God representations and aspects of psychological functioning: A meta-analysis

Abstract Context: Results of meta-analyses show weak associations between religiosity and well-being, but are based on divergent definitions of religiosity. Objective: The aim of this meta-analysis was to examine the magnitude of the associations between God representations and aspects of psychological functioning. Based on object-relations and attachment theory, the study discerns six dimensions of God representations: Two positive affective God representations, three negative affective God representations, and God control. Associations with well-being and distress and with self-concept, relationships with others and neuroticism were examined. Methods: The meta-analysis was based on 123 samples out of 112 primary studies with 348 effect sizes from in total 29,963 adolescent and adult participants, with a vast majority adherent of a theistic religion. Results: The analyses, based on the random-effects model, yielded mostly medium effect sizes (r = .25 to r = .30) for the associations of positive God representations with well-being, and for the associations of two out of three negative God representations with distress. Associations of God representations with self-concept, relationships with others and neuroticism were of the same magnitude. Various moderator variables could not explain the relatively high amount of heterogeneity. The authors found no indications of publication bias. Conclusion: The observed effect sizes are significantly stronger than those generally found in meta-analyses of associations between religiousness and well-being/mental health. Results demonstrate the importance of focusing on God representations instead of on behavioral or rather global aspects of religiosity. Several implications with respect to assessment, clinical practice, and future research are discussed.

Abstract: Context: Results of meta-analyses show weak associations between religiosity and well-being, but are based on divergent definitions of religiosity. Objective: The aim of this meta-analysis was to examine the magnitude of the associations between God representations and aspects of psychological functioning. Based on object-relations and attachment theory, the study discerns six dimensions of God representations: Two positive affective God representations, three negative affective God representations, and God control. Associations with well-being and distress and with self-concept, relationships with others and neuroticism were examined. Methods: The meta-analysis was based on 123 samples out of 112 primary studies with 348 effect sizes from in total 29,963 adolescent and adult participants, with a vast majority adherent of a theistic religion. Results: The analyses, based on the random-effects model, yielded mostly medium effect sizes (r = .25 to r = .30) for the associations of positive God representations with wellbeing, and for the associations of two out of three negative God representations with distress. Associations of God representations with self-concept, relationships with others and neuroticism were of the same magnitude. Various moderator variables could not explain the relatively high amount of heterogeneity. The authors Henk P. Stulp

ABOUT THE AUTHOR
The author is a lecturer at a Dutch Christian university of applied sciences that trains students for work in the domains social work, pastoral work, nursing, and education. This article is the introduction of his PhD project about God representations. The project is based on the "relational spirituality" approach that emphasizes the importance of the personally and emotionally experienced relationship with the Divine. Based on attachment theory and object-relations theory -two dominant theoretical frameworks in religion research-a central assumption of the project is the importance of implicit aspects of God representations. A measure for assessing implicit God representations, the Apperception Test God Representations (ATGR), has been developed. Aim of the project is the validation of this measure. Thereto, in a patient and a nonpatient group, associations of implicit and explicit measures with measures of implicit and explicit distress and object-relational functioning, amongst others, are compared.

PUBLIC INTEREST STATEMENT
This meta-analysis demonstrates the importance of personal God representations for religious people who believe in a personal relationship with the God they believe in. It found that positive God representations are most strongly related to wellbeing and negative God representations most strongly to psychological distress. Subjects with negative God representations also have a more negative perception of self and others and are more susceptible to negative emotions, than subjects with positive God representations. Thereby it underlines that theoretically God representations can be viewed as a special form of mental representations of relationships. This implies that patients with disturbed perceptions of self and others may also have difficulties in benefitting from the support most believers find in the relationship with their God. Results also may imply that therapy should not only focus on strengthening positive, but also on directly dealing with and transforming negative God representations. found no indications of publication bias. Conclusion: The observed effect sizes are significantly stronger than those generally found in meta-analyses of associations between religiousness and well-being/mental health. Results demonstrate the importance of focusing on God representations instead of on behavioral or rather global aspects of religiosity. Several implications with respect to assessment, clinical practice, and future research are discussed. Keywords: meta-analysis; god representations; well-being; mental health; self-concept During the last decades, there has been a significant increase in attention in scientific research for religion in the context of mental health. In mental health care, religion has long been thought to have a negative effect on health (Neeleman & Persaud, 1995). This can be traced back to Sigmund Freud's view that religion is a projection of an infantile need for an authoritative being that can function as a father figure (Freud, 2004). As a consequence, religion was supposed to have a predominantly negative influence on mental health because, according to this view, religion would be accompanied by many restrictive rules that lead to strong feelings of guilt and fear of punishment by an angry god. Other psychologists (Rizzuto, 1979;Winnicott, 1971) have argued that religion may also have a positive influence on psychological functioning because believers may as well project positive attributes to their god. This can give them strength and may contribute to personal growth.
Although convincing evidence-as presented below-exists for the association between religiosity and well-being/mental health, not much is known yet about the underlying mechanisms that explain this relation. More insight is needed, and this is especially important for health professionals working with religious/spiritual patients. It might contribute to the development of interventions that may strengthen the potential positive influences of religion/spirituality (R/S), and to interventions that may lead to diminishing or solving negative influences.
There is a lot of debate about the definitions of religiosity and spirituality (Hill et al., 2000;Zinnbauer & Pargament, 2005). According to Koenig, King, and Carson (2012), the terms religion and religiosity are often used to refer to shared beliefs and rituals and to the membership of a faith community, whereas the term spirituality is often used to emphasize more individualistic beliefs and rituals. However, basically, both concepts share a belief in the sacred and the transcendental. In this meta-analysis, we will therefore use both terms interchangeably. However, the main focus of this study is on a specific aspect of religiosity and spirituality that is based on monotheistic religions (as, e.g. Christianity, Islam, and Judaism) that assume the existence of one personal God to whom the believer can relate (Davis, Granqvist, & Sharp, 2018): the personal God representation.
In this meta-analysis, we will, amongst others, examine if the personal God representation has stronger associations with well-being/mental health than more general aspects of religiosity. There is confusion about the construct of God representations (Gibson, 2008). Terms like God concept, God image, and God representation are often used interchangeably. A useful distinction is that between two dimensions of God representations: cognitive/doctrinal beliefs (about how God is conceptually viewed by a person) and emotional/experiential feelings about God, about the personally experienced relationship with God (Davis, Moriarty, & Mauch, 2013;Zahl & Gibson, 2012). In this study we will focus on the relational/emotional/experiential dimension.
For adherents of a theistic religion, someone's God representation may indicate psychological mechanisms at work that could explain much of the association between religiosity and wellbeing. There are some sound reasons to focus on God representations concerning well-being and mental health. One of them is that findings from studies of the associations between broader defined religiosity and well-being suggest the importance of personal beliefs. Therefore, we will first explore the results of these findings. Another reason is that on theoretical grounds God representations can be viewed as an important explanation for the found associations between religiosity and wellbeing/mental health. We will subsequently discuss these theoretical grounds, based on attachment and object-relations theory. Well-being/mental health and its counterpart, psychological distress are summarized in this study with the term adjustmental psychological functioning, to emphasize the general notion that they can be viewed as indicators of psychological adjustment (Ano & Vasconcelles, 2005;Salsman, Brown, Brechting, & Carlson, 2005).

