Current meta-analysis

In the present meta-analysis, great emphasis is placed on a differentiated view and fine-grained analysis of the R/S variable. In order to obtain more robust results in R/S-mental health connection research, it is important to examine precisely whether and to what extent measurement instruments capture "religion" and/or "spirituality" and which of the multiple aspects of R/S play a dominant role in the (sub-)scales. Therefore, we conduct a qualitative analysis of all questionnaire items and classify the potential R/S measurement instruments on this basis.

For mental health, on the other hand, we selected a single indicator that (a) is widely used in empirical research, for which there is (b) a psychometrically valid measurement instrument, and that (c) is related to R/S from a theoretical point of view. In this systematic review, sense of coherence was chosen as an indicator of mental health because coherence—understood as a meaning- and trust-producing meta-resource—has been considered a mechanism capable of explaining the relationship between R/S and psychological health since the early dawn of R/S-health connection research.

Already 1987, sociologist of religion Ellen L. Idler introduced the coherence hypothesis into the field of study [12]. In the meantime, it is well theorized and researched that meaning-making is a key resource for psychological functioning [13] and that R/S can help people produce meaning privately or collectively [14]. Precisely because the coherence hypothesis emphasizes the meaning dimension of the R/S-health connection, it retains its explanatory power in view of the individualized appropriation of religious/spiritual offers of meaning on the “commodity market of transcendencies” [15] and in view of the increasing diversification of post-secular identities and worldviews [16].

Over time, Idler’s coherence hypothesis has evolved into the sense of coherence hypothesis (see [17] for a detailed treatment of the reception history). While Idler [12] understood coherence as a sense of meaning based on religiously reinforced optimism/fatalism, the sense of coherence hypothesis states that R/S and mental health are mediated by Antonovsky’s concept of sense of coherence (SOC), a basic life orientation characterized by a comprehensive and dynamic sense of trust [18]. Some twenty years ago, George and colleagues [19] asserted that the SOC hypothesis has received the most empirical support and claimed that SOC could explain 20–30% of the relationship between religious involvement and health. That SOC may be a particularly promising variable in the R/S-mental health connection has also been suggested in a recent meta-analysis: “The most important recommendation is to employ outcomes that are theoretically closer to R/S than distress and well-being, such as sense of coherence, […]” [4].

However, hitherto no systematic synthesis of this type of research has been published. Hence, the aim of our meta-analysis is to systematically examine the statistical associations between R/S and Antonovsky’s SOC. In this way, it makes an important contribution to testing the plausibility of the SOC hypothesis within R/S-mental health connection research. In connection with the SOC hypothesis, we are also interested in which form of transcendence reference, religion or spirituality, correlates more strongly with SOC. For this reason, in the following chapter we will heuristically define what we mean by "religion" and "spirituality" in the context of our meta-analysis and which mixed forms of R/S exist—a decision that will also determine how we will classify the R/S measurement instruments used in the included studies. Since we assume that both religion and spirituality are multidimensional constructs, our classification of the included R/S scales will not only decide whether the scale is an R scale, an S scale, a mixed scale, or no R/S scale according to our heuristic definition; we will also classify which dimension or aspect of R/S each R/S scale primarily measures. In this way, we aim to answer the question of which R/S aspects (faith, collective practice, God-relationship, etc.) are most strongly/weakly associated with SOC. However, since not only religion and spirituality can be understood as multidimensional constructs, but also the SOC scale generated by Antonovsky is conceived three-dimensionally, we also aim to answer the question with which SOC subscale (comprehensibility, manageablility, meaningfullness) the R/S scales correlate most clearly.

Another aim of the present meta-analysis is to identify moderator variables that have an effect on the statistical relationship between R/S and SOC. The selection of potential moderators is strongly determined by the type and breadth of data we can extract from the included studies. Nonetheless, we considered in advance which variables we would like to extract on theoretical grounds because we hypothesize that they might have a moderating effect on the R/S-SOC relationship. We formulated the following hypotheses regarding the moderator variables that we will consider in this meta-analysis:

• Age group: Since both religiosity and SOC increase with advancing age [20–22], it can be assumed that the R/S-SOC connection is stronger in older samples.
• Gender: Some studies have shown that the SOC of boys and men is higher than the SOC of girls and women [23,24]. Other studies, however, have found no gender differences in SOC strength among adolescents [25] and adults [26,27]. In the area of religious commitment, a gender gap has been noted for years: By and large, women appear to be more religious/spiritual [28,29]. We assume that these effects average out and no significant gender differences in the R/S-SOC connection come into play.
• Health status: According to Antonovsky’s theory of salutogenesis, people who are mentally or physically ill tend to be near the dis-ease pole of the health ease/dis-ease continuum [18]. Since, according to theory, the position on the continuum depends strongly on a person’s SOC strength, the SOC should generally be lower in clinical samples. On the other hand, R/S may increase in the context of illness, as religious/spiritual coping resources are often mobilized in difficult life situations [30]. Thus, we make the tentative hypothesis that the R/S-SOC connection is somewhat stronger in clinical samples.
• Cultural value orientation: Culture, especially cultural belonging and stability, is an important source for the development of a strong SOC [31,32]. Antonovsky [18] assumed that in countries or (sub-) cultures that have a shared canon of fixed yet flexible religious/spiritual beliefs, practices and values, the R/S-SOC connection is likely to be stronger. Therefore, drawing on the Inglehart-Welzel World Cultural Map (https://www.worldvaluessurvey.org), we hypothesize that the statistical association of R/S and SOC will be lower in countries with secular (post-religious) values than in countries committed to traditional (religious) values.
• Religious affiliation: It stands to reason that the R/S-SOC connection is more pronounced in religious and/or spiritual populations than in non-religious/non-spiritual or heterogeneously composed groups, since only for people of faith does R/S have relevance as a generalized resistance resource or SOC source.
• Time period: Garssen and colleagues [4] identified the publication year of the studies as a significant moderator in the R/S-health connection. In their meta-analysis, studies before 2000 and after 2009 show higher effect sizes. We hypothesize that further development and specification of R/S instruments will lead to stronger correlations being measured in more recent times.

Conceptualizing and measuring sense of coherence

Antonovsky [18] formally defined SOC as

a global orientation that expresses the extent to which one has a pervasive, enduring though dynamic feeling of confidence that (a) the stimuli deriving from one’s internal and external environments in the course of living are structured, predictable, and explicable, (b) the resources are available to one to meet the demands posed by these stimuli, and (c) these demands are challenges, worthy of investment and engagement.

From the definition follows that Antonovsky conceptualized SOC as a three-dimensional construct. He named the three SOC components (a) comprehensibility, (b) manageability, and (c) meaningfulness. They represent the cognitive, instrumental, and motivational or emotional aspects of SOC.

