

To assess the impacts of people’s political views (liberal versus conservative), perceived social tightness (i.e., the strength of social norms), and individualist orientation (both vertical and horizontal individualism are considered) on their privacy concerns (PC), perceived social benefits (SB), and acceptance (AP), we fitted a series of multilevel linear regression models. In the first set of models (Models 2a, 3a and 4a), we used people’s political views (PV) as the predictor variable. In the next step (Models 2b, 3b and 4b), people’s political views (PV) and perceived social tightness (ST) were included. The third set of models (Models 2c, 3c and 4c) included people’s political views (PV), perceived social tightness (ST), vertical individualism (VI), and horizontal individualism (HI). We further included people’s privacy as a key independent variable in Model 3d to predict their perceived social benefits. Meanwhile, both people’s privacy concerns and perceived social benefits were included in Model 4d as the key predictor variables for people’s acceptance. All models also included the same individual-level and country-level covariates as controls. We also used ‘lme4’ (v.1.1–31) and ‘lmerTest’ (v.3.1–3) packages in R statistical software (version 4.1) to fit all multilevel linear regression models. Equations (2), (3), and (4) summarize the models, with β denoting the coefficients, ε and π denoting the random effects of the individual-level and the country-level covariates respectively:

$$_=_+_PV+_ST+_VI+_HI+_Controls+_+_$$

(2)

$$_=_+_PV+_ST+_VI+_HI+_PC+_Controls+_+_$$

(3)

$$\begin AP_ = \alpha_ + \beta_ PV + \beta_ ST + \beta_ VI + \beta_ HI + \beta_ PC + \beta_ SB + \beta_ Controls + \varepsilon_ + \pi_ \hfill \\ }i\left( }} \right) \, = , \, \ldots ,;j\left( }} \right) \, = , \, \ldots , \hfill \\ \end$$

(4)

Finally, we fitted three multilevel structure equation models (SEMs) to explore the direct, indirect, and mediating effects among the variables. The multilevel SEMs were designed to evaluate and validate the robustness of the abovementioned multilevel linear regression models. Therefore, the first model (Model 5a) tested the following hypothesis: (1) People’s political views, perceived social tightness, and individualism have direct effects on their privacy concerns, perceived social benefits, and acceptance of the COVID-19 control measures. The second model (Model 5b) focused on testing hypothesis (1) and the second hypothesis: (2) People’s privacy concerns and perceived social benefits play mediating roles in the effects of people’s political views, individualism, and perceived social tightness on their acceptance of COVID-19 control measures. The third model (Model 5c) further tested hypotheses (1) and (2) and the third hypothesis: (3) People’s perceived social benefits play a mediating role in the effects of people’s privacy concerns on their acceptance of COVID-19 control measures. The path diagrams of the estimated models are presented in Additional file 1: Figures S3-S5. The three SEMs were implemented with MPlus (version 8.3).