Traumatic life experiences and religiosity in eight countries

We present two datasets from a project about the relationship between traumatic life experiences and religiosity. These include data from 1,754 individuals in the United States (n = 322), Brazil (n = 205), China (n = 202), India (n = 205), Indonesia (n = 205), Russia (n = 205), Thailand (n = 205), and Turkey (n = 205). Surveys were consistent across samples: they include measures of traumatic life experiences, negative affective traits, existential security, life satisfaction, death anxiety, and various religious beliefs, attitudes, and behaviours. Psychometric evaluations of measures of supernatural belief and death anxiety were conducted.

1. Demographics. We collected data on gender, age, ethnicity, marital status, employment status/occupation, and socioeconomic status. 2. Traumatic Life Experiences Inventory. We included a 29-item measure of traumatic life experiences-the Traumatic Life Experiences Inventory-adapted from previous batteries 14,15 . Participants indicated whether or not each of the events has ever occurred to them.

Depression Anxiety Stress Scales. Lovibond and Lovibond's 21-item Depression Anxiety Stress Scales
(DASS-21) was used to measure negative affective traits 16 . The DASS-21 is widely used and has been validated in multiple studies across multiple cultural contexts. Psychometric evaluations of the DASS-21 have been positive, with the caveat that its subscales-depression, anxiety, and stress-may not be as distinct as originally theorised 17-20 . 4. Subjective Existential Security. Norris and Inglehart have observed that there is a paucity of subjective measures of existential security 4 . Therefore, we developed a simple 6-item measure that asks participants about how much they worry about financial matters, health, personal safety, personal relationships, and social status. 5. Religiosity. A variety of religiosity measures were included. The first two items pertained to participants' upbringing; this was followed by items asking about their current self-description as religious and spiritual, and their specific religious affiliation/s. We then included Jong and Halberstadt's 6-item Supernatural Belief Scale 1,21 . Next came a 5-item measure of religious behaviour, developed in our previous work. We also included Welton, Adkins, Ingle, and Dixon's 22  www.nature.com/scientificdata www.nature.com/scientificdata/ multidimensional locus of control measure 23 . Finally came Shariff and Norenzayan's 14-item Views of God scale 24 that asked about whether positive (e.g., forgiving, loving) and negative (e.g., vengeful, fearsome) traits applied to God. 6. Satisfaction with Life. As a contrast to the subjective existential security measure, we also included Deiner, Emmons, Larsen, and Griffin's widely used and previously validated Satisfaction with Life Scale (SWLS) 25 . Like the DASS-21 above, the SWLS is well-regarded and has enjoyed extensive cross-cultural use 26 . 7. Death anxiety. Two measures of death anxiety were included: Conte, Weiner, and Plutchik's Death Anxiety Questionnaire (DAQ) 27 and Jong, and Halberstadt's Existential Death Anxiety Scale (EDAS) 1,21 . The order of these two scales was counter-balanced across participants.
The survey, originally composed in English, was translated by Qualtrics's partner Language Connect into Brazilian Portuguese, Chinese (Simplified), Hindi, Tamil, Bahasa Indonesia, Russian, Thai, and Turkish. The translation process involved an initial translation, followed by a check by the same translator; the resulting translation was then independently checked again before being uploaded to Qualtrics. Qualtrics Panels then recruited participants in December 2015 and January 2016 from the USA, Brazil, China, India, Indonesia, Russia, Thailand, and Turkey. The target sample size was n = 200 for each country, except the USA (n = 300). To obtain these samples, Qualtrics Panels collected 1,816 responses. Of these, 59 were excluded as they did not fully provide informed consent; a further 3 were excluded due to missing explicit location (i.e., country) information. This resulted in 1,754 completed surveys.
Several checks were put in place to increase data reliability. A two-stage attention check was included after the Traumatic Life Experiences Inventory. It was disguised as a Positive and Negative Affect Schedule 28 (PANAS): instead of selecting options to indicate their current emotional state, participants were asked to only check the   www.nature.com/scientificdata www.nature.com/scientificdata/ "none of the above" option. If they failed to comply, they were informed that the PANAS was an attention check, asked to try again, and encouraged to pay closer attention to the rest of the survey. Of the 1,754, 935 participants passed the first time, and were automatically retained; 815 participants failed the first time but passed the second time, and were also retained; 4 participants failed twice, and were rejected.
At the end of the survey, participants were asked whether they responded randomly or provided responses that were false at any point during the survey. 1,668 indicated that they did not, 86 indicated that they did. For the purposes of our main analyses, both groups of participants were retained.
Aggregating across countries, approximately 55% of the participants were male and 45% were female: males were over-represented in India, Indonesia, and Turkey. The sample comprised approximately 28% Christians, 19% Muslims, 13% Buddhists, 9% Hindus, 25% nonreligious/atheist/agnostic, and 6% other (see Table 1). For comparison, Pew's estimate of the global distribution of religions in 2020 is 31.1% Christian, 24.9% Muslim, 6.6% Buddhist, 15.2% Hindu, 15.6% unaffiliated, and 6.6% other 29 . An oversampling of nonreligious individuals led to proportions of the dominant religious group in each country being lower than Pew's 2020 projections, especially   www.nature.com/scientificdata www.nature.com/scientificdata/ Turkey (98%). The exception to this trend was China, where the nonreligious were the dominant category in our sample (71%) as well as Pew's estimate (51.8%).