Based on a qualitative pilot study, Antonovsky [18] developed a 29-item-long Orientation to Life Questionnaire (SOC-29) and a short version of this scale (SOC-13) to measure SOC. Each item is rated on a 7-point Likert scale. Thus, the total SOC score ranges from 29 to 203 (SOC-29) and from 13 to 91 (SOC-13) respectively. A high SOC score is associated with the presence of resistance resources, adequate tension and stress management skills and the enduring feeling of confidence that things will work out as well as reasonably expected.

Antonovsky supposed that one’s SOC is developed in the first third of one’s life and that it remains relatively stable over time. Some studies confirm Antonovsky’s stability hypothesis [33], while some other studies suggest that SOC is much more dynamic and changeable than initially assumed [34]. Schnyder and colleagues [35] have tried to propose a middle ground between the positions: “Antonovsky’s SOC can probably be seen as a relatively stable (trait) measure, showing some degree of (state) variability when a person is faced with a drastic life event.”

Even though Antonovsky differentiated between three SOC components on a theoretical basis, he emphasized the holistic idea of his construct [18]. On an empirical basis, too, he underlined that SOC should be dealt with as a measure of one global factor [36]. Confirmatory factor analysis in various studies of SOC confirm that a single-factor model best fits the data [37]. By the same token, other studies suggest that the factorial structure seems to be multi- rather than unidimensional [38].

Antonovsky’s SOC scales show fairly good to excellent reliability. For the SOC-29 scale the Cronbach’s α coefficient of internal consistency has ranged from .82 to .95 in 26 studies analyzed by Antonovsky [36] and from .70 to .95 in 124 studies reviewed by Eriksson and Lindström [39]. The α values for SOC-13 range from .70 to .92. Cronbach’s α is rarely reported for the SOC sub-scales.

High correlations between SOC and some aspects of mental health have led some scholars to argue that SOC is not a discrete construct but rather an inverse measure for psychological distress, negative affectivity or neuroticism. Geyer [40], for example, holds that the “[…] very high negative correlations between SOC and depression/anxiety suggest that the instruments used may assess the same phenomenon, but with inverse signs”. On the other hand, these findings could also be interpreted as validation of Antonovsky’s SOC scale [41], because it is expected that SOC will highly correlate with mental health measures (convergent validity) and that it will predict mental health outcomes (predictive validity).

In our context, it is of particular interest whether the content of some SOC items could be partly regarded as overlapping the content of items measuring R/S. Coming from a methodological point of view, the conceptual and contentual discrimination between SOC and R/S is a premise for our meta-analysis [42]. Antonovsky attested religion an important role as a macrosocial generalized resistance resource [43], but he did not incorporate items dealing with transcendent aspects in his SOC questionnaire [44]. In accordance with this finding, Piedmont and colleagues [45] conclude that “SOC appears more oriented towards a secular-type of meaning-making” (p. 7). This interpretation is backed up by factor analyses showing that R/S and SOC items load on different factors [46].

Conceptualizing religion and spirituality

In the context of this meta-analysis, we conceptualize religion and spirituality as endpoints of a continuum (Fig 1). Hence, the controversial question of whether religion or spirituality is the broader construct no longer arises [47]. Both are equal poles of a dynamic field and therefore in their clearest expression, extreme manifestations of one and the same human mode of perception. In this way, common polarizations invoked to conceptually distinguish religion from spirituality can still be used to make an analytical differentiation. Religion in its "pure form" refers to transcendent attributions of meaning that are usually communicated with recourse to a handed-down system of moral teachings, rituals, practices and precepts, which is supported by an often hierarchically organized system of specialists and is institutionally anchored in a society. Idealtypically, the concept of spirituality describes rather subjective transcendent attributions of meaning, whose communication does not necessarily have to be linked to a fixed religious system of beliefs and practices, but rather is based on the individual-creative appropriation of religious concepts and on personal search and experience. At the same time, the continuum model makes it clear that the separation of religion and spirituality is not conceived as a simple dichotomy, but that there are fluid transitions and various mixed forms between the poles. If one wants to punctuate the continuum for analytical reasons, some ideal-typical hybrids can be delineated between religion (R) and spirituality (S): rather religion but also spirituality (RS), equally religion and spirituality (RS/SR or SR/RS), and rather spirituality but also religion (SR).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Continuum model of religion/spirituality.

https://doi.org/10.1371/journal.pone.0289203.g001

The phenomenal domain covered by the R/S continuum differs from the domain of the non-religious/non-spiritual by communications with reference to something or someone transcendent. The distinction transcendence/immanence is prominently used as “guiding difference” (Leitdifferenz) in a systems theory approach to mark the boundary between the secular and the religious [48]. Luhmann, a prominent representative of systems theory, understood “transcendence” as a point of view by which an intrinsically indeterminate immanent world (e.g., a lightning strike) is observed at the same time as something determinable and interpretable (e.g., Thor’s hammer blow). Or in Luhmann’s words: “a communication is always then religious when it observes immanence from the point of view of transcendence […]. Only when viewed from transcendence do events in this world acquire a religious meaning” ([49]; translated and cited by [50]).

The recourse to Luhmann helped us to clarify our concept of transcendence and thus the distinction between R/S and non-religion/non-spirituality. Whether a transcendent interpretation is rather religious or rather spiritual depends (a) on the reference system used and (b) the degree of definiteness. Religious systems of meaning usually have certain cognitive schemes in order to interpret the immanent in a transcendent way, e.g., as God’s work or the work of the devil in the case of Christian tradition. These schemes of interpretation are often institutionalized (e.g., in the form of theologies), reinforced by collective practices, and perpetuated by a faith community. In comparison, spirituality is characterized by a greater scope of individual interpretation and by greater indeterminacy: the immanent is indeed interpreted transcendently, but the reference system chosen to describe what has been experienced remains rather vague (e.g., a higher power or an invisible energy), may vary, and/or is not based on institutionalized traditions.

Search strategies

Literature search was designed to retrieve studies that report a statistical correlation between an original SOC questionnaire and an R/S measure. In this paper, two different literature search styles were used. For a start we made use of the classical “building blocks” technique [51]: The query was divided into facet A (salutogenesis), facet B (R/S), and facet C (quantitative design). According to this facet structure, Academic Search Ultimate, APA PsycINFO, APA PsycArticles, CINAHL, EMBASE, MEDLINE, PSYINDEX, PubMed, SocINDEX, and Web of Science were searched using the Boolean phrase depicted in S1 Table. In some cases, the search algorithm had to be adapted to the query boxes of the databases in order to obtain a reasonable number of results. The exact search queries for each database are compiled in S2 Table. The last search run was conducted by the first author on March 3, 2023.

In addition to database searches, we drew on the “berry picking” model of information retrieval proposed by Bates [52]. As Booth [51] points out, “berry picking” “is actually a meta-strategy, including six specific tactics”: footnote chasing, citation searching, journal run, area scanning, subject searches, and author searching. Most of the “berry picking” work was done using not only online databases but also online public access catalogs (OPACs), digital libraries, and Web search engines. The big advantage of “berry picking” is the opportunity to track down studies overlooked by systematic search strategies. Disadvantages include high time cost and the lack of replicability and transparency. The “berry picking” process was stopped on March 3, 2023.