The associations between religiosity/spirituality and adjustmental psychological functioning
The available meta-analyses of the associations between religion and adjustmental psychological functioning (Ano & Vasconcelles, 2005;Bergin, 1983;Hackney & Sanders, 2003;Smith, McCullough, & Poll, 2003;Witter, Stock, Okun, & Haring, 1985) suggest that in general being (more) religious is associated with higher well-being and with fewer mental health problems (see Table 1). The found associations are weak, but support the notions of Winnicott (1971) and Rizzuto (1979) about the potential positive influences of religiosity.
Various factors influence the strength and direction of the associations, such as the variety in dimensions and aspects of religiosity. Witter et al. (1985), for example, found stronger positive associations for activities than for beliefs. Hackney and Sanders (2003), in turn, found stronger associations for personal devotion than for institutional membership and ideology, whereas Smith et al. (2003) found that extrinsic religiosity was positively, and other measures of religiosity (e.g. intrinsic religious orientation, religious attitudes, and beliefs), were negatively associated with depressive symptoms. A second factor is the distinction between positive and negative aspects of religiosity and of psychological adjustment. Results of Ano and Vasconcelles (2005), for example, suggest that positive aspects of religiosity (e.g. asking for forgiveness, seeking support from clergy, seeking spiritual connection) are more strongly associated with positive aspects of adjustmental psychological functioning, and negative aspects of religiosity (e.g. spiritual discontent, seeing God as punishing) more strongly with negative aspects of adjustmental psychological functioning. The relevance of these finer distinctions within the concept of religion (and spirituality) is that they may explain some of the ambiguous or inverse associations found in a minority of the included studies.
Most narrative reviews about the association between religiosity and adjustmental psychological functioning (Ellison & Levin, 1998;Gartner, Larson, & Allen, 1991;Koenig et al., 2012;Koenig, McCullough, & Larson, 2001;Larson et al., 1992;Payne, Bergin, Bielema, & Jenkins, 1991) also conclude that religiosity is predominantly positively associated with well-being, and predominantly negatively with mental problems, but that there are also studies with ambiguous or inverse results. One factor that seems related to negative or ambiguous results is psychopathology: Payne et al. (1991) found negative or no associations for the few studies with clinical samples in their review, and Koenig et al. (2012) found relatively more studies with positive associations between religiosity and mental problems for C-cluster Personality Disorders (18 studies, 17% negative, 50% positive) and Bipolar Disorder (4 studies, 0% negative; 50% positive).
1.1.1. Explanations for the associations between R/S and well-being/mental health Koenig et al. (2012) developed various comprehensive models to explain associations between religion and mental health. In their model for monotheistic religions they stress the importance of God representations: The relationship with God has direct effects on wellbeing and mental health, fostering positive emotions caused by a sense of being loved and protected by a beneficial divine being. They also include indirect effects in their model: religion generates social support, offers sources and strategies of coping, influences (good) choices, and diminishes the influence of More specific explanations are offered by attachment and object-relations theory. Both developmental theories assume that a core element of personality and personality pathology, namely how persons view themselves and others (Livesley, 1998(Livesley, , 2013, influence how they see and experience their relationship with God. This approach of religion is known as "relational spirituality" (Davis et al., 2018;Hall, 2007aHall, , 2007b and also integrates findings from stress-coping theory, social cognition theory, and brain research.
Object-relations theory and attachment theory (Hall, 2007a(Hall, , 2007b both assume that mental representations of people are formed during early development, which in turn influence the way God representations are formed. These experiences lead to mostly unconscious relational schemas or internal working models, which comprise representations of self and others, as well as their affective quality.
Less optimal experiences of responsivity and availability, according to attachment theory, may result in insecure attachment styles, such as: (a) anxious attachment: trying to restore disturbed feelings of security by using hyperactivating strategies (e.g. expressing anxiety and anger) to establish the availability of the attachment figure; (b) avoidant attachment: trying to restore this inner sense of felt security by using deactivating strategies (e.g. suppressing disturbing emotions or thoughts Bowlby, 1972Bowlby, , 2008Bretherton & Munholland, 2008Mikulincer & Shaver, 2008). In normal development, internal working models foster the capacity for affect regulation and stress coping (Fonagy, Gergely, Jurist, & Target, 2004;Mikulincer & Shaver, 2008). Insecure working models of attachment relationships may confer risk for physical disease and psychopathology through non-adaptive coping and impaired stress and affect regulation (Maunder & Hunter, 2008). Several studies have confirmed the usefulness of the attachment theory framework in the domain of religion (Granqvist, 1998;Granqvist & Hagekull, 1999;Hall, Fujikawa, Halcrow, Hill, & Delaney, 2009;Kirkpatrick, 1998;Kirkpatrick & Shaver, 1990. According to object relations theory (Fairbairn, 1954;Klein, 1946;Mahler, 1971;Winnicott, 1971), pathological internal working models involve less integrated representations of self and others. On the lowest levels, persons have difficulty in differentiating between the self and others, or in integrating positive and negative feelings about self or others. This often leads to emotional instability and the use of primitive defense mechanisms like splitting and projective identification. On lower levels others are predominantly viewed as less benevolent (affectionate, benevolent, warm, constructive involvement, positive ideal, nurturant) and more punitive (judgmental, punitive, and ambivalent) (Huprich, Auerbach, Porcerelli, & Bupp, 2015;Kernberg & Caligor, 1996). Higher, healthier levels correspond to more integrated and symbolized representations of self and others, involving affect tolerance, regulation, ambivalence and the ability to understand the perspective of others. There is also evidence of the usefulness of object-relations theory in the domain of religion (Brokaw & Edwards, 1994;Hall & Brokaw, 1995;Stalsett, Engedal, & Austad, 2010;Tisdale, Key, Edwards, & Brokaw, 1997).

Dimensions of God representations
Most measures of God representations have been derived from these described theoretical frameworks, and therefore for this meta-analysis we based our dimensions of God representations predominantly on these theories: Secure, anxious and avoidant attachment to God (Granqvist & Hagekull, 1999;Kirkpatrick, 1998;McDonald, Beck, Allison, & Norsworthy, 2005), and positive and negative God representations, which we derived from measures using adjectives/attributes like benevolent, kind, supporting or wrathful, judging/punishing, for how God is perceived, and terms like gratitude, fear, anger etc., for the feelings a person experiences in his or her relationship with God (Benson & Spilka, 1973;Lawrence, 1997;Schaap-Jonker, Eurelings-Bontekoe, Verhagen, & Zock, 2002).
One aspect of God representations is not as clearly related to these theoretical frameworks, and regards the extent to which God-according to the subject-has power, exerts control, gives guidance (Benson & Spilka, 1973;Schieman, 2008). We refer to this aspect as the God control aspect.

