Varieties of Religious Attachment y 1 Running head: VARIETIES OF RELIGIOUS ATTACHMENT
Varieties of Religious Attachment: Working Models in Semantic Space Kevin S. Reimer
Alvin C. Dueck
Azusa Pacific University
Fuller Graduate School of Psychology
Garth Neufeld Alliant International University Sherry Steenwyk
Fuller Graduate School of Psychology
Author Note Kevin S. Reimer, Department of Graduate Psychology, Azusa Pacific University, Azusa, California. Alvin C. Dueck, Fuller Graduate School of Psychology, Pasadena, California. Garth Neufeld, Alliant International University, Fresno, California. Sherry Steenwyk, Tracy Sidesinger, Fuller Graduate School of Psychology, Pasadena, California. This research was supported by a grant from Metanexus Institute/John Templeton Foundation to the first two authors. We thank M. Kyle Matsuba for suggestions to improve the manuscript. Correspondence regarding this article should be addressed to Kevin S. Reimer, Department of Graduate Psychology, Azusa Pacific University, 901 East Alosta Avenue, P.O. Box 7000, Azusa, California 91702. Phone: (626) 815-6000 x5507. Fax: (626) 815-5015. Electronic mail: [email protected]
Varieties of Religious Attachment y 2 Abstract This study considered correspondence between working model representations of self and other in religious attachment. Given the importance of social context in the formation of working models, the project emphasized attachment language in a comparative religious design. In Study 1 (with 431 undergraduates), a global model of correspondence in religious attachment was used to test path relations between positive (e.g., secure) working models and perception of the divine. In Study 2 (with 20 nominated exemplars from Jewish, Muslim, and Christian traditions, respectively), socialization influences in correspondence were explored by unsupervised comparison of working model representations for each religious tradition in a computational semantic space. In general, correspondence proved a better predictor of religious attachment than compensation. When understood within a social intelligence framework, exemplar working models from all three religious traditions reflect socialization patterns integrating semantic and episodic knowledge of self with the divine and peers.
[1,081 characters with spaces]
Keywords: correspondence, internal working models, religious attachment, semantic space, social intelligence
Varieties of Religious Attachment y 3 Varieties of Religious Attachment: Working Models in Semantic Space It was William James in The Varieties of Religious Experience (1902/2002) who famously identified the tendency of people to describe God using the language of interpersonal relationship. More recently, researchers have turned to attachment theory as a means to understanding religious perceptions in relational context (Birgegard & Granqvist, 2004; Granqvist, 1998; Granqvist & Hagekull, 2000; Granqvist & Kirkpatrick, 2004; Kirkpatrick, 1998, 1999; Kirkpatrick & Shaver, 1990, 1992; Rowatt & Kirkpatrick, 2002). Measures of attachment quality have been applied to perceptions of God with the general result that securely attached individuals experience the divine as more loving, less distant, and less controlling than insecurely attached people (Granqvist & Kirkpatrick, 2004; Kirkpatrick, 1998, 1999; Rowatt & Kirkpatrick, 2002). One explanation for this finding is that people perceive themselves in a relationship with God analogous to human attachment figures. However, the argument for correspondence between human and divine attachments is complicated by evidence of compensatory attachment behaviors for individuals with negative working models of self and other. Individuals with insecure attachment styles report more conversion experiences and increased attendance at religious services relative to securely attached people (Birgegard & Granqvist, 2004; Granqvist, 1998; Granqvist & Hagekull, 2000; Granqvist & Kirkpatrick, 2004; Kirkpatrick & Shaver, 1990). Efforts to clarify these findings include examination of socialization influences through parental attachment figures along with religious conversion experience (Birgegard & Granqvist, 2004; Granqvist, 1998; Granqvist & Hagekull, 2000; Granqvist & Kirkpatrick, 2004; Kirkpatrick, 1997, 1999). It is possible that underlying differences in working models of self and other are, at least in part, responsible for this complex picture of religious attachment. Working
Varieties of Religious Attachment y 4 models mediate the adaptive potential of relationships within a goal framework (Bowlby, 1980; Bretherton, 2005; Collins & Allard, 2001; Collins & Feeney, 2000, 2004). Further work on religious attachment should explore how working models are implicated in perception of the divine, including socialization differences by religious tradition. The primary objective for the present research is to evaluate underlying differences in working models toward a fuller, more comprehensive account of personality in religious experience. The two studies reported here consider correspondence in religious attachment through secure working model representations of self and other (Study 1) with the intent of mapping working models to evaluate socialization influences (Study 2). Thus, we examine the nature of correspondence between human and divine attachments for individuals from three world religions (Judaism, Islam, and Christianity). Prior to describing the studies in greater detail, we review research on adult attachment, religion, and working models. Adult Attachment Theory and Religion Adult attachment theory is understood in terms of a behavioral system evolved to facilitate adaptive responses to environmental threats or stress (Bowlby, 1969, 1973, 1980). Individual differences in threat responses are predicated upon attachment styles that reflect behaviors and affect organized from experience with past attachment figures (Ainsworth, Blehar, Waters, & Wall, 1978; Collins & Feeney, 2004; Mikulincer, Dolev, & Shaver, 2004). Attachment styles are widely believed to reflect underlying differences in working models of self and other. Working models are representational schemas that integrate knowledge of self and other along with perceived expectations and goals associated with social networks (Bowlby, 1980; Bretherton, 2005; Collins & Feeney, 2004). Working model schemas tend to function beneath awareness, demonstrating a degree of automaticity in behavior (Collins & Read, 1994).
Varieties of Religious Attachment y 5 Presumably, working models can be extended to include transcendence in the event that God functions as a kind of surrogate attachment figure (Ainsworth, 1985; Granqvist & Kirkpatrick, 2004; Kirkpatrick, 1998). In this view, the representational structure of the working model functions as a stable referent for interpersonal relationship that includes the divine. Working models are characterized by four attachment styles or prototypes based on two dimensions (Brennan, Clark, & Shaver, 1998; Collins & Feeney, 2004). Dimension 1 reflects a continuum of behaviors measured in terms of anxiety. Individuals with high anxiety reflect attachment histories where experiences of ambivalent or inconsistent care are incorporated into working models of self and other. Dimension 2 reflects the extent to which individuals demonstrate avoidance in relationships. People with high avoidance harbor distrust, anticipating rejection or retributive behavior from others. Accordingly, four attachment prototypes represent individual differences as measured along these two dimensions. Fearful-avoidant people are high on both anxiety and avoidance dimensions. They evidence negative perceptions of self and other. Dismissing-avoidant individuals score low on the anxiety dimension with high avoidance. Working models for these people reflect positive self representations with negative perceptions of others. Preoccupied persons have high anxiety with low avoidance. Working models for these individuals tend to be negatively oriented toward the self, but positive toward others. Secure or positively attached individuals score low on both anxiety and avoidance dimensions. Their working models reflect a generally positive view of self and other. Studies of religious attachment feature lively debates regarding the nature of transcendent motivation in the behavioral system. Are perceptions of the divine primarily characterized by compensation for attachment insecurity or correspondence between divine and human attachment figures? The compensation hypothesis for religious attachment emphasizes God as
Varieties of Religious Attachment y 6 mediating distress for persons with insecure (e.g., fearful-avoidant, dismissing-avoidant, preoccupied) attachment styles, often through dramatic conversion experiences (Granqvist & Kirkpatrick, 2004; Kirkpatrick & Shaver, 1990). Evidence for compensation was found in heightened religiosity for single adults as compared with individuals in romantic partnerships (Granqvist & Hagekull, 2000). Compensation was also noted in a well-known longitudinal survey of religious change in college students (Kirkpatrick, 1997). The correspondence hypothesis suggests that individuals with secure attachment have working models that promote positive responses between human and divine attachment figures, and reflect religious socialization processes associated with the social network. Correspondence is noted in a variety of studies beginning with seminal work on religious belief by Kirkpatrick and Shaver (1992), and more recently in parental socialization and subliminal priming research (Birgegard & Granqvist, 2004; Granqvist & Hagekull, 1999). The evidence suggests that both insecurely and securely attached individuals demonstrate proximity seeking behaviors with the divine, but for different reasons. In an effort to streamline this expanded account of religious attachment, the two hypotheses were aligned in a complementary fashion where compensation accounts for longitudinal attachment history and correspondence reflects contemporary adult social networks (Granqvist & Kirkpatrick, 2004). In step with this proposal, allied research shifted to focus on conversion experiences indicative of compensation for those with insecure styles and socialization patterns anticipating contemporary correspondence in religious attachment (Birgegard & Granqvist, 2004; Granqvist & Kirkpatrick, 2004). A conclusion from this work is that, with the exception of conversion experiences and parental religious socialization, the main effects of attachment are too inconsistent to facilitate a comprehensive account of personality in religious context (Granqvist & Kirkpatrick, 2004).