Inclusion and exclusion criteria

Each study found was screened for eligibility. All of the following five criteria (I1-I5) have to be met in order for studies to be included in the review:

1. The study must be written in English or German.
2. The study must be published in a journal, as a book, as a book chapter or as a dissertation/habilitation thesis.
3. The study must use one of Antonovsky’s original questionnaires (SOC-29 or SOC-13) to measure SOC quantitively.
4. The study must use at least one instrument/item to measure R/S.
5. The study must report at least one correlation coefficient between a R/S (sub-)scale and a SOC (sub-) scale.

A study was excluded from the meta-analysis if any of the following criteria (E1-E5) applied:

1. The authors were unable to obtain the full text of the study.
2. The study results were reported elsewhere and that work met the above inclusion criteria.
3. The sample size is less than or equal to ten.
4. The assessment of the study with our critical appraisal tool showed that the study was of limited suitability for the questions of this meta-analysis (“poor fit”).
5. The instrument/item used to assess R/S did not qualify as an R/S measure according to our classification. (Less than 67% of the questionnaire items have no relation to R/S, using our heuristic definition of R/S formulated above to classify the items. The exact methodological procedure is described below.)

A rationale for the choice of the inclusion and exclusion criteria is provided in S3 Table. If certain information (e.g., the correlation coefficient or the wording of a self-developed R/S questionnaire) were missing, that would be necessary for the study not to be excluded from the meta-analysis, the first author contacted the study authors via e-mail, if possible.

Critical appraisal of studies

Each study that met the inclusion criteria and had not been excluded based on one of the first three exclusion criteria was critically appraised to determine whether it has a sufficient “fit” to the research questions of this meta-analysis. We did not use an already validated instrument to assess the “quality” of the studies for two reasons: (1) The Cochrane Collaboration was critical of the use of “quality” appraisal tools because the scales often mixed different concepts of “quality” and therefore the validity of a summary score was limited [53]. (2) Even if an assessed study was methodologically rigorous and met all standard “quality” criteria, such an critical appraisal would say little about the “quality” of the effect size, which is of particular relevance for our purposes. We include many studies with entirely different research contexts, study designs, and objectives that only "coincidentally" report a correlation coefficient between an R/S variable and SOC. Therefore, we did not need an appraisal tool that only asks for general non-specific “quality” criteria, but one that assesses the “fit” to the questions of our meta-analysis. For example, the question of whether the included study uses a valid R/S measurement instrument plays a special role for our assessment

To decide whether a study “fits” the context of our investigation, we developed an appraisal tool with three domains: (a) study population, (b) R/S measure, and (c) data analysis (S1 Tool). The tool incorporates criteria that have already been used [54,55], but also contains criteria that have been developed specifically for our purpose.

The first domain (study population) includes five criteria (e.g., sample size) with a score range from 0 to 10 points, the second domain (R/S measure) includes four criteria (e.g., Cronbach’s α) with a score range from 0 to 8, and the third domain (data analysis) includes three criteria (e.g., precise report of effect size) with a score ranging from 0 to 6. The overall result is recorded in a tripartite form (e.g., 6/5/2). Each study was assessed twice by the first author (intrarater reliability) and then reviewed by another author of the research team (interrater reliability). If there was a disagreement between the raters, it was discussed and a value was agreed upon.

The reviewed studies are classified into three categories according to the following criteria:

• poor fit: The second value (R/S measure) is less than 2 and/or the overall score is less than or equal to 7.
• moderate fit: The second value (R/S measure) is between 2 and 5 and the overall score is between 8 and 15.
• good fit: The second value (R/S measure) is above 5 and the total score is above 16.

Studies with a “poor fit” were excluded from the meta-analysis.

Classification of measuring instruments according to the continuum model of religion and spirituality

The designation of a measuring instrument can be deceptive. Just because terms such as “spiritual” or “religious” are mentioned in the title does not necessarily mean that the questionnaire items also measure religion and/or spirituality. For instance, Koenig [56] has convincingly argued that Peterman and colleagues’ Spiritual Well-Being scale (FACIT-Sp) is a measure of positive feelings rather than a measure of spirituality in the strict sense. The reverse case also occurs: Scales that do not have terms such as “religious” or “spiritual” in the title may nevertheless contain items that measure R/S. Therefore, an item-specific content analysis of the questionnaire items is necessary to ensure that the measures that may be included in our meta-analysis actually capture the R/S variable.

We classified each item of all potential R/S measures used in the included studies according to our heuristic continuum model of R/S (Fig 1). For this purpose, the continuum is divided into the five categories introduced above: religion (R), rather religion but also spirituality (RS), equally religion and spirituality (RS/SR or SR/RS), rather spirituality but also religion (SR), and spirituality (S). Each item is assigned one of these five labels or, alternatively, it is classified with the symbol X, which means that this item has no reference to someone or something transcendent. To determine whether a (sub-) scale as a whole should be considered an R-measure, RS-measure, RS/SR-measure, SR/RS-measure, SR-measure, or S-measure, the percentages of S-items and R-items within the total questionnaire items are calculated. If, in the course of content analysis, a (sub-)scale is found not to include enough R/S-related items (less than 67%) to be classified as an R/S measure in our sense, the study (insofar as it does not include at least one other valid R/S instrument) was excluded from the meta-analysis.

In order to make the classification process intersubjectively more comprehensible, the first author developed a coding aid with “trigger words” (S4 Table). Each potential R/S measure was classified twice by the first author and then reviewed by another author. If different opinions prevailed in the classification, the research team discussed until a consensus was reached. The methodological intricacies of the classification process are explained in more detail in S1 Text.

Classification of measuring instruments according to their religious/spiritual aspects

Previous meta-analyses have made distinctions between different dimensions of R/S and categorized the included R/S measures accordingly. The used coding schemes ranged from three [6] to 21 categories [7]. Garssen and colleagues [4] assigned the R/S scales to eight categories. In the categorization process, their “aim was to stay as close as possible to the names and categories used in the reviewed articles” (p. 4). Zwingmann, Klein and Büssing [57] undertook a more theory-driven classification of R/S instruments according to their primary measurement intention. In this meta-analysis we made use of a qualitative content analysis [58] of the items of all R/S measures used in the reviewed studies as the basis for classification. In accordance with the logic of qualitative coding, no categories were predetermined in advance; instead, the category system emerged stepwise from the analysis of the questionnaire items themselves. In this way, over time, certain categories are consolidated, revised and made more precise, and finally provided with anchor examples and descriptions. Each item was coded twice by the first and then reviewed by the second author. In case of a disagreement, the item in question was discussed by the author team.