God representations and dispositional aspects of psychological functioning
Attachment and object-relations theory both assume that general schemas underlie both interpersonal representations of self and others and God representations. These general schemas or models are supposed to have trait-like characteristics. Traits are general "underlying", not directly observable dispositions that have relative stability over time and are supposed to be related to heredity and upbringing (Fridhandler, 1986;Mischel, 2013;Strelau, 2001). Some scholars, for example, refer to attachment models as relatively stable traits (Green, Furrer, & McAllister, 2007) or chronic general models (Collins & Feeney, 2004). Davis et al. (2013) assume that God representations also have trait-like, chronic characteristics. However, it must be emphasized that these working models are especially determined by interactions with caregivers, and therefore have to be considered less stable than temperament-based traits.
If it is true that relatively stable general schemas underlie both God representations and internal working models of self and others, one would expect God representations and representations of self and others to be associated with each other. In attachment theory research in the domain of religion, this assumed association is known as the correspondence hypothesis (Granqvist, 1998;Kirkpatrick, 1998;Kirkpatrick & Shaver, 1990). But these authors also hypothesize that attachment to God representations may compensate for insecure or negative interpersonal representations (known as the compensation hypothesis). Hall et al. (2009) assume correspondence on the deeper level of (implicit) internal working models, and on a more behavioral level they expect evidence of compensation. This compensation implies that insecurely attached persons may be more actively involved in actions aimed at finding relief in religion and in the relationship with God.
We expect that God representations are not only associated with adjustmental aspects of psychological functioning, but also with relatively stable, trait-like representations of self and others, and with neuroticism as an indicator of trait-like affect (dis)regulation. We will refer to these factors as dispositional aspects of psychological functioning. Existence of associations between God representations and dispositional aspects of psychological functioning can be considered as support for the importance of the ideas of attachment and object relations theory for understanding the development of God representations.

Aim of meta-analysis
In this meta-analysis we examine the associations between God representations and adjustmental aspects of psychological functioning, to see if these associations are stronger than the usually found associations with broader measures of religiosity. We also examine the associations between God representations and dispositional aspects of psychological functioning: theoretically related variables that are connected with internal working models of relationships: self-concept, relationships with others and neuroticism.
The meta-analytic method is suitable to detect sources of diversity (Borenstein, Hedges, Higgins, & Rothstein, 2005). Because we used a wide variety of God representation measures and measures of dispositional and adjustmental aspects of psychological functioning, originating from diverse samples, this meta-analysis especially aims at detecting sources of diversity. Therefore we performed analyses on three levels, starting from the most general level that compromises all God representation dimensions and examining associations with undifferentiated adjustmental and dispositional aspects. On the second level, we split out the God representation measures in the six dimensions (Secure attachment to God, Anxious attachment to God, Avoidant attachment to God, Positive God representations, Negative God representations, and God control) again examining associations with undifferentiated adjustmental and dispositional aspects. On the third level, we examined more specific associations between dimensions of God representations and the adjustmental subdomains of well-being and distress and the dispositional subdomains self-concept, relationships with others, and neuroticism (as an operationalization of the capacity for affect regulation). We compared the strength of associations between these various measures. We also aimed to detect the effect of various moderator variables on the found associations. Finally we addressed the issue of publication bias, to determine whether in the selected studies an underrepresentation of studies with weak or non-significant associations existed.
Hypothesis 1. We expect that (a) positive God representations will be significantly and positively related to well-being and negatively to distress, and that (b) negative God representations will be significantly and negatively related to well-being and positively to distress. The strength of these associations will be larger (>.20) than the weak aggregated association of about r = .10 between religiosity and well-being/mental health that is generally found in the discussed meta-analyses, because we assume that God representations are a more determining aspect of religiosity than many other widely used measures.
Hypothesis 2. We expect that (a) positive God representations will be significantly and positively related to positive self-concept and to positive relationships with others, and negatively to neuroticism, and that (b) negative God representations will be significantly and negatively related to positive self-concept and to positive relationships with others, and positively to neuroticism.

Moderator analyses
To gain more insight into the association between God representations and psychological aspects, it is also important to examine the influence of potential moderator variables on this association. As moderator variables we use the various study-and sample characteristics of the included studies: (a) context/respondent status (samples with subjects with mental health problems or serious life problems); (b) method of measurement (self-report or implicit/indirect measures); (c) religion/denomination; (d) religiosity (the degree of religious involvement); (e) gender; (f) age; (g) quality of the study; (h) year of the study; and (i) quality of God representation measures.

Eligibility criteria
Included were all studies with samples with a mean age of 15 years or older, regardless of design, using a combination of on the one hand a measure for God representations (aimed at a monotheistic belief in a personal god) and on the other hand a measure of an adjustmental or dispositional dimension factor (as defined), and of which we obtained a statistical association measure for one or more association(s) between them. Only scholarly (peer-reviewed) journal articles were included. No language restrictions were imposed. All studies complying with these criteria, dating from 1990 to May, 2015 were included.

Literature search
The search strategy was developed by the first author, in cooperation with an experienced librarian/data information specialist and adjusted for the different search machines/databases. Searches were conducted in Psychology and Behavioral Sciences Collection, MEDLINE, PsycINFO and PsycARTICLES by the comprehensive search machine Academic Search Premiere, and in Science Direct, restricted to journals in the sections Nursing and Health Professions, Psychology and Social Sciences, in May 2015. Search terms for God representations were all possible combinations of the term God with (different forms of) the terms image or representation or concept or attachment. These terms were combined with the terms for the adjustmental or for the dispositional dimension. For the adjustmental dimension the terms anxiety, depression, pathology, distress, therapy, outcome, well-being, happiness, life satisfaction and adjustment were used, and for the dispositional dimension the search consisted of the terms personality, object relation, adult attachment and child attachment.

Study selection and data extraction process
First, two researchers (first and third author) independently screened titles and abstracts for inclusion; articles on which both agreed about exclusion, were excluded. From the remaining articles, the full text was read and independently assessed. Disagreement or doubt was resolved in consensus discussions. This resulted in 135 initial studies to be included.
Fifty-six studies of forty-nine authors did not report (all) correlations. Authors of studies with missing data or without the required data format for any of the relevant associations were approached by email in an attempt to obtain the correct data. Two reminders were sent in case of no response. Twenty-five authors replied (51%), 13 authors (26.5%) provided us with the missing correlations for 20 studies, 12 replied that the data were not available anymore. Twenty-one did not respond to the emails, and from three authors their email address was unknown or no longer operational. From the remaining 36 incomplete reporting studies, 17 studies could be included because they reported about at least one of the associations of this meta-analysis. The remaining (36-17=) 19 studies had to be excluded from the meta-analysis because they did not report about any associations between the measures of this meta-analysis. This resulted in (135-19 =) 116 studies. Four of these studies were excluded because they reported about the same samples and measures, resulting in (116-4 =) 112 studies.
Four studies had the same samples but reported about different measures. These studies were combined, resulting in (112-4 =) 108 separate or combined studies. Ninety-six of these studies consisted of one sample, 10 studies had two samples with appropriate associations, one study had three samples, and one study had four samples with appropriate associations, resulting in (96 × 1 + 10 × 2 + 1 × 3 + 1 x 4) = 123 independent samples ( Figure 1).
Data from selected studies were extracted by the first author. The third author checked the accuracy of extraction on a sample of 22 of the 112 studies. Only one minor incorrect extraction was discovered, implying that the accuracy of data extraction was good.

