Materials and methods

We conducted an online survey to understand how religion influences attitudes toward GPM issues. We conducted the survey with US adults between February 6, 2023, and April 5, 2023. The survey took 30–45 minutes to complete and was administered using Qualtrics. In addition to the AGPM, survey items included a battery of measures examining variables, such as religious affiliation, frequency of attendance at services, personal prayer practices, and views of the body and human origin, and demographics, such as education and political affiliation. The AGPM items were measured on a 7-point Likert scale with the following options: 7 = strongly agree, 6 = agree, 5 = somewhat agree, 4 = neither agree nor disagree, 3 = somewhat disagree, 2 = disagree, and 1 = strongly disagree. The AGPM.v23 survey instrument may be requested at https://bioethicsresearch.org/research-services/testing-services/.

We used two survey panel companies, Prolific and Cloud Research, for participant recruitment. Participants from Prolific were representative of the US population in terms of age, race, and gender, and in the Cloud Research sample, we over-sampled for specific religious affiliations. After removing participants who did not pass preliminary quality checks, our final sample included 2999 from Prolific and 1940 from Cloud Research, for a total of 4939 participants. This study was approved as exempt by the Washington University in St Louis Institutional Review Board (IRB#: 202201153).

Exploratory factor analysis (EFA) reduces a large set of items to a smaller number of factors, each composed of multiple items. We used a scree plot to identify the number of factors to extract and weighted least squares to estimate parameters for the EFA. Because the factors were likely correlated with each other, we used the promax oblique rotation [Reference Yong and Pearce10]. We evaluated the results using global fit indices. Following Brown [Reference Brown11], we examined χ ² and standardized root mean square residual (SRMR) as absolute fit indices, root mean square error of approximation (RMSEA) [Reference Steiger12] as a parsimony correction fit, and comparative fit indices (CFI) [Reference Bentler13] as a form of comparative fit. A good fit would include a non-significant χ ², SRMR ≤ 0.8, CFI ≥ 0.95, and RMSEA < .06. A CFI ≥ 0.90 and ≤ 0.95 along with RMSEA < 0.08 indicate acceptable fit [Reference Bentler13].

Often, EFA is followed by restricted or confirmatory factor analysis (CFA); however, CFA has many restrictive assumptions that may not be realistic [Reference Lorenzo-Seva and Ferrando15,Reference Steenkamp and Maydeu-Olivares16]. For example, CFA requires that all cross-loadings are exactly zero, which is an assumption that is unlikely to be met in applied datasets, particularly in a measure such as the AGPM, which examines two domains of variables at once (attitudes toward specific activities and cross-cutting benefits and concerns) [Reference Steenkamp and Maydeu-Olivares16]. As an alternative to CFA, EFA can be followed by unrestricted factor analysis (UFA) [Reference Steenkamp and Maydeu-Olivares16]. UFA can be used to specify the number of factors a priori and substantive indicators for each factor while allowing for small (aimed to be close to zero as much as possible) cross-loadings. A UFA allows the researcher to incorporate an approximate a priori knowledge about the pattern of factor loadings (based on the EFA results) but not have to resort to restrictive assumptions when estimating measurement model parameters.

To use the EFA followed by UFA strategy, we first divided our full sample into two groups at random (group 1, n = 2460 and group 2, n = 2479). We used descriptive statistics, t-tests, and chi-squared tests to compare demographic characteristics of participants across the two groups to ensure there were no notable differences that might impact the factor analyses (Table 1). Demographics compared were selected based on prior research that found age, gender, race, education, employment status, income, and political affiliation to have a significant association with one or more of the original AGPM factors [Reference DuBois, Mozersky, Antes, English, Parsons and Baldwin5]. After comparing the two groups to ensure they were roughly equivalent, we conducted EFA of the AGPM.v23 in order to identify the underlying factors.

Table 1. Comparison of two groups of participants who completed a survey on values and attitudes toward genomics and genomic healthcare (total n = 4939; survey year = 2023)

Following EFA with group 1, a UFA was conducted with group 2 using weighted least squares as the estimator and a Procrustes rotation [Reference McCrae, Zonderman, Costa, Bond and Paunonen17], which allows for a targeted factor rotation based on the pattern of factor loadings from the group 1 EFA. We compared the group 1 EFA with the rotated group 2 results using Tucker’s congruence coefficient to determine the similarity between the factors from the two groups [Reference Grice18].

Once the factor structure was confirmed, we compared the updated factors with the original factors [Reference DuBois, Mozersky, Antes, English, Parsons and Baldwin5]. Given our large sample size, we set alpha to 0.01, so inferential results with a p-value less than 0.01 were considered statistically significant.

We used R version 4.3.0 and RStudio version 2023.06.2 build 561 to conduct analyses. R packages used included foreign, psych, here, tidyverse, EFA.dimensions, flextable, sjPlot, and knitr.

Results

We found no significant differences between the two groups for race, employment, gender, ethnicity, household income, political party, education, or age (Table 1). We did not include the overall general support items in the factor analyses since they were not written to represent latent factors but computed a Cronbach’s alpha for the six general support items as a measure of reliability (Cronbach’s α = 0.82). The distribution of results for all participants for the items included in the factor analyses is shown in Fig. 1 and grouped by factor.