To calculate the predominance of certain R/S aspects within a measurement instrument, a formula was developed that determines how often an R/S aspect must occur relative to the number of items in order to be considered dominant: If the scale included 1–4 items, 50% of the items had to address the same R/S aspect for that R/S aspect to be classified as dominant. For longer measuring instruments, the percentage was reduced accordingly (33% of 5–13 items; 25% of 14–25 items; 20% of 26 or more items). The percentages chosen are not absolute, of course, but have emerged as sensible benchmarks after a lengthy trial and error process. If more than one-third (33%) of the items addressed negative aspects of R/S (and were not reverse coded), the (sub-) scale was classified as a negative measurement instrument (symbolized with the abbreviation N).

Coding matrix of the included studies

Each included study was tabulated according to the following coding system: (a) serial number (numbering is according to alphabetical sorting by author surname), (b) author/s, (c) year of publication, (d) time period (1988–1999; 2000–2011; 2012–2023), (e) country (usually the nationality of the sample, but if the researchers are from another country, this is added in brackets after it), (f) cultural map (assignment of the country according to the Inglehart-Welzel World Cultural Map [59]: African-Islamic; Catholic Europe; Confucian; English-Speaking; Orthodox Europe; Protestant Europe; Latin America; West & South Asia), (g) UN geoscheme (17 subregional groups), (h) study type (article; book; chapter; dissertation), (i) study design (cross-sectional; longitudinal), (j) survey method (one or several of these codes: administered; in-person; mail; online; telephone), (k) response rate (usually in % and/or in numbers), (l) sample description, (m) sample type (mentally ill; migrants; population; refugees; religious officials; somatic patients; specific group; specific profession; students), (n) major life event (yes; no), (o) health status (clinical; non-clinical), (p) type of disease, (q) sample size, (r) size group (small [10–99]; middle [100–299]; large [>299]), (s) number of women, (t) number of men, (u) number of diverse people, (v) gender group (women; mostly women [at least two thirds of the sample are women]; mixed; mostly men [at least two thirds of the sample are men]; men), (w) average age of the sample, (x) standard deviation age, (y) age range, (z) age group (classification is based on the average age of the sample: <18; 18–30; 31–60; >60), (aa) religious affiliation (If possible, the percentage of the sample belonging to a particular worldview is indicated. Options include not only the “major” religions, but also indicators such as “atheist”, “non-religious” or “spiritual”), (ab) religion group (If 100% of the sample belongs to a belief system, the sample is coded as, for example, “Christian” or “Muslim”. If two thirds, i.e., over 66.6%, belong to a belief system, the sample is coded as, for example, “mostly Christian” or “mostly Muslim”. In the other cases, the code “heterogenous” is used), (ac) R/S measure (The name of the R/S measuring instrument is listed here. If more than one instrument was used in a study, the study is listed in the table a corresponding number of times. However, each study is only included once in every meta-analytic procedure), (ad) classification of the R/S measure (The code corresponds to the classification of the respective R/S instrument using our R/S continuum model, i.e., the codes are: R; RS; RS/SR; SR/RS; SR; S), (ae) dominant R/S aspects (Listed here are which R/S aspects dominate in each R/S measure according to our content analysis), (af) SOC measure (SOC-13; SOC-29), (ag) SOC item mean score (range, 1 to 7), (ah) SOC strength (weak: SOC item mean score < 4.1; moderate: SOC item mean score between 4.1 and 5.3; strong: SOC item mean score > 5.3), (ai) total SOC mean score (SOC-13 range, 13 to 91; SOC-29 range, 29 to 203), (aj) SOC range, (ak) SD SOC, (al) R/S mean score, (am) R/S range, (an) SD R/S score, (ao) correlation SOC-C (comprehensibility) and R/S measure, (ap) correlation SOC-MA (manageability) and R/S measure, (aq) correlation SOC-ME (meaningfulness) and R/S measure, (ar) correlation of SOC total mean score and R/S measure, (as) correlation method (Pearson; Spearman), (at) level of significance (ns; p < .05; p < .01; p < .005; p < .001; p < .0005; or exact p value), (au) score of the 1^st domain of the appraisal tool (range, 0 to 10), (av) score of the 2^nd domain of the appraisal tool (range, 0 to 8), (aw) score of the 3^rd domain of the appraisal tool (range, 0 to 6), (ax) total score of the appraisal tool (range, 0 to 24), (ay) fit assessment (poor; moderate; good), (az) personal communication (yes; no).

Extraction of data was performed twice for each study by the first author and have been checked by authors of the team.

Meta-analytical procedures

Basis for all statistical calculations is a Microsoft® Excel-sheet, on which the extracted data of all included studies are listed. Bivariate correlation coefficients (Pearson’s product-moment correlations or Spearman’s rank correlations) between the R/S (sub-) scales and total SOC are set as effect sizes (ESs) and listed in column AR of the Excel table. If correlations are also reported between R/S measures and one or more of the SOC sub-components (comprehensibility, manageability, meaningfulness), these ESs are also listed. For each subgroup analysis, an extra Excel table (reduced to necessary information) is created from the total Excel sheet to ensure that each study sample (even if reporting multiple effect sizes) is included only once in each analysis. This explains the difference between the number of studies and the number of ESs. A subgroup must consist of at least five different studies and therefore also of at least five ESs. If a study included multiple ESs relevant to a subgroup analysis (e.g., correlations between total SOC and two different S measures), these ESs were combined into an aggregate ES for each study in the respective meta-analysis. However, due to the askew distribution of correlations, we could not simply take the arithmetic mean of the ESs. Thus, we transformed the ESs into Fisher z-values, weighted them based on the number of cases, and then transformed the mean back into an r-value.

Subgroups were analyzed using R statistical software (RStudio, Version 1.4.1106) with the packages robumeta, metafor, and dplyr. The program converts the list of (aggregated) r-values of the uploaded Excel table to z-values and calculates a cumulative ES r+ (including standard error, level of significance, and 95% confidence intervals) using random effects models.

To test whether the difference in ESs between two subgroups is statistically significant, z-tests for two independent samples have been calculated [60].

Heterogeneity was examined using Baujat plots, τ², I² statistics, H² index, and Q test. For the detection of any outlier studies, primarily the Baujat plot was used. Studies that strongly contributed to heterogeneity (horizontal axis) and that strongly biased the overall result (vertical axis) were omitted gradually from the subgroup until no study was visible as an extreme outlier in the upper right corner of the Baujat plot. Each of these omission steps is recorded and the subgroup analysis is recalculated to keep the procedure transparent and the change in values due to the leaving-out of studies traceable. A maximum of three studies were excluded in a subgroup analysis, and these were often the same studies despite different subgroups.

Publication bias was examined using funnel plots. We tested for funnel plot asymmetry using a mixed-effects meta-regression model with standard error as predictor [61] and a rank correlation test (Kendall’s τ).

