The perceived vulnerability to disease scale: Cross-cultural measurement invariance and associations with fear of COVID-19 across 16 countries

Using cross-sectional data from N = 4274 young adults across 16 countries during the COVID-19 pandemic, we examined the cross-cultural measurement invariance of the perceived vulnerability to disease (PVD) scale and tested the hypothesis that the association between PVD and fear of COVID-19 is stronger under high disease threat [that is, absence of COVID-19 vaccination, living in a country with

However, there remains a lack of cross-national investigation on how subjective PVD is associated with disease-avoiding tendencies under conditions where disease threat is low versus high. A core obstacle to study PVD across contexts is related to the PVD measure's low internal consistency (Ahorsu et al., 2022;Miller & Maner, 2012;Stangier et al., 2022) and inconsistent factorial structure across different cultural contexts.
Consequently, many studies utilized a total score of PVD (the sum of PI and GA scores) often skipping tests of its factor structure (e.g., De Pasquale et al., 2021;Mallett et al., 2021;Stevenson et al., 2021). The few studies that examined the factorial structure of the PVD scale were inconclusive, indicating that the two-factor structure consisting of PI and GA may not be consistently replicated, or that certain scale items may be susceptible to cultural bias.
For instance, Díaz et al. (2016) found that a two-factor model of the Spanish PVD with separate PI and GA factors (excluding two GA items) fitted the data better compared to a single-factor 15-item model (see also Moradi-motlagh et al., 2020 for an Iranian, andFerreira et al., 2022 for a Portuguese adaptation).
The present study addresses these shortcomings: first, it tests the dimensionality and cross-cultural measurement invariance of the PVD Scale , and second, it examines how the link between PVD and fear of COVID-19 may be moderated by individual-level and country-level conditions that represent objectively high (vs. low) disease threat. We propose that PVD will associate more strongly with fear of COVID-19, when disease threat is objectively high. For the present study, high disease threat was assumed when one may be more likely to catch a COVID-19 infection; and when getting infected would be more likely to result in a severe course. As getting a COVID-19 vaccine reduces the likelihood of infection (see meta-analysis by Zheng et al., 2022), and the threat of suffering from a life-threatening course is less salient in contexts with more developed health and living standards and lower COVID-19 mortality rates, disease threat was operationalized to be high when individuals had not received a COVID-19 vaccine, and when individuals lived in a country with high COVID-19 mortality rates and low Human Development Index (HDI).
Specifically, we hypothesize that the association between PVD and fear of COVID-19 would be stronger among individuals without a COVID-19 vaccine compared to those who had received at least one dose of a vaccine (Hypothesis 1), stronger in contexts with low (compared to high) HDI (Hypothesis 2), and stronger in contexts with high (vs. low) COVID-19 mortality rates (Hypothesis 3). 1 To keep inter-individual variation in disease threat constant, we conducted the present study with a low-risk population that is unlikely to experience a serious course of COVID-19, namely young adults aged between 18 and 30 years (Ho et al., 2020).

| Participants and procedure
Online self-report data were collected from young adults across 16 countries between September 2021 and March 2022 using convenience sampling. English measures were used in high English proficiency contexts (i.e., the Netherlands, India, and Hong Kong). In all other contexts, measures were adapted into the respondents' local language by using the committee approach for translation, where a small group of knowledgeable individuals translated the measures by discussing and adapting its contents to context-specific meanings (Beaton et al., 2000; see also van de Vijver & Leung, 2021). Ethical approval and informed consent were obtained.

| Vaccination status (individual-level)
Respondents were asked about their COVID-19 vaccination status, with three response options: 0 = no, 1 = yes, partially, and 2 = yes, fully. Based on the time-and context-dependency of "full vaccination" (e.g., different vaccines and number of doses across countries), responses were dichotomized as 0 (no vaccination at all) and 1 (at least one dose of vaccine).

| Human development index (HDI) (country-level)
Each country's most recent HDI score was retrieved (UNDP, 2022). The HDI reflects a composite index comprising affluence, educational level, and life expectancy information. Higher HDI reflects a more developed country.

| COVID-19 mortality rates (country-level)
Cumulative deaths per million were used to measure COVID-19 mortality (Our World in Data, 2022). To ensure comparability, we used the cumulative death count as of October 2021, which was the month where most countries began data collection.