Varieties of Religious Attachment y 7 This conclusion may be premature given that religious attachment studies have tended to interpret working model differences based on self-report assessments from convenience (e.g., Christian or secular) samples (Birgegard & Granqvist, 2004; Granqvist, 1998, 2002; Granqvist & Hagekull, 1999, 2001, 2003; Granqvist & Kirkpatrick, 2004; Kirkpatrick 1998). Given Bowlby’s (1969; 1980) contention that working models are intrinsic to corresponding attachment behaviors in the social network, language used to describe representations of divine other should be considered closely. Attempts have been made to apply attachment language to trait adjective God-image scales or assessment paragraphs (Kirkpatrick & Shaver, 1992; Kirkpatrick, 1998; Rowatt & Kirkpatrick, 2002), but it is unclear how fully these scales capture relational dimensions of corresponding attachments, particularly across different religious traditions. Hutsebaut and Verhoeven (1995) found that open-ended interview questions evoked richly interpersonal perceptions of the divine in participants. Response language did not closely resemble trait adjective self-report measures for God-image. Correspondence might be robustly considered through assessments employing attachment language to detail (a) experience of God’s actions as secure base, and (b) positive or negative feelings about God as secure base (Schaap-Jonker, Eurelings-Bontekoe, Verhagen, & Zock, 2002). The prevalence of convenience samples in the religious attachment literature is problematic. Bowlby’s (1969, 1980) theory of working models is premised upon reciprocity with the environment. By extension, working models are crucially related to the cultural and interpersonal priorities of particular religious contexts. Semantic associations with human and divine attachment representations are likely forged in these contexts. The nature of correspondence between human and divine attachments may vary depending on socialization structures endemic to particular religions (Shweder, Much, Mahapatra, & Park, 1997). In the
Varieties of Religious Attachment y 8 event that correspondence is the best predictor of religious attachment, socialization assessment necessitates comparative religious study of individuals with consistent, demonstrable commitments to their respective traditions. Working models emphasizing a positive view of self and other (e.g., secure attachment) offer the most straightforward means to exploring the predictive strength of correspondence understood through attachment language, and are expected to coincide with consistent religious commitment (Granqvist & Kirkpatrick, 2004). Working Models in Semantic Space How are underlying differences between working models explored in a manner sensitive to language and religious tradition? The social intelligence view of personality understands the self as mental representations reflecting semantic and episodic knowledge (Kihlstrom, Beer, & Klein, 2003; Kihlstrom & Cantor, 2000; Kihlstrom, Cantor, Albright, Chew, Klein, & Niedenthal, 1988; Kihlstrom, Marchese-Foster, & Klein, 1997). Semantic knowledge comprises abstract concepts regarding the self including values, attitudes, traits, and motives (Kihlstrom et al., 1997). Episodic knowledge includes concrete information regarding the self such as events that are situated in autobiographical narrative. In the purview of social intelligence, both elements of self-understanding are related to aspects of memory (e.g., semantic and episodic) associated with cognitive representations (Kihlstrom et al., 1997; Northoff & Bermpohl, 2004; Squire, 1992; Squire & Zola, 1998; Tulving, 1983; Vargha-Khadem, Gadian, Watkins, Connelly, Van Paesschen, & Mishkin, 1997). Following Bowlby’s (1980) ethological claim that working model representations are aligned with memory, working models might be considered as a hierarchy of representational concepts that reflect intrapsychic and interpersonal concerns (Kihlstrom & Cantor, 2000; Kihlstrom et al., 1997). If representations of self and other could be
Varieties of Religious Attachment y 9 studied on the basis of their semantic and episodic knowledge bases, it would be possible to outline working models in a manner specific to differences such as religious tradition. A recent study successfully managed these issues by employing a quantitative method to map cognitive representations of self and other related to identity in moral functioning (Reimer & Wade-Stein, 2004). To account for semantic and episodic dimensions of self-understanding, representations were cued through questions in a semi-structured interview. Narrative responses for various self-representations (e.g., actual self, ideal self, despised self, temporal self) along with others (e.g., caregivers, romantic attachment figures) were compared to one another with the use of Latent Semantic Analysis (LSA; lsa.colorado.edu; Laham, 1997; Landauer, Foltz, & Laham, 1998; Steyvers & Tenenbaum, 2005). LSA is a knowledge representation model that provides unsupervised (e.g., empirically derived) meaning comparisons between words, sentences, or paragraphs in a high-dimensional semantic space (Landauer et al., 1998). The model is premised upon a matrix decomposition technique similar to factor analysis known as singular-value decomposition (SVD). In LSA words, sentences, and paragraphs are assigned a vector as an estimate of meaning in the semantic space. To specify vector trajectories, LSA references an 11 million word corpus of first-year collegiate readers that serves as the model’s global knowledge of the world. Cosine angles between vectors quantify similarity or dissimilarity between texts, resulting in a standard covariance matrix. In the identity study, similarity of self and other representations was calculated and subjected to inferential multivariate techniques to facilitate interpretation (Reimer & Wade-Stein, 2004). Results from this study demonstrated that it is possible to construct perceptual maps of self and other representations in a social intelligence framework using narrative data. The approach might be extended to analyze similarity or dissimilarity between self and other
Varieties of Religious Attachment y 10 representations in working models. Proximities between representations reflect interpersonal histories culminating in the current organization of the attachment behavioral system. Patterns reflecting socialization processes are evident in the degree to which semantic and episodic knowledge is shared between representations in narrative. In general, we expect that greater distances between self and other representations reflect insecure attachment history (e.g., anxiety and/or avoidance) or less prominent socialization influence in the contemporary social network. Shorter distances between self and other representations are typical of secure attachment or more prominent socialization influences in the contemporary social network. The method allows cognitive representations of multiple attachment figures to be considered simultaneously. Working models are empirically derived in terms of cognitive representations ensconced within participant narrative. One advantage of this method is that it preserves semantic and episodic knowledge aspects of attachment experience in the participant’s working model. Because the LSA technique resources semi-structured response narratives, language used to describe self along with various others (e.g., human and divine attachments) can be considered within the cultural and interpersonal context of particular religions. Variations in participant conceptualization of God as a secure base are mapped in relation to social networks using multidimensional scaling (MDS; Cox & Cox, 2000). Set relations between representations imply attachments reflecting corresponding influences in the social network, identified with the use of hierarchical cluster analysis (HCA; Everitt, Landau, & Leese, 2001). If correspondence is globally supported where perceptions of the divine are construed in attachment language, differing socialization patterns in working models for participants from various religions can be quantified in a context-specific manner.
Varieties of Religious Attachment y 11 The ability to specify working models through the language of particular socialization contexts raises a potentially thorny problem. A comparative religious design favors participants that evince religious maturity and prosocial commitment as viable representatives of their respective tradition. However, different religious traditions will likely have disparate ideas of what is meant by “maturity” or what constitutes demonstrable commitment. Although prior work on religious attachment anticipates that positive (secure) working models are associated with correspondence, we cannot assume that religious maturity and prosocial commitment will always coincide with secure working models. Walker and Hennig (2004) offer a solution, indicating that individual differences in personality can be subjected to naturalistic perspectives when study variables are abstract and values-laden. In other words, people’s ordinary conceptions of an abstract construct such as morality or religiousness can be used to establish benchmarks for psychological functioning. In order to create a level playing field for the comparative study of working models between religions, naturalistic perspectives of religious maturity and prosocial commitment might be used to identify exemplar participants for study. Consideration of underlying differences for working models in semantic space is therefore deliberately descriptive and tentative, premised upon criteria of maturity and prosocial commitment defined through consensus of the particular religious traditions under study. Overview of Studies In summary, the main goal for the present research is to consider underlying differences in working models through correspondence in religious attachment. The two studies reported here are exploratory and descriptive given their focus on working models in religious attachment, evaluating how positive (e.g., secure) perceptions of self and other correspond with perceptions of the divine where the latter is understood through attachment language (Study 1). This
Varieties of Religious Attachment y 12 anticipates differences in attachment histories and socialization patterns as represented in working models for exemplars from several world religions nominated for maturity and prosocial commitment (Study 2). Socialization patterns inherent to religious attachments are likely to vary by religious tradition where multiple attachment figures are understood in context and mapped in a social intelligence view of working models. Study 1 models correspondence in religious attachment through positive (secure) working models, using attachment language to describe the divine. The argument for correspondence in working models anticipates continuity in the attachment behavioral system between human and divine figures. Correspondence is observed in the manner by which securely attached individuals have a representational framework amenable to a positive perception of God, and people successfully internalize religious beliefs from attachment figures (Granqvist, 2002; Granqvist & Hagekull, 1999; Granqvist & Kirkpatrick, 2004; Kirkpatrick, 1992). We hypothesize that individuals with positive (secure) representations of self and other in their working models will demonstrate this kind of continuity through positive perceptions of a God understood through human attachment language. Secondarily, we hypothesize that mediating insecure attachment styles will provide positive perception of God as an attachment figure, but with considerably weaker associations. The model proposes that, when controlling for religious tradition, correspondence is a better predictor of religious attachment than compensation where attachment language is used to describe God. The efficacy of the correspondence model tested in Study 1 is extended by examination of underlying differences in working models for nominated exemplars from Jewish, Muslim, and Christian religious backgrounds. Study 2 considers correspondence between human and religious attachment figures from a social intelligence perspective, mapping working model
Varieties of Religious Attachment y 13 representations of self and other by comparing similarity and dissimilarity of narrative responses using LSA. Participants are nominated for religious maturity and prosocial commitment on the basis of consensus criteria outlined by leaders from each religious tradition. Overall, we hypothesize that exemplars will have generally positive (secure) working models of self and other. Based on evidence affirming parental socialization influence in a correspondence view of religious attachment, we hypothesize that the parent representation will be close to self representation in working models for all three religious groups. We additionally hypothesize that the God representation will be close to self and parent representations in all three groups. In general, we expect that exemplar working models will reveal close proximities between self and other that reflect correspondence as modeled in Study 1. Study 1: Correspondence Model Study 1 tested a global correspondence model proposing that individuals with working models reflecting positive perceptions of self and other (e.g., secure attachment) will demonstrate positive perception of God when attachment language is used to describe the divine. The structural equation model was organized in a manner that emphasized perceptions of God for persons with positive working models, mediated through insecure attachment styles. We conceptualized the model on the basis of individuals with low anxiety and low avoidance that suggested positive perceptions of self and other (e.g., secure attachment style), with mediating factors of heightened anxiety and/or avoidance representing some combination of the self and other negatively (e.g., fearful-avoidant, dismissing-avoidant, preoccupied attachment styles). The model tested two hypotheses while controlling for participant religious tradition. Method Participants