Transparency and openness

This meta-analysis follows the standards developed by the Transparency and Openness Committee [62]. All materials are either publicly available on the Open Science Framework (https://osf.io/2smz5/) or are available on request from the first author. With the supporting information, all methodological steps can be traced in detail. Tables for sub-group analyses can be created from the Excel spreadsheet that captures all extracted data (S5 Table). One such sub-table created by us is also available as an example (S6 Table). All other sub-tables used for subgroup analyses will be provided by the first author upon request. Using the R code, which is also publicly available (S1 Code), all our calculations can be reproduced.

In presenting our approach as well as our results, we followed the American Psychological Association (APA) Meta-Analysis Reporting Standards (MARS) [63] and the PRISMA 2020 statement [64]. This meta-analysis was registered with PROSPERO on 2 April 2021 (registration number: CRD42021240380). After discussion of an early draft manuscript, suggestions from commentators were incorporated, resulting in minor modifications of the preregistration plan.

Summary of the included and excluded studies

By searching databases, we identified 943 studies (cleared from duplicates: 508) that may report a statistical correlation between R/S and SOC. Of these studies, 51 met our inclusion and exclusion criteria. The targeted search for additional studies that fell through the grid of the database search (“berry picking”) revealed a further 92 studies. According to our list of criteria, 38 additional studies were included. Thus, in total, our meta-analysis is based on 89 studies with 67,913 participants. Of these 89 studies, 17 studies could only be included in the meta-analysis because missing information could be obtained through email contact (S7 Table). A list of the 444 studies that we sought for retrieval, assessed for eligibility and that were subsequently excluded on the basis of different criteria (I3-I5 and E1-E5) is included as supporting information (S8 Table). The search and selection process is illustrated in a PRISMA flow diagram (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. PRISMA flow diagram.

https://doi.org/10.1371/journal.pone.0289203.g002

In the methods sub-section "Coding matrix of the included studies", we defined 52 codes that are extracted for each included study. The result of this extraction process is listed in its entirety in an Excel spreadsheet provided as S5 Table. Our codes from (a) to (az) correspond to the column labels of the Excel table, ensuring easy traceability. Table 1 briefly presents the 89 included studies.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Measurement and effect size information of study samples included in the meta-analysis.

https://doi.org/10.1371/journal.pone.0289203.t001

Results of the critical appraisal process

Studies that met the inclusion criteria and did not fulfill any of the first three exclusion criteria were reviewed using our critical appraisal tool (S1 Tool) to determine how well they fit the focus of this meta-analysis. Seven studies [151–157] were excluded according to our fourth exclusion criterion because of "poor fit". Nevertheless, the data of these studies was entered in the overall Excel sheet (S5 Table) so that they could be considered in later analyses. The average total appraisal score of the 89 included studies was 15.13 points (range 9–24). Individual appraisal scores can be found in Table 1.

Results of the classification of the religion/spirituality measures

Studies that did not use at least one item to measure R/S were excluded from the meta-analysis (criterion I4). In a further step, the question was whether the R/S questionnaires used in the studies actually measured the variable R/S according to our heuristic definition of R/S (criterion E5), namely as communication with reference to something or someone transcendent (Fig 1). Using the coding aid (S4 Table), each individual questionnaire item was analyzed and given a label of R, RS, RS/SR or SR/RS, SR, S, or X (no reference to transcendence). More than a third (> 67%) of the items had to have a reference to transcendence for us to speak of an R/S measure. A total of 1,318 questionnaire items from 128 different potential R/S (sub-) scales were subjected to qualitative content analysis. The fact that 17 total scales, along with some of their sub-scales, were not considered as R/S measures according to our definition/coding (a detailed overview and rationale for the exclusion process can be found in S9 Table) resulted in 22 studies being excluded from the meta-analysis.

In the 89 studies that finally remained, 111 different positive R/S (sub-) scales were used with a total of 925 codable items. Of the 925 items coded, 45.2% were coded with R (418 items), 23.2% with S (215 items), 15.8% with RS (146 items), 7.4% with SR (68 items), 5.0% with X (46 items), and 3.5% with RS/SR (32 items). Based on the item codings, the scales were classified as a whole: Of the 111 (sub-) scales included, 47.7% were classified as R measures (53 scales), 22.5% as RS measures (25 scales), 17.1% as S measures (19 scales), 6.3% as SR measures (7 scales), 3.6% as RS/SR or SR/RS measures (4 scales), and 3 scales (2.7%) could not be classified. Eleven (sub-) scales were classified as negative R/S measures. Detailed documentation of the item-by-item analysis and classification of all those (sub-) scales that have been used in the included studies can be found in S10 Table.

Dominant aspects of the religion/spirituality measures

Each questionnaire item was not only assigned to a label (R, RS, RS/SR or SR/RS, SR, S, X), but was furthermore examined to determine which aspects of R/S it emphasized. By analyzing the 925 items, a scheme of 61 codes and nine superordinate categories was gradually developed (S11 Table). Over time, each code that emerged from the item-by-item analysis could be provided with a code description, trigger words, and an anchor example. The codes are at an intermediate level of abstraction and each designates an aspect of R/S (e.g., belief, individual practice or God relationship). A formula was then used to decide which R/S aspects of the (sub-)scale stood out as dominant. Which of the included R/S measures emphasize which R/S aspects can be seen in both Tables 1 and S10.

To give an example: The Intellect sub-scale of the Centrality of Religiosity Scale [158] consists of three items. Each item was individually analyzed using our coding scheme. The result was the following: Centrality (1), frequency (2), individual practice (1), intellect (3). This means that all items measure the intellectual aspect of R/S (which, by the way, speaks for the conceptually clean construction of the sub-scale Intellect), two items refer to the frequency of religious/spiritual activities, and one item asks for the centrality of R/S and the individual R/S practice, respectively. The codes in bold indicate that they are R/S aspects that are dominantly represented in this sub-scale.

Sense of coherence

Of the 89 included studies, 47 studies (52.8%) used the short version of Antonovsky’s SOC questionnaire (SOC-13), 38 studies (42.7%) used the long version (SOC-29), and four studies (4.5%) used only a SOC sub-scale or have not indicated which SOC-questionnaire version has been used. Across the 85 studies which used one of the two complete questionnaires, the overall SOC score was reported or could be calculated in 64 cases. In these studies, the total SOC score averaged M = 4.40 on a seven-point Likert scale. The highest average total SOC score measured was 6.02 points in a sample of 192 Jewish male volunteers [150]. The lowest total SOC score was 3.78 points in a group of 1,243 Romanian female medical students [81]. Based on the mean and SD of SOC scores, a three-level categorization was performed. The following tripartite was used: 1.00 to 4.25 points = weak SOC; 4.26 to 5.14 points = moderate SOC; 5.15 to 7.00 points = strong SOC. The majority of the studies (38) measured a moderate SOC, 14 studies measured a weak SOC and 11 studies measured a strong SOC. For the remainder of the studies (26), classification could not be made due to lack of information.