Assessment of methodological quality of studies
Because most studies had an observational design, many of the criteria of a well-known and widely used tool for assessing risk of bias-The Cochrane Collaboration tool-were not applicable. Therefore an adjusted tool was used, based on a selection of the criteria of the Newcastle-Ottawa Scale adapted for cross-sectional studies (Kocsis et al., 2010) and of a checklist for the evaluation of research articles (Durant, 1994). It addresses the following aspects: selection (sample size, method of acquisition and criteria for in-and exclusion, non-response), measurement (method of measurement, reliability and validity), statistics (selection of adequate tests, dealing with confounders) and conclusions (logic, limitations). Every aspect was independently assessed by the first and third author on a three-point scale (0 to 2 points), resulting in a maximum score of 18 points. When scores of both raters differed at least three points, the scores on every criterion were assessed on the basis of consensus (12.6% of the quality scores had to be discussed this way). Total-score differences less than three were averaged. The interrater reliabilities were good to excellent, according to the Intra Class Correlation Coefficients (two-way random effects model, absolute agreement) for the independently scored quality-scores: ICC = .71 (single measure)/.83 (average measure).

Measures
God representation scores were categorized into three groups, consisting of in total six dimensions, based on theoretical distinctions. The first group contained all attachment to God measures, measuring the way the person feels and acts regarding his attachment-based relationship with God. Within this group, three types of measures were distinguished: (1) secure attachment to God (a mix of measures with only secure attachment items and measures with secure and insecure items, placed on one dimension; (2) anxious attachment to God; (3) avoidant attachment to God. The second group of measures is called positive/negative God representations and focuses on the way a person perceives or affectively experiences God; here every measure is reduced to a (4) positive image of God or a (5) negative image of God. The third type of measure, (6) God control measures, regards the extent of control, influence, or power that is attributed to God. This also includes seeing God as a judging/punishing God, as far as it is not taken personally. For the adjustmental aspects of psychological functioning, measures of (1) well-being/adjustment and of (2) distress were chosen. For well-being, studies with a variety of measures have been selected, such as satisfaction (of work, body, marriage, etc.), adjustment (to work, or after trauma), personal growth (after a crisis), therapy outcome, or general measures of well-being. For distress, also studies with a wide range of measures have been used: general distress, anxiety, depression, dissatisfaction, state-anger, etc.
The selection of measures of dispositional aspects of psychological functioning was based on attachment theory and object relations theory. For (1) self-concept, studies with measures of selfconcept and locus of control were selected. For (2) relationships with others, studies with measures of object-relational functioning and interpersonal attachment (partners, parents, friends) were selected. All scores were treated as either secure/positive or insecure/negative representations of self and others. The link with affect regulation was established by selecting studies that measured (3) worrying, and the Big Five dimension neuroticism (negatively); or disposition measures of hope and optimism (positively). In Table 2 we listed the type(s) of measures we extracted from each study.

Pearson's correlation coefficient as effect size
The majority of selected studies (90%) reported the Pearson correlation coefficient for the associations between God representations and the dispositional or adjustmental dimension. For studies reporting                  data in other formats and for which we did not obtain correlation coefficients from the authors, data were transformed using standard meta-analytic calculations (Borenstein et al., 2005). These scores were then imported in the software program for meta-analyses Comprehensive Meta-Analysis Version 2 (CMA, Borenstein, Hedges, Higgins, & Rothstein, 2014), leading to 30 possible outcomes per study: six types of God representation measures x five other measures (two types of adjustment measures + three types of disposition measures). In the present meta-analysis, this resulted in 348 effect sizes from 123 independent samples (average of 2.83 effect sizes (ES's) per sample). Effect sizes were assigned a positive value if they were consistent with the a priori predictions, and a negative value if they were inconsistent with the a priori predictions. All analyses for the present study were performed using the CMA software. Following Cohen (1988), correlations of .10 to .29 are considered as small effect sizes, correlations of .30 to .49 as medium effect sizes, and correlations of at least .50 as large effect sizes.

Calculations of effect sizes on three levels
We calculated effect sizes on three levels of varying abstraction. On the first level, we examined the associations of undifferentiated God representations with respectively undifferentiated adjustmental and undifferentiated dispositional aspects. For calculating effect sizes on this level, multiple correlations per individual study were averaged, to meet the statistical assumption of independence required for meta-analysis. In doing so, we followed standard meta-analytic procedures (Horvath & Symonds, 1991;Martin, Garske, & Davis, 2000).
On the second level, we examined the associations between the six dimensions of God representations and undifferentiated adjustmental measures and the associations between the six dimensions of God representations and undifferentiated dispositional measures, 12 (6 x 2) effect sizes in total.
On the third level, we examined associations between each dimension of God representations and the subdomains of the adjustmental aspects (well-being and distress) and associations between each dimension of God representations and the subdomains of the dispositional aspects (self-concept, relationships with others and neuroticism), 30 effect sizes in total.
For determining the significance of the effect sizes, we lowered the usual 5% level of significance to 0.1% (p = .001) because we calculated 42 (12 + 30) separate effect sizes. This correction was aimed at diminishing the risk of type I errors ("false positives") given the large number of separate tests.

The random-effects model
Calculations of effect sizes were based on the random-effects model, because we expected the true effect size to vary between studies due to varying measures, used within very different populations under various circumstances. This has its effect on the weight assigned to each individual study as a function of the within-study variance.

Heterogeneity analysis
Heterogeneity was examined by inspecting several aspects of the aggregated effect sizes, using forest plots. Differences in effect sizes between individual studies were examined for the presence of heterogeneity using the Q B statistic, and the I 2 -value, which is a measure for the proportion of the total variance that can be addressed to these real differences. For an interpretation of I 2 , the Cochrane website offers the following rules of thumb: 0%-40%: might not be important; 30-60%: may represent moderate heterogeneity; 50%-90%: may represent substantial heterogeneity; 75%-100%: considerable heterogeneity. We considered I 2 -values of 50% and higher as an important indication for the need to examine sources of heterogeneity. However, it should be emphasized that this measure is a relative measure, giving no indication of the absolute magnitude of the heterogeneity, which is better represented by the T-value. This is the standard deviation of the aggregated effect size, which is in the same scale as the chosen measure for all effect sizes: the correlation coefficient (Borenstein et al., 2005). Therefore we considered the heterogeneity of effect sizes with T < 0.10, regardless of the I 2 -value, also as not substantial.

Examining sources of heterogeneity
On both levels of analysis, we used subgroup analyses and meta-regression analyses to examine potential sources of heterogeneity, thereby simultaneously testing our hypotheses.

Subgroup analyses
Our hypotheses are about differences in aggregated effect sizes, caused by differences between dimensions of God representation or caused by differences between subdomains of adjustment or disposition measures (lower level). These differences were examined by subgroup analyses based on the fixed-effects model, as this is the common approach (Cuijpers, 2016).
When examining these differences between subgroups for explaining heterogeneity, studies that had outcomes for both subgroups were excluded, to avoid violating the assumption of independence. This often led to the exclusion of many available effect sizes. Only for a few subgroup analyses, if independent comparisons were impossible, we used all available effect sizes, treating them as independent.