Varieties of Religious Attachment y 14 The sample consisted of 431 undergraduates (301 women and 130 men) enrolled in two general education course sections at a large public university in California. Undergraduates participated in the study in exchange for partial course credit. Participant ages ranged from 17 years to 57 years (M = 21.6 years, SD = 6.15). Ethnicity was 3.0% African American, 1.2% American Indian, 4.2% Asian, 64.7% Caucasian, 18.1% Latino/a, 6.7% Multi-Racial, and 2.1% other ethnic background. Religious tradition included .9% Buddhist, .2% Hindu, .2% Muslim, 75.6% Protestant, 14.8% Roman Catholic, and 4.9% no religion. Measures, Variables, & Proposed Model Selection of measures appropriate to study objectives required compromises reflecting the growing measurement literature on adult attachment (Bartholomew & Shaver, 1998; Fraley & Waller, 1998; Fraley, Waller, & Brennan, 2000; Shaver & Mikulincer, 2004). For Study 1 we needed a general, working model representation of secure attachment amenable to a large sample of university undergraduates. This requirement suggested self-report (as opposed to interview) format, oriented toward generalized attachment experiences that avoided conceptual priming for religious language. We felt that peer-oriented attachment language was more neutral than romantic attachment language and therefore less likely to introduce a priming confound in the means by which participants interpreted items dealing with perception of the divine (Bartholomew & Shaver, 1998; Fraley, Waller, & Brennan, 2000). With its twin emphases on peer attachments and working models of self and other, we used the Relationship Scales Questionnaire (RSQ; Griffin & Bartholomew, 1994a). This instrument measures four attachment prototypes (e.g., secure, fearful-avoidant, dismissing-avoidant, and preoccupied) along with models of self and other. For the present study, RSQ self and other indicators were combined
Varieties of Religious Attachment y 15 with the secure attachment prototype indicator to comprise a general, latent working model variable (Bartholomew & Shaver, 1998). In addition to the RSQ, participants completed the God Image Scales, a measure for perception of the divine using language of human attachments (GIS; Lawrence, 1997). The GIS considers issues of belonging and goodness through acceptance. Prior work on religious attachment emphasized adaptation of existing trait-adjective measures of God-image or attachment paragraphs designed to correspond with single prototypes (Kirkpatrick & Shaver, 1992; Granqvist & Kirkpatrick, 2004). By contrast, the GIS was designed on the primary assumption that God functions as a secure base much like a human caregiver. Following Rizzuto (1979), the GIS differentiates between God concept and God image, using attachment language to provide a composite perception of the divine in an attachment role. The instrument demonstrates adequate construct validity (Lawrence, 1997). Thus, our analysis was premised upon the proposed structural equation model found in Figure 1. The independent (exogenous) variable for the structural model was the latent positive working model. The latent independent variable was comprised of three indicators. Mediating observed variables included fearfulavoidant, dismissing-avoidant, and preoccupied insecure attachment styles. The latent dependent (endogenous) variable for the structural model was perception of divine. As with the independent variable, perception of divine was composed of three indicators and controlled for participant religious tradition. --INSERT FIGURE 1 ABOUT HERE-Attachment.
Attachment was measured with the RSQ (Griffin & Bartholomew, 1994a).
The RSQ is a 30 item self-report measuring four attachment prototypes (e.g., secure, fearfulavoidant, dismissing-avoidant, preoccupied) and two attachment dimensions (e.g., model of self,
Varieties of Religious Attachment y 16 model of other). Each item is scored on a 5-point Likert scale from not at all like me to very much like me. To test the hypothesized model in Figure 1, insecure attachment prototypes were assessed using the RSQ. These prototypes comprised mediating observed variables. The fearfulavoidant scale is made up of 4 items (e.g., “I am uncomfortable being close to others”; α = .72). Additionally, the dismissing-avoidant scale is made up of 4 items (e.g., “I prefer not to depend on others”; α = .60). Finally, the preoccupied scale is made up of 4 items (e.g., “I worry that others don’t value me as much as I value them”; α = .51). In order to outline positive working model as a latent independent (exogenous) variable, we followed guidelines from Griffin and Bartholomew (1994b) on scoring the RSQ for self and other items. Model of self [(preoccupied + fearful) - (secure + dismissive)] was computed where high scores reflect positive view of the self. Model of other [(dismissive + fearful) – (secure + preoccupied)] was computed where high scores suggest openness to intimacy in relationships. These indicators were combined with the 5 item (e.g., “I find it easy to get emotionally close to others”; α = .67) secure prototype scale. Thus, the positive working model variable was comprised of three indicators including model of self, model of other, and secure attachment style. A mean scale was created from the three indicators with an internal consistency (alpha) of .62. Perception of divine. Perception of the divine was measured with the GIS (Lawrence, 1997). The GIS is a 36 item self-report that includes three scales applying attachment language to the divine. The GIS is based on a 4-point Likert scale ranging from strongly disagree to strongly agree. Three scales within the GIS were used as indicators in this latent dependent (endogenous) variable. The first scale, acceptance, deals with the extent to which participants feel worthy of divine attention and love. The acceptance scale was comprised of 6 items (e.g., “I am sometimes anxious about whether God still loves me”; α = .67). The second scale, entitled
Varieties of Religious Attachment y 17 presence, relates to the participant’s sense of whether God is available as a caregiver providing a secure base. The presence scale was also constructed from 6 items (e.g., “I get no feeling of closeness to God, even in prayer”; α = .83). The third scale, challenge, is conceptualized in terms of whether God’s presence supports or demands that the participant move away from the secure base to interact with the world. The challenge scale was derived from 6 items (e.g., “God asks me to keep growing as a person”; α = .76). Thus, perception of the divine was measured as a three indicator latent dependent variable premised upon language reflecting attachments. Again, a mean scale was created from the three indicators, demonstrating internal consistency (alpha) of .78. Preliminary Analyses Means, standard deviations, and bivariate correlations for manifest indicators and variables are given in Tables 1 and 2. The Pearson correlations in these tables are indicative of relations between manifest indicators and variables rather than underlying latent constructs tested in the structural model. The secure attachment manifest indicator was positively correlated (p < .01) with all three dimensions of the GIS (r = .13 to .28). Conversely, the fearful-avoidant (high anxiety, high avoidance) manifest variable was negatively correlated (p < .01) with two of the three GIS dimensions (r = -.23 to -.25). Lowest correlations were observed for the preoccupied manifest variable from the RSQ and the challenge manifest indicator from the GIS. --INSERT TABLES 1 AND 2 ABOUT HERE-SEM We used version 5.0 of the Analysis of Moment Structures (AMOS; Arbuckle & Wothke, 1999) program to test a model for religious attachment that included one latent independent variable, three mediating observed variables, and one latent dependent variable. Direct maximum
Varieties of Religious Attachment y 18 likelihood estimation of missing data facilitated inclusion of the entire sample. Positive working model served as a latent independent variable including model of self, model of other, and secure attachment indicators. Fearful-avoidant, dismissing-avoidant, and preoccupied insecure attachment styles comprised the three mediating observed variables. Perception of the divine was the latent dependent variable identified by acceptance, presence, and challenge indicators of religious attachment. The model was examined for goodness of fit with the use of several indices tempered by recommendations from Hu and Bentler (1999). These included the root mean square error of approximation (RMSEA), an index of discrepancy per degrees of freedom. Acceptable fit for this index should be near or above .06. The goodness of fit index (GFI), adjusted goodness of fit index (AGFI) and comparative fit index (CFI) were also used to assess the overall robustness of the model. With these recommendations, GFI and CFI should be .95 or better, with AGFI at or above .90. We conducted the test of the structural model in two steps. Anderson and Gerbing (1988) recommend a split-half procedure for model fitting. This approach necessitates dividing the data into two randomly selected subsamples. The first subsample is used for an initial test of the model and for the purpose of specifying model modifications as needed. The second subsample serves a confirmatory function for analysis of the final model. Following this procedure, the first randomly selected subsample had 215 participants, with 216 participants in the second subsample (overall N = 431). The strategy was employed to test the primary study hypothesis that individuals with positive (secure) representations of self and other will demonstrate correspondence through positive perceptions of a God understood through human attachment language. Results and Discussion
Varieties of Religious Attachment y 19 Initial analysis of the proposed structural model with the first subsample found poor (not admissible) fit of the data to the model. Modeled variables were examined in order to deduce whether the lack of fit was related to the measurement or structural models. It was determined that the mediating observed variables dismissing-avoidant and preoccupied were poorly fitted. However, the latent independent variable positive working model and mediating observed variable fearful-avoidant demonstrated moderate fit X2 (67, N = 216) = 220.34, p < .01. Consequently, we eliminated the mediating observed variables dismissing-avoidant and preoccupied from the model. We then freed the correlations among the three error terms for the latent independent variable positive working model and for the three error terms for the latent dependent variable perception of the divine. This change caused a significant improvement in overall model fit, X2 (28, N = 215) = 57.0, p < .01; GFI = .94, AGFI = .88, CFI = .96, RMSEA = .10. Confirmatory analysis was conducted on the second subsample and indicated good model fit, X2 (18, N = 216) = 49.26, p < .01; GFI = .95, AGFI = .90, CFI = .97, RMSEA = .09. The modified structural equation model reflecting confirmatory analysis in the second subsample is given in Figure 2. --INSERT FIGURE 2 ABOUT HERE-The purpose of Study 1 was to globally consider correspondence between positive working models and the divine where the latter is described using the language of attachment. Results supported the main hypothesis in that relational continuity implied through Bowlby’s theory of working models is sustained where positive or secure representations of self and other are considered in perception of the divine. The standardized path coefficient between the independent (exogenous) variable and the dependent (endogenous) variable was robust. We additionally hypothesized that mediating variables of insecure attachment styles will reflect a