Correlations between religion/spirituality and sense of coherence

Overall and subgroup meta-analyses.

Based on our coding matrix (S5 Table), subgroup meta-analyses were conducted using a random effects model. The adjusted effect size between SOC and all positive R/S measures was r+ = .120, 95% CI [.092, .149]. In our research context, particularly significant (r+ < -.180 or > .180) were correlations between SOC and negative R/S scales (r+ = -.405, 95% CI [-.476, -.333]), R/S instruments measuring primarily positive emotions (r+ = .212, 95% CI [.170, .253]) or meaning-making (adjusted r+ = .196, 95% CI [.126, .265]). Both sample characteristics (age, culture, gender, health status, religious affiliation) and study characteristics (e.g., publication year) had a moderating effect on the R/S-SOC connection. The correlation was particularly high in eight studies from Southern Asia (adjusted r+ = .226, 95% CI [.156, .297]), the African Islamic cultural value zone (adjusted r+ = .196, 95% CI [.106, .285]), and in a small subgroup of six Iranian studies (adjusted r+ = .194, 95% CI [.117, .271]). The entire results of the subgroup analyses and the heterogeneity tests are summarized in Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Subgroup meta-analyses and heterogeneity tests.

https://doi.org/10.1371/journal.pone.0289203.t002

Detection of outliers.

Baujat plots were produced to identify influential outliers in a subgroup that biased both heterogeneity and the cumulative ES. Based on Baujat diagrams, it was decided that individual studies would be excluded from a subgroup and a new meta-analysis without outliers would be calculated. Recurrent outliers include the studies by Ghazinour et al. [94] and Hammer et al. [98]. The first study is characterized by a very high positive ES (r = .730), and the second study is characterized by a moderate to high ES (r = .373) and a relatively large sample (n = 1,530).

The subgroup analysis of all positive R/S measures can serve as an example. Based on a Baujat plot (Fig A in S2 Fig), the study by Ghazinour and colleagues was detected as an outlier and a second subgroup analysis was performed without this study. A comparison of the two subgroup analyses shows that the cumulative ES decreases from r+ = .128 (95% CI [.095, .160]) to .120 (95% CI [.092, .149]). At the same time, the very high heterogeneity (I² = 91.89%) in the subgroup with Ghazinour et al. only slightly decreases in the subgroup adjusted for the Ghazinour study (I² = 89.04%). However, after excluding Ghazinour et al., another study shows up in the upper right corner, which no longer extremely skews the overall result, but still has a relatively strong effect on the pooled ES (Fig B in S2 Fig). If we also exclude this study (Hammer et al.), the overall field moves closer together, but the basic picture of the Baujat plot no longer changes significantly (Fig C in S2 Fig). For this reason, we have chosen to consider the calculation of the subset of all positive R/S measures without Ghazinour et al. (but with Hammer et al.) as the authoritative variant.

The decision is further supported by the forest plot (Fig 3), which makes Ghazinour and colleagues’ study visible as a clear outlier, while Hammer’s and colleagues’ study—although further to the right of the overall effect estimate than most other studies—does not show major discrepancies. Furthermore, the long confidence intervals of the studies by Edwards and Besseling and Meghani and colleagues are remarkable, which, however, are not so important because of their rather small point estimates.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Forest plot for the subgroup of all positive R/S measures.

https://doi.org/10.1371/journal.pone.0289203.g003

In all cases, individual studies (at maximum two in each subgroup) were omitted gradually from the subgroup analyses until no study is visible as an extreme outlier in the upper right corner of the Baujat plot. The changes in values caused by each step of the leaving-out process can be seen in Table 2, so that every decision made is transparent. Despite the exclusion of outliers, the level of heterogeneity has generally remained high.

Religion/Spirituality and sense of coherence are significantly correlated

Our meta-analysis shows a positive significant relationship between positive R/S measures and SOC. If the outlier study identified by means of Baujat and funnel plot is excluded from the group, the overall ES is r+ = .120. Comparing the adjusted overall ES in our meta-analysis with results of other systematic reviews in the field of R/S-mental health-connection research, the association seems to be slightly higher (see the introductory paragraph above). The impression is strengthened when taking into account that German-language studies were also included in our meta-analysis, since the association between R/S and psychological health has been shown to be lower in samples from the German-speaking area than, for example, in meta-analyses based primarily on U.S. samples [7].

In addition, it appears that the breadth of our inclusion criteria led to a decrease in the cumulative ES. With slightly stricter criteria (only studies from the year 2000 onward; only studies with a sample size ≤ 100; no measures of extrinsic religious orientation), the result would have been even more significant: In the remaining 65 studies the cumulative ES is r+ = .139, 95% CI [.103, .175] based on an overall sample size of 60,608. If Ghazinour et al. [94] is omitted from this subgroup as an outlier study, the overall ES is still r+ = .130, 95% CI [.099, .160], with the calculation based on 64 studies with 60,508 participants. The result supports the timeliness of the SOC hypothesis and confirms the assumption of Garssen and colleagues [4] that SOC is a particularly well-suited indicator of psychological adjustment in the context of R/S-mental health connection research.

The relationship between SOC and positive R/S is also moderated by personal factors (e.g., age, gender, health status). In our analysis, a strong correlation between SOC and R/S was measured in the subgroup of middle-aged people (adj. r+ = .150), whereas the ES was significantly lower in younger (adj. r+ = .077) and older people (r+ = .056). In addition, samples composed exclusively or primarily of men have significantly higher ES than samples composed exclusively or primarily of women (z = 2.617, p = .004). Finally, in non-clinical samples, the ES turns out to be higher (adj. r+ = .118) than in clinical samples (r+ = .112), although the difference is not significant (z = .299, p = .382). In short, the SOC-R/S association is particularly positive in middle-aged men. Given that in most countries men rate themselves as less religious/spiritual than women [159,160], we attribute our result more to the influence of the SOC variable. At least, some studies suggest that the SOC of men is stronger than the SOC of women [24]. Since Antonovsky [18] emphasized that many salutary experiences of comprehensibility, manageability, and meaningfulness are gained in work contexts and are closely related to social recognition, it can be surmised that the stronger ES between SOC and R/S in the subgroup of “men and mostly men” is related to the still higher socio-economic status of middle-aged men (31–60 years) in many countries and cultures. For example, several studies show that there is a clear relationship between the level of income and the level of SOC [161–163]. In addition, religious power and prestige [164] might play a moderating role in the SOC-R/S association because in most traditional religions (e.g., Buddhism, Catholicism or Islam) men play a prominent role in religious organizations and communities, which might be a source of their SOC.