Moderator analyses
For examining the possible effects of moderator variables, meta-regression analyses were conducted on the two highest levels. With these analyses, the influence of three continuous variables (year and quality of study and quality of God representation measures) and of six categorical variables (respondent status, method of measurement, religiosity, religion/denomination, sex, and age) were established.
We included categorical variables for analyses if a variable had at least two categories with four or more studies for the subgroup. This broad approach was chosen to be able to detect potential differences in a majority of the small subgroups.

Publication bias
In meta-analyses there is always the risk of overestimating the strength of the combined effect size because of the well-known "file-drawer effect" (Thornton & Lee, 2000), implying that nonsignificant findings, which are more likely in small studies, are less likely to be published. Therefore it is important to check if small studies with relatively small effect sizes are underrepresented in meta-analyses. A useful method for examining this is looking at the funnel plot. An indication for publication bias are "missing" studies at the lower-left corner of the plot. These "missing" studies are the (smaller) studies with lower standard errors and with lower effect sizes. A more quantitative approach to checking publication bias is by simulating a meta-analysis that corrects for potentially missing effect sizes by making the funnel plot symmetrical and comparing the simulated with the observed results. This is done with Duval and Tweedie (2000) trim and fill analysis. We conducted these trim and fill analyses on all three levels.
To test the robustness of the found effect sizes, we did Orwin's (1983) fail-safe analyses on the first level. With these analyses, we calculated how many studies with a correlation of r = 0 would be needed to lower the found effect size to r = .10, the usually found association between religiosity and well-being/mental health. On the third level, we also examined the robustness of the significant effect sizes of the associations of specific God representations with well-being and distress with r > .20, because they are based on much smaller numbers of studies. Table 3 summarizes the distribution of studies/samples and separate effect sizes across the categories of the moderator variables of all the studies in this meta-analysis. The distributions of studies across the 12 subgroups used in subsequent analyses are shown as well. Table 4 shows the results of the meta-analysis on all three levels of analysis.

Analyses on level 1
The effect size of the association between undifferentiated God representations and undifferentiated adjustmental aspects of psychological functioning was highly significant, r = .196, and approximated the expected effect size of r = .20, as stated in hypothesis 1. We compared this result with a new computation in CMA of Bergin's (1983) studies, which yielded a nonsignificant effect size of r = .072. A test of the difference between the two effect sizes was significant, Q = 5.481, p = .019. Comparing our results with those of Hackney and Sanders (2003), their overall effect size of r = .10, CI 95% [.10, .11] differed significantly from our average effect size, as the not overlapping confidence intervals indicated. At last we compared our results with the meta-analytical outcome of Smith et al. (2003), who found a random-effects weighted average effect size of r = −.096, CI 95% [−.011, −.08]. Converted to positive values, this r = .096, CI 95% [.08, .11] differed significantly from our r = .196, indicated by the clearly not overlapping confidence intervals.
The association between undifferentiated God representations and undifferentiated dispositional aspects was also highly significant, r = .242, as expected by hypothesis 2.
The substantial or considerable heterogeneity of both effect sizes asks for further examination. At the next level, we aim at finding sources of heterogeneity in the differences between the various God representation dimensions in the strength of associations with adjustmental and dispositional measures.

Associations of differentiated God representation measures with undifferentiated adjustmental aspects
Five out of six dimensions of God representations had highly significant associations with undifferentiated adjustmental aspects of psychological functioning (well-being/distress). Anxious attachment to God and negative God representation, with effect sizes of respectively r = .263, and r = .218, had the strongest associations with well-being/distress, in accordance with hypothesis 1, which expected effect sizes > .20. The highly significant associations of positive God representation, r = .194, and secure attachment, r = .189, with well-being/distress were just below the expected strength. The highly significant association of avoidant attachment, r = .099, and the not significant association of God control, r = .068, with well-being/distress were much lower. From the significant associations with wellbeing/distress, the heterogeneity for the association with avoidant attachment to God-according to I 2 -was substantial, but the standard deviation of the effect size was low (T = .076), indicating that differences between effect sizes of individual studies were relatively small. The heterogeneity of the significant effect sizes for the associations between the other God representation measures and well-being/distress was still considerable, asking for further analyses for its potential sources.
The omnibus test for subgroup analysis (see Table 5) detected no significant differences between the effect sizes of the six subgroups of God representation measures. Because there were no studies that used only avoidant attachment to God measures in combination with adjustmental aspects, we could not test these differences by treating the effect sizes as dependent. Therefore we did this subgroup analysis again, treating all available 167 effect sizes as independent. Now the omnibus test yielded highly significant differences between effect sizes, and results of post hoc analyses showed that the associations of God Control and avoidant attachment to God with undifferentiated adjustmental aspects (well-being/distress) were significantly lower than the associations of the other God representation measures with well-being/distress.

Associations of differentiated God representation measures with undifferentiated dispositional aspects
Nearly all effect sizes of the associations between the dimensions of God representations and undifferentiated dispositional aspects were significant, as expected (hypothesis 2). Only the association between God Control and dispositional aspects was not significant. The associations of secure and anxious attachment to God with dispositional aspects had the strongest effect sizes, r = .307 and r = .307, respectively, followed by positive God representation and negative God representations, that had effect sizes of respectively r = .224, and r = .198, for their associations with dispositional aspects. The weakest associations with dispositional aspects were found for the God representation dimensions avoidant attachment to God, r = .159, and God Control, r = .084.
Heterogeneity, based on I 2 , was substantial for the association of dispositional aspects with secure attachment to God, and it was considerable for the association with the other five God representation measures. Only the effect size of the association of dispositional aspects with avoidant attachment to God had a low standard deviation (T = .080), indicating that differences between effect sizes of individual effect studies were relatively small. Sources of potential heterogeneity must be examined for the association of the other God representation dimensions with dispositional aspects.

Subgroup analyses
The omnibus test for subgroup analysis (see Table 6) detected no significant differences between the effect sizes of the six subgroups in their associations with undifferentiated dispositional aspects. To examine the potential difference between avoidant attachment to God versus other God representation dimensions in their associations with dispositional aspects, we used all 181 effect sizes in a new subgroup analysis by treating them as independent. Results of post hoc analyses showed that the association between God control and undifferentiated dispositional aspects was significantly lower than the associations of the secure and anxious attachment to God dimensions and of the positive God representations dimension with undifferentiated dispositional aspects. The associations of the negative God representations dimension and of avoidant attachment to God with the undifferentiated dispositional aspects were significantly lower than the associations of secure and anxious attachment to God with the undifferentiated dispositional aspects.

