Invincibility threatens vaccination intentions during a pandemic

Some people feel they are invincible to the novel coronavirus SARS-CoV-2 (COVID-19). They believe that being infected with COVID-19 would not be a serious threat to their health. While these people may or may not be correct in their personal risk assessment, we find that such perceived invincibility may undermine community efforts to achieve herd immunity. Multi-level analysis of survey respondents across 51 countries finds that perceived invincibility from COVID-19 is negatively associated with believing there is a need to prevent the spread of COVID-19 in one’s community (n = 218,956) and one’s willingness to inoculate against the disease (n = 71,148). These effects are most pronounced among individuals from countries lower in cultural collectivism (e.g., USA, UK, Canada) and highlight the need to consider the interplay of individual and cultural factors in our efforts to understand, predict, and promote preventative health behavior during a pandemic.

Table B. Descriptive statistics and correlations for vaccine intention.  Table E. Multinomial logit results for the effects of perceived invincibility and cultural collectivism on prosocial concern. Table F. Ordinary least square results for the effects of perceived invincibility and cultural collectivism on vaccine intention. Table G. Multinomial logit results for the effects of perceived invincibility and cultural collectivism on vaccine intention.

Additional analysis 2. Age and gender.
Table H. Ordinary least square results for the effects of perceived invincibility and cultural collectivism on prosocial concern by age cohort (50 and under). Table I. Ordinary least square results for the effects of perceived invincibility and cultural collectivism on prosocial concern by age cohort (over 50). Table J. Ordinary least square results for the effects of perceived invincibility and cultural collectivism on vaccine intention by age cohort (50 and under). Table K. Ordinary least square results for the effects of perceived invincibility and cultural collectivism on vaccine intention by age cohort (over 50). Table L. Ordinary least square results for the effects of perceived invincibility and cultural collectivism on prosocial concern by sex (male). Table M. Ordinary least square results for the effects of perceived invincibility and cultural collectivism on prosocial concern by sex (female). Table N. Ordinary least square results for the effects of perceived invincibility and cultural collectivism on vaccine intention by sex (male). Table O. Ordinary least square results for the effects of perceived invincibility and cultural collectivism on vaccine intention by sex (female). S1 File References. References included in the S1 File.

Data 1. Individual-level data
Individual-level data was sourced from the Beliefs, Behaviors, and Norms Survey [1,2].
Collis et al. [2] provide details on the motivation behind the design of the survey and the sampling methodology used to ensure representative samples from each country. The survey dataset (deidentified) is available to academics and nonprofits upon completion of the Facebook Data Use Agreement at https://dataforgood.fb.com/docs/preventive-health-survey-request-fordata-access/. Fig A in

Data 2. Country-level data
The country-level dataset is publicly available from the OSF public repository at https://osf.io/qwn9f/?view_only=4349d3e12ade40b99b10053cf4f8fdf8. Country-level data on cultural collectivism was sourced from Fincher et al. [3] and Webster et al. [4]. The data includes country-level collectivism scores from four prior studies, including Hofstede [5], Suh et al. [6], Gelfand et al. [7], and Kashima and Kashima [8]. The Hofstede [5] data consists of individualism-collectivism scores from surveys conducted on over 100,000 worldwide IBM employees. The Hofstede [5] data provided by Webster [4] was compiled from Fincher et al. [3] and Geert Hofstede's public website [9]. The Suh et al. [6] data consists of individualismcollectivism scores for countries based on estimates from Hofstede's [7] and estimates from Triandis [10]. The Gelfand et al. [9] data consists of collectivism scores based on 17,370 responses to their Global Leadership and Organizational Behavior Effectiveness (GLOBE) Research Program. The Kashima and Kashima [8] data consists of a binary measure of collectivism across countries based on the spoken language of the country and the acceptability of omitting first (e.g., "I") and second (e.g., "you") pronouns.
The country-level data was matched with 51 countries for which there were corresponding individual-level data from the BBNS. For each of the 51 countries (see Table C), a composite collectivism score was created following Fincher et al. [3] and Webster et al. [4].
For the syntax and the corresponding output for this transformation see S1 Syntax and S1 Output, respectively. First, each of the four country-level scores were standardized using zscores. Second, the Hofstede [5] and the Suh et al. [6] scores were reverse scored (z * -1) so that higher scores would indicate higher collectivism to align with the other two scores. Bivariate correlations between these four scores were then assessed, and the mean of the four scores was taken to create a composite measure. Finally, the composite measure was linearly transformed to create a Cultural Collectivism Index score with a mean of about 50 and a SD of about 20 (see Table C) to aide in the interpretability of our findings relative to prior cross-cultural research on collectivism [3][4][5].

Additional analysis 1. Alternative model specifications
To test the robustness of the results presented in the main paper we conducted additional model specifications. Table D in S1 repeats the analysis from Table 1 in the main paper but uses Ordinary Least Squares with clustered standard errors at the country level. Since the variable prosocial concern takes on six possible values, we also conducted multinomial logit regressions treating prosocial concern as a categorical variable. The coefficients, reported in Table D in S1, are relative risk ratios (RRR). Estimates for each column are relative to selecting "Not important at all" for prosocial concern. Both Tables D and E in S1 show that the results are consistent with those in Table 1 in the main paper.  Similarly, Table F in S1 repeats the analysis from Table 2 in the main paper but uses Ordinary Least Squares with clustered standard errors at the country level. and G in S1 show that the results are consistent with those in Table 2 in the main paper.

Additional analysis 2. Age and gender
To test the robustness of the results across age cohorts, we repeated our analysis from