As was to be expected, there is also a clear significant inverse correlation between negative R/S measures and SOC of r+ = -.405. Hodapp and Zwingmann [7] argued that such a pronounced negative association is particularly found in German-speaking countries. The fact that three studies with large German-speaking samples (n+ = 4,960) are part of our subgroup might suggest that this bias partly explains the high negative value. However, the negative ES in our subgroup analysis is twice as high as the ES of r+ = -.200 in Hodapp and Zwingmann’s meta-analysis of 37 negative correlations from the German-speaking area. The strikingly high correlation, although not very robust due to the small number of ESs in the subgroup (11), can probably be attributed to the high fit between religious samples and R/S measures. Two German studies in our subgroup used the Spiritual Dryness Scale [73] in church-active Catholics, and three studies used the Negative Religious Coping subscale [165] in two Iranian Muslim samples and a refugee sample. If we take a closer look at the items of the two scales, we notice that both negative R measures focus on a God relationship that is burdened by negative emotions. SOC as an outcome parameter might be particularly sensitive to the loss of transpersonal trust in these religious samples because, by and large, SOC is by definition a measure of a “pervasive, enduring though dynamic feeling of confidence” [18].

Emotional and meaning aspects of religion/spirituality are most associated with sense of coherence

Salsman and colleagues [166] used a taxonomy of four categories (affective, behavioral, cognitive, other) to classify the R/S measures in their meta-analysis of the R/S and mental health association in cancer patients. They concluded that the affective R/S dimension (with the conceptually problematic Spiritual Well-Being scales excluded) is significantly more strongly associated with mental health (z = .290) than the other dimensions: cognitive R/S (z = .100), other R/S (z = .080), and behavioral R/S (z = .030) [167]. If we assign the 20 dominant aspects of R/S that we were able to subject to subgroup meta-analysis (Table 2) to the higher-order categories “affective”, “cognitive”, and “behavioral”, a strikingly consistent picture emerges (Fig 4). In our meta-analysis, the emotional aspect of R/S exhibits the strongest ES (r+ = .212), whereas the intellectual aspect of R/S is hardly related to SOC (r+ = .062). Other R/S aspects that have a strong affective component (meaning, trust, God concept, God relationship, experience) also have consistently higher r-values than more intellectual aspects of R/S (afterlife, development, belief). In the middle range of the ESs, R/S aspects are found that can largely be categorized as behavioral (e.g., frequency, churchiness, collective practice, centrality, way of life).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Three-dimensional model of aspects of religion/spirituality.

https://doi.org/10.1371/journal.pone.0289203.g004

That emotion and meaning turned out to be the R/S aspects most associated with SOC becomes particularly plausible when one considers this definition of the SOC component “meaningfulness” inspired by Antonvosky: "Meaningfulness, the motivational dimension, refers to the extent to which one feels that life has an emotional meaning, that at least some of the problems faced in life are worth commitment and dedication, and are seen as challenges rather than only as burdens" [168]. It is therefore hardly surprising that the cumulative ES between positive R/S measures and the SOC-subscale Meaningfulness, adjusted for outlier studies, is r+ = .144. This association is strikingly stronger than the non-significant relationship between R/S and the cognitive SOC-subscale Comprehensibility, which, when adjusted, is only r+ = .043. The behavioral SOC component Manageability is also much less correlated with positive R/S measures (r+ = .071), so that it can be assumed that the R/S-SOC connection can be explained primarily by the emotionally stabilizing and meaning-giving function of R/S as a resource of SOC as an “enduring though dynamic feeling of confidence” [18]. This observation echoes the conceptual ideas of German theologian Hans Küng, who defined “faith” as “basic trust in life” (Lebensvertrauen) and which he described as the “cornerstone of a healthy personality” [169] and the “foundation of a global ethic” [170].

Sense of coherence is more correlated with spirituality than religion

As religion and spirituality become more socially and scientifically differentiated, the question arises as to which form of reference to transcendence is more associated with SOC. From a theoretical perspective, Antonovsky viewed the societal processes that accompanied the emergence of spirituality in modernity, namely an increasing de-traditionalization and individualization of faith, rather critically. Because religion, in his view, would rather contribute to group cohesion and worldview congruence, he assumed that institutionalized religion would be a stronger biographical SOC source than individualized spirituality (for a detailed analysis, see [43]).

However, the results of our meta-analysis refute Antonovsky’s hypothesis and show that S measures (r+ = .138) are significantly stronger (z = 3.996, p < .001) associated with SOC than R measures (r+ = .094). As expected, mixed measures, i.e., instruments that include approximately the same number of S as R items, show a correlation strength that lies between the other two subgroups (adj. r+ = .124). The result can thus be reduced to the formula: The more items an R/S scale includes that measure spirituality (as reference to non-institutionalized and conceptually/semantically open transcendent ascriptions of meaning), the higher the correlation with SOC.

The result is probably due to a complex amalgamation of sociocultural developments, academic trends, and methodological factors. One of the main reasons might be the impact of changing study populations (e.g., the growing group of the non-religious or the “spiritual but not religious” [SBNR]) that respond differently to the (older) R and the (newer) S measures. Measurement instruments are constructed not only with certain questions in mind, but also with certain samples. This (to a greater or lesser extent) time, culture, and sample boundedness of many R/S measures can lead to results being biased. For example, it is likely that participants who would rate themselves as non-religious or SBNR would not score well on questionnaires that we classified as R measures. For instance, the study by Bradbury et al. [46] supports this explanation. The significant negative association between the R measure (Traditional Religious Beliefs) and SOC of r = -.220 is most likely due to the sample of a primarily non-religious population of undergraduate students in England. Since the medieval-looking four-item sub-scale of the Revised Paranormal Belief Scale [171] asks primarily for agreement with metaphysical Christian concepts (soul, devil, God, heaven, hell), it will presumably correlate positively with SOC only in a country or at a time when Christian dogmas are culturally strongly anchored and still have credibility. Thus, a strong positive relationship between religion and SOC might depend on whether and to what extent certain concepts of transcendence are supported by socio-cultural plausibility structures.

Religion/Spirituality as social value is strongly associated with sense of coherence

The subgroup analysis results compiled in Table 2 under the sub-headings “Country”, “Geographic zone”, “Cultural value zone”, and “Religious affiliation” suggest: If the religious/spiritual worldviews, values and practices of the subgroup under study are supported by the cultural environment, i.e., if social plausibility structures are broadly intact, the R/S-SOC relationship turns out stronger. In agreement with the findings of Hodapp and Zwingmann [7], a transatlantic comparison shows that the R/S-SOC connection was found to be significantly weaker (z = -6.751, p < .001) in German-speaking samples (r+ = .109) than in populations from the United States (r+ = .171). This may be related not only to the different levels of religiosity of people [172], but also to the massive loss of social relevance of traditional Christian religion in German-speaking countries. However, an intra-European comparison shows that studies carried out in the geographic zone Western Europe (to which the German-speaking countries geographically belong) performed better (r+ = .120) than studies from Southern Europe (r+ = .075) or Northern Europe (r+ = -.014).

