In addition to the relationship between reason and intuition in the decision-making process, there is evidence suggesting significant correlations between personality traits and religiosity or belief (Homayouni, 2011; Khoynezhad et al., 2012). Findings from Faria et al. (2013) and Shenhav and Greene (2011) indicate that a higher inclination toward intuitive thinking is associated with increased religious belief, while analytical thinking, conversely, tends to correlate with heightened religious or magical disbelief.

The discourse surrounding science as a belief, whether in the context of challenging or affirming other beliefs, is not a recent phenomenon, and certain occurrences highlighted by Machado et al. (2019) continue to be apparent in contemporary communication media. Where there exists belief, there is also a counterpart of disbelief. Some indications regarding the origins of scientific disbelief, as noted by Rynes et al. (2018), might arise from analytical errors or even questionable research practices involving the falsification of data or results. Additionally, disbelief may be associated with ideological polarization, prompting individuals to seek guidance from those who share similar beliefs (Marks et al., 2019).

Similar to belief, Rutjens et al. (2018) provides evidence of scientific disbelief, often associated with religious beliefs or even more conservative political views (Lewandowsky & Oberauer, 2016), as well as various conspiratorial ideas (Lewandowsky et al., 2013). In general, there are a considerable number of studies and evidence that analytical thinking is related to religious disbelief (Gervais & Norenzayan, 2012; Stagnaro et al., 2019), while religious belief is more related to intuition and, in turn, to automatic and heuristic processes.

Such findings do not imply that diverse beliefs cannot coexist within the same individual, nor do they suggest that these beliefs are entirely contradictory, so an individual can commonly hold multiple beliefs without internal conflicts. Occurrences like the one described were observed by Irwin et al. (2015) concerning a supposed doublethink. The researcher suggests that this type of doublethink (believing in science and believing in the paranormal) would occur in a compartmentalized manner, maintaining both forms of thinking simultaneously.

Another example in this context is related to the study by Machado and Zangari (2016), where Kardecist spiritists and Catholics used rational argumentation and scientific foundations to justify phenomena and align their beliefs. A similar situation was discussed by Krauss and Colombo (2020), suggesting that ambiguous evidence can be interpreted as supportive of a conclusion.

The importance of adaptation and the implementation of additional methodological procedures, which are much more complex than the literal operationalization of a translation, is emphasized by Borsa et al. (2012) and Sireci (2005). Their guidance revolves around avoiding the compromise of construct validity and promoting the preservation of the concept attributed to the instrument. Therefore, we chose to conduct a new translation procedure as one way to observe evidence that the content of the item has been retained and is suitable for the Brazilian context.

Hence, some procedures guided by Borsa et al. (2012) involve forward and back translation, synthesis of versions, evaluation by judges, evaluation by the target audience, pilot study, and the evaluation of the factorial structure of the instrument. These steps assist the researcher in attempting to retain the context and observe indications of validity. The initial translation was performed by an experienced professional with a Bachelor’s degree in Languages and Translation, along with free translations by other individuals with recognized proficiency in the English language. Thus, the proposals for the use of multiple versions with bilingual translators unfamiliar with the construct were met (Beaton et al., 2000; Cassep-Borges et al., 2010).

To test consistency in a back translation situation (Beaton et al., 2000; Borsa et al., 2012, Sireci, 2005), the consensus version was submitted to two university students-native English speakers residing in the United States with Portuguese as their second language. The context in this process was maintained, and only a few terms were altered without compromising the meaning. It is also of interest to compare the version used by Machado et al. (2019) with this study’s consensus version, as shown in Table 1.

When analyzing Table 1 in terms of convergence, it becomes evident that item 9 maintains an identical translation, while the wording of the other items varies slightly (e.g., from “rationally” to “in a rational manner”), without altering the context. Concerning item 2, the translation performed by the researchers omitted Sagan’s (2006) metaphorical context but still preserved the essential context in the statement, where science is seen as a hopeful way to solve a problem (the light at the end of the tunnel or the candle in the dark).

The panel of judges consisted of three individuals with bachelor’s degrees in Business Administration and one with a bachelor’s degree in Economics, all pursuing master’s degrees in Work Psychology. Additionally, there were three Ph.D. students in Psychology and one individual holding a Ph.D. in Psychology. The significance of the number of individuals participating in the evaluation process and the scores assigned to the quality of the content is emphasized by Alexandre and Coluci (2011) to ensure the relevance, clarity, and pertinence of the items. They suggest a concordance rate above 90% among judges as a criterion for assigning content validity. Experts and judges were invited to assign scores on a scale from 1 to 5, where 5 represented very high concordance regarding the criterion, and 1 represented very low concordance. From there, the Content Validity Index was calculated, resulting in 93%, considering the number of scores of 4 or 5 in relation to the total items and responses.

Data collection took place between 2019 and 2020 in a study titled “What do you think about science?” totaling 662 observations. Among the collected information are sociodemographic data, the administration of the BISS instrument, a version of the IGFP-5 (Big Five), and an instrument measuring rationality/intuition (CRT, Cognitive Reflection Test). Groups with more than 20 individuals sharing the same religion were adopted. Religions with fewer than 20 respondents were consolidated into the “other” category. The grouped data can be observed in Table 2.

Non-normality of the data could be a warning factor if the sample were small and exhibited heteroscedasticity (Chantarangsi et al., 2016; Ghasemi & Zahediasl, 2012; Hair et al., 2009; Miot, 2017; Pino, 2014). However, for this sample, procedures were adopted to proceed with the analysis, considering the size (n > 200 individuals) and indications of homoscedasticity. Another important observation to support this decision is that the skewness is above 1 and below 2 only in item 10, which, according to Miles and Shevlin (2001), could be considered for the continuation of the analyses, as this would not be indicative of significant bias problems. Also, the sample did not have missing values.