Analyses on level 3
3.4.1. Associations between differentiated God representations and differentiated adjustmental aspects 3.4.1.1. Associations of God representations dimensions with well-being. Four out of six God representation dimensions were highly significantly associated with well-being. Secure and anxious attachment to God and positive God representations had the strongest associations, with r > .20, as expected (hypothesis 1). The negative God representation dimension had an association with well-being less than r = .20. The associations of avoidant attachment to God with well-being and of God Control with well-being were non-significant.
Heterogeneity of the significant effect sizes was very low for the association of well-being with secure attachment, according to I 2 and T. For the association with positive God representations it was substantial, but T was smaller than 0.10, indicating that differences between individual effect sizes were relatively small. For the associations of well-being with anxious attachment to God, with positive God representations, and with negative God representations, heterogeneity was considerable or substantial.

Associations of God representations dimensions with distress.
From the associations of the six God representation dimensions with distress, only the dimensions anxious attachment to God and negative God representations were significantly associated with this adjustmental aspect with r > .20, as expected (hypothesis 2). The dimensions avoidant attachment to God and positive God representations were also significantly associated with distress, but here r < .20. The associations of secure attachment to God and God Control with distress were non-significant. Heterogeneity of the significant effect sizes was considerable for the associations of anxious attachment to God, negative God representations, and positive God representations with distress. According to I 2 , heterogeneity was moderate for the association between avoidant attachment to God and distress, with T < 0.10, indicating that this effect size might be a rather precise estimate.
3.4.1.3. Subgroup analyses. Results of subgroup analyses (see Table 7) confirmed significant differences in strength of the associations between well-being and distress on the one hand and the positive and negative God representation dimensions on the other. The positive God representation dimension had significantly stronger associations with well-being than with distress; the negative God representation dimension had significantly stronger associations with distress than with well-being. There were no significant differences between well-being and distress regarding their associations with attachment to God measures.

Associations between differentiated God representations and differentiated dispositional aspects
3.4.2.1. Associations of God representation dimensions with relationships with others. As shown in Table 4, five of the six associations of God representation dimensions with relationships with others were highly significant; only the association of God Control with relationships with others was nonsignificant.
Heterogeneity of the associations was considerable or substantial. The associations of avoidant attachment to God and of the negative God representation dimension with relationships with others had standard deviations of T < 0.10, suggesting valid estimates. 3.4.2.2. Associations of God representation dimensions with self-concept. Three out of six associations of self-concept with the God representation dimensions were (highly) significant: secure attachment to God, anxious attachment to God and positive God representations. Heterogeneity of the effect sizes of all three significant associations was considerable.
3.4.2.3. Associations of God representation dimensions with neuroticism. All six God representation dimensions showed significant associations with neuroticism. Heterogeneity of the aggregated effect sizes was low for the association of secure attachment to God and of God control with neuroticism. It was substantial or considerable for the association of positive God representations and anxious attachment to God with neuroticism. All in all, on the third level all associations were positive, and 73% of the associations were significant at the p = .001 level. From these significant associations, 82% still had substantial or considerable heterogeneity, to be examined further with moderator analyses.

Subgroup analyses.
For the associations with secure attachment to God and God control, studies with measures of the dispositional dimension did not meet the criterion of at least two categories with at least four studies. For the other four God representation dimensions, none of the differences in strength of associations between dispositional aspects and God representations was significant (see Table 8).

Publication bias
To check whether small studies with relatively small effect sizes were underrepresented in these meta-analyses, we generated two funnel plots (see Figure 2), based on separate meta-analyses for the associations between undifferentiated God representation measures and undifferentiated For the effect sizes of God representations with adjustmental aspects, Duval and Tweedie's trim and fill analysis indicated that there were seventeen "missing" studies at the left side of the mean. Inspecting the funnel plot showed that these missing studies were distributed equally over the standard error axis, so there was no overrepresentation of "missing" studies at the lower-left corner of the plot (representing the smaller studies with lower standard errors and with lower effect sizes). According to Orwin's fail-safe analysis, it would take 63 studies with a correlation of r = 0 to lower the aggregated effect size (based on 73 studies) to r = .10 (the generally found association between religiosity and well-being).
For the effect sizes of God representations with dispositional aspects, based on 87 studies, inspection of the funnel plot and Duval and Tweedie's trim and fill analysis indicated that there were no "missing" studies at the left side of the mean. It would take 128 studies with a correlation of r = 0 to lower the aggregated effect size to r = .10.
Therefore, there were no indications of publication bias for the aggregated effect size of undifferentiated God representations with undifferentiated state aspects and with undifferentiated trait aspects, so the estimate of the associations was sufficiently robust.
Because specific God representations on the second and third level of analysis differed in their associations with state and trait aspects from the associations on the first level, we also checked for "missing" effect sizes at the left side on the second and third level and if positive, looked at their distribution accross the standard error axis. Overall, these funnel plots also yielded no indications of publication bias. Only three of the 30 associations on level 3 had a slight underrepresentation of small studies with low effect sizes, with for two of them (the associations of anxious and avoidant attachment to God with well-being) two out of three missing studies with low precision, and for the third (the association between the positive God representation The open diamondshaped symbol represents the overall effect size based on observed effect sizes, the black symol represents the estimated effect size affter imputing the "missing" studies. dimension and well-being) only one missing study, placed at the low precision part of the standard error axe (see Table 4).
At last, we did Orwin's fail-safe analyses on the third level to examine the robustness of the significant effect sizes of the associations of specific God representations with well-being and distress with r > .20, because they are based on much smaller numbers of studies. We again checked the robustness by calculating how many studies with a correlation of r = 0 would be needed to lower the found effect size to r = .10. For the association between secure attachment to God and well-being (based on 5 studies) this would take 10 studies; for the association between anxious attachment to God and well-being (based on 16 studies) it would take 19 studies; for the association between anxious attachment to God and distress (based on 24 studies) it would take 48 studies; for the association between the positive God representations dimension and well-being (based on 12 studies) it would take 25 studies; and for the association between the negative God representations dimension and distress (based on 26 studies) it would take 31 studies with r = 0 to lower the aggregated correlation to r < .10. We consider the results at level three to be sufficiently robust, because also for these associations there were no indications of publication bias.

Moderator analyses
Moderator analyses were performed to further examine heterogeneity. On the third level, the subgroups were too small to do these analyses. In fact, this was also the case for many combinations on level 2. For reasons of limited space, we report only the results of analyses on level 1. On this level, Religion/denomination and Year of study were the only factors that explained some variation.

Religion/denomination
There was a significant effect of religion/denomination on the association between God representations and adjustmental aspects, explaining 12% of the total between-study variance (see Table 9). Post hoc analyses of the differences revealed that mixed religious/not religious samples showed lower associations between God representations and adjustmental aspects, than the orthodox, evangelical/Baptist and mainstream Christian samples. There was also a significant effect on the association between God representations and dispositional aspects, explaining 9% of the total between-study variance. The association was significantly stronger for Evangelical/Baptist and for mixed Christian/other religions samples than for mainstream Christian and for mixed religious/not religious samples.

Year of study
The effect of the continuous moderator variable year of study on the associations between God representations and dispositional measures (see Table 10) was significant, explaining 9% of the total between-study variance. More recent studies showed stronger associations.
Taken together, most of the substantial or considerable heterogeneity of the effect sizes could not be explained by the selected moderator variables.