| Factor structure of the PVD scale
To investigate the PVD's factor structure, an exploratory factor analysis (EFA) was performed with the pooled sample.
The number of factors was decided based on the scree plot (Cattell & Vogelmann, 1977) and Kaiser's criterion. The scree-plot indicated a four-factor solution, which, however, did not provide a good fit to the data, as the fourth factor comprised of only three items (items 1, 4, and 11) all of which showed cross-loadings with at least two other factors.
For details, see Figure S1 and Table S1. Using Kaiser's criterion, three factors were extracted with eigenvalues >1.00, explaining 51.5% of the variance. Due to observed cross-loadings on factors 1 and 3 (>0.30), an Oblimin rotation was performed. Extracting three and then two components, the two-factor model demonstrated the clearest structure. In the two-factor solution, the first component represented the GA subscale, while the second component represented the PI subscale. Both PVD subscales exhibited a positive correlation in the pooled sample (r = .27). For PVD item wording and loadings, refer to Table S2.
We then repeated the two-factor EFA using the Oblimin rotation, separately for each of the 16 countries, and found weak loadings on the GA factor for item 11 in Germany, Bulgaria, and Serbia (<0.30), and item 13 in Spain, Australia, Greece, India and Bulgaria (<0.40). Moreover, item 13 cross-loaded on the PI factor in Bulgaria, Australia, and India. Thus, for all following analyses, items 11 and 13 were excluded which we labelled as PVD-r (i.e., reduced PVD), and GA-r (reduced GA).

| Model comparisons
To investigate the scale's factor structure, we used the Mplus statistical package (Muthén & Muthén, 2021) for structural equation modeling with the maximum likelihood estimator. We compared four models to determine the best fit and interpretation. The first model (M1) was a single-factor CFA model with all items loading on one latent PVD-r factor. The second model (M2) was a two-factor CFA model with the PI and GA-r items loading on separate factors. However, cross-loadings between factors were found in the EFA, so we specified an Exploratory Structural  (M4), where all items loaded on a general factor and group factors representing the two subscales. Orthogonal bi-geomin rotation was applied to this model. The fit indices suggested that M4 showed the best fit (Table 2). Most items (except for items 2 and 8) significantly loaded on the general factor, with loadings ranging from 0.17 to 0.86; and both the PI items and the GA-r items significantly loaded on their respective factors. Thus, the ESEM bi-factor model was retained for further analyses.

| Measurement invariance
Measurement invariance was assessed at three levels: configural invariance (i.e., testing the invariance of the overall factor structure across countries), metric invariance (i.e., testing the invariance of factor loadings across countries), and scalar invariance (i.e., testing the invariance of item intercepts across countries; Furr, 2018). Goodness-of-fit indices at different invariance levels can be found in Table 3. The findings indicate that configural invariance was supported. The metric invariance model was marginally acceptable with a ΔCFI = −0.033 and a ΔRMSEA = 0.008. Scalar invariance was not established, therefore mean differences across groups cannot be computed (Table 3).

| Hypotheses testing
To test whether the association between PVD-r with fear of COVID-19 was moderated by vaccination status, HDI, and COVID-19 mortality rate, a mixed-level linear regression analysis using the ML estimator with the Jamovi 2.0 program (The Jamovi Project, 2022) was conducted. The correlations between the study variables are presented in Table S3. The countries' intercepts were added as random effects, and the scores of vaccination status, HDI, and COVID-19 mortality rates (grand-mean centered), PVD-r (group-mean centered), and their two-way interactions were included as predictors. Gender was added as a covariate, because gender effects are common for PVD (Díaz et al., 2016) and fear of COVID-19 (Nino et al., 2021;Sánchez-Teruel et al., 2022), and the gender distribution strongly varied across the 16 countries of the present research.
As presented in