Analysis of the Box’s M test showed that the null hypothesis was not rejected, considering the significance level of p < 0.05, indicating equality of variance among the items.

For the execution of Exploratory Factor Analysis, a sample of 662 individuals was used, and the factors were extracted using the Factor program version 10.10.03. The analysis employed the Robust Diagonally Weighted Least Squares and Minimum Rank Factor Analysis methods via parallel analysis (Damásio, 2012) with Promax rotation and suppression of values below 0.30, without indicating the number of factors to extract. In the humanities and social sciences, oblique rotations are generally used, given the conceptual assumption of a relationship between factors. This choice is rooted in the understanding that it is impossible to isolate factors adequately and entirely (Damásio, 2012; Tabachnick & Fidel, 2014).

The factorability of the matrix is observed by the Kaiser-Meyer-Olkin test (KMO = 0.92), and the Bartlett’s Sphericity test (significant at χ² = 4252.6; 45; p > 0.000). Both indices signal good evidence in this study, given the adequacy both to the assumption that KMO indices above 0.7 indicate good factorability (Damásio, 2012; Laros, 2004) and the detected significance in the sphericity test (Tabachnick & Fidel, 2014). The extracted factors showed loadings above 0.5 (Hair et al., 2009) and communalities above 0.3 and below 0.8, assumptions observed in the analysis. The proportion of variance explained by the instrument was 0.64 with the indication of only one factor.

A high Cronbach’s Alpha (α = 0.96) and Guttman’s Lambda (λ² = 0.97) are observed, evidence that the instrument has excellent reliability indices (Marocos & Garcia-Marques, 2013; Souza et al., 2017). The same level of reliability index quality was observed for each individual question. Thus, the exclusion of items for the purpose of improving reliability levels becomes unnecessary.

To assess discriminant validity indices, the squared correlations of the items composing the scale were obtained for comparison with the average extracted variance. The mean of the squared correlations observed in the analysis was 0.40. Although the factor loadings are mostly high and above 0.50, their other indices (χ²/ Degrees of Freedom ratio, CFI, TLI, RMSEA, and AVE), presented in Table 3 as model 1, did not show good results from the perspective of confirmatory factor analysis assumptions.

As observed, the exclusion of items provides evidence of improvement in the indices presented in the analysis. Among the indicators, only the Root Mean Square Error of Approximation (RMSEA) would not fit within the expected parameters. However, it can be observed that, despite marginally located at its midpoint, the lower limit of the RMSEA is 0.08 and crosses the indicated value range. The discussion regarding the fit indices acceptance limits in various articles (Lai, 2019; Lai & Green, 2016; Xia & Yang, 2019) converges on the joint use of indices such as RMSEA, Tucker Lewis Index (TLI), Goodness of Fit Index (GFI), and Comparative Fit Index (CFI) to complement the decision-making process for model acceptance. As the other indices indicate good fit, it is possible to support the hypothesis of a good model fit, given the proximity of the RMSEA to the expected value.

In order to observe the relationships between belief in science and personality traits (Homayouni, 2011; Khoynezhad et al., 2012), analytical thinking (Gervais & Norenzayan, 2012; Sanchez et al., 2017), religiosity (Rutjens et al., 2018), biological sex, and level of education (Maraldi et al., 2016), a correlation analysis was conducted considering the variables observed in Table 4. The observations were listed to demonstrate the relationship of belief in science with some of the factors and constructs listed in the literature and present in this study.

The observed correlations range from weak (positively with openness and negatively with age, biological sex, agreeableness, and conscientiousness) to moderate (positively with the atheist/agnostic group and negatively with the evangelical group and religion). The variable “religion” comprises all religious manifestations present in the study. The variable considering the religious group was included in the table to measure its relationship with belief in science individually for some of the religions. The choice to individualize the observation of the relationship between some religious denominations was based on the number of respondents in the study, as well as the quantity and national representativeness of religions observed in the last census conducted in 2010, considering the four largest representations. Although some of the relationships appear almost negligible, they were included to assess some of the hypotheses suggested in the literature, which will be discussed later.

In this study, in addition to grouping religious denominations according to their conceptual similarity, an equalization of the number of individuals per religious group was also carried out to compare the means among Atheists, Evangelicals, Catholics, and Kardecists. The sample was divided to ensure that each group in this study had a minimum of 80 individuals. Four groups were selected: Atheists, Catholics, Evangelicals, and Kardecists. These groupings comprise a significant portion of the religious denomination observations in this study and represent the four largest representations observed in the last Instituto Brasileiro de Geografia e Estatística (IBGE, Brazilian Institute of Geography and Statistics) census in 2010 (Instituto Brasileiro de Geografia e Estatística, 2020).

Through an ANOVA [F(3.316) = 59.73; p < 0.05] considering the robust test of equality of means, the null hypothesis − H0 that the groups have equal means was rejected. In the Post Hoc comparison, it is observed that there is equality only in the comparison between the Spiritist and Catholic groups, as seen in Table 5. What can be seen is that the means are mathematically different and show statistically significant differences in the comparison between the Kardecist and Catholic groups (which have statistically equal means) compared to the Atheist/Agnostic or Evangelical groups.

Although the assumptions of normality and homogeneity of variances are violated, there is evidence that the impacts may be minimized, given that it is a sample with more than 200 individuals (Hair et al., 2009) and that showed good effect size indices (considerable partial eta squared of 0.36) and a power of 0.999 calculated via G*power 3.1.9.4.