Data Records
All three datasets have been anonymised and are available in XLSX and CSV (non-proprietary) formats on the Open Science Framework (OSF) platform 13 together with files of the questionnaires. Abbreviation guides for variable names are also included in each XSLX file as well as in separate CSV files.

technical Validation
As psychometric information has previously been published about the EDAS and SBS-6 used in the USA, Brazil, and Russia under similar recruitment conditions, we compared our results to those previous results 1 . As shown in Table 2, Guttman split-half reliability (λ 4 ) for both SBS and EDAS were similar in our study as compared to the previous study. We also ran principal axis factoring on both the SBS and EDAS to extract a single factor. The proportion of variance explained by the one factor solution varied from country to country, but was always >50%. Furthermore, the proportion of variance was similar in our study as in the previous study, with the exception of the SBS in Russia: the first factor now explains less variance than before.
As this dataset also included other measures of death anxiety and religiosity, we were able to run tests of convergent validity. As the same measures appeared in the previous dataset in the USA, Brazil, and Russia, we were able to compare our findings with previous results in these countries. As before, SBS scores were correlated with other measures of religiosity: furthermore, both in this dataset and in the previous study, the correlation between SBS scores and religious volunteerism stood out as very weak in Russia (see Table 3). This is likely because in both studies, self-reported levels of religious volunteerism in Russia were low. EDAS was correlated with DAQ in every country, and effect sizes were similar between this study and previous research (see Table 4).
Taken together, our internal reliability analyses, factor analyses, and convergent validity analyses indicate that the data are sound. However, as there are ways to apply more stringent exclusion criteria to our data, we compared two subsets for each country: the first subset included only participants who passed the attention check the first time and indicated that they did not respond randomly or falsely ("Pass"), whereas the second subset included participants who failed the first attention check (but passed the second) and/or indicated that they responded randomly or falsely ("Fail"). Aggregating across all countries, this method of division produced approximately equally sized groups (48% Fail): however, subset sizes differed from country to country. Notably, very few participants from the USA failed even on these stringent criteria (6.2%): between-group comparisons for the USA are therefore unreliable. As mentioned above, only 86 participants across all countries indicated that they had responded randomly or falsely: 67% of those who failed here failed the attention check the first time, and all but one passed the second time.
As shown in Table 5 and Fig. 1, the central tendencies and distribution of values for the SBS are similar between the two subsets in most countries. T-tests found a significant difference only in India, though the magnitude of difference was similar in Russia. Differences were more marked for the EDAS: t-tests found significant differences in India, Thailand, and Turkey (Table 6). However, the density plots show that distributions were similar between subsets for all countries (Fig. 2). None of the significant tests were corrected for multiple comparisons. Finally, we inspected correlations between SBS and other religiosity measures, and between EDAS and DAQ. For the former analyses, we correlated SBS with the mean of the seven other religiosity items (cf. www.nature.com/scientificdata www.nature.com/scientificdata/ 95% confidence intervals overlapped in every case. The confidence intervals for the USA are large due to small sample size, as very few participants failed our data quality checks.
In summary, in the full samples for each country, the two key measures-EDAS and SBS-6-behaved as expected, consistent with previous research, demonstrating high levels of internal reliability and convergent validity with related indicators. Applying a more stringent exclusion criterion made minimal difference to the distribution of values of the SBS and EDAS, though some differences in central tendencies were detected. Correlations between these measures and related measures were similar between Pass/Fail subsets in all countries.

Code availability
An R script and wrapper for the data analyses presented here are available as.rmd and.rproj files on the Open Science Framework (OSF) platform together with the questionnaires and data files 13 . These can be used to generate exploratory factor analyses and estimates of split-half reliabilities for the focal scales (SBS and the EDAS) independently in each country. All data presented in tables and figures in the manuscript can be easily reproduced using the provided code. Those interested in following up on our analyses can use our base-script as a starting point.