The country or geographic differences could be attributed to varying cultural value zones. According to The Inglehart-Welzel World Cultural Map [59], most northern European countries belong to the cultural value zone "Protestant Europe", which emphasizes secular-rational and self-expression values and is more critical of religion, so R/S does not have the same plausibility strength as in cultures where R/S still plays a vital role in the societal value system. In many Western countries with Christian background, people of faith face the question of whether to leave or remain in the constituted church, which leads to inner insecurity, religious/spiritual doubts, and an unstable religious identity. While a stable religious affiliation is usually associated with better mental health, those who are undecided suffer more from depressive symptoms, according to an American study [173]. Thus, it is hardly surprising that a strong ES (adj. r+ = .194) was measured in a small subgroup of six Iranian studies, which belongs geographically to Southern Asia (adj. r+ = .226) and to the so-called “African Islamic” cultural value zone (adj. r+ = .196). On the World Cultural Map, the “African Islamic” zone is located in the lower left corner, diametrically opposite to the “Protestant Europe” zone, which means that the countries in this cultural zone emphasize traditional (religious) values. Many countries from this cultural value zone have a binding state religion (e.g., Shiite Islam in Iran or Sunni Islam in Pakistan) and/or the people feel they belong to a relatively stable, political powerful and socially valued majority religion [174]. Sample populations that consist entirely or predominantly of Muslims (adj. r+ = .131) or Christians (adj. r+ = .136) have a significantly stronger R/S-SOC relationship than heterogonous samples (r+ = .071). If this result is not due to a bias caused by a too small subgroup size (e.g., the adjusted Muslim subgroup consists only of 1,550 participants and 11 ESs), it can be surmised that R/S is a stronger SOC source in homogenous religious/spiritual samples.

Overall, our results are thus compatible with the "religiosity as social value hypothesis" (RASV), which states that in religious cultures religiosity is more likely to be a social resource for self-esteem than in secular cultures [175]. By analogy with the RASV hypothesis, we can summarize our assumption as follows: If the culture in which a (religious/spiritual) person lives supports his/her (religious/spiritual) worldview/practice, i.e., if this worldview/practice is culturally embedded, socially plausibilized, and socially valued, then the likelihood that this person will have a strong SOC increases. In technical terms, the variable "cultural value orientation" seems to have a moderating effect on the relationship between R/S and SOC. However, our hypothesis does not make any statement about the causality of the relationship: A strong SOC may lead a person to follow a religion/spirituality that is socio-culturally accepted. Or, conversely, a culturally embedded and socially valued R/S may be a source of a strong SOC. However, since there are many individual pathways to a strong SOC, it must remain open what role R/S plays in a person’s life and whether there are not other, more central socio-cultural and personal SOC sources.

Strengths

Our meta-analysis focused on a well-established mechanism within R/S-health research, namely the SOC hypothesis, which promised robust results from a theoretical perspective [17].

The multitude of possible combinations between the multidimensional construct R/S and negative/positive indicators for mental health quickly leads to an unmanageable number of options (e.g., spiritual struggle/anxiety, religious affiliation/life satisfaction, religious coping/locus of control, etc.) or to simplifying generalizations (e.g., religion/mental health). However, by selecting SOC as a single, widely used and psychometrically valid indicator of psychological adjustment, we avoided the complexity of one variable (mental health) in order to deal more precisely with the intricacy and richness of aspects of the other variable (R/S).

Perhaps the greatest strength of our meta-analysis is that all items of all R/S measures were subjected to a qualitative content analysis (S10 Table). In this way, it was possible to decide in a transparent and systematic way whether a scale measures religion, spirituality, a mixed form, or (despite a scale name that promises otherwise) neither. At the same time, all included R/S instruments were analyzed item-by-item to determine which aspects of R/S they (actually) measure, which in individual cases may mean that a scale on spiritual well-being turns out to be a measure of the relationship with God. The categorization of R/S scales reduced the likelihood that the meta-analysis will be biased by measurement-specific issues and led to more valid and more fine-grained results.

Publication bias was further reduced by the inclusion of dissertations and the use of a self-developed critical appraisal tool to ensure that inappropriate and methodologically problematic studies (“poor fit”) were excluded from the meta-analysis.

Another advantage of our meta-analysis was that both English- and German-language studies were included. This not only expanded the data material, but also allowed cross-cultural comparisons and the analysis of cultural factors as potential moderators of the R/S-SOC connection.

Limitations

Since, strictly speaking, the SOC hypothesis conceptualizes SOC as a mediator between R/S and health, the results of the meta-analysis can only be considered as a partial confirmation of the hypothesis. While the SOC hypothesis assumes an indirect causal effect of R/S (predictor) on health (outcome) interlinked by SOC (mediator), the present meta-analysis does not make any statements about causality. Because primary research on the R/S-SOC connection has used almost exclusively correlational study designs, the causal direction of variables remained unclear. R/S may be a biographical source for a strong SOC. At the same time, a strong SOC may be a source for the mobilization of religious/spiritual resources in coping with stressful events. Thus, a beneficial relationship is presumably bi-directional.

While our decision to choose SOC as the sole indicator for the complex construct of mental health necessarily reduces the complexity of the meta-analysis to a manageable level, it also has some disadvantages. Only tentative conclusions can be drawn about the association between R/S and mental health in general, as SOC is only one positive variable that may interact with the phenomenon of R/S in specific ways. And even though SOC is often considered a transcultural construct [176,177], it is conceivable that there are cultures that would not prioritize SOC as a crucial indicator of mental health, which could limit the generalizability of our results.

The broadness of our inclusion criteria led to a high heterogeneity of the studies (in terms of methodology as well as sample and study characteristics), which in turn made it difficult to identify the exact religious/spiritual factors affecting the R/S-health connection. The systematic differentiation of R/S scales according to our heuristic continuum model of R/S (Fig 1) and the inductively developed taxonomy of R/S aspects (S11 Table) did help in a more precise analysis. But this very approach involved preliminary decisions (e.g., the definition of the constructs “religion” and “spirituality”) and many further choices (e.g., the coding of each individual item) that affected the outcome. For this reason, we tried to increase intersubjective comprehensibility through intra- and interrater agreement and explained in detail our methodological procedures, ensuring transparency (S1 Text).

The validity of our results, as with almost any meta-analysis in this line of research, is limited by the fact that both the SOC and the various R/S measures are self-report instruments. Self-assessments are, of course, highly subjective and susceptible to bias due to social desirability [178,179]. With regard to SOC measurement, there is also the fact that extremely high SOC scores and possibly very high R/S-SOC correlations may be indicative of rigid SOC [180]. Antonovsky has also referred to rigid SOC as inauthentic SOC, as it is based on a simplistic, fundamentalist worldview that can be (but need not be) fragile when subjected to a reality check [18].