Varieties of Religious Attachment y 20 diminished (albeit positive) perception of God as an attachment figure. This hypothesis was also supported, although in conjunction with the unexpected elimination of two mediating observed variables. The standardized path coefficient between the mediating observed variable fearfulavoidant and the dependent (endogenous) variable perception of the divine was significant, but considerably lower than the coefficient indicated for the main hypothesis. In a general sense, correspondence in the model does appear to better predict positive perception of the divine than compensation. Our discussion recalls Granqvist and Kirkpatrick’s (2004) proposal for correspondence and compensation hypotheses. Study emphasis on positive working models in religious attachment validates the contention that correspondence reflects the contemporary adult social network. Beyond immediate issues related to worthiness and divine availability (e.g., acceptance, presence), attachment language in the GIS emphasizes the extent to which participants feel empowered to move away from the divine as a secure base (e.g., challenge). The structural model tested in Study 1 did not account for specific attachment relationships in the social network of participants. However, the use of human relationship categories in divine perception suggests that the attachment behavioral system in working models is extended into the religious domain on the basis of parallel or analogous experiences. The strong predictive utility of correspondence in religious attachment justifies further efforts to specify the nature of contemporary socialization influences in the social network. Study findings regarding compensation were more difficult to interpret. We address two issues salient to the compensation findings, including (a) elimination of dismissing-avoidant and preoccupied mediating observed variables, and (b) the strength of path coefficient between the fearful-avoidant mediating observed variable and the latent dependent variable. One response to
Varieties of Religious Attachment y 21 issue (a) relates to attachment measurement. A potential drawback to the RSQ is its heavy investment in the measurement of attachment prototypes (Fraley & Waller, 1998; Fraley et al., 2000). Psychometric analysis of the RSQ reinforces the ethological underpinnings of Bowlby’s theory, suggesting that attachment measurement can be understood in a manner that does not necessitate typological (e.g., prototype) schemes (Fraley & Waller, 1998). On the basis of their construct validation work, Fraley and Waller (1998) argue that the four prototype categories in the RSQ should be collapsed into two continuous dimensions of anxiety and avoidance. For the present study, secure attachment indicated by the latent independent variable positive working model represents diminished anxiety and avoidance. The remaining observed variable fearfulavoidant represents high anxiety and avoidance. The structural model retains those prototypes that, on the basis of Fraley and Waller’s (1998) psychometric revisions, are anticipated to provide “opposites” or variability that generally distinguishes individuals with low anxiety (or avoidance) from those with high anxiety (or avoidance). With regard to issue (b), we observe that the path for the fearful-avoidant mediating observed variable and the latent dependent variable (perception of the divine) was significantly positive. Fraley and Waller (1998) hypothesize that anxiety and avoidance dimensions may work at different (albeit complementary) levels of social cognition. Anxiety is implicated by limbic activity, reflecting affective thresholds. Avoidance appeals to the organization of knowledge representations. Along these lines, Rowatt and Kirkpatrick (2002) found that heightened anxiety predicted neuroticism and negative affect in religious attachment. High avoidance inversely predicted agreeableness and religious symbolic immortality. Given high anxiety and avoidance intrinsic to the fearful-avoidant observed variable, we can surmise that the positive path coefficient reflects a compensation pattern at least partly interpretable on the basis of the five-
Varieties of Religious Attachment y 22 factor model (McRae & Costa, 1990). Heightened neuroticism in this instance suggests guiltprone compensation for self-perceived inadequacies (McRae, Costa, & Busch, 1986). Diminished agreeableness implies a considerable degree of internal ambivalence in perception of the divine, where guilt coincides with skeptical expectations of transcendent reciprocity. A second interpretation for issue (b) recalls Granqvist and Kirkpatrick’s (2004) notion that compensation is an artifact of developmentally cumulative attachment experiences. Sroufe, Egeland, Carlson and Collins (2005) characterize their decades-long work on attachment in terms of developmental capacity for the individual to relate to peers. These authors offer five bases for effective peer attachments including (a) motivational base or belief that relationships will be rewarding, (b) attitudinal base or belief that one may elicit responses from others, (c) instrumental base or mastery through support for exploration, (d) emotional base or selfregulation of emotion, and (e) relational base or expectations concerning empathic exchange (Sroufe et al., 2005). The present study framed religious attachment in terms of peer relations, making this developmental interpretation potentially relevant to compensation. In addition to measurement issues, it is possible that the elimination of dismissing-avoidant and preoccupied mediating observed variables relates to developmental deficits intrinsic to the five bases. The prominence of high anxiety and high avoidance (fearful-avoidant mediating observed variable) in the structural equation model suggests that compensation is related to attachment deficits that are fairly equitably spread across all five developmental bases. In a longitudinal sense, compensation implied by the model can be characterized in terms of efforts to resource divine transcendence to make up for diminished capacities to facilitate engagement with peers, stay engaged with peer groups, or construct a basic understanding of what positive relationships
Varieties of Religious Attachment y 23 require (Sroufe et al., 2005). The present study did not handle compensation in a longitudinal manner, nor did it account for conversion experiences. These conclusions indicate further study. Study 2: Correspondence & Socialization in Working Models The efficacy of generalized correspondence was further specified by exploratory comparison of underlying differences in working models for nominated exemplars from Jewish, Muslim, and Christian religious backgrounds. The purpose of Study 2 was to employ a comparative religious design to consider working model correspondence between human and religious attachment figures from a social intelligence perspective. This included evaluation of attachment security in nominated exemplars from three religious traditions. The study created maps of working models to facilitate comparison of set relations reflecting socialization. This was accomplished by using latent semantic analysis (LSA) to evaluate semantic and episodic content in participant narratives that correspond with representations of self and various others. Two complementary multivariate techniques were used to further specify representations in the high-dimensional semantic space. Multidimensional scaling (MDS) was used to construct perceptual maps outlining dimensions for representations reflecting correspondence in religious attachment and socialization. Hierarchical cluster analysis (HCA) identified set relations between representations regarded as similar or sharing significant semantic and episodic content. Given the main objective to study correspondence and socialization influences, monotheistic world religions that reference the divine as a singular other (e.g., Jewish, Muslim, and Christian) were targeted. Thus, the present study tested three hypotheses reflecting correspondence and socialization in working models. Method Nominated Exemplar Participants
Varieties of Religious Attachment y 24 Sampling for the second study followed guidelines in a recent project emphasizing naturalistic conceptions of exemplary psychological functioning applicable to participant nomination (Matsuba & Walker, 2004). Three separate focus groups were convened from Jewish, Muslim, and Christian religions, respectively. Focus groups were comprised of 6-12 leaders and clergy from each religion in a diverse California metropolitan region. Leaders were invited from religion subgroups that were numerically well represented in the metropolitan area, including Reform Jewish, Sunni Muslim, and Presbyterian Christian traditions. Focus groups were conducted in English, asking individuals to identify exemplar nomination criteria that reflected religious maturity and prosocial commitment. Focus group process included prioritization of criteria, with similar descriptors typically collapsed into general statements. Groups were asked to ratify final criteria as being representative of their process and religious perspective. The resulting list of nomination criteria included (a) learning and being in continual process, (b) sense of (and acting on) responsibility for one’s religious fellows, (c) sense of one’s own faith that informs daily life, (d) God-consciousness, (e) believes in Torah/Qu’ran/Bible as word of God and follows it in daily life, (f) lives life intentionally, (g) practices faith (e.g., prayer, fasting, observances, charity, declaration of faith, pilgrimage), (h) promotes peace among all peoples, (i) is actively engaged with God and others, (j) lives a joyful, balanced, and humble life, and (k) is interested in helping others grow spiritually in a quietly contagious manner. Consensus nomination criteria were provided to leaders and clergy that participated in the focus groups. Leaders and clergy were asked to nominate individuals from within their respective religious tradition that evinced strong evidence of nomination criteria. Nominated exemplars comprised several official religious leaders, but mainly consisted of everyday individuals from area synagogues, mosques, and church congregations. Nominated exemplars
Varieties of Religious Attachment y 25 were contacted and invited to participate in the study. Interested exemplars were mailed consent forms, a self-report survey, and scheduled for a face-to-face interview. Participants were provided with a $50 honorarium. Of 36 exemplars nominated from Reform Jewish leaders and clergy, 20 participated. This sample group averaged 45.0 years of age (SD = 11.2, range = 25-66). The sample self-identified as ethnically Jewish (82.4%), European (11.8%), or Latino/a (5.9%). Level of education included high school completion (5.0%), bachelor’s degree (30.0%), master’s degree (55.0%) and doctoral degree (10.0%). Out of 27 nominations from Sunni Muslim leaders and clergy, 20 participated. The Muslim sample averaged 34.5 years of age (SD = 11.4, range = 23-79). This sample selfidentified as ethnically European (70.6%) or Turkish (29.4%). Level of education included high school completion (5.0%), bachelor’s degree (35.0%), master’s degree (35.0%) and doctoral degree (25.0%). Of 32 nominations from Presbyterian Christian leaders and clergy, 20 participated. The Christian sample averaged 56.9 years of age (SD = 11.3, range = 33-72). This sample self-identified as ethnically European (80.0%), Latino/a (15.0%), or American Indian (5.0%). Level of education included high school completion (15.0%), trade school or associate’s degree (5.0%), bachelor’s degree (10.0%), master’s degree (50.0%) and doctoral degree (20.0%). The lower mean age for Muslim exemplars reflects the recently emigrated, highly educated profile of Turkish citizens in several local mosques. Overall, nominated exemplars were welleducated individuals working in professional vocations including engineers, nurses, teachers, and doctors. All exemplars were fluent in the English language. Procedure The general correspondence model outlined in Study 1 focused on positive (e.g., secure) internal working models. For Study 2, the first hypothesis was that nominated exemplars would
Varieties of Religious Attachment y 26 demonstrate positive (secure) internal working models. Accordingly, Study 2 participants completed the Relationship Scales Questionnaire (RSQ; Griffin & Bartholomew, 1994a). To address the second and third hypotheses, participants responded to semi-structured interview questions. Questions included references to global self (What kind of person are you?), along with others implicated in working models that reflect the social network past and present. These others included romantic partner (What kind of person are you with your romantic partner?), parents (What kind of person are you with your parents?), best friend (What kind of person are you with your best friend?), and God (What kind of person are you with God?). Interview responses were recorded and transcribed. All identifying content was removed from transcripts. Interview recordings were erased and a numerical code assigned to transcripts to ensure confidentiality. Transcribed responses were collated by interview question for each of the three religious traditions. In this manner, aggregate responses to the interview were constructed for Jewish, Muslim, and Christian sample groups. Each aggregate response served as one of five cognitive representation variables in a social intelligence view of working models (e.g., self, romantic partner, parents, best friend, God). The second and third hypotheses in Study 2 called for analysis of similarity and dissimilarity between cognitive representation variables for each religious sample group as an approximation of socialization in working models of attachment. The basis for LSA analysis of similarity and dissimilarity is any text selected by the researcher on the basis of its appropriateness for evaluating shared semantic and episodic knowledge. Given the current study interest of correspondence in religious attachment, the 30 item stems of the RSQ were used as texts for comparison to the five cognitive representation variables from each religious tradition. The RSQ afforded a continuous measure of attachment between Study 1 and Study 2.