Discussion
The main aim of this meta-analysis was to examine associations between various dimensions and aspects of religiosity, in particular, God Representations, and mental health, from the perspective of attachment theory and object-relations theory. The meta analysis was based on 123 studies with one or more associations between God representations and adjustmental or dispositional aspects of psychological functioning, resulting in 348 effect sizes, of in total 29,816 participants. The most important finding is that medium-sized associations were found for the associations between dimensions of God representations and well-being and distress, as well as for the associations between God representations and self-concept, relationships with others and    neuroticism. These associations are much stronger than those generally reported in studies adopting unidimensional and behavioral measures of religiousness. Because there were no signs of publication bias and the results, based on Orwin's (1983) fail-safe analyses, were sufficiently robust, the effect sizes reported in the current meta-analysis may be considered as valid estimates of the examined associations.

God representations and adjustmental psychological functioning
The results of this meta-analysis predominantly confirmed the first hypothesis: the effect sizes for the association between God representations and measures of well-being/distress were in the expected directions, and the aggregated effect size, r = .20, had the expected strength. It was also significantly stronger than the meta-analytical outcomes from Bergin (1983); Hackney and Sanders (2003); Smith et al. (2003) for the associations between religiosity and well-being/distress. To our knowledge, this is the first study that demonstrates with a meta-analysis such robust associations of structural aspects of religion with well-being and distress. It indicates that the concept of God representation is an important mediating factor in the association between monotheistic religiosity and well-being/mental health and distress. The results are in line with the notion of many scholars in the religious domain, often referred to as relational spirituality, that the relational character of monotheistic religions, the experienced personal relationship with the divine, is a central factor of those religions (Davis et al., 2018;Davis, Hook, & Worthington, 2008;Hall, 2007a;Hill & Hall, 2002;Leffel, 2007aLeffel, , 2007bSandage & Williamson, 2010;Simpson, Newman, & Fuqua, 2008;Verhagen & Schreurs, 2018).

Difference between positive and negative God representations in their associations with well-being and distress
The highly significant findings that positive God representations were more strongly associated with well-being than with distress (and vice versa for negative God representations) clearly demonstrates the complexity of religious/spiritual functioning. Results suggest that they are not just two opposite poles of the same dimension, but should be considered as two different aspects of God representations. Gibson (2008) recognizes this ambiguity with regard to God representations. He emphasized the existence of multiple cognitive schemas for God in one person. These findings also undergird object-relations theory explanations of God representations. This theory made invaluable contributions to the understanding of these phenomena with its concept of integration of good and bad internalized objects. It is considered mature to attribute good as well as bad attributes to the self, to important others and to the relationship with them, and to be able to integrate them in such a way that they can exist together at the same time, to tolerate and to somehow also understand this ambiguity. Apparently, this also applies to God representations.
This notion should have consequences for the operationalization of God representations: besides their content, God representation measures should also assess more structural components as ambiguity, differentiation and integration.

God representations and dispositional psychological functioning
Results also confirmed the second hypothesis: measures of secure attachment to God and of positive God representations were positively associated with positive self-concept and positive relationships with others, and negatively with neuroticism, whereas measures of insecure attachment to God were negatively associated with positive self-concept and positive relationships with others. The aggregated effect size of r = .24 had the expected strength, and we found medium effect sizes for the associations of the dispositional measures with secure and anxious attachment to God.
To the best of our knowledge, this is the first meta-analysis focusing on the associations between God representations and dispositional measures, implying that comparisons with other meta analytic studies on this topic cannot be made. Our findings extend other influential reviews indicating that mental representations of people are associated with psychopathology (Huprich & Greenberg, 2003).

God representations and view of self and others
The results demonstrate that God representations are associated equally strongly with selfconcept, the experienced relationships with others, and neuroticism. The findings are in support of the correspondence hypothesis, demonstrating correspondence of God representations not only with the view of self but also with the experienced relationship with others. Many scholars explain the often found association between God representations and self-concept, or-more specifically -self-esteem (Benson & Spilka, 1973;Lawrence, 1997;McDargh, 1983) by hypothesizing that the God representation is merely or predominantly a projection of the the self. In the domain of attachment-theory inspired research of God representations, the emphasis is more on the perception of others, and here the correspondence hypothesis (Granqvist, 1998;McDonald et al., 2005) assumes that an insecure relationship with God corresponds with an insecure attachment to parents or adults. The observed associations of God representations with neuroticism (as an indication of the capacity for affect regulation) also corroborate theoretical explanations of objectrelations and attachment theory, which both stress the central role of internal working models in affect regulation (Fonagy, Gergely, & Jurist, 2004;Kernberg & Caligor, 1996).

Weak associations with God control
Results also demonstrated that the God control dimension had significantly weaker associations with adjustmental and dispositional aspects than the other God representation dimensions. The only significant association was the positive association between God control and neuroticism. There are several potential explanations for finding hardly any significant associations. First, it may be due to the small statistical power caused by the low number of studies that used this God representation dimension. Second, conceptual confusion about God control may also be a cause: although we aimed at choosing a rather neutral, less affective measure of beliefs about the agency of God, the specific items of questionnaires that measured God control also focussed for example on the protection by a benevolent God, or on the rejection by a judging God. Therefore the items also contained affective aspects. Third, the concept of God control may have different meanings for healthy subjects than for patient and for orthodox and non-orthodox patients. Jonker (2007) found that scores on the Questionnaire God Representations scale perceiving God's actions as ruling/punishing positively related to feelings of anxiety for God, except for non-patient members of the Orthodox-Reformed or Evangelical/Baptist denominations. The Ruling/punishing image of God was also related to positive feelings towards God, but only among non-patients. In a nonclinical sample, Eurelings-Bontekoe, Hekman- Van Steeg, and Verschuur (2005) found this particular concept of God to be rather independent of personality and attachment variables. Therefore the ruling/punishing image of God can be viewed as a double-edged sword (Johnson, Li, Cohen, & Okun, 2013). In future research, in operationalizing the God control dimension it might be important to pay more attention in formulations of items to the distinction between the concept of "God as a judge" both as a non-affective, rather doctrinal phenomenon as well as an affect laden God representation. In addition, it is also important to be aware of differences in interpretation of this concept between adherents of various denominations, and between patients and non-patients.

Moderator analyses
Although subgroup analyses demonstrated some significant differences that enhanced our insight in the associations between God representations and adjustmental and dispositional aspects, they did not contribute much in explaining and reducing statistical heterogeneity. Moderator analyses for the effect of religiosity, religion/denomination, sex, age, year of study, and quality of study and of God representation measures also could not explain the heterogeneity of most effect sizes. With our broad approach, including all studies that reported associations between God representations and adjustmental or dispositional aspects, this was to be expected. Yet, the heterogeneity of these findings remains to be explained.
Undoubtedly, different measures for similar concepts, and different samples, caused much heterogeneity that could not be incorporated as study-level variables and thus could not be explored. Therefore, although we consider the found effect sizes to be valid and robust, future research should aim to explain the remaining heterogeneity in most of the associations.
Attachment-and object-relations theory, with their emphasis on implicit working models, implies that assessment of God representations should (also) focus on implicit aspects thereof. To note, in nearly half of the studies of this meta-analysis, authors mentioned the use of selfreport instruments as a limitation, and half of them thereby pointed at the specific nature of unconscious processes that asked for implicit measurement. However, remarkably, only one study in our meta-analysis used an implicit measure of God representations, and only five studies used other than self-report measures for dispositional or adjustmental aspects. Therefore, the potential important influence of this moderator factor could not be established well.
The notion that the presence of (more severe) psychopathology might moderate the general associations between religion and well-being/mental health or distress, as suggested by the outcomes of meta-analytic studies about the associations between religion and well-being, could also not be established because of a lack of studies that focus on God representations in clinical samples.