Figure 1. Distribution of 4939 participant responses to the attitudes toward genomics and precision medicine items in a survey on values and attitudes toward genomics and genomic healthcare (2023) grouped by factor.

The EFA of group 1 data (n = 2460) identified seven factors; the items and item factor loadings are shown in Table 2. All primary factor loadings had a magnitude of 0.37 or higher; while all items loaded most heavily on their primary factor, six of the items cross-loaded at ≥ 0.3 magnitude on a second factor. The overall R ² of 0.66 indicated the model explained 66% of the total variance in the data. The seven final factors and their variance explained (R ²) and Cronbach’s α were health benefits (R ² = 0.94, α = 0.89), sacredness of life concerns (R ² = 0.92, α = 0.82), privacy concerns (R ² = 0.93, α = 0.88), knowledge benefits (R ² = 0.9, α = 0.79), gene editing concerns (R ² = 0.91, α = 0.83), mRNA vaccine concerns (R ² = 0.95, α = 0.87), and social justice concerns (R ² = 0.86, α = 0.81), indicating that the model was a good fit for the observed data. Cronbach’s αmeasures internal validity or how strongly the items in the factor correlate with each other; values above 0.7 are considered acceptable. The factor R ² estimates the percentage of variance in the latent factors that the estimated factors would explain [Reference Grice18]. The model χ ² was 1334.59 with a p-value less than 0.001, indicating a significant difference between expected and observed values in the correlation matrix and suggesting some lack of fit. Although χ ² is widely reported in applied studies, it is rarely utilized as the sole index of overall fit because of its tendency to reject large-N solutions even when differences are negligible. On the other hand, both RMSEA (0.068; 90% CI: 0.067–0.071) and CFI (0.93) suggested acceptable fit, and SRMR (0.02) indicated good fit.

Table 2. Exploratory factor analysis loadings for responses from 2460 survey participants who completed a survey on values and attitudes toward genomics and genomic healthcare (2023)

Note: Cross-loadings (≤.3) on latent factors other than the primary factor are denoted by parentheses.

^a Item was reverse scored in the revised measure.

^b Item was reverse scored in the original measure.

Following EFA with group 1, we conducted a UFA with group 2 using Procrustes rotation [Reference Lorenzo-Seva and ten Berge19]. Tucker’s congruence coefficient was 0.99, indicating that the factors from the two groups were essentially equivalent. The model χ ² was 1545.84 with a p-value less than .001. The SRMR of 0.02 indicated a good fit. The root mean square error of approximation (RMSEA) was 0.073 (90% CI: 0.071–0.075), and the CFI was 0.93, both suggesting a moderate fit.

The original AGPM factor structure contained five factors: embryo concerns, privacy concerns, perceived benefits, nature concerns, and social justice concerns. The original factor for each item is shown in the last column in Table 2. With the exception of the health benefits factor, items loaded together in a way consistent with the prior factor structure. Eight items were reverse scored in the original AGPM (i.e., a response of strongly agree was originally coded as 7 but reverse coded as 1, and strongly disagree was originally coded as 1 and reversed coded to be 7 with the other categories reversed to match the new order), with six loading on a concerns factor in the original AGPM that now load on the new health benefits factor in the updated AGPM. The other two items originally reverse scored also show consistency between the original AGPM and the updated AGPM. The new mRNA items all loaded together on a single factor. Two of the mRNA items were reverse scored to be consistent with the other mRNA items where agreement with the statement suggests concern about the mRNA vaccine.

Limitations

The AGPM.v23 was validated within a US context in 2023. Responses to the AGPM.v23 may have been affected by societal events and cultural factors. The length of the AGPM completion (estimated at 15 minutes) could present a barrier for some studies. As a global measure of GPM attitudes, the AGPM will need to be updated as technologies evolve.

Discussion

This brief report examined updates to the pre-pandemic AGPM items and factor structure [Reference DuBois, Mozersky, Antes, English, Parsons and Baldwin5] after introducing new items about mRNA vaccines. In the final factor structure, all of the mRNA items loaded together onto one factor. We thought the mRNA items might load on other factors like social justice, health benefits, or privacy. However, given the politicized environment at the time of the survey, and the wide-ranging vaccine hesitancy reasons many of which were not related to genomic medicine (e.g., the perception of rushed development or the belief that the vaccines contain microchips to track people), it is not surprising the mRNA items grouped together into a single factor. It is possible that the mRNA items may load on other factors (e.g., health benefits) in similar ways to other GPM activities when COVID-19 vaccine development is further in the past.

GPM activities play an important role in promoting and protecting public health, as in the case of the COVID-19 vaccine, which was demonstrated effective at reducing the risk of severe disease and death [Reference Beladiya, Kumar and Vasava20]. However, vaccine uptake was lower among people with higher levels of religion, putting religious people at higher risk for more severe disease and death [Reference Imran, Leng and Iqbal3]. The AGPM.v23 instrument provides one option for building a better understanding role of religion in public attitudes toward GPM. In future papers and studies, we will present findings using the AGPM.v23 that help illuminate why religion and other demographic factors are associated with greater concerns about mRNA and other GPM activities, pointing to new opportunities for engagement with communities to achieve greater mutual understanding and identify potential paths for addressing concerns.