Varieties of Religious Attachment y 27 Additionally, the RSQ provided a basis for semantic and episodic comparison that was known to be reliable and valid, at least along anxiety and avoidance dimensions. Five cognitive representation variables were positioned as columns with RSQ items as rows in a 5 X 30 LSA matrix for each religious tradition. Thus, LSA semantic space analyses produced three covariance matrices reflecting similarity and dissimilarity between cognitive representations based on exemplar responses from Jewish, Muslim, and Christian religious traditions. Results Results are organized into three sections following Study 2 hypotheses. The first section examines the prevalence of secure attachment for exemplars from three religious traditions. The second and third sections present findings from multivariate procedures used to explore working model representations in three LSA covariance matrices comparing cognitive representation variables with RSQ item stems. The second section considers multidimensional scaling (MDS) perceptual maps of cognitive representation variables as working models from three religious traditions. The third section presents hierarchical cluster analysis (HCA) clusters identifying set relations between cognitive representation variables in a manner suggestive of correspondence in the contemporary social network. The second and third sections highlight underlying dimensions and cluster relations in attachment correspondence indicative of socialization processes. Secure attachment.
Table 3 provides mean scores for secure prototype along with
model of self and other for exemplars from each religious tradition. Nominated exemplar RSQ scores suggested an overall profile of positive (e.g., secure) working models of attachment. Secure prototype scores were positive for model of self and model of other. High scores on model of self and model of other are indicative of diminished anxiety and avoidance, respectively. To determine whether this profile was typical of all three religious traditions, a one-
Varieties of Religious Attachment y 28 way ANOVA was conducted with secure prototype along with model of self and other as dependent variables. This analysis revealed no main or interaction effects for religious tradition, either through post-hoc or homogeneity of variance tests. --INSERT TABLE 3 ABOUT HERE-MDS. For each religious tradition, MDS was used to construct perceptual maps of cognitive representation variables approximating working models of attachment. MDS outlines structure in distance measures between variables when the basis for comparison is unknown or undefined. Structure is identified along orthogonal axes (e.g., dimensions) reflecting frequency of co-occurrence among variables. The ALSCAL MDS model was used for the present study as an approach suitable for data matrices of group samples (Hair, Black, Babin, Anderson, & Tatham, 2005). Generally, modeled data reflect good fit where Kruskal’s stress value is equal to or below .20, with R2 at or above .60 (Hair et al., 2005). For the three MDS models in the present study, Kruskal’s stress was .03 (Jewish), .02 (Muslim), and .02 (Christian). Final R2 values were .99 for all three models, respectively. These values indicated strong overall model fit in three different solutions by religious tradition. Thus, MDS models were constructed for twodimensional perceptual maps presenting distance measures between cognitive representation variables considered on the basis of similarity or dissimilarity in a LSA semantic space. --INSERT FIGURES 1, 2, & 3 ABOUT HERE-The MDS model solution for Jewish exemplar cognitive representation variables is given in Figure 3. Inspection of variable distribution suggested a self-other dimension and a transcendent dimension. Identification of these dimensions was subjectively based on interpretation of how variables were arrayed across each dimension. The positive pole of the horizontal self-other dimension was anchored by the self variable, with the parent variable at the
Varieties of Religious Attachment y 29 opposing pole. The positive pole of the vertical transcendent dimension was anchored by the romantic partner and self variables, with the God variable at the opposing pole. The MDS model for Muslim exemplar cognitive representation variables is presented in Figure 4. The horizontal dimension was anchored at the positive pole by the self variable, with the parent variable at the negative pole. The distribution mirrored the horizontal dimension in the Jewish MDS model. This suggested a similar self-other interpretation for the Muslim horizontal dimension. However, the vertical dimension also suggested a self-other interpretation, anchored by the best friend variable at the positive pole and the self variable at the opposing pole. The central location of the God variable did not clarify any differentiating role for transcendence in either dimension. For the sake of interpretative consistency, we chose to label the Muslim MDS horizontal dimension as self-other. The vertical dimension suggested a developmental progression of attachment transference, with the parent variable closest to the self variable, and the best friend variable farthest away. The MDS model solution for Christian exemplar cognitive representation variables is given in Figure 5. The horizontal dimension was anchored at the positive pole by the God and self variables, with the parent variable at the negative pole. As with the other models, this suggested a self-other interpretation. The vertical dimension was anchored at the positive pole by the God variable, with the self variable at the negative pole. This mirrored the transcendent interpretation for the vertical dimension in the Jewish MDS model. Thus, we labeled the vertical dimension for the Christian model solution transcendent. HCA. In addition to MDS analyses, LSA covariance matrices comparing similarity and dissimilarity between cognitive representation variables in semantic space were subjected to hierarchical cluster analysis. The purpose of HCA was to explore set relations between cognitive
Varieties of Religious Attachment y 30 representation variables in a manner suggestive of socialization process in working models. HCA for each religious tradition was conducted using Ward’s method along with the squared Euclidean distance interval. At present there is no established basis for identifying an optimal number of clusters for any solution, making this process somewhat open to interpretation. However, the HCA agglomeration schedule contains information that may be used to delimit clusters (Hair et al., 2005). The point where large percentage change between agglomeration coefficients becomes more stable (as reflected in smaller percentage change) is suggestive of how many clusters are indicated in the overall solution. This rule was applied to three HCA analyses run for Jewish, Muslim, and Christian LSA matrices, respectively. To simplify interpretation, HCA cluster solutions are integrated into MDS perceptual maps for each religious tradition in Figures 3, 4, and 5. This integrated presentation reflects a complementary function between HCA and MDS on LSA covariance data, where variable clusters are afforded orientation for visual analysis on the basis of underlying dimensions. Using the percentage change in agglomeration rule, a 3 cluster solution was indicated for all religious traditions. Cluster boundaries are illustrated in MDS perceptual maps with different lines enclosing clustered variables. In Figures 3-5, thick (solid) boundary lines reflect the first cluster, followed by a thin (solid) boundary line for the second cluster, and a dotted line used to enclose the third cluster. Figure 3 presents clusters for the Jewish exemplars. Cluster 1 enclosed the God variable with best friend. Cluster 2 enclosed the parent variable with romantic partner. Cluster 3 enclosed Cluster 1 with the addition of the self variable. Figure 4 presents clusters for the Muslim exemplars. Cluster 1 enclosed the God variable with self. Cluster 2 enclosed the parent variable with romantic partner. Cluster 3 enclosed Cluster 1, with the addition of the best friend variable. Figure 5 presents clusters for Christian exemplars. Cluster 1 enclosed the romantic
Varieties of Religious Attachment y 31 partner variable with best friend. Cluster 2 enclosed the God variable with self. Cluster 3 enclosed Clusters 1 and 2. Discussion The main purpose of the second study was to explore working model correspondence between human and divine attachment figures in a manner suggestive of socialization influences. Using a comparative religious sample design, we obtained nominated exemplar narrative responses to questions related to cognitive representation variables of self, romantic partner, parents, best friend, and God. These narrative responses were analyzed using LSA to consider similarity and dissimilarity of representations in an unsupervised, computational semantic space. The general study hypothesis that nominated exemplars would have positive (secure) working models of self and other was confirmed. This finding further supports Granqvist and Kirkpatrick’s (2004) contention that securely attached individuals with demonstrable religious commitment evince divine attachment that corresponds with the contemporary social network. We additionally hypothesized that parental socialization influence is implied by close proximity between parental and self representation for all three religious groups. Finally, we hypothesized that the God representation will be close to self and parent representations as a ratification of correspondence. The exploratory use of LSA, MDS, and HCA revealed a somewhat unexpected picture of socialization influences in correspondence. In a refutation of the second study hypothesis, working models for all three groups showed the parent representation to be far removed from the self representation. Moreover, in no religious group was the parent representation found in close proximity (as identified in set relations through cluster analysis) with the God representation. However, the third hypothesis that the God representation would be close to the self
Varieties of Religious Attachment y 32 representation was supported through cluster analysis for all three religious traditions. To better account for these findings, we consider results from each religious tradition as unique socialization contexts reflecting underlying differences in working models, followed by summary discussion of similarities between groups. Taken from a social intelligence perspective, the Jewish exemplar working model demonstrated strongest semantic and episodic similarity between best friend and God. The inclusion of self into this set relation on the third cluster suggests that working model representations of divine and human attachment figures are significantly analogous. Socialization may include peer affinities both in terms of how Jewish exemplars conceptualize the divine (e.g., as a close friend), but additionally through shared religious experiences in close peer relationships. The second cluster set relation between parent and romantic partner representations could offer support to recent arguments for intergenerational transmission (e.g., transference) of attachment style between parental and romantic figures in females (Obegi, Morrison, & Shaver, 2004). The lack of inclusion of divine or self representations in the second cluster is somewhat surprising, indicating a fairly discrete parcel of shared semantic and episodic knowledge that links self and divine. The comparative linguistic basis for the LSA method suggests that similar language is used by Jewish exemplars to describe both human and divine attachment figures, and that the divine is conceptualized as an important contemporary socialization influence. The Muslim exemplar working model reflects a strong link between divine and self representations in the first cluster. The primacy of divine attachment correspondence is augmented with inclusion of the best friend representation on the third cluster. The finding raises an issue regarding what attachments are fundamentally resourced as reference points in working model correspondence. Whereas Granqvist and Kirkpatrick’s (2004) correspondence hypothesis