Clinical implications
An important issue is the clinical significance of the statistically significant results of this metaanalysis. The strongest associations in this meta-analysis, the association between the positive God representations dimension and well-being and the association between anxious attachment to God and distress, have medium effect sizes (for both r = .30). If God representations on a general level have this association with well-being and distress, it should have clinical implications. Approximately half of the world population has a theistic belief (Hackett et al., 2012). The World Psychiatric Association officially stated that "A tactful consideration of patients' religious beliefs and practices as well as their spirituality should routinely be considered and will sometimes be an essential component of psychiatric history taking" (Moreira-Almeida, Sharma, van Rensburg, Verhagen, & Cook, 2016). Therefore it is important in clinical intakes to systematically address religion and to pay attention to God representations among patients with a theistic belief. If this is done by self-report questionnaires, results of this meta-analysis indicate that it is important to use questionnaires that treat secure and insecure attachment to God and positive and negative God representations as separate dimensions. Otherwise potential negative God representations, associated with mental health problems, might be overlooked and neglected.
Of course, the relevance of this distinction is dependent on the course of therapeutic treatment. In line with popular trends as positive psychology and solution-focused therapy, the focus in therapy may lie on strengthening a positive God representation, thereby avoiding focusing on negative God representations. However, in a discussion of various modern spiritual approaches to mental Leffel (2007aLeffel ( , 2007b warns for "simple spirituality" that seems to assume that just focusing on positive feelings and positive thinking will make the negative emotions go away, while ignoring the implicit nature of representations. In his view, deep and lasting spiritual (and resulting personality) transformations are possible by focusing on disclosure and integration of negative emotions, directed at changes in the affective implicit and procedural structures of personality. This should be related to a focus on character change and the development of virtues; not on well-being or happiness, instrumentally fostered by religion or spirituality. Our results suggest the importance of focusing in therapy on negative as well as positive God representations.

Limitations
This meta-analysis has several limitations that need to be mentioned when interpreting the results. First, an important limitation, implied by the choice for God representation measures, is our reduction of religion/spirituality to theistic religions. Though in our search we looked at samples from all theistic religions, our final selection contained only two samples with predominantly other than Christian (namely Jewish) subjects. This does not mean that our results are based only on adherents of Christian religions: the study contains 10 samples with a mix of Christian subjects and subjects that adhere to other religions, and 33 samples are a mix of religious and non-religious subjects. Therefore it should be kept in mind that other than Christian religions are underrepresented in this study, which in turn limits the generalizability of the results. A second limitation is the quality of the included studies. Results are based on observational data of predominantly cross-sectional studies, which precludes any conclusions about the direction of the found associations. Third, this meta-analysis is based on published articles only. Although we found no indications of publication bias in our selection of studies, analyses are not based on all potentially available data. Fourth, in this meta-analysis much of the considerable or substantial heterogeneity of the effect sizes could not be explained, meaning that there is still much variation of true effect sizes. Fifth, a limitation is that we categorized the different measures of God representations into six dimensions, thereby ignoring more subtle differences. For example, we did not distinguish more specific negative God representations such as feeling anxious or being mad at God or seeing God as distant, while it seems reasonable that these differences are associated with different personality traits. There is some evidence that these differences are distinctively associated with types of religious struggle (Exline, Grubbs, & Homolka, 2015). Sixth, it must be noted that part of the association between God representations and adjustmental aspects may be the result of a specific same-method effect; the linguistic similarities in God representation items and adjustment-measure items as anger, fear, frustration, etc. More research is needed in this area to clarify these issues. Seventh, a limitation of this meta-analysis is the low number of studies with clinical samples, with samples with subjects with serious life problems, and with implicit measures.

Future research
A meta-analysis with analyses only at study-level variables is not a suitable method for testing pathways between the variables of a model. As a consequence, we cannot give conclusive answers about the nature of the examined relations. Nevertheless, results of this meta-analysis suggest that there may be some direct influence of God representations on well-being and distress that is relatively independent of religious denomination, respondent status (serious life problems or mental health problems), sex or age. It is unclear, however, whether and to what extent God representations impact psychological functioning through an experienced "real" relationship with the God object that may also alter the self-concept, rather than through a mere projection of the self.
Further, to examine causal relationships between God representations and adjustmental aspects and the mediating role of dispositional aspects, is it important to conduct longitudinal studies, ideally examining development from early childhood to adulthood. A major advance would be if meta-analyses could be conducted by synthesizing the available data on respondent level, to be able to examine the pathways and the best fitting model to explain the complex interrelations between the different variables. We recommend the development of systems to be able to aggregate data on this level, and we welcome the development of a scientific culture that makes this possible.
Theoretically, it is assumed that implicit aspects of God representations, especially for subjects that suffer from external stressors such as serious life problems, or from internal stressors such as personality problems, have an important influence on their psychological functioning. This metaanalysis demonstrated two important gaps in this respect. First, there is a lack of studies that examine associations between God representations and well-being/mental health for subjects that suffer from mental health or serious life problems. Future research should take this into account by examining these associations for samples with various mental health problems (under which particularly personality disorders) and samples of subjects undergoing various serious life problems. Second, there are hardly any studies that measure associations of implicit God representations with well-being/mental health. It is unknown if and to what extent discrepancies exist between scores on explicit and implicit measures of God representations, and if these discrepancies differ between healthy and pathological or otherwise seriously stressed subjects. Hall and Fujikawa (2013) assume that different attachment styles are related to specific discrepancies between explicit and implicit God representation measures. We subscribe their statement that advances in the field of God representation research are dependent on the development of implicit God representation measures to examine these discrepancies. Therefore future research should take this into account by examining and comparing explicit and implicit God representations and their associations with adjustmental and dispositional aspects in both clinical and non-clinical samples.
A first step is the development of a reliable and valid instrument for measuring implicit God representations. This meta-analysis is part of a project in which such an implicit measure has been developed and is being validated in both a non-clinical and a clinical sample (Stulp, Glas, & Eurelings-Bontekoe, 2019;Stulp, Koelen, Glas, & Eurelings-Bontekoe, 2018).

Conclusions
This meta-analysis has clearly demonstrated the importance of God representations for research on the association between religiosity and well-being/mental health, at least for adherents of a theistic religion. We demonstrated that narrowing down the general concept of religiosity to specific measures of God representations resulted in stronger associations with well-being and mental health than previously reported. We also demonstrated that object relations and attachment theory may be fruitful approaches in potentially explaining the mechanisms behind this association.