| DISCUSSION
The present study makes two contributions to the literature. First, the bi-factor ESEM model revealed a single global factor of PVD that incorporates the sub-factors of PI and GA-r. This adds to previous studies that did not consistently find a clear structure of two separate factors (e.g., Díaz et al., 2016). Configural invariance was achieved in the cross-cultural measurement invariance analysis, but scalar invariance could not be established, preventing cross-country mean comparisons. Notably, not only in the present research, but also in previous studies items 11 and 13 have been problematic and eventually deleted (e.g., Díaz et al., 2016;Moradi-motlagh et al., 2020). These findings suggest that despite the global impact of COVID-19 and the implementation of risk reduction measures by governments worldwide, significant differences across countries in PVD exist. These differences may stem from genuine variations in PVD or cultural and contextual influences on item interpretation (e.g., "My hands do not feel dirty after touching money": different norms for behaviors may exist across different cultural and temporal contexts) and associated practices. Therefore, caution is necessary when interpreting cross-cultural research involving the concept of PVD.
Second, our study offers substantial insights into the association between PVD and fear of COVID-19 across various levels of disease threat. Surprisingly, the absence of a COVID-19 vaccine did not amplify the link between PVD-r and fear of COVID-19 as we expected. Instead, this association was stronger among individuals who had received at least one dose of the vaccine; plus, our data suggest that vaccinated individuals generally reported higher levels of fear of COVID-19 than those who chose not to get vaccinated. This suggests that individuals who Note: R 2 marginal = 0.19; Gender coded as 1 = male, 2 = female, 3 = other; Vaccine coded as 0 = no, 1 = yes; N = 4263; PVD-r = perceived vulnerability to disease score based on the reduced item set excluding item 11 and 13.

T A B L E 4
Results of the mixed-linear regression to predict fear of COVID-19. experienced significant fear of COVID-19 took proactive measures by getting vaccinated aligning with the protection motivation theory (Rogers & Prentice-Dunn, 1997). Therefore, while PVD reflects adaptive responses to disease threat cues , the association between PVD and feelings of worry appears to be less affected by individual variations.
Regarding country-level effects, we expected that living in regions with higher COVID-19 mortality rates and lower HDI would be associated with higher PVD-r. The latter assumption was not supported, suggesting that HDI and COVID-19 mortality rates may represent two qualitatively different contextual variables in terms of disease threat. As such, our results may account for a (pandemic) specificity (vs. generality) explanation of how each of those country-level variables are associated with PVD-r. Thus, higher levels of disease threat specific to the pandemic (i.e., high COVID-19 mortality rates) may generate a reactive mechanism and strengthen the association between PVD-r 9 of 18

F I G U R E 1
The moderation effect of COVID-19 vaccination on the association between PVD-r and Fear of COVID-19. PVD-r = perceived vulnerability to disease score based on the reduced item set excluding item 11 and 13. COVID-19 vaccine = yes refers to respondents who have received at least one dose of a COVID-19 vaccine.

F I G U R E 2
The moderation effect of COVID-19 mortality rates on the association between PVD-r and Fear of COVID-19. PVD-r = perceived vulnerability to disease score based on the reduced item set excluding item 11 and 13. COVID-19 mortality reflects the level of cumulative COVID-19 deaths per million by October 2021. and COVID-19-related fears (see Safra et al., 2021). When being exposed to contextual threats that are less pandemic specific, however, individuals might not have a clear interpretation of their perceived threat, as healthcare circumstances are worse in low compared to high HDI contexts, while reported mortality rates were lower (r = .18 in our research). This association aligns with previous findings involving over 150 countries (Mirahmadizadeh et al., 2022) and may be attributed to better epidemiological monitoring and more accurate reporting of COVID-19 deaths in high HDI contexts (Shahbazi & Khazaei, 2020), as it is unlikely that more deaths would occur in more developed countries.
Several limitations should be acknowledged, and future research directions can be proposed. Firstly, the interplay between various contextual variables, such as HDI, COVID-19 mortality rates, vaccination accessibility, and government policies, may interact differently and shape the perception of COVID-19 within specific countries and timeframes. Understanding these complex relationships warrants further investigation. Secondly, the limitations in sample variability, including skewed gender and vaccination status distributions, and the reliance on convenience samples of young adults, restrict the generalizability of our findings. Future studies should consider contextual differences more comprehensively, examine gender-specific associations, diversify the age range, and employ alternative sampling methods, such as purposive sampling. Lastly, the cross-sectional design hinders causal interpretations, and longitudinal designs would provide valuable insights in future explanatory studies.
In conclusion, our study contributes by examining the factorial structure of the PVD scale across nations and assessing its cross-national validity. Additionally, our findings highlight the role of PVD and country-level disease threat indicators in the appraisal of COVID-19 fear. Further research on this widely used measure can uncover globally relevant factors associated with health-related attitudes, behaviors, and outcomes.