Varieties of Religious Attachment y 33 assumes that the divine is referenced on the basis of prior human attachment experiences, the Muslim working model suggests that the inverse may be true in the adult’s contemporary social network, namely that human relationships are referenced on the basis of divine attachment experience. Socialization influences in this view might be premised first on the divine attachment representation. The underlying difference observed in the vertical MDS dimension interpreted in terms of developmental attachment transference implicates change whereby adult Muslim exemplars retrospectively view their experiences with human attachments in divine terms. As with the Jewish exemplar group, the second cluster combined parental and romantic attachment representations in a manner that did not suggest immediate or direct socialization in religious attachment. The working model for Christian exemplars reveals close semantic similarity between romantic partner and best friend representations in the first cluster. This difference suggests a variation on how romantic and peer attachments are represented. The second cluster inclusion of God and self representation is in the third cluster combined with the first. The Christian working model implies correspondence that is distributed amongst significant attachment figures excluding parents. Socialization in this instance appears to reflect those who are readily accessible in the contemporary social network. Correspondence between human and divine attachment figures appears to be fairly evenly distributed, with similar semantic and episodic integration of these relationships into the behavioral system. The distal position of the parent representation in the Christian working model is particularly striking given that prior findings suggesting a central role for parents in religious attachment were premised on convenience samples including disproportionately high numbers of Christians. This finding should be interpreted cautiously, however, given that the mean age of the Christian sample was
Varieties of Religious Attachment y 34 considerably higher than exemplars from other religious traditions. It is possible that the immediate socialization influence of parents was framed in different semantic and episodic knowledge bases reflecting the fact that for many, both parents had already passed away. Overall, the social intelligence approach to working models suggests that exemplars nominated for their religious maturity and prosocial commitment make corresponding representational alignments between human and divine attachment figures. Finer variations in the nature of corresponding working model representations are visible between religious groups, but in all cases MDS and HCA analyses support the idea that similar knowledge is employed to describe (and presumably experience) human and divine attachments. The rejected hypothesis regarding proximity of parental representation may be an artifact of the contemporary social network aspect of correspondence in religious attachment. Gradual change in the organization of attachment representation in the working model is developmentally indicated for those with secure attachment profiles (Granqvist & Kirkpatrick, 2004). Clearly, the adult exemplars interviewed in this study did not represent parental attachment figures in a closely analogous position relative to the divine or self, meaning that whatever socialization influences may have existed at earlier developmental stages, the contemporary context suggests alternative socialization processes, notably best friends. We did not consider the religious practices or tradition of best friends for the study, yet the significance of contemporary peer relationships in correspondence is worthy of future investigation in terms of individual differences for personality in religious attachment with concomitant developmental processes. We note that our use of the RSQ in these exploratory analyses may have “front loaded” the efficacy of peer-oriented language in LSA, making the use of other attachment items stems (e.g., romantic) relevant to future investigation of correspondence in working models.
Varieties of Religious Attachment y 35 Despite this limitation, correspondence in the case of Muslim exemplars appeared to reverse the assumption that human attachments form the pre-existing referent for divine attachment representation. This finding suggests revision to the correspondence hypothesis in emphasizing bi-directional reference between human and divine attachment representations in working models. General Discussion How are working models implicated in the manner by which people perceive the divine as an attachment figure? This project was constructed in an effort to provide a more comprehensive understanding of personality in religious attachment for three major world religions. Religious experience is ubiquitous across cultures and peoples. The propensity for people to use relational language to describe that experience suggests a complex interchange between underlying representations of self and other in the attachment behavioral system. Our intent was to model these representations in terms of overall correspondence and on the basis of socialization influences, a strategy predicated on the belief that representations in working models are imbued with semantic and episodic knowledge. This perspective makes the language of attachment particularly important toward an understanding of how attachment experiences are incorporated into religious understanding. The studies reported above were designed on the assumption that working models associated with religious attachment are hardly monolithic structures, but instead reflect changing influences vitally related to the context of particular religious traditions and social networks. To accomplish these objectives, we examined a general model of correspondence (Study 1) along with specific representations of self and other reflecting working models (Study 2). Together, the studies emphasized attachment language in perceptions of the divine along with
Varieties of Religious Attachment y 36 particularities that influenced perceptions endemic to three major religious traditions. The strong path association between positive (e.g., secure) working model and perception of the divine in Study 1 and the prominence of contemporary peer (e.g., best friend) influence in exemplar socialization influences in Study 2 provides support for a recent correspondence proposal (Granqvist & Kirkpatrick, 2004; Kirkpatrick, 1999). Yet overall study findings are more complex than this most forthright conclusion. Presumably, correspondence in securely attached individuals affords importance to parental influence in religious socialization. Correspondence implies close representational similarity between God and parents in religious attachment, even accounting for contemporary social networks that emphasize peer over familial influences. Despite generally secure attachment profiles, nominated exemplars from all three religions did not evince close integration of parent representations with either self or the divine. This surprise finding is worthy of reflection with regard to implications for correspondence in religious attachment, and more generally toward the contextual study of attachment in working models across the life span. Regarding the former, we note that for at least one world religion (e.g., Islam), perceptual map and set relations findings suggest that religious attachment with the divine serves as the preeminent referent for human others, at least in the contemporary social network. One possible explanation for this finding is that correspondence for securely attached people growing up in a Muslim context reflects parental socialization influences that so centrally implicate God in everyday life that divine representations eventually supersede human others in working models. Certainly elements of this kind of perceived immanence were evident in the responses of Muslim exemplars that commonly prefaced their reflections with the refrain “if Allah wills.” It may be that this is a developmental process whereby Muslim exemplars grow to realign attachment intimacy toward the divine. Gradual
Varieties of Religious Attachment y 37 reorganization of working models influences the manner by which individuals perceive the social network past and present. This implies revision for correspondence to accommodate longitudinal aspects of the compensation hypothesis. Correspondence might be better understood not only on the basis of contemporary socialization influences, but with regard to the manner by which prior influences shaped the prominence of God in the contemporary working model, which in turn frames retrospective accounts of relationships in episodic memory. The subtlety of the longitudinal question supports further study on the developmental contours of religious attachment, a position advocated in a review by Kirkpatrick (1999). Also salient to correspondence is the Study 2 finding that God is closely represented with self in exemplar working models for all three religious traditions. While not unexpected for persons nominated by each tradition for their exemplary religious and prosocial commitments, the close proximity between God and self representations may spotlight cracks on the correspondence hypothesis to the extent that transcendent beings are understood in a manner analogous to human relationships. In our formalist estimation, the divine is not human and cannot be physically perceived in time and space. As a consequence, semantic and episodic knowledge regarding God must at some level remain prototypical (Cantor & Mischel, 1979; Fehr, 1988; Walker & Hennig, 2004; Walker & Pitts, 1998). Exemplars from monotheistic religious traditions may base some aspects of their prototypical notion of the divine on human others, but it is likely that this prototype is far more complex, featuring aspects of selfunderstanding that incorporate religious virtues learned from teachings, holy writings, observances, and pilgrimages. Perceptual map and set relation findings for all three traditions support an interpretation for considerable semantic and episodic convergence between God and self representations. Even when ensconced within the practices of particular religious
Varieties of Religious Attachment y 38 communities (e.g., synagogues, mosques, and church congregations), representations of God are likely to become blurred with self insofar as two fundamental dimensions of identity (agency and communion) are not easily distinguished in relation to a divine other that lacks physical properties. In addition to the developmental concerns noted above, the correspondence hypothesis should in some manner be enlarged to accommodate the complexity of prototypicality in religious experience. This discussion raises an interesting possibility. Do the perceptual maps constructed for religious exemplars reflect goal-corrected partnerships with the divine? Bowlby (1969) argued that securely attached children grow to gradually alter their working model representations of caregivers. Working models are similarly revised on the part of the caregiver. With the onset of developmental capacity for perspective-taking (e.g., theory of mind), Bowlby (1969) believed that children and parents construct shared working models reflecting deeper security manifest in reciprocity and mutual negotiation. He defined this process in terms of a goal-corrected partnership. The close proximity between divine and self representations in exemplar working models may reflect similar capacities for perspective-taking on the part of individuals who perceive sustained attachment security in a religious other such as the divine. Even if the divine is prototypically represented in a manner that transcends physicality, individuals may through prayer and ritual understand the divine in a goal-corrected sense, a partnership reflecting dynamic “give-and-take.” Indeed, exemplars from all three religious traditions spoke extensively about their perceptions of God’s current expectations for their behavior, relationships, and even vocation. For adult exemplars with sophisticated capacities for theory of mind, it is possible that goal-corrected partnerships that initially emerged in the context of secure attachments with caregivers are similarly constructed with the divine. The accoutrements of organized religion
Varieties of Religious Attachment y 39 (e.g., worship, prayer, ritual, observances, and pilgrimage) are most likely designed to support this kind of goal-corrected religious attachment. In addition to the representational approach to working models used in the present study, further work on correspondence in religious attachment might explore the goal-corrected issue in terms of individual differences in dispositional traits and adaptations in personality that reflect strivings and goals (McAdams, Anyidoho, Brown, Huang, Kaplan, & Machado, 2004). A final concern of the study relates to the manner by which working models are studied in social context. In particular, underlying differences in working models were considered on the basis of semantic and episodic knowledge in a social intelligence framework. An unsupervised computational methodology was used to explore working models in terms of constituent representations of self and other for exemplars from three world religions. Striking similarities were noted for five cognitive representation variables across three religious traditions. The relative lack of shared semantic and episodic content between parent and self representations for exemplars in the present study supports a flexible, contextually specific interpretation of working models. Implications for correspondence in religious attachment notwithstanding, this finding suggests a potentially useful methodological take on the manner by which working models enable reflection on past and present attachment experiences toward a sense of healthy and adaptive proximity regulation in relationships (Bretherton, 2005; Bretherton & Munholland, 1999). Quantification of representations in working model topography is historically difficult given the complexities of memory, affect, and meaning-making associated with cognitive schema. Semantic and episodic knowledge affiliated with the organization of self and other representations can in semantic space be considered individually or across groups, providing an exploratory basis for understanding working models in terms of influences in the social network.
Varieties of Religious Attachment y 40 The method may represent a useful step toward more intentional study of individual differences in working model representations between groups and longitudinally across the life span. Along with these conclusions, several study limitations must be noted. The study considered religious attachment in nominated exemplars from three world religions. By definition exemplars are rare, limiting the sample size for exploratory analyses of working model representations. Moreover, exemplars were selected from monotheistic world religions owing to the central focus on correspondence with a singular divine other in religious attachment. The three traditions sampled in the present study are hardly representative of the diversity of religious experience in the world. The project did not consider animist, polytheist, pantheist, or panentheist religious traditions. It is entirely possible that correspondence in religious attachment is of diminished significance or outright irrelevance in the working models of individuals from Shinto, Hindu, or tribal religious backgrounds. Finally, exploration of working model representations in religious attachment was premised upon a novel methodology that is experimental in applied behavioral research. Summary evaluation of these findings is tentative and subject to the replication of protocols incorporating computational analyses of natural language for the study of working models in personality. References Ainsworth, M. (1985). Attachments across the life-span. Bulletin of the New York Academy of Medicine, 61, 792-812. Ainsworth, M., Blehar, M., Waters, E., & Wall, S. (1978). Patterns of attachment: A psychological study of the strange situation. Hillsdale, NJ: Erlbaum. Anderson, J., & Gerbing, D. (1988). Structural equation modeling in practice: A review and recommended two-step approach. Psychological Bulletin, 103, 411-423.
Varieties of Religious Attachment y 41 Arbuckle, J., & Wothke, W. (1999). Amos user’s guide. Chicago: SmallWaters. Bartholomew, K., & Shaver, P. (1998). Methods of assessing adult attachment: Do they converge? In J. Simpson & W. Rholes (Eds.), Attachment theory and close relationships (pp. 25-45). New York: Guilford. Birgegard, A., & Granqvist, P. (2004). The correspondence between attachment to parents and God: Three experiments using subliminal separation cues. Personality and Social Psychology Bulletin, 30, 1122-1135. Bowlby, J. (1969). Attachment and loss: Vol. 1. Attachment. New York: Basic. Bowlby, J. (1973). Attachment and loss: Vol. 2. Separation: Anxiety and Anger. New York: Basic. Bowlby, J. (1980). Attachment and loss: Vol. 3. Loss: Sadness and Depression. New York: Basic. Brennan, K., Clark, C., & Shaver, P. (1998). Self-report measurement of adult attachment: An integrative overview. In J. Simpson & W. Rholes (Eds.), Attachment theory and close relationships (pp. 46-76). New York: Guilford. Bretherton, I. (2005). In pursuit of the internal working model construct and its relevance to attachment relationships. In K. Grossmann, K. Grossmann, & E. Waters (Eds.), Attachment from infancy to adulthood (pp. 13-47). New York: Guilford. Bretherton, I., & Munholland K. (1999). Internal working models in attachment relationships: A construct revisited. In J. Cassidy, & P. Shaver (Eds.), Handbook of attachment: Theory, research and clinical applications (pp. 89-111). New York: Guilford.
Varieties of Religious Attachment y 42 Cantor, N., & Mischel, W. (1979). Prototypes in person perceptions. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 12, pp. 3-52). New York: Academic Press. Collins, N., & Allard, L. (2001). Cognitive representations of attachment: The content and function of working models. In G. Fletcher & M. Clark (Eds.), Blackwell handbook of social psychology: Vol. 2 Interpersonal processes (pp. 60-85). Oxford: Blackwell. Collins, N., & Feeney, B. (2000). A safe haven: An attachment theory perspective on supportseeking and caregiving in intimate relationships. Journal of Personality and Social Psychology, 78, 1053-1073. Collins, N., & Feeney, B. (2004). Working models of attachment shape perceptions of social support: Evidence from experimental and observational studies. Journal of Personality and Social Psychology, 87, 363-383. Collins N., & Read, S. (1994). Cognitive representations of attachment: The structure and function of working models. In K. Bartholomew & D. Perlman (Eds.), Advances in personal relationships: Vol. 5. Attachment processes in adulthood (pp. 53-90). London: Jessica Kingsley. Cox, T., & Cox, M. (2000). Multidimensional scaling (2nd edition). Boca Raton, FL: Chapman and Hall/CRC. Everitt, B., Landau, S., & Leese, M. (2001). Cluster analysis (4th edition). London: Arnold. Fehr, B. (1988). Prototype analysis of the concepts of love and commitment. Journal of Personality and Social Psychology, 55, 557-579.
Varieties of Religious Attachment y 43 Fraley, R., & Waller, N. (1998). Adult attachment patterns: A test of the typological model. In J. Simpson & W. Rholes (Eds.), Attachment theory and close relationships (pp. 77-114). New York: Guilford. Fraley, R., Waller, N., & Brennan, K. (2000). An item response theory analysis of self-report measures of adult attachment. Journal of Personality and Social Psychology, 78, 360-375. Granqvist, P. (1998). Religiousness and perceived childhood attachment: On the question of compensation or correspondence. Journal for the Scientific Study of Religion, 37, 350367. Granqvist, P. (2002). Attachment and religiosity in adolescence: Cross-sectional and longitudinal evaluations. Personality and Social Psychology Bulletin, 28, 260-270. Granqvist, P., & Hagekull, B. (1999). Religiousness and perceived childhood attachment: Profiling socialized correspondence and emotional compensation. Journal for the Scientific Study of Religion, 38, 254-273. Granqvist, P., & Hagekull, B. (2000). Religiosity, adult attachment, and why “singles” are more religious. International Journal for the Psychology of Religion, 10, 111-123. Granqvist, P., & Hagekull, B. (2001). Seeking security in the new age: On attachment and emotional compensation. Journal for the Scientific Study of Religion, 40, 529-547. Granqvist, P., & Hagekull, B. (2003). Longitudinal predictions of religious change in adolescence: Contributions from the interaction of attachment and relationship status. Journal of Social and Personal Relationships, 20, 793-817. Granqvist, P., & Kirkpatrick, L. (2004). Religious conversion and perceived childhood attachment: A meta-analysis. International Journal for the Psychology of Religion, 14,
Varieties of Religious Attachment y 44 223-250. Griffin, D., & Bartholomew, K. (1994a). Models of the self and other: Fundamental dimensions underlying measures of adult attachment. Journal of Personality and Social Psychology, 67, 430-445. Griffin, D., & Bartholomew, K. (1994b). The metaphysics of measurement: The case of adult attachment. In K. Bartholomew & D. Perlman (Eds.), Advances in personal relationships: Vol. 5. Attachment processes in adulthood (pp. 17-52). London: Jessica Kingsley. Hair, J., Black, B., Babin, B., Anderson, R., & Tatham, R. (2005). Multivariate data analysis (6th edition). New York: Prentice Hall. Hu, L., & Bentler, P. (1999). Cutoff criteria for fit indices in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1-55. Hutsebaut, D. & Verhoeven, D. (1995). Studying dimensions of God representation: Choosing closed or open-ended research questions. International Journal for the Psychology of Religion, 5, 49-60. James, W. (1902/2002). The varieties of religious experience (Centenary Edition). London: Routledge. Kihlstrom, J., Beer, J., & Klein, S. (2003). Self and identity as memory. In M. Leary & J. Tangney (Eds.), Handbook of self and identity (pp. 68-90). New York: Guilford. Kihlstrom, J., & Cantor, N. (2000). Social intelligence. In R.J. Sternberg (Ed.), Handbook of intelligence (pp. 359-379). New York: Cambridge.
Varieties of Religious Attachment y 45 Kihlstrom, J., Cantor, N., Albright, J., Chew, B., Klein, S., & Niedenthal, P. (1988). Information processing and the study of the self. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 21, pp. 145-177). New York: Academic. Kihlstrom, J., Marchese-Foster, L., & Klein, S. (1997). Situating the self in interpersonal space. In U. Neisser (Ed.), The conceptual self in context: Culture, experience, selfunderstanding (pp. 154-175). New York: Cambridge. Kirkpatrick, L. (1992). An attachment-theory approach to the psychology of religion. International Journal for the Psychology of Religion, 2, 3-28. Kirkpatrick, L. (1997). A longitudinal study of changes in religious belief and behavior as a function of individual differences in adult attachment style. Journal for the Scientific Study of Religion, 36, 207-217. Kirkpatrick, L. (1998). God as a substitute attachment figure: A longitudinal study of adult attachment style and religious change in college students. Personality and Social Psychology Bulletin, 24, 961-973. Kirkpatrick, L. (1999). Attachment and religious representations and behavior. In J. Cassidy, & P. Shaver (Eds.), Handbook of attachment: Theory, research and clinical applications (pp. 803-822). New York: Guilford. Kirkpatrick, L., & Shaver, P. (1990). Attachment theory and religion: Childhood attachments, religious beliefs, and conversion. Journal for the Scientific Study of Religion, 29, 315-334. Kirkpatrick, L., & Shaver, P. (1992). An attachment-theoretical approach to romantic love and religious belief. Personality and Social Psychology Bulletin, 18, 266-275.
Varieties of Religious Attachment y 46 Laham, D. (1997). Latent semantic analysis approaches to categorization. In M. Shafto & P. Langley (Eds.), Proceedings of the 19th Annual Conference of the Cognitive Science Society (p. 979). Hillsdale, NJ: Erlbaum. Landauer, T., Foltz, P., & Laham, D. (1998). An introduction to latent semantic analysis. Discourse Processes, 3, 259-284. Lawrence, R. (1997). Measuring the image of God: The God Image Inventory and the God Image Scales. Journal of Psychology and Theology, 25, 214-226. Matsuba, M., & Walker, L.J. (2004). Extraordinary moral commitment: Young adults involved in social organizations. Journal of Personality, 72, 413-436. McAdams, D., Anyidoho, N., Brown, C., Huang, Y., Kaplan, B., & Machado, M. (2004). Traits and stories: Links between dispositional and narrative features of personality. Journal of Personality, 72, 761-784. McRae, R., & Costa, P. (1990). Personality in adulthood. New York: Guilford. McRae, R., Costa, P., & Busch, C. (1986). Evaluating comprehensiveness in personality systems: The California Q-set and the five-factor model. Journal of Personality, 54, 430446. Mikulincer, M., Dolev, T., & Shaver, P. (2004). Attachment-related strategies during thought suppression: Ironic rebounds and vulnerable self-representations. Journal of Personality and Social Psychology, 87, 940-956. Northoff, G., & Bermpohl, F. (2004). Cortical midline structures and the self. Trends in Cognitive Sciences, 8, 102-107.
Varieties of Religious Attachment y 47 Obegi, J., Morrison, T., & Shaver, P. (2004). Exploring intergenerational transmission of attachment style in young female adults and their mothers. Journal of Social and Personal Relationships, 21, 625-638. Reimer, K., & Wade-Stein, D. (2004). Moral identity in adolescence: Self and other in semantic space. Identity, 4, 229-249. Rizzuto, A. (1979). The birth of the living God: A psychoanalytic study. Chicago: University of Chicago Press. Rowatt, W., & Kirkpatrick, L. (2002). Two dimensions of attachment to God and their relation to affect, religiosity, and personality constructs. Journal for the Scientific Study of Religion, 41, 637-651. Schaap-Jonker, H., Eurelings-Bontekoe, E., Verhagen, P., & Zock, H. (2002). Image of God and personality pathology: An exploratory study among psychiatric patients. Mental Health, Religion & Culture, 5, 55-71. Shaver, P., & Mikulincer, M. (2004). What do self-report attachment measures assess? In W. Rholes & J. Simpson (Eds.), Adult attachment: Theory, research, and clinical implications (pp. 17-54). New York: Guilford. Shweder, R., Much, N., Mahapatra, M., & Park, L. (1997). The “big three” of morality (autonomy, community, divinity) and the “big three” explanations of suffering. In A. Brandt & P. Rozin (Eds.), Morality and health (pp. 119-169). Florence, KY: Taylor & Francis/Routledge. Squire, L. (1992). Memory and the hippocampus: A synthesis from findings with rats, monkeys, and humans. Psychological Review, 99, 195-232.
Varieties of Religious Attachment y 48 Squire, L., & Zola, S. (1998). Episodic memory, semantic memory, and amnesia. Hippocampus, 8, 205-211. Sroufe, L., Egeland, B., Carlson, E., & Collins, W. (2005). Placing early attachment experiences in developmental context: The Minnesota longitudinal study. In K. Grossmann, K. Grossmann, & E. Waters (Eds.), Attachment from infancy to adulthood (pp. 48-70). New York: Guilford. Steyvers, M., & Tenenbaum, J. (2005). The large-scale structure of semantic networks: Statistical analyses and a model of semantic growth. Cognitive Science, 29, 41-78. Tulving, E. (1983). Elements of episodic memory. New York: Oxford. Vargha-Khadem, F., Gadian, D., Watkins, K., Connelly, A., Van Paesschen, W., & Mishkin, M. (1997). Differential effects of early hippocampal pathology on episodic and semantic memory. Science, 277, 376-380. Walker, L.J., & Hennig, K. (2004). Differing conceptions of moral exemplarity: Just, brave, and caring. Journal of Personality and Social Psychology, 86, 629-647. Walker, L.J., & Pitts, R. (1998). Naturalistic conceptions of moral maturity. Developmental Psychology, 34, 403-419.
**p ≤ .01 (2-tailed)
*p ≤ .05 (2-tailed)
Ind./Var. M SD Acc Pres Chal Sec Pre Fear Dis ____________________________________________________________________________________________________________
Table 1 Means, Standard Deviations, and Correlations for GIS and RSQ manifest indicators and variables
Varieties of Religious Attachment y 49
**p ≤ .01 (2-tailed)
Other 2.99 9.23 1.000 ____________________________________________________________________________________________________________
Table 2 Means, Standard Deviations, and Correlations for GIS and RSQ manifest indicators
Varieties of Religious Attachment y 50
Varieties of Religious Attachment y 51 Table 3 Nominated Exemplar Scores on Three RSQ Scales for Positive Working Model Positive Working Model Scales _____________________________________________________ Religious Tradition
Muslim M SD
Christian M SD
Varieties of Religious Attachment y 52 Figure 1. Structural equation model showing relations between positive working model, insecure attachment styles, and perception of the divine. e = error; d = disturbance.
Perception of Divine
d1 e9 e10
Dismissing-Avoidant Preoccupied Positive Working Model d2
Varieties of Religious Attachment y 53 Figure 2. Results of standard maximum likelihood estimates of positive working model, fearful avoidant attachment style, and perception of the divine. e = error; d = disturbance.
Perception of Divine
Positive Working Model d2
X2 (18, N = 216) = 49.26, p < .01; GFI = .95, AGFI = .90, CFI = .97, RMSEA = .09
Varieties of Religious Attachment y 54
Figure 3. Multidimensional scaling dimensions and hierarchical cluster analysis clusters for Jewish exemplar internal working models. Cluster 1 Cluster 2 Cluster 3 0.50
Romantic Partner 0.25
God -0.75 -4
Varieties of Religious Attachment y 55
Figure 4. Multidimensional scaling dimensions and hierarchical cluster analysis clusters for Muslim exemplar internal working models.
Derived Stimulus Configuration
Cluster 1 Cluster 2 Cluster 3
Varieties of Religious Attachment y 56
Figure 5. Multidimensional scaling dimensions and hierarchical cluster analysis clusters for Christian exemplar internal working models. Cluster 1 Cluster 2 Cluster 3
Best Friend Parent Romantic Partner
Self -0.75 -2